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(mddev, 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 /* mddev_suspend makes sure no new requests are submitted
258 * to the device, and that any requests that have been submitted
259 * are completely handled.
260 * Once ->stop is called and completes, the module will be completely
263 static void mddev_suspend(mddev_t *mddev)
265 BUG_ON(mddev->suspended);
266 mddev->suspended = 1;
268 wait_event(mddev->sb_wait, atomic_read(&mddev->active_io) == 0);
269 mddev->pers->quiesce(mddev, 1);
272 static void mddev_resume(mddev_t *mddev)
274 mddev->suspended = 0;
275 wake_up(&mddev->sb_wait);
276 mddev->pers->quiesce(mddev, 0);
279 int mddev_congested(mddev_t *mddev, int bits)
283 return mddev->suspended;
285 EXPORT_SYMBOL(mddev_congested);
288 * Generic barrier handling for md
291 #define POST_REQUEST_BARRIER ((void*)1)
293 static void md_end_barrier(struct bio *bio, int err)
295 mdk_rdev_t *rdev = bio->bi_private;
296 mddev_t *mddev = rdev->mddev;
297 if (err == -EOPNOTSUPP && mddev->barrier != POST_REQUEST_BARRIER)
298 set_bit(BIO_EOPNOTSUPP, &mddev->barrier->bi_flags);
300 rdev_dec_pending(rdev, mddev);
302 if (atomic_dec_and_test(&mddev->flush_pending)) {
303 if (mddev->barrier == POST_REQUEST_BARRIER) {
304 /* This was a post-request barrier */
305 mddev->barrier = NULL;
306 wake_up(&mddev->sb_wait);
308 /* The pre-request barrier has finished */
309 schedule_work(&mddev->barrier_work);
314 static void submit_barriers(mddev_t *mddev)
319 list_for_each_entry_rcu(rdev, &mddev->disks, same_set)
320 if (rdev->raid_disk >= 0 &&
321 !test_bit(Faulty, &rdev->flags)) {
322 /* Take two references, one is dropped
323 * when request finishes, one after
324 * we reclaim rcu_read_lock
327 atomic_inc(&rdev->nr_pending);
328 atomic_inc(&rdev->nr_pending);
330 bi = bio_alloc(GFP_KERNEL, 0);
331 bi->bi_end_io = md_end_barrier;
332 bi->bi_private = rdev;
333 bi->bi_bdev = rdev->bdev;
334 atomic_inc(&mddev->flush_pending);
335 submit_bio(WRITE_BARRIER, bi);
337 rdev_dec_pending(rdev, mddev);
342 static void md_submit_barrier(struct work_struct *ws)
344 mddev_t *mddev = container_of(ws, mddev_t, barrier_work);
345 struct bio *bio = mddev->barrier;
347 atomic_set(&mddev->flush_pending, 1);
349 if (test_bit(BIO_EOPNOTSUPP, &bio->bi_flags))
350 bio_endio(bio, -EOPNOTSUPP);
351 else if (bio->bi_size == 0)
352 /* an empty barrier - all done */
355 bio->bi_rw &= ~(1<<BIO_RW_BARRIER);
356 if (mddev->pers->make_request(mddev, bio))
357 generic_make_request(bio);
358 mddev->barrier = POST_REQUEST_BARRIER;
359 submit_barriers(mddev);
361 if (atomic_dec_and_test(&mddev->flush_pending)) {
362 mddev->barrier = NULL;
363 wake_up(&mddev->sb_wait);
367 void md_barrier_request(mddev_t *mddev, struct bio *bio)
369 spin_lock_irq(&mddev->write_lock);
370 wait_event_lock_irq(mddev->sb_wait,
372 mddev->write_lock, /*nothing*/);
373 mddev->barrier = bio;
374 spin_unlock_irq(&mddev->write_lock);
376 atomic_set(&mddev->flush_pending, 1);
377 INIT_WORK(&mddev->barrier_work, md_submit_barrier);
379 submit_barriers(mddev);
381 if (atomic_dec_and_test(&mddev->flush_pending))
382 schedule_work(&mddev->barrier_work);
384 EXPORT_SYMBOL(md_barrier_request);
386 static inline mddev_t *mddev_get(mddev_t *mddev)
388 atomic_inc(&mddev->active);
392 static void mddev_delayed_delete(struct work_struct *ws);
394 static void mddev_put(mddev_t *mddev)
396 if (!atomic_dec_and_lock(&mddev->active, &all_mddevs_lock))
398 if (!mddev->raid_disks && list_empty(&mddev->disks) &&
399 mddev->ctime == 0 && !mddev->hold_active) {
400 /* Array is not configured at all, and not held active,
402 list_del(&mddev->all_mddevs);
403 if (mddev->gendisk) {
404 /* we did a probe so need to clean up.
405 * Call schedule_work inside the spinlock
406 * so that flush_scheduled_work() after
407 * mddev_find will succeed in waiting for the
410 INIT_WORK(&mddev->del_work, mddev_delayed_delete);
411 schedule_work(&mddev->del_work);
415 spin_unlock(&all_mddevs_lock);
418 static void mddev_init(mddev_t *mddev)
420 mutex_init(&mddev->open_mutex);
421 mutex_init(&mddev->reconfig_mutex);
422 mutex_init(&mddev->bitmap_info.mutex);
423 INIT_LIST_HEAD(&mddev->disks);
424 INIT_LIST_HEAD(&mddev->all_mddevs);
425 init_timer(&mddev->safemode_timer);
426 atomic_set(&mddev->active, 1);
427 atomic_set(&mddev->openers, 0);
428 atomic_set(&mddev->active_io, 0);
429 spin_lock_init(&mddev->write_lock);
430 atomic_set(&mddev->flush_pending, 0);
431 init_waitqueue_head(&mddev->sb_wait);
432 init_waitqueue_head(&mddev->recovery_wait);
433 mddev->reshape_position = MaxSector;
434 mddev->resync_min = 0;
435 mddev->resync_max = MaxSector;
436 mddev->level = LEVEL_NONE;
439 static mddev_t * mddev_find(dev_t unit)
441 mddev_t *mddev, *new = NULL;
444 spin_lock(&all_mddevs_lock);
447 list_for_each_entry(mddev, &all_mddevs, all_mddevs)
448 if (mddev->unit == unit) {
450 spin_unlock(&all_mddevs_lock);
456 list_add(&new->all_mddevs, &all_mddevs);
457 spin_unlock(&all_mddevs_lock);
458 new->hold_active = UNTIL_IOCTL;
462 /* find an unused unit number */
463 static int next_minor = 512;
464 int start = next_minor;
468 dev = MKDEV(MD_MAJOR, next_minor);
470 if (next_minor > MINORMASK)
472 if (next_minor == start) {
473 /* Oh dear, all in use. */
474 spin_unlock(&all_mddevs_lock);
480 list_for_each_entry(mddev, &all_mddevs, all_mddevs)
481 if (mddev->unit == dev) {
487 new->md_minor = MINOR(dev);
488 new->hold_active = UNTIL_STOP;
489 list_add(&new->all_mddevs, &all_mddevs);
490 spin_unlock(&all_mddevs_lock);
493 spin_unlock(&all_mddevs_lock);
495 new = kzalloc(sizeof(*new), GFP_KERNEL);
500 if (MAJOR(unit) == MD_MAJOR)
501 new->md_minor = MINOR(unit);
503 new->md_minor = MINOR(unit) >> MdpMinorShift;
510 static inline int mddev_lock(mddev_t * mddev)
512 return mutex_lock_interruptible(&mddev->reconfig_mutex);
515 static inline int mddev_is_locked(mddev_t *mddev)
517 return mutex_is_locked(&mddev->reconfig_mutex);
520 static inline int mddev_trylock(mddev_t * mddev)
522 return mutex_trylock(&mddev->reconfig_mutex);
525 static struct attribute_group md_redundancy_group;
527 static void mddev_unlock(mddev_t * mddev)
529 if (mddev->to_remove) {
530 /* These cannot be removed under reconfig_mutex as
531 * an access to the files will try to take reconfig_mutex
532 * while holding the file unremovable, which leads to
534 * So hold open_mutex instead - we are allowed to take
535 * it while holding reconfig_mutex, and md_run can
536 * use it to wait for the remove to complete.
538 struct attribute_group *to_remove = mddev->to_remove;
539 mddev->to_remove = NULL;
540 mutex_lock(&mddev->open_mutex);
541 mutex_unlock(&mddev->reconfig_mutex);
543 if (to_remove != &md_redundancy_group)
544 sysfs_remove_group(&mddev->kobj, to_remove);
545 if (mddev->pers == NULL ||
546 mddev->pers->sync_request == NULL) {
547 sysfs_remove_group(&mddev->kobj, &md_redundancy_group);
548 if (mddev->sysfs_action)
549 sysfs_put(mddev->sysfs_action);
550 mddev->sysfs_action = NULL;
552 mutex_unlock(&mddev->open_mutex);
554 mutex_unlock(&mddev->reconfig_mutex);
556 md_wakeup_thread(mddev->thread);
559 static mdk_rdev_t * find_rdev_nr(mddev_t *mddev, int nr)
563 list_for_each_entry(rdev, &mddev->disks, same_set)
564 if (rdev->desc_nr == nr)
570 static mdk_rdev_t * find_rdev(mddev_t * mddev, dev_t dev)
574 list_for_each_entry(rdev, &mddev->disks, same_set)
575 if (rdev->bdev->bd_dev == dev)
581 static struct mdk_personality *find_pers(int level, char *clevel)
583 struct mdk_personality *pers;
584 list_for_each_entry(pers, &pers_list, list) {
585 if (level != LEVEL_NONE && pers->level == level)
587 if (strcmp(pers->name, clevel)==0)
593 /* return the offset of the super block in 512byte sectors */
594 static inline sector_t calc_dev_sboffset(struct block_device *bdev)
596 sector_t num_sectors = bdev->bd_inode->i_size / 512;
597 return MD_NEW_SIZE_SECTORS(num_sectors);
600 static int alloc_disk_sb(mdk_rdev_t * rdev)
605 rdev->sb_page = alloc_page(GFP_KERNEL);
606 if (!rdev->sb_page) {
607 printk(KERN_ALERT "md: out of memory.\n");
614 static void free_disk_sb(mdk_rdev_t * rdev)
617 put_page(rdev->sb_page);
619 rdev->sb_page = NULL;
626 static void super_written(struct bio *bio, int error)
628 mdk_rdev_t *rdev = bio->bi_private;
629 mddev_t *mddev = rdev->mddev;
631 if (error || !test_bit(BIO_UPTODATE, &bio->bi_flags)) {
632 printk("md: super_written gets error=%d, uptodate=%d\n",
633 error, test_bit(BIO_UPTODATE, &bio->bi_flags));
634 WARN_ON(test_bit(BIO_UPTODATE, &bio->bi_flags));
635 md_error(mddev, rdev);
638 if (atomic_dec_and_test(&mddev->pending_writes))
639 wake_up(&mddev->sb_wait);
643 static void super_written_barrier(struct bio *bio, int error)
645 struct bio *bio2 = bio->bi_private;
646 mdk_rdev_t *rdev = bio2->bi_private;
647 mddev_t *mddev = rdev->mddev;
649 if (!test_bit(BIO_UPTODATE, &bio->bi_flags) &&
650 error == -EOPNOTSUPP) {
652 /* barriers don't appear to be supported :-( */
653 set_bit(BarriersNotsupp, &rdev->flags);
654 mddev->barriers_work = 0;
655 spin_lock_irqsave(&mddev->write_lock, flags);
656 bio2->bi_next = mddev->biolist;
657 mddev->biolist = bio2;
658 spin_unlock_irqrestore(&mddev->write_lock, flags);
659 wake_up(&mddev->sb_wait);
663 bio->bi_private = rdev;
664 super_written(bio, error);
668 void md_super_write(mddev_t *mddev, mdk_rdev_t *rdev,
669 sector_t sector, int size, struct page *page)
671 /* write first size bytes of page to sector of rdev
672 * Increment mddev->pending_writes before returning
673 * and decrement it on completion, waking up sb_wait
674 * if zero is reached.
675 * If an error occurred, call md_error
677 * As we might need to resubmit the request if BIO_RW_BARRIER
678 * causes ENOTSUPP, we allocate a spare bio...
680 struct bio *bio = bio_alloc(GFP_NOIO, 1);
681 int rw = (1<<BIO_RW) | (1<<BIO_RW_SYNCIO) | (1<<BIO_RW_UNPLUG);
683 bio->bi_bdev = rdev->bdev;
684 bio->bi_sector = sector;
685 bio_add_page(bio, page, size, 0);
686 bio->bi_private = rdev;
687 bio->bi_end_io = super_written;
690 atomic_inc(&mddev->pending_writes);
691 if (!test_bit(BarriersNotsupp, &rdev->flags)) {
693 rw |= (1<<BIO_RW_BARRIER);
694 rbio = bio_clone(bio, GFP_NOIO);
695 rbio->bi_private = bio;
696 rbio->bi_end_io = super_written_barrier;
697 submit_bio(rw, rbio);
702 void md_super_wait(mddev_t *mddev)
704 /* wait for all superblock writes that were scheduled to complete.
705 * if any had to be retried (due to BARRIER problems), retry them
709 prepare_to_wait(&mddev->sb_wait, &wq, TASK_UNINTERRUPTIBLE);
710 if (atomic_read(&mddev->pending_writes)==0)
712 while (mddev->biolist) {
714 spin_lock_irq(&mddev->write_lock);
715 bio = mddev->biolist;
716 mddev->biolist = bio->bi_next ;
718 spin_unlock_irq(&mddev->write_lock);
719 submit_bio(bio->bi_rw, bio);
723 finish_wait(&mddev->sb_wait, &wq);
726 static void bi_complete(struct bio *bio, int error)
728 complete((struct completion*)bio->bi_private);
731 int sync_page_io(struct block_device *bdev, sector_t sector, int size,
732 struct page *page, int rw)
734 struct bio *bio = bio_alloc(GFP_NOIO, 1);
735 struct completion event;
738 rw |= (1 << BIO_RW_SYNCIO) | (1 << BIO_RW_UNPLUG);
741 bio->bi_sector = sector;
742 bio_add_page(bio, page, size, 0);
743 init_completion(&event);
744 bio->bi_private = &event;
745 bio->bi_end_io = bi_complete;
747 wait_for_completion(&event);
749 ret = test_bit(BIO_UPTODATE, &bio->bi_flags);
753 EXPORT_SYMBOL_GPL(sync_page_io);
755 static int read_disk_sb(mdk_rdev_t * rdev, int size)
757 char b[BDEVNAME_SIZE];
758 if (!rdev->sb_page) {
766 if (!sync_page_io(rdev->bdev, rdev->sb_start, size, rdev->sb_page, READ))
772 printk(KERN_WARNING "md: disabled device %s, could not read superblock.\n",
773 bdevname(rdev->bdev,b));
777 static int uuid_equal(mdp_super_t *sb1, mdp_super_t *sb2)
779 return sb1->set_uuid0 == sb2->set_uuid0 &&
780 sb1->set_uuid1 == sb2->set_uuid1 &&
781 sb1->set_uuid2 == sb2->set_uuid2 &&
782 sb1->set_uuid3 == sb2->set_uuid3;
785 static int sb_equal(mdp_super_t *sb1, mdp_super_t *sb2)
788 mdp_super_t *tmp1, *tmp2;
790 tmp1 = kmalloc(sizeof(*tmp1),GFP_KERNEL);
791 tmp2 = kmalloc(sizeof(*tmp2),GFP_KERNEL);
793 if (!tmp1 || !tmp2) {
795 printk(KERN_INFO "md.c sb_equal(): failed to allocate memory!\n");
803 * nr_disks is not constant
808 ret = (memcmp(tmp1, tmp2, MD_SB_GENERIC_CONSTANT_WORDS * 4) == 0);
816 static u32 md_csum_fold(u32 csum)
818 csum = (csum & 0xffff) + (csum >> 16);
819 return (csum & 0xffff) + (csum >> 16);
822 static unsigned int calc_sb_csum(mdp_super_t * sb)
825 u32 *sb32 = (u32*)sb;
827 unsigned int disk_csum, csum;
829 disk_csum = sb->sb_csum;
832 for (i = 0; i < MD_SB_BYTES/4 ; i++)
834 csum = (newcsum & 0xffffffff) + (newcsum>>32);
838 /* This used to use csum_partial, which was wrong for several
839 * reasons including that different results are returned on
840 * different architectures. It isn't critical that we get exactly
841 * the same return value as before (we always csum_fold before
842 * testing, and that removes any differences). However as we
843 * know that csum_partial always returned a 16bit value on
844 * alphas, do a fold to maximise conformity to previous behaviour.
846 sb->sb_csum = md_csum_fold(disk_csum);
848 sb->sb_csum = disk_csum;
855 * Handle superblock details.
856 * We want to be able to handle multiple superblock formats
857 * so we have a common interface to them all, and an array of
858 * different handlers.
859 * We rely on user-space to write the initial superblock, and support
860 * reading and updating of superblocks.
861 * Interface methods are:
862 * int load_super(mdk_rdev_t *dev, mdk_rdev_t *refdev, int minor_version)
863 * loads and validates a superblock on dev.
864 * if refdev != NULL, compare superblocks on both devices
866 * 0 - dev has a superblock that is compatible with refdev
867 * 1 - dev has a superblock that is compatible and newer than refdev
868 * so dev should be used as the refdev in future
869 * -EINVAL superblock incompatible or invalid
870 * -othererror e.g. -EIO
872 * int validate_super(mddev_t *mddev, mdk_rdev_t *dev)
873 * Verify that dev is acceptable into mddev.
874 * The first time, mddev->raid_disks will be 0, and data from
875 * dev should be merged in. Subsequent calls check that dev
876 * is new enough. Return 0 or -EINVAL
878 * void sync_super(mddev_t *mddev, mdk_rdev_t *dev)
879 * Update the superblock for rdev with data in mddev
880 * This does not write to disc.
886 struct module *owner;
887 int (*load_super)(mdk_rdev_t *rdev, mdk_rdev_t *refdev,
889 int (*validate_super)(mddev_t *mddev, mdk_rdev_t *rdev);
890 void (*sync_super)(mddev_t *mddev, mdk_rdev_t *rdev);
891 unsigned long long (*rdev_size_change)(mdk_rdev_t *rdev,
892 sector_t num_sectors);
896 * Check that the given mddev has no bitmap.
898 * This function is called from the run method of all personalities that do not
899 * support bitmaps. It prints an error message and returns non-zero if mddev
900 * has a bitmap. Otherwise, it returns 0.
903 int md_check_no_bitmap(mddev_t *mddev)
905 if (!mddev->bitmap_info.file && !mddev->bitmap_info.offset)
907 printk(KERN_ERR "%s: bitmaps are not supported for %s\n",
908 mdname(mddev), mddev->pers->name);
911 EXPORT_SYMBOL(md_check_no_bitmap);
914 * load_super for 0.90.0
916 static int super_90_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
918 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
923 * Calculate the position of the superblock (512byte sectors),
924 * it's at the end of the disk.
926 * It also happens to be a multiple of 4Kb.
928 rdev->sb_start = calc_dev_sboffset(rdev->bdev);
930 ret = read_disk_sb(rdev, MD_SB_BYTES);
935 bdevname(rdev->bdev, b);
936 sb = (mdp_super_t*)page_address(rdev->sb_page);
938 if (sb->md_magic != MD_SB_MAGIC) {
939 printk(KERN_ERR "md: invalid raid superblock magic on %s\n",
944 if (sb->major_version != 0 ||
945 sb->minor_version < 90 ||
946 sb->minor_version > 91) {
947 printk(KERN_WARNING "Bad version number %d.%d on %s\n",
948 sb->major_version, sb->minor_version,
953 if (sb->raid_disks <= 0)
956 if (md_csum_fold(calc_sb_csum(sb)) != md_csum_fold(sb->sb_csum)) {
957 printk(KERN_WARNING "md: invalid superblock checksum on %s\n",
962 rdev->preferred_minor = sb->md_minor;
963 rdev->data_offset = 0;
964 rdev->sb_size = MD_SB_BYTES;
966 if (sb->level == LEVEL_MULTIPATH)
969 rdev->desc_nr = sb->this_disk.number;
975 mdp_super_t *refsb = (mdp_super_t*)page_address(refdev->sb_page);
976 if (!uuid_equal(refsb, sb)) {
977 printk(KERN_WARNING "md: %s has different UUID to %s\n",
978 b, bdevname(refdev->bdev,b2));
981 if (!sb_equal(refsb, sb)) {
982 printk(KERN_WARNING "md: %s has same UUID"
983 " but different superblock to %s\n",
984 b, bdevname(refdev->bdev, b2));
988 ev2 = md_event(refsb);
994 rdev->sectors = rdev->sb_start;
996 if (rdev->sectors < sb->size * 2 && sb->level > 1)
997 /* "this cannot possibly happen" ... */
1005 * validate_super for 0.90.0
1007 static int super_90_validate(mddev_t *mddev, mdk_rdev_t *rdev)
1010 mdp_super_t *sb = (mdp_super_t *)page_address(rdev->sb_page);
1011 __u64 ev1 = md_event(sb);
1013 rdev->raid_disk = -1;
1014 clear_bit(Faulty, &rdev->flags);
1015 clear_bit(In_sync, &rdev->flags);
1016 clear_bit(WriteMostly, &rdev->flags);
1017 clear_bit(BarriersNotsupp, &rdev->flags);
1019 if (mddev->raid_disks == 0) {
1020 mddev->major_version = 0;
1021 mddev->minor_version = sb->minor_version;
1022 mddev->patch_version = sb->patch_version;
1023 mddev->external = 0;
1024 mddev->chunk_sectors = sb->chunk_size >> 9;
1025 mddev->ctime = sb->ctime;
1026 mddev->utime = sb->utime;
1027 mddev->level = sb->level;
1028 mddev->clevel[0] = 0;
1029 mddev->layout = sb->layout;
1030 mddev->raid_disks = sb->raid_disks;
1031 mddev->dev_sectors = sb->size * 2;
1032 mddev->events = ev1;
1033 mddev->bitmap_info.offset = 0;
1034 mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9;
1036 if (mddev->minor_version >= 91) {
1037 mddev->reshape_position = sb->reshape_position;
1038 mddev->delta_disks = sb->delta_disks;
1039 mddev->new_level = sb->new_level;
1040 mddev->new_layout = sb->new_layout;
1041 mddev->new_chunk_sectors = sb->new_chunk >> 9;
1043 mddev->reshape_position = MaxSector;
1044 mddev->delta_disks = 0;
1045 mddev->new_level = mddev->level;
1046 mddev->new_layout = mddev->layout;
1047 mddev->new_chunk_sectors = mddev->chunk_sectors;
1050 if (sb->state & (1<<MD_SB_CLEAN))
1051 mddev->recovery_cp = MaxSector;
1053 if (sb->events_hi == sb->cp_events_hi &&
1054 sb->events_lo == sb->cp_events_lo) {
1055 mddev->recovery_cp = sb->recovery_cp;
1057 mddev->recovery_cp = 0;
1060 memcpy(mddev->uuid+0, &sb->set_uuid0, 4);
1061 memcpy(mddev->uuid+4, &sb->set_uuid1, 4);
1062 memcpy(mddev->uuid+8, &sb->set_uuid2, 4);
1063 memcpy(mddev->uuid+12,&sb->set_uuid3, 4);
1065 mddev->max_disks = MD_SB_DISKS;
1067 if (sb->state & (1<<MD_SB_BITMAP_PRESENT) &&
1068 mddev->bitmap_info.file == NULL)
1069 mddev->bitmap_info.offset =
1070 mddev->bitmap_info.default_offset;
1072 } else if (mddev->pers == NULL) {
1073 /* Insist on good event counter while assembling */
1075 if (ev1 < mddev->events)
1077 } else if (mddev->bitmap) {
1078 /* if adding to array with a bitmap, then we can accept an
1079 * older device ... but not too old.
1081 if (ev1 < mddev->bitmap->events_cleared)
1084 if (ev1 < mddev->events)
1085 /* just a hot-add of a new device, leave raid_disk at -1 */
1089 if (mddev->level != LEVEL_MULTIPATH) {
1090 desc = sb->disks + rdev->desc_nr;
1092 if (desc->state & (1<<MD_DISK_FAULTY))
1093 set_bit(Faulty, &rdev->flags);
1094 else if (desc->state & (1<<MD_DISK_SYNC) /* &&
1095 desc->raid_disk < mddev->raid_disks */) {
1096 set_bit(In_sync, &rdev->flags);
1097 rdev->raid_disk = desc->raid_disk;
1098 } else if (desc->state & (1<<MD_DISK_ACTIVE)) {
1099 /* active but not in sync implies recovery up to
1100 * reshape position. We don't know exactly where
1101 * that is, so set to zero for now */
1102 if (mddev->minor_version >= 91) {
1103 rdev->recovery_offset = 0;
1104 rdev->raid_disk = desc->raid_disk;
1107 if (desc->state & (1<<MD_DISK_WRITEMOSTLY))
1108 set_bit(WriteMostly, &rdev->flags);
1109 } else /* MULTIPATH are always insync */
1110 set_bit(In_sync, &rdev->flags);
1115 * sync_super for 0.90.0
1117 static void super_90_sync(mddev_t *mddev, mdk_rdev_t *rdev)
1121 int next_spare = mddev->raid_disks;
1124 /* make rdev->sb match mddev data..
1127 * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
1128 * 3/ any empty disks < next_spare become removed
1130 * disks[0] gets initialised to REMOVED because
1131 * we cannot be sure from other fields if it has
1132 * been initialised or not.
1135 int active=0, working=0,failed=0,spare=0,nr_disks=0;
1137 rdev->sb_size = MD_SB_BYTES;
1139 sb = (mdp_super_t*)page_address(rdev->sb_page);
1141 memset(sb, 0, sizeof(*sb));
1143 sb->md_magic = MD_SB_MAGIC;
1144 sb->major_version = mddev->major_version;
1145 sb->patch_version = mddev->patch_version;
1146 sb->gvalid_words = 0; /* ignored */
1147 memcpy(&sb->set_uuid0, mddev->uuid+0, 4);
1148 memcpy(&sb->set_uuid1, mddev->uuid+4, 4);
1149 memcpy(&sb->set_uuid2, mddev->uuid+8, 4);
1150 memcpy(&sb->set_uuid3, mddev->uuid+12,4);
1152 sb->ctime = mddev->ctime;
1153 sb->level = mddev->level;
1154 sb->size = mddev->dev_sectors / 2;
1155 sb->raid_disks = mddev->raid_disks;
1156 sb->md_minor = mddev->md_minor;
1157 sb->not_persistent = 0;
1158 sb->utime = mddev->utime;
1160 sb->events_hi = (mddev->events>>32);
1161 sb->events_lo = (u32)mddev->events;
1163 if (mddev->reshape_position == MaxSector)
1164 sb->minor_version = 90;
1166 sb->minor_version = 91;
1167 sb->reshape_position = mddev->reshape_position;
1168 sb->new_level = mddev->new_level;
1169 sb->delta_disks = mddev->delta_disks;
1170 sb->new_layout = mddev->new_layout;
1171 sb->new_chunk = mddev->new_chunk_sectors << 9;
1173 mddev->minor_version = sb->minor_version;
1176 sb->recovery_cp = mddev->recovery_cp;
1177 sb->cp_events_hi = (mddev->events>>32);
1178 sb->cp_events_lo = (u32)mddev->events;
1179 if (mddev->recovery_cp == MaxSector)
1180 sb->state = (1<< MD_SB_CLEAN);
1182 sb->recovery_cp = 0;
1184 sb->layout = mddev->layout;
1185 sb->chunk_size = mddev->chunk_sectors << 9;
1187 if (mddev->bitmap && mddev->bitmap_info.file == NULL)
1188 sb->state |= (1<<MD_SB_BITMAP_PRESENT);
1190 sb->disks[0].state = (1<<MD_DISK_REMOVED);
1191 list_for_each_entry(rdev2, &mddev->disks, same_set) {
1194 int is_active = test_bit(In_sync, &rdev2->flags);
1196 if (rdev2->raid_disk >= 0 &&
1197 sb->minor_version >= 91)
1198 /* we have nowhere to store the recovery_offset,
1199 * but if it is not below the reshape_position,
1200 * we can piggy-back on that.
1203 if (rdev2->raid_disk < 0 ||
1204 test_bit(Faulty, &rdev2->flags))
1207 desc_nr = rdev2->raid_disk;
1209 desc_nr = next_spare++;
1210 rdev2->desc_nr = desc_nr;
1211 d = &sb->disks[rdev2->desc_nr];
1213 d->number = rdev2->desc_nr;
1214 d->major = MAJOR(rdev2->bdev->bd_dev);
1215 d->minor = MINOR(rdev2->bdev->bd_dev);
1217 d->raid_disk = rdev2->raid_disk;
1219 d->raid_disk = rdev2->desc_nr; /* compatibility */
1220 if (test_bit(Faulty, &rdev2->flags))
1221 d->state = (1<<MD_DISK_FAULTY);
1222 else if (is_active) {
1223 d->state = (1<<MD_DISK_ACTIVE);
1224 if (test_bit(In_sync, &rdev2->flags))
1225 d->state |= (1<<MD_DISK_SYNC);
1233 if (test_bit(WriteMostly, &rdev2->flags))
1234 d->state |= (1<<MD_DISK_WRITEMOSTLY);
1236 /* now set the "removed" and "faulty" bits on any missing devices */
1237 for (i=0 ; i < mddev->raid_disks ; i++) {
1238 mdp_disk_t *d = &sb->disks[i];
1239 if (d->state == 0 && d->number == 0) {
1242 d->state = (1<<MD_DISK_REMOVED);
1243 d->state |= (1<<MD_DISK_FAULTY);
1247 sb->nr_disks = nr_disks;
1248 sb->active_disks = active;
1249 sb->working_disks = working;
1250 sb->failed_disks = failed;
1251 sb->spare_disks = spare;
1253 sb->this_disk = sb->disks[rdev->desc_nr];
1254 sb->sb_csum = calc_sb_csum(sb);
1258 * rdev_size_change for 0.90.0
1260 static unsigned long long
1261 super_90_rdev_size_change(mdk_rdev_t *rdev, sector_t num_sectors)
1263 if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
1264 return 0; /* component must fit device */
1265 if (rdev->mddev->bitmap_info.offset)
1266 return 0; /* can't move bitmap */
1267 rdev->sb_start = calc_dev_sboffset(rdev->bdev);
1268 if (!num_sectors || num_sectors > rdev->sb_start)
1269 num_sectors = rdev->sb_start;
1270 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
1272 md_super_wait(rdev->mddev);
1273 return num_sectors / 2; /* kB for sysfs */
1278 * version 1 superblock
1281 static __le32 calc_sb_1_csum(struct mdp_superblock_1 * sb)
1285 unsigned long long newcsum;
1286 int size = 256 + le32_to_cpu(sb->max_dev)*2;
1287 __le32 *isuper = (__le32*)sb;
1290 disk_csum = sb->sb_csum;
1293 for (i=0; size>=4; size -= 4 )
1294 newcsum += le32_to_cpu(*isuper++);
1297 newcsum += le16_to_cpu(*(__le16*) isuper);
1299 csum = (newcsum & 0xffffffff) + (newcsum >> 32);
1300 sb->sb_csum = disk_csum;
1301 return cpu_to_le32(csum);
1304 static int super_1_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
1306 struct mdp_superblock_1 *sb;
1309 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
1313 * Calculate the position of the superblock in 512byte sectors.
1314 * It is always aligned to a 4K boundary and
1315 * depeding on minor_version, it can be:
1316 * 0: At least 8K, but less than 12K, from end of device
1317 * 1: At start of device
1318 * 2: 4K from start of device.
1320 switch(minor_version) {
1322 sb_start = rdev->bdev->bd_inode->i_size >> 9;
1324 sb_start &= ~(sector_t)(4*2-1);
1335 rdev->sb_start = sb_start;
1337 /* superblock is rarely larger than 1K, but it can be larger,
1338 * and it is safe to read 4k, so we do that
1340 ret = read_disk_sb(rdev, 4096);
1341 if (ret) return ret;
1344 sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1346 if (sb->magic != cpu_to_le32(MD_SB_MAGIC) ||
1347 sb->major_version != cpu_to_le32(1) ||
1348 le32_to_cpu(sb->max_dev) > (4096-256)/2 ||
1349 le64_to_cpu(sb->super_offset) != rdev->sb_start ||
1350 (le32_to_cpu(sb->feature_map) & ~MD_FEATURE_ALL) != 0)
1353 if (calc_sb_1_csum(sb) != sb->sb_csum) {
1354 printk("md: invalid superblock checksum on %s\n",
1355 bdevname(rdev->bdev,b));
1358 if (le64_to_cpu(sb->data_size) < 10) {
1359 printk("md: data_size too small on %s\n",
1360 bdevname(rdev->bdev,b));
1364 rdev->preferred_minor = 0xffff;
1365 rdev->data_offset = le64_to_cpu(sb->data_offset);
1366 atomic_set(&rdev->corrected_errors, le32_to_cpu(sb->cnt_corrected_read));
1368 rdev->sb_size = le32_to_cpu(sb->max_dev) * 2 + 256;
1369 bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
1370 if (rdev->sb_size & bmask)
1371 rdev->sb_size = (rdev->sb_size | bmask) + 1;
1374 && rdev->data_offset < sb_start + (rdev->sb_size/512))
1377 if (sb->level == cpu_to_le32(LEVEL_MULTIPATH))
1380 rdev->desc_nr = le32_to_cpu(sb->dev_number);
1386 struct mdp_superblock_1 *refsb =
1387 (struct mdp_superblock_1*)page_address(refdev->sb_page);
1389 if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 ||
1390 sb->level != refsb->level ||
1391 sb->layout != refsb->layout ||
1392 sb->chunksize != refsb->chunksize) {
1393 printk(KERN_WARNING "md: %s has strangely different"
1394 " superblock to %s\n",
1395 bdevname(rdev->bdev,b),
1396 bdevname(refdev->bdev,b2));
1399 ev1 = le64_to_cpu(sb->events);
1400 ev2 = le64_to_cpu(refsb->events);
1408 rdev->sectors = (rdev->bdev->bd_inode->i_size >> 9) -
1409 le64_to_cpu(sb->data_offset);
1411 rdev->sectors = rdev->sb_start;
1412 if (rdev->sectors < le64_to_cpu(sb->data_size))
1414 rdev->sectors = le64_to_cpu(sb->data_size);
1415 if (le64_to_cpu(sb->size) > rdev->sectors)
1420 static int super_1_validate(mddev_t *mddev, mdk_rdev_t *rdev)
1422 struct mdp_superblock_1 *sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1423 __u64 ev1 = le64_to_cpu(sb->events);
1425 rdev->raid_disk = -1;
1426 clear_bit(Faulty, &rdev->flags);
1427 clear_bit(In_sync, &rdev->flags);
1428 clear_bit(WriteMostly, &rdev->flags);
1429 clear_bit(BarriersNotsupp, &rdev->flags);
1431 if (mddev->raid_disks == 0) {
1432 mddev->major_version = 1;
1433 mddev->patch_version = 0;
1434 mddev->external = 0;
1435 mddev->chunk_sectors = le32_to_cpu(sb->chunksize);
1436 mddev->ctime = le64_to_cpu(sb->ctime) & ((1ULL << 32)-1);
1437 mddev->utime = le64_to_cpu(sb->utime) & ((1ULL << 32)-1);
1438 mddev->level = le32_to_cpu(sb->level);
1439 mddev->clevel[0] = 0;
1440 mddev->layout = le32_to_cpu(sb->layout);
1441 mddev->raid_disks = le32_to_cpu(sb->raid_disks);
1442 mddev->dev_sectors = le64_to_cpu(sb->size);
1443 mddev->events = ev1;
1444 mddev->bitmap_info.offset = 0;
1445 mddev->bitmap_info.default_offset = 1024 >> 9;
1447 mddev->recovery_cp = le64_to_cpu(sb->resync_offset);
1448 memcpy(mddev->uuid, sb->set_uuid, 16);
1450 mddev->max_disks = (4096-256)/2;
1452 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) &&
1453 mddev->bitmap_info.file == NULL )
1454 mddev->bitmap_info.offset =
1455 (__s32)le32_to_cpu(sb->bitmap_offset);
1457 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) {
1458 mddev->reshape_position = le64_to_cpu(sb->reshape_position);
1459 mddev->delta_disks = le32_to_cpu(sb->delta_disks);
1460 mddev->new_level = le32_to_cpu(sb->new_level);
1461 mddev->new_layout = le32_to_cpu(sb->new_layout);
1462 mddev->new_chunk_sectors = le32_to_cpu(sb->new_chunk);
1464 mddev->reshape_position = MaxSector;
1465 mddev->delta_disks = 0;
1466 mddev->new_level = mddev->level;
1467 mddev->new_layout = mddev->layout;
1468 mddev->new_chunk_sectors = mddev->chunk_sectors;
1471 } else if (mddev->pers == NULL) {
1472 /* Insist of good event counter while assembling */
1474 if (ev1 < mddev->events)
1476 } else if (mddev->bitmap) {
1477 /* If adding to array with a bitmap, then we can accept an
1478 * older device, but not too old.
1480 if (ev1 < mddev->bitmap->events_cleared)
1483 if (ev1 < mddev->events)
1484 /* just a hot-add of a new device, leave raid_disk at -1 */
1487 if (mddev->level != LEVEL_MULTIPATH) {
1489 if (rdev->desc_nr < 0 ||
1490 rdev->desc_nr >= le32_to_cpu(sb->max_dev)) {
1494 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
1496 case 0xffff: /* spare */
1498 case 0xfffe: /* faulty */
1499 set_bit(Faulty, &rdev->flags);
1502 if ((le32_to_cpu(sb->feature_map) &
1503 MD_FEATURE_RECOVERY_OFFSET))
1504 rdev->recovery_offset = le64_to_cpu(sb->recovery_offset);
1506 set_bit(In_sync, &rdev->flags);
1507 rdev->raid_disk = role;
1510 if (sb->devflags & WriteMostly1)
1511 set_bit(WriteMostly, &rdev->flags);
1512 } else /* MULTIPATH are always insync */
1513 set_bit(In_sync, &rdev->flags);
1518 static void super_1_sync(mddev_t *mddev, mdk_rdev_t *rdev)
1520 struct mdp_superblock_1 *sb;
1523 /* make rdev->sb match mddev and rdev data. */
1525 sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1527 sb->feature_map = 0;
1529 sb->recovery_offset = cpu_to_le64(0);
1530 memset(sb->pad1, 0, sizeof(sb->pad1));
1531 memset(sb->pad2, 0, sizeof(sb->pad2));
1532 memset(sb->pad3, 0, sizeof(sb->pad3));
1534 sb->utime = cpu_to_le64((__u64)mddev->utime);
1535 sb->events = cpu_to_le64(mddev->events);
1537 sb->resync_offset = cpu_to_le64(mddev->recovery_cp);
1539 sb->resync_offset = cpu_to_le64(0);
1541 sb->cnt_corrected_read = cpu_to_le32(atomic_read(&rdev->corrected_errors));
1543 sb->raid_disks = cpu_to_le32(mddev->raid_disks);
1544 sb->size = cpu_to_le64(mddev->dev_sectors);
1545 sb->chunksize = cpu_to_le32(mddev->chunk_sectors);
1546 sb->level = cpu_to_le32(mddev->level);
1547 sb->layout = cpu_to_le32(mddev->layout);
1549 if (mddev->bitmap && mddev->bitmap_info.file == NULL) {
1550 sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_info.offset);
1551 sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET);
1554 if (rdev->raid_disk >= 0 &&
1555 !test_bit(In_sync, &rdev->flags)) {
1557 cpu_to_le32(MD_FEATURE_RECOVERY_OFFSET);
1558 sb->recovery_offset =
1559 cpu_to_le64(rdev->recovery_offset);
1562 if (mddev->reshape_position != MaxSector) {
1563 sb->feature_map |= cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE);
1564 sb->reshape_position = cpu_to_le64(mddev->reshape_position);
1565 sb->new_layout = cpu_to_le32(mddev->new_layout);
1566 sb->delta_disks = cpu_to_le32(mddev->delta_disks);
1567 sb->new_level = cpu_to_le32(mddev->new_level);
1568 sb->new_chunk = cpu_to_le32(mddev->new_chunk_sectors);
1572 list_for_each_entry(rdev2, &mddev->disks, same_set)
1573 if (rdev2->desc_nr+1 > max_dev)
1574 max_dev = rdev2->desc_nr+1;
1576 if (max_dev > le32_to_cpu(sb->max_dev)) {
1578 sb->max_dev = cpu_to_le32(max_dev);
1579 rdev->sb_size = max_dev * 2 + 256;
1580 bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
1581 if (rdev->sb_size & bmask)
1582 rdev->sb_size = (rdev->sb_size | bmask) + 1;
1584 for (i=0; i<max_dev;i++)
1585 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1587 list_for_each_entry(rdev2, &mddev->disks, same_set) {
1589 if (test_bit(Faulty, &rdev2->flags))
1590 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1591 else if (test_bit(In_sync, &rdev2->flags))
1592 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1593 else if (rdev2->raid_disk >= 0)
1594 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1596 sb->dev_roles[i] = cpu_to_le16(0xffff);
1599 sb->sb_csum = calc_sb_1_csum(sb);
1602 static unsigned long long
1603 super_1_rdev_size_change(mdk_rdev_t *rdev, sector_t num_sectors)
1605 struct mdp_superblock_1 *sb;
1606 sector_t max_sectors;
1607 if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
1608 return 0; /* component must fit device */
1609 if (rdev->sb_start < rdev->data_offset) {
1610 /* minor versions 1 and 2; superblock before data */
1611 max_sectors = rdev->bdev->bd_inode->i_size >> 9;
1612 max_sectors -= rdev->data_offset;
1613 if (!num_sectors || num_sectors > max_sectors)
1614 num_sectors = max_sectors;
1615 } else if (rdev->mddev->bitmap_info.offset) {
1616 /* minor version 0 with bitmap we can't move */
1619 /* minor version 0; superblock after data */
1621 sb_start = (rdev->bdev->bd_inode->i_size >> 9) - 8*2;
1622 sb_start &= ~(sector_t)(4*2 - 1);
1623 max_sectors = rdev->sectors + sb_start - rdev->sb_start;
1624 if (!num_sectors || num_sectors > max_sectors)
1625 num_sectors = max_sectors;
1626 rdev->sb_start = sb_start;
1628 sb = (struct mdp_superblock_1 *) page_address(rdev->sb_page);
1629 sb->data_size = cpu_to_le64(num_sectors);
1630 sb->super_offset = rdev->sb_start;
1631 sb->sb_csum = calc_sb_1_csum(sb);
1632 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
1634 md_super_wait(rdev->mddev);
1635 return num_sectors / 2; /* kB for sysfs */
1638 static struct super_type super_types[] = {
1641 .owner = THIS_MODULE,
1642 .load_super = super_90_load,
1643 .validate_super = super_90_validate,
1644 .sync_super = super_90_sync,
1645 .rdev_size_change = super_90_rdev_size_change,
1649 .owner = THIS_MODULE,
1650 .load_super = super_1_load,
1651 .validate_super = super_1_validate,
1652 .sync_super = super_1_sync,
1653 .rdev_size_change = super_1_rdev_size_change,
1657 static int match_mddev_units(mddev_t *mddev1, mddev_t *mddev2)
1659 mdk_rdev_t *rdev, *rdev2;
1662 rdev_for_each_rcu(rdev, mddev1)
1663 rdev_for_each_rcu(rdev2, mddev2)
1664 if (rdev->bdev->bd_contains ==
1665 rdev2->bdev->bd_contains) {
1673 static LIST_HEAD(pending_raid_disks);
1676 * Try to register data integrity profile for an mddev
1678 * This is called when an array is started and after a disk has been kicked
1679 * from the array. It only succeeds if all working and active component devices
1680 * are integrity capable with matching profiles.
1682 int md_integrity_register(mddev_t *mddev)
1684 mdk_rdev_t *rdev, *reference = NULL;
1686 if (list_empty(&mddev->disks))
1687 return 0; /* nothing to do */
1688 if (blk_get_integrity(mddev->gendisk))
1689 return 0; /* already registered */
1690 list_for_each_entry(rdev, &mddev->disks, same_set) {
1691 /* skip spares and non-functional disks */
1692 if (test_bit(Faulty, &rdev->flags))
1694 if (rdev->raid_disk < 0)
1697 * If at least one rdev is not integrity capable, we can not
1698 * enable data integrity for the md device.
1700 if (!bdev_get_integrity(rdev->bdev))
1703 /* Use the first rdev as the reference */
1707 /* does this rdev's profile match the reference profile? */
1708 if (blk_integrity_compare(reference->bdev->bd_disk,
1709 rdev->bdev->bd_disk) < 0)
1713 * All component devices are integrity capable and have matching
1714 * profiles, register the common profile for the md device.
1716 if (blk_integrity_register(mddev->gendisk,
1717 bdev_get_integrity(reference->bdev)) != 0) {
1718 printk(KERN_ERR "md: failed to register integrity for %s\n",
1722 printk(KERN_NOTICE "md: data integrity on %s enabled\n",
1726 EXPORT_SYMBOL(md_integrity_register);
1728 /* Disable data integrity if non-capable/non-matching disk is being added */
1729 void md_integrity_add_rdev(mdk_rdev_t *rdev, mddev_t *mddev)
1731 struct blk_integrity *bi_rdev = bdev_get_integrity(rdev->bdev);
1732 struct blk_integrity *bi_mddev = blk_get_integrity(mddev->gendisk);
1734 if (!bi_mddev) /* nothing to do */
1736 if (rdev->raid_disk < 0) /* skip spares */
1738 if (bi_rdev && blk_integrity_compare(mddev->gendisk,
1739 rdev->bdev->bd_disk) >= 0)
1741 printk(KERN_NOTICE "disabling data integrity on %s\n", mdname(mddev));
1742 blk_integrity_unregister(mddev->gendisk);
1744 EXPORT_SYMBOL(md_integrity_add_rdev);
1746 static int bind_rdev_to_array(mdk_rdev_t * rdev, mddev_t * mddev)
1748 char b[BDEVNAME_SIZE];
1758 /* prevent duplicates */
1759 if (find_rdev(mddev, rdev->bdev->bd_dev))
1762 /* make sure rdev->sectors exceeds mddev->dev_sectors */
1763 if (rdev->sectors && (mddev->dev_sectors == 0 ||
1764 rdev->sectors < mddev->dev_sectors)) {
1766 /* Cannot change size, so fail
1767 * If mddev->level <= 0, then we don't care
1768 * about aligning sizes (e.g. linear)
1770 if (mddev->level > 0)
1773 mddev->dev_sectors = rdev->sectors;
1776 /* Verify rdev->desc_nr is unique.
1777 * If it is -1, assign a free number, else
1778 * check number is not in use
1780 if (rdev->desc_nr < 0) {
1782 if (mddev->pers) choice = mddev->raid_disks;
1783 while (find_rdev_nr(mddev, choice))
1785 rdev->desc_nr = choice;
1787 if (find_rdev_nr(mddev, rdev->desc_nr))
1790 if (mddev->max_disks && rdev->desc_nr >= mddev->max_disks) {
1791 printk(KERN_WARNING "md: %s: array is limited to %d devices\n",
1792 mdname(mddev), mddev->max_disks);
1795 bdevname(rdev->bdev,b);
1796 while ( (s=strchr(b, '/')) != NULL)
1799 rdev->mddev = mddev;
1800 printk(KERN_INFO "md: bind<%s>\n", b);
1802 if ((err = kobject_add(&rdev->kobj, &mddev->kobj, "dev-%s", b)))
1805 ko = &part_to_dev(rdev->bdev->bd_part)->kobj;
1806 if ((err = sysfs_create_link(&rdev->kobj, ko, "block"))) {
1807 kobject_del(&rdev->kobj);
1810 rdev->sysfs_state = sysfs_get_dirent(rdev->kobj.sd, "state");
1812 list_add_rcu(&rdev->same_set, &mddev->disks);
1813 bd_claim_by_disk(rdev->bdev, rdev->bdev->bd_holder, mddev->gendisk);
1815 /* May as well allow recovery to be retried once */
1816 mddev->recovery_disabled = 0;
1821 printk(KERN_WARNING "md: failed to register dev-%s for %s\n",
1826 static void md_delayed_delete(struct work_struct *ws)
1828 mdk_rdev_t *rdev = container_of(ws, mdk_rdev_t, del_work);
1829 kobject_del(&rdev->kobj);
1830 kobject_put(&rdev->kobj);
1833 static void unbind_rdev_from_array(mdk_rdev_t * rdev)
1835 char b[BDEVNAME_SIZE];
1840 bd_release_from_disk(rdev->bdev, rdev->mddev->gendisk);
1841 list_del_rcu(&rdev->same_set);
1842 printk(KERN_INFO "md: unbind<%s>\n", bdevname(rdev->bdev,b));
1844 sysfs_remove_link(&rdev->kobj, "block");
1845 sysfs_put(rdev->sysfs_state);
1846 rdev->sysfs_state = NULL;
1847 /* We need to delay this, otherwise we can deadlock when
1848 * writing to 'remove' to "dev/state". We also need
1849 * to delay it due to rcu usage.
1852 INIT_WORK(&rdev->del_work, md_delayed_delete);
1853 kobject_get(&rdev->kobj);
1854 schedule_work(&rdev->del_work);
1858 * prevent the device from being mounted, repartitioned or
1859 * otherwise reused by a RAID array (or any other kernel
1860 * subsystem), by bd_claiming the device.
1862 static int lock_rdev(mdk_rdev_t *rdev, dev_t dev, int shared)
1865 struct block_device *bdev;
1866 char b[BDEVNAME_SIZE];
1868 bdev = open_by_devnum(dev, FMODE_READ|FMODE_WRITE);
1870 printk(KERN_ERR "md: could not open %s.\n",
1871 __bdevname(dev, b));
1872 return PTR_ERR(bdev);
1874 err = bd_claim(bdev, shared ? (mdk_rdev_t *)lock_rdev : rdev);
1876 printk(KERN_ERR "md: could not bd_claim %s.\n",
1878 blkdev_put(bdev, FMODE_READ|FMODE_WRITE);
1882 set_bit(AllReserved, &rdev->flags);
1887 static void unlock_rdev(mdk_rdev_t *rdev)
1889 struct block_device *bdev = rdev->bdev;
1894 blkdev_put(bdev, FMODE_READ|FMODE_WRITE);
1897 void md_autodetect_dev(dev_t dev);
1899 static void export_rdev(mdk_rdev_t * rdev)
1901 char b[BDEVNAME_SIZE];
1902 printk(KERN_INFO "md: export_rdev(%s)\n",
1903 bdevname(rdev->bdev,b));
1908 if (test_bit(AutoDetected, &rdev->flags))
1909 md_autodetect_dev(rdev->bdev->bd_dev);
1912 kobject_put(&rdev->kobj);
1915 static void kick_rdev_from_array(mdk_rdev_t * rdev)
1917 unbind_rdev_from_array(rdev);
1921 static void export_array(mddev_t *mddev)
1923 mdk_rdev_t *rdev, *tmp;
1925 rdev_for_each(rdev, tmp, mddev) {
1930 kick_rdev_from_array(rdev);
1932 if (!list_empty(&mddev->disks))
1934 mddev->raid_disks = 0;
1935 mddev->major_version = 0;
1938 static void print_desc(mdp_disk_t *desc)
1940 printk(" DISK<N:%d,(%d,%d),R:%d,S:%d>\n", desc->number,
1941 desc->major,desc->minor,desc->raid_disk,desc->state);
1944 static void print_sb_90(mdp_super_t *sb)
1949 "md: SB: (V:%d.%d.%d) ID:<%08x.%08x.%08x.%08x> CT:%08x\n",
1950 sb->major_version, sb->minor_version, sb->patch_version,
1951 sb->set_uuid0, sb->set_uuid1, sb->set_uuid2, sb->set_uuid3,
1953 printk(KERN_INFO "md: L%d S%08d ND:%d RD:%d md%d LO:%d CS:%d\n",
1954 sb->level, sb->size, sb->nr_disks, sb->raid_disks,
1955 sb->md_minor, sb->layout, sb->chunk_size);
1956 printk(KERN_INFO "md: UT:%08x ST:%d AD:%d WD:%d"
1957 " FD:%d SD:%d CSUM:%08x E:%08lx\n",
1958 sb->utime, sb->state, sb->active_disks, sb->working_disks,
1959 sb->failed_disks, sb->spare_disks,
1960 sb->sb_csum, (unsigned long)sb->events_lo);
1963 for (i = 0; i < MD_SB_DISKS; i++) {
1966 desc = sb->disks + i;
1967 if (desc->number || desc->major || desc->minor ||
1968 desc->raid_disk || (desc->state && (desc->state != 4))) {
1969 printk(" D %2d: ", i);
1973 printk(KERN_INFO "md: THIS: ");
1974 print_desc(&sb->this_disk);
1977 static void print_sb_1(struct mdp_superblock_1 *sb)
1981 uuid = sb->set_uuid;
1983 "md: SB: (V:%u) (F:0x%08x) Array-ID:<%pU>\n"
1984 "md: Name: \"%s\" CT:%llu\n",
1985 le32_to_cpu(sb->major_version),
1986 le32_to_cpu(sb->feature_map),
1989 (unsigned long long)le64_to_cpu(sb->ctime)
1990 & MD_SUPERBLOCK_1_TIME_SEC_MASK);
1992 uuid = sb->device_uuid;
1994 "md: L%u SZ%llu RD:%u LO:%u CS:%u DO:%llu DS:%llu SO:%llu"
1996 "md: Dev:%08x UUID: %pU\n"
1997 "md: (F:0x%08x) UT:%llu Events:%llu ResyncOffset:%llu CSUM:0x%08x\n"
1998 "md: (MaxDev:%u) \n",
1999 le32_to_cpu(sb->level),
2000 (unsigned long long)le64_to_cpu(sb->size),
2001 le32_to_cpu(sb->raid_disks),
2002 le32_to_cpu(sb->layout),
2003 le32_to_cpu(sb->chunksize),
2004 (unsigned long long)le64_to_cpu(sb->data_offset),
2005 (unsigned long long)le64_to_cpu(sb->data_size),
2006 (unsigned long long)le64_to_cpu(sb->super_offset),
2007 (unsigned long long)le64_to_cpu(sb->recovery_offset),
2008 le32_to_cpu(sb->dev_number),
2011 (unsigned long long)le64_to_cpu(sb->utime) & MD_SUPERBLOCK_1_TIME_SEC_MASK,
2012 (unsigned long long)le64_to_cpu(sb->events),
2013 (unsigned long long)le64_to_cpu(sb->resync_offset),
2014 le32_to_cpu(sb->sb_csum),
2015 le32_to_cpu(sb->max_dev)
2019 static void print_rdev(mdk_rdev_t *rdev, int major_version)
2021 char b[BDEVNAME_SIZE];
2022 printk(KERN_INFO "md: rdev %s, Sect:%08llu F:%d S:%d DN:%u\n",
2023 bdevname(rdev->bdev, b), (unsigned long long)rdev->sectors,
2024 test_bit(Faulty, &rdev->flags), test_bit(In_sync, &rdev->flags),
2026 if (rdev->sb_loaded) {
2027 printk(KERN_INFO "md: rdev superblock (MJ:%d):\n", major_version);
2028 switch (major_version) {
2030 print_sb_90((mdp_super_t*)page_address(rdev->sb_page));
2033 print_sb_1((struct mdp_superblock_1 *)page_address(rdev->sb_page));
2037 printk(KERN_INFO "md: no rdev superblock!\n");
2040 static void md_print_devices(void)
2042 struct list_head *tmp;
2045 char b[BDEVNAME_SIZE];
2048 printk("md: **********************************\n");
2049 printk("md: * <COMPLETE RAID STATE PRINTOUT> *\n");
2050 printk("md: **********************************\n");
2051 for_each_mddev(mddev, tmp) {
2054 bitmap_print_sb(mddev->bitmap);
2056 printk("%s: ", mdname(mddev));
2057 list_for_each_entry(rdev, &mddev->disks, same_set)
2058 printk("<%s>", bdevname(rdev->bdev,b));
2061 list_for_each_entry(rdev, &mddev->disks, same_set)
2062 print_rdev(rdev, mddev->major_version);
2064 printk("md: **********************************\n");
2069 static void sync_sbs(mddev_t * mddev, int nospares)
2071 /* Update each superblock (in-memory image), but
2072 * if we are allowed to, skip spares which already
2073 * have the right event counter, or have one earlier
2074 * (which would mean they aren't being marked as dirty
2075 * with the rest of the array)
2079 /* First make sure individual recovery_offsets are correct */
2080 list_for_each_entry(rdev, &mddev->disks, same_set) {
2081 if (rdev->raid_disk >= 0 &&
2082 !test_bit(In_sync, &rdev->flags) &&
2083 mddev->curr_resync_completed > rdev->recovery_offset)
2084 rdev->recovery_offset = mddev->curr_resync_completed;
2087 list_for_each_entry(rdev, &mddev->disks, same_set) {
2088 if (rdev->sb_events == mddev->events ||
2090 rdev->raid_disk < 0 &&
2091 (rdev->sb_events&1)==0 &&
2092 rdev->sb_events+1 == mddev->events)) {
2093 /* Don't update this superblock */
2094 rdev->sb_loaded = 2;
2096 super_types[mddev->major_version].
2097 sync_super(mddev, rdev);
2098 rdev->sb_loaded = 1;
2103 static void md_update_sb(mddev_t * mddev, int force_change)
2109 mddev->utime = get_seconds();
2110 if (mddev->external)
2113 spin_lock_irq(&mddev->write_lock);
2115 set_bit(MD_CHANGE_PENDING, &mddev->flags);
2116 if (test_and_clear_bit(MD_CHANGE_DEVS, &mddev->flags))
2118 if (test_and_clear_bit(MD_CHANGE_CLEAN, &mddev->flags))
2119 /* just a clean<-> dirty transition, possibly leave spares alone,
2120 * though if events isn't the right even/odd, we will have to do
2126 if (mddev->degraded)
2127 /* If the array is degraded, then skipping spares is both
2128 * dangerous and fairly pointless.
2129 * Dangerous because a device that was removed from the array
2130 * might have a event_count that still looks up-to-date,
2131 * so it can be re-added without a resync.
2132 * Pointless because if there are any spares to skip,
2133 * then a recovery will happen and soon that array won't
2134 * be degraded any more and the spare can go back to sleep then.
2138 sync_req = mddev->in_sync;
2140 /* If this is just a dirty<->clean transition, and the array is clean
2141 * and 'events' is odd, we can roll back to the previous clean state */
2143 && (mddev->in_sync && mddev->recovery_cp == MaxSector)
2144 && (mddev->events & 1)
2145 && mddev->events != 1)
2148 /* otherwise we have to go forward and ... */
2150 if (!mddev->in_sync || mddev->recovery_cp != MaxSector) { /* not clean */
2151 /* .. if the array isn't clean, an 'even' event must also go
2153 if ((mddev->events&1)==0)
2156 /* otherwise an 'odd' event must go to spares */
2157 if ((mddev->events&1))
2162 if (!mddev->events) {
2164 * oops, this 64-bit counter should never wrap.
2165 * Either we are in around ~1 trillion A.C., assuming
2166 * 1 reboot per second, or we have a bug:
2173 * do not write anything to disk if using
2174 * nonpersistent superblocks
2176 if (!mddev->persistent) {
2177 if (!mddev->external)
2178 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
2180 spin_unlock_irq(&mddev->write_lock);
2181 wake_up(&mddev->sb_wait);
2184 sync_sbs(mddev, nospares);
2185 spin_unlock_irq(&mddev->write_lock);
2188 "md: updating %s RAID superblock on device (in sync %d)\n",
2189 mdname(mddev),mddev->in_sync);
2191 bitmap_update_sb(mddev->bitmap);
2192 list_for_each_entry(rdev, &mddev->disks, same_set) {
2193 char b[BDEVNAME_SIZE];
2194 dprintk(KERN_INFO "md: ");
2195 if (rdev->sb_loaded != 1)
2196 continue; /* no noise on spare devices */
2197 if (test_bit(Faulty, &rdev->flags))
2198 dprintk("(skipping faulty ");
2200 dprintk("%s ", bdevname(rdev->bdev,b));
2201 if (!test_bit(Faulty, &rdev->flags)) {
2202 md_super_write(mddev,rdev,
2203 rdev->sb_start, rdev->sb_size,
2205 dprintk(KERN_INFO "(write) %s's sb offset: %llu\n",
2206 bdevname(rdev->bdev,b),
2207 (unsigned long long)rdev->sb_start);
2208 rdev->sb_events = mddev->events;
2212 if (mddev->level == LEVEL_MULTIPATH)
2213 /* only need to write one superblock... */
2216 md_super_wait(mddev);
2217 /* if there was a failure, MD_CHANGE_DEVS was set, and we re-write super */
2219 spin_lock_irq(&mddev->write_lock);
2220 if (mddev->in_sync != sync_req ||
2221 test_bit(MD_CHANGE_DEVS, &mddev->flags)) {
2222 /* have to write it out again */
2223 spin_unlock_irq(&mddev->write_lock);
2226 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
2227 spin_unlock_irq(&mddev->write_lock);
2228 wake_up(&mddev->sb_wait);
2229 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
2230 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
2234 /* words written to sysfs files may, or may not, be \n terminated.
2235 * We want to accept with case. For this we use cmd_match.
2237 static int cmd_match(const char *cmd, const char *str)
2239 /* See if cmd, written into a sysfs file, matches
2240 * str. They must either be the same, or cmd can
2241 * have a trailing newline
2243 while (*cmd && *str && *cmd == *str) {
2254 struct rdev_sysfs_entry {
2255 struct attribute attr;
2256 ssize_t (*show)(mdk_rdev_t *, char *);
2257 ssize_t (*store)(mdk_rdev_t *, const char *, size_t);
2261 state_show(mdk_rdev_t *rdev, char *page)
2266 if (test_bit(Faulty, &rdev->flags)) {
2267 len+= sprintf(page+len, "%sfaulty",sep);
2270 if (test_bit(In_sync, &rdev->flags)) {
2271 len += sprintf(page+len, "%sin_sync",sep);
2274 if (test_bit(WriteMostly, &rdev->flags)) {
2275 len += sprintf(page+len, "%swrite_mostly",sep);
2278 if (test_bit(Blocked, &rdev->flags)) {
2279 len += sprintf(page+len, "%sblocked", sep);
2282 if (!test_bit(Faulty, &rdev->flags) &&
2283 !test_bit(In_sync, &rdev->flags)) {
2284 len += sprintf(page+len, "%sspare", sep);
2287 return len+sprintf(page+len, "\n");
2291 state_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2294 * faulty - simulates and error
2295 * remove - disconnects the device
2296 * writemostly - sets write_mostly
2297 * -writemostly - clears write_mostly
2298 * blocked - sets the Blocked flag
2299 * -blocked - clears the Blocked flag
2300 * insync - sets Insync providing device isn't active
2303 if (cmd_match(buf, "faulty") && rdev->mddev->pers) {
2304 md_error(rdev->mddev, rdev);
2306 } else if (cmd_match(buf, "remove")) {
2307 if (rdev->raid_disk >= 0)
2310 mddev_t *mddev = rdev->mddev;
2311 kick_rdev_from_array(rdev);
2313 md_update_sb(mddev, 1);
2314 md_new_event(mddev);
2317 } else if (cmd_match(buf, "writemostly")) {
2318 set_bit(WriteMostly, &rdev->flags);
2320 } else if (cmd_match(buf, "-writemostly")) {
2321 clear_bit(WriteMostly, &rdev->flags);
2323 } else if (cmd_match(buf, "blocked")) {
2324 set_bit(Blocked, &rdev->flags);
2326 } else if (cmd_match(buf, "-blocked")) {
2327 clear_bit(Blocked, &rdev->flags);
2328 wake_up(&rdev->blocked_wait);
2329 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2330 md_wakeup_thread(rdev->mddev->thread);
2333 } else if (cmd_match(buf, "insync") && rdev->raid_disk == -1) {
2334 set_bit(In_sync, &rdev->flags);
2337 if (!err && rdev->sysfs_state)
2338 sysfs_notify_dirent(rdev->sysfs_state);
2339 return err ? err : len;
2341 static struct rdev_sysfs_entry rdev_state =
2342 __ATTR(state, S_IRUGO|S_IWUSR, state_show, state_store);
2345 errors_show(mdk_rdev_t *rdev, char *page)
2347 return sprintf(page, "%d\n", atomic_read(&rdev->corrected_errors));
2351 errors_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2354 unsigned long n = simple_strtoul(buf, &e, 10);
2355 if (*buf && (*e == 0 || *e == '\n')) {
2356 atomic_set(&rdev->corrected_errors, n);
2361 static struct rdev_sysfs_entry rdev_errors =
2362 __ATTR(errors, S_IRUGO|S_IWUSR, errors_show, errors_store);
2365 slot_show(mdk_rdev_t *rdev, char *page)
2367 if (rdev->raid_disk < 0)
2368 return sprintf(page, "none\n");
2370 return sprintf(page, "%d\n", rdev->raid_disk);
2374 slot_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2379 int slot = simple_strtoul(buf, &e, 10);
2380 if (strncmp(buf, "none", 4)==0)
2382 else if (e==buf || (*e && *e!= '\n'))
2384 if (rdev->mddev->pers && slot == -1) {
2385 /* Setting 'slot' on an active array requires also
2386 * updating the 'rd%d' link, and communicating
2387 * with the personality with ->hot_*_disk.
2388 * For now we only support removing
2389 * failed/spare devices. This normally happens automatically,
2390 * but not when the metadata is externally managed.
2392 if (rdev->raid_disk == -1)
2394 /* personality does all needed checks */
2395 if (rdev->mddev->pers->hot_add_disk == NULL)
2397 err = rdev->mddev->pers->
2398 hot_remove_disk(rdev->mddev, rdev->raid_disk);
2401 sprintf(nm, "rd%d", rdev->raid_disk);
2402 sysfs_remove_link(&rdev->mddev->kobj, nm);
2403 rdev->raid_disk = -1;
2404 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2405 md_wakeup_thread(rdev->mddev->thread);
2406 } else if (rdev->mddev->pers) {
2408 /* Activating a spare .. or possibly reactivating
2409 * if we ever get bitmaps working here.
2412 if (rdev->raid_disk != -1)
2415 if (rdev->mddev->pers->hot_add_disk == NULL)
2418 list_for_each_entry(rdev2, &rdev->mddev->disks, same_set)
2419 if (rdev2->raid_disk == slot)
2422 rdev->raid_disk = slot;
2423 if (test_bit(In_sync, &rdev->flags))
2424 rdev->saved_raid_disk = slot;
2426 rdev->saved_raid_disk = -1;
2427 err = rdev->mddev->pers->
2428 hot_add_disk(rdev->mddev, rdev);
2430 rdev->raid_disk = -1;
2433 sysfs_notify_dirent(rdev->sysfs_state);
2434 sprintf(nm, "rd%d", rdev->raid_disk);
2435 if (sysfs_create_link(&rdev->mddev->kobj, &rdev->kobj, nm))
2437 "md: cannot register "
2439 nm, mdname(rdev->mddev));
2441 /* don't wakeup anyone, leave that to userspace. */
2443 if (slot >= rdev->mddev->raid_disks)
2445 rdev->raid_disk = slot;
2446 /* assume it is working */
2447 clear_bit(Faulty, &rdev->flags);
2448 clear_bit(WriteMostly, &rdev->flags);
2449 set_bit(In_sync, &rdev->flags);
2450 sysfs_notify_dirent(rdev->sysfs_state);
2456 static struct rdev_sysfs_entry rdev_slot =
2457 __ATTR(slot, S_IRUGO|S_IWUSR, slot_show, slot_store);
2460 offset_show(mdk_rdev_t *rdev, char *page)
2462 return sprintf(page, "%llu\n", (unsigned long long)rdev->data_offset);
2466 offset_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2469 unsigned long long offset = simple_strtoull(buf, &e, 10);
2470 if (e==buf || (*e && *e != '\n'))
2472 if (rdev->mddev->pers && rdev->raid_disk >= 0)
2474 if (rdev->sectors && rdev->mddev->external)
2475 /* Must set offset before size, so overlap checks
2478 rdev->data_offset = offset;
2482 static struct rdev_sysfs_entry rdev_offset =
2483 __ATTR(offset, S_IRUGO|S_IWUSR, offset_show, offset_store);
2486 rdev_size_show(mdk_rdev_t *rdev, char *page)
2488 return sprintf(page, "%llu\n", (unsigned long long)rdev->sectors / 2);
2491 static int overlaps(sector_t s1, sector_t l1, sector_t s2, sector_t l2)
2493 /* check if two start/length pairs overlap */
2501 static int strict_blocks_to_sectors(const char *buf, sector_t *sectors)
2503 unsigned long long blocks;
2506 if (strict_strtoull(buf, 10, &blocks) < 0)
2509 if (blocks & 1ULL << (8 * sizeof(blocks) - 1))
2510 return -EINVAL; /* sector conversion overflow */
2513 if (new != blocks * 2)
2514 return -EINVAL; /* unsigned long long to sector_t overflow */
2521 rdev_size_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2523 mddev_t *my_mddev = rdev->mddev;
2524 sector_t oldsectors = rdev->sectors;
2527 if (strict_blocks_to_sectors(buf, §ors) < 0)
2529 if (my_mddev->pers && rdev->raid_disk >= 0) {
2530 if (my_mddev->persistent) {
2531 sectors = super_types[my_mddev->major_version].
2532 rdev_size_change(rdev, sectors);
2535 } else if (!sectors)
2536 sectors = (rdev->bdev->bd_inode->i_size >> 9) -
2539 if (sectors < my_mddev->dev_sectors)
2540 return -EINVAL; /* component must fit device */
2542 rdev->sectors = sectors;
2543 if (sectors > oldsectors && my_mddev->external) {
2544 /* need to check that all other rdevs with the same ->bdev
2545 * do not overlap. We need to unlock the mddev to avoid
2546 * a deadlock. We have already changed rdev->sectors, and if
2547 * we have to change it back, we will have the lock again.
2551 struct list_head *tmp;
2553 mddev_unlock(my_mddev);
2554 for_each_mddev(mddev, tmp) {
2558 list_for_each_entry(rdev2, &mddev->disks, same_set)
2559 if (test_bit(AllReserved, &rdev2->flags) ||
2560 (rdev->bdev == rdev2->bdev &&
2562 overlaps(rdev->data_offset, rdev->sectors,
2568 mddev_unlock(mddev);
2574 mddev_lock(my_mddev);
2576 /* Someone else could have slipped in a size
2577 * change here, but doing so is just silly.
2578 * We put oldsectors back because we *know* it is
2579 * safe, and trust userspace not to race with
2582 rdev->sectors = oldsectors;
2589 static struct rdev_sysfs_entry rdev_size =
2590 __ATTR(size, S_IRUGO|S_IWUSR, rdev_size_show, rdev_size_store);
2593 static ssize_t recovery_start_show(mdk_rdev_t *rdev, char *page)
2595 unsigned long long recovery_start = rdev->recovery_offset;
2597 if (test_bit(In_sync, &rdev->flags) ||
2598 recovery_start == MaxSector)
2599 return sprintf(page, "none\n");
2601 return sprintf(page, "%llu\n", recovery_start);
2604 static ssize_t recovery_start_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2606 unsigned long long recovery_start;
2608 if (cmd_match(buf, "none"))
2609 recovery_start = MaxSector;
2610 else if (strict_strtoull(buf, 10, &recovery_start))
2613 if (rdev->mddev->pers &&
2614 rdev->raid_disk >= 0)
2617 rdev->recovery_offset = recovery_start;
2618 if (recovery_start == MaxSector)
2619 set_bit(In_sync, &rdev->flags);
2621 clear_bit(In_sync, &rdev->flags);
2625 static struct rdev_sysfs_entry rdev_recovery_start =
2626 __ATTR(recovery_start, S_IRUGO|S_IWUSR, recovery_start_show, recovery_start_store);
2628 static struct attribute *rdev_default_attrs[] = {
2634 &rdev_recovery_start.attr,
2638 rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
2640 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
2641 mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
2642 mddev_t *mddev = rdev->mddev;
2648 rv = mddev ? mddev_lock(mddev) : -EBUSY;
2650 if (rdev->mddev == NULL)
2653 rv = entry->show(rdev, page);
2654 mddev_unlock(mddev);
2660 rdev_attr_store(struct kobject *kobj, struct attribute *attr,
2661 const char *page, size_t length)
2663 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
2664 mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
2666 mddev_t *mddev = rdev->mddev;
2670 if (!capable(CAP_SYS_ADMIN))
2672 rv = mddev ? mddev_lock(mddev): -EBUSY;
2674 if (rdev->mddev == NULL)
2677 rv = entry->store(rdev, page, length);
2678 mddev_unlock(mddev);
2683 static void rdev_free(struct kobject *ko)
2685 mdk_rdev_t *rdev = container_of(ko, mdk_rdev_t, kobj);
2688 static struct sysfs_ops rdev_sysfs_ops = {
2689 .show = rdev_attr_show,
2690 .store = rdev_attr_store,
2692 static struct kobj_type rdev_ktype = {
2693 .release = rdev_free,
2694 .sysfs_ops = &rdev_sysfs_ops,
2695 .default_attrs = rdev_default_attrs,
2699 * Import a device. If 'super_format' >= 0, then sanity check the superblock
2701 * mark the device faulty if:
2703 * - the device is nonexistent (zero size)
2704 * - the device has no valid superblock
2706 * a faulty rdev _never_ has rdev->sb set.
2708 static mdk_rdev_t *md_import_device(dev_t newdev, int super_format, int super_minor)
2710 char b[BDEVNAME_SIZE];
2715 rdev = kzalloc(sizeof(*rdev), GFP_KERNEL);
2717 printk(KERN_ERR "md: could not alloc mem for new device!\n");
2718 return ERR_PTR(-ENOMEM);
2721 if ((err = alloc_disk_sb(rdev)))
2724 err = lock_rdev(rdev, newdev, super_format == -2);
2728 kobject_init(&rdev->kobj, &rdev_ktype);
2731 rdev->saved_raid_disk = -1;
2732 rdev->raid_disk = -1;
2734 rdev->data_offset = 0;
2735 rdev->sb_events = 0;
2736 rdev->last_read_error.tv_sec = 0;
2737 rdev->last_read_error.tv_nsec = 0;
2738 atomic_set(&rdev->nr_pending, 0);
2739 atomic_set(&rdev->read_errors, 0);
2740 atomic_set(&rdev->corrected_errors, 0);
2742 size = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
2745 "md: %s has zero or unknown size, marking faulty!\n",
2746 bdevname(rdev->bdev,b));
2751 if (super_format >= 0) {
2752 err = super_types[super_format].
2753 load_super(rdev, NULL, super_minor);
2754 if (err == -EINVAL) {
2756 "md: %s does not have a valid v%d.%d "
2757 "superblock, not importing!\n",
2758 bdevname(rdev->bdev,b),
2759 super_format, super_minor);
2764 "md: could not read %s's sb, not importing!\n",
2765 bdevname(rdev->bdev,b));
2770 INIT_LIST_HEAD(&rdev->same_set);
2771 init_waitqueue_head(&rdev->blocked_wait);
2776 if (rdev->sb_page) {
2782 return ERR_PTR(err);
2786 * Check a full RAID array for plausibility
2790 static void analyze_sbs(mddev_t * mddev)
2793 mdk_rdev_t *rdev, *freshest, *tmp;
2794 char b[BDEVNAME_SIZE];
2797 rdev_for_each(rdev, tmp, mddev)
2798 switch (super_types[mddev->major_version].
2799 load_super(rdev, freshest, mddev->minor_version)) {
2807 "md: fatal superblock inconsistency in %s"
2808 " -- removing from array\n",
2809 bdevname(rdev->bdev,b));
2810 kick_rdev_from_array(rdev);
2814 super_types[mddev->major_version].
2815 validate_super(mddev, freshest);
2818 rdev_for_each(rdev, tmp, mddev) {
2819 if (mddev->max_disks &&
2820 (rdev->desc_nr >= mddev->max_disks ||
2821 i > mddev->max_disks)) {
2823 "md: %s: %s: only %d devices permitted\n",
2824 mdname(mddev), bdevname(rdev->bdev, b),
2826 kick_rdev_from_array(rdev);
2829 if (rdev != freshest)
2830 if (super_types[mddev->major_version].
2831 validate_super(mddev, rdev)) {
2832 printk(KERN_WARNING "md: kicking non-fresh %s"
2834 bdevname(rdev->bdev,b));
2835 kick_rdev_from_array(rdev);
2838 if (mddev->level == LEVEL_MULTIPATH) {
2839 rdev->desc_nr = i++;
2840 rdev->raid_disk = rdev->desc_nr;
2841 set_bit(In_sync, &rdev->flags);
2842 } else if (rdev->raid_disk >= (mddev->raid_disks - min(0, mddev->delta_disks))) {
2843 rdev->raid_disk = -1;
2844 clear_bit(In_sync, &rdev->flags);
2849 /* Read a fixed-point number.
2850 * Numbers in sysfs attributes should be in "standard" units where
2851 * possible, so time should be in seconds.
2852 * However we internally use a a much smaller unit such as
2853 * milliseconds or jiffies.
2854 * This function takes a decimal number with a possible fractional
2855 * component, and produces an integer which is the result of
2856 * multiplying that number by 10^'scale'.
2857 * all without any floating-point arithmetic.
2859 int strict_strtoul_scaled(const char *cp, unsigned long *res, int scale)
2861 unsigned long result = 0;
2863 while (isdigit(*cp) || (*cp == '.' && decimals < 0)) {
2866 else if (decimals < scale) {
2869 result = result * 10 + value;
2881 while (decimals < scale) {
2890 static void md_safemode_timeout(unsigned long data);
2893 safe_delay_show(mddev_t *mddev, char *page)
2895 int msec = (mddev->safemode_delay*1000)/HZ;
2896 return sprintf(page, "%d.%03d\n", msec/1000, msec%1000);
2899 safe_delay_store(mddev_t *mddev, const char *cbuf, size_t len)
2903 if (strict_strtoul_scaled(cbuf, &msec, 3) < 0)
2906 mddev->safemode_delay = 0;
2908 unsigned long old_delay = mddev->safemode_delay;
2909 mddev->safemode_delay = (msec*HZ)/1000;
2910 if (mddev->safemode_delay == 0)
2911 mddev->safemode_delay = 1;
2912 if (mddev->safemode_delay < old_delay)
2913 md_safemode_timeout((unsigned long)mddev);
2917 static struct md_sysfs_entry md_safe_delay =
2918 __ATTR(safe_mode_delay, S_IRUGO|S_IWUSR,safe_delay_show, safe_delay_store);
2921 level_show(mddev_t *mddev, char *page)
2923 struct mdk_personality *p = mddev->pers;
2925 return sprintf(page, "%s\n", p->name);
2926 else if (mddev->clevel[0])
2927 return sprintf(page, "%s\n", mddev->clevel);
2928 else if (mddev->level != LEVEL_NONE)
2929 return sprintf(page, "%d\n", mddev->level);
2935 level_store(mddev_t *mddev, const char *buf, size_t len)
2939 struct mdk_personality *pers;
2943 if (mddev->pers == NULL) {
2946 if (len >= sizeof(mddev->clevel))
2948 strncpy(mddev->clevel, buf, len);
2949 if (mddev->clevel[len-1] == '\n')
2951 mddev->clevel[len] = 0;
2952 mddev->level = LEVEL_NONE;
2956 /* request to change the personality. Need to ensure:
2957 * - array is not engaged in resync/recovery/reshape
2958 * - old personality can be suspended
2959 * - new personality will access other array.
2962 if (mddev->sync_thread || mddev->reshape_position != MaxSector)
2965 if (!mddev->pers->quiesce) {
2966 printk(KERN_WARNING "md: %s: %s does not support online personality change\n",
2967 mdname(mddev), mddev->pers->name);
2971 /* Now find the new personality */
2972 if (len == 0 || len >= sizeof(level))
2974 strncpy(level, buf, len);
2975 if (level[len-1] == '\n')
2979 request_module("md-%s", level);
2980 spin_lock(&pers_lock);
2981 pers = find_pers(LEVEL_NONE, level);
2982 if (!pers || !try_module_get(pers->owner)) {
2983 spin_unlock(&pers_lock);
2984 printk(KERN_WARNING "md: personality %s not loaded\n", level);
2987 spin_unlock(&pers_lock);
2989 if (pers == mddev->pers) {
2990 /* Nothing to do! */
2991 module_put(pers->owner);
2994 if (!pers->takeover) {
2995 module_put(pers->owner);
2996 printk(KERN_WARNING "md: %s: %s does not support personality takeover\n",
2997 mdname(mddev), level);
3001 /* ->takeover must set new_* and/or delta_disks
3002 * if it succeeds, and may set them when it fails.
3004 priv = pers->takeover(mddev);
3006 mddev->new_level = mddev->level;
3007 mddev->new_layout = mddev->layout;
3008 mddev->new_chunk_sectors = mddev->chunk_sectors;
3009 mddev->raid_disks -= mddev->delta_disks;
3010 mddev->delta_disks = 0;
3011 module_put(pers->owner);
3012 printk(KERN_WARNING "md: %s: %s would not accept array\n",
3013 mdname(mddev), level);
3014 return PTR_ERR(priv);
3017 /* Looks like we have a winner */
3018 mddev_suspend(mddev);
3019 mddev->pers->stop(mddev);
3021 if (mddev->pers->sync_request == NULL &&
3022 pers->sync_request != NULL) {
3023 /* need to add the md_redundancy_group */
3024 if (sysfs_create_group(&mddev->kobj, &md_redundancy_group))
3026 "md: cannot register extra attributes for %s\n",
3028 mddev->sysfs_action = sysfs_get_dirent(mddev->kobj.sd, "sync_action");
3030 if (mddev->pers->sync_request != NULL &&
3031 pers->sync_request == NULL) {
3032 /* need to remove the md_redundancy_group */
3033 if (mddev->to_remove == NULL)
3034 mddev->to_remove = &md_redundancy_group;
3037 if (mddev->pers->sync_request == NULL &&
3039 /* We are converting from a no-redundancy array
3040 * to a redundancy array and metadata is managed
3041 * externally so we need to be sure that writes
3042 * won't block due to a need to transition
3044 * until external management is started.
3047 mddev->safemode_delay = 0;
3048 mddev->safemode = 0;
3051 module_put(mddev->pers->owner);
3052 /* Invalidate devices that are now superfluous */
3053 list_for_each_entry(rdev, &mddev->disks, same_set)
3054 if (rdev->raid_disk >= mddev->raid_disks) {
3055 rdev->raid_disk = -1;
3056 clear_bit(In_sync, &rdev->flags);
3059 mddev->private = priv;
3060 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
3061 mddev->level = mddev->new_level;
3062 mddev->layout = mddev->new_layout;
3063 mddev->chunk_sectors = mddev->new_chunk_sectors;
3064 mddev->delta_disks = 0;
3065 if (mddev->pers->sync_request == NULL) {
3066 /* this is now an array without redundancy, so
3067 * it must always be in_sync
3070 del_timer_sync(&mddev->safemode_timer);
3073 mddev_resume(mddev);
3074 set_bit(MD_CHANGE_DEVS, &mddev->flags);
3075 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3076 md_wakeup_thread(mddev->thread);
3077 sysfs_notify(&mddev->kobj, NULL, "level");
3078 md_new_event(mddev);
3082 static struct md_sysfs_entry md_level =
3083 __ATTR(level, S_IRUGO|S_IWUSR, level_show, level_store);
3087 layout_show(mddev_t *mddev, char *page)
3089 /* just a number, not meaningful for all levels */
3090 if (mddev->reshape_position != MaxSector &&
3091 mddev->layout != mddev->new_layout)
3092 return sprintf(page, "%d (%d)\n",
3093 mddev->new_layout, mddev->layout);
3094 return sprintf(page, "%d\n", mddev->layout);
3098 layout_store(mddev_t *mddev, const char *buf, size_t len)
3101 unsigned long n = simple_strtoul(buf, &e, 10);
3103 if (!*buf || (*e && *e != '\n'))
3108 if (mddev->pers->check_reshape == NULL)
3110 mddev->new_layout = n;
3111 err = mddev->pers->check_reshape(mddev);
3113 mddev->new_layout = mddev->layout;
3117 mddev->new_layout = n;
3118 if (mddev->reshape_position == MaxSector)
3123 static struct md_sysfs_entry md_layout =
3124 __ATTR(layout, S_IRUGO|S_IWUSR, layout_show, layout_store);
3128 raid_disks_show(mddev_t *mddev, char *page)
3130 if (mddev->raid_disks == 0)
3132 if (mddev->reshape_position != MaxSector &&
3133 mddev->delta_disks != 0)
3134 return sprintf(page, "%d (%d)\n", mddev->raid_disks,
3135 mddev->raid_disks - mddev->delta_disks);
3136 return sprintf(page, "%d\n", mddev->raid_disks);
3139 static int update_raid_disks(mddev_t *mddev, int raid_disks);
3142 raid_disks_store(mddev_t *mddev, const char *buf, size_t len)
3146 unsigned long n = simple_strtoul(buf, &e, 10);
3148 if (!*buf || (*e && *e != '\n'))
3152 rv = update_raid_disks(mddev, n);
3153 else if (mddev->reshape_position != MaxSector) {
3154 int olddisks = mddev->raid_disks - mddev->delta_disks;
3155 mddev->delta_disks = n - olddisks;
3156 mddev->raid_disks = n;
3158 mddev->raid_disks = n;
3159 return rv ? rv : len;
3161 static struct md_sysfs_entry md_raid_disks =
3162 __ATTR(raid_disks, S_IRUGO|S_IWUSR, raid_disks_show, raid_disks_store);
3165 chunk_size_show(mddev_t *mddev, char *page)
3167 if (mddev->reshape_position != MaxSector &&
3168 mddev->chunk_sectors != mddev->new_chunk_sectors)
3169 return sprintf(page, "%d (%d)\n",
3170 mddev->new_chunk_sectors << 9,
3171 mddev->chunk_sectors << 9);
3172 return sprintf(page, "%d\n", mddev->chunk_sectors << 9);
3176 chunk_size_store(mddev_t *mddev, const char *buf, size_t len)
3179 unsigned long n = simple_strtoul(buf, &e, 10);
3181 if (!*buf || (*e && *e != '\n'))
3186 if (mddev->pers->check_reshape == NULL)
3188 mddev->new_chunk_sectors = n >> 9;
3189 err = mddev->pers->check_reshape(mddev);
3191 mddev->new_chunk_sectors = mddev->chunk_sectors;
3195 mddev->new_chunk_sectors = n >> 9;
3196 if (mddev->reshape_position == MaxSector)
3197 mddev->chunk_sectors = n >> 9;
3201 static struct md_sysfs_entry md_chunk_size =
3202 __ATTR(chunk_size, S_IRUGO|S_IWUSR, chunk_size_show, chunk_size_store);
3205 resync_start_show(mddev_t *mddev, char *page)
3207 if (mddev->recovery_cp == MaxSector)
3208 return sprintf(page, "none\n");
3209 return sprintf(page, "%llu\n", (unsigned long long)mddev->recovery_cp);
3213 resync_start_store(mddev_t *mddev, const char *buf, size_t len)
3216 unsigned long long n = simple_strtoull(buf, &e, 10);
3220 if (cmd_match(buf, "none"))
3222 else if (!*buf || (*e && *e != '\n'))
3225 mddev->recovery_cp = n;
3228 static struct md_sysfs_entry md_resync_start =
3229 __ATTR(resync_start, S_IRUGO|S_IWUSR, resync_start_show, resync_start_store);
3232 * The array state can be:
3235 * No devices, no size, no level
3236 * Equivalent to STOP_ARRAY ioctl
3238 * May have some settings, but array is not active
3239 * all IO results in error
3240 * When written, doesn't tear down array, but just stops it
3241 * suspended (not supported yet)
3242 * All IO requests will block. The array can be reconfigured.
3243 * Writing this, if accepted, will block until array is quiescent
3245 * no resync can happen. no superblocks get written.
3246 * write requests fail
3248 * like readonly, but behaves like 'clean' on a write request.
3250 * clean - no pending writes, but otherwise active.
3251 * When written to inactive array, starts without resync
3252 * If a write request arrives then
3253 * if metadata is known, mark 'dirty' and switch to 'active'.
3254 * if not known, block and switch to write-pending
3255 * If written to an active array that has pending writes, then fails.
3257 * fully active: IO and resync can be happening.
3258 * When written to inactive array, starts with resync
3261 * clean, but writes are blocked waiting for 'active' to be written.
3264 * like active, but no writes have been seen for a while (100msec).
3267 enum array_state { clear, inactive, suspended, readonly, read_auto, clean, active,
3268 write_pending, active_idle, bad_word};
3269 static char *array_states[] = {
3270 "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active",
3271 "write-pending", "active-idle", NULL };
3273 static int match_word(const char *word, char **list)
3276 for (n=0; list[n]; n++)
3277 if (cmd_match(word, list[n]))
3283 array_state_show(mddev_t *mddev, char *page)
3285 enum array_state st = inactive;
3298 else if (test_bit(MD_CHANGE_CLEAN, &mddev->flags))
3300 else if (mddev->safemode)
3306 if (list_empty(&mddev->disks) &&
3307 mddev->raid_disks == 0 &&
3308 mddev->dev_sectors == 0)
3313 return sprintf(page, "%s\n", array_states[st]);
3316 static int do_md_stop(mddev_t * mddev, int ro, int is_open);
3317 static int md_set_readonly(mddev_t * mddev, int is_open);
3318 static int do_md_run(mddev_t * mddev);
3319 static int restart_array(mddev_t *mddev);
3322 array_state_store(mddev_t *mddev, const char *buf, size_t len)
3325 enum array_state st = match_word(buf, array_states);
3330 /* stopping an active array */
3331 if (atomic_read(&mddev->openers) > 0)
3333 err = do_md_stop(mddev, 0, 0);
3336 /* stopping an active array */
3338 if (atomic_read(&mddev->openers) > 0)
3340 err = do_md_stop(mddev, 2, 0);
3342 err = 0; /* already inactive */
3345 break; /* not supported yet */
3348 err = md_set_readonly(mddev, 0);
3351 set_disk_ro(mddev->gendisk, 1);
3352 err = do_md_run(mddev);
3358 err = md_set_readonly(mddev, 0);
3359 else if (mddev->ro == 1)
3360 err = restart_array(mddev);
3363 set_disk_ro(mddev->gendisk, 0);
3367 err = do_md_run(mddev);
3372 restart_array(mddev);
3373 spin_lock_irq(&mddev->write_lock);
3374 if (atomic_read(&mddev->writes_pending) == 0) {
3375 if (mddev->in_sync == 0) {
3377 if (mddev->safemode == 1)
3378 mddev->safemode = 0;
3379 if (mddev->persistent)
3380 set_bit(MD_CHANGE_CLEAN,
3386 spin_unlock_irq(&mddev->write_lock);
3392 restart_array(mddev);
3393 if (mddev->external)
3394 clear_bit(MD_CHANGE_CLEAN, &mddev->flags);
3395 wake_up(&mddev->sb_wait);
3399 set_disk_ro(mddev->gendisk, 0);
3400 err = do_md_run(mddev);
3405 /* these cannot be set */
3411 sysfs_notify_dirent(mddev->sysfs_state);
3415 static struct md_sysfs_entry md_array_state =
3416 __ATTR(array_state, S_IRUGO|S_IWUSR, array_state_show, array_state_store);
3419 max_corrected_read_errors_show(mddev_t *mddev, char *page) {
3420 return sprintf(page, "%d\n",
3421 atomic_read(&mddev->max_corr_read_errors));
3425 max_corrected_read_errors_store(mddev_t *mddev, const char *buf, size_t len)
3428 unsigned long n = simple_strtoul(buf, &e, 10);
3430 if (*buf && (*e == 0 || *e == '\n')) {
3431 atomic_set(&mddev->max_corr_read_errors, n);
3437 static struct md_sysfs_entry max_corr_read_errors =
3438 __ATTR(max_read_errors, S_IRUGO|S_IWUSR, max_corrected_read_errors_show,
3439 max_corrected_read_errors_store);
3442 null_show(mddev_t *mddev, char *page)
3448 new_dev_store(mddev_t *mddev, const char *buf, size_t len)
3450 /* buf must be %d:%d\n? giving major and minor numbers */
3451 /* The new device is added to the array.
3452 * If the array has a persistent superblock, we read the
3453 * superblock to initialise info and check validity.
3454 * Otherwise, only checking done is that in bind_rdev_to_array,
3455 * which mainly checks size.
3458 int major = simple_strtoul(buf, &e, 10);
3464 if (!*buf || *e != ':' || !e[1] || e[1] == '\n')
3466 minor = simple_strtoul(e+1, &e, 10);
3467 if (*e && *e != '\n')
3469 dev = MKDEV(major, minor);
3470 if (major != MAJOR(dev) ||
3471 minor != MINOR(dev))
3475 if (mddev->persistent) {
3476 rdev = md_import_device(dev, mddev->major_version,
3477 mddev->minor_version);
3478 if (!IS_ERR(rdev) && !list_empty(&mddev->disks)) {
3479 mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
3480 mdk_rdev_t, same_set);
3481 err = super_types[mddev->major_version]
3482 .load_super(rdev, rdev0, mddev->minor_version);
3486 } else if (mddev->external)
3487 rdev = md_import_device(dev, -2, -1);
3489 rdev = md_import_device(dev, -1, -1);
3492 return PTR_ERR(rdev);
3493 err = bind_rdev_to_array(rdev, mddev);
3497 return err ? err : len;
3500 static struct md_sysfs_entry md_new_device =
3501 __ATTR(new_dev, S_IWUSR, null_show, new_dev_store);
3504 bitmap_store(mddev_t *mddev, const char *buf, size_t len)
3507 unsigned long chunk, end_chunk;
3511 /* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */
3513 chunk = end_chunk = simple_strtoul(buf, &end, 0);
3514 if (buf == end) break;
3515 if (*end == '-') { /* range */
3517 end_chunk = simple_strtoul(buf, &end, 0);
3518 if (buf == end) break;
3520 if (*end && !isspace(*end)) break;
3521 bitmap_dirty_bits(mddev->bitmap, chunk, end_chunk);
3522 buf = skip_spaces(end);
3524 bitmap_unplug(mddev->bitmap); /* flush the bits to disk */
3529 static struct md_sysfs_entry md_bitmap =
3530 __ATTR(bitmap_set_bits, S_IWUSR, null_show, bitmap_store);
3533 size_show(mddev_t *mddev, char *page)
3535 return sprintf(page, "%llu\n",
3536 (unsigned long long)mddev->dev_sectors / 2);
3539 static int update_size(mddev_t *mddev, sector_t num_sectors);
3542 size_store(mddev_t *mddev, const char *buf, size_t len)
3544 /* If array is inactive, we can reduce the component size, but
3545 * not increase it (except from 0).
3546 * If array is active, we can try an on-line resize
3549 int err = strict_blocks_to_sectors(buf, §ors);
3554 err = update_size(mddev, sectors);
3555 md_update_sb(mddev, 1);
3557 if (mddev->dev_sectors == 0 ||
3558 mddev->dev_sectors > sectors)
3559 mddev->dev_sectors = sectors;
3563 return err ? err : len;
3566 static struct md_sysfs_entry md_size =
3567 __ATTR(component_size, S_IRUGO|S_IWUSR, size_show, size_store);
3572 * 'none' for arrays with no metadata (good luck...)
3573 * 'external' for arrays with externally managed metadata,
3574 * or N.M for internally known formats
3577 metadata_show(mddev_t *mddev, char *page)
3579 if (mddev->persistent)
3580 return sprintf(page, "%d.%d\n",
3581 mddev->major_version, mddev->minor_version);
3582 else if (mddev->external)
3583 return sprintf(page, "external:%s\n", mddev->metadata_type);
3585 return sprintf(page, "none\n");
3589 metadata_store(mddev_t *mddev, const char *buf, size_t len)
3593 /* Changing the details of 'external' metadata is
3594 * always permitted. Otherwise there must be
3595 * no devices attached to the array.
3597 if (mddev->external && strncmp(buf, "external:", 9) == 0)
3599 else if (!list_empty(&mddev->disks))
3602 if (cmd_match(buf, "none")) {
3603 mddev->persistent = 0;
3604 mddev->external = 0;
3605 mddev->major_version = 0;
3606 mddev->minor_version = 90;
3609 if (strncmp(buf, "external:", 9) == 0) {
3610 size_t namelen = len-9;
3611 if (namelen >= sizeof(mddev->metadata_type))
3612 namelen = sizeof(mddev->metadata_type)-1;
3613 strncpy(mddev->metadata_type, buf+9, namelen);
3614 mddev->metadata_type[namelen] = 0;
3615 if (namelen && mddev->metadata_type[namelen-1] == '\n')
3616 mddev->metadata_type[--namelen] = 0;
3617 mddev->persistent = 0;
3618 mddev->external = 1;
3619 mddev->major_version = 0;
3620 mddev->minor_version = 90;
3623 major = simple_strtoul(buf, &e, 10);
3624 if (e==buf || *e != '.')
3627 minor = simple_strtoul(buf, &e, 10);
3628 if (e==buf || (*e && *e != '\n') )
3630 if (major >= ARRAY_SIZE(super_types) || super_types[major].name == NULL)
3632 mddev->major_version = major;
3633 mddev->minor_version = minor;
3634 mddev->persistent = 1;
3635 mddev->external = 0;
3639 static struct md_sysfs_entry md_metadata =
3640 __ATTR(metadata_version, S_IRUGO|S_IWUSR, metadata_show, metadata_store);
3643 action_show(mddev_t *mddev, char *page)
3645 char *type = "idle";
3646 if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
3648 else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
3649 (!mddev->ro && test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))) {
3650 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
3652 else if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
3653 if (!test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
3655 else if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
3659 } else if (test_bit(MD_RECOVERY_RECOVER, &mddev->recovery))
3662 return sprintf(page, "%s\n", type);
3666 action_store(mddev_t *mddev, const char *page, size_t len)
3668 if (!mddev->pers || !mddev->pers->sync_request)
3671 if (cmd_match(page, "frozen"))
3672 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
3674 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
3676 if (cmd_match(page, "idle") || cmd_match(page, "frozen")) {
3677 if (mddev->sync_thread) {
3678 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
3679 md_unregister_thread(mddev->sync_thread);
3680 mddev->sync_thread = NULL;
3681 mddev->recovery = 0;
3683 } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
3684 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
3686 else if (cmd_match(page, "resync"))
3687 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3688 else if (cmd_match(page, "recover")) {
3689 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
3690 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3691 } else if (cmd_match(page, "reshape")) {
3693 if (mddev->pers->start_reshape == NULL)
3695 err = mddev->pers->start_reshape(mddev);
3698 sysfs_notify(&mddev->kobj, NULL, "degraded");
3700 if (cmd_match(page, "check"))
3701 set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
3702 else if (!cmd_match(page, "repair"))
3704 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
3705 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
3707 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3708 md_wakeup_thread(mddev->thread);
3709 sysfs_notify_dirent(mddev->sysfs_action);
3714 mismatch_cnt_show(mddev_t *mddev, char *page)
3716 return sprintf(page, "%llu\n",
3717 (unsigned long long) mddev->resync_mismatches);
3720 static struct md_sysfs_entry md_scan_mode =
3721 __ATTR(sync_action, S_IRUGO|S_IWUSR, action_show, action_store);
3724 static struct md_sysfs_entry md_mismatches = __ATTR_RO(mismatch_cnt);
3727 sync_min_show(mddev_t *mddev, char *page)
3729 return sprintf(page, "%d (%s)\n", speed_min(mddev),
3730 mddev->sync_speed_min ? "local": "system");
3734 sync_min_store(mddev_t *mddev, const char *buf, size_t len)
3738 if (strncmp(buf, "system", 6)==0) {
3739 mddev->sync_speed_min = 0;
3742 min = simple_strtoul(buf, &e, 10);
3743 if (buf == e || (*e && *e != '\n') || min <= 0)
3745 mddev->sync_speed_min = min;
3749 static struct md_sysfs_entry md_sync_min =
3750 __ATTR(sync_speed_min, S_IRUGO|S_IWUSR, sync_min_show, sync_min_store);
3753 sync_max_show(mddev_t *mddev, char *page)
3755 return sprintf(page, "%d (%s)\n", speed_max(mddev),
3756 mddev->sync_speed_max ? "local": "system");
3760 sync_max_store(mddev_t *mddev, const char *buf, size_t len)
3764 if (strncmp(buf, "system", 6)==0) {
3765 mddev->sync_speed_max = 0;
3768 max = simple_strtoul(buf, &e, 10);
3769 if (buf == e || (*e && *e != '\n') || max <= 0)
3771 mddev->sync_speed_max = max;
3775 static struct md_sysfs_entry md_sync_max =
3776 __ATTR(sync_speed_max, S_IRUGO|S_IWUSR, sync_max_show, sync_max_store);
3779 degraded_show(mddev_t *mddev, char *page)
3781 return sprintf(page, "%d\n", mddev->degraded);
3783 static struct md_sysfs_entry md_degraded = __ATTR_RO(degraded);
3786 sync_force_parallel_show(mddev_t *mddev, char *page)
3788 return sprintf(page, "%d\n", mddev->parallel_resync);
3792 sync_force_parallel_store(mddev_t *mddev, const char *buf, size_t len)
3796 if (strict_strtol(buf, 10, &n))
3799 if (n != 0 && n != 1)
3802 mddev->parallel_resync = n;
3804 if (mddev->sync_thread)
3805 wake_up(&resync_wait);
3810 /* force parallel resync, even with shared block devices */
3811 static struct md_sysfs_entry md_sync_force_parallel =
3812 __ATTR(sync_force_parallel, S_IRUGO|S_IWUSR,
3813 sync_force_parallel_show, sync_force_parallel_store);
3816 sync_speed_show(mddev_t *mddev, char *page)
3818 unsigned long resync, dt, db;
3819 if (mddev->curr_resync == 0)
3820 return sprintf(page, "none\n");
3821 resync = mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active);
3822 dt = (jiffies - mddev->resync_mark) / HZ;
3824 db = resync - mddev->resync_mark_cnt;
3825 return sprintf(page, "%lu\n", db/dt/2); /* K/sec */
3828 static struct md_sysfs_entry md_sync_speed = __ATTR_RO(sync_speed);
3831 sync_completed_show(mddev_t *mddev, char *page)
3833 unsigned long max_sectors, resync;
3835 if (!test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
3836 return sprintf(page, "none\n");
3838 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
3839 max_sectors = mddev->resync_max_sectors;
3841 max_sectors = mddev->dev_sectors;
3843 resync = mddev->curr_resync_completed;
3844 return sprintf(page, "%lu / %lu\n", resync, max_sectors);
3847 static struct md_sysfs_entry md_sync_completed = __ATTR_RO(sync_completed);
3850 min_sync_show(mddev_t *mddev, char *page)
3852 return sprintf(page, "%llu\n",
3853 (unsigned long long)mddev->resync_min);
3856 min_sync_store(mddev_t *mddev, const char *buf, size_t len)
3858 unsigned long long min;
3859 if (strict_strtoull(buf, 10, &min))
3861 if (min > mddev->resync_max)
3863 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
3866 /* Must be a multiple of chunk_size */
3867 if (mddev->chunk_sectors) {
3868 sector_t temp = min;
3869 if (sector_div(temp, mddev->chunk_sectors))
3872 mddev->resync_min = min;
3877 static struct md_sysfs_entry md_min_sync =
3878 __ATTR(sync_min, S_IRUGO|S_IWUSR, min_sync_show, min_sync_store);
3881 max_sync_show(mddev_t *mddev, char *page)
3883 if (mddev->resync_max == MaxSector)
3884 return sprintf(page, "max\n");
3886 return sprintf(page, "%llu\n",
3887 (unsigned long long)mddev->resync_max);
3890 max_sync_store(mddev_t *mddev, const char *buf, size_t len)
3892 if (strncmp(buf, "max", 3) == 0)
3893 mddev->resync_max = MaxSector;
3895 unsigned long long max;
3896 if (strict_strtoull(buf, 10, &max))
3898 if (max < mddev->resync_min)
3900 if (max < mddev->resync_max &&
3902 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
3905 /* Must be a multiple of chunk_size */
3906 if (mddev->chunk_sectors) {
3907 sector_t temp = max;
3908 if (sector_div(temp, mddev->chunk_sectors))
3911 mddev->resync_max = max;
3913 wake_up(&mddev->recovery_wait);
3917 static struct md_sysfs_entry md_max_sync =
3918 __ATTR(sync_max, S_IRUGO|S_IWUSR, max_sync_show, max_sync_store);
3921 suspend_lo_show(mddev_t *mddev, char *page)
3923 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_lo);
3927 suspend_lo_store(mddev_t *mddev, const char *buf, size_t len)
3930 unsigned long long new = simple_strtoull(buf, &e, 10);
3932 if (mddev->pers == NULL ||
3933 mddev->pers->quiesce == NULL)
3935 if (buf == e || (*e && *e != '\n'))
3937 if (new >= mddev->suspend_hi ||
3938 (new > mddev->suspend_lo && new < mddev->suspend_hi)) {
3939 mddev->suspend_lo = new;
3940 mddev->pers->quiesce(mddev, 2);
3945 static struct md_sysfs_entry md_suspend_lo =
3946 __ATTR(suspend_lo, S_IRUGO|S_IWUSR, suspend_lo_show, suspend_lo_store);
3950 suspend_hi_show(mddev_t *mddev, char *page)
3952 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_hi);
3956 suspend_hi_store(mddev_t *mddev, const char *buf, size_t len)
3959 unsigned long long new = simple_strtoull(buf, &e, 10);
3961 if (mddev->pers == NULL ||
3962 mddev->pers->quiesce == NULL)
3964 if (buf == e || (*e && *e != '\n'))
3966 if ((new <= mddev->suspend_lo && mddev->suspend_lo >= mddev->suspend_hi) ||
3967 (new > mddev->suspend_lo && new > mddev->suspend_hi)) {
3968 mddev->suspend_hi = new;
3969 mddev->pers->quiesce(mddev, 1);
3970 mddev->pers->quiesce(mddev, 0);
3975 static struct md_sysfs_entry md_suspend_hi =
3976 __ATTR(suspend_hi, S_IRUGO|S_IWUSR, suspend_hi_show, suspend_hi_store);
3979 reshape_position_show(mddev_t *mddev, char *page)
3981 if (mddev->reshape_position != MaxSector)
3982 return sprintf(page, "%llu\n",
3983 (unsigned long long)mddev->reshape_position);
3984 strcpy(page, "none\n");
3989 reshape_position_store(mddev_t *mddev, const char *buf, size_t len)
3992 unsigned long long new = simple_strtoull(buf, &e, 10);
3995 if (buf == e || (*e && *e != '\n'))
3997 mddev->reshape_position = new;
3998 mddev->delta_disks = 0;
3999 mddev->new_level = mddev->level;
4000 mddev->new_layout = mddev->layout;
4001 mddev->new_chunk_sectors = mddev->chunk_sectors;
4005 static struct md_sysfs_entry md_reshape_position =
4006 __ATTR(reshape_position, S_IRUGO|S_IWUSR, reshape_position_show,
4007 reshape_position_store);
4010 array_size_show(mddev_t *mddev, char *page)
4012 if (mddev->external_size)
4013 return sprintf(page, "%llu\n",
4014 (unsigned long long)mddev->array_sectors/2);
4016 return sprintf(page, "default\n");
4020 array_size_store(mddev_t *mddev, const char *buf, size_t len)
4024 if (strncmp(buf, "default", 7) == 0) {
4026 sectors = mddev->pers->size(mddev, 0, 0);
4028 sectors = mddev->array_sectors;
4030 mddev->external_size = 0;
4032 if (strict_blocks_to_sectors(buf, §ors) < 0)
4034 if (mddev->pers && mddev->pers->size(mddev, 0, 0) < sectors)
4037 mddev->external_size = 1;
4040 mddev->array_sectors = sectors;
4041 set_capacity(mddev->gendisk, mddev->array_sectors);
4043 revalidate_disk(mddev->gendisk);
4048 static struct md_sysfs_entry md_array_size =
4049 __ATTR(array_size, S_IRUGO|S_IWUSR, array_size_show,
4052 static struct attribute *md_default_attrs[] = {
4055 &md_raid_disks.attr,
4056 &md_chunk_size.attr,
4058 &md_resync_start.attr,
4060 &md_new_device.attr,
4061 &md_safe_delay.attr,
4062 &md_array_state.attr,
4063 &md_reshape_position.attr,
4064 &md_array_size.attr,
4065 &max_corr_read_errors.attr,
4069 static struct attribute *md_redundancy_attrs[] = {
4071 &md_mismatches.attr,
4074 &md_sync_speed.attr,
4075 &md_sync_force_parallel.attr,
4076 &md_sync_completed.attr,
4079 &md_suspend_lo.attr,
4080 &md_suspend_hi.attr,
4085 static struct attribute_group md_redundancy_group = {
4087 .attrs = md_redundancy_attrs,
4092 md_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
4094 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
4095 mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
4100 rv = mddev_lock(mddev);
4102 rv = entry->show(mddev, page);
4103 mddev_unlock(mddev);
4109 md_attr_store(struct kobject *kobj, struct attribute *attr,
4110 const char *page, size_t length)
4112 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
4113 mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
4118 if (!capable(CAP_SYS_ADMIN))
4120 rv = mddev_lock(mddev);
4121 if (mddev->hold_active == UNTIL_IOCTL)
4122 mddev->hold_active = 0;
4124 rv = entry->store(mddev, page, length);
4125 mddev_unlock(mddev);
4130 static void md_free(struct kobject *ko)
4132 mddev_t *mddev = container_of(ko, mddev_t, kobj);
4134 if (mddev->sysfs_state)
4135 sysfs_put(mddev->sysfs_state);
4137 if (mddev->gendisk) {
4138 del_gendisk(mddev->gendisk);
4139 put_disk(mddev->gendisk);
4142 blk_cleanup_queue(mddev->queue);
4147 static struct sysfs_ops md_sysfs_ops = {
4148 .show = md_attr_show,
4149 .store = md_attr_store,
4151 static struct kobj_type md_ktype = {
4153 .sysfs_ops = &md_sysfs_ops,
4154 .default_attrs = md_default_attrs,
4159 static void mddev_delayed_delete(struct work_struct *ws)
4161 mddev_t *mddev = container_of(ws, mddev_t, del_work);
4163 sysfs_remove_group(&mddev->kobj, &md_bitmap_group);
4164 kobject_del(&mddev->kobj);
4165 kobject_put(&mddev->kobj);
4168 static int md_alloc(dev_t dev, char *name)
4170 static DEFINE_MUTEX(disks_mutex);
4171 mddev_t *mddev = mddev_find(dev);
4172 struct gendisk *disk;
4181 partitioned = (MAJOR(mddev->unit) != MD_MAJOR);
4182 shift = partitioned ? MdpMinorShift : 0;
4183 unit = MINOR(mddev->unit) >> shift;
4185 /* wait for any previous instance if this device
4186 * to be completed removed (mddev_delayed_delete).
4188 flush_scheduled_work();
4190 mutex_lock(&disks_mutex);
4196 /* Need to ensure that 'name' is not a duplicate.
4199 spin_lock(&all_mddevs_lock);
4201 list_for_each_entry(mddev2, &all_mddevs, all_mddevs)
4202 if (mddev2->gendisk &&
4203 strcmp(mddev2->gendisk->disk_name, name) == 0) {
4204 spin_unlock(&all_mddevs_lock);
4207 spin_unlock(&all_mddevs_lock);
4211 mddev->queue = blk_alloc_queue(GFP_KERNEL);
4214 mddev->queue->queuedata = mddev;
4216 /* Can be unlocked because the queue is new: no concurrency */
4217 queue_flag_set_unlocked(QUEUE_FLAG_CLUSTER, mddev->queue);
4219 blk_queue_make_request(mddev->queue, md_make_request);
4221 disk = alloc_disk(1 << shift);
4223 blk_cleanup_queue(mddev->queue);
4224 mddev->queue = NULL;
4227 disk->major = MAJOR(mddev->unit);
4228 disk->first_minor = unit << shift;
4230 strcpy(disk->disk_name, name);
4231 else if (partitioned)
4232 sprintf(disk->disk_name, "md_d%d", unit);
4234 sprintf(disk->disk_name, "md%d", unit);
4235 disk->fops = &md_fops;
4236 disk->private_data = mddev;
4237 disk->queue = mddev->queue;
4238 /* Allow extended partitions. This makes the
4239 * 'mdp' device redundant, but we can't really
4242 disk->flags |= GENHD_FL_EXT_DEVT;
4244 mddev->gendisk = disk;
4245 error = kobject_init_and_add(&mddev->kobj, &md_ktype,
4246 &disk_to_dev(disk)->kobj, "%s", "md");
4248 /* This isn't possible, but as kobject_init_and_add is marked
4249 * __must_check, we must do something with the result
4251 printk(KERN_WARNING "md: cannot register %s/md - name in use\n",
4255 if (sysfs_create_group(&mddev->kobj, &md_bitmap_group))
4256 printk(KERN_DEBUG "pointless warning\n");
4258 mutex_unlock(&disks_mutex);
4260 kobject_uevent(&mddev->kobj, KOBJ_ADD);
4261 mddev->sysfs_state = sysfs_get_dirent(mddev->kobj.sd, "array_state");
4267 static struct kobject *md_probe(dev_t dev, int *part, void *data)
4269 md_alloc(dev, NULL);
4273 static int add_named_array(const char *val, struct kernel_param *kp)
4275 /* val must be "md_*" where * is not all digits.
4276 * We allocate an array with a large free minor number, and
4277 * set the name to val. val must not already be an active name.
4279 int len = strlen(val);
4280 char buf[DISK_NAME_LEN];
4282 while (len && val[len-1] == '\n')
4284 if (len >= DISK_NAME_LEN)
4286 strlcpy(buf, val, len+1);
4287 if (strncmp(buf, "md_", 3) != 0)
4289 return md_alloc(0, buf);
4292 static void md_safemode_timeout(unsigned long data)
4294 mddev_t *mddev = (mddev_t *) data;
4296 if (!atomic_read(&mddev->writes_pending)) {
4297 mddev->safemode = 1;
4298 if (mddev->external)
4299 sysfs_notify_dirent(mddev->sysfs_state);
4301 md_wakeup_thread(mddev->thread);
4304 static int start_dirty_degraded;
4306 static int md_run(mddev_t *mddev)
4310 struct mdk_personality *pers;
4312 if (list_empty(&mddev->disks))
4313 /* cannot run an array with no devices.. */
4319 /* These two calls synchronise us with the
4320 * sysfs_remove_group calls in mddev_unlock,
4321 * so they must have completed.
4323 mutex_lock(&mddev->open_mutex);
4324 mutex_unlock(&mddev->open_mutex);
4327 * Analyze all RAID superblock(s)
4329 if (!mddev->raid_disks) {
4330 if (!mddev->persistent)
4335 if (mddev->level != LEVEL_NONE)
4336 request_module("md-level-%d", mddev->level);
4337 else if (mddev->clevel[0])
4338 request_module("md-%s", mddev->clevel);
4341 * Drop all container device buffers, from now on
4342 * the only valid external interface is through the md
4345 list_for_each_entry(rdev, &mddev->disks, same_set) {
4346 if (test_bit(Faulty, &rdev->flags))
4348 sync_blockdev(rdev->bdev);
4349 invalidate_bdev(rdev->bdev);
4351 /* perform some consistency tests on the device.
4352 * We don't want the data to overlap the metadata,
4353 * Internal Bitmap issues have been handled elsewhere.
4355 if (rdev->data_offset < rdev->sb_start) {
4356 if (mddev->dev_sectors &&
4357 rdev->data_offset + mddev->dev_sectors
4359 printk("md: %s: data overlaps metadata\n",
4364 if (rdev->sb_start + rdev->sb_size/512
4365 > rdev->data_offset) {
4366 printk("md: %s: metadata overlaps data\n",
4371 sysfs_notify_dirent(rdev->sysfs_state);
4374 spin_lock(&pers_lock);
4375 pers = find_pers(mddev->level, mddev->clevel);
4376 if (!pers || !try_module_get(pers->owner)) {
4377 spin_unlock(&pers_lock);
4378 if (mddev->level != LEVEL_NONE)
4379 printk(KERN_WARNING "md: personality for level %d is not loaded!\n",
4382 printk(KERN_WARNING "md: personality for level %s is not loaded!\n",
4387 spin_unlock(&pers_lock);
4388 if (mddev->level != pers->level) {
4389 mddev->level = pers->level;
4390 mddev->new_level = pers->level;
4392 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
4394 if (mddev->reshape_position != MaxSector &&
4395 pers->start_reshape == NULL) {
4396 /* This personality cannot handle reshaping... */
4398 module_put(pers->owner);
4402 if (pers->sync_request) {
4403 /* Warn if this is a potentially silly
4406 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
4410 list_for_each_entry(rdev, &mddev->disks, same_set)
4411 list_for_each_entry(rdev2, &mddev->disks, same_set) {
4413 rdev->bdev->bd_contains ==
4414 rdev2->bdev->bd_contains) {
4416 "%s: WARNING: %s appears to be"
4417 " on the same physical disk as"
4420 bdevname(rdev->bdev,b),
4421 bdevname(rdev2->bdev,b2));
4428 "True protection against single-disk"
4429 " failure might be compromised.\n");
4432 mddev->recovery = 0;
4433 /* may be over-ridden by personality */
4434 mddev->resync_max_sectors = mddev->dev_sectors;
4436 mddev->barriers_work = 1;
4437 mddev->ok_start_degraded = start_dirty_degraded;
4439 if (start_readonly && mddev->ro == 0)
4440 mddev->ro = 2; /* read-only, but switch on first write */
4442 err = mddev->pers->run(mddev);
4444 printk(KERN_ERR "md: pers->run() failed ...\n");
4445 else if (mddev->pers->size(mddev, 0, 0) < mddev->array_sectors) {
4446 WARN_ONCE(!mddev->external_size, "%s: default size too small,"
4447 " but 'external_size' not in effect?\n", __func__);
4449 "md: invalid array_size %llu > default size %llu\n",
4450 (unsigned long long)mddev->array_sectors / 2,
4451 (unsigned long long)mddev->pers->size(mddev, 0, 0) / 2);
4453 mddev->pers->stop(mddev);
4455 if (err == 0 && mddev->pers->sync_request) {
4456 err = bitmap_create(mddev);
4458 printk(KERN_ERR "%s: failed to create bitmap (%d)\n",
4459 mdname(mddev), err);
4460 mddev->pers->stop(mddev);
4464 module_put(mddev->pers->owner);
4466 bitmap_destroy(mddev);
4469 if (mddev->pers->sync_request) {
4470 if (sysfs_create_group(&mddev->kobj, &md_redundancy_group))
4472 "md: cannot register extra attributes for %s\n",
4474 mddev->sysfs_action = sysfs_get_dirent(mddev->kobj.sd, "sync_action");
4475 } else if (mddev->ro == 2) /* auto-readonly not meaningful */
4478 atomic_set(&mddev->writes_pending,0);
4479 atomic_set(&mddev->max_corr_read_errors,
4480 MD_DEFAULT_MAX_CORRECTED_READ_ERRORS);
4481 mddev->safemode = 0;
4482 mddev->safemode_timer.function = md_safemode_timeout;
4483 mddev->safemode_timer.data = (unsigned long) mddev;
4484 mddev->safemode_delay = (200 * HZ)/1000 +1; /* 200 msec delay */
4487 list_for_each_entry(rdev, &mddev->disks, same_set)
4488 if (rdev->raid_disk >= 0) {
4490 sprintf(nm, "rd%d", rdev->raid_disk);
4491 if (sysfs_create_link(&mddev->kobj, &rdev->kobj, nm))
4492 printk("md: cannot register %s for %s\n",
4496 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4499 md_update_sb(mddev, 0);
4501 md_wakeup_thread(mddev->thread);
4502 md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
4504 md_new_event(mddev);
4505 sysfs_notify_dirent(mddev->sysfs_state);
4506 if (mddev->sysfs_action)
4507 sysfs_notify_dirent(mddev->sysfs_action);
4508 sysfs_notify(&mddev->kobj, NULL, "degraded");
4512 static int do_md_run(mddev_t *mddev)
4516 err = md_run(mddev);
4520 set_capacity(mddev->gendisk, mddev->array_sectors);
4521 revalidate_disk(mddev->gendisk);
4522 kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
4527 static int restart_array(mddev_t *mddev)
4529 struct gendisk *disk = mddev->gendisk;
4531 /* Complain if it has no devices */
4532 if (list_empty(&mddev->disks))
4538 mddev->safemode = 0;
4540 set_disk_ro(disk, 0);
4541 printk(KERN_INFO "md: %s switched to read-write mode.\n",
4543 /* Kick recovery or resync if necessary */
4544 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4545 md_wakeup_thread(mddev->thread);
4546 md_wakeup_thread(mddev->sync_thread);
4547 sysfs_notify_dirent(mddev->sysfs_state);
4551 /* similar to deny_write_access, but accounts for our holding a reference
4552 * to the file ourselves */
4553 static int deny_bitmap_write_access(struct file * file)
4555 struct inode *inode = file->f_mapping->host;
4557 spin_lock(&inode->i_lock);
4558 if (atomic_read(&inode->i_writecount) > 1) {
4559 spin_unlock(&inode->i_lock);
4562 atomic_set(&inode->i_writecount, -1);
4563 spin_unlock(&inode->i_lock);
4568 void restore_bitmap_write_access(struct file *file)
4570 struct inode *inode = file->f_mapping->host;
4572 spin_lock(&inode->i_lock);
4573 atomic_set(&inode->i_writecount, 1);
4574 spin_unlock(&inode->i_lock);
4577 static void md_clean(mddev_t *mddev)
4579 mddev->array_sectors = 0;
4580 mddev->external_size = 0;
4581 mddev->dev_sectors = 0;
4582 mddev->raid_disks = 0;
4583 mddev->recovery_cp = 0;
4584 mddev->resync_min = 0;
4585 mddev->resync_max = MaxSector;
4586 mddev->reshape_position = MaxSector;
4587 mddev->external = 0;
4588 mddev->persistent = 0;
4589 mddev->level = LEVEL_NONE;
4590 mddev->clevel[0] = 0;
4593 mddev->metadata_type[0] = 0;
4594 mddev->chunk_sectors = 0;
4595 mddev->ctime = mddev->utime = 0;
4597 mddev->max_disks = 0;
4599 mddev->delta_disks = 0;
4600 mddev->new_level = LEVEL_NONE;
4601 mddev->new_layout = 0;
4602 mddev->new_chunk_sectors = 0;
4603 mddev->curr_resync = 0;
4604 mddev->resync_mismatches = 0;
4605 mddev->suspend_lo = mddev->suspend_hi = 0;
4606 mddev->sync_speed_min = mddev->sync_speed_max = 0;
4607 mddev->recovery = 0;
4609 mddev->degraded = 0;
4610 mddev->barriers_work = 0;
4611 mddev->safemode = 0;
4612 mddev->bitmap_info.offset = 0;
4613 mddev->bitmap_info.default_offset = 0;
4614 mddev->bitmap_info.chunksize = 0;
4615 mddev->bitmap_info.daemon_sleep = 0;
4616 mddev->bitmap_info.max_write_behind = 0;
4619 static void md_stop_writes(mddev_t *mddev)
4621 if (mddev->sync_thread) {
4622 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4623 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
4624 md_unregister_thread(mddev->sync_thread);
4625 mddev->sync_thread = NULL;
4628 del_timer_sync(&mddev->safemode_timer);
4630 bitmap_flush(mddev);
4631 md_super_wait(mddev);
4633 if (!mddev->in_sync || mddev->flags) {
4634 /* mark array as shutdown cleanly */
4636 md_update_sb(mddev, 1);
4640 static void md_stop(mddev_t *mddev)
4642 md_stop_writes(mddev);
4644 mddev->pers->stop(mddev);
4645 if (mddev->pers->sync_request && mddev->to_remove == NULL)
4646 mddev->to_remove = &md_redundancy_group;
4647 module_put(mddev->pers->owner);
4649 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4652 static int md_set_readonly(mddev_t *mddev, int is_open)
4655 mutex_lock(&mddev->open_mutex);
4656 if (atomic_read(&mddev->openers) > is_open) {
4657 printk("md: %s still in use.\n",mdname(mddev));
4662 md_stop_writes(mddev);
4668 set_disk_ro(mddev->gendisk, 1);
4669 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4670 sysfs_notify_dirent(mddev->sysfs_state);
4674 mutex_unlock(&mddev->open_mutex);
4679 * 0 - completely stop and dis-assemble array
4680 * 2 - stop but do not disassemble array
4682 static int do_md_stop(mddev_t * mddev, int mode, int is_open)
4685 struct gendisk *disk = mddev->gendisk;
4688 mutex_lock(&mddev->open_mutex);
4689 if (atomic_read(&mddev->openers) > is_open) {
4690 printk("md: %s still in use.\n",mdname(mddev));
4692 } else if (mddev->pers) {
4695 set_disk_ro(disk, 0);
4698 mddev->queue->merge_bvec_fn = NULL;
4699 mddev->queue->unplug_fn = NULL;
4700 mddev->queue->backing_dev_info.congested_fn = NULL;
4702 /* tell userspace to handle 'inactive' */
4703 sysfs_notify_dirent(mddev->sysfs_state);
4705 list_for_each_entry(rdev, &mddev->disks, same_set)
4706 if (rdev->raid_disk >= 0) {
4708 sprintf(nm, "rd%d", rdev->raid_disk);
4709 sysfs_remove_link(&mddev->kobj, nm);
4712 set_capacity(disk, 0);
4713 revalidate_disk(disk);
4720 mutex_unlock(&mddev->open_mutex);
4724 * Free resources if final stop
4728 printk(KERN_INFO "md: %s stopped.\n", mdname(mddev));
4730 bitmap_destroy(mddev);
4731 if (mddev->bitmap_info.file) {
4732 restore_bitmap_write_access(mddev->bitmap_info.file);
4733 fput(mddev->bitmap_info.file);
4734 mddev->bitmap_info.file = NULL;
4736 mddev->bitmap_info.offset = 0;
4738 export_array(mddev);
4741 kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
4742 if (mddev->hold_active == UNTIL_STOP)
4743 mddev->hold_active = 0;
4747 blk_integrity_unregister(disk);
4748 md_new_event(mddev);
4749 sysfs_notify_dirent(mddev->sysfs_state);
4754 static void autorun_array(mddev_t *mddev)
4759 if (list_empty(&mddev->disks))
4762 printk(KERN_INFO "md: running: ");
4764 list_for_each_entry(rdev, &mddev->disks, same_set) {
4765 char b[BDEVNAME_SIZE];
4766 printk("<%s>", bdevname(rdev->bdev,b));
4770 err = do_md_run(mddev);
4772 printk(KERN_WARNING "md: do_md_run() returned %d\n", err);
4773 do_md_stop(mddev, 0, 0);
4778 * lets try to run arrays based on all disks that have arrived
4779 * until now. (those are in pending_raid_disks)
4781 * the method: pick the first pending disk, collect all disks with
4782 * the same UUID, remove all from the pending list and put them into
4783 * the 'same_array' list. Then order this list based on superblock
4784 * update time (freshest comes first), kick out 'old' disks and
4785 * compare superblocks. If everything's fine then run it.
4787 * If "unit" is allocated, then bump its reference count
4789 static void autorun_devices(int part)
4791 mdk_rdev_t *rdev0, *rdev, *tmp;
4793 char b[BDEVNAME_SIZE];
4795 printk(KERN_INFO "md: autorun ...\n");
4796 while (!list_empty(&pending_raid_disks)) {
4799 LIST_HEAD(candidates);
4800 rdev0 = list_entry(pending_raid_disks.next,
4801 mdk_rdev_t, same_set);
4803 printk(KERN_INFO "md: considering %s ...\n",
4804 bdevname(rdev0->bdev,b));
4805 INIT_LIST_HEAD(&candidates);
4806 rdev_for_each_list(rdev, tmp, &pending_raid_disks)
4807 if (super_90_load(rdev, rdev0, 0) >= 0) {
4808 printk(KERN_INFO "md: adding %s ...\n",
4809 bdevname(rdev->bdev,b));
4810 list_move(&rdev->same_set, &candidates);
4813 * now we have a set of devices, with all of them having
4814 * mostly sane superblocks. It's time to allocate the
4818 dev = MKDEV(mdp_major,
4819 rdev0->preferred_minor << MdpMinorShift);
4820 unit = MINOR(dev) >> MdpMinorShift;
4822 dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
4825 if (rdev0->preferred_minor != unit) {
4826 printk(KERN_INFO "md: unit number in %s is bad: %d\n",
4827 bdevname(rdev0->bdev, b), rdev0->preferred_minor);
4831 md_probe(dev, NULL, NULL);
4832 mddev = mddev_find(dev);
4833 if (!mddev || !mddev->gendisk) {
4837 "md: cannot allocate memory for md drive.\n");
4840 if (mddev_lock(mddev))
4841 printk(KERN_WARNING "md: %s locked, cannot run\n",
4843 else if (mddev->raid_disks || mddev->major_version
4844 || !list_empty(&mddev->disks)) {
4846 "md: %s already running, cannot run %s\n",
4847 mdname(mddev), bdevname(rdev0->bdev,b));
4848 mddev_unlock(mddev);
4850 printk(KERN_INFO "md: created %s\n", mdname(mddev));
4851 mddev->persistent = 1;
4852 rdev_for_each_list(rdev, tmp, &candidates) {
4853 list_del_init(&rdev->same_set);
4854 if (bind_rdev_to_array(rdev, mddev))
4857 autorun_array(mddev);
4858 mddev_unlock(mddev);
4860 /* on success, candidates will be empty, on error
4863 rdev_for_each_list(rdev, tmp, &candidates) {
4864 list_del_init(&rdev->same_set);
4869 printk(KERN_INFO "md: ... autorun DONE.\n");
4871 #endif /* !MODULE */
4873 static int get_version(void __user * arg)
4877 ver.major = MD_MAJOR_VERSION;
4878 ver.minor = MD_MINOR_VERSION;
4879 ver.patchlevel = MD_PATCHLEVEL_VERSION;
4881 if (copy_to_user(arg, &ver, sizeof(ver)))
4887 static int get_array_info(mddev_t * mddev, void __user * arg)
4889 mdu_array_info_t info;
4890 int nr,working,insync,failed,spare;
4893 nr=working=insync=failed=spare=0;
4894 list_for_each_entry(rdev, &mddev->disks, same_set) {
4896 if (test_bit(Faulty, &rdev->flags))
4900 if (test_bit(In_sync, &rdev->flags))
4907 info.major_version = mddev->major_version;
4908 info.minor_version = mddev->minor_version;
4909 info.patch_version = MD_PATCHLEVEL_VERSION;
4910 info.ctime = mddev->ctime;
4911 info.level = mddev->level;
4912 info.size = mddev->dev_sectors / 2;
4913 if (info.size != mddev->dev_sectors / 2) /* overflow */
4916 info.raid_disks = mddev->raid_disks;
4917 info.md_minor = mddev->md_minor;
4918 info.not_persistent= !mddev->persistent;
4920 info.utime = mddev->utime;
4923 info.state = (1<<MD_SB_CLEAN);
4924 if (mddev->bitmap && mddev->bitmap_info.offset)
4925 info.state = (1<<MD_SB_BITMAP_PRESENT);
4926 info.active_disks = insync;
4927 info.working_disks = working;
4928 info.failed_disks = failed;
4929 info.spare_disks = spare;
4931 info.layout = mddev->layout;
4932 info.chunk_size = mddev->chunk_sectors << 9;
4934 if (copy_to_user(arg, &info, sizeof(info)))
4940 static int get_bitmap_file(mddev_t * mddev, void __user * arg)
4942 mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */
4943 char *ptr, *buf = NULL;
4946 if (md_allow_write(mddev))
4947 file = kmalloc(sizeof(*file), GFP_NOIO);
4949 file = kmalloc(sizeof(*file), GFP_KERNEL);
4954 /* bitmap disabled, zero the first byte and copy out */
4955 if (!mddev->bitmap || !mddev->bitmap->file) {
4956 file->pathname[0] = '\0';
4960 buf = kmalloc(sizeof(file->pathname), GFP_KERNEL);
4964 ptr = d_path(&mddev->bitmap->file->f_path, buf, sizeof(file->pathname));
4968 strcpy(file->pathname, ptr);
4972 if (copy_to_user(arg, file, sizeof(*file)))
4980 static int get_disk_info(mddev_t * mddev, void __user * arg)
4982 mdu_disk_info_t info;
4985 if (copy_from_user(&info, arg, sizeof(info)))
4988 rdev = find_rdev_nr(mddev, info.number);
4990 info.major = MAJOR(rdev->bdev->bd_dev);
4991 info.minor = MINOR(rdev->bdev->bd_dev);
4992 info.raid_disk = rdev->raid_disk;
4994 if (test_bit(Faulty, &rdev->flags))
4995 info.state |= (1<<MD_DISK_FAULTY);
4996 else if (test_bit(In_sync, &rdev->flags)) {
4997 info.state |= (1<<MD_DISK_ACTIVE);
4998 info.state |= (1<<MD_DISK_SYNC);
5000 if (test_bit(WriteMostly, &rdev->flags))
5001 info.state |= (1<<MD_DISK_WRITEMOSTLY);
5003 info.major = info.minor = 0;
5004 info.raid_disk = -1;
5005 info.state = (1<<MD_DISK_REMOVED);
5008 if (copy_to_user(arg, &info, sizeof(info)))
5014 static int add_new_disk(mddev_t * mddev, mdu_disk_info_t *info)
5016 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
5018 dev_t dev = MKDEV(info->major,info->minor);
5020 if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
5023 if (!mddev->raid_disks) {
5025 /* expecting a device which has a superblock */
5026 rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
5029 "md: md_import_device returned %ld\n",
5031 return PTR_ERR(rdev);
5033 if (!list_empty(&mddev->disks)) {
5034 mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
5035 mdk_rdev_t, same_set);
5036 err = super_types[mddev->major_version]
5037 .load_super(rdev, rdev0, mddev->minor_version);
5040 "md: %s has different UUID to %s\n",
5041 bdevname(rdev->bdev,b),
5042 bdevname(rdev0->bdev,b2));
5047 err = bind_rdev_to_array(rdev, mddev);
5054 * add_new_disk can be used once the array is assembled
5055 * to add "hot spares". They must already have a superblock
5060 if (!mddev->pers->hot_add_disk) {
5062 "%s: personality does not support diskops!\n",
5066 if (mddev->persistent)
5067 rdev = md_import_device(dev, mddev->major_version,
5068 mddev->minor_version);
5070 rdev = md_import_device(dev, -1, -1);
5073 "md: md_import_device returned %ld\n",
5075 return PTR_ERR(rdev);
5077 /* set save_raid_disk if appropriate */
5078 if (!mddev->persistent) {
5079 if (info->state & (1<<MD_DISK_SYNC) &&
5080 info->raid_disk < mddev->raid_disks)
5081 rdev->raid_disk = info->raid_disk;
5083 rdev->raid_disk = -1;
5085 super_types[mddev->major_version].
5086 validate_super(mddev, rdev);
5087 rdev->saved_raid_disk = rdev->raid_disk;
5089 clear_bit(In_sync, &rdev->flags); /* just to be sure */
5090 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
5091 set_bit(WriteMostly, &rdev->flags);
5093 clear_bit(WriteMostly, &rdev->flags);
5095 rdev->raid_disk = -1;
5096 err = bind_rdev_to_array(rdev, mddev);
5097 if (!err && !mddev->pers->hot_remove_disk) {
5098 /* If there is hot_add_disk but no hot_remove_disk
5099 * then added disks for geometry changes,
5100 * and should be added immediately.
5102 super_types[mddev->major_version].
5103 validate_super(mddev, rdev);
5104 err = mddev->pers->hot_add_disk(mddev, rdev);
5106 unbind_rdev_from_array(rdev);
5111 sysfs_notify_dirent(rdev->sysfs_state);
5113 md_update_sb(mddev, 1);
5114 if (mddev->degraded)
5115 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
5116 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5117 md_wakeup_thread(mddev->thread);
5121 /* otherwise, add_new_disk is only allowed
5122 * for major_version==0 superblocks
5124 if (mddev->major_version != 0) {
5125 printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n",
5130 if (!(info->state & (1<<MD_DISK_FAULTY))) {
5132 rdev = md_import_device(dev, -1, 0);
5135 "md: error, md_import_device() returned %ld\n",
5137 return PTR_ERR(rdev);
5139 rdev->desc_nr = info->number;
5140 if (info->raid_disk < mddev->raid_disks)
5141 rdev->raid_disk = info->raid_disk;
5143 rdev->raid_disk = -1;
5145 if (rdev->raid_disk < mddev->raid_disks)
5146 if (info->state & (1<<MD_DISK_SYNC))
5147 set_bit(In_sync, &rdev->flags);
5149 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
5150 set_bit(WriteMostly, &rdev->flags);
5152 if (!mddev->persistent) {
5153 printk(KERN_INFO "md: nonpersistent superblock ...\n");
5154 rdev->sb_start = rdev->bdev->bd_inode->i_size / 512;
5156 rdev->sb_start = calc_dev_sboffset(rdev->bdev);
5157 rdev->sectors = rdev->sb_start;
5159 err = bind_rdev_to_array(rdev, mddev);
5169 static int hot_remove_disk(mddev_t * mddev, dev_t dev)
5171 char b[BDEVNAME_SIZE];
5174 rdev = find_rdev(mddev, dev);
5178 if (rdev->raid_disk >= 0)
5181 kick_rdev_from_array(rdev);
5182 md_update_sb(mddev, 1);
5183 md_new_event(mddev);
5187 printk(KERN_WARNING "md: cannot remove active disk %s from %s ...\n",
5188 bdevname(rdev->bdev,b), mdname(mddev));
5192 static int hot_add_disk(mddev_t * mddev, dev_t dev)
5194 char b[BDEVNAME_SIZE];
5201 if (mddev->major_version != 0) {
5202 printk(KERN_WARNING "%s: HOT_ADD may only be used with"
5203 " version-0 superblocks.\n",
5207 if (!mddev->pers->hot_add_disk) {
5209 "%s: personality does not support diskops!\n",
5214 rdev = md_import_device(dev, -1, 0);
5217 "md: error, md_import_device() returned %ld\n",
5222 if (mddev->persistent)
5223 rdev->sb_start = calc_dev_sboffset(rdev->bdev);
5225 rdev->sb_start = rdev->bdev->bd_inode->i_size / 512;
5227 rdev->sectors = rdev->sb_start;
5229 if (test_bit(Faulty, &rdev->flags)) {
5231 "md: can not hot-add faulty %s disk to %s!\n",
5232 bdevname(rdev->bdev,b), mdname(mddev));
5236 clear_bit(In_sync, &rdev->flags);
5238 rdev->saved_raid_disk = -1;
5239 err = bind_rdev_to_array(rdev, mddev);
5244 * The rest should better be atomic, we can have disk failures
5245 * noticed in interrupt contexts ...
5248 rdev->raid_disk = -1;
5250 md_update_sb(mddev, 1);
5253 * Kick recovery, maybe this spare has to be added to the
5254 * array immediately.
5256 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5257 md_wakeup_thread(mddev->thread);
5258 md_new_event(mddev);
5266 static int set_bitmap_file(mddev_t *mddev, int fd)
5271 if (!mddev->pers->quiesce)
5273 if (mddev->recovery || mddev->sync_thread)
5275 /* we should be able to change the bitmap.. */
5281 return -EEXIST; /* cannot add when bitmap is present */
5282 mddev->bitmap_info.file = fget(fd);
5284 if (mddev->bitmap_info.file == NULL) {
5285 printk(KERN_ERR "%s: error: failed to get bitmap file\n",
5290 err = deny_bitmap_write_access(mddev->bitmap_info.file);
5292 printk(KERN_ERR "%s: error: bitmap file is already in use\n",
5294 fput(mddev->bitmap_info.file);
5295 mddev->bitmap_info.file = NULL;
5298 mddev->bitmap_info.offset = 0; /* file overrides offset */
5299 } else if (mddev->bitmap == NULL)
5300 return -ENOENT; /* cannot remove what isn't there */
5303 mddev->pers->quiesce(mddev, 1);
5305 err = bitmap_create(mddev);
5306 if (fd < 0 || err) {
5307 bitmap_destroy(mddev);
5308 fd = -1; /* make sure to put the file */
5310 mddev->pers->quiesce(mddev, 0);
5313 if (mddev->bitmap_info.file) {
5314 restore_bitmap_write_access(mddev->bitmap_info.file);
5315 fput(mddev->bitmap_info.file);
5317 mddev->bitmap_info.file = NULL;
5324 * set_array_info is used two different ways
5325 * The original usage is when creating a new array.
5326 * In this usage, raid_disks is > 0 and it together with
5327 * level, size, not_persistent,layout,chunksize determine the
5328 * shape of the array.
5329 * This will always create an array with a type-0.90.0 superblock.
5330 * The newer usage is when assembling an array.
5331 * In this case raid_disks will be 0, and the major_version field is
5332 * use to determine which style super-blocks are to be found on the devices.
5333 * The minor and patch _version numbers are also kept incase the
5334 * super_block handler wishes to interpret them.
5336 static int set_array_info(mddev_t * mddev, mdu_array_info_t *info)
5339 if (info->raid_disks == 0) {
5340 /* just setting version number for superblock loading */
5341 if (info->major_version < 0 ||
5342 info->major_version >= ARRAY_SIZE(super_types) ||
5343 super_types[info->major_version].name == NULL) {
5344 /* maybe try to auto-load a module? */
5346 "md: superblock version %d not known\n",
5347 info->major_version);
5350 mddev->major_version = info->major_version;
5351 mddev->minor_version = info->minor_version;
5352 mddev->patch_version = info->patch_version;
5353 mddev->persistent = !info->not_persistent;
5354 /* ensure mddev_put doesn't delete this now that there
5355 * is some minimal configuration.
5357 mddev->ctime = get_seconds();
5360 mddev->major_version = MD_MAJOR_VERSION;
5361 mddev->minor_version = MD_MINOR_VERSION;
5362 mddev->patch_version = MD_PATCHLEVEL_VERSION;
5363 mddev->ctime = get_seconds();
5365 mddev->level = info->level;
5366 mddev->clevel[0] = 0;
5367 mddev->dev_sectors = 2 * (sector_t)info->size;
5368 mddev->raid_disks = info->raid_disks;
5369 /* don't set md_minor, it is determined by which /dev/md* was
5372 if (info->state & (1<<MD_SB_CLEAN))
5373 mddev->recovery_cp = MaxSector;
5375 mddev->recovery_cp = 0;
5376 mddev->persistent = ! info->not_persistent;
5377 mddev->external = 0;
5379 mddev->layout = info->layout;
5380 mddev->chunk_sectors = info->chunk_size >> 9;
5382 mddev->max_disks = MD_SB_DISKS;
5384 if (mddev->persistent)
5386 set_bit(MD_CHANGE_DEVS, &mddev->flags);
5388 mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9;
5389 mddev->bitmap_info.offset = 0;
5391 mddev->reshape_position = MaxSector;
5394 * Generate a 128 bit UUID
5396 get_random_bytes(mddev->uuid, 16);
5398 mddev->new_level = mddev->level;
5399 mddev->new_chunk_sectors = mddev->chunk_sectors;
5400 mddev->new_layout = mddev->layout;
5401 mddev->delta_disks = 0;
5406 void md_set_array_sectors(mddev_t *mddev, sector_t array_sectors)
5408 WARN(!mddev_is_locked(mddev), "%s: unlocked mddev!\n", __func__);
5410 if (mddev->external_size)
5413 mddev->array_sectors = array_sectors;
5415 EXPORT_SYMBOL(md_set_array_sectors);
5417 static int update_size(mddev_t *mddev, sector_t num_sectors)
5421 int fit = (num_sectors == 0);
5423 if (mddev->pers->resize == NULL)
5425 /* The "num_sectors" is the number of sectors of each device that
5426 * is used. This can only make sense for arrays with redundancy.
5427 * linear and raid0 always use whatever space is available. We can only
5428 * consider changing this number if no resync or reconstruction is
5429 * happening, and if the new size is acceptable. It must fit before the
5430 * sb_start or, if that is <data_offset, it must fit before the size
5431 * of each device. If num_sectors is zero, we find the largest size
5435 if (mddev->sync_thread)
5438 /* Sorry, cannot grow a bitmap yet, just remove it,
5442 list_for_each_entry(rdev, &mddev->disks, same_set) {
5443 sector_t avail = rdev->sectors;
5445 if (fit && (num_sectors == 0 || num_sectors > avail))
5446 num_sectors = avail;
5447 if (avail < num_sectors)
5450 rv = mddev->pers->resize(mddev, num_sectors);
5452 revalidate_disk(mddev->gendisk);
5456 static int update_raid_disks(mddev_t *mddev, int raid_disks)
5459 /* change the number of raid disks */
5460 if (mddev->pers->check_reshape == NULL)
5462 if (raid_disks <= 0 ||
5463 (mddev->max_disks && raid_disks >= mddev->max_disks))
5465 if (mddev->sync_thread || mddev->reshape_position != MaxSector)
5467 mddev->delta_disks = raid_disks - mddev->raid_disks;
5469 rv = mddev->pers->check_reshape(mddev);
5475 * update_array_info is used to change the configuration of an
5477 * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
5478 * fields in the info are checked against the array.
5479 * Any differences that cannot be handled will cause an error.
5480 * Normally, only one change can be managed at a time.
5482 static int update_array_info(mddev_t *mddev, mdu_array_info_t *info)
5488 /* calculate expected state,ignoring low bits */
5489 if (mddev->bitmap && mddev->bitmap_info.offset)
5490 state |= (1 << MD_SB_BITMAP_PRESENT);
5492 if (mddev->major_version != info->major_version ||
5493 mddev->minor_version != info->minor_version ||
5494 /* mddev->patch_version != info->patch_version || */
5495 mddev->ctime != info->ctime ||
5496 mddev->level != info->level ||
5497 /* mddev->layout != info->layout || */
5498 !mddev->persistent != info->not_persistent||
5499 mddev->chunk_sectors != info->chunk_size >> 9 ||
5500 /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
5501 ((state^info->state) & 0xfffffe00)
5504 /* Check there is only one change */
5505 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
5507 if (mddev->raid_disks != info->raid_disks)
5509 if (mddev->layout != info->layout)
5511 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT))
5518 if (mddev->layout != info->layout) {
5520 * we don't need to do anything at the md level, the
5521 * personality will take care of it all.
5523 if (mddev->pers->check_reshape == NULL)
5526 mddev->new_layout = info->layout;
5527 rv = mddev->pers->check_reshape(mddev);
5529 mddev->new_layout = mddev->layout;
5533 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
5534 rv = update_size(mddev, (sector_t)info->size * 2);
5536 if (mddev->raid_disks != info->raid_disks)
5537 rv = update_raid_disks(mddev, info->raid_disks);
5539 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) {
5540 if (mddev->pers->quiesce == NULL)
5542 if (mddev->recovery || mddev->sync_thread)
5544 if (info->state & (1<<MD_SB_BITMAP_PRESENT)) {
5545 /* add the bitmap */
5548 if (mddev->bitmap_info.default_offset == 0)
5550 mddev->bitmap_info.offset =
5551 mddev->bitmap_info.default_offset;
5552 mddev->pers->quiesce(mddev, 1);
5553 rv = bitmap_create(mddev);
5555 bitmap_destroy(mddev);
5556 mddev->pers->quiesce(mddev, 0);
5558 /* remove the bitmap */
5561 if (mddev->bitmap->file)
5563 mddev->pers->quiesce(mddev, 1);
5564 bitmap_destroy(mddev);
5565 mddev->pers->quiesce(mddev, 0);
5566 mddev->bitmap_info.offset = 0;
5569 md_update_sb(mddev, 1);
5573 static int set_disk_faulty(mddev_t *mddev, dev_t dev)
5577 if (mddev->pers == NULL)
5580 rdev = find_rdev(mddev, dev);
5584 md_error(mddev, rdev);
5589 * We have a problem here : there is no easy way to give a CHS
5590 * virtual geometry. We currently pretend that we have a 2 heads
5591 * 4 sectors (with a BIG number of cylinders...). This drives
5592 * dosfs just mad... ;-)
5594 static int md_getgeo(struct block_device *bdev, struct hd_geometry *geo)
5596 mddev_t *mddev = bdev->bd_disk->private_data;
5600 geo->cylinders = mddev->array_sectors / 8;
5604 static int md_ioctl(struct block_device *bdev, fmode_t mode,
5605 unsigned int cmd, unsigned long arg)
5608 void __user *argp = (void __user *)arg;
5609 mddev_t *mddev = NULL;
5612 if (!capable(CAP_SYS_ADMIN))
5616 * Commands dealing with the RAID driver but not any
5622 err = get_version(argp);
5625 case PRINT_RAID_DEBUG:
5633 autostart_arrays(arg);
5640 * Commands creating/starting a new array:
5643 mddev = bdev->bd_disk->private_data;
5650 err = mddev_lock(mddev);
5653 "md: ioctl lock interrupted, reason %d, cmd %d\n",
5660 case SET_ARRAY_INFO:
5662 mdu_array_info_t info;
5664 memset(&info, 0, sizeof(info));
5665 else if (copy_from_user(&info, argp, sizeof(info))) {
5670 err = update_array_info(mddev, &info);
5672 printk(KERN_WARNING "md: couldn't update"
5673 " array info. %d\n", err);
5678 if (!list_empty(&mddev->disks)) {
5680 "md: array %s already has disks!\n",
5685 if (mddev->raid_disks) {
5687 "md: array %s already initialised!\n",
5692 err = set_array_info(mddev, &info);
5694 printk(KERN_WARNING "md: couldn't set"
5695 " array info. %d\n", err);
5705 * Commands querying/configuring an existing array:
5707 /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
5708 * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */
5709 if ((!mddev->raid_disks && !mddev->external)
5710 && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY
5711 && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE
5712 && cmd != GET_BITMAP_FILE) {
5718 * Commands even a read-only array can execute:
5722 case GET_ARRAY_INFO:
5723 err = get_array_info(mddev, argp);
5726 case GET_BITMAP_FILE:
5727 err = get_bitmap_file(mddev, argp);
5731 err = get_disk_info(mddev, argp);
5734 case RESTART_ARRAY_RW:
5735 err = restart_array(mddev);
5739 err = do_md_stop(mddev, 0, 1);
5743 err = md_set_readonly(mddev, 1);
5747 if (get_user(ro, (int __user *)(arg))) {
5753 /* if the bdev is going readonly the value of mddev->ro
5754 * does not matter, no writes are coming
5759 /* are we are already prepared for writes? */
5763 /* transitioning to readauto need only happen for
5764 * arrays that call md_write_start
5767 err = restart_array(mddev);
5770 set_disk_ro(mddev->gendisk, 0);
5777 * The remaining ioctls are changing the state of the
5778 * superblock, so we do not allow them on read-only arrays.
5779 * However non-MD ioctls (e.g. get-size) will still come through
5780 * here and hit the 'default' below, so only disallow
5781 * 'md' ioctls, and switch to rw mode if started auto-readonly.
5783 if (_IOC_TYPE(cmd) == MD_MAJOR && mddev->ro && mddev->pers) {
5784 if (mddev->ro == 2) {
5786 sysfs_notify_dirent(mddev->sysfs_state);
5787 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5788 md_wakeup_thread(mddev->thread);
5799 mdu_disk_info_t info;
5800 if (copy_from_user(&info, argp, sizeof(info)))
5803 err = add_new_disk(mddev, &info);
5807 case HOT_REMOVE_DISK:
5808 err = hot_remove_disk(mddev, new_decode_dev(arg));
5812 err = hot_add_disk(mddev, new_decode_dev(arg));
5815 case SET_DISK_FAULTY:
5816 err = set_disk_faulty(mddev, new_decode_dev(arg));
5820 err = do_md_run(mddev);
5823 case SET_BITMAP_FILE:
5824 err = set_bitmap_file(mddev, (int)arg);
5834 if (mddev->hold_active == UNTIL_IOCTL &&
5836 mddev->hold_active = 0;
5837 mddev_unlock(mddev);
5846 #ifdef CONFIG_COMPAT
5847 static int md_compat_ioctl(struct block_device *bdev, fmode_t mode,
5848 unsigned int cmd, unsigned long arg)
5851 case HOT_REMOVE_DISK:
5853 case SET_DISK_FAULTY:
5854 case SET_BITMAP_FILE:
5855 /* These take in integer arg, do not convert */
5858 arg = (unsigned long)compat_ptr(arg);
5862 return md_ioctl(bdev, mode, cmd, arg);
5864 #endif /* CONFIG_COMPAT */
5866 static int md_open(struct block_device *bdev, fmode_t mode)
5869 * Succeed if we can lock the mddev, which confirms that
5870 * it isn't being stopped right now.
5872 mddev_t *mddev = mddev_find(bdev->bd_dev);
5875 if (mddev->gendisk != bdev->bd_disk) {
5876 /* we are racing with mddev_put which is discarding this
5880 /* Wait until bdev->bd_disk is definitely gone */
5881 flush_scheduled_work();
5882 /* Then retry the open from the top */
5883 return -ERESTARTSYS;
5885 BUG_ON(mddev != bdev->bd_disk->private_data);
5887 if ((err = mutex_lock_interruptible(&mddev->open_mutex)))
5891 atomic_inc(&mddev->openers);
5892 mutex_unlock(&mddev->open_mutex);
5898 static int md_release(struct gendisk *disk, fmode_t mode)
5900 mddev_t *mddev = disk->private_data;
5903 atomic_dec(&mddev->openers);
5908 static const struct block_device_operations md_fops =
5910 .owner = THIS_MODULE,
5912 .release = md_release,
5914 #ifdef CONFIG_COMPAT
5915 .compat_ioctl = md_compat_ioctl,
5917 .getgeo = md_getgeo,
5920 static int md_thread(void * arg)
5922 mdk_thread_t *thread = arg;
5925 * md_thread is a 'system-thread', it's priority should be very
5926 * high. We avoid resource deadlocks individually in each
5927 * raid personality. (RAID5 does preallocation) We also use RR and
5928 * the very same RT priority as kswapd, thus we will never get
5929 * into a priority inversion deadlock.
5931 * we definitely have to have equal or higher priority than
5932 * bdflush, otherwise bdflush will deadlock if there are too
5933 * many dirty RAID5 blocks.
5936 allow_signal(SIGKILL);
5937 while (!kthread_should_stop()) {
5939 /* We need to wait INTERRUPTIBLE so that
5940 * we don't add to the load-average.
5941 * That means we need to be sure no signals are
5944 if (signal_pending(current))
5945 flush_signals(current);
5947 wait_event_interruptible_timeout
5949 test_bit(THREAD_WAKEUP, &thread->flags)
5950 || kthread_should_stop(),
5953 clear_bit(THREAD_WAKEUP, &thread->flags);
5955 thread->run(thread->mddev);
5961 void md_wakeup_thread(mdk_thread_t *thread)
5964 dprintk("md: waking up MD thread %s.\n", thread->tsk->comm);
5965 set_bit(THREAD_WAKEUP, &thread->flags);
5966 wake_up(&thread->wqueue);
5970 mdk_thread_t *md_register_thread(void (*run) (mddev_t *), mddev_t *mddev,
5973 mdk_thread_t *thread;
5975 thread = kzalloc(sizeof(mdk_thread_t), GFP_KERNEL);
5979 init_waitqueue_head(&thread->wqueue);
5982 thread->mddev = mddev;
5983 thread->timeout = MAX_SCHEDULE_TIMEOUT;
5984 thread->tsk = kthread_run(md_thread, thread,
5986 mdname(thread->mddev),
5987 name ?: mddev->pers->name);
5988 if (IS_ERR(thread->tsk)) {
5995 void md_unregister_thread(mdk_thread_t *thread)
5999 dprintk("interrupting MD-thread pid %d\n", task_pid_nr(thread->tsk));
6001 kthread_stop(thread->tsk);
6005 void md_error(mddev_t *mddev, mdk_rdev_t *rdev)
6012 if (!rdev || test_bit(Faulty, &rdev->flags))
6015 if (mddev->external)
6016 set_bit(Blocked, &rdev->flags);
6018 dprintk("md_error dev:%s, rdev:(%d:%d), (caller: %p,%p,%p,%p).\n",
6020 MAJOR(rdev->bdev->bd_dev), MINOR(rdev->bdev->bd_dev),
6021 __builtin_return_address(0),__builtin_return_address(1),
6022 __builtin_return_address(2),__builtin_return_address(3));
6026 if (!mddev->pers->error_handler)
6028 mddev->pers->error_handler(mddev,rdev);
6029 if (mddev->degraded)
6030 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
6031 sysfs_notify_dirent(rdev->sysfs_state);
6032 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6033 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6034 md_wakeup_thread(mddev->thread);
6035 md_new_event_inintr(mddev);
6038 /* seq_file implementation /proc/mdstat */
6040 static void status_unused(struct seq_file *seq)
6045 seq_printf(seq, "unused devices: ");
6047 list_for_each_entry(rdev, &pending_raid_disks, same_set) {
6048 char b[BDEVNAME_SIZE];
6050 seq_printf(seq, "%s ",
6051 bdevname(rdev->bdev,b));
6054 seq_printf(seq, "<none>");
6056 seq_printf(seq, "\n");
6060 static void status_resync(struct seq_file *seq, mddev_t * mddev)
6062 sector_t max_sectors, resync, res;
6063 unsigned long dt, db;
6066 unsigned int per_milli;
6068 resync = mddev->curr_resync - atomic_read(&mddev->recovery_active);
6070 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
6071 max_sectors = mddev->resync_max_sectors;
6073 max_sectors = mddev->dev_sectors;
6076 * Should not happen.
6082 /* Pick 'scale' such that (resync>>scale)*1000 will fit
6083 * in a sector_t, and (max_sectors>>scale) will fit in a
6084 * u32, as those are the requirements for sector_div.
6085 * Thus 'scale' must be at least 10
6088 if (sizeof(sector_t) > sizeof(unsigned long)) {
6089 while ( max_sectors/2 > (1ULL<<(scale+32)))
6092 res = (resync>>scale)*1000;
6093 sector_div(res, (u32)((max_sectors>>scale)+1));
6097 int i, x = per_milli/50, y = 20-x;
6098 seq_printf(seq, "[");
6099 for (i = 0; i < x; i++)
6100 seq_printf(seq, "=");
6101 seq_printf(seq, ">");
6102 for (i = 0; i < y; i++)
6103 seq_printf(seq, ".");
6104 seq_printf(seq, "] ");
6106 seq_printf(seq, " %s =%3u.%u%% (%llu/%llu)",
6107 (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)?
6109 (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)?
6111 (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
6112 "resync" : "recovery"))),
6113 per_milli/10, per_milli % 10,
6114 (unsigned long long) resync/2,
6115 (unsigned long long) max_sectors/2);
6118 * dt: time from mark until now
6119 * db: blocks written from mark until now
6120 * rt: remaining time
6122 * rt is a sector_t, so could be 32bit or 64bit.
6123 * So we divide before multiply in case it is 32bit and close
6125 * We scale the divisor (db) by 32 to avoid loosing precision
6126 * near the end of resync when the number of remaining sectors
6128 * We then divide rt by 32 after multiplying by db to compensate.
6129 * The '+1' avoids division by zero if db is very small.
6131 dt = ((jiffies - mddev->resync_mark) / HZ);
6133 db = (mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active))
6134 - mddev->resync_mark_cnt;
6136 rt = max_sectors - resync; /* number of remaining sectors */
6137 sector_div(rt, db/32+1);
6141 seq_printf(seq, " finish=%lu.%lumin", (unsigned long)rt / 60,
6142 ((unsigned long)rt % 60)/6);
6144 seq_printf(seq, " speed=%ldK/sec", db/2/dt);
6147 static void *md_seq_start(struct seq_file *seq, loff_t *pos)
6149 struct list_head *tmp;
6159 spin_lock(&all_mddevs_lock);
6160 list_for_each(tmp,&all_mddevs)
6162 mddev = list_entry(tmp, mddev_t, all_mddevs);
6164 spin_unlock(&all_mddevs_lock);
6167 spin_unlock(&all_mddevs_lock);
6169 return (void*)2;/* tail */
6173 static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
6175 struct list_head *tmp;
6176 mddev_t *next_mddev, *mddev = v;
6182 spin_lock(&all_mddevs_lock);
6184 tmp = all_mddevs.next;
6186 tmp = mddev->all_mddevs.next;
6187 if (tmp != &all_mddevs)
6188 next_mddev = mddev_get(list_entry(tmp,mddev_t,all_mddevs));
6190 next_mddev = (void*)2;
6193 spin_unlock(&all_mddevs_lock);
6201 static void md_seq_stop(struct seq_file *seq, void *v)
6205 if (mddev && v != (void*)1 && v != (void*)2)
6209 struct mdstat_info {
6213 static int md_seq_show(struct seq_file *seq, void *v)
6218 struct mdstat_info *mi = seq->private;
6219 struct bitmap *bitmap;
6221 if (v == (void*)1) {
6222 struct mdk_personality *pers;
6223 seq_printf(seq, "Personalities : ");
6224 spin_lock(&pers_lock);
6225 list_for_each_entry(pers, &pers_list, list)
6226 seq_printf(seq, "[%s] ", pers->name);
6228 spin_unlock(&pers_lock);
6229 seq_printf(seq, "\n");
6230 mi->event = atomic_read(&md_event_count);
6233 if (v == (void*)2) {
6238 if (mddev_lock(mddev) < 0)
6241 if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
6242 seq_printf(seq, "%s : %sactive", mdname(mddev),
6243 mddev->pers ? "" : "in");
6246 seq_printf(seq, " (read-only)");
6248 seq_printf(seq, " (auto-read-only)");
6249 seq_printf(seq, " %s", mddev->pers->name);
6253 list_for_each_entry(rdev, &mddev->disks, same_set) {
6254 char b[BDEVNAME_SIZE];
6255 seq_printf(seq, " %s[%d]",
6256 bdevname(rdev->bdev,b), rdev->desc_nr);
6257 if (test_bit(WriteMostly, &rdev->flags))
6258 seq_printf(seq, "(W)");
6259 if (test_bit(Faulty, &rdev->flags)) {
6260 seq_printf(seq, "(F)");
6262 } else if (rdev->raid_disk < 0)
6263 seq_printf(seq, "(S)"); /* spare */
6264 sectors += rdev->sectors;
6267 if (!list_empty(&mddev->disks)) {
6269 seq_printf(seq, "\n %llu blocks",
6270 (unsigned long long)
6271 mddev->array_sectors / 2);
6273 seq_printf(seq, "\n %llu blocks",
6274 (unsigned long long)sectors / 2);
6276 if (mddev->persistent) {
6277 if (mddev->major_version != 0 ||
6278 mddev->minor_version != 90) {
6279 seq_printf(seq," super %d.%d",
6280 mddev->major_version,
6281 mddev->minor_version);
6283 } else if (mddev->external)
6284 seq_printf(seq, " super external:%s",
6285 mddev->metadata_type);
6287 seq_printf(seq, " super non-persistent");
6290 mddev->pers->status(seq, mddev);
6291 seq_printf(seq, "\n ");
6292 if (mddev->pers->sync_request) {
6293 if (mddev->curr_resync > 2) {
6294 status_resync(seq, mddev);
6295 seq_printf(seq, "\n ");
6296 } else if (mddev->curr_resync == 1 || mddev->curr_resync == 2)
6297 seq_printf(seq, "\tresync=DELAYED\n ");
6298 else if (mddev->recovery_cp < MaxSector)
6299 seq_printf(seq, "\tresync=PENDING\n ");
6302 seq_printf(seq, "\n ");
6304 if ((bitmap = mddev->bitmap)) {
6305 unsigned long chunk_kb;
6306 unsigned long flags;
6307 spin_lock_irqsave(&bitmap->lock, flags);
6308 chunk_kb = mddev->bitmap_info.chunksize >> 10;
6309 seq_printf(seq, "bitmap: %lu/%lu pages [%luKB], "
6311 bitmap->pages - bitmap->missing_pages,
6313 (bitmap->pages - bitmap->missing_pages)
6314 << (PAGE_SHIFT - 10),
6315 chunk_kb ? chunk_kb : mddev->bitmap_info.chunksize,
6316 chunk_kb ? "KB" : "B");
6318 seq_printf(seq, ", file: ");
6319 seq_path(seq, &bitmap->file->f_path, " \t\n");
6322 seq_printf(seq, "\n");
6323 spin_unlock_irqrestore(&bitmap->lock, flags);
6326 seq_printf(seq, "\n");
6328 mddev_unlock(mddev);
6333 static const struct seq_operations md_seq_ops = {
6334 .start = md_seq_start,
6335 .next = md_seq_next,
6336 .stop = md_seq_stop,
6337 .show = md_seq_show,
6340 static int md_seq_open(struct inode *inode, struct file *file)
6343 struct mdstat_info *mi = kmalloc(sizeof(*mi), GFP_KERNEL);
6347 error = seq_open(file, &md_seq_ops);
6351 struct seq_file *p = file->private_data;
6353 mi->event = atomic_read(&md_event_count);
6358 static unsigned int mdstat_poll(struct file *filp, poll_table *wait)
6360 struct seq_file *m = filp->private_data;
6361 struct mdstat_info *mi = m->private;
6364 poll_wait(filp, &md_event_waiters, wait);
6366 /* always allow read */
6367 mask = POLLIN | POLLRDNORM;
6369 if (mi->event != atomic_read(&md_event_count))
6370 mask |= POLLERR | POLLPRI;
6374 static const struct file_operations md_seq_fops = {
6375 .owner = THIS_MODULE,
6376 .open = md_seq_open,
6378 .llseek = seq_lseek,
6379 .release = seq_release_private,
6380 .poll = mdstat_poll,
6383 int register_md_personality(struct mdk_personality *p)
6385 spin_lock(&pers_lock);
6386 list_add_tail(&p->list, &pers_list);
6387 printk(KERN_INFO "md: %s personality registered for level %d\n", p->name, p->level);
6388 spin_unlock(&pers_lock);
6392 int unregister_md_personality(struct mdk_personality *p)
6394 printk(KERN_INFO "md: %s personality unregistered\n", p->name);
6395 spin_lock(&pers_lock);
6396 list_del_init(&p->list);
6397 spin_unlock(&pers_lock);
6401 static int is_mddev_idle(mddev_t *mddev, int init)
6409 rdev_for_each_rcu(rdev, mddev) {
6410 struct gendisk *disk = rdev->bdev->bd_contains->bd_disk;
6411 curr_events = (int)part_stat_read(&disk->part0, sectors[0]) +
6412 (int)part_stat_read(&disk->part0, sectors[1]) -
6413 atomic_read(&disk->sync_io);
6414 /* sync IO will cause sync_io to increase before the disk_stats
6415 * as sync_io is counted when a request starts, and
6416 * disk_stats is counted when it completes.
6417 * So resync activity will cause curr_events to be smaller than
6418 * when there was no such activity.
6419 * non-sync IO will cause disk_stat to increase without
6420 * increasing sync_io so curr_events will (eventually)
6421 * be larger than it was before. Once it becomes
6422 * substantially larger, the test below will cause
6423 * the array to appear non-idle, and resync will slow
6425 * If there is a lot of outstanding resync activity when
6426 * we set last_event to curr_events, then all that activity
6427 * completing might cause the array to appear non-idle
6428 * and resync will be slowed down even though there might
6429 * not have been non-resync activity. This will only
6430 * happen once though. 'last_events' will soon reflect
6431 * the state where there is little or no outstanding
6432 * resync requests, and further resync activity will
6433 * always make curr_events less than last_events.
6436 if (init || curr_events - rdev->last_events > 64) {
6437 rdev->last_events = curr_events;
6445 void md_done_sync(mddev_t *mddev, int blocks, int ok)
6447 /* another "blocks" (512byte) blocks have been synced */
6448 atomic_sub(blocks, &mddev->recovery_active);
6449 wake_up(&mddev->recovery_wait);
6451 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6452 md_wakeup_thread(mddev->thread);
6453 // stop recovery, signal do_sync ....
6458 /* md_write_start(mddev, bi)
6459 * If we need to update some array metadata (e.g. 'active' flag
6460 * in superblock) before writing, schedule a superblock update
6461 * and wait for it to complete.
6463 void md_write_start(mddev_t *mddev, struct bio *bi)
6466 if (bio_data_dir(bi) != WRITE)
6469 BUG_ON(mddev->ro == 1);
6470 if (mddev->ro == 2) {
6471 /* need to switch to read/write */
6473 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6474 md_wakeup_thread(mddev->thread);
6475 md_wakeup_thread(mddev->sync_thread);
6478 atomic_inc(&mddev->writes_pending);
6479 if (mddev->safemode == 1)
6480 mddev->safemode = 0;
6481 if (mddev->in_sync) {
6482 spin_lock_irq(&mddev->write_lock);
6483 if (mddev->in_sync) {
6485 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
6486 md_wakeup_thread(mddev->thread);
6489 spin_unlock_irq(&mddev->write_lock);
6492 sysfs_notify_dirent(mddev->sysfs_state);
6493 wait_event(mddev->sb_wait,
6494 !test_bit(MD_CHANGE_CLEAN, &mddev->flags) &&
6495 !test_bit(MD_CHANGE_PENDING, &mddev->flags));
6498 void md_write_end(mddev_t *mddev)
6500 if (atomic_dec_and_test(&mddev->writes_pending)) {
6501 if (mddev->safemode == 2)
6502 md_wakeup_thread(mddev->thread);
6503 else if (mddev->safemode_delay)
6504 mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay);
6508 /* md_allow_write(mddev)
6509 * Calling this ensures that the array is marked 'active' so that writes
6510 * may proceed without blocking. It is important to call this before
6511 * attempting a GFP_KERNEL allocation while holding the mddev lock.
6512 * Must be called with mddev_lock held.
6514 * In the ->external case MD_CHANGE_CLEAN can not be cleared until mddev->lock
6515 * is dropped, so return -EAGAIN after notifying userspace.
6517 int md_allow_write(mddev_t *mddev)
6523 if (!mddev->pers->sync_request)
6526 spin_lock_irq(&mddev->write_lock);
6527 if (mddev->in_sync) {
6529 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
6530 if (mddev->safemode_delay &&
6531 mddev->safemode == 0)
6532 mddev->safemode = 1;
6533 spin_unlock_irq(&mddev->write_lock);
6534 md_update_sb(mddev, 0);
6535 sysfs_notify_dirent(mddev->sysfs_state);
6537 spin_unlock_irq(&mddev->write_lock);
6539 if (test_bit(MD_CHANGE_CLEAN, &mddev->flags))
6544 EXPORT_SYMBOL_GPL(md_allow_write);
6546 #define SYNC_MARKS 10
6547 #define SYNC_MARK_STEP (3*HZ)
6548 void md_do_sync(mddev_t *mddev)
6551 unsigned int currspeed = 0,
6553 sector_t max_sectors,j, io_sectors;
6554 unsigned long mark[SYNC_MARKS];
6555 sector_t mark_cnt[SYNC_MARKS];
6557 struct list_head *tmp;
6558 sector_t last_check;
6563 /* just incase thread restarts... */
6564 if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
6566 if (mddev->ro) /* never try to sync a read-only array */
6569 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
6570 if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
6571 desc = "data-check";
6572 else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
6573 desc = "requested-resync";
6576 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
6581 /* we overload curr_resync somewhat here.
6582 * 0 == not engaged in resync at all
6583 * 2 == checking that there is no conflict with another sync
6584 * 1 == like 2, but have yielded to allow conflicting resync to
6586 * other == active in resync - this many blocks
6588 * Before starting a resync we must have set curr_resync to
6589 * 2, and then checked that every "conflicting" array has curr_resync
6590 * less than ours. When we find one that is the same or higher
6591 * we wait on resync_wait. To avoid deadlock, we reduce curr_resync
6592 * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
6593 * This will mean we have to start checking from the beginning again.
6598 mddev->curr_resync = 2;
6601 if (kthread_should_stop())
6602 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6604 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
6606 for_each_mddev(mddev2, tmp) {
6607 if (mddev2 == mddev)
6609 if (!mddev->parallel_resync
6610 && mddev2->curr_resync
6611 && match_mddev_units(mddev, mddev2)) {
6613 if (mddev < mddev2 && mddev->curr_resync == 2) {
6614 /* arbitrarily yield */
6615 mddev->curr_resync = 1;
6616 wake_up(&resync_wait);
6618 if (mddev > mddev2 && mddev->curr_resync == 1)
6619 /* no need to wait here, we can wait the next
6620 * time 'round when curr_resync == 2
6623 /* We need to wait 'interruptible' so as not to
6624 * contribute to the load average, and not to
6625 * be caught by 'softlockup'
6627 prepare_to_wait(&resync_wait, &wq, TASK_INTERRUPTIBLE);
6628 if (!kthread_should_stop() &&
6629 mddev2->curr_resync >= mddev->curr_resync) {
6630 printk(KERN_INFO "md: delaying %s of %s"
6631 " until %s has finished (they"
6632 " share one or more physical units)\n",
6633 desc, mdname(mddev), mdname(mddev2));
6635 if (signal_pending(current))
6636 flush_signals(current);
6638 finish_wait(&resync_wait, &wq);
6641 finish_wait(&resync_wait, &wq);
6644 } while (mddev->curr_resync < 2);
6647 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
6648 /* resync follows the size requested by the personality,
6649 * which defaults to physical size, but can be virtual size
6651 max_sectors = mddev->resync_max_sectors;
6652 mddev->resync_mismatches = 0;
6653 /* we don't use the checkpoint if there's a bitmap */
6654 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
6655 j = mddev->resync_min;
6656 else if (!mddev->bitmap)
6657 j = mddev->recovery_cp;
6659 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
6660 max_sectors = mddev->dev_sectors;
6662 /* recovery follows the physical size of devices */
6663 max_sectors = mddev->dev_sectors;
6666 list_for_each_entry_rcu(rdev, &mddev->disks, same_set)
6667 if (rdev->raid_disk >= 0 &&
6668 !test_bit(Faulty, &rdev->flags) &&
6669 !test_bit(In_sync, &rdev->flags) &&
6670 rdev->recovery_offset < j)
6671 j = rdev->recovery_offset;
6675 printk(KERN_INFO "md: %s of RAID array %s\n", desc, mdname(mddev));
6676 printk(KERN_INFO "md: minimum _guaranteed_ speed:"
6677 " %d KB/sec/disk.\n", speed_min(mddev));
6678 printk(KERN_INFO "md: using maximum available idle IO bandwidth "
6679 "(but not more than %d KB/sec) for %s.\n",
6680 speed_max(mddev), desc);
6682 is_mddev_idle(mddev, 1); /* this initializes IO event counters */
6685 for (m = 0; m < SYNC_MARKS; m++) {
6687 mark_cnt[m] = io_sectors;
6690 mddev->resync_mark = mark[last_mark];
6691 mddev->resync_mark_cnt = mark_cnt[last_mark];
6694 * Tune reconstruction:
6696 window = 32*(PAGE_SIZE/512);
6697 printk(KERN_INFO "md: using %dk window, over a total of %llu blocks.\n",
6698 window/2,(unsigned long long) max_sectors/2);
6700 atomic_set(&mddev->recovery_active, 0);
6705 "md: resuming %s of %s from checkpoint.\n",
6706 desc, mdname(mddev));
6707 mddev->curr_resync = j;
6709 mddev->curr_resync_completed = mddev->curr_resync;
6711 while (j < max_sectors) {
6716 if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
6717 ((mddev->curr_resync > mddev->curr_resync_completed &&
6718 (mddev->curr_resync - mddev->curr_resync_completed)
6719 > (max_sectors >> 4)) ||
6720 (j - mddev->curr_resync_completed)*2
6721 >= mddev->resync_max - mddev->curr_resync_completed
6723 /* time to update curr_resync_completed */
6724 blk_unplug(mddev->queue);
6725 wait_event(mddev->recovery_wait,
6726 atomic_read(&mddev->recovery_active) == 0);
6727 mddev->curr_resync_completed =
6729 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
6730 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
6733 while (j >= mddev->resync_max && !kthread_should_stop()) {
6734 /* As this condition is controlled by user-space,
6735 * we can block indefinitely, so use '_interruptible'
6736 * to avoid triggering warnings.
6738 flush_signals(current); /* just in case */
6739 wait_event_interruptible(mddev->recovery_wait,
6740 mddev->resync_max > j
6741 || kthread_should_stop());
6744 if (kthread_should_stop())
6747 sectors = mddev->pers->sync_request(mddev, j, &skipped,
6748 currspeed < speed_min(mddev));
6750 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6754 if (!skipped) { /* actual IO requested */
6755 io_sectors += sectors;
6756 atomic_add(sectors, &mddev->recovery_active);
6760 if (j>1) mddev->curr_resync = j;
6761 mddev->curr_mark_cnt = io_sectors;
6762 if (last_check == 0)
6763 /* this is the earliers that rebuilt will be
6764 * visible in /proc/mdstat
6766 md_new_event(mddev);
6768 if (last_check + window > io_sectors || j == max_sectors)
6771 last_check = io_sectors;
6773 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
6777 if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) {
6779 int next = (last_mark+1) % SYNC_MARKS;
6781 mddev->resync_mark = mark[next];
6782 mddev->resync_mark_cnt = mark_cnt[next];
6783 mark[next] = jiffies;
6784 mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active);
6789 if (kthread_should_stop())
6794 * this loop exits only if either when we are slower than
6795 * the 'hard' speed limit, or the system was IO-idle for
6797 * the system might be non-idle CPU-wise, but we only care
6798 * about not overloading the IO subsystem. (things like an
6799 * e2fsck being done on the RAID array should execute fast)
6801 blk_unplug(mddev->queue);
6804 currspeed = ((unsigned long)(io_sectors-mddev->resync_mark_cnt))/2
6805 /((jiffies-mddev->resync_mark)/HZ +1) +1;
6807 if (currspeed > speed_min(mddev)) {
6808 if ((currspeed > speed_max(mddev)) ||
6809 !is_mddev_idle(mddev, 0)) {
6815 printk(KERN_INFO "md: %s: %s done.\n",mdname(mddev), desc);
6817 * this also signals 'finished resyncing' to md_stop
6820 blk_unplug(mddev->queue);
6822 wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
6824 /* tell personality that we are finished */
6825 mddev->pers->sync_request(mddev, max_sectors, &skipped, 1);
6827 if (!test_bit(MD_RECOVERY_CHECK, &mddev->recovery) &&
6828 mddev->curr_resync > 2) {
6829 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
6830 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
6831 if (mddev->curr_resync >= mddev->recovery_cp) {
6833 "md: checkpointing %s of %s.\n",
6834 desc, mdname(mddev));
6835 mddev->recovery_cp = mddev->curr_resync;
6838 mddev->recovery_cp = MaxSector;
6840 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery))
6841 mddev->curr_resync = MaxSector;
6843 list_for_each_entry_rcu(rdev, &mddev->disks, same_set)
6844 if (rdev->raid_disk >= 0 &&
6845 !test_bit(Faulty, &rdev->flags) &&
6846 !test_bit(In_sync, &rdev->flags) &&
6847 rdev->recovery_offset < mddev->curr_resync)
6848 rdev->recovery_offset = mddev->curr_resync;
6852 set_bit(MD_CHANGE_DEVS, &mddev->flags);
6855 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
6856 /* We completed so min/max setting can be forgotten if used. */
6857 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
6858 mddev->resync_min = 0;
6859 mddev->resync_max = MaxSector;
6860 } else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
6861 mddev->resync_min = mddev->curr_resync_completed;
6862 mddev->curr_resync = 0;
6863 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery))
6864 mddev->curr_resync_completed = 0;
6865 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
6866 wake_up(&resync_wait);
6867 set_bit(MD_RECOVERY_DONE, &mddev->recovery);
6868 md_wakeup_thread(mddev->thread);
6873 * got a signal, exit.
6876 "md: md_do_sync() got signal ... exiting\n");
6877 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6881 EXPORT_SYMBOL_GPL(md_do_sync);
6884 static int remove_and_add_spares(mddev_t *mddev)
6889 mddev->curr_resync_completed = 0;
6891 list_for_each_entry(rdev, &mddev->disks, same_set)
6892 if (rdev->raid_disk >= 0 &&
6893 !test_bit(Blocked, &rdev->flags) &&
6894 (test_bit(Faulty, &rdev->flags) ||
6895 ! test_bit(In_sync, &rdev->flags)) &&
6896 atomic_read(&rdev->nr_pending)==0) {
6897 if (mddev->pers->hot_remove_disk(
6898 mddev, rdev->raid_disk)==0) {
6900 sprintf(nm,"rd%d", rdev->raid_disk);
6901 sysfs_remove_link(&mddev->kobj, nm);
6902 rdev->raid_disk = -1;
6906 if (mddev->degraded && ! mddev->ro && !mddev->recovery_disabled) {
6907 list_for_each_entry(rdev, &mddev->disks, same_set) {
6908 if (rdev->raid_disk >= 0 &&
6909 !test_bit(In_sync, &rdev->flags) &&
6910 !test_bit(Blocked, &rdev->flags))
6912 if (rdev->raid_disk < 0
6913 && !test_bit(Faulty, &rdev->flags)) {
6914 rdev->recovery_offset = 0;
6916 hot_add_disk(mddev, rdev) == 0) {
6918 sprintf(nm, "rd%d", rdev->raid_disk);
6919 if (sysfs_create_link(&mddev->kobj,
6922 "md: cannot register "
6926 md_new_event(mddev);
6927 set_bit(MD_CHANGE_DEVS, &mddev->flags);
6936 * This routine is regularly called by all per-raid-array threads to
6937 * deal with generic issues like resync and super-block update.
6938 * Raid personalities that don't have a thread (linear/raid0) do not
6939 * need this as they never do any recovery or update the superblock.
6941 * It does not do any resync itself, but rather "forks" off other threads
6942 * to do that as needed.
6943 * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
6944 * "->recovery" and create a thread at ->sync_thread.
6945 * When the thread finishes it sets MD_RECOVERY_DONE
6946 * and wakeups up this thread which will reap the thread and finish up.
6947 * This thread also removes any faulty devices (with nr_pending == 0).
6949 * The overall approach is:
6950 * 1/ if the superblock needs updating, update it.
6951 * 2/ If a recovery thread is running, don't do anything else.
6952 * 3/ If recovery has finished, clean up, possibly marking spares active.
6953 * 4/ If there are any faulty devices, remove them.
6954 * 5/ If array is degraded, try to add spares devices
6955 * 6/ If array has spares or is not in-sync, start a resync thread.
6957 void md_check_recovery(mddev_t *mddev)
6963 bitmap_daemon_work(mddev);
6968 if (signal_pending(current)) {
6969 if (mddev->pers->sync_request && !mddev->external) {
6970 printk(KERN_INFO "md: %s in immediate safe mode\n",
6972 mddev->safemode = 2;
6974 flush_signals(current);
6977 if (mddev->ro && !test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
6980 (mddev->flags && !mddev->external) ||
6981 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
6982 test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
6983 (mddev->external == 0 && mddev->safemode == 1) ||
6984 (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending)
6985 && !mddev->in_sync && mddev->recovery_cp == MaxSector)
6989 if (mddev_trylock(mddev)) {
6993 /* Only thing we do on a ro array is remove
6996 remove_and_add_spares(mddev);
6997 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7001 if (!mddev->external) {
7003 spin_lock_irq(&mddev->write_lock);
7004 if (mddev->safemode &&
7005 !atomic_read(&mddev->writes_pending) &&
7007 mddev->recovery_cp == MaxSector) {
7010 if (mddev->persistent)
7011 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
7013 if (mddev->safemode == 1)
7014 mddev->safemode = 0;
7015 spin_unlock_irq(&mddev->write_lock);
7017 sysfs_notify_dirent(mddev->sysfs_state);
7021 md_update_sb(mddev, 0);
7023 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
7024 !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
7025 /* resync/recovery still happening */
7026 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7029 if (mddev->sync_thread) {
7030 /* resync has finished, collect result */
7031 md_unregister_thread(mddev->sync_thread);
7032 mddev->sync_thread = NULL;
7033 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
7034 !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
7036 /* activate any spares */
7037 if (mddev->pers->spare_active(mddev))
7038 sysfs_notify(&mddev->kobj, NULL,
7041 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
7042 mddev->pers->finish_reshape)
7043 mddev->pers->finish_reshape(mddev);
7044 md_update_sb(mddev, 1);
7046 /* if array is no-longer degraded, then any saved_raid_disk
7047 * information must be scrapped
7049 if (!mddev->degraded)
7050 list_for_each_entry(rdev, &mddev->disks, same_set)
7051 rdev->saved_raid_disk = -1;
7053 mddev->recovery = 0;
7054 /* flag recovery needed just to double check */
7055 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7056 sysfs_notify_dirent(mddev->sysfs_action);
7057 md_new_event(mddev);
7060 /* Set RUNNING before clearing NEEDED to avoid
7061 * any transients in the value of "sync_action".
7063 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
7064 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7065 /* Clear some bits that don't mean anything, but
7068 clear_bit(MD_RECOVERY_INTR, &mddev->recovery);
7069 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
7071 if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
7073 /* no recovery is running.
7074 * remove any failed drives, then
7075 * add spares if possible.
7076 * Spare are also removed and re-added, to allow
7077 * the personality to fail the re-add.
7080 if (mddev->reshape_position != MaxSector) {
7081 if (mddev->pers->check_reshape == NULL ||
7082 mddev->pers->check_reshape(mddev) != 0)
7083 /* Cannot proceed */
7085 set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
7086 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
7087 } else if ((spares = remove_and_add_spares(mddev))) {
7088 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
7089 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
7090 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
7091 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
7092 } else if (mddev->recovery_cp < MaxSector) {
7093 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
7094 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
7095 } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
7096 /* nothing to be done ... */
7099 if (mddev->pers->sync_request) {
7100 if (spares && mddev->bitmap && ! mddev->bitmap->file) {
7101 /* We are adding a device or devices to an array
7102 * which has the bitmap stored on all devices.
7103 * So make sure all bitmap pages get written
7105 bitmap_write_all(mddev->bitmap);
7107 mddev->sync_thread = md_register_thread(md_do_sync,
7110 if (!mddev->sync_thread) {
7111 printk(KERN_ERR "%s: could not start resync"
7114 /* leave the spares where they are, it shouldn't hurt */
7115 mddev->recovery = 0;
7117 md_wakeup_thread(mddev->sync_thread);
7118 sysfs_notify_dirent(mddev->sysfs_action);
7119 md_new_event(mddev);
7122 if (!mddev->sync_thread) {
7123 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
7124 if (test_and_clear_bit(MD_RECOVERY_RECOVER,
7126 if (mddev->sysfs_action)
7127 sysfs_notify_dirent(mddev->sysfs_action);
7129 mddev_unlock(mddev);
7133 void md_wait_for_blocked_rdev(mdk_rdev_t *rdev, mddev_t *mddev)
7135 sysfs_notify_dirent(rdev->sysfs_state);
7136 wait_event_timeout(rdev->blocked_wait,
7137 !test_bit(Blocked, &rdev->flags),
7138 msecs_to_jiffies(5000));
7139 rdev_dec_pending(rdev, mddev);
7141 EXPORT_SYMBOL(md_wait_for_blocked_rdev);
7143 static int md_notify_reboot(struct notifier_block *this,
7144 unsigned long code, void *x)
7146 struct list_head *tmp;
7149 if ((code == SYS_DOWN) || (code == SYS_HALT) || (code == SYS_POWER_OFF)) {
7151 printk(KERN_INFO "md: stopping all md devices.\n");
7153 for_each_mddev(mddev, tmp)
7154 if (mddev_trylock(mddev)) {
7155 /* Force a switch to readonly even array
7156 * appears to still be in use. Hence
7159 md_set_readonly(mddev, 100);
7160 mddev_unlock(mddev);
7163 * certain more exotic SCSI devices are known to be
7164 * volatile wrt too early system reboots. While the
7165 * right place to handle this issue is the given
7166 * driver, we do want to have a safe RAID driver ...
7173 static struct notifier_block md_notifier = {
7174 .notifier_call = md_notify_reboot,
7176 .priority = INT_MAX, /* before any real devices */
7179 static void md_geninit(void)
7181 dprintk("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
7183 proc_create("mdstat", S_IRUGO, NULL, &md_seq_fops);
7186 static int __init md_init(void)
7188 if (register_blkdev(MD_MAJOR, "md"))
7190 if ((mdp_major=register_blkdev(0, "mdp"))<=0) {
7191 unregister_blkdev(MD_MAJOR, "md");
7194 blk_register_region(MKDEV(MD_MAJOR, 0), 1UL<<MINORBITS, THIS_MODULE,
7195 md_probe, NULL, NULL);
7196 blk_register_region(MKDEV(mdp_major, 0), 1UL<<MINORBITS, THIS_MODULE,
7197 md_probe, NULL, NULL);
7199 register_reboot_notifier(&md_notifier);
7200 raid_table_header = register_sysctl_table(raid_root_table);
7210 * Searches all registered partitions for autorun RAID arrays
7214 static LIST_HEAD(all_detected_devices);
7215 struct detected_devices_node {
7216 struct list_head list;
7220 void md_autodetect_dev(dev_t dev)
7222 struct detected_devices_node *node_detected_dev;
7224 node_detected_dev = kzalloc(sizeof(*node_detected_dev), GFP_KERNEL);
7225 if (node_detected_dev) {
7226 node_detected_dev->dev = dev;
7227 list_add_tail(&node_detected_dev->list, &all_detected_devices);
7229 printk(KERN_CRIT "md: md_autodetect_dev: kzalloc failed"
7230 ", skipping dev(%d,%d)\n", MAJOR(dev), MINOR(dev));
7235 static void autostart_arrays(int part)
7238 struct detected_devices_node *node_detected_dev;
7240 int i_scanned, i_passed;
7245 printk(KERN_INFO "md: Autodetecting RAID arrays.\n");
7247 while (!list_empty(&all_detected_devices) && i_scanned < INT_MAX) {
7249 node_detected_dev = list_entry(all_detected_devices.next,
7250 struct detected_devices_node, list);
7251 list_del(&node_detected_dev->list);
7252 dev = node_detected_dev->dev;
7253 kfree(node_detected_dev);
7254 rdev = md_import_device(dev,0, 90);
7258 if (test_bit(Faulty, &rdev->flags)) {
7262 set_bit(AutoDetected, &rdev->flags);
7263 list_add(&rdev->same_set, &pending_raid_disks);
7267 printk(KERN_INFO "md: Scanned %d and added %d devices.\n",
7268 i_scanned, i_passed);
7270 autorun_devices(part);
7273 #endif /* !MODULE */
7275 static __exit void md_exit(void)
7278 struct list_head *tmp;
7280 blk_unregister_region(MKDEV(MD_MAJOR,0), 1U << MINORBITS);
7281 blk_unregister_region(MKDEV(mdp_major,0), 1U << MINORBITS);
7283 unregister_blkdev(MD_MAJOR,"md");
7284 unregister_blkdev(mdp_major, "mdp");
7285 unregister_reboot_notifier(&md_notifier);
7286 unregister_sysctl_table(raid_table_header);
7287 remove_proc_entry("mdstat", NULL);
7288 for_each_mddev(mddev, tmp) {
7289 export_array(mddev);
7290 mddev->hold_active = 0;
7294 subsys_initcall(md_init);
7295 module_exit(md_exit)
7297 static int get_ro(char *buffer, struct kernel_param *kp)
7299 return sprintf(buffer, "%d", start_readonly);
7301 static int set_ro(const char *val, struct kernel_param *kp)
7304 int num = simple_strtoul(val, &e, 10);
7305 if (*val && (*e == '\0' || *e == '\n')) {
7306 start_readonly = num;
7312 module_param_call(start_ro, set_ro, get_ro, NULL, S_IRUSR|S_IWUSR);
7313 module_param(start_dirty_degraded, int, S_IRUGO|S_IWUSR);
7315 module_param_call(new_array, add_named_array, NULL, NULL, S_IWUSR);
7317 EXPORT_SYMBOL(register_md_personality);
7318 EXPORT_SYMBOL(unregister_md_personality);
7319 EXPORT_SYMBOL(md_error);
7320 EXPORT_SYMBOL(md_done_sync);
7321 EXPORT_SYMBOL(md_write_start);
7322 EXPORT_SYMBOL(md_write_end);
7323 EXPORT_SYMBOL(md_register_thread);
7324 EXPORT_SYMBOL(md_unregister_thread);
7325 EXPORT_SYMBOL(md_wakeup_thread);
7326 EXPORT_SYMBOL(md_check_recovery);
7327 MODULE_LICENSE("GPL");
7328 MODULE_DESCRIPTION("MD RAID framework");
7330 MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR);