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/string.h>
43 #include <linux/hdreg.h>
44 #include <linux/proc_fs.h>
45 #include <linux/random.h>
46 #include <linux/reboot.h>
47 #include <linux/file.h>
48 #include <linux/compat.h>
49 #include <linux/delay.h>
50 #include <linux/raid/md_p.h>
51 #include <linux/raid/md_u.h>
52 #include <linux/slab.h>
57 #define dprintk(x...) ((void)(DEBUG && printk(x)))
61 static void autostart_arrays(int part);
64 static LIST_HEAD(pers_list);
65 static DEFINE_SPINLOCK(pers_lock);
67 static void md_print_devices(void);
69 static DECLARE_WAIT_QUEUE_HEAD(resync_wait);
71 #define MD_BUG(x...) { printk("md: bug in file %s, line %d\n", __FILE__, __LINE__); md_print_devices(); }
74 * Default number of read corrections we'll attempt on an rdev
75 * before ejecting it from the array. We divide the read error
76 * count by 2 for every hour elapsed between read errors.
78 #define MD_DEFAULT_MAX_CORRECTED_READ_ERRORS 20
80 * Current RAID-1,4,5 parallel reconstruction 'guaranteed speed limit'
81 * is 1000 KB/sec, so the extra system load does not show up that much.
82 * Increase it if you want to have more _guaranteed_ speed. Note that
83 * the RAID driver will use the maximum available bandwidth if the IO
84 * subsystem is idle. There is also an 'absolute maximum' reconstruction
85 * speed limit - in case reconstruction slows down your system despite
88 * you can change it via /proc/sys/dev/raid/speed_limit_min and _max.
89 * or /sys/block/mdX/md/sync_speed_{min,max}
92 static int sysctl_speed_limit_min = 1000;
93 static int sysctl_speed_limit_max = 200000;
94 static inline int speed_min(mddev_t *mddev)
96 return mddev->sync_speed_min ?
97 mddev->sync_speed_min : sysctl_speed_limit_min;
100 static inline int speed_max(mddev_t *mddev)
102 return mddev->sync_speed_max ?
103 mddev->sync_speed_max : sysctl_speed_limit_max;
106 static struct ctl_table_header *raid_table_header;
108 static ctl_table raid_table[] = {
110 .procname = "speed_limit_min",
111 .data = &sysctl_speed_limit_min,
112 .maxlen = sizeof(int),
113 .mode = S_IRUGO|S_IWUSR,
114 .proc_handler = proc_dointvec,
117 .procname = "speed_limit_max",
118 .data = &sysctl_speed_limit_max,
119 .maxlen = sizeof(int),
120 .mode = S_IRUGO|S_IWUSR,
121 .proc_handler = proc_dointvec,
126 static ctl_table raid_dir_table[] = {
130 .mode = S_IRUGO|S_IXUGO,
136 static ctl_table raid_root_table[] = {
141 .child = raid_dir_table,
146 static const struct block_device_operations md_fops;
148 static int start_readonly;
151 * We have a system wide 'event count' that is incremented
152 * on any 'interesting' event, and readers of /proc/mdstat
153 * can use 'poll' or 'select' to find out when the event
157 * start array, stop array, error, add device, remove device,
158 * start build, activate spare
160 static DECLARE_WAIT_QUEUE_HEAD(md_event_waiters);
161 static atomic_t md_event_count;
162 void md_new_event(mddev_t *mddev)
164 atomic_inc(&md_event_count);
165 wake_up(&md_event_waiters);
167 EXPORT_SYMBOL_GPL(md_new_event);
169 /* Alternate version that can be called from interrupts
170 * when calling sysfs_notify isn't needed.
172 static void md_new_event_inintr(mddev_t *mddev)
174 atomic_inc(&md_event_count);
175 wake_up(&md_event_waiters);
179 * Enables to iterate over all existing md arrays
180 * all_mddevs_lock protects this list.
182 static LIST_HEAD(all_mddevs);
183 static DEFINE_SPINLOCK(all_mddevs_lock);
187 * iterates through all used mddevs in the system.
188 * We take care to grab the all_mddevs_lock whenever navigating
189 * the list, and to always hold a refcount when unlocked.
190 * Any code which breaks out of this loop while own
191 * a reference to the current mddev and must mddev_put it.
193 #define for_each_mddev(mddev,tmp) \
195 for (({ spin_lock(&all_mddevs_lock); \
196 tmp = all_mddevs.next; \
198 ({ if (tmp != &all_mddevs) \
199 mddev_get(list_entry(tmp, mddev_t, all_mddevs));\
200 spin_unlock(&all_mddevs_lock); \
201 if (mddev) mddev_put(mddev); \
202 mddev = list_entry(tmp, mddev_t, all_mddevs); \
203 tmp != &all_mddevs;}); \
204 ({ spin_lock(&all_mddevs_lock); \
209 /* Rather than calling directly into the personality make_request function,
210 * IO requests come here first so that we can check if the device is
211 * being suspended pending a reconfiguration.
212 * We hold a refcount over the call to ->make_request. By the time that
213 * call has finished, the bio has been linked into some internal structure
214 * and so is visible to ->quiesce(), so we don't need the refcount any more.
216 static int md_make_request(struct request_queue *q, struct bio *bio)
218 const int rw = bio_data_dir(bio);
219 mddev_t *mddev = q->queuedata;
223 if (mddev == NULL || mddev->pers == NULL) {
228 if (mddev->suspended || mddev->barrier) {
231 prepare_to_wait(&mddev->sb_wait, &__wait,
232 TASK_UNINTERRUPTIBLE);
233 if (!mddev->suspended && !mddev->barrier)
239 finish_wait(&mddev->sb_wait, &__wait);
241 atomic_inc(&mddev->active_io);
244 rv = mddev->pers->make_request(mddev, bio);
246 cpu = part_stat_lock();
247 part_stat_inc(cpu, &mddev->gendisk->part0, ios[rw]);
248 part_stat_add(cpu, &mddev->gendisk->part0, sectors[rw],
252 if (atomic_dec_and_test(&mddev->active_io) && mddev->suspended)
253 wake_up(&mddev->sb_wait);
258 /* mddev_suspend makes sure no new requests are submitted
259 * to the device, and that any requests that have been submitted
260 * are completely handled.
261 * Once ->stop is called and completes, the module will be completely
264 static void mddev_suspend(mddev_t *mddev)
266 BUG_ON(mddev->suspended);
267 mddev->suspended = 1;
269 wait_event(mddev->sb_wait, atomic_read(&mddev->active_io) == 0);
270 mddev->pers->quiesce(mddev, 1);
273 static void mddev_resume(mddev_t *mddev)
275 mddev->suspended = 0;
276 wake_up(&mddev->sb_wait);
277 mddev->pers->quiesce(mddev, 0);
280 int mddev_congested(mddev_t *mddev, int bits)
284 return mddev->suspended;
286 EXPORT_SYMBOL(mddev_congested);
289 * Generic barrier handling for md
292 #define POST_REQUEST_BARRIER ((void*)1)
294 static void md_end_barrier(struct bio *bio, int err)
296 mdk_rdev_t *rdev = bio->bi_private;
297 mddev_t *mddev = rdev->mddev;
298 if (err == -EOPNOTSUPP && mddev->barrier != POST_REQUEST_BARRIER)
299 set_bit(BIO_EOPNOTSUPP, &mddev->barrier->bi_flags);
301 rdev_dec_pending(rdev, mddev);
303 if (atomic_dec_and_test(&mddev->flush_pending)) {
304 if (mddev->barrier == POST_REQUEST_BARRIER) {
305 /* This was a post-request barrier */
306 mddev->barrier = NULL;
307 wake_up(&mddev->sb_wait);
309 /* The pre-request barrier has finished */
310 schedule_work(&mddev->barrier_work);
315 static void submit_barriers(mddev_t *mddev)
320 list_for_each_entry_rcu(rdev, &mddev->disks, same_set)
321 if (rdev->raid_disk >= 0 &&
322 !test_bit(Faulty, &rdev->flags)) {
323 /* Take two references, one is dropped
324 * when request finishes, one after
325 * we reclaim rcu_read_lock
328 atomic_inc(&rdev->nr_pending);
329 atomic_inc(&rdev->nr_pending);
331 bi = bio_alloc(GFP_KERNEL, 0);
332 bi->bi_end_io = md_end_barrier;
333 bi->bi_private = rdev;
334 bi->bi_bdev = rdev->bdev;
335 atomic_inc(&mddev->flush_pending);
336 submit_bio(WRITE_BARRIER, bi);
338 rdev_dec_pending(rdev, mddev);
343 static void md_submit_barrier(struct work_struct *ws)
345 mddev_t *mddev = container_of(ws, mddev_t, barrier_work);
346 struct bio *bio = mddev->barrier;
348 atomic_set(&mddev->flush_pending, 1);
350 if (test_bit(BIO_EOPNOTSUPP, &bio->bi_flags))
351 bio_endio(bio, -EOPNOTSUPP);
352 else if (bio->bi_size == 0)
353 /* an empty barrier - all done */
356 bio->bi_rw &= ~(1<<BIO_RW_BARRIER);
357 if (mddev->pers->make_request(mddev, bio))
358 generic_make_request(bio);
359 mddev->barrier = POST_REQUEST_BARRIER;
360 submit_barriers(mddev);
362 if (atomic_dec_and_test(&mddev->flush_pending)) {
363 mddev->barrier = NULL;
364 wake_up(&mddev->sb_wait);
368 void md_barrier_request(mddev_t *mddev, struct bio *bio)
370 spin_lock_irq(&mddev->write_lock);
371 wait_event_lock_irq(mddev->sb_wait,
373 mddev->write_lock, /*nothing*/);
374 mddev->barrier = bio;
375 spin_unlock_irq(&mddev->write_lock);
377 atomic_set(&mddev->flush_pending, 1);
378 INIT_WORK(&mddev->barrier_work, md_submit_barrier);
380 submit_barriers(mddev);
382 if (atomic_dec_and_test(&mddev->flush_pending))
383 schedule_work(&mddev->barrier_work);
385 EXPORT_SYMBOL(md_barrier_request);
387 static inline mddev_t *mddev_get(mddev_t *mddev)
389 atomic_inc(&mddev->active);
393 static void mddev_delayed_delete(struct work_struct *ws);
395 static void mddev_put(mddev_t *mddev)
397 if (!atomic_dec_and_lock(&mddev->active, &all_mddevs_lock))
399 if (!mddev->raid_disks && list_empty(&mddev->disks) &&
400 mddev->ctime == 0 && !mddev->hold_active) {
401 /* Array is not configured at all, and not held active,
403 list_del(&mddev->all_mddevs);
404 if (mddev->gendisk) {
405 /* we did a probe so need to clean up.
406 * Call schedule_work inside the spinlock
407 * so that flush_scheduled_work() after
408 * mddev_find will succeed in waiting for the
411 INIT_WORK(&mddev->del_work, mddev_delayed_delete);
412 schedule_work(&mddev->del_work);
416 spin_unlock(&all_mddevs_lock);
419 static void mddev_init(mddev_t *mddev)
421 mutex_init(&mddev->open_mutex);
422 mutex_init(&mddev->reconfig_mutex);
423 mutex_init(&mddev->bitmap_info.mutex);
424 INIT_LIST_HEAD(&mddev->disks);
425 INIT_LIST_HEAD(&mddev->all_mddevs);
426 init_timer(&mddev->safemode_timer);
427 atomic_set(&mddev->active, 1);
428 atomic_set(&mddev->openers, 0);
429 atomic_set(&mddev->active_io, 0);
430 spin_lock_init(&mddev->write_lock);
431 atomic_set(&mddev->flush_pending, 0);
432 init_waitqueue_head(&mddev->sb_wait);
433 init_waitqueue_head(&mddev->recovery_wait);
434 mddev->reshape_position = MaxSector;
435 mddev->resync_min = 0;
436 mddev->resync_max = MaxSector;
437 mddev->level = LEVEL_NONE;
440 static mddev_t * mddev_find(dev_t unit)
442 mddev_t *mddev, *new = NULL;
445 spin_lock(&all_mddevs_lock);
448 list_for_each_entry(mddev, &all_mddevs, all_mddevs)
449 if (mddev->unit == unit) {
451 spin_unlock(&all_mddevs_lock);
457 list_add(&new->all_mddevs, &all_mddevs);
458 spin_unlock(&all_mddevs_lock);
459 new->hold_active = UNTIL_IOCTL;
463 /* find an unused unit number */
464 static int next_minor = 512;
465 int start = next_minor;
469 dev = MKDEV(MD_MAJOR, next_minor);
471 if (next_minor > MINORMASK)
473 if (next_minor == start) {
474 /* Oh dear, all in use. */
475 spin_unlock(&all_mddevs_lock);
481 list_for_each_entry(mddev, &all_mddevs, all_mddevs)
482 if (mddev->unit == dev) {
488 new->md_minor = MINOR(dev);
489 new->hold_active = UNTIL_STOP;
490 list_add(&new->all_mddevs, &all_mddevs);
491 spin_unlock(&all_mddevs_lock);
494 spin_unlock(&all_mddevs_lock);
496 new = kzalloc(sizeof(*new), GFP_KERNEL);
501 if (MAJOR(unit) == MD_MAJOR)
502 new->md_minor = MINOR(unit);
504 new->md_minor = MINOR(unit) >> MdpMinorShift;
511 static inline int mddev_lock(mddev_t * mddev)
513 return mutex_lock_interruptible(&mddev->reconfig_mutex);
516 static inline int mddev_is_locked(mddev_t *mddev)
518 return mutex_is_locked(&mddev->reconfig_mutex);
521 static inline int mddev_trylock(mddev_t * mddev)
523 return mutex_trylock(&mddev->reconfig_mutex);
526 static struct attribute_group md_redundancy_group;
528 static void mddev_unlock(mddev_t * mddev)
530 if (mddev->to_remove) {
531 /* These cannot be removed under reconfig_mutex as
532 * an access to the files will try to take reconfig_mutex
533 * while holding the file unremovable, which leads to
535 * So hold open_mutex instead - we are allowed to take
536 * it while holding reconfig_mutex, and md_run can
537 * use it to wait for the remove to complete.
539 struct attribute_group *to_remove = mddev->to_remove;
540 mddev->to_remove = NULL;
541 mutex_lock(&mddev->open_mutex);
542 mutex_unlock(&mddev->reconfig_mutex);
544 if (mddev->kobj.sd) {
545 if (to_remove != &md_redundancy_group)
546 sysfs_remove_group(&mddev->kobj, to_remove);
547 if (mddev->pers == NULL ||
548 mddev->pers->sync_request == NULL) {
549 sysfs_remove_group(&mddev->kobj, &md_redundancy_group);
550 if (mddev->sysfs_action)
551 sysfs_put(mddev->sysfs_action);
552 mddev->sysfs_action = NULL;
555 mutex_unlock(&mddev->open_mutex);
557 mutex_unlock(&mddev->reconfig_mutex);
559 md_wakeup_thread(mddev->thread);
562 static mdk_rdev_t * find_rdev_nr(mddev_t *mddev, int nr)
566 list_for_each_entry(rdev, &mddev->disks, same_set)
567 if (rdev->desc_nr == nr)
573 static mdk_rdev_t * find_rdev(mddev_t * mddev, dev_t dev)
577 list_for_each_entry(rdev, &mddev->disks, same_set)
578 if (rdev->bdev->bd_dev == dev)
584 static struct mdk_personality *find_pers(int level, char *clevel)
586 struct mdk_personality *pers;
587 list_for_each_entry(pers, &pers_list, list) {
588 if (level != LEVEL_NONE && pers->level == level)
590 if (strcmp(pers->name, clevel)==0)
596 /* return the offset of the super block in 512byte sectors */
597 static inline sector_t calc_dev_sboffset(struct block_device *bdev)
599 sector_t num_sectors = bdev->bd_inode->i_size / 512;
600 return MD_NEW_SIZE_SECTORS(num_sectors);
603 static int alloc_disk_sb(mdk_rdev_t * rdev)
608 rdev->sb_page = alloc_page(GFP_KERNEL);
609 if (!rdev->sb_page) {
610 printk(KERN_ALERT "md: out of memory.\n");
617 static void free_disk_sb(mdk_rdev_t * rdev)
620 put_page(rdev->sb_page);
622 rdev->sb_page = NULL;
629 static void super_written(struct bio *bio, int error)
631 mdk_rdev_t *rdev = bio->bi_private;
632 mddev_t *mddev = rdev->mddev;
634 if (error || !test_bit(BIO_UPTODATE, &bio->bi_flags)) {
635 printk("md: super_written gets error=%d, uptodate=%d\n",
636 error, test_bit(BIO_UPTODATE, &bio->bi_flags));
637 WARN_ON(test_bit(BIO_UPTODATE, &bio->bi_flags));
638 md_error(mddev, rdev);
641 if (atomic_dec_and_test(&mddev->pending_writes))
642 wake_up(&mddev->sb_wait);
646 static void super_written_barrier(struct bio *bio, int error)
648 struct bio *bio2 = bio->bi_private;
649 mdk_rdev_t *rdev = bio2->bi_private;
650 mddev_t *mddev = rdev->mddev;
652 if (!test_bit(BIO_UPTODATE, &bio->bi_flags) &&
653 error == -EOPNOTSUPP) {
655 /* barriers don't appear to be supported :-( */
656 set_bit(BarriersNotsupp, &rdev->flags);
657 mddev->barriers_work = 0;
658 spin_lock_irqsave(&mddev->write_lock, flags);
659 bio2->bi_next = mddev->biolist;
660 mddev->biolist = bio2;
661 spin_unlock_irqrestore(&mddev->write_lock, flags);
662 wake_up(&mddev->sb_wait);
666 bio->bi_private = rdev;
667 super_written(bio, error);
671 void md_super_write(mddev_t *mddev, mdk_rdev_t *rdev,
672 sector_t sector, int size, struct page *page)
674 /* write first size bytes of page to sector of rdev
675 * Increment mddev->pending_writes before returning
676 * and decrement it on completion, waking up sb_wait
677 * if zero is reached.
678 * If an error occurred, call md_error
680 * As we might need to resubmit the request if BIO_RW_BARRIER
681 * causes ENOTSUPP, we allocate a spare bio...
683 struct bio *bio = bio_alloc(GFP_NOIO, 1);
684 int rw = (1<<BIO_RW) | (1<<BIO_RW_SYNCIO) | (1<<BIO_RW_UNPLUG);
686 bio->bi_bdev = rdev->bdev;
687 bio->bi_sector = sector;
688 bio_add_page(bio, page, size, 0);
689 bio->bi_private = rdev;
690 bio->bi_end_io = super_written;
693 atomic_inc(&mddev->pending_writes);
694 if (!test_bit(BarriersNotsupp, &rdev->flags)) {
696 rw |= (1<<BIO_RW_BARRIER);
697 rbio = bio_clone(bio, GFP_NOIO);
698 rbio->bi_private = bio;
699 rbio->bi_end_io = super_written_barrier;
700 submit_bio(rw, rbio);
705 void md_super_wait(mddev_t *mddev)
707 /* wait for all superblock writes that were scheduled to complete.
708 * if any had to be retried (due to BARRIER problems), retry them
712 prepare_to_wait(&mddev->sb_wait, &wq, TASK_UNINTERRUPTIBLE);
713 if (atomic_read(&mddev->pending_writes)==0)
715 while (mddev->biolist) {
717 spin_lock_irq(&mddev->write_lock);
718 bio = mddev->biolist;
719 mddev->biolist = bio->bi_next ;
721 spin_unlock_irq(&mddev->write_lock);
722 submit_bio(bio->bi_rw, bio);
726 finish_wait(&mddev->sb_wait, &wq);
729 static void bi_complete(struct bio *bio, int error)
731 complete((struct completion*)bio->bi_private);
734 int sync_page_io(struct block_device *bdev, sector_t sector, int size,
735 struct page *page, int rw)
737 struct bio *bio = bio_alloc(GFP_NOIO, 1);
738 struct completion event;
741 rw |= (1 << BIO_RW_SYNCIO) | (1 << BIO_RW_UNPLUG);
744 bio->bi_sector = sector;
745 bio_add_page(bio, page, size, 0);
746 init_completion(&event);
747 bio->bi_private = &event;
748 bio->bi_end_io = bi_complete;
750 wait_for_completion(&event);
752 ret = test_bit(BIO_UPTODATE, &bio->bi_flags);
756 EXPORT_SYMBOL_GPL(sync_page_io);
758 static int read_disk_sb(mdk_rdev_t * rdev, int size)
760 char b[BDEVNAME_SIZE];
761 if (!rdev->sb_page) {
769 if (!sync_page_io(rdev->bdev, rdev->sb_start, size, rdev->sb_page, READ))
775 printk(KERN_WARNING "md: disabled device %s, could not read superblock.\n",
776 bdevname(rdev->bdev,b));
780 static int uuid_equal(mdp_super_t *sb1, mdp_super_t *sb2)
782 return sb1->set_uuid0 == sb2->set_uuid0 &&
783 sb1->set_uuid1 == sb2->set_uuid1 &&
784 sb1->set_uuid2 == sb2->set_uuid2 &&
785 sb1->set_uuid3 == sb2->set_uuid3;
788 static int sb_equal(mdp_super_t *sb1, mdp_super_t *sb2)
791 mdp_super_t *tmp1, *tmp2;
793 tmp1 = kmalloc(sizeof(*tmp1),GFP_KERNEL);
794 tmp2 = kmalloc(sizeof(*tmp2),GFP_KERNEL);
796 if (!tmp1 || !tmp2) {
798 printk(KERN_INFO "md.c sb_equal(): failed to allocate memory!\n");
806 * nr_disks is not constant
811 ret = (memcmp(tmp1, tmp2, MD_SB_GENERIC_CONSTANT_WORDS * 4) == 0);
819 static u32 md_csum_fold(u32 csum)
821 csum = (csum & 0xffff) + (csum >> 16);
822 return (csum & 0xffff) + (csum >> 16);
825 static unsigned int calc_sb_csum(mdp_super_t * sb)
828 u32 *sb32 = (u32*)sb;
830 unsigned int disk_csum, csum;
832 disk_csum = sb->sb_csum;
835 for (i = 0; i < MD_SB_BYTES/4 ; i++)
837 csum = (newcsum & 0xffffffff) + (newcsum>>32);
841 /* This used to use csum_partial, which was wrong for several
842 * reasons including that different results are returned on
843 * different architectures. It isn't critical that we get exactly
844 * the same return value as before (we always csum_fold before
845 * testing, and that removes any differences). However as we
846 * know that csum_partial always returned a 16bit value on
847 * alphas, do a fold to maximise conformity to previous behaviour.
849 sb->sb_csum = md_csum_fold(disk_csum);
851 sb->sb_csum = disk_csum;
858 * Handle superblock details.
859 * We want to be able to handle multiple superblock formats
860 * so we have a common interface to them all, and an array of
861 * different handlers.
862 * We rely on user-space to write the initial superblock, and support
863 * reading and updating of superblocks.
864 * Interface methods are:
865 * int load_super(mdk_rdev_t *dev, mdk_rdev_t *refdev, int minor_version)
866 * loads and validates a superblock on dev.
867 * if refdev != NULL, compare superblocks on both devices
869 * 0 - dev has a superblock that is compatible with refdev
870 * 1 - dev has a superblock that is compatible and newer than refdev
871 * so dev should be used as the refdev in future
872 * -EINVAL superblock incompatible or invalid
873 * -othererror e.g. -EIO
875 * int validate_super(mddev_t *mddev, mdk_rdev_t *dev)
876 * Verify that dev is acceptable into mddev.
877 * The first time, mddev->raid_disks will be 0, and data from
878 * dev should be merged in. Subsequent calls check that dev
879 * is new enough. Return 0 or -EINVAL
881 * void sync_super(mddev_t *mddev, mdk_rdev_t *dev)
882 * Update the superblock for rdev with data in mddev
883 * This does not write to disc.
889 struct module *owner;
890 int (*load_super)(mdk_rdev_t *rdev, mdk_rdev_t *refdev,
892 int (*validate_super)(mddev_t *mddev, mdk_rdev_t *rdev);
893 void (*sync_super)(mddev_t *mddev, mdk_rdev_t *rdev);
894 unsigned long long (*rdev_size_change)(mdk_rdev_t *rdev,
895 sector_t num_sectors);
899 * Check that the given mddev has no bitmap.
901 * This function is called from the run method of all personalities that do not
902 * support bitmaps. It prints an error message and returns non-zero if mddev
903 * has a bitmap. Otherwise, it returns 0.
906 int md_check_no_bitmap(mddev_t *mddev)
908 if (!mddev->bitmap_info.file && !mddev->bitmap_info.offset)
910 printk(KERN_ERR "%s: bitmaps are not supported for %s\n",
911 mdname(mddev), mddev->pers->name);
914 EXPORT_SYMBOL(md_check_no_bitmap);
917 * load_super for 0.90.0
919 static int super_90_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
921 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
926 * Calculate the position of the superblock (512byte sectors),
927 * it's at the end of the disk.
929 * It also happens to be a multiple of 4Kb.
931 rdev->sb_start = calc_dev_sboffset(rdev->bdev);
933 ret = read_disk_sb(rdev, MD_SB_BYTES);
938 bdevname(rdev->bdev, b);
939 sb = (mdp_super_t*)page_address(rdev->sb_page);
941 if (sb->md_magic != MD_SB_MAGIC) {
942 printk(KERN_ERR "md: invalid raid superblock magic on %s\n",
947 if (sb->major_version != 0 ||
948 sb->minor_version < 90 ||
949 sb->minor_version > 91) {
950 printk(KERN_WARNING "Bad version number %d.%d on %s\n",
951 sb->major_version, sb->minor_version,
956 if (sb->raid_disks <= 0)
959 if (md_csum_fold(calc_sb_csum(sb)) != md_csum_fold(sb->sb_csum)) {
960 printk(KERN_WARNING "md: invalid superblock checksum on %s\n",
965 rdev->preferred_minor = sb->md_minor;
966 rdev->data_offset = 0;
967 rdev->sb_size = MD_SB_BYTES;
969 if (sb->level == LEVEL_MULTIPATH)
972 rdev->desc_nr = sb->this_disk.number;
978 mdp_super_t *refsb = (mdp_super_t*)page_address(refdev->sb_page);
979 if (!uuid_equal(refsb, sb)) {
980 printk(KERN_WARNING "md: %s has different UUID to %s\n",
981 b, bdevname(refdev->bdev,b2));
984 if (!sb_equal(refsb, sb)) {
985 printk(KERN_WARNING "md: %s has same UUID"
986 " but different superblock to %s\n",
987 b, bdevname(refdev->bdev, b2));
991 ev2 = md_event(refsb);
997 rdev->sectors = rdev->sb_start;
999 if (rdev->sectors < sb->size * 2 && sb->level > 1)
1000 /* "this cannot possibly happen" ... */
1008 * validate_super for 0.90.0
1010 static int super_90_validate(mddev_t *mddev, mdk_rdev_t *rdev)
1013 mdp_super_t *sb = (mdp_super_t *)page_address(rdev->sb_page);
1014 __u64 ev1 = md_event(sb);
1016 rdev->raid_disk = -1;
1017 clear_bit(Faulty, &rdev->flags);
1018 clear_bit(In_sync, &rdev->flags);
1019 clear_bit(WriteMostly, &rdev->flags);
1020 clear_bit(BarriersNotsupp, &rdev->flags);
1022 if (mddev->raid_disks == 0) {
1023 mddev->major_version = 0;
1024 mddev->minor_version = sb->minor_version;
1025 mddev->patch_version = sb->patch_version;
1026 mddev->external = 0;
1027 mddev->chunk_sectors = sb->chunk_size >> 9;
1028 mddev->ctime = sb->ctime;
1029 mddev->utime = sb->utime;
1030 mddev->level = sb->level;
1031 mddev->clevel[0] = 0;
1032 mddev->layout = sb->layout;
1033 mddev->raid_disks = sb->raid_disks;
1034 mddev->dev_sectors = sb->size * 2;
1035 mddev->events = ev1;
1036 mddev->bitmap_info.offset = 0;
1037 mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9;
1039 if (mddev->minor_version >= 91) {
1040 mddev->reshape_position = sb->reshape_position;
1041 mddev->delta_disks = sb->delta_disks;
1042 mddev->new_level = sb->new_level;
1043 mddev->new_layout = sb->new_layout;
1044 mddev->new_chunk_sectors = sb->new_chunk >> 9;
1046 mddev->reshape_position = MaxSector;
1047 mddev->delta_disks = 0;
1048 mddev->new_level = mddev->level;
1049 mddev->new_layout = mddev->layout;
1050 mddev->new_chunk_sectors = mddev->chunk_sectors;
1053 if (sb->state & (1<<MD_SB_CLEAN))
1054 mddev->recovery_cp = MaxSector;
1056 if (sb->events_hi == sb->cp_events_hi &&
1057 sb->events_lo == sb->cp_events_lo) {
1058 mddev->recovery_cp = sb->recovery_cp;
1060 mddev->recovery_cp = 0;
1063 memcpy(mddev->uuid+0, &sb->set_uuid0, 4);
1064 memcpy(mddev->uuid+4, &sb->set_uuid1, 4);
1065 memcpy(mddev->uuid+8, &sb->set_uuid2, 4);
1066 memcpy(mddev->uuid+12,&sb->set_uuid3, 4);
1068 mddev->max_disks = MD_SB_DISKS;
1070 if (sb->state & (1<<MD_SB_BITMAP_PRESENT) &&
1071 mddev->bitmap_info.file == NULL)
1072 mddev->bitmap_info.offset =
1073 mddev->bitmap_info.default_offset;
1075 } else if (mddev->pers == NULL) {
1076 /* Insist on good event counter while assembling, except
1077 * for spares (which don't need an event count) */
1079 if (sb->disks[rdev->desc_nr].state & (
1080 (1<<MD_DISK_SYNC) | (1 << MD_DISK_ACTIVE)))
1081 if (ev1 < mddev->events)
1083 } else if (mddev->bitmap) {
1084 /* if adding to array with a bitmap, then we can accept an
1085 * older device ... but not too old.
1087 if (ev1 < mddev->bitmap->events_cleared)
1090 if (ev1 < mddev->events)
1091 /* just a hot-add of a new device, leave raid_disk at -1 */
1095 if (mddev->level != LEVEL_MULTIPATH) {
1096 desc = sb->disks + rdev->desc_nr;
1098 if (desc->state & (1<<MD_DISK_FAULTY))
1099 set_bit(Faulty, &rdev->flags);
1100 else if (desc->state & (1<<MD_DISK_SYNC) /* &&
1101 desc->raid_disk < mddev->raid_disks */) {
1102 set_bit(In_sync, &rdev->flags);
1103 rdev->raid_disk = desc->raid_disk;
1104 } else if (desc->state & (1<<MD_DISK_ACTIVE)) {
1105 /* active but not in sync implies recovery up to
1106 * reshape position. We don't know exactly where
1107 * that is, so set to zero for now */
1108 if (mddev->minor_version >= 91) {
1109 rdev->recovery_offset = 0;
1110 rdev->raid_disk = desc->raid_disk;
1113 if (desc->state & (1<<MD_DISK_WRITEMOSTLY))
1114 set_bit(WriteMostly, &rdev->flags);
1115 } else /* MULTIPATH are always insync */
1116 set_bit(In_sync, &rdev->flags);
1121 * sync_super for 0.90.0
1123 static void super_90_sync(mddev_t *mddev, mdk_rdev_t *rdev)
1127 int next_spare = mddev->raid_disks;
1130 /* make rdev->sb match mddev data..
1133 * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
1134 * 3/ any empty disks < next_spare become removed
1136 * disks[0] gets initialised to REMOVED because
1137 * we cannot be sure from other fields if it has
1138 * been initialised or not.
1141 int active=0, working=0,failed=0,spare=0,nr_disks=0;
1143 rdev->sb_size = MD_SB_BYTES;
1145 sb = (mdp_super_t*)page_address(rdev->sb_page);
1147 memset(sb, 0, sizeof(*sb));
1149 sb->md_magic = MD_SB_MAGIC;
1150 sb->major_version = mddev->major_version;
1151 sb->patch_version = mddev->patch_version;
1152 sb->gvalid_words = 0; /* ignored */
1153 memcpy(&sb->set_uuid0, mddev->uuid+0, 4);
1154 memcpy(&sb->set_uuid1, mddev->uuid+4, 4);
1155 memcpy(&sb->set_uuid2, mddev->uuid+8, 4);
1156 memcpy(&sb->set_uuid3, mddev->uuid+12,4);
1158 sb->ctime = mddev->ctime;
1159 sb->level = mddev->level;
1160 sb->size = mddev->dev_sectors / 2;
1161 sb->raid_disks = mddev->raid_disks;
1162 sb->md_minor = mddev->md_minor;
1163 sb->not_persistent = 0;
1164 sb->utime = mddev->utime;
1166 sb->events_hi = (mddev->events>>32);
1167 sb->events_lo = (u32)mddev->events;
1169 if (mddev->reshape_position == MaxSector)
1170 sb->minor_version = 90;
1172 sb->minor_version = 91;
1173 sb->reshape_position = mddev->reshape_position;
1174 sb->new_level = mddev->new_level;
1175 sb->delta_disks = mddev->delta_disks;
1176 sb->new_layout = mddev->new_layout;
1177 sb->new_chunk = mddev->new_chunk_sectors << 9;
1179 mddev->minor_version = sb->minor_version;
1182 sb->recovery_cp = mddev->recovery_cp;
1183 sb->cp_events_hi = (mddev->events>>32);
1184 sb->cp_events_lo = (u32)mddev->events;
1185 if (mddev->recovery_cp == MaxSector)
1186 sb->state = (1<< MD_SB_CLEAN);
1188 sb->recovery_cp = 0;
1190 sb->layout = mddev->layout;
1191 sb->chunk_size = mddev->chunk_sectors << 9;
1193 if (mddev->bitmap && mddev->bitmap_info.file == NULL)
1194 sb->state |= (1<<MD_SB_BITMAP_PRESENT);
1196 sb->disks[0].state = (1<<MD_DISK_REMOVED);
1197 list_for_each_entry(rdev2, &mddev->disks, same_set) {
1200 int is_active = test_bit(In_sync, &rdev2->flags);
1202 if (rdev2->raid_disk >= 0 &&
1203 sb->minor_version >= 91)
1204 /* we have nowhere to store the recovery_offset,
1205 * but if it is not below the reshape_position,
1206 * we can piggy-back on that.
1209 if (rdev2->raid_disk < 0 ||
1210 test_bit(Faulty, &rdev2->flags))
1213 desc_nr = rdev2->raid_disk;
1215 desc_nr = next_spare++;
1216 rdev2->desc_nr = desc_nr;
1217 d = &sb->disks[rdev2->desc_nr];
1219 d->number = rdev2->desc_nr;
1220 d->major = MAJOR(rdev2->bdev->bd_dev);
1221 d->minor = MINOR(rdev2->bdev->bd_dev);
1223 d->raid_disk = rdev2->raid_disk;
1225 d->raid_disk = rdev2->desc_nr; /* compatibility */
1226 if (test_bit(Faulty, &rdev2->flags))
1227 d->state = (1<<MD_DISK_FAULTY);
1228 else if (is_active) {
1229 d->state = (1<<MD_DISK_ACTIVE);
1230 if (test_bit(In_sync, &rdev2->flags))
1231 d->state |= (1<<MD_DISK_SYNC);
1239 if (test_bit(WriteMostly, &rdev2->flags))
1240 d->state |= (1<<MD_DISK_WRITEMOSTLY);
1242 /* now set the "removed" and "faulty" bits on any missing devices */
1243 for (i=0 ; i < mddev->raid_disks ; i++) {
1244 mdp_disk_t *d = &sb->disks[i];
1245 if (d->state == 0 && d->number == 0) {
1248 d->state = (1<<MD_DISK_REMOVED);
1249 d->state |= (1<<MD_DISK_FAULTY);
1253 sb->nr_disks = nr_disks;
1254 sb->active_disks = active;
1255 sb->working_disks = working;
1256 sb->failed_disks = failed;
1257 sb->spare_disks = spare;
1259 sb->this_disk = sb->disks[rdev->desc_nr];
1260 sb->sb_csum = calc_sb_csum(sb);
1264 * rdev_size_change for 0.90.0
1266 static unsigned long long
1267 super_90_rdev_size_change(mdk_rdev_t *rdev, sector_t num_sectors)
1269 if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
1270 return 0; /* component must fit device */
1271 if (rdev->mddev->bitmap_info.offset)
1272 return 0; /* can't move bitmap */
1273 rdev->sb_start = calc_dev_sboffset(rdev->bdev);
1274 if (!num_sectors || num_sectors > rdev->sb_start)
1275 num_sectors = rdev->sb_start;
1276 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
1278 md_super_wait(rdev->mddev);
1279 return num_sectors / 2; /* kB for sysfs */
1284 * version 1 superblock
1287 static __le32 calc_sb_1_csum(struct mdp_superblock_1 * sb)
1291 unsigned long long newcsum;
1292 int size = 256 + le32_to_cpu(sb->max_dev)*2;
1293 __le32 *isuper = (__le32*)sb;
1296 disk_csum = sb->sb_csum;
1299 for (i=0; size>=4; size -= 4 )
1300 newcsum += le32_to_cpu(*isuper++);
1303 newcsum += le16_to_cpu(*(__le16*) isuper);
1305 csum = (newcsum & 0xffffffff) + (newcsum >> 32);
1306 sb->sb_csum = disk_csum;
1307 return cpu_to_le32(csum);
1310 static int super_1_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
1312 struct mdp_superblock_1 *sb;
1315 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
1319 * Calculate the position of the superblock in 512byte sectors.
1320 * It is always aligned to a 4K boundary and
1321 * depeding on minor_version, it can be:
1322 * 0: At least 8K, but less than 12K, from end of device
1323 * 1: At start of device
1324 * 2: 4K from start of device.
1326 switch(minor_version) {
1328 sb_start = rdev->bdev->bd_inode->i_size >> 9;
1330 sb_start &= ~(sector_t)(4*2-1);
1341 rdev->sb_start = sb_start;
1343 /* superblock is rarely larger than 1K, but it can be larger,
1344 * and it is safe to read 4k, so we do that
1346 ret = read_disk_sb(rdev, 4096);
1347 if (ret) return ret;
1350 sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1352 if (sb->magic != cpu_to_le32(MD_SB_MAGIC) ||
1353 sb->major_version != cpu_to_le32(1) ||
1354 le32_to_cpu(sb->max_dev) > (4096-256)/2 ||
1355 le64_to_cpu(sb->super_offset) != rdev->sb_start ||
1356 (le32_to_cpu(sb->feature_map) & ~MD_FEATURE_ALL) != 0)
1359 if (calc_sb_1_csum(sb) != sb->sb_csum) {
1360 printk("md: invalid superblock checksum on %s\n",
1361 bdevname(rdev->bdev,b));
1364 if (le64_to_cpu(sb->data_size) < 10) {
1365 printk("md: data_size too small on %s\n",
1366 bdevname(rdev->bdev,b));
1370 rdev->preferred_minor = 0xffff;
1371 rdev->data_offset = le64_to_cpu(sb->data_offset);
1372 atomic_set(&rdev->corrected_errors, le32_to_cpu(sb->cnt_corrected_read));
1374 rdev->sb_size = le32_to_cpu(sb->max_dev) * 2 + 256;
1375 bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
1376 if (rdev->sb_size & bmask)
1377 rdev->sb_size = (rdev->sb_size | bmask) + 1;
1380 && rdev->data_offset < sb_start + (rdev->sb_size/512))
1383 if (sb->level == cpu_to_le32(LEVEL_MULTIPATH))
1386 rdev->desc_nr = le32_to_cpu(sb->dev_number);
1392 struct mdp_superblock_1 *refsb =
1393 (struct mdp_superblock_1*)page_address(refdev->sb_page);
1395 if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 ||
1396 sb->level != refsb->level ||
1397 sb->layout != refsb->layout ||
1398 sb->chunksize != refsb->chunksize) {
1399 printk(KERN_WARNING "md: %s has strangely different"
1400 " superblock to %s\n",
1401 bdevname(rdev->bdev,b),
1402 bdevname(refdev->bdev,b2));
1405 ev1 = le64_to_cpu(sb->events);
1406 ev2 = le64_to_cpu(refsb->events);
1414 rdev->sectors = (rdev->bdev->bd_inode->i_size >> 9) -
1415 le64_to_cpu(sb->data_offset);
1417 rdev->sectors = rdev->sb_start;
1418 if (rdev->sectors < le64_to_cpu(sb->data_size))
1420 rdev->sectors = le64_to_cpu(sb->data_size);
1421 if (le64_to_cpu(sb->size) > rdev->sectors)
1426 static int super_1_validate(mddev_t *mddev, mdk_rdev_t *rdev)
1428 struct mdp_superblock_1 *sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1429 __u64 ev1 = le64_to_cpu(sb->events);
1431 rdev->raid_disk = -1;
1432 clear_bit(Faulty, &rdev->flags);
1433 clear_bit(In_sync, &rdev->flags);
1434 clear_bit(WriteMostly, &rdev->flags);
1435 clear_bit(BarriersNotsupp, &rdev->flags);
1437 if (mddev->raid_disks == 0) {
1438 mddev->major_version = 1;
1439 mddev->patch_version = 0;
1440 mddev->external = 0;
1441 mddev->chunk_sectors = le32_to_cpu(sb->chunksize);
1442 mddev->ctime = le64_to_cpu(sb->ctime) & ((1ULL << 32)-1);
1443 mddev->utime = le64_to_cpu(sb->utime) & ((1ULL << 32)-1);
1444 mddev->level = le32_to_cpu(sb->level);
1445 mddev->clevel[0] = 0;
1446 mddev->layout = le32_to_cpu(sb->layout);
1447 mddev->raid_disks = le32_to_cpu(sb->raid_disks);
1448 mddev->dev_sectors = le64_to_cpu(sb->size);
1449 mddev->events = ev1;
1450 mddev->bitmap_info.offset = 0;
1451 mddev->bitmap_info.default_offset = 1024 >> 9;
1453 mddev->recovery_cp = le64_to_cpu(sb->resync_offset);
1454 memcpy(mddev->uuid, sb->set_uuid, 16);
1456 mddev->max_disks = (4096-256)/2;
1458 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) &&
1459 mddev->bitmap_info.file == NULL )
1460 mddev->bitmap_info.offset =
1461 (__s32)le32_to_cpu(sb->bitmap_offset);
1463 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) {
1464 mddev->reshape_position = le64_to_cpu(sb->reshape_position);
1465 mddev->delta_disks = le32_to_cpu(sb->delta_disks);
1466 mddev->new_level = le32_to_cpu(sb->new_level);
1467 mddev->new_layout = le32_to_cpu(sb->new_layout);
1468 mddev->new_chunk_sectors = le32_to_cpu(sb->new_chunk);
1470 mddev->reshape_position = MaxSector;
1471 mddev->delta_disks = 0;
1472 mddev->new_level = mddev->level;
1473 mddev->new_layout = mddev->layout;
1474 mddev->new_chunk_sectors = mddev->chunk_sectors;
1477 } else if (mddev->pers == NULL) {
1478 /* Insist of good event counter while assembling, except for
1479 * spares (which don't need an event count) */
1481 if (rdev->desc_nr >= 0 &&
1482 rdev->desc_nr < le32_to_cpu(sb->max_dev) &&
1483 le16_to_cpu(sb->dev_roles[rdev->desc_nr]) < 0xfffe)
1484 if (ev1 < mddev->events)
1486 } else if (mddev->bitmap) {
1487 /* If adding to array with a bitmap, then we can accept an
1488 * older device, but not too old.
1490 if (ev1 < mddev->bitmap->events_cleared)
1493 if (ev1 < mddev->events)
1494 /* just a hot-add of a new device, leave raid_disk at -1 */
1497 if (mddev->level != LEVEL_MULTIPATH) {
1499 if (rdev->desc_nr < 0 ||
1500 rdev->desc_nr >= le32_to_cpu(sb->max_dev)) {
1504 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
1506 case 0xffff: /* spare */
1508 case 0xfffe: /* faulty */
1509 set_bit(Faulty, &rdev->flags);
1512 if ((le32_to_cpu(sb->feature_map) &
1513 MD_FEATURE_RECOVERY_OFFSET))
1514 rdev->recovery_offset = le64_to_cpu(sb->recovery_offset);
1516 set_bit(In_sync, &rdev->flags);
1517 rdev->raid_disk = role;
1520 if (sb->devflags & WriteMostly1)
1521 set_bit(WriteMostly, &rdev->flags);
1522 } else /* MULTIPATH are always insync */
1523 set_bit(In_sync, &rdev->flags);
1528 static void super_1_sync(mddev_t *mddev, mdk_rdev_t *rdev)
1530 struct mdp_superblock_1 *sb;
1533 /* make rdev->sb match mddev and rdev data. */
1535 sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1537 sb->feature_map = 0;
1539 sb->recovery_offset = cpu_to_le64(0);
1540 memset(sb->pad1, 0, sizeof(sb->pad1));
1541 memset(sb->pad2, 0, sizeof(sb->pad2));
1542 memset(sb->pad3, 0, sizeof(sb->pad3));
1544 sb->utime = cpu_to_le64((__u64)mddev->utime);
1545 sb->events = cpu_to_le64(mddev->events);
1547 sb->resync_offset = cpu_to_le64(mddev->recovery_cp);
1549 sb->resync_offset = cpu_to_le64(0);
1551 sb->cnt_corrected_read = cpu_to_le32(atomic_read(&rdev->corrected_errors));
1553 sb->raid_disks = cpu_to_le32(mddev->raid_disks);
1554 sb->size = cpu_to_le64(mddev->dev_sectors);
1555 sb->chunksize = cpu_to_le32(mddev->chunk_sectors);
1556 sb->level = cpu_to_le32(mddev->level);
1557 sb->layout = cpu_to_le32(mddev->layout);
1559 if (mddev->bitmap && mddev->bitmap_info.file == NULL) {
1560 sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_info.offset);
1561 sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET);
1564 if (rdev->raid_disk >= 0 &&
1565 !test_bit(In_sync, &rdev->flags)) {
1567 cpu_to_le32(MD_FEATURE_RECOVERY_OFFSET);
1568 sb->recovery_offset =
1569 cpu_to_le64(rdev->recovery_offset);
1572 if (mddev->reshape_position != MaxSector) {
1573 sb->feature_map |= cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE);
1574 sb->reshape_position = cpu_to_le64(mddev->reshape_position);
1575 sb->new_layout = cpu_to_le32(mddev->new_layout);
1576 sb->delta_disks = cpu_to_le32(mddev->delta_disks);
1577 sb->new_level = cpu_to_le32(mddev->new_level);
1578 sb->new_chunk = cpu_to_le32(mddev->new_chunk_sectors);
1582 list_for_each_entry(rdev2, &mddev->disks, same_set)
1583 if (rdev2->desc_nr+1 > max_dev)
1584 max_dev = rdev2->desc_nr+1;
1586 if (max_dev > le32_to_cpu(sb->max_dev)) {
1588 sb->max_dev = cpu_to_le32(max_dev);
1589 rdev->sb_size = max_dev * 2 + 256;
1590 bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
1591 if (rdev->sb_size & bmask)
1592 rdev->sb_size = (rdev->sb_size | bmask) + 1;
1594 for (i=0; i<max_dev;i++)
1595 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1597 list_for_each_entry(rdev2, &mddev->disks, same_set) {
1599 if (test_bit(Faulty, &rdev2->flags))
1600 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1601 else if (test_bit(In_sync, &rdev2->flags))
1602 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1603 else if (rdev2->raid_disk >= 0)
1604 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1606 sb->dev_roles[i] = cpu_to_le16(0xffff);
1609 sb->sb_csum = calc_sb_1_csum(sb);
1612 static unsigned long long
1613 super_1_rdev_size_change(mdk_rdev_t *rdev, sector_t num_sectors)
1615 struct mdp_superblock_1 *sb;
1616 sector_t max_sectors;
1617 if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
1618 return 0; /* component must fit device */
1619 if (rdev->sb_start < rdev->data_offset) {
1620 /* minor versions 1 and 2; superblock before data */
1621 max_sectors = rdev->bdev->bd_inode->i_size >> 9;
1622 max_sectors -= rdev->data_offset;
1623 if (!num_sectors || num_sectors > max_sectors)
1624 num_sectors = max_sectors;
1625 } else if (rdev->mddev->bitmap_info.offset) {
1626 /* minor version 0 with bitmap we can't move */
1629 /* minor version 0; superblock after data */
1631 sb_start = (rdev->bdev->bd_inode->i_size >> 9) - 8*2;
1632 sb_start &= ~(sector_t)(4*2 - 1);
1633 max_sectors = rdev->sectors + sb_start - rdev->sb_start;
1634 if (!num_sectors || num_sectors > max_sectors)
1635 num_sectors = max_sectors;
1636 rdev->sb_start = sb_start;
1638 sb = (struct mdp_superblock_1 *) page_address(rdev->sb_page);
1639 sb->data_size = cpu_to_le64(num_sectors);
1640 sb->super_offset = rdev->sb_start;
1641 sb->sb_csum = calc_sb_1_csum(sb);
1642 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
1644 md_super_wait(rdev->mddev);
1645 return num_sectors / 2; /* kB for sysfs */
1648 static struct super_type super_types[] = {
1651 .owner = THIS_MODULE,
1652 .load_super = super_90_load,
1653 .validate_super = super_90_validate,
1654 .sync_super = super_90_sync,
1655 .rdev_size_change = super_90_rdev_size_change,
1659 .owner = THIS_MODULE,
1660 .load_super = super_1_load,
1661 .validate_super = super_1_validate,
1662 .sync_super = super_1_sync,
1663 .rdev_size_change = super_1_rdev_size_change,
1667 static int match_mddev_units(mddev_t *mddev1, mddev_t *mddev2)
1669 mdk_rdev_t *rdev, *rdev2;
1672 rdev_for_each_rcu(rdev, mddev1)
1673 rdev_for_each_rcu(rdev2, mddev2)
1674 if (rdev->bdev->bd_contains ==
1675 rdev2->bdev->bd_contains) {
1683 static LIST_HEAD(pending_raid_disks);
1686 * Try to register data integrity profile for an mddev
1688 * This is called when an array is started and after a disk has been kicked
1689 * from the array. It only succeeds if all working and active component devices
1690 * are integrity capable with matching profiles.
1692 int md_integrity_register(mddev_t *mddev)
1694 mdk_rdev_t *rdev, *reference = NULL;
1696 if (list_empty(&mddev->disks))
1697 return 0; /* nothing to do */
1698 if (blk_get_integrity(mddev->gendisk))
1699 return 0; /* already registered */
1700 list_for_each_entry(rdev, &mddev->disks, same_set) {
1701 /* skip spares and non-functional disks */
1702 if (test_bit(Faulty, &rdev->flags))
1704 if (rdev->raid_disk < 0)
1707 * If at least one rdev is not integrity capable, we can not
1708 * enable data integrity for the md device.
1710 if (!bdev_get_integrity(rdev->bdev))
1713 /* Use the first rdev as the reference */
1717 /* does this rdev's profile match the reference profile? */
1718 if (blk_integrity_compare(reference->bdev->bd_disk,
1719 rdev->bdev->bd_disk) < 0)
1723 * All component devices are integrity capable and have matching
1724 * profiles, register the common profile for the md device.
1726 if (blk_integrity_register(mddev->gendisk,
1727 bdev_get_integrity(reference->bdev)) != 0) {
1728 printk(KERN_ERR "md: failed to register integrity for %s\n",
1732 printk(KERN_NOTICE "md: data integrity on %s enabled\n",
1736 EXPORT_SYMBOL(md_integrity_register);
1738 /* Disable data integrity if non-capable/non-matching disk is being added */
1739 void md_integrity_add_rdev(mdk_rdev_t *rdev, mddev_t *mddev)
1741 struct blk_integrity *bi_rdev = bdev_get_integrity(rdev->bdev);
1742 struct blk_integrity *bi_mddev = blk_get_integrity(mddev->gendisk);
1744 if (!bi_mddev) /* nothing to do */
1746 if (rdev->raid_disk < 0) /* skip spares */
1748 if (bi_rdev && blk_integrity_compare(mddev->gendisk,
1749 rdev->bdev->bd_disk) >= 0)
1751 printk(KERN_NOTICE "disabling data integrity on %s\n", mdname(mddev));
1752 blk_integrity_unregister(mddev->gendisk);
1754 EXPORT_SYMBOL(md_integrity_add_rdev);
1756 static int bind_rdev_to_array(mdk_rdev_t * rdev, mddev_t * mddev)
1758 char b[BDEVNAME_SIZE];
1768 /* prevent duplicates */
1769 if (find_rdev(mddev, rdev->bdev->bd_dev))
1772 /* make sure rdev->sectors exceeds mddev->dev_sectors */
1773 if (rdev->sectors && (mddev->dev_sectors == 0 ||
1774 rdev->sectors < mddev->dev_sectors)) {
1776 /* Cannot change size, so fail
1777 * If mddev->level <= 0, then we don't care
1778 * about aligning sizes (e.g. linear)
1780 if (mddev->level > 0)
1783 mddev->dev_sectors = rdev->sectors;
1786 /* Verify rdev->desc_nr is unique.
1787 * If it is -1, assign a free number, else
1788 * check number is not in use
1790 if (rdev->desc_nr < 0) {
1792 if (mddev->pers) choice = mddev->raid_disks;
1793 while (find_rdev_nr(mddev, choice))
1795 rdev->desc_nr = choice;
1797 if (find_rdev_nr(mddev, rdev->desc_nr))
1800 if (mddev->max_disks && rdev->desc_nr >= mddev->max_disks) {
1801 printk(KERN_WARNING "md: %s: array is limited to %d devices\n",
1802 mdname(mddev), mddev->max_disks);
1805 bdevname(rdev->bdev,b);
1806 while ( (s=strchr(b, '/')) != NULL)
1809 rdev->mddev = mddev;
1810 printk(KERN_INFO "md: bind<%s>\n", b);
1812 if ((err = kobject_add(&rdev->kobj, &mddev->kobj, "dev-%s", b)))
1815 ko = &part_to_dev(rdev->bdev->bd_part)->kobj;
1816 if (sysfs_create_link(&rdev->kobj, ko, "block"))
1817 /* failure here is OK */;
1818 rdev->sysfs_state = sysfs_get_dirent_safe(rdev->kobj.sd, "state");
1820 list_add_rcu(&rdev->same_set, &mddev->disks);
1821 bd_claim_by_disk(rdev->bdev, rdev->bdev->bd_holder, mddev->gendisk);
1823 /* May as well allow recovery to be retried once */
1824 mddev->recovery_disabled = 0;
1829 printk(KERN_WARNING "md: failed to register dev-%s for %s\n",
1834 static void md_delayed_delete(struct work_struct *ws)
1836 mdk_rdev_t *rdev = container_of(ws, mdk_rdev_t, del_work);
1837 kobject_del(&rdev->kobj);
1838 kobject_put(&rdev->kobj);
1841 static void unbind_rdev_from_array(mdk_rdev_t * rdev)
1843 char b[BDEVNAME_SIZE];
1848 bd_release_from_disk(rdev->bdev, rdev->mddev->gendisk);
1849 list_del_rcu(&rdev->same_set);
1850 printk(KERN_INFO "md: unbind<%s>\n", bdevname(rdev->bdev,b));
1852 sysfs_remove_link(&rdev->kobj, "block");
1853 sysfs_put(rdev->sysfs_state);
1854 rdev->sysfs_state = NULL;
1855 /* We need to delay this, otherwise we can deadlock when
1856 * writing to 'remove' to "dev/state". We also need
1857 * to delay it due to rcu usage.
1860 INIT_WORK(&rdev->del_work, md_delayed_delete);
1861 kobject_get(&rdev->kobj);
1862 schedule_work(&rdev->del_work);
1866 * prevent the device from being mounted, repartitioned or
1867 * otherwise reused by a RAID array (or any other kernel
1868 * subsystem), by bd_claiming the device.
1870 static int lock_rdev(mdk_rdev_t *rdev, dev_t dev, int shared)
1873 struct block_device *bdev;
1874 char b[BDEVNAME_SIZE];
1876 bdev = open_by_devnum(dev, FMODE_READ|FMODE_WRITE);
1878 printk(KERN_ERR "md: could not open %s.\n",
1879 __bdevname(dev, b));
1880 return PTR_ERR(bdev);
1882 err = bd_claim(bdev, shared ? (mdk_rdev_t *)lock_rdev : rdev);
1884 printk(KERN_ERR "md: could not bd_claim %s.\n",
1886 blkdev_put(bdev, FMODE_READ|FMODE_WRITE);
1890 set_bit(AllReserved, &rdev->flags);
1895 static void unlock_rdev(mdk_rdev_t *rdev)
1897 struct block_device *bdev = rdev->bdev;
1902 blkdev_put(bdev, FMODE_READ|FMODE_WRITE);
1905 void md_autodetect_dev(dev_t dev);
1907 static void export_rdev(mdk_rdev_t * rdev)
1909 char b[BDEVNAME_SIZE];
1910 printk(KERN_INFO "md: export_rdev(%s)\n",
1911 bdevname(rdev->bdev,b));
1916 if (test_bit(AutoDetected, &rdev->flags))
1917 md_autodetect_dev(rdev->bdev->bd_dev);
1920 kobject_put(&rdev->kobj);
1923 static void kick_rdev_from_array(mdk_rdev_t * rdev)
1925 unbind_rdev_from_array(rdev);
1929 static void export_array(mddev_t *mddev)
1931 mdk_rdev_t *rdev, *tmp;
1933 rdev_for_each(rdev, tmp, mddev) {
1938 kick_rdev_from_array(rdev);
1940 if (!list_empty(&mddev->disks))
1942 mddev->raid_disks = 0;
1943 mddev->major_version = 0;
1946 static void print_desc(mdp_disk_t *desc)
1948 printk(" DISK<N:%d,(%d,%d),R:%d,S:%d>\n", desc->number,
1949 desc->major,desc->minor,desc->raid_disk,desc->state);
1952 static void print_sb_90(mdp_super_t *sb)
1957 "md: SB: (V:%d.%d.%d) ID:<%08x.%08x.%08x.%08x> CT:%08x\n",
1958 sb->major_version, sb->minor_version, sb->patch_version,
1959 sb->set_uuid0, sb->set_uuid1, sb->set_uuid2, sb->set_uuid3,
1961 printk(KERN_INFO "md: L%d S%08d ND:%d RD:%d md%d LO:%d CS:%d\n",
1962 sb->level, sb->size, sb->nr_disks, sb->raid_disks,
1963 sb->md_minor, sb->layout, sb->chunk_size);
1964 printk(KERN_INFO "md: UT:%08x ST:%d AD:%d WD:%d"
1965 " FD:%d SD:%d CSUM:%08x E:%08lx\n",
1966 sb->utime, sb->state, sb->active_disks, sb->working_disks,
1967 sb->failed_disks, sb->spare_disks,
1968 sb->sb_csum, (unsigned long)sb->events_lo);
1971 for (i = 0; i < MD_SB_DISKS; i++) {
1974 desc = sb->disks + i;
1975 if (desc->number || desc->major || desc->minor ||
1976 desc->raid_disk || (desc->state && (desc->state != 4))) {
1977 printk(" D %2d: ", i);
1981 printk(KERN_INFO "md: THIS: ");
1982 print_desc(&sb->this_disk);
1985 static void print_sb_1(struct mdp_superblock_1 *sb)
1989 uuid = sb->set_uuid;
1991 "md: SB: (V:%u) (F:0x%08x) Array-ID:<%pU>\n"
1992 "md: Name: \"%s\" CT:%llu\n",
1993 le32_to_cpu(sb->major_version),
1994 le32_to_cpu(sb->feature_map),
1997 (unsigned long long)le64_to_cpu(sb->ctime)
1998 & MD_SUPERBLOCK_1_TIME_SEC_MASK);
2000 uuid = sb->device_uuid;
2002 "md: L%u SZ%llu RD:%u LO:%u CS:%u DO:%llu DS:%llu SO:%llu"
2004 "md: Dev:%08x UUID: %pU\n"
2005 "md: (F:0x%08x) UT:%llu Events:%llu ResyncOffset:%llu CSUM:0x%08x\n"
2006 "md: (MaxDev:%u) \n",
2007 le32_to_cpu(sb->level),
2008 (unsigned long long)le64_to_cpu(sb->size),
2009 le32_to_cpu(sb->raid_disks),
2010 le32_to_cpu(sb->layout),
2011 le32_to_cpu(sb->chunksize),
2012 (unsigned long long)le64_to_cpu(sb->data_offset),
2013 (unsigned long long)le64_to_cpu(sb->data_size),
2014 (unsigned long long)le64_to_cpu(sb->super_offset),
2015 (unsigned long long)le64_to_cpu(sb->recovery_offset),
2016 le32_to_cpu(sb->dev_number),
2019 (unsigned long long)le64_to_cpu(sb->utime) & MD_SUPERBLOCK_1_TIME_SEC_MASK,
2020 (unsigned long long)le64_to_cpu(sb->events),
2021 (unsigned long long)le64_to_cpu(sb->resync_offset),
2022 le32_to_cpu(sb->sb_csum),
2023 le32_to_cpu(sb->max_dev)
2027 static void print_rdev(mdk_rdev_t *rdev, int major_version)
2029 char b[BDEVNAME_SIZE];
2030 printk(KERN_INFO "md: rdev %s, Sect:%08llu F:%d S:%d DN:%u\n",
2031 bdevname(rdev->bdev, b), (unsigned long long)rdev->sectors,
2032 test_bit(Faulty, &rdev->flags), test_bit(In_sync, &rdev->flags),
2034 if (rdev->sb_loaded) {
2035 printk(KERN_INFO "md: rdev superblock (MJ:%d):\n", major_version);
2036 switch (major_version) {
2038 print_sb_90((mdp_super_t*)page_address(rdev->sb_page));
2041 print_sb_1((struct mdp_superblock_1 *)page_address(rdev->sb_page));
2045 printk(KERN_INFO "md: no rdev superblock!\n");
2048 static void md_print_devices(void)
2050 struct list_head *tmp;
2053 char b[BDEVNAME_SIZE];
2056 printk("md: **********************************\n");
2057 printk("md: * <COMPLETE RAID STATE PRINTOUT> *\n");
2058 printk("md: **********************************\n");
2059 for_each_mddev(mddev, tmp) {
2062 bitmap_print_sb(mddev->bitmap);
2064 printk("%s: ", mdname(mddev));
2065 list_for_each_entry(rdev, &mddev->disks, same_set)
2066 printk("<%s>", bdevname(rdev->bdev,b));
2069 list_for_each_entry(rdev, &mddev->disks, same_set)
2070 print_rdev(rdev, mddev->major_version);
2072 printk("md: **********************************\n");
2077 static void sync_sbs(mddev_t * mddev, int nospares)
2079 /* Update each superblock (in-memory image), but
2080 * if we are allowed to, skip spares which already
2081 * have the right event counter, or have one earlier
2082 * (which would mean they aren't being marked as dirty
2083 * with the rest of the array)
2087 /* First make sure individual recovery_offsets are correct */
2088 list_for_each_entry(rdev, &mddev->disks, same_set) {
2089 if (rdev->raid_disk >= 0 &&
2090 mddev->delta_disks >= 0 &&
2091 !test_bit(In_sync, &rdev->flags) &&
2092 mddev->curr_resync_completed > rdev->recovery_offset)
2093 rdev->recovery_offset = mddev->curr_resync_completed;
2096 list_for_each_entry(rdev, &mddev->disks, same_set) {
2097 if (rdev->sb_events == mddev->events ||
2099 rdev->raid_disk < 0 &&
2100 rdev->sb_events+1 == mddev->events)) {
2101 /* Don't update this superblock */
2102 rdev->sb_loaded = 2;
2104 super_types[mddev->major_version].
2105 sync_super(mddev, rdev);
2106 rdev->sb_loaded = 1;
2111 static void md_update_sb(mddev_t * mddev, int force_change)
2117 mddev->utime = get_seconds();
2118 if (mddev->external)
2121 spin_lock_irq(&mddev->write_lock);
2123 set_bit(MD_CHANGE_PENDING, &mddev->flags);
2124 if (test_and_clear_bit(MD_CHANGE_DEVS, &mddev->flags))
2126 if (test_and_clear_bit(MD_CHANGE_CLEAN, &mddev->flags))
2127 /* just a clean<-> dirty transition, possibly leave spares alone,
2128 * though if events isn't the right even/odd, we will have to do
2134 if (mddev->degraded)
2135 /* If the array is degraded, then skipping spares is both
2136 * dangerous and fairly pointless.
2137 * Dangerous because a device that was removed from the array
2138 * might have a event_count that still looks up-to-date,
2139 * so it can be re-added without a resync.
2140 * Pointless because if there are any spares to skip,
2141 * then a recovery will happen and soon that array won't
2142 * be degraded any more and the spare can go back to sleep then.
2146 sync_req = mddev->in_sync;
2148 /* If this is just a dirty<->clean transition, and the array is clean
2149 * and 'events' is odd, we can roll back to the previous clean state */
2151 && (mddev->in_sync && mddev->recovery_cp == MaxSector)
2152 && mddev->can_decrease_events
2153 && mddev->events != 1) {
2155 mddev->can_decrease_events = 0;
2157 /* otherwise we have to go forward and ... */
2159 mddev->can_decrease_events = nospares;
2162 if (!mddev->events) {
2164 * oops, this 64-bit counter should never wrap.
2165 * Either we are in around ~1 trillion A.C., assuming
2166 * 1 reboot per second, or we have a bug:
2173 * do not write anything to disk if using
2174 * nonpersistent superblocks
2176 if (!mddev->persistent) {
2177 if (!mddev->external)
2178 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
2180 spin_unlock_irq(&mddev->write_lock);
2181 wake_up(&mddev->sb_wait);
2184 sync_sbs(mddev, nospares);
2185 spin_unlock_irq(&mddev->write_lock);
2188 "md: updating %s RAID superblock on device (in sync %d)\n",
2189 mdname(mddev),mddev->in_sync);
2191 bitmap_update_sb(mddev->bitmap);
2192 list_for_each_entry(rdev, &mddev->disks, same_set) {
2193 char b[BDEVNAME_SIZE];
2194 dprintk(KERN_INFO "md: ");
2195 if (rdev->sb_loaded != 1)
2196 continue; /* no noise on spare devices */
2197 if (test_bit(Faulty, &rdev->flags))
2198 dprintk("(skipping faulty ");
2200 dprintk("%s ", bdevname(rdev->bdev,b));
2201 if (!test_bit(Faulty, &rdev->flags)) {
2202 md_super_write(mddev,rdev,
2203 rdev->sb_start, rdev->sb_size,
2205 dprintk(KERN_INFO "(write) %s's sb offset: %llu\n",
2206 bdevname(rdev->bdev,b),
2207 (unsigned long long)rdev->sb_start);
2208 rdev->sb_events = mddev->events;
2212 if (mddev->level == LEVEL_MULTIPATH)
2213 /* only need to write one superblock... */
2216 md_super_wait(mddev);
2217 /* if there was a failure, MD_CHANGE_DEVS was set, and we re-write super */
2219 spin_lock_irq(&mddev->write_lock);
2220 if (mddev->in_sync != sync_req ||
2221 test_bit(MD_CHANGE_DEVS, &mddev->flags)) {
2222 /* have to write it out again */
2223 spin_unlock_irq(&mddev->write_lock);
2226 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
2227 spin_unlock_irq(&mddev->write_lock);
2228 wake_up(&mddev->sb_wait);
2229 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
2230 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
2234 /* words written to sysfs files may, or may not, be \n terminated.
2235 * We want to accept with case. For this we use cmd_match.
2237 static int cmd_match(const char *cmd, const char *str)
2239 /* See if cmd, written into a sysfs file, matches
2240 * str. They must either be the same, or cmd can
2241 * have a trailing newline
2243 while (*cmd && *str && *cmd == *str) {
2254 struct rdev_sysfs_entry {
2255 struct attribute attr;
2256 ssize_t (*show)(mdk_rdev_t *, char *);
2257 ssize_t (*store)(mdk_rdev_t *, const char *, size_t);
2261 state_show(mdk_rdev_t *rdev, char *page)
2266 if (test_bit(Faulty, &rdev->flags)) {
2267 len+= sprintf(page+len, "%sfaulty",sep);
2270 if (test_bit(In_sync, &rdev->flags)) {
2271 len += sprintf(page+len, "%sin_sync",sep);
2274 if (test_bit(WriteMostly, &rdev->flags)) {
2275 len += sprintf(page+len, "%swrite_mostly",sep);
2278 if (test_bit(Blocked, &rdev->flags)) {
2279 len += sprintf(page+len, "%sblocked", sep);
2282 if (!test_bit(Faulty, &rdev->flags) &&
2283 !test_bit(In_sync, &rdev->flags)) {
2284 len += sprintf(page+len, "%sspare", sep);
2287 return len+sprintf(page+len, "\n");
2291 state_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2294 * faulty - simulates and error
2295 * remove - disconnects the device
2296 * writemostly - sets write_mostly
2297 * -writemostly - clears write_mostly
2298 * blocked - sets the Blocked flag
2299 * -blocked - clears the Blocked flag
2300 * insync - sets Insync providing device isn't active
2303 if (cmd_match(buf, "faulty") && rdev->mddev->pers) {
2304 md_error(rdev->mddev, rdev);
2306 } else if (cmd_match(buf, "remove")) {
2307 if (rdev->raid_disk >= 0)
2310 mddev_t *mddev = rdev->mddev;
2311 kick_rdev_from_array(rdev);
2313 md_update_sb(mddev, 1);
2314 md_new_event(mddev);
2317 } else if (cmd_match(buf, "writemostly")) {
2318 set_bit(WriteMostly, &rdev->flags);
2320 } else if (cmd_match(buf, "-writemostly")) {
2321 clear_bit(WriteMostly, &rdev->flags);
2323 } else if (cmd_match(buf, "blocked")) {
2324 set_bit(Blocked, &rdev->flags);
2326 } else if (cmd_match(buf, "-blocked")) {
2327 clear_bit(Blocked, &rdev->flags);
2328 wake_up(&rdev->blocked_wait);
2329 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2330 md_wakeup_thread(rdev->mddev->thread);
2333 } else if (cmd_match(buf, "insync") && rdev->raid_disk == -1) {
2334 set_bit(In_sync, &rdev->flags);
2338 sysfs_notify_dirent_safe(rdev->sysfs_state);
2339 return err ? err : len;
2341 static struct rdev_sysfs_entry rdev_state =
2342 __ATTR(state, S_IRUGO|S_IWUSR, state_show, state_store);
2345 errors_show(mdk_rdev_t *rdev, char *page)
2347 return sprintf(page, "%d\n", atomic_read(&rdev->corrected_errors));
2351 errors_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2354 unsigned long n = simple_strtoul(buf, &e, 10);
2355 if (*buf && (*e == 0 || *e == '\n')) {
2356 atomic_set(&rdev->corrected_errors, n);
2361 static struct rdev_sysfs_entry rdev_errors =
2362 __ATTR(errors, S_IRUGO|S_IWUSR, errors_show, errors_store);
2365 slot_show(mdk_rdev_t *rdev, char *page)
2367 if (rdev->raid_disk < 0)
2368 return sprintf(page, "none\n");
2370 return sprintf(page, "%d\n", rdev->raid_disk);
2374 slot_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2379 int slot = simple_strtoul(buf, &e, 10);
2380 if (strncmp(buf, "none", 4)==0)
2382 else if (e==buf || (*e && *e!= '\n'))
2384 if (rdev->mddev->pers && slot == -1) {
2385 /* Setting 'slot' on an active array requires also
2386 * updating the 'rd%d' link, and communicating
2387 * with the personality with ->hot_*_disk.
2388 * For now we only support removing
2389 * failed/spare devices. This normally happens automatically,
2390 * but not when the metadata is externally managed.
2392 if (rdev->raid_disk == -1)
2394 /* personality does all needed checks */
2395 if (rdev->mddev->pers->hot_add_disk == NULL)
2397 err = rdev->mddev->pers->
2398 hot_remove_disk(rdev->mddev, rdev->raid_disk);
2401 sprintf(nm, "rd%d", rdev->raid_disk);
2402 sysfs_remove_link(&rdev->mddev->kobj, nm);
2403 rdev->raid_disk = -1;
2404 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2405 md_wakeup_thread(rdev->mddev->thread);
2406 } else if (rdev->mddev->pers) {
2408 /* Activating a spare .. or possibly reactivating
2409 * if we ever get bitmaps working here.
2412 if (rdev->raid_disk != -1)
2415 if (rdev->mddev->pers->hot_add_disk == NULL)
2418 list_for_each_entry(rdev2, &rdev->mddev->disks, same_set)
2419 if (rdev2->raid_disk == slot)
2422 rdev->raid_disk = slot;
2423 if (test_bit(In_sync, &rdev->flags))
2424 rdev->saved_raid_disk = slot;
2426 rdev->saved_raid_disk = -1;
2427 err = rdev->mddev->pers->
2428 hot_add_disk(rdev->mddev, rdev);
2430 rdev->raid_disk = -1;
2433 sysfs_notify_dirent_safe(rdev->sysfs_state);
2434 sprintf(nm, "rd%d", rdev->raid_disk);
2435 if (sysfs_create_link(&rdev->mddev->kobj, &rdev->kobj, nm))
2436 /* failure here is OK */;
2437 /* don't wakeup anyone, leave that to userspace. */
2439 if (slot >= rdev->mddev->raid_disks)
2441 rdev->raid_disk = slot;
2442 /* assume it is working */
2443 clear_bit(Faulty, &rdev->flags);
2444 clear_bit(WriteMostly, &rdev->flags);
2445 set_bit(In_sync, &rdev->flags);
2446 sysfs_notify_dirent_safe(rdev->sysfs_state);
2452 static struct rdev_sysfs_entry rdev_slot =
2453 __ATTR(slot, S_IRUGO|S_IWUSR, slot_show, slot_store);
2456 offset_show(mdk_rdev_t *rdev, char *page)
2458 return sprintf(page, "%llu\n", (unsigned long long)rdev->data_offset);
2462 offset_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2465 unsigned long long offset = simple_strtoull(buf, &e, 10);
2466 if (e==buf || (*e && *e != '\n'))
2468 if (rdev->mddev->pers && rdev->raid_disk >= 0)
2470 if (rdev->sectors && rdev->mddev->external)
2471 /* Must set offset before size, so overlap checks
2474 rdev->data_offset = offset;
2478 static struct rdev_sysfs_entry rdev_offset =
2479 __ATTR(offset, S_IRUGO|S_IWUSR, offset_show, offset_store);
2482 rdev_size_show(mdk_rdev_t *rdev, char *page)
2484 return sprintf(page, "%llu\n", (unsigned long long)rdev->sectors / 2);
2487 static int overlaps(sector_t s1, sector_t l1, sector_t s2, sector_t l2)
2489 /* check if two start/length pairs overlap */
2497 static int strict_blocks_to_sectors(const char *buf, sector_t *sectors)
2499 unsigned long long blocks;
2502 if (strict_strtoull(buf, 10, &blocks) < 0)
2505 if (blocks & 1ULL << (8 * sizeof(blocks) - 1))
2506 return -EINVAL; /* sector conversion overflow */
2509 if (new != blocks * 2)
2510 return -EINVAL; /* unsigned long long to sector_t overflow */
2517 rdev_size_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2519 mddev_t *my_mddev = rdev->mddev;
2520 sector_t oldsectors = rdev->sectors;
2523 if (strict_blocks_to_sectors(buf, §ors) < 0)
2525 if (my_mddev->pers && rdev->raid_disk >= 0) {
2526 if (my_mddev->persistent) {
2527 sectors = super_types[my_mddev->major_version].
2528 rdev_size_change(rdev, sectors);
2531 } else if (!sectors)
2532 sectors = (rdev->bdev->bd_inode->i_size >> 9) -
2535 if (sectors < my_mddev->dev_sectors)
2536 return -EINVAL; /* component must fit device */
2538 rdev->sectors = sectors;
2539 if (sectors > oldsectors && my_mddev->external) {
2540 /* need to check that all other rdevs with the same ->bdev
2541 * do not overlap. We need to unlock the mddev to avoid
2542 * a deadlock. We have already changed rdev->sectors, and if
2543 * we have to change it back, we will have the lock again.
2547 struct list_head *tmp;
2549 mddev_unlock(my_mddev);
2550 for_each_mddev(mddev, tmp) {
2554 list_for_each_entry(rdev2, &mddev->disks, same_set)
2555 if (test_bit(AllReserved, &rdev2->flags) ||
2556 (rdev->bdev == rdev2->bdev &&
2558 overlaps(rdev->data_offset, rdev->sectors,
2564 mddev_unlock(mddev);
2570 mddev_lock(my_mddev);
2572 /* Someone else could have slipped in a size
2573 * change here, but doing so is just silly.
2574 * We put oldsectors back because we *know* it is
2575 * safe, and trust userspace not to race with
2578 rdev->sectors = oldsectors;
2585 static struct rdev_sysfs_entry rdev_size =
2586 __ATTR(size, S_IRUGO|S_IWUSR, rdev_size_show, rdev_size_store);
2589 static ssize_t recovery_start_show(mdk_rdev_t *rdev, char *page)
2591 unsigned long long recovery_start = rdev->recovery_offset;
2593 if (test_bit(In_sync, &rdev->flags) ||
2594 recovery_start == MaxSector)
2595 return sprintf(page, "none\n");
2597 return sprintf(page, "%llu\n", recovery_start);
2600 static ssize_t recovery_start_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2602 unsigned long long recovery_start;
2604 if (cmd_match(buf, "none"))
2605 recovery_start = MaxSector;
2606 else if (strict_strtoull(buf, 10, &recovery_start))
2609 if (rdev->mddev->pers &&
2610 rdev->raid_disk >= 0)
2613 rdev->recovery_offset = recovery_start;
2614 if (recovery_start == MaxSector)
2615 set_bit(In_sync, &rdev->flags);
2617 clear_bit(In_sync, &rdev->flags);
2621 static struct rdev_sysfs_entry rdev_recovery_start =
2622 __ATTR(recovery_start, S_IRUGO|S_IWUSR, recovery_start_show, recovery_start_store);
2624 static struct attribute *rdev_default_attrs[] = {
2630 &rdev_recovery_start.attr,
2634 rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
2636 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
2637 mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
2638 mddev_t *mddev = rdev->mddev;
2644 rv = mddev ? mddev_lock(mddev) : -EBUSY;
2646 if (rdev->mddev == NULL)
2649 rv = entry->show(rdev, page);
2650 mddev_unlock(mddev);
2656 rdev_attr_store(struct kobject *kobj, struct attribute *attr,
2657 const char *page, size_t length)
2659 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
2660 mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
2662 mddev_t *mddev = rdev->mddev;
2666 if (!capable(CAP_SYS_ADMIN))
2668 rv = mddev ? mddev_lock(mddev): -EBUSY;
2670 if (rdev->mddev == NULL)
2673 rv = entry->store(rdev, page, length);
2674 mddev_unlock(mddev);
2679 static void rdev_free(struct kobject *ko)
2681 mdk_rdev_t *rdev = container_of(ko, mdk_rdev_t, kobj);
2684 static const struct sysfs_ops rdev_sysfs_ops = {
2685 .show = rdev_attr_show,
2686 .store = rdev_attr_store,
2688 static struct kobj_type rdev_ktype = {
2689 .release = rdev_free,
2690 .sysfs_ops = &rdev_sysfs_ops,
2691 .default_attrs = rdev_default_attrs,
2695 * Import a device. If 'super_format' >= 0, then sanity check the superblock
2697 * mark the device faulty if:
2699 * - the device is nonexistent (zero size)
2700 * - the device has no valid superblock
2702 * a faulty rdev _never_ has rdev->sb set.
2704 static mdk_rdev_t *md_import_device(dev_t newdev, int super_format, int super_minor)
2706 char b[BDEVNAME_SIZE];
2711 rdev = kzalloc(sizeof(*rdev), GFP_KERNEL);
2713 printk(KERN_ERR "md: could not alloc mem for new device!\n");
2714 return ERR_PTR(-ENOMEM);
2717 if ((err = alloc_disk_sb(rdev)))
2720 err = lock_rdev(rdev, newdev, super_format == -2);
2724 kobject_init(&rdev->kobj, &rdev_ktype);
2727 rdev->saved_raid_disk = -1;
2728 rdev->raid_disk = -1;
2730 rdev->data_offset = 0;
2731 rdev->sb_events = 0;
2732 rdev->last_read_error.tv_sec = 0;
2733 rdev->last_read_error.tv_nsec = 0;
2734 atomic_set(&rdev->nr_pending, 0);
2735 atomic_set(&rdev->read_errors, 0);
2736 atomic_set(&rdev->corrected_errors, 0);
2738 size = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
2741 "md: %s has zero or unknown size, marking faulty!\n",
2742 bdevname(rdev->bdev,b));
2747 if (super_format >= 0) {
2748 err = super_types[super_format].
2749 load_super(rdev, NULL, super_minor);
2750 if (err == -EINVAL) {
2752 "md: %s does not have a valid v%d.%d "
2753 "superblock, not importing!\n",
2754 bdevname(rdev->bdev,b),
2755 super_format, super_minor);
2760 "md: could not read %s's sb, not importing!\n",
2761 bdevname(rdev->bdev,b));
2766 INIT_LIST_HEAD(&rdev->same_set);
2767 init_waitqueue_head(&rdev->blocked_wait);
2772 if (rdev->sb_page) {
2778 return ERR_PTR(err);
2782 * Check a full RAID array for plausibility
2786 static void analyze_sbs(mddev_t * mddev)
2789 mdk_rdev_t *rdev, *freshest, *tmp;
2790 char b[BDEVNAME_SIZE];
2793 rdev_for_each(rdev, tmp, mddev)
2794 switch (super_types[mddev->major_version].
2795 load_super(rdev, freshest, mddev->minor_version)) {
2803 "md: fatal superblock inconsistency in %s"
2804 " -- removing from array\n",
2805 bdevname(rdev->bdev,b));
2806 kick_rdev_from_array(rdev);
2810 super_types[mddev->major_version].
2811 validate_super(mddev, freshest);
2814 rdev_for_each(rdev, tmp, mddev) {
2815 if (mddev->max_disks &&
2816 (rdev->desc_nr >= mddev->max_disks ||
2817 i > mddev->max_disks)) {
2819 "md: %s: %s: only %d devices permitted\n",
2820 mdname(mddev), bdevname(rdev->bdev, b),
2822 kick_rdev_from_array(rdev);
2825 if (rdev != freshest)
2826 if (super_types[mddev->major_version].
2827 validate_super(mddev, rdev)) {
2828 printk(KERN_WARNING "md: kicking non-fresh %s"
2830 bdevname(rdev->bdev,b));
2831 kick_rdev_from_array(rdev);
2834 if (mddev->level == LEVEL_MULTIPATH) {
2835 rdev->desc_nr = i++;
2836 rdev->raid_disk = rdev->desc_nr;
2837 set_bit(In_sync, &rdev->flags);
2838 } else if (rdev->raid_disk >= (mddev->raid_disks - min(0, mddev->delta_disks))) {
2839 rdev->raid_disk = -1;
2840 clear_bit(In_sync, &rdev->flags);
2845 /* Read a fixed-point number.
2846 * Numbers in sysfs attributes should be in "standard" units where
2847 * possible, so time should be in seconds.
2848 * However we internally use a a much smaller unit such as
2849 * milliseconds or jiffies.
2850 * This function takes a decimal number with a possible fractional
2851 * component, and produces an integer which is the result of
2852 * multiplying that number by 10^'scale'.
2853 * all without any floating-point arithmetic.
2855 int strict_strtoul_scaled(const char *cp, unsigned long *res, int scale)
2857 unsigned long result = 0;
2859 while (isdigit(*cp) || (*cp == '.' && decimals < 0)) {
2862 else if (decimals < scale) {
2865 result = result * 10 + value;
2877 while (decimals < scale) {
2886 static void md_safemode_timeout(unsigned long data);
2889 safe_delay_show(mddev_t *mddev, char *page)
2891 int msec = (mddev->safemode_delay*1000)/HZ;
2892 return sprintf(page, "%d.%03d\n", msec/1000, msec%1000);
2895 safe_delay_store(mddev_t *mddev, const char *cbuf, size_t len)
2899 if (strict_strtoul_scaled(cbuf, &msec, 3) < 0)
2902 mddev->safemode_delay = 0;
2904 unsigned long old_delay = mddev->safemode_delay;
2905 mddev->safemode_delay = (msec*HZ)/1000;
2906 if (mddev->safemode_delay == 0)
2907 mddev->safemode_delay = 1;
2908 if (mddev->safemode_delay < old_delay)
2909 md_safemode_timeout((unsigned long)mddev);
2913 static struct md_sysfs_entry md_safe_delay =
2914 __ATTR(safe_mode_delay, S_IRUGO|S_IWUSR,safe_delay_show, safe_delay_store);
2917 level_show(mddev_t *mddev, char *page)
2919 struct mdk_personality *p = mddev->pers;
2921 return sprintf(page, "%s\n", p->name);
2922 else if (mddev->clevel[0])
2923 return sprintf(page, "%s\n", mddev->clevel);
2924 else if (mddev->level != LEVEL_NONE)
2925 return sprintf(page, "%d\n", mddev->level);
2931 level_store(mddev_t *mddev, const char *buf, size_t len)
2935 struct mdk_personality *pers;
2940 if (mddev->pers == NULL) {
2943 if (len >= sizeof(mddev->clevel))
2945 strncpy(mddev->clevel, buf, len);
2946 if (mddev->clevel[len-1] == '\n')
2948 mddev->clevel[len] = 0;
2949 mddev->level = LEVEL_NONE;
2953 /* request to change the personality. Need to ensure:
2954 * - array is not engaged in resync/recovery/reshape
2955 * - old personality can be suspended
2956 * - new personality will access other array.
2959 if (mddev->sync_thread || mddev->reshape_position != MaxSector)
2962 if (!mddev->pers->quiesce) {
2963 printk(KERN_WARNING "md: %s: %s does not support online personality change\n",
2964 mdname(mddev), mddev->pers->name);
2968 /* Now find the new personality */
2969 if (len == 0 || len >= sizeof(clevel))
2971 strncpy(clevel, buf, len);
2972 if (clevel[len-1] == '\n')
2975 if (strict_strtol(clevel, 10, &level))
2978 if (request_module("md-%s", clevel) != 0)
2979 request_module("md-level-%s", clevel);
2980 spin_lock(&pers_lock);
2981 pers = find_pers(level, clevel);
2982 if (!pers || !try_module_get(pers->owner)) {
2983 spin_unlock(&pers_lock);
2984 printk(KERN_WARNING "md: personality %s not loaded\n", clevel);
2987 spin_unlock(&pers_lock);
2989 if (pers == mddev->pers) {
2990 /* Nothing to do! */
2991 module_put(pers->owner);
2994 if (!pers->takeover) {
2995 module_put(pers->owner);
2996 printk(KERN_WARNING "md: %s: %s does not support personality takeover\n",
2997 mdname(mddev), clevel);
3001 list_for_each_entry(rdev, &mddev->disks, same_set)
3002 rdev->new_raid_disk = rdev->raid_disk;
3004 /* ->takeover must set new_* and/or delta_disks
3005 * if it succeeds, and may set them when it fails.
3007 priv = pers->takeover(mddev);
3009 mddev->new_level = mddev->level;
3010 mddev->new_layout = mddev->layout;
3011 mddev->new_chunk_sectors = mddev->chunk_sectors;
3012 mddev->raid_disks -= mddev->delta_disks;
3013 mddev->delta_disks = 0;
3014 module_put(pers->owner);
3015 printk(KERN_WARNING "md: %s: %s would not accept array\n",
3016 mdname(mddev), clevel);
3017 return PTR_ERR(priv);
3020 /* Looks like we have a winner */
3021 mddev_suspend(mddev);
3022 mddev->pers->stop(mddev);
3024 if (mddev->pers->sync_request == NULL &&
3025 pers->sync_request != NULL) {
3026 /* need to add the md_redundancy_group */
3027 if (sysfs_create_group(&mddev->kobj, &md_redundancy_group))
3029 "md: cannot register extra attributes for %s\n",
3031 mddev->sysfs_action = sysfs_get_dirent(mddev->kobj.sd, NULL, "sync_action");
3033 if (mddev->pers->sync_request != NULL &&
3034 pers->sync_request == NULL) {
3035 /* need to remove the md_redundancy_group */
3036 if (mddev->to_remove == NULL)
3037 mddev->to_remove = &md_redundancy_group;
3040 if (mddev->pers->sync_request == NULL &&
3042 /* We are converting from a no-redundancy array
3043 * to a redundancy array and metadata is managed
3044 * externally so we need to be sure that writes
3045 * won't block due to a need to transition
3047 * until external management is started.
3050 mddev->safemode_delay = 0;
3051 mddev->safemode = 0;
3054 list_for_each_entry(rdev, &mddev->disks, same_set) {
3056 if (rdev->raid_disk < 0)
3058 if (rdev->new_raid_disk > mddev->raid_disks)
3059 rdev->new_raid_disk = -1;
3060 if (rdev->new_raid_disk == rdev->raid_disk)
3062 sprintf(nm, "rd%d", rdev->raid_disk);
3063 sysfs_remove_link(&mddev->kobj, nm);
3065 list_for_each_entry(rdev, &mddev->disks, same_set) {
3066 if (rdev->raid_disk < 0)
3068 if (rdev->new_raid_disk == rdev->raid_disk)
3070 rdev->raid_disk = rdev->new_raid_disk;
3071 if (rdev->raid_disk < 0)
3072 clear_bit(In_sync, &rdev->flags);
3075 sprintf(nm, "rd%d", rdev->raid_disk);
3076 if(sysfs_create_link(&mddev->kobj, &rdev->kobj, nm))
3077 printk("md: cannot register %s for %s after level change\n",
3082 module_put(mddev->pers->owner);
3084 mddev->private = priv;
3085 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
3086 mddev->level = mddev->new_level;
3087 mddev->layout = mddev->new_layout;
3088 mddev->chunk_sectors = mddev->new_chunk_sectors;
3089 mddev->delta_disks = 0;
3090 if (mddev->pers->sync_request == NULL) {
3091 /* this is now an array without redundancy, so
3092 * it must always be in_sync
3095 del_timer_sync(&mddev->safemode_timer);
3098 mddev_resume(mddev);
3099 set_bit(MD_CHANGE_DEVS, &mddev->flags);
3100 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3101 md_wakeup_thread(mddev->thread);
3102 sysfs_notify(&mddev->kobj, NULL, "level");
3103 md_new_event(mddev);
3107 static struct md_sysfs_entry md_level =
3108 __ATTR(level, S_IRUGO|S_IWUSR, level_show, level_store);
3112 layout_show(mddev_t *mddev, char *page)
3114 /* just a number, not meaningful for all levels */
3115 if (mddev->reshape_position != MaxSector &&
3116 mddev->layout != mddev->new_layout)
3117 return sprintf(page, "%d (%d)\n",
3118 mddev->new_layout, mddev->layout);
3119 return sprintf(page, "%d\n", mddev->layout);
3123 layout_store(mddev_t *mddev, const char *buf, size_t len)
3126 unsigned long n = simple_strtoul(buf, &e, 10);
3128 if (!*buf || (*e && *e != '\n'))
3133 if (mddev->pers->check_reshape == NULL)
3135 mddev->new_layout = n;
3136 err = mddev->pers->check_reshape(mddev);
3138 mddev->new_layout = mddev->layout;
3142 mddev->new_layout = n;
3143 if (mddev->reshape_position == MaxSector)
3148 static struct md_sysfs_entry md_layout =
3149 __ATTR(layout, S_IRUGO|S_IWUSR, layout_show, layout_store);
3153 raid_disks_show(mddev_t *mddev, char *page)
3155 if (mddev->raid_disks == 0)
3157 if (mddev->reshape_position != MaxSector &&
3158 mddev->delta_disks != 0)
3159 return sprintf(page, "%d (%d)\n", mddev->raid_disks,
3160 mddev->raid_disks - mddev->delta_disks);
3161 return sprintf(page, "%d\n", mddev->raid_disks);
3164 static int update_raid_disks(mddev_t *mddev, int raid_disks);
3167 raid_disks_store(mddev_t *mddev, const char *buf, size_t len)
3171 unsigned long n = simple_strtoul(buf, &e, 10);
3173 if (!*buf || (*e && *e != '\n'))
3177 rv = update_raid_disks(mddev, n);
3178 else if (mddev->reshape_position != MaxSector) {
3179 int olddisks = mddev->raid_disks - mddev->delta_disks;
3180 mddev->delta_disks = n - olddisks;
3181 mddev->raid_disks = n;
3183 mddev->raid_disks = n;
3184 return rv ? rv : len;
3186 static struct md_sysfs_entry md_raid_disks =
3187 __ATTR(raid_disks, S_IRUGO|S_IWUSR, raid_disks_show, raid_disks_store);
3190 chunk_size_show(mddev_t *mddev, char *page)
3192 if (mddev->reshape_position != MaxSector &&
3193 mddev->chunk_sectors != mddev->new_chunk_sectors)
3194 return sprintf(page, "%d (%d)\n",
3195 mddev->new_chunk_sectors << 9,
3196 mddev->chunk_sectors << 9);
3197 return sprintf(page, "%d\n", mddev->chunk_sectors << 9);
3201 chunk_size_store(mddev_t *mddev, const char *buf, size_t len)
3204 unsigned long n = simple_strtoul(buf, &e, 10);
3206 if (!*buf || (*e && *e != '\n'))
3211 if (mddev->pers->check_reshape == NULL)
3213 mddev->new_chunk_sectors = n >> 9;
3214 err = mddev->pers->check_reshape(mddev);
3216 mddev->new_chunk_sectors = mddev->chunk_sectors;
3220 mddev->new_chunk_sectors = n >> 9;
3221 if (mddev->reshape_position == MaxSector)
3222 mddev->chunk_sectors = n >> 9;
3226 static struct md_sysfs_entry md_chunk_size =
3227 __ATTR(chunk_size, S_IRUGO|S_IWUSR, chunk_size_show, chunk_size_store);
3230 resync_start_show(mddev_t *mddev, char *page)
3232 if (mddev->recovery_cp == MaxSector)
3233 return sprintf(page, "none\n");
3234 return sprintf(page, "%llu\n", (unsigned long long)mddev->recovery_cp);
3238 resync_start_store(mddev_t *mddev, const char *buf, size_t len)
3241 unsigned long long n = simple_strtoull(buf, &e, 10);
3245 if (cmd_match(buf, "none"))
3247 else if (!*buf || (*e && *e != '\n'))
3250 mddev->recovery_cp = n;
3253 static struct md_sysfs_entry md_resync_start =
3254 __ATTR(resync_start, S_IRUGO|S_IWUSR, resync_start_show, resync_start_store);
3257 * The array state can be:
3260 * No devices, no size, no level
3261 * Equivalent to STOP_ARRAY ioctl
3263 * May have some settings, but array is not active
3264 * all IO results in error
3265 * When written, doesn't tear down array, but just stops it
3266 * suspended (not supported yet)
3267 * All IO requests will block. The array can be reconfigured.
3268 * Writing this, if accepted, will block until array is quiescent
3270 * no resync can happen. no superblocks get written.
3271 * write requests fail
3273 * like readonly, but behaves like 'clean' on a write request.
3275 * clean - no pending writes, but otherwise active.
3276 * When written to inactive array, starts without resync
3277 * If a write request arrives then
3278 * if metadata is known, mark 'dirty' and switch to 'active'.
3279 * if not known, block and switch to write-pending
3280 * If written to an active array that has pending writes, then fails.
3282 * fully active: IO and resync can be happening.
3283 * When written to inactive array, starts with resync
3286 * clean, but writes are blocked waiting for 'active' to be written.
3289 * like active, but no writes have been seen for a while (100msec).
3292 enum array_state { clear, inactive, suspended, readonly, read_auto, clean, active,
3293 write_pending, active_idle, bad_word};
3294 static char *array_states[] = {
3295 "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active",
3296 "write-pending", "active-idle", NULL };
3298 static int match_word(const char *word, char **list)
3301 for (n=0; list[n]; n++)
3302 if (cmd_match(word, list[n]))
3308 array_state_show(mddev_t *mddev, char *page)
3310 enum array_state st = inactive;
3323 else if (test_bit(MD_CHANGE_CLEAN, &mddev->flags))
3325 else if (mddev->safemode)
3331 if (list_empty(&mddev->disks) &&
3332 mddev->raid_disks == 0 &&
3333 mddev->dev_sectors == 0)
3338 return sprintf(page, "%s\n", array_states[st]);
3341 static int do_md_stop(mddev_t * mddev, int ro, int is_open);
3342 static int md_set_readonly(mddev_t * mddev, int is_open);
3343 static int do_md_run(mddev_t * mddev);
3344 static int restart_array(mddev_t *mddev);
3347 array_state_store(mddev_t *mddev, const char *buf, size_t len)
3350 enum array_state st = match_word(buf, array_states);
3355 /* stopping an active array */
3356 if (atomic_read(&mddev->openers) > 0)
3358 err = do_md_stop(mddev, 0, 0);
3361 /* stopping an active array */
3363 if (atomic_read(&mddev->openers) > 0)
3365 err = do_md_stop(mddev, 2, 0);
3367 err = 0; /* already inactive */
3370 break; /* not supported yet */
3373 err = md_set_readonly(mddev, 0);
3376 set_disk_ro(mddev->gendisk, 1);
3377 err = do_md_run(mddev);
3383 err = md_set_readonly(mddev, 0);
3384 else if (mddev->ro == 1)
3385 err = restart_array(mddev);
3388 set_disk_ro(mddev->gendisk, 0);
3392 err = do_md_run(mddev);
3397 restart_array(mddev);
3398 spin_lock_irq(&mddev->write_lock);
3399 if (atomic_read(&mddev->writes_pending) == 0) {
3400 if (mddev->in_sync == 0) {
3402 if (mddev->safemode == 1)
3403 mddev->safemode = 0;
3404 if (mddev->persistent)
3405 set_bit(MD_CHANGE_CLEAN,
3411 spin_unlock_irq(&mddev->write_lock);
3417 restart_array(mddev);
3418 if (mddev->external)
3419 clear_bit(MD_CHANGE_CLEAN, &mddev->flags);
3420 wake_up(&mddev->sb_wait);
3424 set_disk_ro(mddev->gendisk, 0);
3425 err = do_md_run(mddev);
3430 /* these cannot be set */
3436 sysfs_notify_dirent_safe(mddev->sysfs_state);
3440 static struct md_sysfs_entry md_array_state =
3441 __ATTR(array_state, S_IRUGO|S_IWUSR, array_state_show, array_state_store);
3444 max_corrected_read_errors_show(mddev_t *mddev, char *page) {
3445 return sprintf(page, "%d\n",
3446 atomic_read(&mddev->max_corr_read_errors));
3450 max_corrected_read_errors_store(mddev_t *mddev, const char *buf, size_t len)
3453 unsigned long n = simple_strtoul(buf, &e, 10);
3455 if (*buf && (*e == 0 || *e == '\n')) {
3456 atomic_set(&mddev->max_corr_read_errors, n);
3462 static struct md_sysfs_entry max_corr_read_errors =
3463 __ATTR(max_read_errors, S_IRUGO|S_IWUSR, max_corrected_read_errors_show,
3464 max_corrected_read_errors_store);
3467 null_show(mddev_t *mddev, char *page)
3473 new_dev_store(mddev_t *mddev, const char *buf, size_t len)
3475 /* buf must be %d:%d\n? giving major and minor numbers */
3476 /* The new device is added to the array.
3477 * If the array has a persistent superblock, we read the
3478 * superblock to initialise info and check validity.
3479 * Otherwise, only checking done is that in bind_rdev_to_array,
3480 * which mainly checks size.
3483 int major = simple_strtoul(buf, &e, 10);
3489 if (!*buf || *e != ':' || !e[1] || e[1] == '\n')
3491 minor = simple_strtoul(e+1, &e, 10);
3492 if (*e && *e != '\n')
3494 dev = MKDEV(major, minor);
3495 if (major != MAJOR(dev) ||
3496 minor != MINOR(dev))
3500 if (mddev->persistent) {
3501 rdev = md_import_device(dev, mddev->major_version,
3502 mddev->minor_version);
3503 if (!IS_ERR(rdev) && !list_empty(&mddev->disks)) {
3504 mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
3505 mdk_rdev_t, same_set);
3506 err = super_types[mddev->major_version]
3507 .load_super(rdev, rdev0, mddev->minor_version);
3511 } else if (mddev->external)
3512 rdev = md_import_device(dev, -2, -1);
3514 rdev = md_import_device(dev, -1, -1);
3517 return PTR_ERR(rdev);
3518 err = bind_rdev_to_array(rdev, mddev);
3522 return err ? err : len;
3525 static struct md_sysfs_entry md_new_device =
3526 __ATTR(new_dev, S_IWUSR, null_show, new_dev_store);
3529 bitmap_store(mddev_t *mddev, const char *buf, size_t len)
3532 unsigned long chunk, end_chunk;
3536 /* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */
3538 chunk = end_chunk = simple_strtoul(buf, &end, 0);
3539 if (buf == end) break;
3540 if (*end == '-') { /* range */
3542 end_chunk = simple_strtoul(buf, &end, 0);
3543 if (buf == end) break;
3545 if (*end && !isspace(*end)) break;
3546 bitmap_dirty_bits(mddev->bitmap, chunk, end_chunk);
3547 buf = skip_spaces(end);
3549 bitmap_unplug(mddev->bitmap); /* flush the bits to disk */
3554 static struct md_sysfs_entry md_bitmap =
3555 __ATTR(bitmap_set_bits, S_IWUSR, null_show, bitmap_store);
3558 size_show(mddev_t *mddev, char *page)
3560 return sprintf(page, "%llu\n",
3561 (unsigned long long)mddev->dev_sectors / 2);
3564 static int update_size(mddev_t *mddev, sector_t num_sectors);
3567 size_store(mddev_t *mddev, const char *buf, size_t len)
3569 /* If array is inactive, we can reduce the component size, but
3570 * not increase it (except from 0).
3571 * If array is active, we can try an on-line resize
3574 int err = strict_blocks_to_sectors(buf, §ors);
3579 err = update_size(mddev, sectors);
3580 md_update_sb(mddev, 1);
3582 if (mddev->dev_sectors == 0 ||
3583 mddev->dev_sectors > sectors)
3584 mddev->dev_sectors = sectors;
3588 return err ? err : len;
3591 static struct md_sysfs_entry md_size =
3592 __ATTR(component_size, S_IRUGO|S_IWUSR, size_show, size_store);
3597 * 'none' for arrays with no metadata (good luck...)
3598 * 'external' for arrays with externally managed metadata,
3599 * or N.M for internally known formats
3602 metadata_show(mddev_t *mddev, char *page)
3604 if (mddev->persistent)
3605 return sprintf(page, "%d.%d\n",
3606 mddev->major_version, mddev->minor_version);
3607 else if (mddev->external)
3608 return sprintf(page, "external:%s\n", mddev->metadata_type);
3610 return sprintf(page, "none\n");
3614 metadata_store(mddev_t *mddev, const char *buf, size_t len)
3618 /* Changing the details of 'external' metadata is
3619 * always permitted. Otherwise there must be
3620 * no devices attached to the array.
3622 if (mddev->external && strncmp(buf, "external:", 9) == 0)
3624 else if (!list_empty(&mddev->disks))
3627 if (cmd_match(buf, "none")) {
3628 mddev->persistent = 0;
3629 mddev->external = 0;
3630 mddev->major_version = 0;
3631 mddev->minor_version = 90;
3634 if (strncmp(buf, "external:", 9) == 0) {
3635 size_t namelen = len-9;
3636 if (namelen >= sizeof(mddev->metadata_type))
3637 namelen = sizeof(mddev->metadata_type)-1;
3638 strncpy(mddev->metadata_type, buf+9, namelen);
3639 mddev->metadata_type[namelen] = 0;
3640 if (namelen && mddev->metadata_type[namelen-1] == '\n')
3641 mddev->metadata_type[--namelen] = 0;
3642 mddev->persistent = 0;
3643 mddev->external = 1;
3644 mddev->major_version = 0;
3645 mddev->minor_version = 90;
3648 major = simple_strtoul(buf, &e, 10);
3649 if (e==buf || *e != '.')
3652 minor = simple_strtoul(buf, &e, 10);
3653 if (e==buf || (*e && *e != '\n') )
3655 if (major >= ARRAY_SIZE(super_types) || super_types[major].name == NULL)
3657 mddev->major_version = major;
3658 mddev->minor_version = minor;
3659 mddev->persistent = 1;
3660 mddev->external = 0;
3664 static struct md_sysfs_entry md_metadata =
3665 __ATTR(metadata_version, S_IRUGO|S_IWUSR, metadata_show, metadata_store);
3668 action_show(mddev_t *mddev, char *page)
3670 char *type = "idle";
3671 if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
3673 else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
3674 (!mddev->ro && test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))) {
3675 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
3677 else if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
3678 if (!test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
3680 else if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
3684 } else if (test_bit(MD_RECOVERY_RECOVER, &mddev->recovery))
3687 return sprintf(page, "%s\n", type);
3691 action_store(mddev_t *mddev, const char *page, size_t len)
3693 if (!mddev->pers || !mddev->pers->sync_request)
3696 if (cmd_match(page, "frozen"))
3697 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
3699 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
3701 if (cmd_match(page, "idle") || cmd_match(page, "frozen")) {
3702 if (mddev->sync_thread) {
3703 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
3704 md_unregister_thread(mddev->sync_thread);
3705 mddev->sync_thread = NULL;
3706 mddev->recovery = 0;
3708 } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
3709 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
3711 else if (cmd_match(page, "resync"))
3712 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3713 else if (cmd_match(page, "recover")) {
3714 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
3715 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3716 } else if (cmd_match(page, "reshape")) {
3718 if (mddev->pers->start_reshape == NULL)
3720 err = mddev->pers->start_reshape(mddev);
3723 sysfs_notify(&mddev->kobj, NULL, "degraded");
3725 if (cmd_match(page, "check"))
3726 set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
3727 else if (!cmd_match(page, "repair"))
3729 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
3730 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
3732 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3733 md_wakeup_thread(mddev->thread);
3734 sysfs_notify_dirent_safe(mddev->sysfs_action);
3739 mismatch_cnt_show(mddev_t *mddev, char *page)
3741 return sprintf(page, "%llu\n",
3742 (unsigned long long) mddev->resync_mismatches);
3745 static struct md_sysfs_entry md_scan_mode =
3746 __ATTR(sync_action, S_IRUGO|S_IWUSR, action_show, action_store);
3749 static struct md_sysfs_entry md_mismatches = __ATTR_RO(mismatch_cnt);
3752 sync_min_show(mddev_t *mddev, char *page)
3754 return sprintf(page, "%d (%s)\n", speed_min(mddev),
3755 mddev->sync_speed_min ? "local": "system");
3759 sync_min_store(mddev_t *mddev, const char *buf, size_t len)
3763 if (strncmp(buf, "system", 6)==0) {
3764 mddev->sync_speed_min = 0;
3767 min = simple_strtoul(buf, &e, 10);
3768 if (buf == e || (*e && *e != '\n') || min <= 0)
3770 mddev->sync_speed_min = min;
3774 static struct md_sysfs_entry md_sync_min =
3775 __ATTR(sync_speed_min, S_IRUGO|S_IWUSR, sync_min_show, sync_min_store);
3778 sync_max_show(mddev_t *mddev, char *page)
3780 return sprintf(page, "%d (%s)\n", speed_max(mddev),
3781 mddev->sync_speed_max ? "local": "system");
3785 sync_max_store(mddev_t *mddev, const char *buf, size_t len)
3789 if (strncmp(buf, "system", 6)==0) {
3790 mddev->sync_speed_max = 0;
3793 max = simple_strtoul(buf, &e, 10);
3794 if (buf == e || (*e && *e != '\n') || max <= 0)
3796 mddev->sync_speed_max = max;
3800 static struct md_sysfs_entry md_sync_max =
3801 __ATTR(sync_speed_max, S_IRUGO|S_IWUSR, sync_max_show, sync_max_store);
3804 degraded_show(mddev_t *mddev, char *page)
3806 return sprintf(page, "%d\n", mddev->degraded);
3808 static struct md_sysfs_entry md_degraded = __ATTR_RO(degraded);
3811 sync_force_parallel_show(mddev_t *mddev, char *page)
3813 return sprintf(page, "%d\n", mddev->parallel_resync);
3817 sync_force_parallel_store(mddev_t *mddev, const char *buf, size_t len)
3821 if (strict_strtol(buf, 10, &n))
3824 if (n != 0 && n != 1)
3827 mddev->parallel_resync = n;
3829 if (mddev->sync_thread)
3830 wake_up(&resync_wait);
3835 /* force parallel resync, even with shared block devices */
3836 static struct md_sysfs_entry md_sync_force_parallel =
3837 __ATTR(sync_force_parallel, S_IRUGO|S_IWUSR,
3838 sync_force_parallel_show, sync_force_parallel_store);
3841 sync_speed_show(mddev_t *mddev, char *page)
3843 unsigned long resync, dt, db;
3844 if (mddev->curr_resync == 0)
3845 return sprintf(page, "none\n");
3846 resync = mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active);
3847 dt = (jiffies - mddev->resync_mark) / HZ;
3849 db = resync - mddev->resync_mark_cnt;
3850 return sprintf(page, "%lu\n", db/dt/2); /* K/sec */
3853 static struct md_sysfs_entry md_sync_speed = __ATTR_RO(sync_speed);
3856 sync_completed_show(mddev_t *mddev, char *page)
3858 unsigned long max_sectors, resync;
3860 if (!test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
3861 return sprintf(page, "none\n");
3863 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
3864 max_sectors = mddev->resync_max_sectors;
3866 max_sectors = mddev->dev_sectors;
3868 resync = mddev->curr_resync_completed;
3869 return sprintf(page, "%lu / %lu\n", resync, max_sectors);
3872 static struct md_sysfs_entry md_sync_completed = __ATTR_RO(sync_completed);
3875 min_sync_show(mddev_t *mddev, char *page)
3877 return sprintf(page, "%llu\n",
3878 (unsigned long long)mddev->resync_min);
3881 min_sync_store(mddev_t *mddev, const char *buf, size_t len)
3883 unsigned long long min;
3884 if (strict_strtoull(buf, 10, &min))
3886 if (min > mddev->resync_max)
3888 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
3891 /* Must be a multiple of chunk_size */
3892 if (mddev->chunk_sectors) {
3893 sector_t temp = min;
3894 if (sector_div(temp, mddev->chunk_sectors))
3897 mddev->resync_min = min;
3902 static struct md_sysfs_entry md_min_sync =
3903 __ATTR(sync_min, S_IRUGO|S_IWUSR, min_sync_show, min_sync_store);
3906 max_sync_show(mddev_t *mddev, char *page)
3908 if (mddev->resync_max == MaxSector)
3909 return sprintf(page, "max\n");
3911 return sprintf(page, "%llu\n",
3912 (unsigned long long)mddev->resync_max);
3915 max_sync_store(mddev_t *mddev, const char *buf, size_t len)
3917 if (strncmp(buf, "max", 3) == 0)
3918 mddev->resync_max = MaxSector;
3920 unsigned long long max;
3921 if (strict_strtoull(buf, 10, &max))
3923 if (max < mddev->resync_min)
3925 if (max < mddev->resync_max &&
3927 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
3930 /* Must be a multiple of chunk_size */
3931 if (mddev->chunk_sectors) {
3932 sector_t temp = max;
3933 if (sector_div(temp, mddev->chunk_sectors))
3936 mddev->resync_max = max;
3938 wake_up(&mddev->recovery_wait);
3942 static struct md_sysfs_entry md_max_sync =
3943 __ATTR(sync_max, S_IRUGO|S_IWUSR, max_sync_show, max_sync_store);
3946 suspend_lo_show(mddev_t *mddev, char *page)
3948 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_lo);
3952 suspend_lo_store(mddev_t *mddev, const char *buf, size_t len)
3955 unsigned long long new = simple_strtoull(buf, &e, 10);
3957 if (mddev->pers == NULL ||
3958 mddev->pers->quiesce == NULL)
3960 if (buf == e || (*e && *e != '\n'))
3962 if (new >= mddev->suspend_hi ||
3963 (new > mddev->suspend_lo && new < mddev->suspend_hi)) {
3964 mddev->suspend_lo = new;
3965 mddev->pers->quiesce(mddev, 2);
3970 static struct md_sysfs_entry md_suspend_lo =
3971 __ATTR(suspend_lo, S_IRUGO|S_IWUSR, suspend_lo_show, suspend_lo_store);
3975 suspend_hi_show(mddev_t *mddev, char *page)
3977 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_hi);
3981 suspend_hi_store(mddev_t *mddev, const char *buf, size_t len)
3984 unsigned long long new = simple_strtoull(buf, &e, 10);
3986 if (mddev->pers == NULL ||
3987 mddev->pers->quiesce == NULL)
3989 if (buf == e || (*e && *e != '\n'))
3991 if ((new <= mddev->suspend_lo && mddev->suspend_lo >= mddev->suspend_hi) ||
3992 (new > mddev->suspend_lo && new > mddev->suspend_hi)) {
3993 mddev->suspend_hi = new;
3994 mddev->pers->quiesce(mddev, 1);
3995 mddev->pers->quiesce(mddev, 0);
4000 static struct md_sysfs_entry md_suspend_hi =
4001 __ATTR(suspend_hi, S_IRUGO|S_IWUSR, suspend_hi_show, suspend_hi_store);
4004 reshape_position_show(mddev_t *mddev, char *page)
4006 if (mddev->reshape_position != MaxSector)
4007 return sprintf(page, "%llu\n",
4008 (unsigned long long)mddev->reshape_position);
4009 strcpy(page, "none\n");
4014 reshape_position_store(mddev_t *mddev, const char *buf, size_t len)
4017 unsigned long long new = simple_strtoull(buf, &e, 10);
4020 if (buf == e || (*e && *e != '\n'))
4022 mddev->reshape_position = new;
4023 mddev->delta_disks = 0;
4024 mddev->new_level = mddev->level;
4025 mddev->new_layout = mddev->layout;
4026 mddev->new_chunk_sectors = mddev->chunk_sectors;
4030 static struct md_sysfs_entry md_reshape_position =
4031 __ATTR(reshape_position, S_IRUGO|S_IWUSR, reshape_position_show,
4032 reshape_position_store);
4035 array_size_show(mddev_t *mddev, char *page)
4037 if (mddev->external_size)
4038 return sprintf(page, "%llu\n",
4039 (unsigned long long)mddev->array_sectors/2);
4041 return sprintf(page, "default\n");
4045 array_size_store(mddev_t *mddev, const char *buf, size_t len)
4049 if (strncmp(buf, "default", 7) == 0) {
4051 sectors = mddev->pers->size(mddev, 0, 0);
4053 sectors = mddev->array_sectors;
4055 mddev->external_size = 0;
4057 if (strict_blocks_to_sectors(buf, §ors) < 0)
4059 if (mddev->pers && mddev->pers->size(mddev, 0, 0) < sectors)
4062 mddev->external_size = 1;
4065 mddev->array_sectors = sectors;
4066 set_capacity(mddev->gendisk, mddev->array_sectors);
4068 revalidate_disk(mddev->gendisk);
4073 static struct md_sysfs_entry md_array_size =
4074 __ATTR(array_size, S_IRUGO|S_IWUSR, array_size_show,
4077 static struct attribute *md_default_attrs[] = {
4080 &md_raid_disks.attr,
4081 &md_chunk_size.attr,
4083 &md_resync_start.attr,
4085 &md_new_device.attr,
4086 &md_safe_delay.attr,
4087 &md_array_state.attr,
4088 &md_reshape_position.attr,
4089 &md_array_size.attr,
4090 &max_corr_read_errors.attr,
4094 static struct attribute *md_redundancy_attrs[] = {
4096 &md_mismatches.attr,
4099 &md_sync_speed.attr,
4100 &md_sync_force_parallel.attr,
4101 &md_sync_completed.attr,
4104 &md_suspend_lo.attr,
4105 &md_suspend_hi.attr,
4110 static struct attribute_group md_redundancy_group = {
4112 .attrs = md_redundancy_attrs,
4117 md_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
4119 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
4120 mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
4125 rv = mddev_lock(mddev);
4127 rv = entry->show(mddev, page);
4128 mddev_unlock(mddev);
4134 md_attr_store(struct kobject *kobj, struct attribute *attr,
4135 const char *page, size_t length)
4137 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
4138 mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
4143 if (!capable(CAP_SYS_ADMIN))
4145 rv = mddev_lock(mddev);
4146 if (mddev->hold_active == UNTIL_IOCTL)
4147 mddev->hold_active = 0;
4149 rv = entry->store(mddev, page, length);
4150 mddev_unlock(mddev);
4155 static void md_free(struct kobject *ko)
4157 mddev_t *mddev = container_of(ko, mddev_t, kobj);
4159 if (mddev->sysfs_state)
4160 sysfs_put(mddev->sysfs_state);
4162 if (mddev->gendisk) {
4163 del_gendisk(mddev->gendisk);
4164 put_disk(mddev->gendisk);
4167 blk_cleanup_queue(mddev->queue);
4172 static const struct sysfs_ops md_sysfs_ops = {
4173 .show = md_attr_show,
4174 .store = md_attr_store,
4176 static struct kobj_type md_ktype = {
4178 .sysfs_ops = &md_sysfs_ops,
4179 .default_attrs = md_default_attrs,
4184 static void mddev_delayed_delete(struct work_struct *ws)
4186 mddev_t *mddev = container_of(ws, mddev_t, del_work);
4188 sysfs_remove_group(&mddev->kobj, &md_bitmap_group);
4189 kobject_del(&mddev->kobj);
4190 kobject_put(&mddev->kobj);
4193 static int md_alloc(dev_t dev, char *name)
4195 static DEFINE_MUTEX(disks_mutex);
4196 mddev_t *mddev = mddev_find(dev);
4197 struct gendisk *disk;
4206 partitioned = (MAJOR(mddev->unit) != MD_MAJOR);
4207 shift = partitioned ? MdpMinorShift : 0;
4208 unit = MINOR(mddev->unit) >> shift;
4210 /* wait for any previous instance if this device
4211 * to be completed removed (mddev_delayed_delete).
4213 flush_scheduled_work();
4215 mutex_lock(&disks_mutex);
4221 /* Need to ensure that 'name' is not a duplicate.
4224 spin_lock(&all_mddevs_lock);
4226 list_for_each_entry(mddev2, &all_mddevs, all_mddevs)
4227 if (mddev2->gendisk &&
4228 strcmp(mddev2->gendisk->disk_name, name) == 0) {
4229 spin_unlock(&all_mddevs_lock);
4232 spin_unlock(&all_mddevs_lock);
4236 mddev->queue = blk_alloc_queue(GFP_KERNEL);
4239 mddev->queue->queuedata = mddev;
4241 /* Can be unlocked because the queue is new: no concurrency */
4242 queue_flag_set_unlocked(QUEUE_FLAG_CLUSTER, mddev->queue);
4244 blk_queue_make_request(mddev->queue, md_make_request);
4246 disk = alloc_disk(1 << shift);
4248 blk_cleanup_queue(mddev->queue);
4249 mddev->queue = NULL;
4252 disk->major = MAJOR(mddev->unit);
4253 disk->first_minor = unit << shift;
4255 strcpy(disk->disk_name, name);
4256 else if (partitioned)
4257 sprintf(disk->disk_name, "md_d%d", unit);
4259 sprintf(disk->disk_name, "md%d", unit);
4260 disk->fops = &md_fops;
4261 disk->private_data = mddev;
4262 disk->queue = mddev->queue;
4263 /* Allow extended partitions. This makes the
4264 * 'mdp' device redundant, but we can't really
4267 disk->flags |= GENHD_FL_EXT_DEVT;
4269 mddev->gendisk = disk;
4270 error = kobject_init_and_add(&mddev->kobj, &md_ktype,
4271 &disk_to_dev(disk)->kobj, "%s", "md");
4273 /* This isn't possible, but as kobject_init_and_add is marked
4274 * __must_check, we must do something with the result
4276 printk(KERN_WARNING "md: cannot register %s/md - name in use\n",
4280 if (mddev->kobj.sd &&
4281 sysfs_create_group(&mddev->kobj, &md_bitmap_group))
4282 printk(KERN_DEBUG "pointless warning\n");
4284 mutex_unlock(&disks_mutex);
4285 if (!error && mddev->kobj.sd) {
4286 kobject_uevent(&mddev->kobj, KOBJ_ADD);
4287 mddev->sysfs_state = sysfs_get_dirent_safe(mddev->kobj.sd, "array_state");
4293 static struct kobject *md_probe(dev_t dev, int *part, void *data)
4295 md_alloc(dev, NULL);
4299 static int add_named_array(const char *val, struct kernel_param *kp)
4301 /* val must be "md_*" where * is not all digits.
4302 * We allocate an array with a large free minor number, and
4303 * set the name to val. val must not already be an active name.
4305 int len = strlen(val);
4306 char buf[DISK_NAME_LEN];
4308 while (len && val[len-1] == '\n')
4310 if (len >= DISK_NAME_LEN)
4312 strlcpy(buf, val, len+1);
4313 if (strncmp(buf, "md_", 3) != 0)
4315 return md_alloc(0, buf);
4318 static void md_safemode_timeout(unsigned long data)
4320 mddev_t *mddev = (mddev_t *) data;
4322 if (!atomic_read(&mddev->writes_pending)) {
4323 mddev->safemode = 1;
4324 if (mddev->external)
4325 sysfs_notify_dirent_safe(mddev->sysfs_state);
4327 md_wakeup_thread(mddev->thread);
4330 static int start_dirty_degraded;
4332 static int md_run(mddev_t *mddev)
4336 struct mdk_personality *pers;
4338 if (list_empty(&mddev->disks))
4339 /* cannot run an array with no devices.. */
4345 /* These two calls synchronise us with the
4346 * sysfs_remove_group calls in mddev_unlock,
4347 * so they must have completed.
4349 mutex_lock(&mddev->open_mutex);
4350 mutex_unlock(&mddev->open_mutex);
4353 * Analyze all RAID superblock(s)
4355 if (!mddev->raid_disks) {
4356 if (!mddev->persistent)
4361 if (mddev->level != LEVEL_NONE)
4362 request_module("md-level-%d", mddev->level);
4363 else if (mddev->clevel[0])
4364 request_module("md-%s", mddev->clevel);
4367 * Drop all container device buffers, from now on
4368 * the only valid external interface is through the md
4371 list_for_each_entry(rdev, &mddev->disks, same_set) {
4372 if (test_bit(Faulty, &rdev->flags))
4374 sync_blockdev(rdev->bdev);
4375 invalidate_bdev(rdev->bdev);
4377 /* perform some consistency tests on the device.
4378 * We don't want the data to overlap the metadata,
4379 * Internal Bitmap issues have been handled elsewhere.
4381 if (rdev->data_offset < rdev->sb_start) {
4382 if (mddev->dev_sectors &&
4383 rdev->data_offset + mddev->dev_sectors
4385 printk("md: %s: data overlaps metadata\n",
4390 if (rdev->sb_start + rdev->sb_size/512
4391 > rdev->data_offset) {
4392 printk("md: %s: metadata overlaps data\n",
4397 sysfs_notify_dirent_safe(rdev->sysfs_state);
4400 spin_lock(&pers_lock);
4401 pers = find_pers(mddev->level, mddev->clevel);
4402 if (!pers || !try_module_get(pers->owner)) {
4403 spin_unlock(&pers_lock);
4404 if (mddev->level != LEVEL_NONE)
4405 printk(KERN_WARNING "md: personality for level %d is not loaded!\n",
4408 printk(KERN_WARNING "md: personality for level %s is not loaded!\n",
4413 spin_unlock(&pers_lock);
4414 if (mddev->level != pers->level) {
4415 mddev->level = pers->level;
4416 mddev->new_level = pers->level;
4418 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
4420 if (mddev->reshape_position != MaxSector &&
4421 pers->start_reshape == NULL) {
4422 /* This personality cannot handle reshaping... */
4424 module_put(pers->owner);
4428 if (pers->sync_request) {
4429 /* Warn if this is a potentially silly
4432 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
4436 list_for_each_entry(rdev, &mddev->disks, same_set)
4437 list_for_each_entry(rdev2, &mddev->disks, same_set) {
4439 rdev->bdev->bd_contains ==
4440 rdev2->bdev->bd_contains) {
4442 "%s: WARNING: %s appears to be"
4443 " on the same physical disk as"
4446 bdevname(rdev->bdev,b),
4447 bdevname(rdev2->bdev,b2));
4454 "True protection against single-disk"
4455 " failure might be compromised.\n");
4458 mddev->recovery = 0;
4459 /* may be over-ridden by personality */
4460 mddev->resync_max_sectors = mddev->dev_sectors;
4462 mddev->barriers_work = 1;
4463 mddev->ok_start_degraded = start_dirty_degraded;
4465 if (start_readonly && mddev->ro == 0)
4466 mddev->ro = 2; /* read-only, but switch on first write */
4468 err = mddev->pers->run(mddev);
4470 printk(KERN_ERR "md: pers->run() failed ...\n");
4471 else if (mddev->pers->size(mddev, 0, 0) < mddev->array_sectors) {
4472 WARN_ONCE(!mddev->external_size, "%s: default size too small,"
4473 " but 'external_size' not in effect?\n", __func__);
4475 "md: invalid array_size %llu > default size %llu\n",
4476 (unsigned long long)mddev->array_sectors / 2,
4477 (unsigned long long)mddev->pers->size(mddev, 0, 0) / 2);
4479 mddev->pers->stop(mddev);
4481 if (err == 0 && mddev->pers->sync_request) {
4482 err = bitmap_create(mddev);
4484 printk(KERN_ERR "%s: failed to create bitmap (%d)\n",
4485 mdname(mddev), err);
4486 mddev->pers->stop(mddev);
4490 module_put(mddev->pers->owner);
4492 bitmap_destroy(mddev);
4495 if (mddev->pers->sync_request) {
4496 if (mddev->kobj.sd &&
4497 sysfs_create_group(&mddev->kobj, &md_redundancy_group))
4499 "md: cannot register extra attributes for %s\n",
4501 mddev->sysfs_action = sysfs_get_dirent_safe(mddev->kobj.sd, "sync_action");
4502 } else if (mddev->ro == 2) /* auto-readonly not meaningful */
4505 atomic_set(&mddev->writes_pending,0);
4506 atomic_set(&mddev->max_corr_read_errors,
4507 MD_DEFAULT_MAX_CORRECTED_READ_ERRORS);
4508 mddev->safemode = 0;
4509 mddev->safemode_timer.function = md_safemode_timeout;
4510 mddev->safemode_timer.data = (unsigned long) mddev;
4511 mddev->safemode_delay = (200 * HZ)/1000 +1; /* 200 msec delay */
4514 list_for_each_entry(rdev, &mddev->disks, same_set)
4515 if (rdev->raid_disk >= 0) {
4517 sprintf(nm, "rd%d", rdev->raid_disk);
4518 if (sysfs_create_link(&mddev->kobj, &rdev->kobj, nm))
4519 /* failure here is OK */;
4522 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4525 md_update_sb(mddev, 0);
4527 md_wakeup_thread(mddev->thread);
4528 md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
4530 md_new_event(mddev);
4531 sysfs_notify_dirent_safe(mddev->sysfs_state);
4532 sysfs_notify_dirent_safe(mddev->sysfs_action);
4533 sysfs_notify(&mddev->kobj, NULL, "degraded");
4537 static int do_md_run(mddev_t *mddev)
4541 err = md_run(mddev);
4545 set_capacity(mddev->gendisk, mddev->array_sectors);
4546 revalidate_disk(mddev->gendisk);
4547 kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
4552 static int restart_array(mddev_t *mddev)
4554 struct gendisk *disk = mddev->gendisk;
4556 /* Complain if it has no devices */
4557 if (list_empty(&mddev->disks))
4563 mddev->safemode = 0;
4565 set_disk_ro(disk, 0);
4566 printk(KERN_INFO "md: %s switched to read-write mode.\n",
4568 /* Kick recovery or resync if necessary */
4569 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4570 md_wakeup_thread(mddev->thread);
4571 md_wakeup_thread(mddev->sync_thread);
4572 sysfs_notify_dirent_safe(mddev->sysfs_state);
4576 /* similar to deny_write_access, but accounts for our holding a reference
4577 * to the file ourselves */
4578 static int deny_bitmap_write_access(struct file * file)
4580 struct inode *inode = file->f_mapping->host;
4582 spin_lock(&inode->i_lock);
4583 if (atomic_read(&inode->i_writecount) > 1) {
4584 spin_unlock(&inode->i_lock);
4587 atomic_set(&inode->i_writecount, -1);
4588 spin_unlock(&inode->i_lock);
4593 void restore_bitmap_write_access(struct file *file)
4595 struct inode *inode = file->f_mapping->host;
4597 spin_lock(&inode->i_lock);
4598 atomic_set(&inode->i_writecount, 1);
4599 spin_unlock(&inode->i_lock);
4602 static void md_clean(mddev_t *mddev)
4604 mddev->array_sectors = 0;
4605 mddev->external_size = 0;
4606 mddev->dev_sectors = 0;
4607 mddev->raid_disks = 0;
4608 mddev->recovery_cp = 0;
4609 mddev->resync_min = 0;
4610 mddev->resync_max = MaxSector;
4611 mddev->reshape_position = MaxSector;
4612 mddev->external = 0;
4613 mddev->persistent = 0;
4614 mddev->level = LEVEL_NONE;
4615 mddev->clevel[0] = 0;
4618 mddev->metadata_type[0] = 0;
4619 mddev->chunk_sectors = 0;
4620 mddev->ctime = mddev->utime = 0;
4622 mddev->max_disks = 0;
4624 mddev->can_decrease_events = 0;
4625 mddev->delta_disks = 0;
4626 mddev->new_level = LEVEL_NONE;
4627 mddev->new_layout = 0;
4628 mddev->new_chunk_sectors = 0;
4629 mddev->curr_resync = 0;
4630 mddev->resync_mismatches = 0;
4631 mddev->suspend_lo = mddev->suspend_hi = 0;
4632 mddev->sync_speed_min = mddev->sync_speed_max = 0;
4633 mddev->recovery = 0;
4635 mddev->degraded = 0;
4636 mddev->barriers_work = 0;
4637 mddev->safemode = 0;
4638 mddev->bitmap_info.offset = 0;
4639 mddev->bitmap_info.default_offset = 0;
4640 mddev->bitmap_info.chunksize = 0;
4641 mddev->bitmap_info.daemon_sleep = 0;
4642 mddev->bitmap_info.max_write_behind = 0;
4645 static void md_stop_writes(mddev_t *mddev)
4647 if (mddev->sync_thread) {
4648 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4649 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
4650 md_unregister_thread(mddev->sync_thread);
4651 mddev->sync_thread = NULL;
4654 del_timer_sync(&mddev->safemode_timer);
4656 bitmap_flush(mddev);
4657 md_super_wait(mddev);
4659 if (!mddev->in_sync || mddev->flags) {
4660 /* mark array as shutdown cleanly */
4662 md_update_sb(mddev, 1);
4666 static void md_stop(mddev_t *mddev)
4668 md_stop_writes(mddev);
4670 mddev->pers->stop(mddev);
4671 if (mddev->pers->sync_request && mddev->to_remove == NULL)
4672 mddev->to_remove = &md_redundancy_group;
4673 module_put(mddev->pers->owner);
4675 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4678 static int md_set_readonly(mddev_t *mddev, int is_open)
4681 mutex_lock(&mddev->open_mutex);
4682 if (atomic_read(&mddev->openers) > is_open) {
4683 printk("md: %s still in use.\n",mdname(mddev));
4688 md_stop_writes(mddev);
4694 set_disk_ro(mddev->gendisk, 1);
4695 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4696 sysfs_notify_dirent_safe(mddev->sysfs_state);
4700 mutex_unlock(&mddev->open_mutex);
4705 * 0 - completely stop and dis-assemble array
4706 * 2 - stop but do not disassemble array
4708 static int do_md_stop(mddev_t * mddev, int mode, int is_open)
4711 struct gendisk *disk = mddev->gendisk;
4714 mutex_lock(&mddev->open_mutex);
4715 if (atomic_read(&mddev->openers) > is_open) {
4716 printk("md: %s still in use.\n",mdname(mddev));
4718 } else if (mddev->pers) {
4721 set_disk_ro(disk, 0);
4724 mddev->queue->merge_bvec_fn = NULL;
4725 mddev->queue->unplug_fn = NULL;
4726 mddev->queue->backing_dev_info.congested_fn = NULL;
4728 /* tell userspace to handle 'inactive' */
4729 sysfs_notify_dirent_safe(mddev->sysfs_state);
4731 list_for_each_entry(rdev, &mddev->disks, same_set)
4732 if (rdev->raid_disk >= 0) {
4734 sprintf(nm, "rd%d", rdev->raid_disk);
4735 sysfs_remove_link(&mddev->kobj, nm);
4738 set_capacity(disk, 0);
4739 revalidate_disk(disk);
4746 mutex_unlock(&mddev->open_mutex);
4750 * Free resources if final stop
4754 printk(KERN_INFO "md: %s stopped.\n", mdname(mddev));
4756 bitmap_destroy(mddev);
4757 if (mddev->bitmap_info.file) {
4758 restore_bitmap_write_access(mddev->bitmap_info.file);
4759 fput(mddev->bitmap_info.file);
4760 mddev->bitmap_info.file = NULL;
4762 mddev->bitmap_info.offset = 0;
4764 export_array(mddev);
4767 kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
4768 if (mddev->hold_active == UNTIL_STOP)
4769 mddev->hold_active = 0;
4773 blk_integrity_unregister(disk);
4774 md_new_event(mddev);
4775 sysfs_notify_dirent_safe(mddev->sysfs_state);
4780 static void autorun_array(mddev_t *mddev)
4785 if (list_empty(&mddev->disks))
4788 printk(KERN_INFO "md: running: ");
4790 list_for_each_entry(rdev, &mddev->disks, same_set) {
4791 char b[BDEVNAME_SIZE];
4792 printk("<%s>", bdevname(rdev->bdev,b));
4796 err = do_md_run(mddev);
4798 printk(KERN_WARNING "md: do_md_run() returned %d\n", err);
4799 do_md_stop(mddev, 0, 0);
4804 * lets try to run arrays based on all disks that have arrived
4805 * until now. (those are in pending_raid_disks)
4807 * the method: pick the first pending disk, collect all disks with
4808 * the same UUID, remove all from the pending list and put them into
4809 * the 'same_array' list. Then order this list based on superblock
4810 * update time (freshest comes first), kick out 'old' disks and
4811 * compare superblocks. If everything's fine then run it.
4813 * If "unit" is allocated, then bump its reference count
4815 static void autorun_devices(int part)
4817 mdk_rdev_t *rdev0, *rdev, *tmp;
4819 char b[BDEVNAME_SIZE];
4821 printk(KERN_INFO "md: autorun ...\n");
4822 while (!list_empty(&pending_raid_disks)) {
4825 LIST_HEAD(candidates);
4826 rdev0 = list_entry(pending_raid_disks.next,
4827 mdk_rdev_t, same_set);
4829 printk(KERN_INFO "md: considering %s ...\n",
4830 bdevname(rdev0->bdev,b));
4831 INIT_LIST_HEAD(&candidates);
4832 rdev_for_each_list(rdev, tmp, &pending_raid_disks)
4833 if (super_90_load(rdev, rdev0, 0) >= 0) {
4834 printk(KERN_INFO "md: adding %s ...\n",
4835 bdevname(rdev->bdev,b));
4836 list_move(&rdev->same_set, &candidates);
4839 * now we have a set of devices, with all of them having
4840 * mostly sane superblocks. It's time to allocate the
4844 dev = MKDEV(mdp_major,
4845 rdev0->preferred_minor << MdpMinorShift);
4846 unit = MINOR(dev) >> MdpMinorShift;
4848 dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
4851 if (rdev0->preferred_minor != unit) {
4852 printk(KERN_INFO "md: unit number in %s is bad: %d\n",
4853 bdevname(rdev0->bdev, b), rdev0->preferred_minor);
4857 md_probe(dev, NULL, NULL);
4858 mddev = mddev_find(dev);
4859 if (!mddev || !mddev->gendisk) {
4863 "md: cannot allocate memory for md drive.\n");
4866 if (mddev_lock(mddev))
4867 printk(KERN_WARNING "md: %s locked, cannot run\n",
4869 else if (mddev->raid_disks || mddev->major_version
4870 || !list_empty(&mddev->disks)) {
4872 "md: %s already running, cannot run %s\n",
4873 mdname(mddev), bdevname(rdev0->bdev,b));
4874 mddev_unlock(mddev);
4876 printk(KERN_INFO "md: created %s\n", mdname(mddev));
4877 mddev->persistent = 1;
4878 rdev_for_each_list(rdev, tmp, &candidates) {
4879 list_del_init(&rdev->same_set);
4880 if (bind_rdev_to_array(rdev, mddev))
4883 autorun_array(mddev);
4884 mddev_unlock(mddev);
4886 /* on success, candidates will be empty, on error
4889 rdev_for_each_list(rdev, tmp, &candidates) {
4890 list_del_init(&rdev->same_set);
4895 printk(KERN_INFO "md: ... autorun DONE.\n");
4897 #endif /* !MODULE */
4899 static int get_version(void __user * arg)
4903 ver.major = MD_MAJOR_VERSION;
4904 ver.minor = MD_MINOR_VERSION;
4905 ver.patchlevel = MD_PATCHLEVEL_VERSION;
4907 if (copy_to_user(arg, &ver, sizeof(ver)))
4913 static int get_array_info(mddev_t * mddev, void __user * arg)
4915 mdu_array_info_t info;
4916 int nr,working,insync,failed,spare;
4919 nr=working=insync=failed=spare=0;
4920 list_for_each_entry(rdev, &mddev->disks, same_set) {
4922 if (test_bit(Faulty, &rdev->flags))
4926 if (test_bit(In_sync, &rdev->flags))
4933 info.major_version = mddev->major_version;
4934 info.minor_version = mddev->minor_version;
4935 info.patch_version = MD_PATCHLEVEL_VERSION;
4936 info.ctime = mddev->ctime;
4937 info.level = mddev->level;
4938 info.size = mddev->dev_sectors / 2;
4939 if (info.size != mddev->dev_sectors / 2) /* overflow */
4942 info.raid_disks = mddev->raid_disks;
4943 info.md_minor = mddev->md_minor;
4944 info.not_persistent= !mddev->persistent;
4946 info.utime = mddev->utime;
4949 info.state = (1<<MD_SB_CLEAN);
4950 if (mddev->bitmap && mddev->bitmap_info.offset)
4951 info.state = (1<<MD_SB_BITMAP_PRESENT);
4952 info.active_disks = insync;
4953 info.working_disks = working;
4954 info.failed_disks = failed;
4955 info.spare_disks = spare;
4957 info.layout = mddev->layout;
4958 info.chunk_size = mddev->chunk_sectors << 9;
4960 if (copy_to_user(arg, &info, sizeof(info)))
4966 static int get_bitmap_file(mddev_t * mddev, void __user * arg)
4968 mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */
4969 char *ptr, *buf = NULL;
4972 if (md_allow_write(mddev))
4973 file = kmalloc(sizeof(*file), GFP_NOIO);
4975 file = kmalloc(sizeof(*file), GFP_KERNEL);
4980 /* bitmap disabled, zero the first byte and copy out */
4981 if (!mddev->bitmap || !mddev->bitmap->file) {
4982 file->pathname[0] = '\0';
4986 buf = kmalloc(sizeof(file->pathname), GFP_KERNEL);
4990 ptr = d_path(&mddev->bitmap->file->f_path, buf, sizeof(file->pathname));
4994 strcpy(file->pathname, ptr);
4998 if (copy_to_user(arg, file, sizeof(*file)))
5006 static int get_disk_info(mddev_t * mddev, void __user * arg)
5008 mdu_disk_info_t info;
5011 if (copy_from_user(&info, arg, sizeof(info)))
5014 rdev = find_rdev_nr(mddev, info.number);
5016 info.major = MAJOR(rdev->bdev->bd_dev);
5017 info.minor = MINOR(rdev->bdev->bd_dev);
5018 info.raid_disk = rdev->raid_disk;
5020 if (test_bit(Faulty, &rdev->flags))
5021 info.state |= (1<<MD_DISK_FAULTY);
5022 else if (test_bit(In_sync, &rdev->flags)) {
5023 info.state |= (1<<MD_DISK_ACTIVE);
5024 info.state |= (1<<MD_DISK_SYNC);
5026 if (test_bit(WriteMostly, &rdev->flags))
5027 info.state |= (1<<MD_DISK_WRITEMOSTLY);
5029 info.major = info.minor = 0;
5030 info.raid_disk = -1;
5031 info.state = (1<<MD_DISK_REMOVED);
5034 if (copy_to_user(arg, &info, sizeof(info)))
5040 static int add_new_disk(mddev_t * mddev, mdu_disk_info_t *info)
5042 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
5044 dev_t dev = MKDEV(info->major,info->minor);
5046 if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
5049 if (!mddev->raid_disks) {
5051 /* expecting a device which has a superblock */
5052 rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
5055 "md: md_import_device returned %ld\n",
5057 return PTR_ERR(rdev);
5059 if (!list_empty(&mddev->disks)) {
5060 mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
5061 mdk_rdev_t, same_set);
5062 err = super_types[mddev->major_version]
5063 .load_super(rdev, rdev0, mddev->minor_version);
5066 "md: %s has different UUID to %s\n",
5067 bdevname(rdev->bdev,b),
5068 bdevname(rdev0->bdev,b2));
5073 err = bind_rdev_to_array(rdev, mddev);
5080 * add_new_disk can be used once the array is assembled
5081 * to add "hot spares". They must already have a superblock
5086 if (!mddev->pers->hot_add_disk) {
5088 "%s: personality does not support diskops!\n",
5092 if (mddev->persistent)
5093 rdev = md_import_device(dev, mddev->major_version,
5094 mddev->minor_version);
5096 rdev = md_import_device(dev, -1, -1);
5099 "md: md_import_device returned %ld\n",
5101 return PTR_ERR(rdev);
5103 /* set save_raid_disk if appropriate */
5104 if (!mddev->persistent) {
5105 if (info->state & (1<<MD_DISK_SYNC) &&
5106 info->raid_disk < mddev->raid_disks)
5107 rdev->raid_disk = info->raid_disk;
5109 rdev->raid_disk = -1;
5111 super_types[mddev->major_version].
5112 validate_super(mddev, rdev);
5113 rdev->saved_raid_disk = rdev->raid_disk;
5115 clear_bit(In_sync, &rdev->flags); /* just to be sure */
5116 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
5117 set_bit(WriteMostly, &rdev->flags);
5119 clear_bit(WriteMostly, &rdev->flags);
5121 rdev->raid_disk = -1;
5122 err = bind_rdev_to_array(rdev, mddev);
5123 if (!err && !mddev->pers->hot_remove_disk) {
5124 /* If there is hot_add_disk but no hot_remove_disk
5125 * then added disks for geometry changes,
5126 * and should be added immediately.
5128 super_types[mddev->major_version].
5129 validate_super(mddev, rdev);
5130 err = mddev->pers->hot_add_disk(mddev, rdev);
5132 unbind_rdev_from_array(rdev);
5137 sysfs_notify_dirent_safe(rdev->sysfs_state);
5139 md_update_sb(mddev, 1);
5140 if (mddev->degraded)
5141 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
5142 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5143 md_wakeup_thread(mddev->thread);
5147 /* otherwise, add_new_disk is only allowed
5148 * for major_version==0 superblocks
5150 if (mddev->major_version != 0) {
5151 printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n",
5156 if (!(info->state & (1<<MD_DISK_FAULTY))) {
5158 rdev = md_import_device(dev, -1, 0);
5161 "md: error, md_import_device() returned %ld\n",
5163 return PTR_ERR(rdev);
5165 rdev->desc_nr = info->number;
5166 if (info->raid_disk < mddev->raid_disks)
5167 rdev->raid_disk = info->raid_disk;
5169 rdev->raid_disk = -1;
5171 if (rdev->raid_disk < mddev->raid_disks)
5172 if (info->state & (1<<MD_DISK_SYNC))
5173 set_bit(In_sync, &rdev->flags);
5175 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
5176 set_bit(WriteMostly, &rdev->flags);
5178 if (!mddev->persistent) {
5179 printk(KERN_INFO "md: nonpersistent superblock ...\n");
5180 rdev->sb_start = rdev->bdev->bd_inode->i_size / 512;
5182 rdev->sb_start = calc_dev_sboffset(rdev->bdev);
5183 rdev->sectors = rdev->sb_start;
5185 err = bind_rdev_to_array(rdev, mddev);
5195 static int hot_remove_disk(mddev_t * mddev, dev_t dev)
5197 char b[BDEVNAME_SIZE];
5200 rdev = find_rdev(mddev, dev);
5204 if (rdev->raid_disk >= 0)
5207 kick_rdev_from_array(rdev);
5208 md_update_sb(mddev, 1);
5209 md_new_event(mddev);
5213 printk(KERN_WARNING "md: cannot remove active disk %s from %s ...\n",
5214 bdevname(rdev->bdev,b), mdname(mddev));
5218 static int hot_add_disk(mddev_t * mddev, dev_t dev)
5220 char b[BDEVNAME_SIZE];
5227 if (mddev->major_version != 0) {
5228 printk(KERN_WARNING "%s: HOT_ADD may only be used with"
5229 " version-0 superblocks.\n",
5233 if (!mddev->pers->hot_add_disk) {
5235 "%s: personality does not support diskops!\n",
5240 rdev = md_import_device(dev, -1, 0);
5243 "md: error, md_import_device() returned %ld\n",
5248 if (mddev->persistent)
5249 rdev->sb_start = calc_dev_sboffset(rdev->bdev);
5251 rdev->sb_start = rdev->bdev->bd_inode->i_size / 512;
5253 rdev->sectors = rdev->sb_start;
5255 if (test_bit(Faulty, &rdev->flags)) {
5257 "md: can not hot-add faulty %s disk to %s!\n",
5258 bdevname(rdev->bdev,b), mdname(mddev));
5262 clear_bit(In_sync, &rdev->flags);
5264 rdev->saved_raid_disk = -1;
5265 err = bind_rdev_to_array(rdev, mddev);
5270 * The rest should better be atomic, we can have disk failures
5271 * noticed in interrupt contexts ...
5274 rdev->raid_disk = -1;
5276 md_update_sb(mddev, 1);
5279 * Kick recovery, maybe this spare has to be added to the
5280 * array immediately.
5282 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5283 md_wakeup_thread(mddev->thread);
5284 md_new_event(mddev);
5292 static int set_bitmap_file(mddev_t *mddev, int fd)
5297 if (!mddev->pers->quiesce)
5299 if (mddev->recovery || mddev->sync_thread)
5301 /* we should be able to change the bitmap.. */
5307 return -EEXIST; /* cannot add when bitmap is present */
5308 mddev->bitmap_info.file = fget(fd);
5310 if (mddev->bitmap_info.file == NULL) {
5311 printk(KERN_ERR "%s: error: failed to get bitmap file\n",
5316 err = deny_bitmap_write_access(mddev->bitmap_info.file);
5318 printk(KERN_ERR "%s: error: bitmap file is already in use\n",
5320 fput(mddev->bitmap_info.file);
5321 mddev->bitmap_info.file = NULL;
5324 mddev->bitmap_info.offset = 0; /* file overrides offset */
5325 } else if (mddev->bitmap == NULL)
5326 return -ENOENT; /* cannot remove what isn't there */
5329 mddev->pers->quiesce(mddev, 1);
5331 err = bitmap_create(mddev);
5332 if (fd < 0 || err) {
5333 bitmap_destroy(mddev);
5334 fd = -1; /* make sure to put the file */
5336 mddev->pers->quiesce(mddev, 0);
5339 if (mddev->bitmap_info.file) {
5340 restore_bitmap_write_access(mddev->bitmap_info.file);
5341 fput(mddev->bitmap_info.file);
5343 mddev->bitmap_info.file = NULL;
5350 * set_array_info is used two different ways
5351 * The original usage is when creating a new array.
5352 * In this usage, raid_disks is > 0 and it together with
5353 * level, size, not_persistent,layout,chunksize determine the
5354 * shape of the array.
5355 * This will always create an array with a type-0.90.0 superblock.
5356 * The newer usage is when assembling an array.
5357 * In this case raid_disks will be 0, and the major_version field is
5358 * use to determine which style super-blocks are to be found on the devices.
5359 * The minor and patch _version numbers are also kept incase the
5360 * super_block handler wishes to interpret them.
5362 static int set_array_info(mddev_t * mddev, mdu_array_info_t *info)
5365 if (info->raid_disks == 0) {
5366 /* just setting version number for superblock loading */
5367 if (info->major_version < 0 ||
5368 info->major_version >= ARRAY_SIZE(super_types) ||
5369 super_types[info->major_version].name == NULL) {
5370 /* maybe try to auto-load a module? */
5372 "md: superblock version %d not known\n",
5373 info->major_version);
5376 mddev->major_version = info->major_version;
5377 mddev->minor_version = info->minor_version;
5378 mddev->patch_version = info->patch_version;
5379 mddev->persistent = !info->not_persistent;
5380 /* ensure mddev_put doesn't delete this now that there
5381 * is some minimal configuration.
5383 mddev->ctime = get_seconds();
5386 mddev->major_version = MD_MAJOR_VERSION;
5387 mddev->minor_version = MD_MINOR_VERSION;
5388 mddev->patch_version = MD_PATCHLEVEL_VERSION;
5389 mddev->ctime = get_seconds();
5391 mddev->level = info->level;
5392 mddev->clevel[0] = 0;
5393 mddev->dev_sectors = 2 * (sector_t)info->size;
5394 mddev->raid_disks = info->raid_disks;
5395 /* don't set md_minor, it is determined by which /dev/md* was
5398 if (info->state & (1<<MD_SB_CLEAN))
5399 mddev->recovery_cp = MaxSector;
5401 mddev->recovery_cp = 0;
5402 mddev->persistent = ! info->not_persistent;
5403 mddev->external = 0;
5405 mddev->layout = info->layout;
5406 mddev->chunk_sectors = info->chunk_size >> 9;
5408 mddev->max_disks = MD_SB_DISKS;
5410 if (mddev->persistent)
5412 set_bit(MD_CHANGE_DEVS, &mddev->flags);
5414 mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9;
5415 mddev->bitmap_info.offset = 0;
5417 mddev->reshape_position = MaxSector;
5420 * Generate a 128 bit UUID
5422 get_random_bytes(mddev->uuid, 16);
5424 mddev->new_level = mddev->level;
5425 mddev->new_chunk_sectors = mddev->chunk_sectors;
5426 mddev->new_layout = mddev->layout;
5427 mddev->delta_disks = 0;
5432 void md_set_array_sectors(mddev_t *mddev, sector_t array_sectors)
5434 WARN(!mddev_is_locked(mddev), "%s: unlocked mddev!\n", __func__);
5436 if (mddev->external_size)
5439 mddev->array_sectors = array_sectors;
5441 EXPORT_SYMBOL(md_set_array_sectors);
5443 static int update_size(mddev_t *mddev, sector_t num_sectors)
5447 int fit = (num_sectors == 0);
5449 if (mddev->pers->resize == NULL)
5451 /* The "num_sectors" is the number of sectors of each device that
5452 * is used. This can only make sense for arrays with redundancy.
5453 * linear and raid0 always use whatever space is available. We can only
5454 * consider changing this number if no resync or reconstruction is
5455 * happening, and if the new size is acceptable. It must fit before the
5456 * sb_start or, if that is <data_offset, it must fit before the size
5457 * of each device. If num_sectors is zero, we find the largest size
5461 if (mddev->sync_thread)
5464 /* Sorry, cannot grow a bitmap yet, just remove it,
5468 list_for_each_entry(rdev, &mddev->disks, same_set) {
5469 sector_t avail = rdev->sectors;
5471 if (fit && (num_sectors == 0 || num_sectors > avail))
5472 num_sectors = avail;
5473 if (avail < num_sectors)
5476 rv = mddev->pers->resize(mddev, num_sectors);
5478 revalidate_disk(mddev->gendisk);
5482 static int update_raid_disks(mddev_t *mddev, int raid_disks)
5485 /* change the number of raid disks */
5486 if (mddev->pers->check_reshape == NULL)
5488 if (raid_disks <= 0 ||
5489 (mddev->max_disks && raid_disks >= mddev->max_disks))
5491 if (mddev->sync_thread || mddev->reshape_position != MaxSector)
5493 mddev->delta_disks = raid_disks - mddev->raid_disks;
5495 rv = mddev->pers->check_reshape(mddev);
5501 * update_array_info is used to change the configuration of an
5503 * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
5504 * fields in the info are checked against the array.
5505 * Any differences that cannot be handled will cause an error.
5506 * Normally, only one change can be managed at a time.
5508 static int update_array_info(mddev_t *mddev, mdu_array_info_t *info)
5514 /* calculate expected state,ignoring low bits */
5515 if (mddev->bitmap && mddev->bitmap_info.offset)
5516 state |= (1 << MD_SB_BITMAP_PRESENT);
5518 if (mddev->major_version != info->major_version ||
5519 mddev->minor_version != info->minor_version ||
5520 /* mddev->patch_version != info->patch_version || */
5521 mddev->ctime != info->ctime ||
5522 mddev->level != info->level ||
5523 /* mddev->layout != info->layout || */
5524 !mddev->persistent != info->not_persistent||
5525 mddev->chunk_sectors != info->chunk_size >> 9 ||
5526 /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
5527 ((state^info->state) & 0xfffffe00)
5530 /* Check there is only one change */
5531 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
5533 if (mddev->raid_disks != info->raid_disks)
5535 if (mddev->layout != info->layout)
5537 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT))
5544 if (mddev->layout != info->layout) {
5546 * we don't need to do anything at the md level, the
5547 * personality will take care of it all.
5549 if (mddev->pers->check_reshape == NULL)
5552 mddev->new_layout = info->layout;
5553 rv = mddev->pers->check_reshape(mddev);
5555 mddev->new_layout = mddev->layout;
5559 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
5560 rv = update_size(mddev, (sector_t)info->size * 2);
5562 if (mddev->raid_disks != info->raid_disks)
5563 rv = update_raid_disks(mddev, info->raid_disks);
5565 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) {
5566 if (mddev->pers->quiesce == NULL)
5568 if (mddev->recovery || mddev->sync_thread)
5570 if (info->state & (1<<MD_SB_BITMAP_PRESENT)) {
5571 /* add the bitmap */
5574 if (mddev->bitmap_info.default_offset == 0)
5576 mddev->bitmap_info.offset =
5577 mddev->bitmap_info.default_offset;
5578 mddev->pers->quiesce(mddev, 1);
5579 rv = bitmap_create(mddev);
5581 bitmap_destroy(mddev);
5582 mddev->pers->quiesce(mddev, 0);
5584 /* remove the bitmap */
5587 if (mddev->bitmap->file)
5589 mddev->pers->quiesce(mddev, 1);
5590 bitmap_destroy(mddev);
5591 mddev->pers->quiesce(mddev, 0);
5592 mddev->bitmap_info.offset = 0;
5595 md_update_sb(mddev, 1);
5599 static int set_disk_faulty(mddev_t *mddev, dev_t dev)
5603 if (mddev->pers == NULL)
5606 rdev = find_rdev(mddev, dev);
5610 md_error(mddev, rdev);
5615 * We have a problem here : there is no easy way to give a CHS
5616 * virtual geometry. We currently pretend that we have a 2 heads
5617 * 4 sectors (with a BIG number of cylinders...). This drives
5618 * dosfs just mad... ;-)
5620 static int md_getgeo(struct block_device *bdev, struct hd_geometry *geo)
5622 mddev_t *mddev = bdev->bd_disk->private_data;
5626 geo->cylinders = mddev->array_sectors / 8;
5630 static int md_ioctl(struct block_device *bdev, fmode_t mode,
5631 unsigned int cmd, unsigned long arg)
5634 void __user *argp = (void __user *)arg;
5635 mddev_t *mddev = NULL;
5638 if (!capable(CAP_SYS_ADMIN))
5642 * Commands dealing with the RAID driver but not any
5648 err = get_version(argp);
5651 case PRINT_RAID_DEBUG:
5659 autostart_arrays(arg);
5666 * Commands creating/starting a new array:
5669 mddev = bdev->bd_disk->private_data;
5676 err = mddev_lock(mddev);
5679 "md: ioctl lock interrupted, reason %d, cmd %d\n",
5686 case SET_ARRAY_INFO:
5688 mdu_array_info_t info;
5690 memset(&info, 0, sizeof(info));
5691 else if (copy_from_user(&info, argp, sizeof(info))) {
5696 err = update_array_info(mddev, &info);
5698 printk(KERN_WARNING "md: couldn't update"
5699 " array info. %d\n", err);
5704 if (!list_empty(&mddev->disks)) {
5706 "md: array %s already has disks!\n",
5711 if (mddev->raid_disks) {
5713 "md: array %s already initialised!\n",
5718 err = set_array_info(mddev, &info);
5720 printk(KERN_WARNING "md: couldn't set"
5721 " array info. %d\n", err);
5731 * Commands querying/configuring an existing array:
5733 /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
5734 * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */
5735 if ((!mddev->raid_disks && !mddev->external)
5736 && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY
5737 && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE
5738 && cmd != GET_BITMAP_FILE) {
5744 * Commands even a read-only array can execute:
5748 case GET_ARRAY_INFO:
5749 err = get_array_info(mddev, argp);
5752 case GET_BITMAP_FILE:
5753 err = get_bitmap_file(mddev, argp);
5757 err = get_disk_info(mddev, argp);
5760 case RESTART_ARRAY_RW:
5761 err = restart_array(mddev);
5765 err = do_md_stop(mddev, 0, 1);
5769 err = md_set_readonly(mddev, 1);
5773 if (get_user(ro, (int __user *)(arg))) {
5779 /* if the bdev is going readonly the value of mddev->ro
5780 * does not matter, no writes are coming
5785 /* are we are already prepared for writes? */
5789 /* transitioning to readauto need only happen for
5790 * arrays that call md_write_start
5793 err = restart_array(mddev);
5796 set_disk_ro(mddev->gendisk, 0);
5803 * The remaining ioctls are changing the state of the
5804 * superblock, so we do not allow them on read-only arrays.
5805 * However non-MD ioctls (e.g. get-size) will still come through
5806 * here and hit the 'default' below, so only disallow
5807 * 'md' ioctls, and switch to rw mode if started auto-readonly.
5809 if (_IOC_TYPE(cmd) == MD_MAJOR && mddev->ro && mddev->pers) {
5810 if (mddev->ro == 2) {
5812 sysfs_notify_dirent_safe(mddev->sysfs_state);
5813 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5814 md_wakeup_thread(mddev->thread);
5825 mdu_disk_info_t info;
5826 if (copy_from_user(&info, argp, sizeof(info)))
5829 err = add_new_disk(mddev, &info);
5833 case HOT_REMOVE_DISK:
5834 err = hot_remove_disk(mddev, new_decode_dev(arg));
5838 err = hot_add_disk(mddev, new_decode_dev(arg));
5841 case SET_DISK_FAULTY:
5842 err = set_disk_faulty(mddev, new_decode_dev(arg));
5846 err = do_md_run(mddev);
5849 case SET_BITMAP_FILE:
5850 err = set_bitmap_file(mddev, (int)arg);
5860 if (mddev->hold_active == UNTIL_IOCTL &&
5862 mddev->hold_active = 0;
5863 mddev_unlock(mddev);
5872 #ifdef CONFIG_COMPAT
5873 static int md_compat_ioctl(struct block_device *bdev, fmode_t mode,
5874 unsigned int cmd, unsigned long arg)
5877 case HOT_REMOVE_DISK:
5879 case SET_DISK_FAULTY:
5880 case SET_BITMAP_FILE:
5881 /* These take in integer arg, do not convert */
5884 arg = (unsigned long)compat_ptr(arg);
5888 return md_ioctl(bdev, mode, cmd, arg);
5890 #endif /* CONFIG_COMPAT */
5892 static int md_open(struct block_device *bdev, fmode_t mode)
5895 * Succeed if we can lock the mddev, which confirms that
5896 * it isn't being stopped right now.
5898 mddev_t *mddev = mddev_find(bdev->bd_dev);
5901 if (mddev->gendisk != bdev->bd_disk) {
5902 /* we are racing with mddev_put which is discarding this
5906 /* Wait until bdev->bd_disk is definitely gone */
5907 flush_scheduled_work();
5908 /* Then retry the open from the top */
5909 return -ERESTARTSYS;
5911 BUG_ON(mddev != bdev->bd_disk->private_data);
5913 if ((err = mutex_lock_interruptible(&mddev->open_mutex)))
5917 atomic_inc(&mddev->openers);
5918 mutex_unlock(&mddev->open_mutex);
5920 check_disk_size_change(mddev->gendisk, bdev);
5925 static int md_release(struct gendisk *disk, fmode_t mode)
5927 mddev_t *mddev = disk->private_data;
5930 atomic_dec(&mddev->openers);
5935 static const struct block_device_operations md_fops =
5937 .owner = THIS_MODULE,
5939 .release = md_release,
5941 #ifdef CONFIG_COMPAT
5942 .compat_ioctl = md_compat_ioctl,
5944 .getgeo = md_getgeo,
5947 static int md_thread(void * arg)
5949 mdk_thread_t *thread = arg;
5952 * md_thread is a 'system-thread', it's priority should be very
5953 * high. We avoid resource deadlocks individually in each
5954 * raid personality. (RAID5 does preallocation) We also use RR and
5955 * the very same RT priority as kswapd, thus we will never get
5956 * into a priority inversion deadlock.
5958 * we definitely have to have equal or higher priority than
5959 * bdflush, otherwise bdflush will deadlock if there are too
5960 * many dirty RAID5 blocks.
5963 allow_signal(SIGKILL);
5964 while (!kthread_should_stop()) {
5966 /* We need to wait INTERRUPTIBLE so that
5967 * we don't add to the load-average.
5968 * That means we need to be sure no signals are
5971 if (signal_pending(current))
5972 flush_signals(current);
5974 wait_event_interruptible_timeout
5976 test_bit(THREAD_WAKEUP, &thread->flags)
5977 || kthread_should_stop(),
5980 clear_bit(THREAD_WAKEUP, &thread->flags);
5982 thread->run(thread->mddev);
5988 void md_wakeup_thread(mdk_thread_t *thread)
5991 dprintk("md: waking up MD thread %s.\n", thread->tsk->comm);
5992 set_bit(THREAD_WAKEUP, &thread->flags);
5993 wake_up(&thread->wqueue);
5997 mdk_thread_t *md_register_thread(void (*run) (mddev_t *), mddev_t *mddev,
6000 mdk_thread_t *thread;
6002 thread = kzalloc(sizeof(mdk_thread_t), GFP_KERNEL);
6006 init_waitqueue_head(&thread->wqueue);
6009 thread->mddev = mddev;
6010 thread->timeout = MAX_SCHEDULE_TIMEOUT;
6011 thread->tsk = kthread_run(md_thread, thread,
6013 mdname(thread->mddev),
6014 name ?: mddev->pers->name);
6015 if (IS_ERR(thread->tsk)) {
6022 void md_unregister_thread(mdk_thread_t *thread)
6026 dprintk("interrupting MD-thread pid %d\n", task_pid_nr(thread->tsk));
6028 kthread_stop(thread->tsk);
6032 void md_error(mddev_t *mddev, mdk_rdev_t *rdev)
6039 if (!rdev || test_bit(Faulty, &rdev->flags))
6042 if (mddev->external)
6043 set_bit(Blocked, &rdev->flags);
6045 dprintk("md_error dev:%s, rdev:(%d:%d), (caller: %p,%p,%p,%p).\n",
6047 MAJOR(rdev->bdev->bd_dev), MINOR(rdev->bdev->bd_dev),
6048 __builtin_return_address(0),__builtin_return_address(1),
6049 __builtin_return_address(2),__builtin_return_address(3));
6053 if (!mddev->pers->error_handler)
6055 mddev->pers->error_handler(mddev,rdev);
6056 if (mddev->degraded)
6057 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
6058 sysfs_notify_dirent_safe(rdev->sysfs_state);
6059 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6060 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6061 md_wakeup_thread(mddev->thread);
6062 md_new_event_inintr(mddev);
6065 /* seq_file implementation /proc/mdstat */
6067 static void status_unused(struct seq_file *seq)
6072 seq_printf(seq, "unused devices: ");
6074 list_for_each_entry(rdev, &pending_raid_disks, same_set) {
6075 char b[BDEVNAME_SIZE];
6077 seq_printf(seq, "%s ",
6078 bdevname(rdev->bdev,b));
6081 seq_printf(seq, "<none>");
6083 seq_printf(seq, "\n");
6087 static void status_resync(struct seq_file *seq, mddev_t * mddev)
6089 sector_t max_sectors, resync, res;
6090 unsigned long dt, db;
6093 unsigned int per_milli;
6095 resync = mddev->curr_resync - atomic_read(&mddev->recovery_active);
6097 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
6098 max_sectors = mddev->resync_max_sectors;
6100 max_sectors = mddev->dev_sectors;
6103 * Should not happen.
6109 /* Pick 'scale' such that (resync>>scale)*1000 will fit
6110 * in a sector_t, and (max_sectors>>scale) will fit in a
6111 * u32, as those are the requirements for sector_div.
6112 * Thus 'scale' must be at least 10
6115 if (sizeof(sector_t) > sizeof(unsigned long)) {
6116 while ( max_sectors/2 > (1ULL<<(scale+32)))
6119 res = (resync>>scale)*1000;
6120 sector_div(res, (u32)((max_sectors>>scale)+1));
6124 int i, x = per_milli/50, y = 20-x;
6125 seq_printf(seq, "[");
6126 for (i = 0; i < x; i++)
6127 seq_printf(seq, "=");
6128 seq_printf(seq, ">");
6129 for (i = 0; i < y; i++)
6130 seq_printf(seq, ".");
6131 seq_printf(seq, "] ");
6133 seq_printf(seq, " %s =%3u.%u%% (%llu/%llu)",
6134 (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)?
6136 (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)?
6138 (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
6139 "resync" : "recovery"))),
6140 per_milli/10, per_milli % 10,
6141 (unsigned long long) resync/2,
6142 (unsigned long long) max_sectors/2);
6145 * dt: time from mark until now
6146 * db: blocks written from mark until now
6147 * rt: remaining time
6149 * rt is a sector_t, so could be 32bit or 64bit.
6150 * So we divide before multiply in case it is 32bit and close
6152 * We scale the divisor (db) by 32 to avoid loosing precision
6153 * near the end of resync when the number of remaining sectors
6155 * We then divide rt by 32 after multiplying by db to compensate.
6156 * The '+1' avoids division by zero if db is very small.
6158 dt = ((jiffies - mddev->resync_mark) / HZ);
6160 db = (mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active))
6161 - mddev->resync_mark_cnt;
6163 rt = max_sectors - resync; /* number of remaining sectors */
6164 sector_div(rt, db/32+1);
6168 seq_printf(seq, " finish=%lu.%lumin", (unsigned long)rt / 60,
6169 ((unsigned long)rt % 60)/6);
6171 seq_printf(seq, " speed=%ldK/sec", db/2/dt);
6174 static void *md_seq_start(struct seq_file *seq, loff_t *pos)
6176 struct list_head *tmp;
6186 spin_lock(&all_mddevs_lock);
6187 list_for_each(tmp,&all_mddevs)
6189 mddev = list_entry(tmp, mddev_t, all_mddevs);
6191 spin_unlock(&all_mddevs_lock);
6194 spin_unlock(&all_mddevs_lock);
6196 return (void*)2;/* tail */
6200 static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
6202 struct list_head *tmp;
6203 mddev_t *next_mddev, *mddev = v;
6209 spin_lock(&all_mddevs_lock);
6211 tmp = all_mddevs.next;
6213 tmp = mddev->all_mddevs.next;
6214 if (tmp != &all_mddevs)
6215 next_mddev = mddev_get(list_entry(tmp,mddev_t,all_mddevs));
6217 next_mddev = (void*)2;
6220 spin_unlock(&all_mddevs_lock);
6228 static void md_seq_stop(struct seq_file *seq, void *v)
6232 if (mddev && v != (void*)1 && v != (void*)2)
6236 struct mdstat_info {
6240 static int md_seq_show(struct seq_file *seq, void *v)
6245 struct mdstat_info *mi = seq->private;
6246 struct bitmap *bitmap;
6248 if (v == (void*)1) {
6249 struct mdk_personality *pers;
6250 seq_printf(seq, "Personalities : ");
6251 spin_lock(&pers_lock);
6252 list_for_each_entry(pers, &pers_list, list)
6253 seq_printf(seq, "[%s] ", pers->name);
6255 spin_unlock(&pers_lock);
6256 seq_printf(seq, "\n");
6257 mi->event = atomic_read(&md_event_count);
6260 if (v == (void*)2) {
6265 if (mddev_lock(mddev) < 0)
6268 if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
6269 seq_printf(seq, "%s : %sactive", mdname(mddev),
6270 mddev->pers ? "" : "in");
6273 seq_printf(seq, " (read-only)");
6275 seq_printf(seq, " (auto-read-only)");
6276 seq_printf(seq, " %s", mddev->pers->name);
6280 list_for_each_entry(rdev, &mddev->disks, same_set) {
6281 char b[BDEVNAME_SIZE];
6282 seq_printf(seq, " %s[%d]",
6283 bdevname(rdev->bdev,b), rdev->desc_nr);
6284 if (test_bit(WriteMostly, &rdev->flags))
6285 seq_printf(seq, "(W)");
6286 if (test_bit(Faulty, &rdev->flags)) {
6287 seq_printf(seq, "(F)");
6289 } else if (rdev->raid_disk < 0)
6290 seq_printf(seq, "(S)"); /* spare */
6291 sectors += rdev->sectors;
6294 if (!list_empty(&mddev->disks)) {
6296 seq_printf(seq, "\n %llu blocks",
6297 (unsigned long long)
6298 mddev->array_sectors / 2);
6300 seq_printf(seq, "\n %llu blocks",
6301 (unsigned long long)sectors / 2);
6303 if (mddev->persistent) {
6304 if (mddev->major_version != 0 ||
6305 mddev->minor_version != 90) {
6306 seq_printf(seq," super %d.%d",
6307 mddev->major_version,
6308 mddev->minor_version);
6310 } else if (mddev->external)
6311 seq_printf(seq, " super external:%s",
6312 mddev->metadata_type);
6314 seq_printf(seq, " super non-persistent");
6317 mddev->pers->status(seq, mddev);
6318 seq_printf(seq, "\n ");
6319 if (mddev->pers->sync_request) {
6320 if (mddev->curr_resync > 2) {
6321 status_resync(seq, mddev);
6322 seq_printf(seq, "\n ");
6323 } else if (mddev->curr_resync == 1 || mddev->curr_resync == 2)
6324 seq_printf(seq, "\tresync=DELAYED\n ");
6325 else if (mddev->recovery_cp < MaxSector)
6326 seq_printf(seq, "\tresync=PENDING\n ");
6329 seq_printf(seq, "\n ");
6331 if ((bitmap = mddev->bitmap)) {
6332 unsigned long chunk_kb;
6333 unsigned long flags;
6334 spin_lock_irqsave(&bitmap->lock, flags);
6335 chunk_kb = mddev->bitmap_info.chunksize >> 10;
6336 seq_printf(seq, "bitmap: %lu/%lu pages [%luKB], "
6338 bitmap->pages - bitmap->missing_pages,
6340 (bitmap->pages - bitmap->missing_pages)
6341 << (PAGE_SHIFT - 10),
6342 chunk_kb ? chunk_kb : mddev->bitmap_info.chunksize,
6343 chunk_kb ? "KB" : "B");
6345 seq_printf(seq, ", file: ");
6346 seq_path(seq, &bitmap->file->f_path, " \t\n");
6349 seq_printf(seq, "\n");
6350 spin_unlock_irqrestore(&bitmap->lock, flags);
6353 seq_printf(seq, "\n");
6355 mddev_unlock(mddev);
6360 static const struct seq_operations md_seq_ops = {
6361 .start = md_seq_start,
6362 .next = md_seq_next,
6363 .stop = md_seq_stop,
6364 .show = md_seq_show,
6367 static int md_seq_open(struct inode *inode, struct file *file)
6370 struct mdstat_info *mi = kmalloc(sizeof(*mi), GFP_KERNEL);
6374 error = seq_open(file, &md_seq_ops);
6378 struct seq_file *p = file->private_data;
6380 mi->event = atomic_read(&md_event_count);
6385 static unsigned int mdstat_poll(struct file *filp, poll_table *wait)
6387 struct seq_file *m = filp->private_data;
6388 struct mdstat_info *mi = m->private;
6391 poll_wait(filp, &md_event_waiters, wait);
6393 /* always allow read */
6394 mask = POLLIN | POLLRDNORM;
6396 if (mi->event != atomic_read(&md_event_count))
6397 mask |= POLLERR | POLLPRI;
6401 static const struct file_operations md_seq_fops = {
6402 .owner = THIS_MODULE,
6403 .open = md_seq_open,
6405 .llseek = seq_lseek,
6406 .release = seq_release_private,
6407 .poll = mdstat_poll,
6410 int register_md_personality(struct mdk_personality *p)
6412 spin_lock(&pers_lock);
6413 list_add_tail(&p->list, &pers_list);
6414 printk(KERN_INFO "md: %s personality registered for level %d\n", p->name, p->level);
6415 spin_unlock(&pers_lock);
6419 int unregister_md_personality(struct mdk_personality *p)
6421 printk(KERN_INFO "md: %s personality unregistered\n", p->name);
6422 spin_lock(&pers_lock);
6423 list_del_init(&p->list);
6424 spin_unlock(&pers_lock);
6428 static int is_mddev_idle(mddev_t *mddev, int init)
6436 rdev_for_each_rcu(rdev, mddev) {
6437 struct gendisk *disk = rdev->bdev->bd_contains->bd_disk;
6438 curr_events = (int)part_stat_read(&disk->part0, sectors[0]) +
6439 (int)part_stat_read(&disk->part0, sectors[1]) -
6440 atomic_read(&disk->sync_io);
6441 /* sync IO will cause sync_io to increase before the disk_stats
6442 * as sync_io is counted when a request starts, and
6443 * disk_stats is counted when it completes.
6444 * So resync activity will cause curr_events to be smaller than
6445 * when there was no such activity.
6446 * non-sync IO will cause disk_stat to increase without
6447 * increasing sync_io so curr_events will (eventually)
6448 * be larger than it was before. Once it becomes
6449 * substantially larger, the test below will cause
6450 * the array to appear non-idle, and resync will slow
6452 * If there is a lot of outstanding resync activity when
6453 * we set last_event to curr_events, then all that activity
6454 * completing might cause the array to appear non-idle
6455 * and resync will be slowed down even though there might
6456 * not have been non-resync activity. This will only
6457 * happen once though. 'last_events' will soon reflect
6458 * the state where there is little or no outstanding
6459 * resync requests, and further resync activity will
6460 * always make curr_events less than last_events.
6463 if (init || curr_events - rdev->last_events > 64) {
6464 rdev->last_events = curr_events;
6472 void md_done_sync(mddev_t *mddev, int blocks, int ok)
6474 /* another "blocks" (512byte) blocks have been synced */
6475 atomic_sub(blocks, &mddev->recovery_active);
6476 wake_up(&mddev->recovery_wait);
6478 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6479 md_wakeup_thread(mddev->thread);
6480 // stop recovery, signal do_sync ....
6485 /* md_write_start(mddev, bi)
6486 * If we need to update some array metadata (e.g. 'active' flag
6487 * in superblock) before writing, schedule a superblock update
6488 * and wait for it to complete.
6490 void md_write_start(mddev_t *mddev, struct bio *bi)
6493 if (bio_data_dir(bi) != WRITE)
6496 BUG_ON(mddev->ro == 1);
6497 if (mddev->ro == 2) {
6498 /* need to switch to read/write */
6500 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6501 md_wakeup_thread(mddev->thread);
6502 md_wakeup_thread(mddev->sync_thread);
6505 atomic_inc(&mddev->writes_pending);
6506 if (mddev->safemode == 1)
6507 mddev->safemode = 0;
6508 if (mddev->in_sync) {
6509 spin_lock_irq(&mddev->write_lock);
6510 if (mddev->in_sync) {
6512 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
6513 md_wakeup_thread(mddev->thread);
6516 spin_unlock_irq(&mddev->write_lock);
6519 sysfs_notify_dirent_safe(mddev->sysfs_state);
6520 wait_event(mddev->sb_wait,
6521 !test_bit(MD_CHANGE_CLEAN, &mddev->flags) &&
6522 !test_bit(MD_CHANGE_PENDING, &mddev->flags));
6525 void md_write_end(mddev_t *mddev)
6527 if (atomic_dec_and_test(&mddev->writes_pending)) {
6528 if (mddev->safemode == 2)
6529 md_wakeup_thread(mddev->thread);
6530 else if (mddev->safemode_delay)
6531 mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay);
6535 /* md_allow_write(mddev)
6536 * Calling this ensures that the array is marked 'active' so that writes
6537 * may proceed without blocking. It is important to call this before
6538 * attempting a GFP_KERNEL allocation while holding the mddev lock.
6539 * Must be called with mddev_lock held.
6541 * In the ->external case MD_CHANGE_CLEAN can not be cleared until mddev->lock
6542 * is dropped, so return -EAGAIN after notifying userspace.
6544 int md_allow_write(mddev_t *mddev)
6550 if (!mddev->pers->sync_request)
6553 spin_lock_irq(&mddev->write_lock);
6554 if (mddev->in_sync) {
6556 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
6557 if (mddev->safemode_delay &&
6558 mddev->safemode == 0)
6559 mddev->safemode = 1;
6560 spin_unlock_irq(&mddev->write_lock);
6561 md_update_sb(mddev, 0);
6562 sysfs_notify_dirent_safe(mddev->sysfs_state);
6564 spin_unlock_irq(&mddev->write_lock);
6566 if (test_bit(MD_CHANGE_CLEAN, &mddev->flags))
6571 EXPORT_SYMBOL_GPL(md_allow_write);
6573 #define SYNC_MARKS 10
6574 #define SYNC_MARK_STEP (3*HZ)
6575 void md_do_sync(mddev_t *mddev)
6578 unsigned int currspeed = 0,
6580 sector_t max_sectors,j, io_sectors;
6581 unsigned long mark[SYNC_MARKS];
6582 sector_t mark_cnt[SYNC_MARKS];
6584 struct list_head *tmp;
6585 sector_t last_check;
6590 /* just incase thread restarts... */
6591 if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
6593 if (mddev->ro) /* never try to sync a read-only array */
6596 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
6597 if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
6598 desc = "data-check";
6599 else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
6600 desc = "requested-resync";
6603 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
6608 /* we overload curr_resync somewhat here.
6609 * 0 == not engaged in resync at all
6610 * 2 == checking that there is no conflict with another sync
6611 * 1 == like 2, but have yielded to allow conflicting resync to
6613 * other == active in resync - this many blocks
6615 * Before starting a resync we must have set curr_resync to
6616 * 2, and then checked that every "conflicting" array has curr_resync
6617 * less than ours. When we find one that is the same or higher
6618 * we wait on resync_wait. To avoid deadlock, we reduce curr_resync
6619 * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
6620 * This will mean we have to start checking from the beginning again.
6625 mddev->curr_resync = 2;
6628 if (kthread_should_stop())
6629 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6631 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
6633 for_each_mddev(mddev2, tmp) {
6634 if (mddev2 == mddev)
6636 if (!mddev->parallel_resync
6637 && mddev2->curr_resync
6638 && match_mddev_units(mddev, mddev2)) {
6640 if (mddev < mddev2 && mddev->curr_resync == 2) {
6641 /* arbitrarily yield */
6642 mddev->curr_resync = 1;
6643 wake_up(&resync_wait);
6645 if (mddev > mddev2 && mddev->curr_resync == 1)
6646 /* no need to wait here, we can wait the next
6647 * time 'round when curr_resync == 2
6650 /* We need to wait 'interruptible' so as not to
6651 * contribute to the load average, and not to
6652 * be caught by 'softlockup'
6654 prepare_to_wait(&resync_wait, &wq, TASK_INTERRUPTIBLE);
6655 if (!kthread_should_stop() &&
6656 mddev2->curr_resync >= mddev->curr_resync) {
6657 printk(KERN_INFO "md: delaying %s of %s"
6658 " until %s has finished (they"
6659 " share one or more physical units)\n",
6660 desc, mdname(mddev), mdname(mddev2));
6662 if (signal_pending(current))
6663 flush_signals(current);
6665 finish_wait(&resync_wait, &wq);
6668 finish_wait(&resync_wait, &wq);
6671 } while (mddev->curr_resync < 2);
6674 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
6675 /* resync follows the size requested by the personality,
6676 * which defaults to physical size, but can be virtual size
6678 max_sectors = mddev->resync_max_sectors;
6679 mddev->resync_mismatches = 0;
6680 /* we don't use the checkpoint if there's a bitmap */
6681 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
6682 j = mddev->resync_min;
6683 else if (!mddev->bitmap)
6684 j = mddev->recovery_cp;
6686 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
6687 max_sectors = mddev->dev_sectors;
6689 /* recovery follows the physical size of devices */
6690 max_sectors = mddev->dev_sectors;
6693 list_for_each_entry_rcu(rdev, &mddev->disks, same_set)
6694 if (rdev->raid_disk >= 0 &&
6695 !test_bit(Faulty, &rdev->flags) &&
6696 !test_bit(In_sync, &rdev->flags) &&
6697 rdev->recovery_offset < j)
6698 j = rdev->recovery_offset;
6702 printk(KERN_INFO "md: %s of RAID array %s\n", desc, mdname(mddev));
6703 printk(KERN_INFO "md: minimum _guaranteed_ speed:"
6704 " %d KB/sec/disk.\n", speed_min(mddev));
6705 printk(KERN_INFO "md: using maximum available idle IO bandwidth "
6706 "(but not more than %d KB/sec) for %s.\n",
6707 speed_max(mddev), desc);
6709 is_mddev_idle(mddev, 1); /* this initializes IO event counters */
6712 for (m = 0; m < SYNC_MARKS; m++) {
6714 mark_cnt[m] = io_sectors;
6717 mddev->resync_mark = mark[last_mark];
6718 mddev->resync_mark_cnt = mark_cnt[last_mark];
6721 * Tune reconstruction:
6723 window = 32*(PAGE_SIZE/512);
6724 printk(KERN_INFO "md: using %dk window, over a total of %llu blocks.\n",
6725 window/2,(unsigned long long) max_sectors/2);
6727 atomic_set(&mddev->recovery_active, 0);
6732 "md: resuming %s of %s from checkpoint.\n",
6733 desc, mdname(mddev));
6734 mddev->curr_resync = j;
6736 mddev->curr_resync_completed = mddev->curr_resync;
6738 while (j < max_sectors) {
6743 if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
6744 ((mddev->curr_resync > mddev->curr_resync_completed &&
6745 (mddev->curr_resync - mddev->curr_resync_completed)
6746 > (max_sectors >> 4)) ||
6747 (j - mddev->curr_resync_completed)*2
6748 >= mddev->resync_max - mddev->curr_resync_completed
6750 /* time to update curr_resync_completed */
6751 blk_unplug(mddev->queue);
6752 wait_event(mddev->recovery_wait,
6753 atomic_read(&mddev->recovery_active) == 0);
6754 mddev->curr_resync_completed =
6756 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
6757 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
6760 while (j >= mddev->resync_max && !kthread_should_stop()) {
6761 /* As this condition is controlled by user-space,
6762 * we can block indefinitely, so use '_interruptible'
6763 * to avoid triggering warnings.
6765 flush_signals(current); /* just in case */
6766 wait_event_interruptible(mddev->recovery_wait,
6767 mddev->resync_max > j
6768 || kthread_should_stop());
6771 if (kthread_should_stop())
6774 sectors = mddev->pers->sync_request(mddev, j, &skipped,
6775 currspeed < speed_min(mddev));
6777 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6781 if (!skipped) { /* actual IO requested */
6782 io_sectors += sectors;
6783 atomic_add(sectors, &mddev->recovery_active);
6787 if (j>1) mddev->curr_resync = j;
6788 mddev->curr_mark_cnt = io_sectors;
6789 if (last_check == 0)
6790 /* this is the earliers that rebuilt will be
6791 * visible in /proc/mdstat
6793 md_new_event(mddev);
6795 if (last_check + window > io_sectors || j == max_sectors)
6798 last_check = io_sectors;
6800 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
6804 if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) {
6806 int next = (last_mark+1) % SYNC_MARKS;
6808 mddev->resync_mark = mark[next];
6809 mddev->resync_mark_cnt = mark_cnt[next];
6810 mark[next] = jiffies;
6811 mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active);
6816 if (kthread_should_stop())
6821 * this loop exits only if either when we are slower than
6822 * the 'hard' speed limit, or the system was IO-idle for
6824 * the system might be non-idle CPU-wise, but we only care
6825 * about not overloading the IO subsystem. (things like an
6826 * e2fsck being done on the RAID array should execute fast)
6828 blk_unplug(mddev->queue);
6831 currspeed = ((unsigned long)(io_sectors-mddev->resync_mark_cnt))/2
6832 /((jiffies-mddev->resync_mark)/HZ +1) +1;
6834 if (currspeed > speed_min(mddev)) {
6835 if ((currspeed > speed_max(mddev)) ||
6836 !is_mddev_idle(mddev, 0)) {
6842 printk(KERN_INFO "md: %s: %s done.\n",mdname(mddev), desc);
6844 * this also signals 'finished resyncing' to md_stop
6847 blk_unplug(mddev->queue);
6849 wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
6851 /* tell personality that we are finished */
6852 mddev->pers->sync_request(mddev, max_sectors, &skipped, 1);
6854 if (!test_bit(MD_RECOVERY_CHECK, &mddev->recovery) &&
6855 mddev->curr_resync > 2) {
6856 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
6857 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
6858 if (mddev->curr_resync >= mddev->recovery_cp) {
6860 "md: checkpointing %s of %s.\n",
6861 desc, mdname(mddev));
6862 mddev->recovery_cp = mddev->curr_resync;
6865 mddev->recovery_cp = MaxSector;
6867 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery))
6868 mddev->curr_resync = MaxSector;
6870 list_for_each_entry_rcu(rdev, &mddev->disks, same_set)
6871 if (rdev->raid_disk >= 0 &&
6872 mddev->delta_disks >= 0 &&
6873 !test_bit(Faulty, &rdev->flags) &&
6874 !test_bit(In_sync, &rdev->flags) &&
6875 rdev->recovery_offset < mddev->curr_resync)
6876 rdev->recovery_offset = mddev->curr_resync;
6880 set_bit(MD_CHANGE_DEVS, &mddev->flags);
6883 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
6884 /* We completed so min/max setting can be forgotten if used. */
6885 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
6886 mddev->resync_min = 0;
6887 mddev->resync_max = MaxSector;
6888 } else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
6889 mddev->resync_min = mddev->curr_resync_completed;
6890 mddev->curr_resync = 0;
6891 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery))
6892 mddev->curr_resync_completed = 0;
6893 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
6894 wake_up(&resync_wait);
6895 set_bit(MD_RECOVERY_DONE, &mddev->recovery);
6896 md_wakeup_thread(mddev->thread);
6901 * got a signal, exit.
6904 "md: md_do_sync() got signal ... exiting\n");
6905 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6909 EXPORT_SYMBOL_GPL(md_do_sync);
6912 static int remove_and_add_spares(mddev_t *mddev)
6917 mddev->curr_resync_completed = 0;
6919 list_for_each_entry(rdev, &mddev->disks, same_set)
6920 if (rdev->raid_disk >= 0 &&
6921 !test_bit(Blocked, &rdev->flags) &&
6922 (test_bit(Faulty, &rdev->flags) ||
6923 ! test_bit(In_sync, &rdev->flags)) &&
6924 atomic_read(&rdev->nr_pending)==0) {
6925 if (mddev->pers->hot_remove_disk(
6926 mddev, rdev->raid_disk)==0) {
6928 sprintf(nm,"rd%d", rdev->raid_disk);
6929 sysfs_remove_link(&mddev->kobj, nm);
6930 rdev->raid_disk = -1;
6934 if (mddev->degraded && ! mddev->ro && !mddev->recovery_disabled) {
6935 list_for_each_entry(rdev, &mddev->disks, same_set) {
6936 if (rdev->raid_disk >= 0 &&
6937 !test_bit(In_sync, &rdev->flags) &&
6938 !test_bit(Blocked, &rdev->flags))
6940 if (rdev->raid_disk < 0
6941 && !test_bit(Faulty, &rdev->flags)) {
6942 rdev->recovery_offset = 0;
6944 hot_add_disk(mddev, rdev) == 0) {
6946 sprintf(nm, "rd%d", rdev->raid_disk);
6947 if (sysfs_create_link(&mddev->kobj,
6949 /* failure here is OK */;
6951 md_new_event(mddev);
6952 set_bit(MD_CHANGE_DEVS, &mddev->flags);
6961 * This routine is regularly called by all per-raid-array threads to
6962 * deal with generic issues like resync and super-block update.
6963 * Raid personalities that don't have a thread (linear/raid0) do not
6964 * need this as they never do any recovery or update the superblock.
6966 * It does not do any resync itself, but rather "forks" off other threads
6967 * to do that as needed.
6968 * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
6969 * "->recovery" and create a thread at ->sync_thread.
6970 * When the thread finishes it sets MD_RECOVERY_DONE
6971 * and wakeups up this thread which will reap the thread and finish up.
6972 * This thread also removes any faulty devices (with nr_pending == 0).
6974 * The overall approach is:
6975 * 1/ if the superblock needs updating, update it.
6976 * 2/ If a recovery thread is running, don't do anything else.
6977 * 3/ If recovery has finished, clean up, possibly marking spares active.
6978 * 4/ If there are any faulty devices, remove them.
6979 * 5/ If array is degraded, try to add spares devices
6980 * 6/ If array has spares or is not in-sync, start a resync thread.
6982 void md_check_recovery(mddev_t *mddev)
6988 bitmap_daemon_work(mddev);
6993 if (signal_pending(current)) {
6994 if (mddev->pers->sync_request && !mddev->external) {
6995 printk(KERN_INFO "md: %s in immediate safe mode\n",
6997 mddev->safemode = 2;
6999 flush_signals(current);
7002 if (mddev->ro && !test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
7005 (mddev->flags && !mddev->external) ||
7006 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
7007 test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
7008 (mddev->external == 0 && mddev->safemode == 1) ||
7009 (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending)
7010 && !mddev->in_sync && mddev->recovery_cp == MaxSector)
7014 if (mddev_trylock(mddev)) {
7018 /* Only thing we do on a ro array is remove
7021 remove_and_add_spares(mddev);
7022 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7026 if (!mddev->external) {
7028 spin_lock_irq(&mddev->write_lock);
7029 if (mddev->safemode &&
7030 !atomic_read(&mddev->writes_pending) &&
7032 mddev->recovery_cp == MaxSector) {
7035 if (mddev->persistent)
7036 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
7038 if (mddev->safemode == 1)
7039 mddev->safemode = 0;
7040 spin_unlock_irq(&mddev->write_lock);
7042 sysfs_notify_dirent_safe(mddev->sysfs_state);
7046 md_update_sb(mddev, 0);
7048 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
7049 !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
7050 /* resync/recovery still happening */
7051 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7054 if (mddev->sync_thread) {
7055 /* resync has finished, collect result */
7056 md_unregister_thread(mddev->sync_thread);
7057 mddev->sync_thread = NULL;
7058 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
7059 !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
7061 /* activate any spares */
7062 if (mddev->pers->spare_active(mddev))
7063 sysfs_notify(&mddev->kobj, NULL,
7066 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
7067 mddev->pers->finish_reshape)
7068 mddev->pers->finish_reshape(mddev);
7069 md_update_sb(mddev, 1);
7071 /* if array is no-longer degraded, then any saved_raid_disk
7072 * information must be scrapped
7074 if (!mddev->degraded)
7075 list_for_each_entry(rdev, &mddev->disks, same_set)
7076 rdev->saved_raid_disk = -1;
7078 mddev->recovery = 0;
7079 /* flag recovery needed just to double check */
7080 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7081 sysfs_notify_dirent_safe(mddev->sysfs_action);
7082 md_new_event(mddev);
7085 /* Set RUNNING before clearing NEEDED to avoid
7086 * any transients in the value of "sync_action".
7088 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
7089 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7090 /* Clear some bits that don't mean anything, but
7093 clear_bit(MD_RECOVERY_INTR, &mddev->recovery);
7094 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
7096 if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
7098 /* no recovery is running.
7099 * remove any failed drives, then
7100 * add spares if possible.
7101 * Spare are also removed and re-added, to allow
7102 * the personality to fail the re-add.
7105 if (mddev->reshape_position != MaxSector) {
7106 if (mddev->pers->check_reshape == NULL ||
7107 mddev->pers->check_reshape(mddev) != 0)
7108 /* Cannot proceed */
7110 set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
7111 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
7112 } else if ((spares = remove_and_add_spares(mddev))) {
7113 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
7114 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
7115 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
7116 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
7117 } else if (mddev->recovery_cp < MaxSector) {
7118 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
7119 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
7120 } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
7121 /* nothing to be done ... */
7124 if (mddev->pers->sync_request) {
7125 if (spares && mddev->bitmap && ! mddev->bitmap->file) {
7126 /* We are adding a device or devices to an array
7127 * which has the bitmap stored on all devices.
7128 * So make sure all bitmap pages get written
7130 bitmap_write_all(mddev->bitmap);
7132 mddev->sync_thread = md_register_thread(md_do_sync,
7135 if (!mddev->sync_thread) {
7136 printk(KERN_ERR "%s: could not start resync"
7139 /* leave the spares where they are, it shouldn't hurt */
7140 mddev->recovery = 0;
7142 md_wakeup_thread(mddev->sync_thread);
7143 sysfs_notify_dirent_safe(mddev->sysfs_action);
7144 md_new_event(mddev);
7147 if (!mddev->sync_thread) {
7148 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
7149 if (test_and_clear_bit(MD_RECOVERY_RECOVER,
7151 if (mddev->sysfs_action)
7152 sysfs_notify_dirent_safe(mddev->sysfs_action);
7154 mddev_unlock(mddev);
7158 void md_wait_for_blocked_rdev(mdk_rdev_t *rdev, mddev_t *mddev)
7160 sysfs_notify_dirent_safe(rdev->sysfs_state);
7161 wait_event_timeout(rdev->blocked_wait,
7162 !test_bit(Blocked, &rdev->flags),
7163 msecs_to_jiffies(5000));
7164 rdev_dec_pending(rdev, mddev);
7166 EXPORT_SYMBOL(md_wait_for_blocked_rdev);
7168 static int md_notify_reboot(struct notifier_block *this,
7169 unsigned long code, void *x)
7171 struct list_head *tmp;
7174 if ((code == SYS_DOWN) || (code == SYS_HALT) || (code == SYS_POWER_OFF)) {
7176 printk(KERN_INFO "md: stopping all md devices.\n");
7178 for_each_mddev(mddev, tmp)
7179 if (mddev_trylock(mddev)) {
7180 /* Force a switch to readonly even array
7181 * appears to still be in use. Hence
7184 md_set_readonly(mddev, 100);
7185 mddev_unlock(mddev);
7188 * certain more exotic SCSI devices are known to be
7189 * volatile wrt too early system reboots. While the
7190 * right place to handle this issue is the given
7191 * driver, we do want to have a safe RAID driver ...
7198 static struct notifier_block md_notifier = {
7199 .notifier_call = md_notify_reboot,
7201 .priority = INT_MAX, /* before any real devices */
7204 static void md_geninit(void)
7206 dprintk("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
7208 proc_create("mdstat", S_IRUGO, NULL, &md_seq_fops);
7211 static int __init md_init(void)
7213 if (register_blkdev(MD_MAJOR, "md"))
7215 if ((mdp_major=register_blkdev(0, "mdp"))<=0) {
7216 unregister_blkdev(MD_MAJOR, "md");
7219 blk_register_region(MKDEV(MD_MAJOR, 0), 1UL<<MINORBITS, THIS_MODULE,
7220 md_probe, NULL, NULL);
7221 blk_register_region(MKDEV(mdp_major, 0), 1UL<<MINORBITS, THIS_MODULE,
7222 md_probe, NULL, NULL);
7224 register_reboot_notifier(&md_notifier);
7225 raid_table_header = register_sysctl_table(raid_root_table);
7235 * Searches all registered partitions for autorun RAID arrays
7239 static LIST_HEAD(all_detected_devices);
7240 struct detected_devices_node {
7241 struct list_head list;
7245 void md_autodetect_dev(dev_t dev)
7247 struct detected_devices_node *node_detected_dev;
7249 node_detected_dev = kzalloc(sizeof(*node_detected_dev), GFP_KERNEL);
7250 if (node_detected_dev) {
7251 node_detected_dev->dev = dev;
7252 list_add_tail(&node_detected_dev->list, &all_detected_devices);
7254 printk(KERN_CRIT "md: md_autodetect_dev: kzalloc failed"
7255 ", skipping dev(%d,%d)\n", MAJOR(dev), MINOR(dev));
7260 static void autostart_arrays(int part)
7263 struct detected_devices_node *node_detected_dev;
7265 int i_scanned, i_passed;
7270 printk(KERN_INFO "md: Autodetecting RAID arrays.\n");
7272 while (!list_empty(&all_detected_devices) && i_scanned < INT_MAX) {
7274 node_detected_dev = list_entry(all_detected_devices.next,
7275 struct detected_devices_node, list);
7276 list_del(&node_detected_dev->list);
7277 dev = node_detected_dev->dev;
7278 kfree(node_detected_dev);
7279 rdev = md_import_device(dev,0, 90);
7283 if (test_bit(Faulty, &rdev->flags)) {
7287 set_bit(AutoDetected, &rdev->flags);
7288 list_add(&rdev->same_set, &pending_raid_disks);
7292 printk(KERN_INFO "md: Scanned %d and added %d devices.\n",
7293 i_scanned, i_passed);
7295 autorun_devices(part);
7298 #endif /* !MODULE */
7300 static __exit void md_exit(void)
7303 struct list_head *tmp;
7305 blk_unregister_region(MKDEV(MD_MAJOR,0), 1U << MINORBITS);
7306 blk_unregister_region(MKDEV(mdp_major,0), 1U << MINORBITS);
7308 unregister_blkdev(MD_MAJOR,"md");
7309 unregister_blkdev(mdp_major, "mdp");
7310 unregister_reboot_notifier(&md_notifier);
7311 unregister_sysctl_table(raid_table_header);
7312 remove_proc_entry("mdstat", NULL);
7313 for_each_mddev(mddev, tmp) {
7314 export_array(mddev);
7315 mddev->hold_active = 0;
7319 subsys_initcall(md_init);
7320 module_exit(md_exit)
7322 static int get_ro(char *buffer, struct kernel_param *kp)
7324 return sprintf(buffer, "%d", start_readonly);
7326 static int set_ro(const char *val, struct kernel_param *kp)
7329 int num = simple_strtoul(val, &e, 10);
7330 if (*val && (*e == '\0' || *e == '\n')) {
7331 start_readonly = num;
7337 module_param_call(start_ro, set_ro, get_ro, NULL, S_IRUSR|S_IWUSR);
7338 module_param(start_dirty_degraded, int, S_IRUGO|S_IWUSR);
7340 module_param_call(new_array, add_named_array, NULL, NULL, S_IWUSR);
7342 EXPORT_SYMBOL(register_md_personality);
7343 EXPORT_SYMBOL(unregister_md_personality);
7344 EXPORT_SYMBOL(md_error);
7345 EXPORT_SYMBOL(md_done_sync);
7346 EXPORT_SYMBOL(md_write_start);
7347 EXPORT_SYMBOL(md_write_end);
7348 EXPORT_SYMBOL(md_register_thread);
7349 EXPORT_SYMBOL(md_unregister_thread);
7350 EXPORT_SYMBOL(md_wakeup_thread);
7351 EXPORT_SYMBOL(md_check_recovery);
7352 MODULE_LICENSE("GPL");
7353 MODULE_DESCRIPTION("MD RAID framework");
7355 MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR);