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>
56 #define dprintk(x...) ((void)(DEBUG && printk(x)))
60 static void autostart_arrays(int part);
63 static LIST_HEAD(pers_list);
64 static DEFINE_SPINLOCK(pers_lock);
66 static void md_print_devices(void);
68 static DECLARE_WAIT_QUEUE_HEAD(resync_wait);
70 #define MD_BUG(x...) { printk("md: bug in file %s, line %d\n", __FILE__, __LINE__); md_print_devices(); }
73 * Default number of read corrections we'll attempt on an rdev
74 * before ejecting it from the array. We divide the read error
75 * count by 2 for every hour elapsed between read errors.
77 #define MD_DEFAULT_MAX_CORRECTED_READ_ERRORS 20
79 * Current RAID-1,4,5 parallel reconstruction 'guaranteed speed limit'
80 * is 1000 KB/sec, so the extra system load does not show up that much.
81 * Increase it if you want to have more _guaranteed_ speed. Note that
82 * the RAID driver will use the maximum available bandwidth if the IO
83 * subsystem is idle. There is also an 'absolute maximum' reconstruction
84 * speed limit - in case reconstruction slows down your system despite
87 * you can change it via /proc/sys/dev/raid/speed_limit_min and _max.
88 * or /sys/block/mdX/md/sync_speed_{min,max}
91 static int sysctl_speed_limit_min = 1000;
92 static int sysctl_speed_limit_max = 200000;
93 static inline int speed_min(mddev_t *mddev)
95 return mddev->sync_speed_min ?
96 mddev->sync_speed_min : sysctl_speed_limit_min;
99 static inline int speed_max(mddev_t *mddev)
101 return mddev->sync_speed_max ?
102 mddev->sync_speed_max : sysctl_speed_limit_max;
105 static struct ctl_table_header *raid_table_header;
107 static ctl_table raid_table[] = {
109 .procname = "speed_limit_min",
110 .data = &sysctl_speed_limit_min,
111 .maxlen = sizeof(int),
112 .mode = S_IRUGO|S_IWUSR,
113 .proc_handler = proc_dointvec,
116 .procname = "speed_limit_max",
117 .data = &sysctl_speed_limit_max,
118 .maxlen = sizeof(int),
119 .mode = S_IRUGO|S_IWUSR,
120 .proc_handler = proc_dointvec,
125 static ctl_table raid_dir_table[] = {
129 .mode = S_IRUGO|S_IXUGO,
135 static ctl_table raid_root_table[] = {
140 .child = raid_dir_table,
145 static const struct block_device_operations md_fops;
147 static int start_readonly;
150 * We have a system wide 'event count' that is incremented
151 * on any 'interesting' event, and readers of /proc/mdstat
152 * can use 'poll' or 'select' to find out when the event
156 * start array, stop array, error, add device, remove device,
157 * start build, activate spare
159 static DECLARE_WAIT_QUEUE_HEAD(md_event_waiters);
160 static atomic_t md_event_count;
161 void md_new_event(mddev_t *mddev)
163 atomic_inc(&md_event_count);
164 wake_up(&md_event_waiters);
166 EXPORT_SYMBOL_GPL(md_new_event);
168 /* Alternate version that can be called from interrupts
169 * when calling sysfs_notify isn't needed.
171 static void md_new_event_inintr(mddev_t *mddev)
173 atomic_inc(&md_event_count);
174 wake_up(&md_event_waiters);
178 * Enables to iterate over all existing md arrays
179 * all_mddevs_lock protects this list.
181 static LIST_HEAD(all_mddevs);
182 static DEFINE_SPINLOCK(all_mddevs_lock);
186 * iterates through all used mddevs in the system.
187 * We take care to grab the all_mddevs_lock whenever navigating
188 * the list, and to always hold a refcount when unlocked.
189 * Any code which breaks out of this loop while own
190 * a reference to the current mddev and must mddev_put it.
192 #define for_each_mddev(mddev,tmp) \
194 for (({ spin_lock(&all_mddevs_lock); \
195 tmp = all_mddevs.next; \
197 ({ if (tmp != &all_mddevs) \
198 mddev_get(list_entry(tmp, mddev_t, all_mddevs));\
199 spin_unlock(&all_mddevs_lock); \
200 if (mddev) mddev_put(mddev); \
201 mddev = list_entry(tmp, mddev_t, all_mddevs); \
202 tmp != &all_mddevs;}); \
203 ({ spin_lock(&all_mddevs_lock); \
208 /* Rather than calling directly into the personality make_request function,
209 * IO requests come here first so that we can check if the device is
210 * being suspended pending a reconfiguration.
211 * We hold a refcount over the call to ->make_request. By the time that
212 * call has finished, the bio has been linked into some internal structure
213 * and so is visible to ->quiesce(), so we don't need the refcount any more.
215 static int md_make_request(struct request_queue *q, struct bio *bio)
217 const int rw = bio_data_dir(bio);
218 mddev_t *mddev = q->queuedata;
222 if (mddev == NULL || mddev->pers == NULL) {
227 if (mddev->suspended || mddev->barrier) {
230 prepare_to_wait(&mddev->sb_wait, &__wait,
231 TASK_UNINTERRUPTIBLE);
232 if (!mddev->suspended && !mddev->barrier)
238 finish_wait(&mddev->sb_wait, &__wait);
240 atomic_inc(&mddev->active_io);
243 rv = mddev->pers->make_request(q, bio);
245 cpu = part_stat_lock();
246 part_stat_inc(cpu, &mddev->gendisk->part0, ios[rw]);
247 part_stat_add(cpu, &mddev->gendisk->part0, sectors[rw],
251 if (atomic_dec_and_test(&mddev->active_io) && mddev->suspended)
252 wake_up(&mddev->sb_wait);
257 static void mddev_suspend(mddev_t *mddev)
259 BUG_ON(mddev->suspended);
260 mddev->suspended = 1;
262 wait_event(mddev->sb_wait, atomic_read(&mddev->active_io) == 0);
263 mddev->pers->quiesce(mddev, 1);
264 md_unregister_thread(mddev->thread);
265 mddev->thread = NULL;
266 /* we now know that no code is executing in the personality module,
267 * except possibly the tail end of a ->bi_end_io function, but that
268 * is certain to complete before the module has a chance to get
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->queue, 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 mddev_t * mddev_find(dev_t unit)
421 mddev_t *mddev, *new = NULL;
424 spin_lock(&all_mddevs_lock);
427 list_for_each_entry(mddev, &all_mddevs, all_mddevs)
428 if (mddev->unit == unit) {
430 spin_unlock(&all_mddevs_lock);
436 list_add(&new->all_mddevs, &all_mddevs);
437 spin_unlock(&all_mddevs_lock);
438 new->hold_active = UNTIL_IOCTL;
442 /* find an unused unit number */
443 static int next_minor = 512;
444 int start = next_minor;
448 dev = MKDEV(MD_MAJOR, next_minor);
450 if (next_minor > MINORMASK)
452 if (next_minor == start) {
453 /* Oh dear, all in use. */
454 spin_unlock(&all_mddevs_lock);
460 list_for_each_entry(mddev, &all_mddevs, all_mddevs)
461 if (mddev->unit == dev) {
467 new->md_minor = MINOR(dev);
468 new->hold_active = UNTIL_STOP;
469 list_add(&new->all_mddevs, &all_mddevs);
470 spin_unlock(&all_mddevs_lock);
473 spin_unlock(&all_mddevs_lock);
475 new = kzalloc(sizeof(*new), GFP_KERNEL);
480 if (MAJOR(unit) == MD_MAJOR)
481 new->md_minor = MINOR(unit);
483 new->md_minor = MINOR(unit) >> MdpMinorShift;
485 mutex_init(&new->open_mutex);
486 mutex_init(&new->reconfig_mutex);
487 mutex_init(&new->bitmap_info.mutex);
488 INIT_LIST_HEAD(&new->disks);
489 INIT_LIST_HEAD(&new->all_mddevs);
490 init_timer(&new->safemode_timer);
491 atomic_set(&new->active, 1);
492 atomic_set(&new->openers, 0);
493 atomic_set(&new->active_io, 0);
494 spin_lock_init(&new->write_lock);
495 atomic_set(&new->flush_pending, 0);
496 init_waitqueue_head(&new->sb_wait);
497 init_waitqueue_head(&new->recovery_wait);
498 new->reshape_position = MaxSector;
500 new->resync_max = MaxSector;
501 new->level = LEVEL_NONE;
506 static inline int mddev_lock(mddev_t * mddev)
508 return mutex_lock_interruptible(&mddev->reconfig_mutex);
511 static inline int mddev_is_locked(mddev_t *mddev)
513 return mutex_is_locked(&mddev->reconfig_mutex);
516 static inline int mddev_trylock(mddev_t * mddev)
518 return mutex_trylock(&mddev->reconfig_mutex);
521 static struct attribute_group md_redundancy_group;
523 static void mddev_unlock(mddev_t * mddev)
525 if (mddev->to_remove) {
526 /* These cannot be removed under reconfig_mutex as
527 * an access to the files will try to take reconfig_mutex
528 * while holding the file unremovable, which leads to
530 * So hold open_mutex instead - we are allowed to take
531 * it while holding reconfig_mutex, and md_run can
532 * use it to wait for the remove to complete.
534 struct attribute_group *to_remove = mddev->to_remove;
535 mddev->to_remove = NULL;
536 mutex_lock(&mddev->open_mutex);
537 mutex_unlock(&mddev->reconfig_mutex);
539 if (to_remove != &md_redundancy_group)
540 sysfs_remove_group(&mddev->kobj, to_remove);
541 if (mddev->pers == NULL ||
542 mddev->pers->sync_request == NULL) {
543 sysfs_remove_group(&mddev->kobj, &md_redundancy_group);
544 if (mddev->sysfs_action)
545 sysfs_put(mddev->sysfs_action);
546 mddev->sysfs_action = NULL;
548 mutex_unlock(&mddev->open_mutex);
550 mutex_unlock(&mddev->reconfig_mutex);
552 md_wakeup_thread(mddev->thread);
555 static mdk_rdev_t * find_rdev_nr(mddev_t *mddev, int nr)
559 list_for_each_entry(rdev, &mddev->disks, same_set)
560 if (rdev->desc_nr == nr)
566 static mdk_rdev_t * find_rdev(mddev_t * mddev, dev_t dev)
570 list_for_each_entry(rdev, &mddev->disks, same_set)
571 if (rdev->bdev->bd_dev == dev)
577 static struct mdk_personality *find_pers(int level, char *clevel)
579 struct mdk_personality *pers;
580 list_for_each_entry(pers, &pers_list, list) {
581 if (level != LEVEL_NONE && pers->level == level)
583 if (strcmp(pers->name, clevel)==0)
589 /* return the offset of the super block in 512byte sectors */
590 static inline sector_t calc_dev_sboffset(struct block_device *bdev)
592 sector_t num_sectors = bdev->bd_inode->i_size / 512;
593 return MD_NEW_SIZE_SECTORS(num_sectors);
596 static int alloc_disk_sb(mdk_rdev_t * rdev)
601 rdev->sb_page = alloc_page(GFP_KERNEL);
602 if (!rdev->sb_page) {
603 printk(KERN_ALERT "md: out of memory.\n");
610 static void free_disk_sb(mdk_rdev_t * rdev)
613 put_page(rdev->sb_page);
615 rdev->sb_page = NULL;
622 static void super_written(struct bio *bio, int error)
624 mdk_rdev_t *rdev = bio->bi_private;
625 mddev_t *mddev = rdev->mddev;
627 if (error || !test_bit(BIO_UPTODATE, &bio->bi_flags)) {
628 printk("md: super_written gets error=%d, uptodate=%d\n",
629 error, test_bit(BIO_UPTODATE, &bio->bi_flags));
630 WARN_ON(test_bit(BIO_UPTODATE, &bio->bi_flags));
631 md_error(mddev, rdev);
634 if (atomic_dec_and_test(&mddev->pending_writes))
635 wake_up(&mddev->sb_wait);
639 static void super_written_barrier(struct bio *bio, int error)
641 struct bio *bio2 = bio->bi_private;
642 mdk_rdev_t *rdev = bio2->bi_private;
643 mddev_t *mddev = rdev->mddev;
645 if (!test_bit(BIO_UPTODATE, &bio->bi_flags) &&
646 error == -EOPNOTSUPP) {
648 /* barriers don't appear to be supported :-( */
649 set_bit(BarriersNotsupp, &rdev->flags);
650 mddev->barriers_work = 0;
651 spin_lock_irqsave(&mddev->write_lock, flags);
652 bio2->bi_next = mddev->biolist;
653 mddev->biolist = bio2;
654 spin_unlock_irqrestore(&mddev->write_lock, flags);
655 wake_up(&mddev->sb_wait);
659 bio->bi_private = rdev;
660 super_written(bio, error);
664 void md_super_write(mddev_t *mddev, mdk_rdev_t *rdev,
665 sector_t sector, int size, struct page *page)
667 /* write first size bytes of page to sector of rdev
668 * Increment mddev->pending_writes before returning
669 * and decrement it on completion, waking up sb_wait
670 * if zero is reached.
671 * If an error occurred, call md_error
673 * As we might need to resubmit the request if BIO_RW_BARRIER
674 * causes ENOTSUPP, we allocate a spare bio...
676 struct bio *bio = bio_alloc(GFP_NOIO, 1);
677 int rw = (1<<BIO_RW) | (1<<BIO_RW_SYNCIO) | (1<<BIO_RW_UNPLUG);
679 bio->bi_bdev = rdev->bdev;
680 bio->bi_sector = sector;
681 bio_add_page(bio, page, size, 0);
682 bio->bi_private = rdev;
683 bio->bi_end_io = super_written;
686 atomic_inc(&mddev->pending_writes);
687 if (!test_bit(BarriersNotsupp, &rdev->flags)) {
689 rw |= (1<<BIO_RW_BARRIER);
690 rbio = bio_clone(bio, GFP_NOIO);
691 rbio->bi_private = bio;
692 rbio->bi_end_io = super_written_barrier;
693 submit_bio(rw, rbio);
698 void md_super_wait(mddev_t *mddev)
700 /* wait for all superblock writes that were scheduled to complete.
701 * if any had to be retried (due to BARRIER problems), retry them
705 prepare_to_wait(&mddev->sb_wait, &wq, TASK_UNINTERRUPTIBLE);
706 if (atomic_read(&mddev->pending_writes)==0)
708 while (mddev->biolist) {
710 spin_lock_irq(&mddev->write_lock);
711 bio = mddev->biolist;
712 mddev->biolist = bio->bi_next ;
714 spin_unlock_irq(&mddev->write_lock);
715 submit_bio(bio->bi_rw, bio);
719 finish_wait(&mddev->sb_wait, &wq);
722 static void bi_complete(struct bio *bio, int error)
724 complete((struct completion*)bio->bi_private);
727 int sync_page_io(struct block_device *bdev, sector_t sector, int size,
728 struct page *page, int rw)
730 struct bio *bio = bio_alloc(GFP_NOIO, 1);
731 struct completion event;
734 rw |= (1 << BIO_RW_SYNCIO) | (1 << BIO_RW_UNPLUG);
737 bio->bi_sector = sector;
738 bio_add_page(bio, page, size, 0);
739 init_completion(&event);
740 bio->bi_private = &event;
741 bio->bi_end_io = bi_complete;
743 wait_for_completion(&event);
745 ret = test_bit(BIO_UPTODATE, &bio->bi_flags);
749 EXPORT_SYMBOL_GPL(sync_page_io);
751 static int read_disk_sb(mdk_rdev_t * rdev, int size)
753 char b[BDEVNAME_SIZE];
754 if (!rdev->sb_page) {
762 if (!sync_page_io(rdev->bdev, rdev->sb_start, size, rdev->sb_page, READ))
768 printk(KERN_WARNING "md: disabled device %s, could not read superblock.\n",
769 bdevname(rdev->bdev,b));
773 static int uuid_equal(mdp_super_t *sb1, mdp_super_t *sb2)
775 return sb1->set_uuid0 == sb2->set_uuid0 &&
776 sb1->set_uuid1 == sb2->set_uuid1 &&
777 sb1->set_uuid2 == sb2->set_uuid2 &&
778 sb1->set_uuid3 == sb2->set_uuid3;
781 static int sb_equal(mdp_super_t *sb1, mdp_super_t *sb2)
784 mdp_super_t *tmp1, *tmp2;
786 tmp1 = kmalloc(sizeof(*tmp1),GFP_KERNEL);
787 tmp2 = kmalloc(sizeof(*tmp2),GFP_KERNEL);
789 if (!tmp1 || !tmp2) {
791 printk(KERN_INFO "md.c sb_equal(): failed to allocate memory!\n");
799 * nr_disks is not constant
804 ret = (memcmp(tmp1, tmp2, MD_SB_GENERIC_CONSTANT_WORDS * 4) == 0);
812 static u32 md_csum_fold(u32 csum)
814 csum = (csum & 0xffff) + (csum >> 16);
815 return (csum & 0xffff) + (csum >> 16);
818 static unsigned int calc_sb_csum(mdp_super_t * sb)
821 u32 *sb32 = (u32*)sb;
823 unsigned int disk_csum, csum;
825 disk_csum = sb->sb_csum;
828 for (i = 0; i < MD_SB_BYTES/4 ; i++)
830 csum = (newcsum & 0xffffffff) + (newcsum>>32);
834 /* This used to use csum_partial, which was wrong for several
835 * reasons including that different results are returned on
836 * different architectures. It isn't critical that we get exactly
837 * the same return value as before (we always csum_fold before
838 * testing, and that removes any differences). However as we
839 * know that csum_partial always returned a 16bit value on
840 * alphas, do a fold to maximise conformity to previous behaviour.
842 sb->sb_csum = md_csum_fold(disk_csum);
844 sb->sb_csum = disk_csum;
851 * Handle superblock details.
852 * We want to be able to handle multiple superblock formats
853 * so we have a common interface to them all, and an array of
854 * different handlers.
855 * We rely on user-space to write the initial superblock, and support
856 * reading and updating of superblocks.
857 * Interface methods are:
858 * int load_super(mdk_rdev_t *dev, mdk_rdev_t *refdev, int minor_version)
859 * loads and validates a superblock on dev.
860 * if refdev != NULL, compare superblocks on both devices
862 * 0 - dev has a superblock that is compatible with refdev
863 * 1 - dev has a superblock that is compatible and newer than refdev
864 * so dev should be used as the refdev in future
865 * -EINVAL superblock incompatible or invalid
866 * -othererror e.g. -EIO
868 * int validate_super(mddev_t *mddev, mdk_rdev_t *dev)
869 * Verify that dev is acceptable into mddev.
870 * The first time, mddev->raid_disks will be 0, and data from
871 * dev should be merged in. Subsequent calls check that dev
872 * is new enough. Return 0 or -EINVAL
874 * void sync_super(mddev_t *mddev, mdk_rdev_t *dev)
875 * Update the superblock for rdev with data in mddev
876 * This does not write to disc.
882 struct module *owner;
883 int (*load_super)(mdk_rdev_t *rdev, mdk_rdev_t *refdev,
885 int (*validate_super)(mddev_t *mddev, mdk_rdev_t *rdev);
886 void (*sync_super)(mddev_t *mddev, mdk_rdev_t *rdev);
887 unsigned long long (*rdev_size_change)(mdk_rdev_t *rdev,
888 sector_t num_sectors);
892 * Check that the given mddev has no bitmap.
894 * This function is called from the run method of all personalities that do not
895 * support bitmaps. It prints an error message and returns non-zero if mddev
896 * has a bitmap. Otherwise, it returns 0.
899 int md_check_no_bitmap(mddev_t *mddev)
901 if (!mddev->bitmap_info.file && !mddev->bitmap_info.offset)
903 printk(KERN_ERR "%s: bitmaps are not supported for %s\n",
904 mdname(mddev), mddev->pers->name);
907 EXPORT_SYMBOL(md_check_no_bitmap);
910 * load_super for 0.90.0
912 static int super_90_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
914 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
919 * Calculate the position of the superblock (512byte sectors),
920 * it's at the end of the disk.
922 * It also happens to be a multiple of 4Kb.
924 rdev->sb_start = calc_dev_sboffset(rdev->bdev);
926 ret = read_disk_sb(rdev, MD_SB_BYTES);
931 bdevname(rdev->bdev, b);
932 sb = (mdp_super_t*)page_address(rdev->sb_page);
934 if (sb->md_magic != MD_SB_MAGIC) {
935 printk(KERN_ERR "md: invalid raid superblock magic on %s\n",
940 if (sb->major_version != 0 ||
941 sb->minor_version < 90 ||
942 sb->minor_version > 91) {
943 printk(KERN_WARNING "Bad version number %d.%d on %s\n",
944 sb->major_version, sb->minor_version,
949 if (sb->raid_disks <= 0)
952 if (md_csum_fold(calc_sb_csum(sb)) != md_csum_fold(sb->sb_csum)) {
953 printk(KERN_WARNING "md: invalid superblock checksum on %s\n",
958 rdev->preferred_minor = sb->md_minor;
959 rdev->data_offset = 0;
960 rdev->sb_size = MD_SB_BYTES;
962 if (sb->level == LEVEL_MULTIPATH)
965 rdev->desc_nr = sb->this_disk.number;
971 mdp_super_t *refsb = (mdp_super_t*)page_address(refdev->sb_page);
972 if (!uuid_equal(refsb, sb)) {
973 printk(KERN_WARNING "md: %s has different UUID to %s\n",
974 b, bdevname(refdev->bdev,b2));
977 if (!sb_equal(refsb, sb)) {
978 printk(KERN_WARNING "md: %s has same UUID"
979 " but different superblock to %s\n",
980 b, bdevname(refdev->bdev, b2));
984 ev2 = md_event(refsb);
990 rdev->sectors = rdev->sb_start;
992 if (rdev->sectors < sb->size * 2 && sb->level > 1)
993 /* "this cannot possibly happen" ... */
1001 * validate_super for 0.90.0
1003 static int super_90_validate(mddev_t *mddev, mdk_rdev_t *rdev)
1006 mdp_super_t *sb = (mdp_super_t *)page_address(rdev->sb_page);
1007 __u64 ev1 = md_event(sb);
1009 rdev->raid_disk = -1;
1010 clear_bit(Faulty, &rdev->flags);
1011 clear_bit(In_sync, &rdev->flags);
1012 clear_bit(WriteMostly, &rdev->flags);
1013 clear_bit(BarriersNotsupp, &rdev->flags);
1015 if (mddev->raid_disks == 0) {
1016 mddev->major_version = 0;
1017 mddev->minor_version = sb->minor_version;
1018 mddev->patch_version = sb->patch_version;
1019 mddev->external = 0;
1020 mddev->chunk_sectors = sb->chunk_size >> 9;
1021 mddev->ctime = sb->ctime;
1022 mddev->utime = sb->utime;
1023 mddev->level = sb->level;
1024 mddev->clevel[0] = 0;
1025 mddev->layout = sb->layout;
1026 mddev->raid_disks = sb->raid_disks;
1027 mddev->dev_sectors = sb->size * 2;
1028 mddev->events = ev1;
1029 mddev->bitmap_info.offset = 0;
1030 mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9;
1032 if (mddev->minor_version >= 91) {
1033 mddev->reshape_position = sb->reshape_position;
1034 mddev->delta_disks = sb->delta_disks;
1035 mddev->new_level = sb->new_level;
1036 mddev->new_layout = sb->new_layout;
1037 mddev->new_chunk_sectors = sb->new_chunk >> 9;
1039 mddev->reshape_position = MaxSector;
1040 mddev->delta_disks = 0;
1041 mddev->new_level = mddev->level;
1042 mddev->new_layout = mddev->layout;
1043 mddev->new_chunk_sectors = mddev->chunk_sectors;
1046 if (sb->state & (1<<MD_SB_CLEAN))
1047 mddev->recovery_cp = MaxSector;
1049 if (sb->events_hi == sb->cp_events_hi &&
1050 sb->events_lo == sb->cp_events_lo) {
1051 mddev->recovery_cp = sb->recovery_cp;
1053 mddev->recovery_cp = 0;
1056 memcpy(mddev->uuid+0, &sb->set_uuid0, 4);
1057 memcpy(mddev->uuid+4, &sb->set_uuid1, 4);
1058 memcpy(mddev->uuid+8, &sb->set_uuid2, 4);
1059 memcpy(mddev->uuid+12,&sb->set_uuid3, 4);
1061 mddev->max_disks = MD_SB_DISKS;
1063 if (sb->state & (1<<MD_SB_BITMAP_PRESENT) &&
1064 mddev->bitmap_info.file == NULL)
1065 mddev->bitmap_info.offset =
1066 mddev->bitmap_info.default_offset;
1068 } else if (mddev->pers == NULL) {
1069 /* Insist on good event counter while assembling */
1071 if (ev1 < mddev->events)
1073 } else if (mddev->bitmap) {
1074 /* if adding to array with a bitmap, then we can accept an
1075 * older device ... but not too old.
1077 if (ev1 < mddev->bitmap->events_cleared)
1080 if (ev1 < mddev->events)
1081 /* just a hot-add of a new device, leave raid_disk at -1 */
1085 if (mddev->level != LEVEL_MULTIPATH) {
1086 desc = sb->disks + rdev->desc_nr;
1088 if (desc->state & (1<<MD_DISK_FAULTY))
1089 set_bit(Faulty, &rdev->flags);
1090 else if (desc->state & (1<<MD_DISK_SYNC) /* &&
1091 desc->raid_disk < mddev->raid_disks */) {
1092 set_bit(In_sync, &rdev->flags);
1093 rdev->raid_disk = desc->raid_disk;
1094 } else if (desc->state & (1<<MD_DISK_ACTIVE)) {
1095 /* active but not in sync implies recovery up to
1096 * reshape position. We don't know exactly where
1097 * that is, so set to zero for now */
1098 if (mddev->minor_version >= 91) {
1099 rdev->recovery_offset = 0;
1100 rdev->raid_disk = desc->raid_disk;
1103 if (desc->state & (1<<MD_DISK_WRITEMOSTLY))
1104 set_bit(WriteMostly, &rdev->flags);
1105 } else /* MULTIPATH are always insync */
1106 set_bit(In_sync, &rdev->flags);
1111 * sync_super for 0.90.0
1113 static void super_90_sync(mddev_t *mddev, mdk_rdev_t *rdev)
1117 int next_spare = mddev->raid_disks;
1120 /* make rdev->sb match mddev data..
1123 * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
1124 * 3/ any empty disks < next_spare become removed
1126 * disks[0] gets initialised to REMOVED because
1127 * we cannot be sure from other fields if it has
1128 * been initialised or not.
1131 int active=0, working=0,failed=0,spare=0,nr_disks=0;
1133 rdev->sb_size = MD_SB_BYTES;
1135 sb = (mdp_super_t*)page_address(rdev->sb_page);
1137 memset(sb, 0, sizeof(*sb));
1139 sb->md_magic = MD_SB_MAGIC;
1140 sb->major_version = mddev->major_version;
1141 sb->patch_version = mddev->patch_version;
1142 sb->gvalid_words = 0; /* ignored */
1143 memcpy(&sb->set_uuid0, mddev->uuid+0, 4);
1144 memcpy(&sb->set_uuid1, mddev->uuid+4, 4);
1145 memcpy(&sb->set_uuid2, mddev->uuid+8, 4);
1146 memcpy(&sb->set_uuid3, mddev->uuid+12,4);
1148 sb->ctime = mddev->ctime;
1149 sb->level = mddev->level;
1150 sb->size = mddev->dev_sectors / 2;
1151 sb->raid_disks = mddev->raid_disks;
1152 sb->md_minor = mddev->md_minor;
1153 sb->not_persistent = 0;
1154 sb->utime = mddev->utime;
1156 sb->events_hi = (mddev->events>>32);
1157 sb->events_lo = (u32)mddev->events;
1159 if (mddev->reshape_position == MaxSector)
1160 sb->minor_version = 90;
1162 sb->minor_version = 91;
1163 sb->reshape_position = mddev->reshape_position;
1164 sb->new_level = mddev->new_level;
1165 sb->delta_disks = mddev->delta_disks;
1166 sb->new_layout = mddev->new_layout;
1167 sb->new_chunk = mddev->new_chunk_sectors << 9;
1169 mddev->minor_version = sb->minor_version;
1172 sb->recovery_cp = mddev->recovery_cp;
1173 sb->cp_events_hi = (mddev->events>>32);
1174 sb->cp_events_lo = (u32)mddev->events;
1175 if (mddev->recovery_cp == MaxSector)
1176 sb->state = (1<< MD_SB_CLEAN);
1178 sb->recovery_cp = 0;
1180 sb->layout = mddev->layout;
1181 sb->chunk_size = mddev->chunk_sectors << 9;
1183 if (mddev->bitmap && mddev->bitmap_info.file == NULL)
1184 sb->state |= (1<<MD_SB_BITMAP_PRESENT);
1186 sb->disks[0].state = (1<<MD_DISK_REMOVED);
1187 list_for_each_entry(rdev2, &mddev->disks, same_set) {
1190 int is_active = test_bit(In_sync, &rdev2->flags);
1192 if (rdev2->raid_disk >= 0 &&
1193 sb->minor_version >= 91)
1194 /* we have nowhere to store the recovery_offset,
1195 * but if it is not below the reshape_position,
1196 * we can piggy-back on that.
1199 if (rdev2->raid_disk < 0 ||
1200 test_bit(Faulty, &rdev2->flags))
1203 desc_nr = rdev2->raid_disk;
1205 desc_nr = next_spare++;
1206 rdev2->desc_nr = desc_nr;
1207 d = &sb->disks[rdev2->desc_nr];
1209 d->number = rdev2->desc_nr;
1210 d->major = MAJOR(rdev2->bdev->bd_dev);
1211 d->minor = MINOR(rdev2->bdev->bd_dev);
1213 d->raid_disk = rdev2->raid_disk;
1215 d->raid_disk = rdev2->desc_nr; /* compatibility */
1216 if (test_bit(Faulty, &rdev2->flags))
1217 d->state = (1<<MD_DISK_FAULTY);
1218 else if (is_active) {
1219 d->state = (1<<MD_DISK_ACTIVE);
1220 if (test_bit(In_sync, &rdev2->flags))
1221 d->state |= (1<<MD_DISK_SYNC);
1229 if (test_bit(WriteMostly, &rdev2->flags))
1230 d->state |= (1<<MD_DISK_WRITEMOSTLY);
1232 /* now set the "removed" and "faulty" bits on any missing devices */
1233 for (i=0 ; i < mddev->raid_disks ; i++) {
1234 mdp_disk_t *d = &sb->disks[i];
1235 if (d->state == 0 && d->number == 0) {
1238 d->state = (1<<MD_DISK_REMOVED);
1239 d->state |= (1<<MD_DISK_FAULTY);
1243 sb->nr_disks = nr_disks;
1244 sb->active_disks = active;
1245 sb->working_disks = working;
1246 sb->failed_disks = failed;
1247 sb->spare_disks = spare;
1249 sb->this_disk = sb->disks[rdev->desc_nr];
1250 sb->sb_csum = calc_sb_csum(sb);
1254 * rdev_size_change for 0.90.0
1256 static unsigned long long
1257 super_90_rdev_size_change(mdk_rdev_t *rdev, sector_t num_sectors)
1259 if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
1260 return 0; /* component must fit device */
1261 if (rdev->mddev->bitmap_info.offset)
1262 return 0; /* can't move bitmap */
1263 rdev->sb_start = calc_dev_sboffset(rdev->bdev);
1264 if (!num_sectors || num_sectors > rdev->sb_start)
1265 num_sectors = rdev->sb_start;
1266 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
1268 md_super_wait(rdev->mddev);
1269 return num_sectors / 2; /* kB for sysfs */
1274 * version 1 superblock
1277 static __le32 calc_sb_1_csum(struct mdp_superblock_1 * sb)
1281 unsigned long long newcsum;
1282 int size = 256 + le32_to_cpu(sb->max_dev)*2;
1283 __le32 *isuper = (__le32*)sb;
1286 disk_csum = sb->sb_csum;
1289 for (i=0; size>=4; size -= 4 )
1290 newcsum += le32_to_cpu(*isuper++);
1293 newcsum += le16_to_cpu(*(__le16*) isuper);
1295 csum = (newcsum & 0xffffffff) + (newcsum >> 32);
1296 sb->sb_csum = disk_csum;
1297 return cpu_to_le32(csum);
1300 static int super_1_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
1302 struct mdp_superblock_1 *sb;
1305 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
1309 * Calculate the position of the superblock in 512byte sectors.
1310 * It is always aligned to a 4K boundary and
1311 * depeding on minor_version, it can be:
1312 * 0: At least 8K, but less than 12K, from end of device
1313 * 1: At start of device
1314 * 2: 4K from start of device.
1316 switch(minor_version) {
1318 sb_start = rdev->bdev->bd_inode->i_size >> 9;
1320 sb_start &= ~(sector_t)(4*2-1);
1331 rdev->sb_start = sb_start;
1333 /* superblock is rarely larger than 1K, but it can be larger,
1334 * and it is safe to read 4k, so we do that
1336 ret = read_disk_sb(rdev, 4096);
1337 if (ret) return ret;
1340 sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1342 if (sb->magic != cpu_to_le32(MD_SB_MAGIC) ||
1343 sb->major_version != cpu_to_le32(1) ||
1344 le32_to_cpu(sb->max_dev) > (4096-256)/2 ||
1345 le64_to_cpu(sb->super_offset) != rdev->sb_start ||
1346 (le32_to_cpu(sb->feature_map) & ~MD_FEATURE_ALL) != 0)
1349 if (calc_sb_1_csum(sb) != sb->sb_csum) {
1350 printk("md: invalid superblock checksum on %s\n",
1351 bdevname(rdev->bdev,b));
1354 if (le64_to_cpu(sb->data_size) < 10) {
1355 printk("md: data_size too small on %s\n",
1356 bdevname(rdev->bdev,b));
1360 rdev->preferred_minor = 0xffff;
1361 rdev->data_offset = le64_to_cpu(sb->data_offset);
1362 atomic_set(&rdev->corrected_errors, le32_to_cpu(sb->cnt_corrected_read));
1364 rdev->sb_size = le32_to_cpu(sb->max_dev) * 2 + 256;
1365 bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
1366 if (rdev->sb_size & bmask)
1367 rdev->sb_size = (rdev->sb_size | bmask) + 1;
1370 && rdev->data_offset < sb_start + (rdev->sb_size/512))
1373 if (sb->level == cpu_to_le32(LEVEL_MULTIPATH))
1376 rdev->desc_nr = le32_to_cpu(sb->dev_number);
1382 struct mdp_superblock_1 *refsb =
1383 (struct mdp_superblock_1*)page_address(refdev->sb_page);
1385 if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 ||
1386 sb->level != refsb->level ||
1387 sb->layout != refsb->layout ||
1388 sb->chunksize != refsb->chunksize) {
1389 printk(KERN_WARNING "md: %s has strangely different"
1390 " superblock to %s\n",
1391 bdevname(rdev->bdev,b),
1392 bdevname(refdev->bdev,b2));
1395 ev1 = le64_to_cpu(sb->events);
1396 ev2 = le64_to_cpu(refsb->events);
1404 rdev->sectors = (rdev->bdev->bd_inode->i_size >> 9) -
1405 le64_to_cpu(sb->data_offset);
1407 rdev->sectors = rdev->sb_start;
1408 if (rdev->sectors < le64_to_cpu(sb->data_size))
1410 rdev->sectors = le64_to_cpu(sb->data_size);
1411 if (le64_to_cpu(sb->size) > rdev->sectors)
1416 static int super_1_validate(mddev_t *mddev, mdk_rdev_t *rdev)
1418 struct mdp_superblock_1 *sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1419 __u64 ev1 = le64_to_cpu(sb->events);
1421 rdev->raid_disk = -1;
1422 clear_bit(Faulty, &rdev->flags);
1423 clear_bit(In_sync, &rdev->flags);
1424 clear_bit(WriteMostly, &rdev->flags);
1425 clear_bit(BarriersNotsupp, &rdev->flags);
1427 if (mddev->raid_disks == 0) {
1428 mddev->major_version = 1;
1429 mddev->patch_version = 0;
1430 mddev->external = 0;
1431 mddev->chunk_sectors = le32_to_cpu(sb->chunksize);
1432 mddev->ctime = le64_to_cpu(sb->ctime) & ((1ULL << 32)-1);
1433 mddev->utime = le64_to_cpu(sb->utime) & ((1ULL << 32)-1);
1434 mddev->level = le32_to_cpu(sb->level);
1435 mddev->clevel[0] = 0;
1436 mddev->layout = le32_to_cpu(sb->layout);
1437 mddev->raid_disks = le32_to_cpu(sb->raid_disks);
1438 mddev->dev_sectors = le64_to_cpu(sb->size);
1439 mddev->events = ev1;
1440 mddev->bitmap_info.offset = 0;
1441 mddev->bitmap_info.default_offset = 1024 >> 9;
1443 mddev->recovery_cp = le64_to_cpu(sb->resync_offset);
1444 memcpy(mddev->uuid, sb->set_uuid, 16);
1446 mddev->max_disks = (4096-256)/2;
1448 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) &&
1449 mddev->bitmap_info.file == NULL )
1450 mddev->bitmap_info.offset =
1451 (__s32)le32_to_cpu(sb->bitmap_offset);
1453 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) {
1454 mddev->reshape_position = le64_to_cpu(sb->reshape_position);
1455 mddev->delta_disks = le32_to_cpu(sb->delta_disks);
1456 mddev->new_level = le32_to_cpu(sb->new_level);
1457 mddev->new_layout = le32_to_cpu(sb->new_layout);
1458 mddev->new_chunk_sectors = le32_to_cpu(sb->new_chunk);
1460 mddev->reshape_position = MaxSector;
1461 mddev->delta_disks = 0;
1462 mddev->new_level = mddev->level;
1463 mddev->new_layout = mddev->layout;
1464 mddev->new_chunk_sectors = mddev->chunk_sectors;
1467 } else if (mddev->pers == NULL) {
1468 /* Insist of good event counter while assembling */
1470 if (ev1 < mddev->events)
1472 } else if (mddev->bitmap) {
1473 /* If adding to array with a bitmap, then we can accept an
1474 * older device, but not too old.
1476 if (ev1 < mddev->bitmap->events_cleared)
1479 if (ev1 < mddev->events)
1480 /* just a hot-add of a new device, leave raid_disk at -1 */
1483 if (mddev->level != LEVEL_MULTIPATH) {
1485 if (rdev->desc_nr < 0 ||
1486 rdev->desc_nr >= le32_to_cpu(sb->max_dev)) {
1490 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
1492 case 0xffff: /* spare */
1494 case 0xfffe: /* faulty */
1495 set_bit(Faulty, &rdev->flags);
1498 if ((le32_to_cpu(sb->feature_map) &
1499 MD_FEATURE_RECOVERY_OFFSET))
1500 rdev->recovery_offset = le64_to_cpu(sb->recovery_offset);
1502 set_bit(In_sync, &rdev->flags);
1503 rdev->raid_disk = role;
1506 if (sb->devflags & WriteMostly1)
1507 set_bit(WriteMostly, &rdev->flags);
1508 } else /* MULTIPATH are always insync */
1509 set_bit(In_sync, &rdev->flags);
1514 static void super_1_sync(mddev_t *mddev, mdk_rdev_t *rdev)
1516 struct mdp_superblock_1 *sb;
1519 /* make rdev->sb match mddev and rdev data. */
1521 sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1523 sb->feature_map = 0;
1525 sb->recovery_offset = cpu_to_le64(0);
1526 memset(sb->pad1, 0, sizeof(sb->pad1));
1527 memset(sb->pad2, 0, sizeof(sb->pad2));
1528 memset(sb->pad3, 0, sizeof(sb->pad3));
1530 sb->utime = cpu_to_le64((__u64)mddev->utime);
1531 sb->events = cpu_to_le64(mddev->events);
1533 sb->resync_offset = cpu_to_le64(mddev->recovery_cp);
1535 sb->resync_offset = cpu_to_le64(0);
1537 sb->cnt_corrected_read = cpu_to_le32(atomic_read(&rdev->corrected_errors));
1539 sb->raid_disks = cpu_to_le32(mddev->raid_disks);
1540 sb->size = cpu_to_le64(mddev->dev_sectors);
1541 sb->chunksize = cpu_to_le32(mddev->chunk_sectors);
1542 sb->level = cpu_to_le32(mddev->level);
1543 sb->layout = cpu_to_le32(mddev->layout);
1545 if (mddev->bitmap && mddev->bitmap_info.file == NULL) {
1546 sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_info.offset);
1547 sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET);
1550 if (rdev->raid_disk >= 0 &&
1551 !test_bit(In_sync, &rdev->flags)) {
1553 cpu_to_le32(MD_FEATURE_RECOVERY_OFFSET);
1554 sb->recovery_offset =
1555 cpu_to_le64(rdev->recovery_offset);
1558 if (mddev->reshape_position != MaxSector) {
1559 sb->feature_map |= cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE);
1560 sb->reshape_position = cpu_to_le64(mddev->reshape_position);
1561 sb->new_layout = cpu_to_le32(mddev->new_layout);
1562 sb->delta_disks = cpu_to_le32(mddev->delta_disks);
1563 sb->new_level = cpu_to_le32(mddev->new_level);
1564 sb->new_chunk = cpu_to_le32(mddev->new_chunk_sectors);
1568 list_for_each_entry(rdev2, &mddev->disks, same_set)
1569 if (rdev2->desc_nr+1 > max_dev)
1570 max_dev = rdev2->desc_nr+1;
1572 if (max_dev > le32_to_cpu(sb->max_dev)) {
1574 sb->max_dev = cpu_to_le32(max_dev);
1575 rdev->sb_size = max_dev * 2 + 256;
1576 bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
1577 if (rdev->sb_size & bmask)
1578 rdev->sb_size = (rdev->sb_size | bmask) + 1;
1580 for (i=0; i<max_dev;i++)
1581 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1583 list_for_each_entry(rdev2, &mddev->disks, same_set) {
1585 if (test_bit(Faulty, &rdev2->flags))
1586 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1587 else if (test_bit(In_sync, &rdev2->flags))
1588 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1589 else if (rdev2->raid_disk >= 0)
1590 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1592 sb->dev_roles[i] = cpu_to_le16(0xffff);
1595 sb->sb_csum = calc_sb_1_csum(sb);
1598 static unsigned long long
1599 super_1_rdev_size_change(mdk_rdev_t *rdev, sector_t num_sectors)
1601 struct mdp_superblock_1 *sb;
1602 sector_t max_sectors;
1603 if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
1604 return 0; /* component must fit device */
1605 if (rdev->sb_start < rdev->data_offset) {
1606 /* minor versions 1 and 2; superblock before data */
1607 max_sectors = rdev->bdev->bd_inode->i_size >> 9;
1608 max_sectors -= rdev->data_offset;
1609 if (!num_sectors || num_sectors > max_sectors)
1610 num_sectors = max_sectors;
1611 } else if (rdev->mddev->bitmap_info.offset) {
1612 /* minor version 0 with bitmap we can't move */
1615 /* minor version 0; superblock after data */
1617 sb_start = (rdev->bdev->bd_inode->i_size >> 9) - 8*2;
1618 sb_start &= ~(sector_t)(4*2 - 1);
1619 max_sectors = rdev->sectors + sb_start - rdev->sb_start;
1620 if (!num_sectors || num_sectors > max_sectors)
1621 num_sectors = max_sectors;
1622 rdev->sb_start = sb_start;
1624 sb = (struct mdp_superblock_1 *) page_address(rdev->sb_page);
1625 sb->data_size = cpu_to_le64(num_sectors);
1626 sb->super_offset = rdev->sb_start;
1627 sb->sb_csum = calc_sb_1_csum(sb);
1628 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
1630 md_super_wait(rdev->mddev);
1631 return num_sectors / 2; /* kB for sysfs */
1634 static struct super_type super_types[] = {
1637 .owner = THIS_MODULE,
1638 .load_super = super_90_load,
1639 .validate_super = super_90_validate,
1640 .sync_super = super_90_sync,
1641 .rdev_size_change = super_90_rdev_size_change,
1645 .owner = THIS_MODULE,
1646 .load_super = super_1_load,
1647 .validate_super = super_1_validate,
1648 .sync_super = super_1_sync,
1649 .rdev_size_change = super_1_rdev_size_change,
1653 static int match_mddev_units(mddev_t *mddev1, mddev_t *mddev2)
1655 mdk_rdev_t *rdev, *rdev2;
1658 rdev_for_each_rcu(rdev, mddev1)
1659 rdev_for_each_rcu(rdev2, mddev2)
1660 if (rdev->bdev->bd_contains ==
1661 rdev2->bdev->bd_contains) {
1669 static LIST_HEAD(pending_raid_disks);
1672 * Try to register data integrity profile for an mddev
1674 * This is called when an array is started and after a disk has been kicked
1675 * from the array. It only succeeds if all working and active component devices
1676 * are integrity capable with matching profiles.
1678 int md_integrity_register(mddev_t *mddev)
1680 mdk_rdev_t *rdev, *reference = NULL;
1682 if (list_empty(&mddev->disks))
1683 return 0; /* nothing to do */
1684 if (blk_get_integrity(mddev->gendisk))
1685 return 0; /* already registered */
1686 list_for_each_entry(rdev, &mddev->disks, same_set) {
1687 /* skip spares and non-functional disks */
1688 if (test_bit(Faulty, &rdev->flags))
1690 if (rdev->raid_disk < 0)
1693 * If at least one rdev is not integrity capable, we can not
1694 * enable data integrity for the md device.
1696 if (!bdev_get_integrity(rdev->bdev))
1699 /* Use the first rdev as the reference */
1703 /* does this rdev's profile match the reference profile? */
1704 if (blk_integrity_compare(reference->bdev->bd_disk,
1705 rdev->bdev->bd_disk) < 0)
1709 * All component devices are integrity capable and have matching
1710 * profiles, register the common profile for the md device.
1712 if (blk_integrity_register(mddev->gendisk,
1713 bdev_get_integrity(reference->bdev)) != 0) {
1714 printk(KERN_ERR "md: failed to register integrity for %s\n",
1718 printk(KERN_NOTICE "md: data integrity on %s enabled\n",
1722 EXPORT_SYMBOL(md_integrity_register);
1724 /* Disable data integrity if non-capable/non-matching disk is being added */
1725 void md_integrity_add_rdev(mdk_rdev_t *rdev, mddev_t *mddev)
1727 struct blk_integrity *bi_rdev = bdev_get_integrity(rdev->bdev);
1728 struct blk_integrity *bi_mddev = blk_get_integrity(mddev->gendisk);
1730 if (!bi_mddev) /* nothing to do */
1732 if (rdev->raid_disk < 0) /* skip spares */
1734 if (bi_rdev && blk_integrity_compare(mddev->gendisk,
1735 rdev->bdev->bd_disk) >= 0)
1737 printk(KERN_NOTICE "disabling data integrity on %s\n", mdname(mddev));
1738 blk_integrity_unregister(mddev->gendisk);
1740 EXPORT_SYMBOL(md_integrity_add_rdev);
1742 static int bind_rdev_to_array(mdk_rdev_t * rdev, mddev_t * mddev)
1744 char b[BDEVNAME_SIZE];
1754 /* prevent duplicates */
1755 if (find_rdev(mddev, rdev->bdev->bd_dev))
1758 /* make sure rdev->sectors exceeds mddev->dev_sectors */
1759 if (rdev->sectors && (mddev->dev_sectors == 0 ||
1760 rdev->sectors < mddev->dev_sectors)) {
1762 /* Cannot change size, so fail
1763 * If mddev->level <= 0, then we don't care
1764 * about aligning sizes (e.g. linear)
1766 if (mddev->level > 0)
1769 mddev->dev_sectors = rdev->sectors;
1772 /* Verify rdev->desc_nr is unique.
1773 * If it is -1, assign a free number, else
1774 * check number is not in use
1776 if (rdev->desc_nr < 0) {
1778 if (mddev->pers) choice = mddev->raid_disks;
1779 while (find_rdev_nr(mddev, choice))
1781 rdev->desc_nr = choice;
1783 if (find_rdev_nr(mddev, rdev->desc_nr))
1786 if (mddev->max_disks && rdev->desc_nr >= mddev->max_disks) {
1787 printk(KERN_WARNING "md: %s: array is limited to %d devices\n",
1788 mdname(mddev), mddev->max_disks);
1791 bdevname(rdev->bdev,b);
1792 while ( (s=strchr(b, '/')) != NULL)
1795 rdev->mddev = mddev;
1796 printk(KERN_INFO "md: bind<%s>\n", b);
1798 if ((err = kobject_add(&rdev->kobj, &mddev->kobj, "dev-%s", b)))
1801 ko = &part_to_dev(rdev->bdev->bd_part)->kobj;
1802 if ((err = sysfs_create_link(&rdev->kobj, ko, "block"))) {
1803 kobject_del(&rdev->kobj);
1806 rdev->sysfs_state = sysfs_get_dirent(rdev->kobj.sd, "state");
1808 list_add_rcu(&rdev->same_set, &mddev->disks);
1809 bd_claim_by_disk(rdev->bdev, rdev->bdev->bd_holder, mddev->gendisk);
1811 /* May as well allow recovery to be retried once */
1812 mddev->recovery_disabled = 0;
1817 printk(KERN_WARNING "md: failed to register dev-%s for %s\n",
1822 static void md_delayed_delete(struct work_struct *ws)
1824 mdk_rdev_t *rdev = container_of(ws, mdk_rdev_t, del_work);
1825 kobject_del(&rdev->kobj);
1826 kobject_put(&rdev->kobj);
1829 static void unbind_rdev_from_array(mdk_rdev_t * rdev)
1831 char b[BDEVNAME_SIZE];
1836 bd_release_from_disk(rdev->bdev, rdev->mddev->gendisk);
1837 list_del_rcu(&rdev->same_set);
1838 printk(KERN_INFO "md: unbind<%s>\n", bdevname(rdev->bdev,b));
1840 sysfs_remove_link(&rdev->kobj, "block");
1841 sysfs_put(rdev->sysfs_state);
1842 rdev->sysfs_state = NULL;
1843 /* We need to delay this, otherwise we can deadlock when
1844 * writing to 'remove' to "dev/state". We also need
1845 * to delay it due to rcu usage.
1848 INIT_WORK(&rdev->del_work, md_delayed_delete);
1849 kobject_get(&rdev->kobj);
1850 schedule_work(&rdev->del_work);
1854 * prevent the device from being mounted, repartitioned or
1855 * otherwise reused by a RAID array (or any other kernel
1856 * subsystem), by bd_claiming the device.
1858 static int lock_rdev(mdk_rdev_t *rdev, dev_t dev, int shared)
1861 struct block_device *bdev;
1862 char b[BDEVNAME_SIZE];
1864 bdev = open_by_devnum(dev, FMODE_READ|FMODE_WRITE);
1866 printk(KERN_ERR "md: could not open %s.\n",
1867 __bdevname(dev, b));
1868 return PTR_ERR(bdev);
1870 err = bd_claim(bdev, shared ? (mdk_rdev_t *)lock_rdev : rdev);
1872 printk(KERN_ERR "md: could not bd_claim %s.\n",
1874 blkdev_put(bdev, FMODE_READ|FMODE_WRITE);
1878 set_bit(AllReserved, &rdev->flags);
1883 static void unlock_rdev(mdk_rdev_t *rdev)
1885 struct block_device *bdev = rdev->bdev;
1890 blkdev_put(bdev, FMODE_READ|FMODE_WRITE);
1893 void md_autodetect_dev(dev_t dev);
1895 static void export_rdev(mdk_rdev_t * rdev)
1897 char b[BDEVNAME_SIZE];
1898 printk(KERN_INFO "md: export_rdev(%s)\n",
1899 bdevname(rdev->bdev,b));
1904 if (test_bit(AutoDetected, &rdev->flags))
1905 md_autodetect_dev(rdev->bdev->bd_dev);
1908 kobject_put(&rdev->kobj);
1911 static void kick_rdev_from_array(mdk_rdev_t * rdev)
1913 unbind_rdev_from_array(rdev);
1917 static void export_array(mddev_t *mddev)
1919 mdk_rdev_t *rdev, *tmp;
1921 rdev_for_each(rdev, tmp, mddev) {
1926 kick_rdev_from_array(rdev);
1928 if (!list_empty(&mddev->disks))
1930 mddev->raid_disks = 0;
1931 mddev->major_version = 0;
1934 static void print_desc(mdp_disk_t *desc)
1936 printk(" DISK<N:%d,(%d,%d),R:%d,S:%d>\n", desc->number,
1937 desc->major,desc->minor,desc->raid_disk,desc->state);
1940 static void print_sb_90(mdp_super_t *sb)
1945 "md: SB: (V:%d.%d.%d) ID:<%08x.%08x.%08x.%08x> CT:%08x\n",
1946 sb->major_version, sb->minor_version, sb->patch_version,
1947 sb->set_uuid0, sb->set_uuid1, sb->set_uuid2, sb->set_uuid3,
1949 printk(KERN_INFO "md: L%d S%08d ND:%d RD:%d md%d LO:%d CS:%d\n",
1950 sb->level, sb->size, sb->nr_disks, sb->raid_disks,
1951 sb->md_minor, sb->layout, sb->chunk_size);
1952 printk(KERN_INFO "md: UT:%08x ST:%d AD:%d WD:%d"
1953 " FD:%d SD:%d CSUM:%08x E:%08lx\n",
1954 sb->utime, sb->state, sb->active_disks, sb->working_disks,
1955 sb->failed_disks, sb->spare_disks,
1956 sb->sb_csum, (unsigned long)sb->events_lo);
1959 for (i = 0; i < MD_SB_DISKS; i++) {
1962 desc = sb->disks + i;
1963 if (desc->number || desc->major || desc->minor ||
1964 desc->raid_disk || (desc->state && (desc->state != 4))) {
1965 printk(" D %2d: ", i);
1969 printk(KERN_INFO "md: THIS: ");
1970 print_desc(&sb->this_disk);
1973 static void print_sb_1(struct mdp_superblock_1 *sb)
1977 uuid = sb->set_uuid;
1979 "md: SB: (V:%u) (F:0x%08x) Array-ID:<%pU>\n"
1980 "md: Name: \"%s\" CT:%llu\n",
1981 le32_to_cpu(sb->major_version),
1982 le32_to_cpu(sb->feature_map),
1985 (unsigned long long)le64_to_cpu(sb->ctime)
1986 & MD_SUPERBLOCK_1_TIME_SEC_MASK);
1988 uuid = sb->device_uuid;
1990 "md: L%u SZ%llu RD:%u LO:%u CS:%u DO:%llu DS:%llu SO:%llu"
1992 "md: Dev:%08x UUID: %pU\n"
1993 "md: (F:0x%08x) UT:%llu Events:%llu ResyncOffset:%llu CSUM:0x%08x\n"
1994 "md: (MaxDev:%u) \n",
1995 le32_to_cpu(sb->level),
1996 (unsigned long long)le64_to_cpu(sb->size),
1997 le32_to_cpu(sb->raid_disks),
1998 le32_to_cpu(sb->layout),
1999 le32_to_cpu(sb->chunksize),
2000 (unsigned long long)le64_to_cpu(sb->data_offset),
2001 (unsigned long long)le64_to_cpu(sb->data_size),
2002 (unsigned long long)le64_to_cpu(sb->super_offset),
2003 (unsigned long long)le64_to_cpu(sb->recovery_offset),
2004 le32_to_cpu(sb->dev_number),
2007 (unsigned long long)le64_to_cpu(sb->utime) & MD_SUPERBLOCK_1_TIME_SEC_MASK,
2008 (unsigned long long)le64_to_cpu(sb->events),
2009 (unsigned long long)le64_to_cpu(sb->resync_offset),
2010 le32_to_cpu(sb->sb_csum),
2011 le32_to_cpu(sb->max_dev)
2015 static void print_rdev(mdk_rdev_t *rdev, int major_version)
2017 char b[BDEVNAME_SIZE];
2018 printk(KERN_INFO "md: rdev %s, Sect:%08llu F:%d S:%d DN:%u\n",
2019 bdevname(rdev->bdev, b), (unsigned long long)rdev->sectors,
2020 test_bit(Faulty, &rdev->flags), test_bit(In_sync, &rdev->flags),
2022 if (rdev->sb_loaded) {
2023 printk(KERN_INFO "md: rdev superblock (MJ:%d):\n", major_version);
2024 switch (major_version) {
2026 print_sb_90((mdp_super_t*)page_address(rdev->sb_page));
2029 print_sb_1((struct mdp_superblock_1 *)page_address(rdev->sb_page));
2033 printk(KERN_INFO "md: no rdev superblock!\n");
2036 static void md_print_devices(void)
2038 struct list_head *tmp;
2041 char b[BDEVNAME_SIZE];
2044 printk("md: **********************************\n");
2045 printk("md: * <COMPLETE RAID STATE PRINTOUT> *\n");
2046 printk("md: **********************************\n");
2047 for_each_mddev(mddev, tmp) {
2050 bitmap_print_sb(mddev->bitmap);
2052 printk("%s: ", mdname(mddev));
2053 list_for_each_entry(rdev, &mddev->disks, same_set)
2054 printk("<%s>", bdevname(rdev->bdev,b));
2057 list_for_each_entry(rdev, &mddev->disks, same_set)
2058 print_rdev(rdev, mddev->major_version);
2060 printk("md: **********************************\n");
2065 static void sync_sbs(mddev_t * mddev, int nospares)
2067 /* Update each superblock (in-memory image), but
2068 * if we are allowed to, skip spares which already
2069 * have the right event counter, or have one earlier
2070 * (which would mean they aren't being marked as dirty
2071 * with the rest of the array)
2075 /* First make sure individual recovery_offsets are correct */
2076 list_for_each_entry(rdev, &mddev->disks, same_set) {
2077 if (rdev->raid_disk >= 0 &&
2078 !test_bit(In_sync, &rdev->flags) &&
2079 mddev->curr_resync_completed > rdev->recovery_offset)
2080 rdev->recovery_offset = mddev->curr_resync_completed;
2083 list_for_each_entry(rdev, &mddev->disks, same_set) {
2084 if (rdev->sb_events == mddev->events ||
2086 rdev->raid_disk < 0 &&
2087 (rdev->sb_events&1)==0 &&
2088 rdev->sb_events+1 == mddev->events)) {
2089 /* Don't update this superblock */
2090 rdev->sb_loaded = 2;
2092 super_types[mddev->major_version].
2093 sync_super(mddev, rdev);
2094 rdev->sb_loaded = 1;
2099 static void md_update_sb(mddev_t * mddev, int force_change)
2105 mddev->utime = get_seconds();
2106 if (mddev->external)
2109 spin_lock_irq(&mddev->write_lock);
2111 set_bit(MD_CHANGE_PENDING, &mddev->flags);
2112 if (test_and_clear_bit(MD_CHANGE_DEVS, &mddev->flags))
2114 if (test_and_clear_bit(MD_CHANGE_CLEAN, &mddev->flags))
2115 /* just a clean<-> dirty transition, possibly leave spares alone,
2116 * though if events isn't the right even/odd, we will have to do
2122 if (mddev->degraded)
2123 /* If the array is degraded, then skipping spares is both
2124 * dangerous and fairly pointless.
2125 * Dangerous because a device that was removed from the array
2126 * might have a event_count that still looks up-to-date,
2127 * so it can be re-added without a resync.
2128 * Pointless because if there are any spares to skip,
2129 * then a recovery will happen and soon that array won't
2130 * be degraded any more and the spare can go back to sleep then.
2134 sync_req = mddev->in_sync;
2136 /* If this is just a dirty<->clean transition, and the array is clean
2137 * and 'events' is odd, we can roll back to the previous clean state */
2139 && (mddev->in_sync && mddev->recovery_cp == MaxSector)
2140 && (mddev->events & 1)
2141 && mddev->events != 1)
2144 /* otherwise we have to go forward and ... */
2146 if (!mddev->in_sync || mddev->recovery_cp != MaxSector) { /* not clean */
2147 /* .. if the array isn't clean, an 'even' event must also go
2149 if ((mddev->events&1)==0)
2152 /* otherwise an 'odd' event must go to spares */
2153 if ((mddev->events&1))
2158 if (!mddev->events) {
2160 * oops, this 64-bit counter should never wrap.
2161 * Either we are in around ~1 trillion A.C., assuming
2162 * 1 reboot per second, or we have a bug:
2169 * do not write anything to disk if using
2170 * nonpersistent superblocks
2172 if (!mddev->persistent) {
2173 if (!mddev->external)
2174 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
2176 spin_unlock_irq(&mddev->write_lock);
2177 wake_up(&mddev->sb_wait);
2180 sync_sbs(mddev, nospares);
2181 spin_unlock_irq(&mddev->write_lock);
2184 "md: updating %s RAID superblock on device (in sync %d)\n",
2185 mdname(mddev),mddev->in_sync);
2187 bitmap_update_sb(mddev->bitmap);
2188 list_for_each_entry(rdev, &mddev->disks, same_set) {
2189 char b[BDEVNAME_SIZE];
2190 dprintk(KERN_INFO "md: ");
2191 if (rdev->sb_loaded != 1)
2192 continue; /* no noise on spare devices */
2193 if (test_bit(Faulty, &rdev->flags))
2194 dprintk("(skipping faulty ");
2196 dprintk("%s ", bdevname(rdev->bdev,b));
2197 if (!test_bit(Faulty, &rdev->flags)) {
2198 md_super_write(mddev,rdev,
2199 rdev->sb_start, rdev->sb_size,
2201 dprintk(KERN_INFO "(write) %s's sb offset: %llu\n",
2202 bdevname(rdev->bdev,b),
2203 (unsigned long long)rdev->sb_start);
2204 rdev->sb_events = mddev->events;
2208 if (mddev->level == LEVEL_MULTIPATH)
2209 /* only need to write one superblock... */
2212 md_super_wait(mddev);
2213 /* if there was a failure, MD_CHANGE_DEVS was set, and we re-write super */
2215 spin_lock_irq(&mddev->write_lock);
2216 if (mddev->in_sync != sync_req ||
2217 test_bit(MD_CHANGE_DEVS, &mddev->flags)) {
2218 /* have to write it out again */
2219 spin_unlock_irq(&mddev->write_lock);
2222 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
2223 spin_unlock_irq(&mddev->write_lock);
2224 wake_up(&mddev->sb_wait);
2225 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
2226 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
2230 /* words written to sysfs files may, or may not, be \n terminated.
2231 * We want to accept with case. For this we use cmd_match.
2233 static int cmd_match(const char *cmd, const char *str)
2235 /* See if cmd, written into a sysfs file, matches
2236 * str. They must either be the same, or cmd can
2237 * have a trailing newline
2239 while (*cmd && *str && *cmd == *str) {
2250 struct rdev_sysfs_entry {
2251 struct attribute attr;
2252 ssize_t (*show)(mdk_rdev_t *, char *);
2253 ssize_t (*store)(mdk_rdev_t *, const char *, size_t);
2257 state_show(mdk_rdev_t *rdev, char *page)
2262 if (test_bit(Faulty, &rdev->flags)) {
2263 len+= sprintf(page+len, "%sfaulty",sep);
2266 if (test_bit(In_sync, &rdev->flags)) {
2267 len += sprintf(page+len, "%sin_sync",sep);
2270 if (test_bit(WriteMostly, &rdev->flags)) {
2271 len += sprintf(page+len, "%swrite_mostly",sep);
2274 if (test_bit(Blocked, &rdev->flags)) {
2275 len += sprintf(page+len, "%sblocked", sep);
2278 if (!test_bit(Faulty, &rdev->flags) &&
2279 !test_bit(In_sync, &rdev->flags)) {
2280 len += sprintf(page+len, "%sspare", sep);
2283 return len+sprintf(page+len, "\n");
2287 state_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2290 * faulty - simulates and error
2291 * remove - disconnects the device
2292 * writemostly - sets write_mostly
2293 * -writemostly - clears write_mostly
2294 * blocked - sets the Blocked flag
2295 * -blocked - clears the Blocked flag
2296 * insync - sets Insync providing device isn't active
2299 if (cmd_match(buf, "faulty") && rdev->mddev->pers) {
2300 md_error(rdev->mddev, rdev);
2302 } else if (cmd_match(buf, "remove")) {
2303 if (rdev->raid_disk >= 0)
2306 mddev_t *mddev = rdev->mddev;
2307 kick_rdev_from_array(rdev);
2309 md_update_sb(mddev, 1);
2310 md_new_event(mddev);
2313 } else if (cmd_match(buf, "writemostly")) {
2314 set_bit(WriteMostly, &rdev->flags);
2316 } else if (cmd_match(buf, "-writemostly")) {
2317 clear_bit(WriteMostly, &rdev->flags);
2319 } else if (cmd_match(buf, "blocked")) {
2320 set_bit(Blocked, &rdev->flags);
2322 } else if (cmd_match(buf, "-blocked")) {
2323 clear_bit(Blocked, &rdev->flags);
2324 wake_up(&rdev->blocked_wait);
2325 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2326 md_wakeup_thread(rdev->mddev->thread);
2329 } else if (cmd_match(buf, "insync") && rdev->raid_disk == -1) {
2330 set_bit(In_sync, &rdev->flags);
2333 if (!err && rdev->sysfs_state)
2334 sysfs_notify_dirent(rdev->sysfs_state);
2335 return err ? err : len;
2337 static struct rdev_sysfs_entry rdev_state =
2338 __ATTR(state, S_IRUGO|S_IWUSR, state_show, state_store);
2341 errors_show(mdk_rdev_t *rdev, char *page)
2343 return sprintf(page, "%d\n", atomic_read(&rdev->corrected_errors));
2347 errors_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2350 unsigned long n = simple_strtoul(buf, &e, 10);
2351 if (*buf && (*e == 0 || *e == '\n')) {
2352 atomic_set(&rdev->corrected_errors, n);
2357 static struct rdev_sysfs_entry rdev_errors =
2358 __ATTR(errors, S_IRUGO|S_IWUSR, errors_show, errors_store);
2361 slot_show(mdk_rdev_t *rdev, char *page)
2363 if (rdev->raid_disk < 0)
2364 return sprintf(page, "none\n");
2366 return sprintf(page, "%d\n", rdev->raid_disk);
2370 slot_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2375 int slot = simple_strtoul(buf, &e, 10);
2376 if (strncmp(buf, "none", 4)==0)
2378 else if (e==buf || (*e && *e!= '\n'))
2380 if (rdev->mddev->pers && slot == -1) {
2381 /* Setting 'slot' on an active array requires also
2382 * updating the 'rd%d' link, and communicating
2383 * with the personality with ->hot_*_disk.
2384 * For now we only support removing
2385 * failed/spare devices. This normally happens automatically,
2386 * but not when the metadata is externally managed.
2388 if (rdev->raid_disk == -1)
2390 /* personality does all needed checks */
2391 if (rdev->mddev->pers->hot_add_disk == NULL)
2393 err = rdev->mddev->pers->
2394 hot_remove_disk(rdev->mddev, rdev->raid_disk);
2397 sprintf(nm, "rd%d", rdev->raid_disk);
2398 sysfs_remove_link(&rdev->mddev->kobj, nm);
2399 rdev->raid_disk = -1;
2400 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2401 md_wakeup_thread(rdev->mddev->thread);
2402 } else if (rdev->mddev->pers) {
2404 /* Activating a spare .. or possibly reactivating
2405 * if we ever get bitmaps working here.
2408 if (rdev->raid_disk != -1)
2411 if (rdev->mddev->pers->hot_add_disk == NULL)
2414 list_for_each_entry(rdev2, &rdev->mddev->disks, same_set)
2415 if (rdev2->raid_disk == slot)
2418 rdev->raid_disk = slot;
2419 if (test_bit(In_sync, &rdev->flags))
2420 rdev->saved_raid_disk = slot;
2422 rdev->saved_raid_disk = -1;
2423 err = rdev->mddev->pers->
2424 hot_add_disk(rdev->mddev, rdev);
2426 rdev->raid_disk = -1;
2429 sysfs_notify_dirent(rdev->sysfs_state);
2430 sprintf(nm, "rd%d", rdev->raid_disk);
2431 if (sysfs_create_link(&rdev->mddev->kobj, &rdev->kobj, nm))
2433 "md: cannot register "
2435 nm, mdname(rdev->mddev));
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(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 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;
2939 if (mddev->pers == NULL) {
2942 if (len >= sizeof(mddev->clevel))
2944 strncpy(mddev->clevel, buf, len);
2945 if (mddev->clevel[len-1] == '\n')
2947 mddev->clevel[len] = 0;
2948 mddev->level = LEVEL_NONE;
2952 /* request to change the personality. Need to ensure:
2953 * - array is not engaged in resync/recovery/reshape
2954 * - old personality can be suspended
2955 * - new personality will access other array.
2958 if (mddev->sync_thread || mddev->reshape_position != MaxSector)
2961 if (!mddev->pers->quiesce) {
2962 printk(KERN_WARNING "md: %s: %s does not support online personality change\n",
2963 mdname(mddev), mddev->pers->name);
2967 /* Now find the new personality */
2968 if (len == 0 || len >= sizeof(level))
2970 strncpy(level, buf, len);
2971 if (level[len-1] == '\n')
2975 request_module("md-%s", level);
2976 spin_lock(&pers_lock);
2977 pers = find_pers(LEVEL_NONE, level);
2978 if (!pers || !try_module_get(pers->owner)) {
2979 spin_unlock(&pers_lock);
2980 printk(KERN_WARNING "md: personality %s not loaded\n", level);
2983 spin_unlock(&pers_lock);
2985 if (pers == mddev->pers) {
2986 /* Nothing to do! */
2987 module_put(pers->owner);
2990 if (!pers->takeover) {
2991 module_put(pers->owner);
2992 printk(KERN_WARNING "md: %s: %s does not support personality takeover\n",
2993 mdname(mddev), level);
2997 /* ->takeover must set new_* and/or delta_disks
2998 * if it succeeds, and may set them when it fails.
3000 priv = pers->takeover(mddev);
3002 mddev->new_level = mddev->level;
3003 mddev->new_layout = mddev->layout;
3004 mddev->new_chunk_sectors = mddev->chunk_sectors;
3005 mddev->raid_disks -= mddev->delta_disks;
3006 mddev->delta_disks = 0;
3007 module_put(pers->owner);
3008 printk(KERN_WARNING "md: %s: %s would not accept array\n",
3009 mdname(mddev), level);
3010 return PTR_ERR(priv);
3013 /* Looks like we have a winner */
3014 mddev_suspend(mddev);
3015 mddev->pers->stop(mddev);
3017 if (mddev->pers->sync_request == NULL &&
3018 pers->sync_request != NULL) {
3019 /* need to add the md_redundancy_group */
3020 if (sysfs_create_group(&mddev->kobj, &md_redundancy_group))
3022 "md: cannot register extra attributes for %s\n",
3024 mddev->sysfs_action = sysfs_get_dirent(mddev->kobj.sd, "sync_action");
3026 if (mddev->pers->sync_request != NULL &&
3027 pers->sync_request == NULL) {
3028 /* need to remove the md_redundancy_group */
3029 if (mddev->to_remove == NULL)
3030 mddev->to_remove = &md_redundancy_group;
3033 if (mddev->pers->sync_request == NULL &&
3035 /* We are converting from a no-redundancy array
3036 * to a redundancy array and metadata is managed
3037 * externally so we need to be sure that writes
3038 * won't block due to a need to transition
3040 * until external management is started.
3043 mddev->safemode_delay = 0;
3044 mddev->safemode = 0;
3047 module_put(mddev->pers->owner);
3048 /* Invalidate devices that are now superfluous */
3049 list_for_each_entry(rdev, &mddev->disks, same_set)
3050 if (rdev->raid_disk >= mddev->raid_disks) {
3051 rdev->raid_disk = -1;
3052 clear_bit(In_sync, &rdev->flags);
3055 mddev->private = priv;
3056 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
3057 mddev->level = mddev->new_level;
3058 mddev->layout = mddev->new_layout;
3059 mddev->chunk_sectors = mddev->new_chunk_sectors;
3060 mddev->delta_disks = 0;
3061 if (mddev->pers->sync_request == NULL) {
3062 /* this is now an array without redundancy, so
3063 * it must always be in_sync
3066 del_timer_sync(&mddev->safemode_timer);
3069 mddev_resume(mddev);
3070 set_bit(MD_CHANGE_DEVS, &mddev->flags);
3071 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3072 md_wakeup_thread(mddev->thread);
3073 sysfs_notify(&mddev->kobj, NULL, "level");
3077 static struct md_sysfs_entry md_level =
3078 __ATTR(level, S_IRUGO|S_IWUSR, level_show, level_store);
3082 layout_show(mddev_t *mddev, char *page)
3084 /* just a number, not meaningful for all levels */
3085 if (mddev->reshape_position != MaxSector &&
3086 mddev->layout != mddev->new_layout)
3087 return sprintf(page, "%d (%d)\n",
3088 mddev->new_layout, mddev->layout);
3089 return sprintf(page, "%d\n", mddev->layout);
3093 layout_store(mddev_t *mddev, const char *buf, size_t len)
3096 unsigned long n = simple_strtoul(buf, &e, 10);
3098 if (!*buf || (*e && *e != '\n'))
3103 if (mddev->pers->check_reshape == NULL)
3105 mddev->new_layout = n;
3106 err = mddev->pers->check_reshape(mddev);
3108 mddev->new_layout = mddev->layout;
3112 mddev->new_layout = n;
3113 if (mddev->reshape_position == MaxSector)
3118 static struct md_sysfs_entry md_layout =
3119 __ATTR(layout, S_IRUGO|S_IWUSR, layout_show, layout_store);
3123 raid_disks_show(mddev_t *mddev, char *page)
3125 if (mddev->raid_disks == 0)
3127 if (mddev->reshape_position != MaxSector &&
3128 mddev->delta_disks != 0)
3129 return sprintf(page, "%d (%d)\n", mddev->raid_disks,
3130 mddev->raid_disks - mddev->delta_disks);
3131 return sprintf(page, "%d\n", mddev->raid_disks);
3134 static int update_raid_disks(mddev_t *mddev, int raid_disks);
3137 raid_disks_store(mddev_t *mddev, const char *buf, size_t len)
3141 unsigned long n = simple_strtoul(buf, &e, 10);
3143 if (!*buf || (*e && *e != '\n'))
3147 rv = update_raid_disks(mddev, n);
3148 else if (mddev->reshape_position != MaxSector) {
3149 int olddisks = mddev->raid_disks - mddev->delta_disks;
3150 mddev->delta_disks = n - olddisks;
3151 mddev->raid_disks = n;
3153 mddev->raid_disks = n;
3154 return rv ? rv : len;
3156 static struct md_sysfs_entry md_raid_disks =
3157 __ATTR(raid_disks, S_IRUGO|S_IWUSR, raid_disks_show, raid_disks_store);
3160 chunk_size_show(mddev_t *mddev, char *page)
3162 if (mddev->reshape_position != MaxSector &&
3163 mddev->chunk_sectors != mddev->new_chunk_sectors)
3164 return sprintf(page, "%d (%d)\n",
3165 mddev->new_chunk_sectors << 9,
3166 mddev->chunk_sectors << 9);
3167 return sprintf(page, "%d\n", mddev->chunk_sectors << 9);
3171 chunk_size_store(mddev_t *mddev, const char *buf, size_t len)
3174 unsigned long n = simple_strtoul(buf, &e, 10);
3176 if (!*buf || (*e && *e != '\n'))
3181 if (mddev->pers->check_reshape == NULL)
3183 mddev->new_chunk_sectors = n >> 9;
3184 err = mddev->pers->check_reshape(mddev);
3186 mddev->new_chunk_sectors = mddev->chunk_sectors;
3190 mddev->new_chunk_sectors = n >> 9;
3191 if (mddev->reshape_position == MaxSector)
3192 mddev->chunk_sectors = n >> 9;
3196 static struct md_sysfs_entry md_chunk_size =
3197 __ATTR(chunk_size, S_IRUGO|S_IWUSR, chunk_size_show, chunk_size_store);
3200 resync_start_show(mddev_t *mddev, char *page)
3202 if (mddev->recovery_cp == MaxSector)
3203 return sprintf(page, "none\n");
3204 return sprintf(page, "%llu\n", (unsigned long long)mddev->recovery_cp);
3208 resync_start_store(mddev_t *mddev, const char *buf, size_t len)
3211 unsigned long long n = simple_strtoull(buf, &e, 10);
3215 if (cmd_match(buf, "none"))
3217 else if (!*buf || (*e && *e != '\n'))
3220 mddev->recovery_cp = n;
3223 static struct md_sysfs_entry md_resync_start =
3224 __ATTR(resync_start, S_IRUGO|S_IWUSR, resync_start_show, resync_start_store);
3227 * The array state can be:
3230 * No devices, no size, no level
3231 * Equivalent to STOP_ARRAY ioctl
3233 * May have some settings, but array is not active
3234 * all IO results in error
3235 * When written, doesn't tear down array, but just stops it
3236 * suspended (not supported yet)
3237 * All IO requests will block. The array can be reconfigured.
3238 * Writing this, if accepted, will block until array is quiescent
3240 * no resync can happen. no superblocks get written.
3241 * write requests fail
3243 * like readonly, but behaves like 'clean' on a write request.
3245 * clean - no pending writes, but otherwise active.
3246 * When written to inactive array, starts without resync
3247 * If a write request arrives then
3248 * if metadata is known, mark 'dirty' and switch to 'active'.
3249 * if not known, block and switch to write-pending
3250 * If written to an active array that has pending writes, then fails.
3252 * fully active: IO and resync can be happening.
3253 * When written to inactive array, starts with resync
3256 * clean, but writes are blocked waiting for 'active' to be written.
3259 * like active, but no writes have been seen for a while (100msec).
3262 enum array_state { clear, inactive, suspended, readonly, read_auto, clean, active,
3263 write_pending, active_idle, bad_word};
3264 static char *array_states[] = {
3265 "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active",
3266 "write-pending", "active-idle", NULL };
3268 static int match_word(const char *word, char **list)
3271 for (n=0; list[n]; n++)
3272 if (cmd_match(word, list[n]))
3278 array_state_show(mddev_t *mddev, char *page)
3280 enum array_state st = inactive;
3293 else if (test_bit(MD_CHANGE_CLEAN, &mddev->flags))
3295 else if (mddev->safemode)
3301 if (list_empty(&mddev->disks) &&
3302 mddev->raid_disks == 0 &&
3303 mddev->dev_sectors == 0)
3308 return sprintf(page, "%s\n", array_states[st]);
3311 static int do_md_stop(mddev_t * mddev, int ro, int is_open);
3312 static int do_md_run(mddev_t * mddev);
3313 static int restart_array(mddev_t *mddev);
3316 array_state_store(mddev_t *mddev, const char *buf, size_t len)
3319 enum array_state st = match_word(buf, array_states);
3324 /* stopping an active array */
3325 if (atomic_read(&mddev->openers) > 0)
3327 err = do_md_stop(mddev, 0, 0);
3330 /* stopping an active array */
3332 if (atomic_read(&mddev->openers) > 0)
3334 err = do_md_stop(mddev, 2, 0);
3336 err = 0; /* already inactive */
3339 break; /* not supported yet */
3342 err = do_md_stop(mddev, 1, 0);
3345 set_disk_ro(mddev->gendisk, 1);
3346 err = do_md_run(mddev);
3352 err = do_md_stop(mddev, 1, 0);
3353 else if (mddev->ro == 1)
3354 err = restart_array(mddev);
3357 set_disk_ro(mddev->gendisk, 0);
3361 err = do_md_run(mddev);
3366 restart_array(mddev);
3367 spin_lock_irq(&mddev->write_lock);
3368 if (atomic_read(&mddev->writes_pending) == 0) {
3369 if (mddev->in_sync == 0) {
3371 if (mddev->safemode == 1)
3372 mddev->safemode = 0;
3373 if (mddev->persistent)
3374 set_bit(MD_CHANGE_CLEAN,
3380 spin_unlock_irq(&mddev->write_lock);
3386 restart_array(mddev);
3387 if (mddev->external)
3388 clear_bit(MD_CHANGE_CLEAN, &mddev->flags);
3389 wake_up(&mddev->sb_wait);
3393 set_disk_ro(mddev->gendisk, 0);
3394 err = do_md_run(mddev);
3399 /* these cannot be set */
3405 sysfs_notify_dirent(mddev->sysfs_state);
3409 static struct md_sysfs_entry md_array_state =
3410 __ATTR(array_state, S_IRUGO|S_IWUSR, array_state_show, array_state_store);
3413 max_corrected_read_errors_show(mddev_t *mddev, char *page) {
3414 return sprintf(page, "%d\n",
3415 atomic_read(&mddev->max_corr_read_errors));
3419 max_corrected_read_errors_store(mddev_t *mddev, const char *buf, size_t len)
3422 unsigned long n = simple_strtoul(buf, &e, 10);
3424 if (*buf && (*e == 0 || *e == '\n')) {
3425 atomic_set(&mddev->max_corr_read_errors, n);
3431 static struct md_sysfs_entry max_corr_read_errors =
3432 __ATTR(max_read_errors, S_IRUGO|S_IWUSR, max_corrected_read_errors_show,
3433 max_corrected_read_errors_store);
3436 null_show(mddev_t *mddev, char *page)
3442 new_dev_store(mddev_t *mddev, const char *buf, size_t len)
3444 /* buf must be %d:%d\n? giving major and minor numbers */
3445 /* The new device is added to the array.
3446 * If the array has a persistent superblock, we read the
3447 * superblock to initialise info and check validity.
3448 * Otherwise, only checking done is that in bind_rdev_to_array,
3449 * which mainly checks size.
3452 int major = simple_strtoul(buf, &e, 10);
3458 if (!*buf || *e != ':' || !e[1] || e[1] == '\n')
3460 minor = simple_strtoul(e+1, &e, 10);
3461 if (*e && *e != '\n')
3463 dev = MKDEV(major, minor);
3464 if (major != MAJOR(dev) ||
3465 minor != MINOR(dev))
3469 if (mddev->persistent) {
3470 rdev = md_import_device(dev, mddev->major_version,
3471 mddev->minor_version);
3472 if (!IS_ERR(rdev) && !list_empty(&mddev->disks)) {
3473 mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
3474 mdk_rdev_t, same_set);
3475 err = super_types[mddev->major_version]
3476 .load_super(rdev, rdev0, mddev->minor_version);
3480 } else if (mddev->external)
3481 rdev = md_import_device(dev, -2, -1);
3483 rdev = md_import_device(dev, -1, -1);
3486 return PTR_ERR(rdev);
3487 err = bind_rdev_to_array(rdev, mddev);
3491 return err ? err : len;
3494 static struct md_sysfs_entry md_new_device =
3495 __ATTR(new_dev, S_IWUSR, null_show, new_dev_store);
3498 bitmap_store(mddev_t *mddev, const char *buf, size_t len)
3501 unsigned long chunk, end_chunk;
3505 /* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */
3507 chunk = end_chunk = simple_strtoul(buf, &end, 0);
3508 if (buf == end) break;
3509 if (*end == '-') { /* range */
3511 end_chunk = simple_strtoul(buf, &end, 0);
3512 if (buf == end) break;
3514 if (*end && !isspace(*end)) break;
3515 bitmap_dirty_bits(mddev->bitmap, chunk, end_chunk);
3516 buf = skip_spaces(end);
3518 bitmap_unplug(mddev->bitmap); /* flush the bits to disk */
3523 static struct md_sysfs_entry md_bitmap =
3524 __ATTR(bitmap_set_bits, S_IWUSR, null_show, bitmap_store);
3527 size_show(mddev_t *mddev, char *page)
3529 return sprintf(page, "%llu\n",
3530 (unsigned long long)mddev->dev_sectors / 2);
3533 static int update_size(mddev_t *mddev, sector_t num_sectors);
3536 size_store(mddev_t *mddev, const char *buf, size_t len)
3538 /* If array is inactive, we can reduce the component size, but
3539 * not increase it (except from 0).
3540 * If array is active, we can try an on-line resize
3543 int err = strict_blocks_to_sectors(buf, §ors);
3548 err = update_size(mddev, sectors);
3549 md_update_sb(mddev, 1);
3551 if (mddev->dev_sectors == 0 ||
3552 mddev->dev_sectors > sectors)
3553 mddev->dev_sectors = sectors;
3557 return err ? err : len;
3560 static struct md_sysfs_entry md_size =
3561 __ATTR(component_size, S_IRUGO|S_IWUSR, size_show, size_store);
3566 * 'none' for arrays with no metadata (good luck...)
3567 * 'external' for arrays with externally managed metadata,
3568 * or N.M for internally known formats
3571 metadata_show(mddev_t *mddev, char *page)
3573 if (mddev->persistent)
3574 return sprintf(page, "%d.%d\n",
3575 mddev->major_version, mddev->minor_version);
3576 else if (mddev->external)
3577 return sprintf(page, "external:%s\n", mddev->metadata_type);
3579 return sprintf(page, "none\n");
3583 metadata_store(mddev_t *mddev, const char *buf, size_t len)
3587 /* Changing the details of 'external' metadata is
3588 * always permitted. Otherwise there must be
3589 * no devices attached to the array.
3591 if (mddev->external && strncmp(buf, "external:", 9) == 0)
3593 else if (!list_empty(&mddev->disks))
3596 if (cmd_match(buf, "none")) {
3597 mddev->persistent = 0;
3598 mddev->external = 0;
3599 mddev->major_version = 0;
3600 mddev->minor_version = 90;
3603 if (strncmp(buf, "external:", 9) == 0) {
3604 size_t namelen = len-9;
3605 if (namelen >= sizeof(mddev->metadata_type))
3606 namelen = sizeof(mddev->metadata_type)-1;
3607 strncpy(mddev->metadata_type, buf+9, namelen);
3608 mddev->metadata_type[namelen] = 0;
3609 if (namelen && mddev->metadata_type[namelen-1] == '\n')
3610 mddev->metadata_type[--namelen] = 0;
3611 mddev->persistent = 0;
3612 mddev->external = 1;
3613 mddev->major_version = 0;
3614 mddev->minor_version = 90;
3617 major = simple_strtoul(buf, &e, 10);
3618 if (e==buf || *e != '.')
3621 minor = simple_strtoul(buf, &e, 10);
3622 if (e==buf || (*e && *e != '\n') )
3624 if (major >= ARRAY_SIZE(super_types) || super_types[major].name == NULL)
3626 mddev->major_version = major;
3627 mddev->minor_version = minor;
3628 mddev->persistent = 1;
3629 mddev->external = 0;
3633 static struct md_sysfs_entry md_metadata =
3634 __ATTR(metadata_version, S_IRUGO|S_IWUSR, metadata_show, metadata_store);
3637 action_show(mddev_t *mddev, char *page)
3639 char *type = "idle";
3640 if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
3642 else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
3643 (!mddev->ro && test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))) {
3644 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
3646 else if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
3647 if (!test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
3649 else if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
3653 } else if (test_bit(MD_RECOVERY_RECOVER, &mddev->recovery))
3656 return sprintf(page, "%s\n", type);
3660 action_store(mddev_t *mddev, const char *page, size_t len)
3662 if (!mddev->pers || !mddev->pers->sync_request)
3665 if (cmd_match(page, "frozen"))
3666 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
3668 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
3670 if (cmd_match(page, "idle") || cmd_match(page, "frozen")) {
3671 if (mddev->sync_thread) {
3672 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
3673 md_unregister_thread(mddev->sync_thread);
3674 mddev->sync_thread = NULL;
3675 mddev->recovery = 0;
3677 } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
3678 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
3680 else if (cmd_match(page, "resync"))
3681 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3682 else if (cmd_match(page, "recover")) {
3683 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
3684 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3685 } else if (cmd_match(page, "reshape")) {
3687 if (mddev->pers->start_reshape == NULL)
3689 err = mddev->pers->start_reshape(mddev);
3692 sysfs_notify(&mddev->kobj, NULL, "degraded");
3694 if (cmd_match(page, "check"))
3695 set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
3696 else if (!cmd_match(page, "repair"))
3698 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
3699 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
3701 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3702 md_wakeup_thread(mddev->thread);
3703 sysfs_notify_dirent(mddev->sysfs_action);
3708 mismatch_cnt_show(mddev_t *mddev, char *page)
3710 return sprintf(page, "%llu\n",
3711 (unsigned long long) mddev->resync_mismatches);
3714 static struct md_sysfs_entry md_scan_mode =
3715 __ATTR(sync_action, S_IRUGO|S_IWUSR, action_show, action_store);
3718 static struct md_sysfs_entry md_mismatches = __ATTR_RO(mismatch_cnt);
3721 sync_min_show(mddev_t *mddev, char *page)
3723 return sprintf(page, "%d (%s)\n", speed_min(mddev),
3724 mddev->sync_speed_min ? "local": "system");
3728 sync_min_store(mddev_t *mddev, const char *buf, size_t len)
3732 if (strncmp(buf, "system", 6)==0) {
3733 mddev->sync_speed_min = 0;
3736 min = simple_strtoul(buf, &e, 10);
3737 if (buf == e || (*e && *e != '\n') || min <= 0)
3739 mddev->sync_speed_min = min;
3743 static struct md_sysfs_entry md_sync_min =
3744 __ATTR(sync_speed_min, S_IRUGO|S_IWUSR, sync_min_show, sync_min_store);
3747 sync_max_show(mddev_t *mddev, char *page)
3749 return sprintf(page, "%d (%s)\n", speed_max(mddev),
3750 mddev->sync_speed_max ? "local": "system");
3754 sync_max_store(mddev_t *mddev, const char *buf, size_t len)
3758 if (strncmp(buf, "system", 6)==0) {
3759 mddev->sync_speed_max = 0;
3762 max = simple_strtoul(buf, &e, 10);
3763 if (buf == e || (*e && *e != '\n') || max <= 0)
3765 mddev->sync_speed_max = max;
3769 static struct md_sysfs_entry md_sync_max =
3770 __ATTR(sync_speed_max, S_IRUGO|S_IWUSR, sync_max_show, sync_max_store);
3773 degraded_show(mddev_t *mddev, char *page)
3775 return sprintf(page, "%d\n", mddev->degraded);
3777 static struct md_sysfs_entry md_degraded = __ATTR_RO(degraded);
3780 sync_force_parallel_show(mddev_t *mddev, char *page)
3782 return sprintf(page, "%d\n", mddev->parallel_resync);
3786 sync_force_parallel_store(mddev_t *mddev, const char *buf, size_t len)
3790 if (strict_strtol(buf, 10, &n))
3793 if (n != 0 && n != 1)
3796 mddev->parallel_resync = n;
3798 if (mddev->sync_thread)
3799 wake_up(&resync_wait);
3804 /* force parallel resync, even with shared block devices */
3805 static struct md_sysfs_entry md_sync_force_parallel =
3806 __ATTR(sync_force_parallel, S_IRUGO|S_IWUSR,
3807 sync_force_parallel_show, sync_force_parallel_store);
3810 sync_speed_show(mddev_t *mddev, char *page)
3812 unsigned long resync, dt, db;
3813 if (mddev->curr_resync == 0)
3814 return sprintf(page, "none\n");
3815 resync = mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active);
3816 dt = (jiffies - mddev->resync_mark) / HZ;
3818 db = resync - mddev->resync_mark_cnt;
3819 return sprintf(page, "%lu\n", db/dt/2); /* K/sec */
3822 static struct md_sysfs_entry md_sync_speed = __ATTR_RO(sync_speed);
3825 sync_completed_show(mddev_t *mddev, char *page)
3827 unsigned long max_sectors, resync;
3829 if (!test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
3830 return sprintf(page, "none\n");
3832 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
3833 max_sectors = mddev->resync_max_sectors;
3835 max_sectors = mddev->dev_sectors;
3837 resync = mddev->curr_resync_completed;
3838 return sprintf(page, "%lu / %lu\n", resync, max_sectors);
3841 static struct md_sysfs_entry md_sync_completed = __ATTR_RO(sync_completed);
3844 min_sync_show(mddev_t *mddev, char *page)
3846 return sprintf(page, "%llu\n",
3847 (unsigned long long)mddev->resync_min);
3850 min_sync_store(mddev_t *mddev, const char *buf, size_t len)
3852 unsigned long long min;
3853 if (strict_strtoull(buf, 10, &min))
3855 if (min > mddev->resync_max)
3857 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
3860 /* Must be a multiple of chunk_size */
3861 if (mddev->chunk_sectors) {
3862 sector_t temp = min;
3863 if (sector_div(temp, mddev->chunk_sectors))
3866 mddev->resync_min = min;
3871 static struct md_sysfs_entry md_min_sync =
3872 __ATTR(sync_min, S_IRUGO|S_IWUSR, min_sync_show, min_sync_store);
3875 max_sync_show(mddev_t *mddev, char *page)
3877 if (mddev->resync_max == MaxSector)
3878 return sprintf(page, "max\n");
3880 return sprintf(page, "%llu\n",
3881 (unsigned long long)mddev->resync_max);
3884 max_sync_store(mddev_t *mddev, const char *buf, size_t len)
3886 if (strncmp(buf, "max", 3) == 0)
3887 mddev->resync_max = MaxSector;
3889 unsigned long long max;
3890 if (strict_strtoull(buf, 10, &max))
3892 if (max < mddev->resync_min)
3894 if (max < mddev->resync_max &&
3896 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
3899 /* Must be a multiple of chunk_size */
3900 if (mddev->chunk_sectors) {
3901 sector_t temp = max;
3902 if (sector_div(temp, mddev->chunk_sectors))
3905 mddev->resync_max = max;
3907 wake_up(&mddev->recovery_wait);
3911 static struct md_sysfs_entry md_max_sync =
3912 __ATTR(sync_max, S_IRUGO|S_IWUSR, max_sync_show, max_sync_store);
3915 suspend_lo_show(mddev_t *mddev, char *page)
3917 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_lo);
3921 suspend_lo_store(mddev_t *mddev, const char *buf, size_t len)
3924 unsigned long long new = simple_strtoull(buf, &e, 10);
3926 if (mddev->pers == NULL ||
3927 mddev->pers->quiesce == NULL)
3929 if (buf == e || (*e && *e != '\n'))
3931 if (new >= mddev->suspend_hi ||
3932 (new > mddev->suspend_lo && new < mddev->suspend_hi)) {
3933 mddev->suspend_lo = new;
3934 mddev->pers->quiesce(mddev, 2);
3939 static struct md_sysfs_entry md_suspend_lo =
3940 __ATTR(suspend_lo, S_IRUGO|S_IWUSR, suspend_lo_show, suspend_lo_store);
3944 suspend_hi_show(mddev_t *mddev, char *page)
3946 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_hi);
3950 suspend_hi_store(mddev_t *mddev, const char *buf, size_t len)
3953 unsigned long long new = simple_strtoull(buf, &e, 10);
3955 if (mddev->pers == NULL ||
3956 mddev->pers->quiesce == NULL)
3958 if (buf == e || (*e && *e != '\n'))
3960 if ((new <= mddev->suspend_lo && mddev->suspend_lo >= mddev->suspend_hi) ||
3961 (new > mddev->suspend_lo && new > mddev->suspend_hi)) {
3962 mddev->suspend_hi = new;
3963 mddev->pers->quiesce(mddev, 1);
3964 mddev->pers->quiesce(mddev, 0);
3969 static struct md_sysfs_entry md_suspend_hi =
3970 __ATTR(suspend_hi, S_IRUGO|S_IWUSR, suspend_hi_show, suspend_hi_store);
3973 reshape_position_show(mddev_t *mddev, char *page)
3975 if (mddev->reshape_position != MaxSector)
3976 return sprintf(page, "%llu\n",
3977 (unsigned long long)mddev->reshape_position);
3978 strcpy(page, "none\n");
3983 reshape_position_store(mddev_t *mddev, const char *buf, size_t len)
3986 unsigned long long new = simple_strtoull(buf, &e, 10);
3989 if (buf == e || (*e && *e != '\n'))
3991 mddev->reshape_position = new;
3992 mddev->delta_disks = 0;
3993 mddev->new_level = mddev->level;
3994 mddev->new_layout = mddev->layout;
3995 mddev->new_chunk_sectors = mddev->chunk_sectors;
3999 static struct md_sysfs_entry md_reshape_position =
4000 __ATTR(reshape_position, S_IRUGO|S_IWUSR, reshape_position_show,
4001 reshape_position_store);
4004 array_size_show(mddev_t *mddev, char *page)
4006 if (mddev->external_size)
4007 return sprintf(page, "%llu\n",
4008 (unsigned long long)mddev->array_sectors/2);
4010 return sprintf(page, "default\n");
4014 array_size_store(mddev_t *mddev, const char *buf, size_t len)
4018 if (strncmp(buf, "default", 7) == 0) {
4020 sectors = mddev->pers->size(mddev, 0, 0);
4022 sectors = mddev->array_sectors;
4024 mddev->external_size = 0;
4026 if (strict_blocks_to_sectors(buf, §ors) < 0)
4028 if (mddev->pers && mddev->pers->size(mddev, 0, 0) < sectors)
4031 mddev->external_size = 1;
4034 mddev->array_sectors = sectors;
4035 set_capacity(mddev->gendisk, mddev->array_sectors);
4037 revalidate_disk(mddev->gendisk);
4042 static struct md_sysfs_entry md_array_size =
4043 __ATTR(array_size, S_IRUGO|S_IWUSR, array_size_show,
4046 static struct attribute *md_default_attrs[] = {
4049 &md_raid_disks.attr,
4050 &md_chunk_size.attr,
4052 &md_resync_start.attr,
4054 &md_new_device.attr,
4055 &md_safe_delay.attr,
4056 &md_array_state.attr,
4057 &md_reshape_position.attr,
4058 &md_array_size.attr,
4059 &max_corr_read_errors.attr,
4063 static struct attribute *md_redundancy_attrs[] = {
4065 &md_mismatches.attr,
4068 &md_sync_speed.attr,
4069 &md_sync_force_parallel.attr,
4070 &md_sync_completed.attr,
4073 &md_suspend_lo.attr,
4074 &md_suspend_hi.attr,
4079 static struct attribute_group md_redundancy_group = {
4081 .attrs = md_redundancy_attrs,
4086 md_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
4088 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
4089 mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
4094 rv = mddev_lock(mddev);
4096 rv = entry->show(mddev, page);
4097 mddev_unlock(mddev);
4103 md_attr_store(struct kobject *kobj, struct attribute *attr,
4104 const char *page, size_t length)
4106 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
4107 mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
4112 if (!capable(CAP_SYS_ADMIN))
4114 rv = mddev_lock(mddev);
4115 if (mddev->hold_active == UNTIL_IOCTL)
4116 mddev->hold_active = 0;
4118 rv = entry->store(mddev, page, length);
4119 mddev_unlock(mddev);
4124 static void md_free(struct kobject *ko)
4126 mddev_t *mddev = container_of(ko, mddev_t, kobj);
4128 if (mddev->sysfs_state)
4129 sysfs_put(mddev->sysfs_state);
4131 if (mddev->gendisk) {
4132 del_gendisk(mddev->gendisk);
4133 put_disk(mddev->gendisk);
4136 blk_cleanup_queue(mddev->queue);
4141 static struct sysfs_ops md_sysfs_ops = {
4142 .show = md_attr_show,
4143 .store = md_attr_store,
4145 static struct kobj_type md_ktype = {
4147 .sysfs_ops = &md_sysfs_ops,
4148 .default_attrs = md_default_attrs,
4153 static void mddev_delayed_delete(struct work_struct *ws)
4155 mddev_t *mddev = container_of(ws, mddev_t, del_work);
4157 sysfs_remove_group(&mddev->kobj, &md_bitmap_group);
4158 kobject_del(&mddev->kobj);
4159 kobject_put(&mddev->kobj);
4162 static int md_alloc(dev_t dev, char *name)
4164 static DEFINE_MUTEX(disks_mutex);
4165 mddev_t *mddev = mddev_find(dev);
4166 struct gendisk *disk;
4175 partitioned = (MAJOR(mddev->unit) != MD_MAJOR);
4176 shift = partitioned ? MdpMinorShift : 0;
4177 unit = MINOR(mddev->unit) >> shift;
4179 /* wait for any previous instance if this device
4180 * to be completed removed (mddev_delayed_delete).
4182 flush_scheduled_work();
4184 mutex_lock(&disks_mutex);
4190 /* Need to ensure that 'name' is not a duplicate.
4193 spin_lock(&all_mddevs_lock);
4195 list_for_each_entry(mddev2, &all_mddevs, all_mddevs)
4196 if (mddev2->gendisk &&
4197 strcmp(mddev2->gendisk->disk_name, name) == 0) {
4198 spin_unlock(&all_mddevs_lock);
4201 spin_unlock(&all_mddevs_lock);
4205 mddev->queue = blk_alloc_queue(GFP_KERNEL);
4208 mddev->queue->queuedata = mddev;
4210 /* Can be unlocked because the queue is new: no concurrency */
4211 queue_flag_set_unlocked(QUEUE_FLAG_CLUSTER, mddev->queue);
4213 blk_queue_make_request(mddev->queue, md_make_request);
4215 disk = alloc_disk(1 << shift);
4217 blk_cleanup_queue(mddev->queue);
4218 mddev->queue = NULL;
4221 disk->major = MAJOR(mddev->unit);
4222 disk->first_minor = unit << shift;
4224 strcpy(disk->disk_name, name);
4225 else if (partitioned)
4226 sprintf(disk->disk_name, "md_d%d", unit);
4228 sprintf(disk->disk_name, "md%d", unit);
4229 disk->fops = &md_fops;
4230 disk->private_data = mddev;
4231 disk->queue = mddev->queue;
4232 /* Allow extended partitions. This makes the
4233 * 'mdp' device redundant, but we can't really
4236 disk->flags |= GENHD_FL_EXT_DEVT;
4238 mddev->gendisk = disk;
4239 error = kobject_init_and_add(&mddev->kobj, &md_ktype,
4240 &disk_to_dev(disk)->kobj, "%s", "md");
4242 /* This isn't possible, but as kobject_init_and_add is marked
4243 * __must_check, we must do something with the result
4245 printk(KERN_WARNING "md: cannot register %s/md - name in use\n",
4249 if (sysfs_create_group(&mddev->kobj, &md_bitmap_group))
4250 printk(KERN_DEBUG "pointless warning\n");
4252 mutex_unlock(&disks_mutex);
4254 kobject_uevent(&mddev->kobj, KOBJ_ADD);
4255 mddev->sysfs_state = sysfs_get_dirent(mddev->kobj.sd, "array_state");
4261 static struct kobject *md_probe(dev_t dev, int *part, void *data)
4263 md_alloc(dev, NULL);
4267 static int add_named_array(const char *val, struct kernel_param *kp)
4269 /* val must be "md_*" where * is not all digits.
4270 * We allocate an array with a large free minor number, and
4271 * set the name to val. val must not already be an active name.
4273 int len = strlen(val);
4274 char buf[DISK_NAME_LEN];
4276 while (len && val[len-1] == '\n')
4278 if (len >= DISK_NAME_LEN)
4280 strlcpy(buf, val, len+1);
4281 if (strncmp(buf, "md_", 3) != 0)
4283 return md_alloc(0, buf);
4286 static void md_safemode_timeout(unsigned long data)
4288 mddev_t *mddev = (mddev_t *) data;
4290 if (!atomic_read(&mddev->writes_pending)) {
4291 mddev->safemode = 1;
4292 if (mddev->external)
4293 sysfs_notify_dirent(mddev->sysfs_state);
4295 md_wakeup_thread(mddev->thread);
4298 static int start_dirty_degraded;
4300 static int md_run(mddev_t *mddev)
4304 struct mdk_personality *pers;
4306 if (list_empty(&mddev->disks))
4307 /* cannot run an array with no devices.. */
4313 /* These two calls synchronise us with the
4314 * sysfs_remove_group calls in mddev_unlock,
4315 * so they must have completed.
4317 mutex_lock(&mddev->open_mutex);
4318 mutex_unlock(&mddev->open_mutex);
4321 * Analyze all RAID superblock(s)
4323 if (!mddev->raid_disks) {
4324 if (!mddev->persistent)
4329 if (mddev->level != LEVEL_NONE)
4330 request_module("md-level-%d", mddev->level);
4331 else if (mddev->clevel[0])
4332 request_module("md-%s", mddev->clevel);
4335 * Drop all container device buffers, from now on
4336 * the only valid external interface is through the md
4339 list_for_each_entry(rdev, &mddev->disks, same_set) {
4340 if (test_bit(Faulty, &rdev->flags))
4342 sync_blockdev(rdev->bdev);
4343 invalidate_bdev(rdev->bdev);
4345 /* perform some consistency tests on the device.
4346 * We don't want the data to overlap the metadata,
4347 * Internal Bitmap issues have been handled elsewhere.
4349 if (rdev->data_offset < rdev->sb_start) {
4350 if (mddev->dev_sectors &&
4351 rdev->data_offset + mddev->dev_sectors
4353 printk("md: %s: data overlaps metadata\n",
4358 if (rdev->sb_start + rdev->sb_size/512
4359 > rdev->data_offset) {
4360 printk("md: %s: metadata overlaps data\n",
4365 sysfs_notify_dirent(rdev->sysfs_state);
4368 spin_lock(&pers_lock);
4369 pers = find_pers(mddev->level, mddev->clevel);
4370 if (!pers || !try_module_get(pers->owner)) {
4371 spin_unlock(&pers_lock);
4372 if (mddev->level != LEVEL_NONE)
4373 printk(KERN_WARNING "md: personality for level %d is not loaded!\n",
4376 printk(KERN_WARNING "md: personality for level %s is not loaded!\n",
4381 spin_unlock(&pers_lock);
4382 if (mddev->level != pers->level) {
4383 mddev->level = pers->level;
4384 mddev->new_level = pers->level;
4386 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
4388 if (mddev->reshape_position != MaxSector &&
4389 pers->start_reshape == NULL) {
4390 /* This personality cannot handle reshaping... */
4392 module_put(pers->owner);
4396 if (pers->sync_request) {
4397 /* Warn if this is a potentially silly
4400 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
4404 list_for_each_entry(rdev, &mddev->disks, same_set)
4405 list_for_each_entry(rdev2, &mddev->disks, same_set) {
4407 rdev->bdev->bd_contains ==
4408 rdev2->bdev->bd_contains) {
4410 "%s: WARNING: %s appears to be"
4411 " on the same physical disk as"
4414 bdevname(rdev->bdev,b),
4415 bdevname(rdev2->bdev,b2));
4422 "True protection against single-disk"
4423 " failure might be compromised.\n");
4426 mddev->recovery = 0;
4427 /* may be over-ridden by personality */
4428 mddev->resync_max_sectors = mddev->dev_sectors;
4430 mddev->barriers_work = 1;
4431 mddev->ok_start_degraded = start_dirty_degraded;
4433 if (start_readonly && mddev->ro == 0)
4434 mddev->ro = 2; /* read-only, but switch on first write */
4436 err = mddev->pers->run(mddev);
4438 printk(KERN_ERR "md: pers->run() failed ...\n");
4439 else if (mddev->pers->size(mddev, 0, 0) < mddev->array_sectors) {
4440 WARN_ONCE(!mddev->external_size, "%s: default size too small,"
4441 " but 'external_size' not in effect?\n", __func__);
4443 "md: invalid array_size %llu > default size %llu\n",
4444 (unsigned long long)mddev->array_sectors / 2,
4445 (unsigned long long)mddev->pers->size(mddev, 0, 0) / 2);
4447 mddev->pers->stop(mddev);
4449 if (err == 0 && mddev->pers->sync_request) {
4450 err = bitmap_create(mddev);
4452 printk(KERN_ERR "%s: failed to create bitmap (%d)\n",
4453 mdname(mddev), err);
4454 mddev->pers->stop(mddev);
4458 module_put(mddev->pers->owner);
4460 bitmap_destroy(mddev);
4463 if (mddev->pers->sync_request) {
4464 if (sysfs_create_group(&mddev->kobj, &md_redundancy_group))
4466 "md: cannot register extra attributes for %s\n",
4468 mddev->sysfs_action = sysfs_get_dirent(mddev->kobj.sd, "sync_action");
4469 } else if (mddev->ro == 2) /* auto-readonly not meaningful */
4472 atomic_set(&mddev->writes_pending,0);
4473 atomic_set(&mddev->max_corr_read_errors,
4474 MD_DEFAULT_MAX_CORRECTED_READ_ERRORS);
4475 mddev->safemode = 0;
4476 mddev->safemode_timer.function = md_safemode_timeout;
4477 mddev->safemode_timer.data = (unsigned long) mddev;
4478 mddev->safemode_delay = (200 * HZ)/1000 +1; /* 200 msec delay */
4481 list_for_each_entry(rdev, &mddev->disks, same_set)
4482 if (rdev->raid_disk >= 0) {
4484 sprintf(nm, "rd%d", rdev->raid_disk);
4485 if (sysfs_create_link(&mddev->kobj, &rdev->kobj, nm))
4486 printk("md: cannot register %s for %s\n",
4490 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4493 md_update_sb(mddev, 0);
4495 md_wakeup_thread(mddev->thread);
4496 md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
4498 md_new_event(mddev);
4499 sysfs_notify_dirent(mddev->sysfs_state);
4500 if (mddev->sysfs_action)
4501 sysfs_notify_dirent(mddev->sysfs_action);
4502 sysfs_notify(&mddev->kobj, NULL, "degraded");
4506 static int do_md_run(mddev_t *mddev)
4510 err = md_run(mddev);
4514 set_capacity(mddev->gendisk, mddev->array_sectors);
4515 revalidate_disk(mddev->gendisk);
4516 kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
4521 static int restart_array(mddev_t *mddev)
4523 struct gendisk *disk = mddev->gendisk;
4525 /* Complain if it has no devices */
4526 if (list_empty(&mddev->disks))
4532 mddev->safemode = 0;
4534 set_disk_ro(disk, 0);
4535 printk(KERN_INFO "md: %s switched to read-write mode.\n",
4537 /* Kick recovery or resync if necessary */
4538 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4539 md_wakeup_thread(mddev->thread);
4540 md_wakeup_thread(mddev->sync_thread);
4541 sysfs_notify_dirent(mddev->sysfs_state);
4545 /* similar to deny_write_access, but accounts for our holding a reference
4546 * to the file ourselves */
4547 static int deny_bitmap_write_access(struct file * file)
4549 struct inode *inode = file->f_mapping->host;
4551 spin_lock(&inode->i_lock);
4552 if (atomic_read(&inode->i_writecount) > 1) {
4553 spin_unlock(&inode->i_lock);
4556 atomic_set(&inode->i_writecount, -1);
4557 spin_unlock(&inode->i_lock);
4562 void restore_bitmap_write_access(struct file *file)
4564 struct inode *inode = file->f_mapping->host;
4566 spin_lock(&inode->i_lock);
4567 atomic_set(&inode->i_writecount, 1);
4568 spin_unlock(&inode->i_lock);
4571 static void md_clean(mddev_t *mddev)
4573 mddev->array_sectors = 0;
4574 mddev->external_size = 0;
4575 mddev->dev_sectors = 0;
4576 mddev->raid_disks = 0;
4577 mddev->recovery_cp = 0;
4578 mddev->resync_min = 0;
4579 mddev->resync_max = MaxSector;
4580 mddev->reshape_position = MaxSector;
4581 mddev->external = 0;
4582 mddev->persistent = 0;
4583 mddev->level = LEVEL_NONE;
4584 mddev->clevel[0] = 0;
4587 mddev->metadata_type[0] = 0;
4588 mddev->chunk_sectors = 0;
4589 mddev->ctime = mddev->utime = 0;
4591 mddev->max_disks = 0;
4593 mddev->delta_disks = 0;
4594 mddev->new_level = LEVEL_NONE;
4595 mddev->new_layout = 0;
4596 mddev->new_chunk_sectors = 0;
4597 mddev->curr_resync = 0;
4598 mddev->resync_mismatches = 0;
4599 mddev->suspend_lo = mddev->suspend_hi = 0;
4600 mddev->sync_speed_min = mddev->sync_speed_max = 0;
4601 mddev->recovery = 0;
4603 mddev->degraded = 0;
4604 mddev->barriers_work = 0;
4605 mddev->safemode = 0;
4606 mddev->bitmap_info.offset = 0;
4607 mddev->bitmap_info.default_offset = 0;
4608 mddev->bitmap_info.chunksize = 0;
4609 mddev->bitmap_info.daemon_sleep = 0;
4610 mddev->bitmap_info.max_write_behind = 0;
4613 static void md_stop(mddev_t *mddev)
4615 mddev->pers->stop(mddev);
4616 if (mddev->pers->sync_request && mddev->to_remove == NULL)
4617 mddev->to_remove = &md_redundancy_group;
4618 module_put(mddev->pers->owner);
4624 * 0 - completely stop and dis-assemble array
4625 * 1 - switch to readonly
4626 * 2 - stop but do not disassemble array
4628 static int do_md_stop(mddev_t * mddev, int mode, int is_open)
4631 struct gendisk *disk = mddev->gendisk;
4634 mutex_lock(&mddev->open_mutex);
4635 if (atomic_read(&mddev->openers) > is_open) {
4636 printk("md: %s still in use.\n",mdname(mddev));
4638 } else if (mddev->pers) {
4640 if (mddev->sync_thread) {
4641 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4642 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
4643 md_unregister_thread(mddev->sync_thread);
4644 mddev->sync_thread = NULL;
4647 del_timer_sync(&mddev->safemode_timer);
4650 case 1: /* readonly */
4656 case 0: /* disassemble */
4658 bitmap_flush(mddev);
4659 md_super_wait(mddev);
4661 set_disk_ro(disk, 0);
4664 mddev->queue->merge_bvec_fn = NULL;
4665 mddev->queue->unplug_fn = NULL;
4666 mddev->queue->backing_dev_info.congested_fn = NULL;
4668 /* tell userspace to handle 'inactive' */
4669 sysfs_notify_dirent(mddev->sysfs_state);
4671 list_for_each_entry(rdev, &mddev->disks, same_set)
4672 if (rdev->raid_disk >= 0) {
4674 sprintf(nm, "rd%d", rdev->raid_disk);
4675 sysfs_remove_link(&mddev->kobj, nm);
4678 set_capacity(disk, 0);
4679 revalidate_disk(disk);
4684 if (!mddev->in_sync || mddev->flags) {
4685 /* mark array as shutdown cleanly */
4687 md_update_sb(mddev, 1);
4690 set_disk_ro(disk, 1);
4691 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4695 mutex_unlock(&mddev->open_mutex);
4699 * Free resources if final stop
4703 printk(KERN_INFO "md: %s stopped.\n", mdname(mddev));
4705 bitmap_destroy(mddev);
4706 if (mddev->bitmap_info.file) {
4707 restore_bitmap_write_access(mddev->bitmap_info.file);
4708 fput(mddev->bitmap_info.file);
4709 mddev->bitmap_info.file = NULL;
4711 mddev->bitmap_info.offset = 0;
4713 export_array(mddev);
4716 kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
4717 if (mddev->hold_active == UNTIL_STOP)
4718 mddev->hold_active = 0;
4720 } else if (mddev->pers)
4721 printk(KERN_INFO "md: %s switched to read-only mode.\n",
4724 blk_integrity_unregister(disk);
4725 md_new_event(mddev);
4726 sysfs_notify_dirent(mddev->sysfs_state);
4731 static void autorun_array(mddev_t *mddev)
4736 if (list_empty(&mddev->disks))
4739 printk(KERN_INFO "md: running: ");
4741 list_for_each_entry(rdev, &mddev->disks, same_set) {
4742 char b[BDEVNAME_SIZE];
4743 printk("<%s>", bdevname(rdev->bdev,b));
4747 err = do_md_run(mddev);
4749 printk(KERN_WARNING "md: do_md_run() returned %d\n", err);
4750 do_md_stop(mddev, 0, 0);
4755 * lets try to run arrays based on all disks that have arrived
4756 * until now. (those are in pending_raid_disks)
4758 * the method: pick the first pending disk, collect all disks with
4759 * the same UUID, remove all from the pending list and put them into
4760 * the 'same_array' list. Then order this list based on superblock
4761 * update time (freshest comes first), kick out 'old' disks and
4762 * compare superblocks. If everything's fine then run it.
4764 * If "unit" is allocated, then bump its reference count
4766 static void autorun_devices(int part)
4768 mdk_rdev_t *rdev0, *rdev, *tmp;
4770 char b[BDEVNAME_SIZE];
4772 printk(KERN_INFO "md: autorun ...\n");
4773 while (!list_empty(&pending_raid_disks)) {
4776 LIST_HEAD(candidates);
4777 rdev0 = list_entry(pending_raid_disks.next,
4778 mdk_rdev_t, same_set);
4780 printk(KERN_INFO "md: considering %s ...\n",
4781 bdevname(rdev0->bdev,b));
4782 INIT_LIST_HEAD(&candidates);
4783 rdev_for_each_list(rdev, tmp, &pending_raid_disks)
4784 if (super_90_load(rdev, rdev0, 0) >= 0) {
4785 printk(KERN_INFO "md: adding %s ...\n",
4786 bdevname(rdev->bdev,b));
4787 list_move(&rdev->same_set, &candidates);
4790 * now we have a set of devices, with all of them having
4791 * mostly sane superblocks. It's time to allocate the
4795 dev = MKDEV(mdp_major,
4796 rdev0->preferred_minor << MdpMinorShift);
4797 unit = MINOR(dev) >> MdpMinorShift;
4799 dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
4802 if (rdev0->preferred_minor != unit) {
4803 printk(KERN_INFO "md: unit number in %s is bad: %d\n",
4804 bdevname(rdev0->bdev, b), rdev0->preferred_minor);
4808 md_probe(dev, NULL, NULL);
4809 mddev = mddev_find(dev);
4810 if (!mddev || !mddev->gendisk) {
4814 "md: cannot allocate memory for md drive.\n");
4817 if (mddev_lock(mddev))
4818 printk(KERN_WARNING "md: %s locked, cannot run\n",
4820 else if (mddev->raid_disks || mddev->major_version
4821 || !list_empty(&mddev->disks)) {
4823 "md: %s already running, cannot run %s\n",
4824 mdname(mddev), bdevname(rdev0->bdev,b));
4825 mddev_unlock(mddev);
4827 printk(KERN_INFO "md: created %s\n", mdname(mddev));
4828 mddev->persistent = 1;
4829 rdev_for_each_list(rdev, tmp, &candidates) {
4830 list_del_init(&rdev->same_set);
4831 if (bind_rdev_to_array(rdev, mddev))
4834 autorun_array(mddev);
4835 mddev_unlock(mddev);
4837 /* on success, candidates will be empty, on error
4840 rdev_for_each_list(rdev, tmp, &candidates) {
4841 list_del_init(&rdev->same_set);
4846 printk(KERN_INFO "md: ... autorun DONE.\n");
4848 #endif /* !MODULE */
4850 static int get_version(void __user * arg)
4854 ver.major = MD_MAJOR_VERSION;
4855 ver.minor = MD_MINOR_VERSION;
4856 ver.patchlevel = MD_PATCHLEVEL_VERSION;
4858 if (copy_to_user(arg, &ver, sizeof(ver)))
4864 static int get_array_info(mddev_t * mddev, void __user * arg)
4866 mdu_array_info_t info;
4867 int nr,working,insync,failed,spare;
4870 nr=working=insync=failed=spare=0;
4871 list_for_each_entry(rdev, &mddev->disks, same_set) {
4873 if (test_bit(Faulty, &rdev->flags))
4877 if (test_bit(In_sync, &rdev->flags))
4884 info.major_version = mddev->major_version;
4885 info.minor_version = mddev->minor_version;
4886 info.patch_version = MD_PATCHLEVEL_VERSION;
4887 info.ctime = mddev->ctime;
4888 info.level = mddev->level;
4889 info.size = mddev->dev_sectors / 2;
4890 if (info.size != mddev->dev_sectors / 2) /* overflow */
4893 info.raid_disks = mddev->raid_disks;
4894 info.md_minor = mddev->md_minor;
4895 info.not_persistent= !mddev->persistent;
4897 info.utime = mddev->utime;
4900 info.state = (1<<MD_SB_CLEAN);
4901 if (mddev->bitmap && mddev->bitmap_info.offset)
4902 info.state = (1<<MD_SB_BITMAP_PRESENT);
4903 info.active_disks = insync;
4904 info.working_disks = working;
4905 info.failed_disks = failed;
4906 info.spare_disks = spare;
4908 info.layout = mddev->layout;
4909 info.chunk_size = mddev->chunk_sectors << 9;
4911 if (copy_to_user(arg, &info, sizeof(info)))
4917 static int get_bitmap_file(mddev_t * mddev, void __user * arg)
4919 mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */
4920 char *ptr, *buf = NULL;
4923 if (md_allow_write(mddev))
4924 file = kmalloc(sizeof(*file), GFP_NOIO);
4926 file = kmalloc(sizeof(*file), GFP_KERNEL);
4931 /* bitmap disabled, zero the first byte and copy out */
4932 if (!mddev->bitmap || !mddev->bitmap->file) {
4933 file->pathname[0] = '\0';
4937 buf = kmalloc(sizeof(file->pathname), GFP_KERNEL);
4941 ptr = d_path(&mddev->bitmap->file->f_path, buf, sizeof(file->pathname));
4945 strcpy(file->pathname, ptr);
4949 if (copy_to_user(arg, file, sizeof(*file)))
4957 static int get_disk_info(mddev_t * mddev, void __user * arg)
4959 mdu_disk_info_t info;
4962 if (copy_from_user(&info, arg, sizeof(info)))
4965 rdev = find_rdev_nr(mddev, info.number);
4967 info.major = MAJOR(rdev->bdev->bd_dev);
4968 info.minor = MINOR(rdev->bdev->bd_dev);
4969 info.raid_disk = rdev->raid_disk;
4971 if (test_bit(Faulty, &rdev->flags))
4972 info.state |= (1<<MD_DISK_FAULTY);
4973 else if (test_bit(In_sync, &rdev->flags)) {
4974 info.state |= (1<<MD_DISK_ACTIVE);
4975 info.state |= (1<<MD_DISK_SYNC);
4977 if (test_bit(WriteMostly, &rdev->flags))
4978 info.state |= (1<<MD_DISK_WRITEMOSTLY);
4980 info.major = info.minor = 0;
4981 info.raid_disk = -1;
4982 info.state = (1<<MD_DISK_REMOVED);
4985 if (copy_to_user(arg, &info, sizeof(info)))
4991 static int add_new_disk(mddev_t * mddev, mdu_disk_info_t *info)
4993 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
4995 dev_t dev = MKDEV(info->major,info->minor);
4997 if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
5000 if (!mddev->raid_disks) {
5002 /* expecting a device which has a superblock */
5003 rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
5006 "md: md_import_device returned %ld\n",
5008 return PTR_ERR(rdev);
5010 if (!list_empty(&mddev->disks)) {
5011 mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
5012 mdk_rdev_t, same_set);
5013 err = super_types[mddev->major_version]
5014 .load_super(rdev, rdev0, mddev->minor_version);
5017 "md: %s has different UUID to %s\n",
5018 bdevname(rdev->bdev,b),
5019 bdevname(rdev0->bdev,b2));
5024 err = bind_rdev_to_array(rdev, mddev);
5031 * add_new_disk can be used once the array is assembled
5032 * to add "hot spares". They must already have a superblock
5037 if (!mddev->pers->hot_add_disk) {
5039 "%s: personality does not support diskops!\n",
5043 if (mddev->persistent)
5044 rdev = md_import_device(dev, mddev->major_version,
5045 mddev->minor_version);
5047 rdev = md_import_device(dev, -1, -1);
5050 "md: md_import_device returned %ld\n",
5052 return PTR_ERR(rdev);
5054 /* set save_raid_disk if appropriate */
5055 if (!mddev->persistent) {
5056 if (info->state & (1<<MD_DISK_SYNC) &&
5057 info->raid_disk < mddev->raid_disks)
5058 rdev->raid_disk = info->raid_disk;
5060 rdev->raid_disk = -1;
5062 super_types[mddev->major_version].
5063 validate_super(mddev, rdev);
5064 rdev->saved_raid_disk = rdev->raid_disk;
5066 clear_bit(In_sync, &rdev->flags); /* just to be sure */
5067 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
5068 set_bit(WriteMostly, &rdev->flags);
5070 clear_bit(WriteMostly, &rdev->flags);
5072 rdev->raid_disk = -1;
5073 err = bind_rdev_to_array(rdev, mddev);
5074 if (!err && !mddev->pers->hot_remove_disk) {
5075 /* If there is hot_add_disk but no hot_remove_disk
5076 * then added disks for geometry changes,
5077 * and should be added immediately.
5079 super_types[mddev->major_version].
5080 validate_super(mddev, rdev);
5081 err = mddev->pers->hot_add_disk(mddev, rdev);
5083 unbind_rdev_from_array(rdev);
5088 sysfs_notify_dirent(rdev->sysfs_state);
5090 md_update_sb(mddev, 1);
5091 if (mddev->degraded)
5092 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
5093 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5094 md_wakeup_thread(mddev->thread);
5098 /* otherwise, add_new_disk is only allowed
5099 * for major_version==0 superblocks
5101 if (mddev->major_version != 0) {
5102 printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n",
5107 if (!(info->state & (1<<MD_DISK_FAULTY))) {
5109 rdev = md_import_device(dev, -1, 0);
5112 "md: error, md_import_device() returned %ld\n",
5114 return PTR_ERR(rdev);
5116 rdev->desc_nr = info->number;
5117 if (info->raid_disk < mddev->raid_disks)
5118 rdev->raid_disk = info->raid_disk;
5120 rdev->raid_disk = -1;
5122 if (rdev->raid_disk < mddev->raid_disks)
5123 if (info->state & (1<<MD_DISK_SYNC))
5124 set_bit(In_sync, &rdev->flags);
5126 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
5127 set_bit(WriteMostly, &rdev->flags);
5129 if (!mddev->persistent) {
5130 printk(KERN_INFO "md: nonpersistent superblock ...\n");
5131 rdev->sb_start = rdev->bdev->bd_inode->i_size / 512;
5133 rdev->sb_start = calc_dev_sboffset(rdev->bdev);
5134 rdev->sectors = rdev->sb_start;
5136 err = bind_rdev_to_array(rdev, mddev);
5146 static int hot_remove_disk(mddev_t * mddev, dev_t dev)
5148 char b[BDEVNAME_SIZE];
5151 rdev = find_rdev(mddev, dev);
5155 if (rdev->raid_disk >= 0)
5158 kick_rdev_from_array(rdev);
5159 md_update_sb(mddev, 1);
5160 md_new_event(mddev);
5164 printk(KERN_WARNING "md: cannot remove active disk %s from %s ...\n",
5165 bdevname(rdev->bdev,b), mdname(mddev));
5169 static int hot_add_disk(mddev_t * mddev, dev_t dev)
5171 char b[BDEVNAME_SIZE];
5178 if (mddev->major_version != 0) {
5179 printk(KERN_WARNING "%s: HOT_ADD may only be used with"
5180 " version-0 superblocks.\n",
5184 if (!mddev->pers->hot_add_disk) {
5186 "%s: personality does not support diskops!\n",
5191 rdev = md_import_device(dev, -1, 0);
5194 "md: error, md_import_device() returned %ld\n",
5199 if (mddev->persistent)
5200 rdev->sb_start = calc_dev_sboffset(rdev->bdev);
5202 rdev->sb_start = rdev->bdev->bd_inode->i_size / 512;
5204 rdev->sectors = rdev->sb_start;
5206 if (test_bit(Faulty, &rdev->flags)) {
5208 "md: can not hot-add faulty %s disk to %s!\n",
5209 bdevname(rdev->bdev,b), mdname(mddev));
5213 clear_bit(In_sync, &rdev->flags);
5215 rdev->saved_raid_disk = -1;
5216 err = bind_rdev_to_array(rdev, mddev);
5221 * The rest should better be atomic, we can have disk failures
5222 * noticed in interrupt contexts ...
5225 rdev->raid_disk = -1;
5227 md_update_sb(mddev, 1);
5230 * Kick recovery, maybe this spare has to be added to the
5231 * array immediately.
5233 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5234 md_wakeup_thread(mddev->thread);
5235 md_new_event(mddev);
5243 static int set_bitmap_file(mddev_t *mddev, int fd)
5248 if (!mddev->pers->quiesce)
5250 if (mddev->recovery || mddev->sync_thread)
5252 /* we should be able to change the bitmap.. */
5258 return -EEXIST; /* cannot add when bitmap is present */
5259 mddev->bitmap_info.file = fget(fd);
5261 if (mddev->bitmap_info.file == NULL) {
5262 printk(KERN_ERR "%s: error: failed to get bitmap file\n",
5267 err = deny_bitmap_write_access(mddev->bitmap_info.file);
5269 printk(KERN_ERR "%s: error: bitmap file is already in use\n",
5271 fput(mddev->bitmap_info.file);
5272 mddev->bitmap_info.file = NULL;
5275 mddev->bitmap_info.offset = 0; /* file overrides offset */
5276 } else if (mddev->bitmap == NULL)
5277 return -ENOENT; /* cannot remove what isn't there */
5280 mddev->pers->quiesce(mddev, 1);
5282 err = bitmap_create(mddev);
5283 if (fd < 0 || err) {
5284 bitmap_destroy(mddev);
5285 fd = -1; /* make sure to put the file */
5287 mddev->pers->quiesce(mddev, 0);
5290 if (mddev->bitmap_info.file) {
5291 restore_bitmap_write_access(mddev->bitmap_info.file);
5292 fput(mddev->bitmap_info.file);
5294 mddev->bitmap_info.file = NULL;
5301 * set_array_info is used two different ways
5302 * The original usage is when creating a new array.
5303 * In this usage, raid_disks is > 0 and it together with
5304 * level, size, not_persistent,layout,chunksize determine the
5305 * shape of the array.
5306 * This will always create an array with a type-0.90.0 superblock.
5307 * The newer usage is when assembling an array.
5308 * In this case raid_disks will be 0, and the major_version field is
5309 * use to determine which style super-blocks are to be found on the devices.
5310 * The minor and patch _version numbers are also kept incase the
5311 * super_block handler wishes to interpret them.
5313 static int set_array_info(mddev_t * mddev, mdu_array_info_t *info)
5316 if (info->raid_disks == 0) {
5317 /* just setting version number for superblock loading */
5318 if (info->major_version < 0 ||
5319 info->major_version >= ARRAY_SIZE(super_types) ||
5320 super_types[info->major_version].name == NULL) {
5321 /* maybe try to auto-load a module? */
5323 "md: superblock version %d not known\n",
5324 info->major_version);
5327 mddev->major_version = info->major_version;
5328 mddev->minor_version = info->minor_version;
5329 mddev->patch_version = info->patch_version;
5330 mddev->persistent = !info->not_persistent;
5331 /* ensure mddev_put doesn't delete this now that there
5332 * is some minimal configuration.
5334 mddev->ctime = get_seconds();
5337 mddev->major_version = MD_MAJOR_VERSION;
5338 mddev->minor_version = MD_MINOR_VERSION;
5339 mddev->patch_version = MD_PATCHLEVEL_VERSION;
5340 mddev->ctime = get_seconds();
5342 mddev->level = info->level;
5343 mddev->clevel[0] = 0;
5344 mddev->dev_sectors = 2 * (sector_t)info->size;
5345 mddev->raid_disks = info->raid_disks;
5346 /* don't set md_minor, it is determined by which /dev/md* was
5349 if (info->state & (1<<MD_SB_CLEAN))
5350 mddev->recovery_cp = MaxSector;
5352 mddev->recovery_cp = 0;
5353 mddev->persistent = ! info->not_persistent;
5354 mddev->external = 0;
5356 mddev->layout = info->layout;
5357 mddev->chunk_sectors = info->chunk_size >> 9;
5359 mddev->max_disks = MD_SB_DISKS;
5361 if (mddev->persistent)
5363 set_bit(MD_CHANGE_DEVS, &mddev->flags);
5365 mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9;
5366 mddev->bitmap_info.offset = 0;
5368 mddev->reshape_position = MaxSector;
5371 * Generate a 128 bit UUID
5373 get_random_bytes(mddev->uuid, 16);
5375 mddev->new_level = mddev->level;
5376 mddev->new_chunk_sectors = mddev->chunk_sectors;
5377 mddev->new_layout = mddev->layout;
5378 mddev->delta_disks = 0;
5383 void md_set_array_sectors(mddev_t *mddev, sector_t array_sectors)
5385 WARN(!mddev_is_locked(mddev), "%s: unlocked mddev!\n", __func__);
5387 if (mddev->external_size)
5390 mddev->array_sectors = array_sectors;
5392 EXPORT_SYMBOL(md_set_array_sectors);
5394 static int update_size(mddev_t *mddev, sector_t num_sectors)
5398 int fit = (num_sectors == 0);
5400 if (mddev->pers->resize == NULL)
5402 /* The "num_sectors" is the number of sectors of each device that
5403 * is used. This can only make sense for arrays with redundancy.
5404 * linear and raid0 always use whatever space is available. We can only
5405 * consider changing this number if no resync or reconstruction is
5406 * happening, and if the new size is acceptable. It must fit before the
5407 * sb_start or, if that is <data_offset, it must fit before the size
5408 * of each device. If num_sectors is zero, we find the largest size
5412 if (mddev->sync_thread)
5415 /* Sorry, cannot grow a bitmap yet, just remove it,
5419 list_for_each_entry(rdev, &mddev->disks, same_set) {
5420 sector_t avail = rdev->sectors;
5422 if (fit && (num_sectors == 0 || num_sectors > avail))
5423 num_sectors = avail;
5424 if (avail < num_sectors)
5427 rv = mddev->pers->resize(mddev, num_sectors);
5429 revalidate_disk(mddev->gendisk);
5433 static int update_raid_disks(mddev_t *mddev, int raid_disks)
5436 /* change the number of raid disks */
5437 if (mddev->pers->check_reshape == NULL)
5439 if (raid_disks <= 0 ||
5440 (mddev->max_disks && raid_disks >= mddev->max_disks))
5442 if (mddev->sync_thread || mddev->reshape_position != MaxSector)
5444 mddev->delta_disks = raid_disks - mddev->raid_disks;
5446 rv = mddev->pers->check_reshape(mddev);
5452 * update_array_info is used to change the configuration of an
5454 * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
5455 * fields in the info are checked against the array.
5456 * Any differences that cannot be handled will cause an error.
5457 * Normally, only one change can be managed at a time.
5459 static int update_array_info(mddev_t *mddev, mdu_array_info_t *info)
5465 /* calculate expected state,ignoring low bits */
5466 if (mddev->bitmap && mddev->bitmap_info.offset)
5467 state |= (1 << MD_SB_BITMAP_PRESENT);
5469 if (mddev->major_version != info->major_version ||
5470 mddev->minor_version != info->minor_version ||
5471 /* mddev->patch_version != info->patch_version || */
5472 mddev->ctime != info->ctime ||
5473 mddev->level != info->level ||
5474 /* mddev->layout != info->layout || */
5475 !mddev->persistent != info->not_persistent||
5476 mddev->chunk_sectors != info->chunk_size >> 9 ||
5477 /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
5478 ((state^info->state) & 0xfffffe00)
5481 /* Check there is only one change */
5482 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
5484 if (mddev->raid_disks != info->raid_disks)
5486 if (mddev->layout != info->layout)
5488 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT))
5495 if (mddev->layout != info->layout) {
5497 * we don't need to do anything at the md level, the
5498 * personality will take care of it all.
5500 if (mddev->pers->check_reshape == NULL)
5503 mddev->new_layout = info->layout;
5504 rv = mddev->pers->check_reshape(mddev);
5506 mddev->new_layout = mddev->layout;
5510 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
5511 rv = update_size(mddev, (sector_t)info->size * 2);
5513 if (mddev->raid_disks != info->raid_disks)
5514 rv = update_raid_disks(mddev, info->raid_disks);
5516 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) {
5517 if (mddev->pers->quiesce == NULL)
5519 if (mddev->recovery || mddev->sync_thread)
5521 if (info->state & (1<<MD_SB_BITMAP_PRESENT)) {
5522 /* add the bitmap */
5525 if (mddev->bitmap_info.default_offset == 0)
5527 mddev->bitmap_info.offset =
5528 mddev->bitmap_info.default_offset;
5529 mddev->pers->quiesce(mddev, 1);
5530 rv = bitmap_create(mddev);
5532 bitmap_destroy(mddev);
5533 mddev->pers->quiesce(mddev, 0);
5535 /* remove the bitmap */
5538 if (mddev->bitmap->file)
5540 mddev->pers->quiesce(mddev, 1);
5541 bitmap_destroy(mddev);
5542 mddev->pers->quiesce(mddev, 0);
5543 mddev->bitmap_info.offset = 0;
5546 md_update_sb(mddev, 1);
5550 static int set_disk_faulty(mddev_t *mddev, dev_t dev)
5554 if (mddev->pers == NULL)
5557 rdev = find_rdev(mddev, dev);
5561 md_error(mddev, rdev);
5566 * We have a problem here : there is no easy way to give a CHS
5567 * virtual geometry. We currently pretend that we have a 2 heads
5568 * 4 sectors (with a BIG number of cylinders...). This drives
5569 * dosfs just mad... ;-)
5571 static int md_getgeo(struct block_device *bdev, struct hd_geometry *geo)
5573 mddev_t *mddev = bdev->bd_disk->private_data;
5577 geo->cylinders = mddev->array_sectors / 8;
5581 static int md_ioctl(struct block_device *bdev, fmode_t mode,
5582 unsigned int cmd, unsigned long arg)
5585 void __user *argp = (void __user *)arg;
5586 mddev_t *mddev = NULL;
5589 if (!capable(CAP_SYS_ADMIN))
5593 * Commands dealing with the RAID driver but not any
5599 err = get_version(argp);
5602 case PRINT_RAID_DEBUG:
5610 autostart_arrays(arg);
5617 * Commands creating/starting a new array:
5620 mddev = bdev->bd_disk->private_data;
5627 err = mddev_lock(mddev);
5630 "md: ioctl lock interrupted, reason %d, cmd %d\n",
5637 case SET_ARRAY_INFO:
5639 mdu_array_info_t info;
5641 memset(&info, 0, sizeof(info));
5642 else if (copy_from_user(&info, argp, sizeof(info))) {
5647 err = update_array_info(mddev, &info);
5649 printk(KERN_WARNING "md: couldn't update"
5650 " array info. %d\n", err);
5655 if (!list_empty(&mddev->disks)) {
5657 "md: array %s already has disks!\n",
5662 if (mddev->raid_disks) {
5664 "md: array %s already initialised!\n",
5669 err = set_array_info(mddev, &info);
5671 printk(KERN_WARNING "md: couldn't set"
5672 " array info. %d\n", err);
5682 * Commands querying/configuring an existing array:
5684 /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
5685 * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */
5686 if ((!mddev->raid_disks && !mddev->external)
5687 && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY
5688 && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE
5689 && cmd != GET_BITMAP_FILE) {
5695 * Commands even a read-only array can execute:
5699 case GET_ARRAY_INFO:
5700 err = get_array_info(mddev, argp);
5703 case GET_BITMAP_FILE:
5704 err = get_bitmap_file(mddev, argp);
5708 err = get_disk_info(mddev, argp);
5711 case RESTART_ARRAY_RW:
5712 err = restart_array(mddev);
5716 err = do_md_stop(mddev, 0, 1);
5720 err = do_md_stop(mddev, 1, 1);
5724 if (get_user(ro, (int __user *)(arg))) {
5730 /* if the bdev is going readonly the value of mddev->ro
5731 * does not matter, no writes are coming
5736 /* are we are already prepared for writes? */
5740 /* transitioning to readauto need only happen for
5741 * arrays that call md_write_start
5744 err = restart_array(mddev);
5747 set_disk_ro(mddev->gendisk, 0);
5754 * The remaining ioctls are changing the state of the
5755 * superblock, so we do not allow them on read-only arrays.
5756 * However non-MD ioctls (e.g. get-size) will still come through
5757 * here and hit the 'default' below, so only disallow
5758 * 'md' ioctls, and switch to rw mode if started auto-readonly.
5760 if (_IOC_TYPE(cmd) == MD_MAJOR && mddev->ro && mddev->pers) {
5761 if (mddev->ro == 2) {
5763 sysfs_notify_dirent(mddev->sysfs_state);
5764 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5765 md_wakeup_thread(mddev->thread);
5776 mdu_disk_info_t info;
5777 if (copy_from_user(&info, argp, sizeof(info)))
5780 err = add_new_disk(mddev, &info);
5784 case HOT_REMOVE_DISK:
5785 err = hot_remove_disk(mddev, new_decode_dev(arg));
5789 err = hot_add_disk(mddev, new_decode_dev(arg));
5792 case SET_DISK_FAULTY:
5793 err = set_disk_faulty(mddev, new_decode_dev(arg));
5797 err = do_md_run(mddev);
5800 case SET_BITMAP_FILE:
5801 err = set_bitmap_file(mddev, (int)arg);
5811 if (mddev->hold_active == UNTIL_IOCTL &&
5813 mddev->hold_active = 0;
5814 mddev_unlock(mddev);
5823 #ifdef CONFIG_COMPAT
5824 static int md_compat_ioctl(struct block_device *bdev, fmode_t mode,
5825 unsigned int cmd, unsigned long arg)
5828 case HOT_REMOVE_DISK:
5830 case SET_DISK_FAULTY:
5831 case SET_BITMAP_FILE:
5832 /* These take in integer arg, do not convert */
5835 arg = (unsigned long)compat_ptr(arg);
5839 return md_ioctl(bdev, mode, cmd, arg);
5841 #endif /* CONFIG_COMPAT */
5843 static int md_open(struct block_device *bdev, fmode_t mode)
5846 * Succeed if we can lock the mddev, which confirms that
5847 * it isn't being stopped right now.
5849 mddev_t *mddev = mddev_find(bdev->bd_dev);
5852 if (mddev->gendisk != bdev->bd_disk) {
5853 /* we are racing with mddev_put which is discarding this
5857 /* Wait until bdev->bd_disk is definitely gone */
5858 flush_scheduled_work();
5859 /* Then retry the open from the top */
5860 return -ERESTARTSYS;
5862 BUG_ON(mddev != bdev->bd_disk->private_data);
5864 if ((err = mutex_lock_interruptible(&mddev->open_mutex)))
5868 atomic_inc(&mddev->openers);
5869 mutex_unlock(&mddev->open_mutex);
5875 static int md_release(struct gendisk *disk, fmode_t mode)
5877 mddev_t *mddev = disk->private_data;
5880 atomic_dec(&mddev->openers);
5885 static const struct block_device_operations md_fops =
5887 .owner = THIS_MODULE,
5889 .release = md_release,
5891 #ifdef CONFIG_COMPAT
5892 .compat_ioctl = md_compat_ioctl,
5894 .getgeo = md_getgeo,
5897 static int md_thread(void * arg)
5899 mdk_thread_t *thread = arg;
5902 * md_thread is a 'system-thread', it's priority should be very
5903 * high. We avoid resource deadlocks individually in each
5904 * raid personality. (RAID5 does preallocation) We also use RR and
5905 * the very same RT priority as kswapd, thus we will never get
5906 * into a priority inversion deadlock.
5908 * we definitely have to have equal or higher priority than
5909 * bdflush, otherwise bdflush will deadlock if there are too
5910 * many dirty RAID5 blocks.
5913 allow_signal(SIGKILL);
5914 while (!kthread_should_stop()) {
5916 /* We need to wait INTERRUPTIBLE so that
5917 * we don't add to the load-average.
5918 * That means we need to be sure no signals are
5921 if (signal_pending(current))
5922 flush_signals(current);
5924 wait_event_interruptible_timeout
5926 test_bit(THREAD_WAKEUP, &thread->flags)
5927 || kthread_should_stop(),
5930 clear_bit(THREAD_WAKEUP, &thread->flags);
5932 thread->run(thread->mddev);
5938 void md_wakeup_thread(mdk_thread_t *thread)
5941 dprintk("md: waking up MD thread %s.\n", thread->tsk->comm);
5942 set_bit(THREAD_WAKEUP, &thread->flags);
5943 wake_up(&thread->wqueue);
5947 mdk_thread_t *md_register_thread(void (*run) (mddev_t *), mddev_t *mddev,
5950 mdk_thread_t *thread;
5952 thread = kzalloc(sizeof(mdk_thread_t), GFP_KERNEL);
5956 init_waitqueue_head(&thread->wqueue);
5959 thread->mddev = mddev;
5960 thread->timeout = MAX_SCHEDULE_TIMEOUT;
5961 thread->tsk = kthread_run(md_thread, thread,
5963 mdname(thread->mddev),
5964 name ?: mddev->pers->name);
5965 if (IS_ERR(thread->tsk)) {
5972 void md_unregister_thread(mdk_thread_t *thread)
5976 dprintk("interrupting MD-thread pid %d\n", task_pid_nr(thread->tsk));
5978 kthread_stop(thread->tsk);
5982 void md_error(mddev_t *mddev, mdk_rdev_t *rdev)
5989 if (!rdev || test_bit(Faulty, &rdev->flags))
5992 if (mddev->external)
5993 set_bit(Blocked, &rdev->flags);
5995 dprintk("md_error dev:%s, rdev:(%d:%d), (caller: %p,%p,%p,%p).\n",
5997 MAJOR(rdev->bdev->bd_dev), MINOR(rdev->bdev->bd_dev),
5998 __builtin_return_address(0),__builtin_return_address(1),
5999 __builtin_return_address(2),__builtin_return_address(3));
6003 if (!mddev->pers->error_handler)
6005 mddev->pers->error_handler(mddev,rdev);
6006 if (mddev->degraded)
6007 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
6008 sysfs_notify_dirent(rdev->sysfs_state);
6009 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6010 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6011 md_wakeup_thread(mddev->thread);
6012 md_new_event_inintr(mddev);
6015 /* seq_file implementation /proc/mdstat */
6017 static void status_unused(struct seq_file *seq)
6022 seq_printf(seq, "unused devices: ");
6024 list_for_each_entry(rdev, &pending_raid_disks, same_set) {
6025 char b[BDEVNAME_SIZE];
6027 seq_printf(seq, "%s ",
6028 bdevname(rdev->bdev,b));
6031 seq_printf(seq, "<none>");
6033 seq_printf(seq, "\n");
6037 static void status_resync(struct seq_file *seq, mddev_t * mddev)
6039 sector_t max_sectors, resync, res;
6040 unsigned long dt, db;
6043 unsigned int per_milli;
6045 resync = mddev->curr_resync - atomic_read(&mddev->recovery_active);
6047 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
6048 max_sectors = mddev->resync_max_sectors;
6050 max_sectors = mddev->dev_sectors;
6053 * Should not happen.
6059 /* Pick 'scale' such that (resync>>scale)*1000 will fit
6060 * in a sector_t, and (max_sectors>>scale) will fit in a
6061 * u32, as those are the requirements for sector_div.
6062 * Thus 'scale' must be at least 10
6065 if (sizeof(sector_t) > sizeof(unsigned long)) {
6066 while ( max_sectors/2 > (1ULL<<(scale+32)))
6069 res = (resync>>scale)*1000;
6070 sector_div(res, (u32)((max_sectors>>scale)+1));
6074 int i, x = per_milli/50, y = 20-x;
6075 seq_printf(seq, "[");
6076 for (i = 0; i < x; i++)
6077 seq_printf(seq, "=");
6078 seq_printf(seq, ">");
6079 for (i = 0; i < y; i++)
6080 seq_printf(seq, ".");
6081 seq_printf(seq, "] ");
6083 seq_printf(seq, " %s =%3u.%u%% (%llu/%llu)",
6084 (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)?
6086 (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)?
6088 (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
6089 "resync" : "recovery"))),
6090 per_milli/10, per_milli % 10,
6091 (unsigned long long) resync/2,
6092 (unsigned long long) max_sectors/2);
6095 * dt: time from mark until now
6096 * db: blocks written from mark until now
6097 * rt: remaining time
6099 * rt is a sector_t, so could be 32bit or 64bit.
6100 * So we divide before multiply in case it is 32bit and close
6102 * We scale the divisor (db) by 32 to avoid loosing precision
6103 * near the end of resync when the number of remaining sectors
6105 * We then divide rt by 32 after multiplying by db to compensate.
6106 * The '+1' avoids division by zero if db is very small.
6108 dt = ((jiffies - mddev->resync_mark) / HZ);
6110 db = (mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active))
6111 - mddev->resync_mark_cnt;
6113 rt = max_sectors - resync; /* number of remaining sectors */
6114 sector_div(rt, db/32+1);
6118 seq_printf(seq, " finish=%lu.%lumin", (unsigned long)rt / 60,
6119 ((unsigned long)rt % 60)/6);
6121 seq_printf(seq, " speed=%ldK/sec", db/2/dt);
6124 static void *md_seq_start(struct seq_file *seq, loff_t *pos)
6126 struct list_head *tmp;
6136 spin_lock(&all_mddevs_lock);
6137 list_for_each(tmp,&all_mddevs)
6139 mddev = list_entry(tmp, mddev_t, all_mddevs);
6141 spin_unlock(&all_mddevs_lock);
6144 spin_unlock(&all_mddevs_lock);
6146 return (void*)2;/* tail */
6150 static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
6152 struct list_head *tmp;
6153 mddev_t *next_mddev, *mddev = v;
6159 spin_lock(&all_mddevs_lock);
6161 tmp = all_mddevs.next;
6163 tmp = mddev->all_mddevs.next;
6164 if (tmp != &all_mddevs)
6165 next_mddev = mddev_get(list_entry(tmp,mddev_t,all_mddevs));
6167 next_mddev = (void*)2;
6170 spin_unlock(&all_mddevs_lock);
6178 static void md_seq_stop(struct seq_file *seq, void *v)
6182 if (mddev && v != (void*)1 && v != (void*)2)
6186 struct mdstat_info {
6190 static int md_seq_show(struct seq_file *seq, void *v)
6195 struct mdstat_info *mi = seq->private;
6196 struct bitmap *bitmap;
6198 if (v == (void*)1) {
6199 struct mdk_personality *pers;
6200 seq_printf(seq, "Personalities : ");
6201 spin_lock(&pers_lock);
6202 list_for_each_entry(pers, &pers_list, list)
6203 seq_printf(seq, "[%s] ", pers->name);
6205 spin_unlock(&pers_lock);
6206 seq_printf(seq, "\n");
6207 mi->event = atomic_read(&md_event_count);
6210 if (v == (void*)2) {
6215 if (mddev_lock(mddev) < 0)
6218 if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
6219 seq_printf(seq, "%s : %sactive", mdname(mddev),
6220 mddev->pers ? "" : "in");
6223 seq_printf(seq, " (read-only)");
6225 seq_printf(seq, " (auto-read-only)");
6226 seq_printf(seq, " %s", mddev->pers->name);
6230 list_for_each_entry(rdev, &mddev->disks, same_set) {
6231 char b[BDEVNAME_SIZE];
6232 seq_printf(seq, " %s[%d]",
6233 bdevname(rdev->bdev,b), rdev->desc_nr);
6234 if (test_bit(WriteMostly, &rdev->flags))
6235 seq_printf(seq, "(W)");
6236 if (test_bit(Faulty, &rdev->flags)) {
6237 seq_printf(seq, "(F)");
6239 } else if (rdev->raid_disk < 0)
6240 seq_printf(seq, "(S)"); /* spare */
6241 sectors += rdev->sectors;
6244 if (!list_empty(&mddev->disks)) {
6246 seq_printf(seq, "\n %llu blocks",
6247 (unsigned long long)
6248 mddev->array_sectors / 2);
6250 seq_printf(seq, "\n %llu blocks",
6251 (unsigned long long)sectors / 2);
6253 if (mddev->persistent) {
6254 if (mddev->major_version != 0 ||
6255 mddev->minor_version != 90) {
6256 seq_printf(seq," super %d.%d",
6257 mddev->major_version,
6258 mddev->minor_version);
6260 } else if (mddev->external)
6261 seq_printf(seq, " super external:%s",
6262 mddev->metadata_type);
6264 seq_printf(seq, " super non-persistent");
6267 mddev->pers->status(seq, mddev);
6268 seq_printf(seq, "\n ");
6269 if (mddev->pers->sync_request) {
6270 if (mddev->curr_resync > 2) {
6271 status_resync(seq, mddev);
6272 seq_printf(seq, "\n ");
6273 } else if (mddev->curr_resync == 1 || mddev->curr_resync == 2)
6274 seq_printf(seq, "\tresync=DELAYED\n ");
6275 else if (mddev->recovery_cp < MaxSector)
6276 seq_printf(seq, "\tresync=PENDING\n ");
6279 seq_printf(seq, "\n ");
6281 if ((bitmap = mddev->bitmap)) {
6282 unsigned long chunk_kb;
6283 unsigned long flags;
6284 spin_lock_irqsave(&bitmap->lock, flags);
6285 chunk_kb = mddev->bitmap_info.chunksize >> 10;
6286 seq_printf(seq, "bitmap: %lu/%lu pages [%luKB], "
6288 bitmap->pages - bitmap->missing_pages,
6290 (bitmap->pages - bitmap->missing_pages)
6291 << (PAGE_SHIFT - 10),
6292 chunk_kb ? chunk_kb : mddev->bitmap_info.chunksize,
6293 chunk_kb ? "KB" : "B");
6295 seq_printf(seq, ", file: ");
6296 seq_path(seq, &bitmap->file->f_path, " \t\n");
6299 seq_printf(seq, "\n");
6300 spin_unlock_irqrestore(&bitmap->lock, flags);
6303 seq_printf(seq, "\n");
6305 mddev_unlock(mddev);
6310 static const struct seq_operations md_seq_ops = {
6311 .start = md_seq_start,
6312 .next = md_seq_next,
6313 .stop = md_seq_stop,
6314 .show = md_seq_show,
6317 static int md_seq_open(struct inode *inode, struct file *file)
6320 struct mdstat_info *mi = kmalloc(sizeof(*mi), GFP_KERNEL);
6324 error = seq_open(file, &md_seq_ops);
6328 struct seq_file *p = file->private_data;
6330 mi->event = atomic_read(&md_event_count);
6335 static unsigned int mdstat_poll(struct file *filp, poll_table *wait)
6337 struct seq_file *m = filp->private_data;
6338 struct mdstat_info *mi = m->private;
6341 poll_wait(filp, &md_event_waiters, wait);
6343 /* always allow read */
6344 mask = POLLIN | POLLRDNORM;
6346 if (mi->event != atomic_read(&md_event_count))
6347 mask |= POLLERR | POLLPRI;
6351 static const struct file_operations md_seq_fops = {
6352 .owner = THIS_MODULE,
6353 .open = md_seq_open,
6355 .llseek = seq_lseek,
6356 .release = seq_release_private,
6357 .poll = mdstat_poll,
6360 int register_md_personality(struct mdk_personality *p)
6362 spin_lock(&pers_lock);
6363 list_add_tail(&p->list, &pers_list);
6364 printk(KERN_INFO "md: %s personality registered for level %d\n", p->name, p->level);
6365 spin_unlock(&pers_lock);
6369 int unregister_md_personality(struct mdk_personality *p)
6371 printk(KERN_INFO "md: %s personality unregistered\n", p->name);
6372 spin_lock(&pers_lock);
6373 list_del_init(&p->list);
6374 spin_unlock(&pers_lock);
6378 static int is_mddev_idle(mddev_t *mddev, int init)
6386 rdev_for_each_rcu(rdev, mddev) {
6387 struct gendisk *disk = rdev->bdev->bd_contains->bd_disk;
6388 curr_events = (int)part_stat_read(&disk->part0, sectors[0]) +
6389 (int)part_stat_read(&disk->part0, sectors[1]) -
6390 atomic_read(&disk->sync_io);
6391 /* sync IO will cause sync_io to increase before the disk_stats
6392 * as sync_io is counted when a request starts, and
6393 * disk_stats is counted when it completes.
6394 * So resync activity will cause curr_events to be smaller than
6395 * when there was no such activity.
6396 * non-sync IO will cause disk_stat to increase without
6397 * increasing sync_io so curr_events will (eventually)
6398 * be larger than it was before. Once it becomes
6399 * substantially larger, the test below will cause
6400 * the array to appear non-idle, and resync will slow
6402 * If there is a lot of outstanding resync activity when
6403 * we set last_event to curr_events, then all that activity
6404 * completing might cause the array to appear non-idle
6405 * and resync will be slowed down even though there might
6406 * not have been non-resync activity. This will only
6407 * happen once though. 'last_events' will soon reflect
6408 * the state where there is little or no outstanding
6409 * resync requests, and further resync activity will
6410 * always make curr_events less than last_events.
6413 if (init || curr_events - rdev->last_events > 64) {
6414 rdev->last_events = curr_events;
6422 void md_done_sync(mddev_t *mddev, int blocks, int ok)
6424 /* another "blocks" (512byte) blocks have been synced */
6425 atomic_sub(blocks, &mddev->recovery_active);
6426 wake_up(&mddev->recovery_wait);
6428 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6429 md_wakeup_thread(mddev->thread);
6430 // stop recovery, signal do_sync ....
6435 /* md_write_start(mddev, bi)
6436 * If we need to update some array metadata (e.g. 'active' flag
6437 * in superblock) before writing, schedule a superblock update
6438 * and wait for it to complete.
6440 void md_write_start(mddev_t *mddev, struct bio *bi)
6443 if (bio_data_dir(bi) != WRITE)
6446 BUG_ON(mddev->ro == 1);
6447 if (mddev->ro == 2) {
6448 /* need to switch to read/write */
6450 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6451 md_wakeup_thread(mddev->thread);
6452 md_wakeup_thread(mddev->sync_thread);
6455 atomic_inc(&mddev->writes_pending);
6456 if (mddev->safemode == 1)
6457 mddev->safemode = 0;
6458 if (mddev->in_sync) {
6459 spin_lock_irq(&mddev->write_lock);
6460 if (mddev->in_sync) {
6462 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
6463 md_wakeup_thread(mddev->thread);
6466 spin_unlock_irq(&mddev->write_lock);
6469 sysfs_notify_dirent(mddev->sysfs_state);
6470 wait_event(mddev->sb_wait,
6471 !test_bit(MD_CHANGE_CLEAN, &mddev->flags) &&
6472 !test_bit(MD_CHANGE_PENDING, &mddev->flags));
6475 void md_write_end(mddev_t *mddev)
6477 if (atomic_dec_and_test(&mddev->writes_pending)) {
6478 if (mddev->safemode == 2)
6479 md_wakeup_thread(mddev->thread);
6480 else if (mddev->safemode_delay)
6481 mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay);
6485 /* md_allow_write(mddev)
6486 * Calling this ensures that the array is marked 'active' so that writes
6487 * may proceed without blocking. It is important to call this before
6488 * attempting a GFP_KERNEL allocation while holding the mddev lock.
6489 * Must be called with mddev_lock held.
6491 * In the ->external case MD_CHANGE_CLEAN can not be cleared until mddev->lock
6492 * is dropped, so return -EAGAIN after notifying userspace.
6494 int md_allow_write(mddev_t *mddev)
6500 if (!mddev->pers->sync_request)
6503 spin_lock_irq(&mddev->write_lock);
6504 if (mddev->in_sync) {
6506 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
6507 if (mddev->safemode_delay &&
6508 mddev->safemode == 0)
6509 mddev->safemode = 1;
6510 spin_unlock_irq(&mddev->write_lock);
6511 md_update_sb(mddev, 0);
6512 sysfs_notify_dirent(mddev->sysfs_state);
6514 spin_unlock_irq(&mddev->write_lock);
6516 if (test_bit(MD_CHANGE_CLEAN, &mddev->flags))
6521 EXPORT_SYMBOL_GPL(md_allow_write);
6523 #define SYNC_MARKS 10
6524 #define SYNC_MARK_STEP (3*HZ)
6525 void md_do_sync(mddev_t *mddev)
6528 unsigned int currspeed = 0,
6530 sector_t max_sectors,j, io_sectors;
6531 unsigned long mark[SYNC_MARKS];
6532 sector_t mark_cnt[SYNC_MARKS];
6534 struct list_head *tmp;
6535 sector_t last_check;
6540 /* just incase thread restarts... */
6541 if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
6543 if (mddev->ro) /* never try to sync a read-only array */
6546 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
6547 if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
6548 desc = "data-check";
6549 else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
6550 desc = "requested-resync";
6553 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
6558 /* we overload curr_resync somewhat here.
6559 * 0 == not engaged in resync at all
6560 * 2 == checking that there is no conflict with another sync
6561 * 1 == like 2, but have yielded to allow conflicting resync to
6563 * other == active in resync - this many blocks
6565 * Before starting a resync we must have set curr_resync to
6566 * 2, and then checked that every "conflicting" array has curr_resync
6567 * less than ours. When we find one that is the same or higher
6568 * we wait on resync_wait. To avoid deadlock, we reduce curr_resync
6569 * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
6570 * This will mean we have to start checking from the beginning again.
6575 mddev->curr_resync = 2;
6578 if (kthread_should_stop())
6579 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6581 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
6583 for_each_mddev(mddev2, tmp) {
6584 if (mddev2 == mddev)
6586 if (!mddev->parallel_resync
6587 && mddev2->curr_resync
6588 && match_mddev_units(mddev, mddev2)) {
6590 if (mddev < mddev2 && mddev->curr_resync == 2) {
6591 /* arbitrarily yield */
6592 mddev->curr_resync = 1;
6593 wake_up(&resync_wait);
6595 if (mddev > mddev2 && mddev->curr_resync == 1)
6596 /* no need to wait here, we can wait the next
6597 * time 'round when curr_resync == 2
6600 /* We need to wait 'interruptible' so as not to
6601 * contribute to the load average, and not to
6602 * be caught by 'softlockup'
6604 prepare_to_wait(&resync_wait, &wq, TASK_INTERRUPTIBLE);
6605 if (!kthread_should_stop() &&
6606 mddev2->curr_resync >= mddev->curr_resync) {
6607 printk(KERN_INFO "md: delaying %s of %s"
6608 " until %s has finished (they"
6609 " share one or more physical units)\n",
6610 desc, mdname(mddev), mdname(mddev2));
6612 if (signal_pending(current))
6613 flush_signals(current);
6615 finish_wait(&resync_wait, &wq);
6618 finish_wait(&resync_wait, &wq);
6621 } while (mddev->curr_resync < 2);
6624 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
6625 /* resync follows the size requested by the personality,
6626 * which defaults to physical size, but can be virtual size
6628 max_sectors = mddev->resync_max_sectors;
6629 mddev->resync_mismatches = 0;
6630 /* we don't use the checkpoint if there's a bitmap */
6631 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
6632 j = mddev->resync_min;
6633 else if (!mddev->bitmap)
6634 j = mddev->recovery_cp;
6636 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
6637 max_sectors = mddev->dev_sectors;
6639 /* recovery follows the physical size of devices */
6640 max_sectors = mddev->dev_sectors;
6643 list_for_each_entry_rcu(rdev, &mddev->disks, same_set)
6644 if (rdev->raid_disk >= 0 &&
6645 !test_bit(Faulty, &rdev->flags) &&
6646 !test_bit(In_sync, &rdev->flags) &&
6647 rdev->recovery_offset < j)
6648 j = rdev->recovery_offset;
6652 printk(KERN_INFO "md: %s of RAID array %s\n", desc, mdname(mddev));
6653 printk(KERN_INFO "md: minimum _guaranteed_ speed:"
6654 " %d KB/sec/disk.\n", speed_min(mddev));
6655 printk(KERN_INFO "md: using maximum available idle IO bandwidth "
6656 "(but not more than %d KB/sec) for %s.\n",
6657 speed_max(mddev), desc);
6659 is_mddev_idle(mddev, 1); /* this initializes IO event counters */
6662 for (m = 0; m < SYNC_MARKS; m++) {
6664 mark_cnt[m] = io_sectors;
6667 mddev->resync_mark = mark[last_mark];
6668 mddev->resync_mark_cnt = mark_cnt[last_mark];
6671 * Tune reconstruction:
6673 window = 32*(PAGE_SIZE/512);
6674 printk(KERN_INFO "md: using %dk window, over a total of %llu blocks.\n",
6675 window/2,(unsigned long long) max_sectors/2);
6677 atomic_set(&mddev->recovery_active, 0);
6682 "md: resuming %s of %s from checkpoint.\n",
6683 desc, mdname(mddev));
6684 mddev->curr_resync = j;
6686 mddev->curr_resync_completed = mddev->curr_resync;
6688 while (j < max_sectors) {
6693 if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
6694 ((mddev->curr_resync > mddev->curr_resync_completed &&
6695 (mddev->curr_resync - mddev->curr_resync_completed)
6696 > (max_sectors >> 4)) ||
6697 (j - mddev->curr_resync_completed)*2
6698 >= mddev->resync_max - mddev->curr_resync_completed
6700 /* time to update curr_resync_completed */
6701 blk_unplug(mddev->queue);
6702 wait_event(mddev->recovery_wait,
6703 atomic_read(&mddev->recovery_active) == 0);
6704 mddev->curr_resync_completed =
6706 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
6707 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
6710 while (j >= mddev->resync_max && !kthread_should_stop()) {
6711 /* As this condition is controlled by user-space,
6712 * we can block indefinitely, so use '_interruptible'
6713 * to avoid triggering warnings.
6715 flush_signals(current); /* just in case */
6716 wait_event_interruptible(mddev->recovery_wait,
6717 mddev->resync_max > j
6718 || kthread_should_stop());
6721 if (kthread_should_stop())
6724 sectors = mddev->pers->sync_request(mddev, j, &skipped,
6725 currspeed < speed_min(mddev));
6727 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6731 if (!skipped) { /* actual IO requested */
6732 io_sectors += sectors;
6733 atomic_add(sectors, &mddev->recovery_active);
6737 if (j>1) mddev->curr_resync = j;
6738 mddev->curr_mark_cnt = io_sectors;
6739 if (last_check == 0)
6740 /* this is the earliers that rebuilt will be
6741 * visible in /proc/mdstat
6743 md_new_event(mddev);
6745 if (last_check + window > io_sectors || j == max_sectors)
6748 last_check = io_sectors;
6750 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
6754 if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) {
6756 int next = (last_mark+1) % SYNC_MARKS;
6758 mddev->resync_mark = mark[next];
6759 mddev->resync_mark_cnt = mark_cnt[next];
6760 mark[next] = jiffies;
6761 mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active);
6766 if (kthread_should_stop())
6771 * this loop exits only if either when we are slower than
6772 * the 'hard' speed limit, or the system was IO-idle for
6774 * the system might be non-idle CPU-wise, but we only care
6775 * about not overloading the IO subsystem. (things like an
6776 * e2fsck being done on the RAID array should execute fast)
6778 blk_unplug(mddev->queue);
6781 currspeed = ((unsigned long)(io_sectors-mddev->resync_mark_cnt))/2
6782 /((jiffies-mddev->resync_mark)/HZ +1) +1;
6784 if (currspeed > speed_min(mddev)) {
6785 if ((currspeed > speed_max(mddev)) ||
6786 !is_mddev_idle(mddev, 0)) {
6792 printk(KERN_INFO "md: %s: %s done.\n",mdname(mddev), desc);
6794 * this also signals 'finished resyncing' to md_stop
6797 blk_unplug(mddev->queue);
6799 wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
6801 /* tell personality that we are finished */
6802 mddev->pers->sync_request(mddev, max_sectors, &skipped, 1);
6804 if (!test_bit(MD_RECOVERY_CHECK, &mddev->recovery) &&
6805 mddev->curr_resync > 2) {
6806 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
6807 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
6808 if (mddev->curr_resync >= mddev->recovery_cp) {
6810 "md: checkpointing %s of %s.\n",
6811 desc, mdname(mddev));
6812 mddev->recovery_cp = mddev->curr_resync;
6815 mddev->recovery_cp = MaxSector;
6817 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery))
6818 mddev->curr_resync = MaxSector;
6820 list_for_each_entry_rcu(rdev, &mddev->disks, same_set)
6821 if (rdev->raid_disk >= 0 &&
6822 !test_bit(Faulty, &rdev->flags) &&
6823 !test_bit(In_sync, &rdev->flags) &&
6824 rdev->recovery_offset < mddev->curr_resync)
6825 rdev->recovery_offset = mddev->curr_resync;
6829 set_bit(MD_CHANGE_DEVS, &mddev->flags);
6832 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
6833 /* We completed so min/max setting can be forgotten if used. */
6834 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
6835 mddev->resync_min = 0;
6836 mddev->resync_max = MaxSector;
6837 } else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
6838 mddev->resync_min = mddev->curr_resync_completed;
6839 mddev->curr_resync = 0;
6840 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery))
6841 mddev->curr_resync_completed = 0;
6842 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
6843 wake_up(&resync_wait);
6844 set_bit(MD_RECOVERY_DONE, &mddev->recovery);
6845 md_wakeup_thread(mddev->thread);
6850 * got a signal, exit.
6853 "md: md_do_sync() got signal ... exiting\n");
6854 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6858 EXPORT_SYMBOL_GPL(md_do_sync);
6861 static int remove_and_add_spares(mddev_t *mddev)
6866 mddev->curr_resync_completed = 0;
6868 list_for_each_entry(rdev, &mddev->disks, same_set)
6869 if (rdev->raid_disk >= 0 &&
6870 !test_bit(Blocked, &rdev->flags) &&
6871 (test_bit(Faulty, &rdev->flags) ||
6872 ! test_bit(In_sync, &rdev->flags)) &&
6873 atomic_read(&rdev->nr_pending)==0) {
6874 if (mddev->pers->hot_remove_disk(
6875 mddev, rdev->raid_disk)==0) {
6877 sprintf(nm,"rd%d", rdev->raid_disk);
6878 sysfs_remove_link(&mddev->kobj, nm);
6879 rdev->raid_disk = -1;
6883 if (mddev->degraded && ! mddev->ro && !mddev->recovery_disabled) {
6884 list_for_each_entry(rdev, &mddev->disks, same_set) {
6885 if (rdev->raid_disk >= 0 &&
6886 !test_bit(In_sync, &rdev->flags) &&
6887 !test_bit(Blocked, &rdev->flags))
6889 if (rdev->raid_disk < 0
6890 && !test_bit(Faulty, &rdev->flags)) {
6891 rdev->recovery_offset = 0;
6893 hot_add_disk(mddev, rdev) == 0) {
6895 sprintf(nm, "rd%d", rdev->raid_disk);
6896 if (sysfs_create_link(&mddev->kobj,
6899 "md: cannot register "
6903 md_new_event(mddev);
6904 set_bit(MD_CHANGE_DEVS, &mddev->flags);
6913 * This routine is regularly called by all per-raid-array threads to
6914 * deal with generic issues like resync and super-block update.
6915 * Raid personalities that don't have a thread (linear/raid0) do not
6916 * need this as they never do any recovery or update the superblock.
6918 * It does not do any resync itself, but rather "forks" off other threads
6919 * to do that as needed.
6920 * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
6921 * "->recovery" and create a thread at ->sync_thread.
6922 * When the thread finishes it sets MD_RECOVERY_DONE
6923 * and wakeups up this thread which will reap the thread and finish up.
6924 * This thread also removes any faulty devices (with nr_pending == 0).
6926 * The overall approach is:
6927 * 1/ if the superblock needs updating, update it.
6928 * 2/ If a recovery thread is running, don't do anything else.
6929 * 3/ If recovery has finished, clean up, possibly marking spares active.
6930 * 4/ If there are any faulty devices, remove them.
6931 * 5/ If array is degraded, try to add spares devices
6932 * 6/ If array has spares or is not in-sync, start a resync thread.
6934 void md_check_recovery(mddev_t *mddev)
6940 bitmap_daemon_work(mddev);
6945 if (signal_pending(current)) {
6946 if (mddev->pers->sync_request && !mddev->external) {
6947 printk(KERN_INFO "md: %s in immediate safe mode\n",
6949 mddev->safemode = 2;
6951 flush_signals(current);
6954 if (mddev->ro && !test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
6957 (mddev->flags && !mddev->external) ||
6958 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
6959 test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
6960 (mddev->external == 0 && mddev->safemode == 1) ||
6961 (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending)
6962 && !mddev->in_sync && mddev->recovery_cp == MaxSector)
6966 if (mddev_trylock(mddev)) {
6970 /* Only thing we do on a ro array is remove
6973 remove_and_add_spares(mddev);
6974 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6978 if (!mddev->external) {
6980 spin_lock_irq(&mddev->write_lock);
6981 if (mddev->safemode &&
6982 !atomic_read(&mddev->writes_pending) &&
6984 mddev->recovery_cp == MaxSector) {
6987 if (mddev->persistent)
6988 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
6990 if (mddev->safemode == 1)
6991 mddev->safemode = 0;
6992 spin_unlock_irq(&mddev->write_lock);
6994 sysfs_notify_dirent(mddev->sysfs_state);
6998 md_update_sb(mddev, 0);
7000 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
7001 !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
7002 /* resync/recovery still happening */
7003 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7006 if (mddev->sync_thread) {
7007 /* resync has finished, collect result */
7008 md_unregister_thread(mddev->sync_thread);
7009 mddev->sync_thread = NULL;
7010 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
7011 !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
7013 /* activate any spares */
7014 if (mddev->pers->spare_active(mddev))
7015 sysfs_notify(&mddev->kobj, NULL,
7018 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
7019 mddev->pers->finish_reshape)
7020 mddev->pers->finish_reshape(mddev);
7021 md_update_sb(mddev, 1);
7023 /* if array is no-longer degraded, then any saved_raid_disk
7024 * information must be scrapped
7026 if (!mddev->degraded)
7027 list_for_each_entry(rdev, &mddev->disks, same_set)
7028 rdev->saved_raid_disk = -1;
7030 mddev->recovery = 0;
7031 /* flag recovery needed just to double check */
7032 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7033 sysfs_notify_dirent(mddev->sysfs_action);
7034 md_new_event(mddev);
7037 /* Set RUNNING before clearing NEEDED to avoid
7038 * any transients in the value of "sync_action".
7040 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
7041 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7042 /* Clear some bits that don't mean anything, but
7045 clear_bit(MD_RECOVERY_INTR, &mddev->recovery);
7046 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
7048 if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
7050 /* no recovery is running.
7051 * remove any failed drives, then
7052 * add spares if possible.
7053 * Spare are also removed and re-added, to allow
7054 * the personality to fail the re-add.
7057 if (mddev->reshape_position != MaxSector) {
7058 if (mddev->pers->check_reshape == NULL ||
7059 mddev->pers->check_reshape(mddev) != 0)
7060 /* Cannot proceed */
7062 set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
7063 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
7064 } else if ((spares = remove_and_add_spares(mddev))) {
7065 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
7066 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
7067 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
7068 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
7069 } else if (mddev->recovery_cp < MaxSector) {
7070 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
7071 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
7072 } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
7073 /* nothing to be done ... */
7076 if (mddev->pers->sync_request) {
7077 if (spares && mddev->bitmap && ! mddev->bitmap->file) {
7078 /* We are adding a device or devices to an array
7079 * which has the bitmap stored on all devices.
7080 * So make sure all bitmap pages get written
7082 bitmap_write_all(mddev->bitmap);
7084 mddev->sync_thread = md_register_thread(md_do_sync,
7087 if (!mddev->sync_thread) {
7088 printk(KERN_ERR "%s: could not start resync"
7091 /* leave the spares where they are, it shouldn't hurt */
7092 mddev->recovery = 0;
7094 md_wakeup_thread(mddev->sync_thread);
7095 sysfs_notify_dirent(mddev->sysfs_action);
7096 md_new_event(mddev);
7099 if (!mddev->sync_thread) {
7100 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
7101 if (test_and_clear_bit(MD_RECOVERY_RECOVER,
7103 if (mddev->sysfs_action)
7104 sysfs_notify_dirent(mddev->sysfs_action);
7106 mddev_unlock(mddev);
7110 void md_wait_for_blocked_rdev(mdk_rdev_t *rdev, mddev_t *mddev)
7112 sysfs_notify_dirent(rdev->sysfs_state);
7113 wait_event_timeout(rdev->blocked_wait,
7114 !test_bit(Blocked, &rdev->flags),
7115 msecs_to_jiffies(5000));
7116 rdev_dec_pending(rdev, mddev);
7118 EXPORT_SYMBOL(md_wait_for_blocked_rdev);
7120 static int md_notify_reboot(struct notifier_block *this,
7121 unsigned long code, void *x)
7123 struct list_head *tmp;
7126 if ((code == SYS_DOWN) || (code == SYS_HALT) || (code == SYS_POWER_OFF)) {
7128 printk(KERN_INFO "md: stopping all md devices.\n");
7130 for_each_mddev(mddev, tmp)
7131 if (mddev_trylock(mddev)) {
7132 /* Force a switch to readonly even array
7133 * appears to still be in use. Hence
7136 do_md_stop(mddev, 1, 100);
7137 mddev_unlock(mddev);
7140 * certain more exotic SCSI devices are known to be
7141 * volatile wrt too early system reboots. While the
7142 * right place to handle this issue is the given
7143 * driver, we do want to have a safe RAID driver ...
7150 static struct notifier_block md_notifier = {
7151 .notifier_call = md_notify_reboot,
7153 .priority = INT_MAX, /* before any real devices */
7156 static void md_geninit(void)
7158 dprintk("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
7160 proc_create("mdstat", S_IRUGO, NULL, &md_seq_fops);
7163 static int __init md_init(void)
7165 if (register_blkdev(MD_MAJOR, "md"))
7167 if ((mdp_major=register_blkdev(0, "mdp"))<=0) {
7168 unregister_blkdev(MD_MAJOR, "md");
7171 blk_register_region(MKDEV(MD_MAJOR, 0), 1UL<<MINORBITS, THIS_MODULE,
7172 md_probe, NULL, NULL);
7173 blk_register_region(MKDEV(mdp_major, 0), 1UL<<MINORBITS, THIS_MODULE,
7174 md_probe, NULL, NULL);
7176 register_reboot_notifier(&md_notifier);
7177 raid_table_header = register_sysctl_table(raid_root_table);
7187 * Searches all registered partitions for autorun RAID arrays
7191 static LIST_HEAD(all_detected_devices);
7192 struct detected_devices_node {
7193 struct list_head list;
7197 void md_autodetect_dev(dev_t dev)
7199 struct detected_devices_node *node_detected_dev;
7201 node_detected_dev = kzalloc(sizeof(*node_detected_dev), GFP_KERNEL);
7202 if (node_detected_dev) {
7203 node_detected_dev->dev = dev;
7204 list_add_tail(&node_detected_dev->list, &all_detected_devices);
7206 printk(KERN_CRIT "md: md_autodetect_dev: kzalloc failed"
7207 ", skipping dev(%d,%d)\n", MAJOR(dev), MINOR(dev));
7212 static void autostart_arrays(int part)
7215 struct detected_devices_node *node_detected_dev;
7217 int i_scanned, i_passed;
7222 printk(KERN_INFO "md: Autodetecting RAID arrays.\n");
7224 while (!list_empty(&all_detected_devices) && i_scanned < INT_MAX) {
7226 node_detected_dev = list_entry(all_detected_devices.next,
7227 struct detected_devices_node, list);
7228 list_del(&node_detected_dev->list);
7229 dev = node_detected_dev->dev;
7230 kfree(node_detected_dev);
7231 rdev = md_import_device(dev,0, 90);
7235 if (test_bit(Faulty, &rdev->flags)) {
7239 set_bit(AutoDetected, &rdev->flags);
7240 list_add(&rdev->same_set, &pending_raid_disks);
7244 printk(KERN_INFO "md: Scanned %d and added %d devices.\n",
7245 i_scanned, i_passed);
7247 autorun_devices(part);
7250 #endif /* !MODULE */
7252 static __exit void md_exit(void)
7255 struct list_head *tmp;
7257 blk_unregister_region(MKDEV(MD_MAJOR,0), 1U << MINORBITS);
7258 blk_unregister_region(MKDEV(mdp_major,0), 1U << MINORBITS);
7260 unregister_blkdev(MD_MAJOR,"md");
7261 unregister_blkdev(mdp_major, "mdp");
7262 unregister_reboot_notifier(&md_notifier);
7263 unregister_sysctl_table(raid_table_header);
7264 remove_proc_entry("mdstat", NULL);
7265 for_each_mddev(mddev, tmp) {
7266 export_array(mddev);
7267 mddev->hold_active = 0;
7271 subsys_initcall(md_init);
7272 module_exit(md_exit)
7274 static int get_ro(char *buffer, struct kernel_param *kp)
7276 return sprintf(buffer, "%d", start_readonly);
7278 static int set_ro(const char *val, struct kernel_param *kp)
7281 int num = simple_strtoul(val, &e, 10);
7282 if (*val && (*e == '\0' || *e == '\n')) {
7283 start_readonly = num;
7289 module_param_call(start_ro, set_ro, get_ro, NULL, S_IRUSR|S_IWUSR);
7290 module_param(start_dirty_degraded, int, S_IRUGO|S_IWUSR);
7292 module_param_call(new_array, add_named_array, NULL, NULL, S_IWUSR);
7294 EXPORT_SYMBOL(register_md_personality);
7295 EXPORT_SYMBOL(unregister_md_personality);
7296 EXPORT_SYMBOL(md_error);
7297 EXPORT_SYMBOL(md_done_sync);
7298 EXPORT_SYMBOL(md_write_start);
7299 EXPORT_SYMBOL(md_write_end);
7300 EXPORT_SYMBOL(md_register_thread);
7301 EXPORT_SYMBOL(md_unregister_thread);
7302 EXPORT_SYMBOL(md_wakeup_thread);
7303 EXPORT_SYMBOL(md_check_recovery);
7304 MODULE_LICENSE("GPL");
7305 MODULE_DESCRIPTION("MD RAID framework");
7307 MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR);