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>
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/module.h>
47 #include <linux/reboot.h>
48 #include <linux/file.h>
49 #include <linux/compat.h>
50 #include <linux/delay.h>
51 #include <linux/raid/md_p.h>
52 #include <linux/raid/md_u.h>
53 #include <linux/slab.h>
58 static void autostart_arrays(int part);
61 /* pers_list is a list of registered personalities protected
63 * pers_lock does extra service to protect accesses to
64 * mddev->thread when the mutex cannot be held.
66 static LIST_HEAD(pers_list);
67 static DEFINE_SPINLOCK(pers_lock);
69 static void md_print_devices(void);
71 static DECLARE_WAIT_QUEUE_HEAD(resync_wait);
72 static struct workqueue_struct *md_wq;
73 static struct workqueue_struct *md_misc_wq;
75 static int remove_and_add_spares(struct mddev *mddev,
76 struct md_rdev *this);
78 #define MD_BUG(x...) { printk("md: bug in file %s, line %d\n", __FILE__, __LINE__); md_print_devices(); }
81 * Default number of read corrections we'll attempt on an rdev
82 * before ejecting it from the array. We divide the read error
83 * count by 2 for every hour elapsed between read errors.
85 #define MD_DEFAULT_MAX_CORRECTED_READ_ERRORS 20
87 * Current RAID-1,4,5 parallel reconstruction 'guaranteed speed limit'
88 * is 1000 KB/sec, so the extra system load does not show up that much.
89 * Increase it if you want to have more _guaranteed_ speed. Note that
90 * the RAID driver will use the maximum available bandwidth if the IO
91 * subsystem is idle. There is also an 'absolute maximum' reconstruction
92 * speed limit - in case reconstruction slows down your system despite
95 * you can change it via /proc/sys/dev/raid/speed_limit_min and _max.
96 * or /sys/block/mdX/md/sync_speed_{min,max}
99 static int sysctl_speed_limit_min = 1000;
100 static int sysctl_speed_limit_max = 200000;
101 static inline int speed_min(struct mddev *mddev)
103 return mddev->sync_speed_min ?
104 mddev->sync_speed_min : sysctl_speed_limit_min;
107 static inline int speed_max(struct mddev *mddev)
109 return mddev->sync_speed_max ?
110 mddev->sync_speed_max : sysctl_speed_limit_max;
113 static struct ctl_table_header *raid_table_header;
115 static struct ctl_table raid_table[] = {
117 .procname = "speed_limit_min",
118 .data = &sysctl_speed_limit_min,
119 .maxlen = sizeof(int),
120 .mode = S_IRUGO|S_IWUSR,
121 .proc_handler = proc_dointvec,
124 .procname = "speed_limit_max",
125 .data = &sysctl_speed_limit_max,
126 .maxlen = sizeof(int),
127 .mode = S_IRUGO|S_IWUSR,
128 .proc_handler = proc_dointvec,
133 static struct ctl_table raid_dir_table[] = {
137 .mode = S_IRUGO|S_IXUGO,
143 static struct ctl_table raid_root_table[] = {
148 .child = raid_dir_table,
153 static const struct block_device_operations md_fops;
155 static int start_readonly;
158 * like bio_clone, but with a local bio set
161 struct bio *bio_alloc_mddev(gfp_t gfp_mask, int nr_iovecs,
166 if (!mddev || !mddev->bio_set)
167 return bio_alloc(gfp_mask, nr_iovecs);
169 b = bio_alloc_bioset(gfp_mask, nr_iovecs, mddev->bio_set);
174 EXPORT_SYMBOL_GPL(bio_alloc_mddev);
176 struct bio *bio_clone_mddev(struct bio *bio, gfp_t gfp_mask,
179 if (!mddev || !mddev->bio_set)
180 return bio_clone(bio, gfp_mask);
182 return bio_clone_bioset(bio, gfp_mask, mddev->bio_set);
184 EXPORT_SYMBOL_GPL(bio_clone_mddev);
187 * We have a system wide 'event count' that is incremented
188 * on any 'interesting' event, and readers of /proc/mdstat
189 * can use 'poll' or 'select' to find out when the event
193 * start array, stop array, error, add device, remove device,
194 * start build, activate spare
196 static DECLARE_WAIT_QUEUE_HEAD(md_event_waiters);
197 static atomic_t md_event_count;
198 void md_new_event(struct mddev *mddev)
200 atomic_inc(&md_event_count);
201 wake_up(&md_event_waiters);
203 EXPORT_SYMBOL_GPL(md_new_event);
205 /* Alternate version that can be called from interrupts
206 * when calling sysfs_notify isn't needed.
208 static void md_new_event_inintr(struct mddev *mddev)
210 atomic_inc(&md_event_count);
211 wake_up(&md_event_waiters);
215 * Enables to iterate over all existing md arrays
216 * all_mddevs_lock protects this list.
218 static LIST_HEAD(all_mddevs);
219 static DEFINE_SPINLOCK(all_mddevs_lock);
223 * iterates through all used mddevs in the system.
224 * We take care to grab the all_mddevs_lock whenever navigating
225 * the list, and to always hold a refcount when unlocked.
226 * Any code which breaks out of this loop while own
227 * a reference to the current mddev and must mddev_put it.
229 #define for_each_mddev(_mddev,_tmp) \
231 for (({ spin_lock(&all_mddevs_lock); \
232 _tmp = all_mddevs.next; \
234 ({ if (_tmp != &all_mddevs) \
235 mddev_get(list_entry(_tmp, struct mddev, all_mddevs));\
236 spin_unlock(&all_mddevs_lock); \
237 if (_mddev) mddev_put(_mddev); \
238 _mddev = list_entry(_tmp, struct mddev, all_mddevs); \
239 _tmp != &all_mddevs;}); \
240 ({ spin_lock(&all_mddevs_lock); \
241 _tmp = _tmp->next;}) \
245 /* Rather than calling directly into the personality make_request function,
246 * IO requests come here first so that we can check if the device is
247 * being suspended pending a reconfiguration.
248 * We hold a refcount over the call to ->make_request. By the time that
249 * call has finished, the bio has been linked into some internal structure
250 * and so is visible to ->quiesce(), so we don't need the refcount any more.
252 static void md_make_request(struct request_queue *q, struct bio *bio)
254 const int rw = bio_data_dir(bio);
255 struct mddev *mddev = q->queuedata;
257 unsigned int sectors;
259 if (mddev == NULL || mddev->pers == NULL
264 if (mddev->ro == 1 && unlikely(rw == WRITE)) {
265 bio_endio(bio, bio_sectors(bio) == 0 ? 0 : -EROFS);
268 smp_rmb(); /* Ensure implications of 'active' are visible */
270 if (mddev->suspended) {
273 prepare_to_wait(&mddev->sb_wait, &__wait,
274 TASK_UNINTERRUPTIBLE);
275 if (!mddev->suspended)
281 finish_wait(&mddev->sb_wait, &__wait);
283 atomic_inc(&mddev->active_io);
287 * save the sectors now since our bio can
288 * go away inside make_request
290 sectors = bio_sectors(bio);
291 mddev->pers->make_request(mddev, bio);
293 cpu = part_stat_lock();
294 part_stat_inc(cpu, &mddev->gendisk->part0, ios[rw]);
295 part_stat_add(cpu, &mddev->gendisk->part0, sectors[rw], sectors);
298 if (atomic_dec_and_test(&mddev->active_io) && mddev->suspended)
299 wake_up(&mddev->sb_wait);
302 /* mddev_suspend makes sure no new requests are submitted
303 * to the device, and that any requests that have been submitted
304 * are completely handled.
305 * Once ->stop is called and completes, the module will be completely
308 void mddev_suspend(struct mddev *mddev)
310 BUG_ON(mddev->suspended);
311 mddev->suspended = 1;
313 wait_event(mddev->sb_wait, atomic_read(&mddev->active_io) == 0);
314 mddev->pers->quiesce(mddev, 1);
316 del_timer_sync(&mddev->safemode_timer);
318 EXPORT_SYMBOL_GPL(mddev_suspend);
320 void mddev_resume(struct mddev *mddev)
322 mddev->suspended = 0;
323 wake_up(&mddev->sb_wait);
324 mddev->pers->quiesce(mddev, 0);
326 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
327 md_wakeup_thread(mddev->thread);
328 md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
330 EXPORT_SYMBOL_GPL(mddev_resume);
332 int mddev_congested(struct mddev *mddev, int bits)
334 return mddev->suspended;
336 EXPORT_SYMBOL(mddev_congested);
339 * Generic flush handling for md
342 static void md_end_flush(struct bio *bio, int err)
344 struct md_rdev *rdev = bio->bi_private;
345 struct mddev *mddev = rdev->mddev;
347 rdev_dec_pending(rdev, mddev);
349 if (atomic_dec_and_test(&mddev->flush_pending)) {
350 /* The pre-request flush has finished */
351 queue_work(md_wq, &mddev->flush_work);
356 static void md_submit_flush_data(struct work_struct *ws);
358 static void submit_flushes(struct work_struct *ws)
360 struct mddev *mddev = container_of(ws, struct mddev, flush_work);
361 struct md_rdev *rdev;
363 INIT_WORK(&mddev->flush_work, md_submit_flush_data);
364 atomic_set(&mddev->flush_pending, 1);
366 rdev_for_each_rcu(rdev, mddev)
367 if (rdev->raid_disk >= 0 &&
368 !test_bit(Faulty, &rdev->flags)) {
369 /* Take two references, one is dropped
370 * when request finishes, one after
371 * we reclaim rcu_read_lock
374 atomic_inc(&rdev->nr_pending);
375 atomic_inc(&rdev->nr_pending);
377 bi = bio_alloc_mddev(GFP_NOIO, 0, mddev);
378 bi->bi_end_io = md_end_flush;
379 bi->bi_private = rdev;
380 bi->bi_bdev = rdev->bdev;
381 atomic_inc(&mddev->flush_pending);
382 submit_bio(WRITE_FLUSH, bi);
384 rdev_dec_pending(rdev, mddev);
387 if (atomic_dec_and_test(&mddev->flush_pending))
388 queue_work(md_wq, &mddev->flush_work);
391 static void md_submit_flush_data(struct work_struct *ws)
393 struct mddev *mddev = container_of(ws, struct mddev, flush_work);
394 struct bio *bio = mddev->flush_bio;
396 if (bio->bi_iter.bi_size == 0)
397 /* an empty barrier - all done */
400 bio->bi_rw &= ~REQ_FLUSH;
401 mddev->pers->make_request(mddev, bio);
404 mddev->flush_bio = NULL;
405 wake_up(&mddev->sb_wait);
408 void md_flush_request(struct mddev *mddev, struct bio *bio)
410 spin_lock_irq(&mddev->write_lock);
411 wait_event_lock_irq(mddev->sb_wait,
414 mddev->flush_bio = bio;
415 spin_unlock_irq(&mddev->write_lock);
417 INIT_WORK(&mddev->flush_work, submit_flushes);
418 queue_work(md_wq, &mddev->flush_work);
420 EXPORT_SYMBOL(md_flush_request);
422 void md_unplug(struct blk_plug_cb *cb, bool from_schedule)
424 struct mddev *mddev = cb->data;
425 md_wakeup_thread(mddev->thread);
428 EXPORT_SYMBOL(md_unplug);
430 static inline struct mddev *mddev_get(struct mddev *mddev)
432 atomic_inc(&mddev->active);
436 static void mddev_delayed_delete(struct work_struct *ws);
438 static void mddev_put(struct mddev *mddev)
440 struct bio_set *bs = NULL;
442 if (!atomic_dec_and_lock(&mddev->active, &all_mddevs_lock))
444 if (!mddev->raid_disks && list_empty(&mddev->disks) &&
445 mddev->ctime == 0 && !mddev->hold_active) {
446 /* Array is not configured at all, and not held active,
448 list_del_init(&mddev->all_mddevs);
450 mddev->bio_set = NULL;
451 if (mddev->gendisk) {
452 /* We did a probe so need to clean up. Call
453 * queue_work inside the spinlock so that
454 * flush_workqueue() after mddev_find will
455 * succeed in waiting for the work to be done.
457 INIT_WORK(&mddev->del_work, mddev_delayed_delete);
458 queue_work(md_misc_wq, &mddev->del_work);
462 spin_unlock(&all_mddevs_lock);
467 void mddev_init(struct mddev *mddev)
469 mutex_init(&mddev->open_mutex);
470 mutex_init(&mddev->reconfig_mutex);
471 mutex_init(&mddev->bitmap_info.mutex);
472 INIT_LIST_HEAD(&mddev->disks);
473 INIT_LIST_HEAD(&mddev->all_mddevs);
474 init_timer(&mddev->safemode_timer);
475 atomic_set(&mddev->active, 1);
476 atomic_set(&mddev->openers, 0);
477 atomic_set(&mddev->active_io, 0);
478 spin_lock_init(&mddev->write_lock);
479 atomic_set(&mddev->flush_pending, 0);
480 init_waitqueue_head(&mddev->sb_wait);
481 init_waitqueue_head(&mddev->recovery_wait);
482 mddev->reshape_position = MaxSector;
483 mddev->reshape_backwards = 0;
484 mddev->last_sync_action = "none";
485 mddev->resync_min = 0;
486 mddev->resync_max = MaxSector;
487 mddev->level = LEVEL_NONE;
489 EXPORT_SYMBOL_GPL(mddev_init);
491 static struct mddev * mddev_find(dev_t unit)
493 struct mddev *mddev, *new = NULL;
495 if (unit && MAJOR(unit) != MD_MAJOR)
496 unit &= ~((1<<MdpMinorShift)-1);
499 spin_lock(&all_mddevs_lock);
502 list_for_each_entry(mddev, &all_mddevs, all_mddevs)
503 if (mddev->unit == unit) {
505 spin_unlock(&all_mddevs_lock);
511 list_add(&new->all_mddevs, &all_mddevs);
512 spin_unlock(&all_mddevs_lock);
513 new->hold_active = UNTIL_IOCTL;
517 /* find an unused unit number */
518 static int next_minor = 512;
519 int start = next_minor;
523 dev = MKDEV(MD_MAJOR, next_minor);
525 if (next_minor > MINORMASK)
527 if (next_minor == start) {
528 /* Oh dear, all in use. */
529 spin_unlock(&all_mddevs_lock);
535 list_for_each_entry(mddev, &all_mddevs, all_mddevs)
536 if (mddev->unit == dev) {
542 new->md_minor = MINOR(dev);
543 new->hold_active = UNTIL_STOP;
544 list_add(&new->all_mddevs, &all_mddevs);
545 spin_unlock(&all_mddevs_lock);
548 spin_unlock(&all_mddevs_lock);
550 new = kzalloc(sizeof(*new), GFP_KERNEL);
555 if (MAJOR(unit) == MD_MAJOR)
556 new->md_minor = MINOR(unit);
558 new->md_minor = MINOR(unit) >> MdpMinorShift;
565 static inline int __must_check mddev_lock(struct mddev * mddev)
567 return mutex_lock_interruptible(&mddev->reconfig_mutex);
570 /* Sometimes we need to take the lock in a situation where
571 * failure due to interrupts is not acceptable.
573 static inline void mddev_lock_nointr(struct mddev * mddev)
575 mutex_lock(&mddev->reconfig_mutex);
578 static inline int mddev_is_locked(struct mddev *mddev)
580 return mutex_is_locked(&mddev->reconfig_mutex);
583 static inline int mddev_trylock(struct mddev * mddev)
585 return mutex_trylock(&mddev->reconfig_mutex);
588 static struct attribute_group md_redundancy_group;
590 static void mddev_unlock(struct mddev * mddev)
592 if (mddev->to_remove) {
593 /* These cannot be removed under reconfig_mutex as
594 * an access to the files will try to take reconfig_mutex
595 * while holding the file unremovable, which leads to
597 * So hold set sysfs_active while the remove in happeing,
598 * and anything else which might set ->to_remove or my
599 * otherwise change the sysfs namespace will fail with
600 * -EBUSY if sysfs_active is still set.
601 * We set sysfs_active under reconfig_mutex and elsewhere
602 * test it under the same mutex to ensure its correct value
605 struct attribute_group *to_remove = mddev->to_remove;
606 mddev->to_remove = NULL;
607 mddev->sysfs_active = 1;
608 mutex_unlock(&mddev->reconfig_mutex);
610 if (mddev->kobj.sd) {
611 if (to_remove != &md_redundancy_group)
612 sysfs_remove_group(&mddev->kobj, to_remove);
613 if (mddev->pers == NULL ||
614 mddev->pers->sync_request == NULL) {
615 sysfs_remove_group(&mddev->kobj, &md_redundancy_group);
616 if (mddev->sysfs_action)
617 sysfs_put(mddev->sysfs_action);
618 mddev->sysfs_action = NULL;
621 mddev->sysfs_active = 0;
623 mutex_unlock(&mddev->reconfig_mutex);
625 /* As we've dropped the mutex we need a spinlock to
626 * make sure the thread doesn't disappear
628 spin_lock(&pers_lock);
629 md_wakeup_thread(mddev->thread);
630 spin_unlock(&pers_lock);
633 static struct md_rdev *find_rdev_nr_rcu(struct mddev *mddev, int nr)
635 struct md_rdev *rdev;
637 rdev_for_each_rcu(rdev, mddev)
638 if (rdev->desc_nr == nr)
644 static struct md_rdev *find_rdev(struct mddev *mddev, dev_t dev)
646 struct md_rdev *rdev;
648 rdev_for_each(rdev, mddev)
649 if (rdev->bdev->bd_dev == dev)
655 static struct md_rdev *find_rdev_rcu(struct mddev *mddev, dev_t dev)
657 struct md_rdev *rdev;
659 rdev_for_each_rcu(rdev, mddev)
660 if (rdev->bdev->bd_dev == dev)
666 static struct md_personality *find_pers(int level, char *clevel)
668 struct md_personality *pers;
669 list_for_each_entry(pers, &pers_list, list) {
670 if (level != LEVEL_NONE && pers->level == level)
672 if (strcmp(pers->name, clevel)==0)
678 /* return the offset of the super block in 512byte sectors */
679 static inline sector_t calc_dev_sboffset(struct md_rdev *rdev)
681 sector_t num_sectors = i_size_read(rdev->bdev->bd_inode) / 512;
682 return MD_NEW_SIZE_SECTORS(num_sectors);
685 static int alloc_disk_sb(struct md_rdev * rdev)
690 rdev->sb_page = alloc_page(GFP_KERNEL);
691 if (!rdev->sb_page) {
692 printk(KERN_ALERT "md: out of memory.\n");
699 void md_rdev_clear(struct md_rdev *rdev)
702 put_page(rdev->sb_page);
704 rdev->sb_page = NULL;
709 put_page(rdev->bb_page);
710 rdev->bb_page = NULL;
712 kfree(rdev->badblocks.page);
713 rdev->badblocks.page = NULL;
715 EXPORT_SYMBOL_GPL(md_rdev_clear);
717 static void super_written(struct bio *bio, int error)
719 struct md_rdev *rdev = bio->bi_private;
720 struct mddev *mddev = rdev->mddev;
722 if (error || !test_bit(BIO_UPTODATE, &bio->bi_flags)) {
723 printk("md: super_written gets error=%d, uptodate=%d\n",
724 error, test_bit(BIO_UPTODATE, &bio->bi_flags));
725 WARN_ON(test_bit(BIO_UPTODATE, &bio->bi_flags));
726 md_error(mddev, rdev);
729 if (atomic_dec_and_test(&mddev->pending_writes))
730 wake_up(&mddev->sb_wait);
734 void md_super_write(struct mddev *mddev, struct md_rdev *rdev,
735 sector_t sector, int size, struct page *page)
737 /* write first size bytes of page to sector of rdev
738 * Increment mddev->pending_writes before returning
739 * and decrement it on completion, waking up sb_wait
740 * if zero is reached.
741 * If an error occurred, call md_error
743 struct bio *bio = bio_alloc_mddev(GFP_NOIO, 1, mddev);
745 bio->bi_bdev = rdev->meta_bdev ? rdev->meta_bdev : rdev->bdev;
746 bio->bi_iter.bi_sector = sector;
747 bio_add_page(bio, page, size, 0);
748 bio->bi_private = rdev;
749 bio->bi_end_io = super_written;
751 atomic_inc(&mddev->pending_writes);
752 submit_bio(WRITE_FLUSH_FUA, bio);
755 void md_super_wait(struct mddev *mddev)
757 /* wait for all superblock writes that were scheduled to complete */
758 wait_event(mddev->sb_wait, atomic_read(&mddev->pending_writes)==0);
761 int sync_page_io(struct md_rdev *rdev, sector_t sector, int size,
762 struct page *page, int rw, bool metadata_op)
764 struct bio *bio = bio_alloc_mddev(GFP_NOIO, 1, rdev->mddev);
767 bio->bi_bdev = (metadata_op && rdev->meta_bdev) ?
768 rdev->meta_bdev : rdev->bdev;
770 bio->bi_iter.bi_sector = sector + rdev->sb_start;
771 else if (rdev->mddev->reshape_position != MaxSector &&
772 (rdev->mddev->reshape_backwards ==
773 (sector >= rdev->mddev->reshape_position)))
774 bio->bi_iter.bi_sector = sector + rdev->new_data_offset;
776 bio->bi_iter.bi_sector = sector + rdev->data_offset;
777 bio_add_page(bio, page, size, 0);
778 submit_bio_wait(rw, bio);
780 ret = test_bit(BIO_UPTODATE, &bio->bi_flags);
784 EXPORT_SYMBOL_GPL(sync_page_io);
786 static int read_disk_sb(struct md_rdev * rdev, int size)
788 char b[BDEVNAME_SIZE];
789 if (!rdev->sb_page) {
797 if (!sync_page_io(rdev, 0, size, rdev->sb_page, READ, true))
803 printk(KERN_WARNING "md: disabled device %s, could not read superblock.\n",
804 bdevname(rdev->bdev,b));
808 static int uuid_equal(mdp_super_t *sb1, mdp_super_t *sb2)
810 return sb1->set_uuid0 == sb2->set_uuid0 &&
811 sb1->set_uuid1 == sb2->set_uuid1 &&
812 sb1->set_uuid2 == sb2->set_uuid2 &&
813 sb1->set_uuid3 == sb2->set_uuid3;
816 static int sb_equal(mdp_super_t *sb1, mdp_super_t *sb2)
819 mdp_super_t *tmp1, *tmp2;
821 tmp1 = kmalloc(sizeof(*tmp1),GFP_KERNEL);
822 tmp2 = kmalloc(sizeof(*tmp2),GFP_KERNEL);
824 if (!tmp1 || !tmp2) {
826 printk(KERN_INFO "md.c sb_equal(): failed to allocate memory!\n");
834 * nr_disks is not constant
839 ret = (memcmp(tmp1, tmp2, MD_SB_GENERIC_CONSTANT_WORDS * 4) == 0);
847 static u32 md_csum_fold(u32 csum)
849 csum = (csum & 0xffff) + (csum >> 16);
850 return (csum & 0xffff) + (csum >> 16);
853 static unsigned int calc_sb_csum(mdp_super_t * sb)
856 u32 *sb32 = (u32*)sb;
858 unsigned int disk_csum, csum;
860 disk_csum = sb->sb_csum;
863 for (i = 0; i < MD_SB_BYTES/4 ; i++)
865 csum = (newcsum & 0xffffffff) + (newcsum>>32);
869 /* This used to use csum_partial, which was wrong for several
870 * reasons including that different results are returned on
871 * different architectures. It isn't critical that we get exactly
872 * the same return value as before (we always csum_fold before
873 * testing, and that removes any differences). However as we
874 * know that csum_partial always returned a 16bit value on
875 * alphas, do a fold to maximise conformity to previous behaviour.
877 sb->sb_csum = md_csum_fold(disk_csum);
879 sb->sb_csum = disk_csum;
886 * Handle superblock details.
887 * We want to be able to handle multiple superblock formats
888 * so we have a common interface to them all, and an array of
889 * different handlers.
890 * We rely on user-space to write the initial superblock, and support
891 * reading and updating of superblocks.
892 * Interface methods are:
893 * int load_super(struct md_rdev *dev, struct md_rdev *refdev, int minor_version)
894 * loads and validates a superblock on dev.
895 * if refdev != NULL, compare superblocks on both devices
897 * 0 - dev has a superblock that is compatible with refdev
898 * 1 - dev has a superblock that is compatible and newer than refdev
899 * so dev should be used as the refdev in future
900 * -EINVAL superblock incompatible or invalid
901 * -othererror e.g. -EIO
903 * int validate_super(struct mddev *mddev, struct md_rdev *dev)
904 * Verify that dev is acceptable into mddev.
905 * The first time, mddev->raid_disks will be 0, and data from
906 * dev should be merged in. Subsequent calls check that dev
907 * is new enough. Return 0 or -EINVAL
909 * void sync_super(struct mddev *mddev, struct md_rdev *dev)
910 * Update the superblock for rdev with data in mddev
911 * This does not write to disc.
917 struct module *owner;
918 int (*load_super)(struct md_rdev *rdev,
919 struct md_rdev *refdev,
921 int (*validate_super)(struct mddev *mddev,
922 struct md_rdev *rdev);
923 void (*sync_super)(struct mddev *mddev,
924 struct md_rdev *rdev);
925 unsigned long long (*rdev_size_change)(struct md_rdev *rdev,
926 sector_t num_sectors);
927 int (*allow_new_offset)(struct md_rdev *rdev,
928 unsigned long long new_offset);
932 * Check that the given mddev has no bitmap.
934 * This function is called from the run method of all personalities that do not
935 * support bitmaps. It prints an error message and returns non-zero if mddev
936 * has a bitmap. Otherwise, it returns 0.
939 int md_check_no_bitmap(struct mddev *mddev)
941 if (!mddev->bitmap_info.file && !mddev->bitmap_info.offset)
943 printk(KERN_ERR "%s: bitmaps are not supported for %s\n",
944 mdname(mddev), mddev->pers->name);
947 EXPORT_SYMBOL(md_check_no_bitmap);
950 * load_super for 0.90.0
952 static int super_90_load(struct md_rdev *rdev, struct md_rdev *refdev, int minor_version)
954 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
959 * Calculate the position of the superblock (512byte sectors),
960 * it's at the end of the disk.
962 * It also happens to be a multiple of 4Kb.
964 rdev->sb_start = calc_dev_sboffset(rdev);
966 ret = read_disk_sb(rdev, MD_SB_BYTES);
971 bdevname(rdev->bdev, b);
972 sb = page_address(rdev->sb_page);
974 if (sb->md_magic != MD_SB_MAGIC) {
975 printk(KERN_ERR "md: invalid raid superblock magic on %s\n",
980 if (sb->major_version != 0 ||
981 sb->minor_version < 90 ||
982 sb->minor_version > 91) {
983 printk(KERN_WARNING "Bad version number %d.%d on %s\n",
984 sb->major_version, sb->minor_version,
989 if (sb->raid_disks <= 0)
992 if (md_csum_fold(calc_sb_csum(sb)) != md_csum_fold(sb->sb_csum)) {
993 printk(KERN_WARNING "md: invalid superblock checksum on %s\n",
998 rdev->preferred_minor = sb->md_minor;
999 rdev->data_offset = 0;
1000 rdev->new_data_offset = 0;
1001 rdev->sb_size = MD_SB_BYTES;
1002 rdev->badblocks.shift = -1;
1004 if (sb->level == LEVEL_MULTIPATH)
1007 rdev->desc_nr = sb->this_disk.number;
1013 mdp_super_t *refsb = page_address(refdev->sb_page);
1014 if (!uuid_equal(refsb, sb)) {
1015 printk(KERN_WARNING "md: %s has different UUID to %s\n",
1016 b, bdevname(refdev->bdev,b2));
1019 if (!sb_equal(refsb, sb)) {
1020 printk(KERN_WARNING "md: %s has same UUID"
1021 " but different superblock to %s\n",
1022 b, bdevname(refdev->bdev, b2));
1026 ev2 = md_event(refsb);
1032 rdev->sectors = rdev->sb_start;
1033 /* Limit to 4TB as metadata cannot record more than that.
1034 * (not needed for Linear and RAID0 as metadata doesn't
1037 if (rdev->sectors >= (2ULL << 32) && sb->level >= 1)
1038 rdev->sectors = (2ULL << 32) - 2;
1040 if (rdev->sectors < ((sector_t)sb->size) * 2 && sb->level >= 1)
1041 /* "this cannot possibly happen" ... */
1049 * validate_super for 0.90.0
1051 static int super_90_validate(struct mddev *mddev, struct md_rdev *rdev)
1054 mdp_super_t *sb = page_address(rdev->sb_page);
1055 __u64 ev1 = md_event(sb);
1057 rdev->raid_disk = -1;
1058 clear_bit(Faulty, &rdev->flags);
1059 clear_bit(In_sync, &rdev->flags);
1060 clear_bit(Bitmap_sync, &rdev->flags);
1061 clear_bit(WriteMostly, &rdev->flags);
1063 if (mddev->raid_disks == 0) {
1064 mddev->major_version = 0;
1065 mddev->minor_version = sb->minor_version;
1066 mddev->patch_version = sb->patch_version;
1067 mddev->external = 0;
1068 mddev->chunk_sectors = sb->chunk_size >> 9;
1069 mddev->ctime = sb->ctime;
1070 mddev->utime = sb->utime;
1071 mddev->level = sb->level;
1072 mddev->clevel[0] = 0;
1073 mddev->layout = sb->layout;
1074 mddev->raid_disks = sb->raid_disks;
1075 mddev->dev_sectors = ((sector_t)sb->size) * 2;
1076 mddev->events = ev1;
1077 mddev->bitmap_info.offset = 0;
1078 mddev->bitmap_info.space = 0;
1079 /* bitmap can use 60 K after the 4K superblocks */
1080 mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9;
1081 mddev->bitmap_info.default_space = 64*2 - (MD_SB_BYTES >> 9);
1082 mddev->reshape_backwards = 0;
1084 if (mddev->minor_version >= 91) {
1085 mddev->reshape_position = sb->reshape_position;
1086 mddev->delta_disks = sb->delta_disks;
1087 mddev->new_level = sb->new_level;
1088 mddev->new_layout = sb->new_layout;
1089 mddev->new_chunk_sectors = sb->new_chunk >> 9;
1090 if (mddev->delta_disks < 0)
1091 mddev->reshape_backwards = 1;
1093 mddev->reshape_position = MaxSector;
1094 mddev->delta_disks = 0;
1095 mddev->new_level = mddev->level;
1096 mddev->new_layout = mddev->layout;
1097 mddev->new_chunk_sectors = mddev->chunk_sectors;
1100 if (sb->state & (1<<MD_SB_CLEAN))
1101 mddev->recovery_cp = MaxSector;
1103 if (sb->events_hi == sb->cp_events_hi &&
1104 sb->events_lo == sb->cp_events_lo) {
1105 mddev->recovery_cp = sb->recovery_cp;
1107 mddev->recovery_cp = 0;
1110 memcpy(mddev->uuid+0, &sb->set_uuid0, 4);
1111 memcpy(mddev->uuid+4, &sb->set_uuid1, 4);
1112 memcpy(mddev->uuid+8, &sb->set_uuid2, 4);
1113 memcpy(mddev->uuid+12,&sb->set_uuid3, 4);
1115 mddev->max_disks = MD_SB_DISKS;
1117 if (sb->state & (1<<MD_SB_BITMAP_PRESENT) &&
1118 mddev->bitmap_info.file == NULL) {
1119 mddev->bitmap_info.offset =
1120 mddev->bitmap_info.default_offset;
1121 mddev->bitmap_info.space =
1122 mddev->bitmap_info.default_space;
1125 } else if (mddev->pers == NULL) {
1126 /* Insist on good event counter while assembling, except
1127 * for spares (which don't need an event count) */
1129 if (sb->disks[rdev->desc_nr].state & (
1130 (1<<MD_DISK_SYNC) | (1 << MD_DISK_ACTIVE)))
1131 if (ev1 < mddev->events)
1133 } else if (mddev->bitmap) {
1134 /* if adding to array with a bitmap, then we can accept an
1135 * older device ... but not too old.
1137 if (ev1 < mddev->bitmap->events_cleared)
1139 if (ev1 < mddev->events)
1140 set_bit(Bitmap_sync, &rdev->flags);
1142 if (ev1 < mddev->events)
1143 /* just a hot-add of a new device, leave raid_disk at -1 */
1147 if (mddev->level != LEVEL_MULTIPATH) {
1148 desc = sb->disks + rdev->desc_nr;
1150 if (desc->state & (1<<MD_DISK_FAULTY))
1151 set_bit(Faulty, &rdev->flags);
1152 else if (desc->state & (1<<MD_DISK_SYNC) /* &&
1153 desc->raid_disk < mddev->raid_disks */) {
1154 set_bit(In_sync, &rdev->flags);
1155 rdev->raid_disk = desc->raid_disk;
1156 rdev->saved_raid_disk = desc->raid_disk;
1157 } else if (desc->state & (1<<MD_DISK_ACTIVE)) {
1158 /* active but not in sync implies recovery up to
1159 * reshape position. We don't know exactly where
1160 * that is, so set to zero for now */
1161 if (mddev->minor_version >= 91) {
1162 rdev->recovery_offset = 0;
1163 rdev->raid_disk = desc->raid_disk;
1166 if (desc->state & (1<<MD_DISK_WRITEMOSTLY))
1167 set_bit(WriteMostly, &rdev->flags);
1168 } else /* MULTIPATH are always insync */
1169 set_bit(In_sync, &rdev->flags);
1174 * sync_super for 0.90.0
1176 static void super_90_sync(struct mddev *mddev, struct md_rdev *rdev)
1179 struct md_rdev *rdev2;
1180 int next_spare = mddev->raid_disks;
1183 /* make rdev->sb match mddev data..
1186 * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
1187 * 3/ any empty disks < next_spare become removed
1189 * disks[0] gets initialised to REMOVED because
1190 * we cannot be sure from other fields if it has
1191 * been initialised or not.
1194 int active=0, working=0,failed=0,spare=0,nr_disks=0;
1196 rdev->sb_size = MD_SB_BYTES;
1198 sb = page_address(rdev->sb_page);
1200 memset(sb, 0, sizeof(*sb));
1202 sb->md_magic = MD_SB_MAGIC;
1203 sb->major_version = mddev->major_version;
1204 sb->patch_version = mddev->patch_version;
1205 sb->gvalid_words = 0; /* ignored */
1206 memcpy(&sb->set_uuid0, mddev->uuid+0, 4);
1207 memcpy(&sb->set_uuid1, mddev->uuid+4, 4);
1208 memcpy(&sb->set_uuid2, mddev->uuid+8, 4);
1209 memcpy(&sb->set_uuid3, mddev->uuid+12,4);
1211 sb->ctime = mddev->ctime;
1212 sb->level = mddev->level;
1213 sb->size = mddev->dev_sectors / 2;
1214 sb->raid_disks = mddev->raid_disks;
1215 sb->md_minor = mddev->md_minor;
1216 sb->not_persistent = 0;
1217 sb->utime = mddev->utime;
1219 sb->events_hi = (mddev->events>>32);
1220 sb->events_lo = (u32)mddev->events;
1222 if (mddev->reshape_position == MaxSector)
1223 sb->minor_version = 90;
1225 sb->minor_version = 91;
1226 sb->reshape_position = mddev->reshape_position;
1227 sb->new_level = mddev->new_level;
1228 sb->delta_disks = mddev->delta_disks;
1229 sb->new_layout = mddev->new_layout;
1230 sb->new_chunk = mddev->new_chunk_sectors << 9;
1232 mddev->minor_version = sb->minor_version;
1235 sb->recovery_cp = mddev->recovery_cp;
1236 sb->cp_events_hi = (mddev->events>>32);
1237 sb->cp_events_lo = (u32)mddev->events;
1238 if (mddev->recovery_cp == MaxSector)
1239 sb->state = (1<< MD_SB_CLEAN);
1241 sb->recovery_cp = 0;
1243 sb->layout = mddev->layout;
1244 sb->chunk_size = mddev->chunk_sectors << 9;
1246 if (mddev->bitmap && mddev->bitmap_info.file == NULL)
1247 sb->state |= (1<<MD_SB_BITMAP_PRESENT);
1249 sb->disks[0].state = (1<<MD_DISK_REMOVED);
1250 rdev_for_each(rdev2, mddev) {
1253 int is_active = test_bit(In_sync, &rdev2->flags);
1255 if (rdev2->raid_disk >= 0 &&
1256 sb->minor_version >= 91)
1257 /* we have nowhere to store the recovery_offset,
1258 * but if it is not below the reshape_position,
1259 * we can piggy-back on that.
1262 if (rdev2->raid_disk < 0 ||
1263 test_bit(Faulty, &rdev2->flags))
1266 desc_nr = rdev2->raid_disk;
1268 desc_nr = next_spare++;
1269 rdev2->desc_nr = desc_nr;
1270 d = &sb->disks[rdev2->desc_nr];
1272 d->number = rdev2->desc_nr;
1273 d->major = MAJOR(rdev2->bdev->bd_dev);
1274 d->minor = MINOR(rdev2->bdev->bd_dev);
1276 d->raid_disk = rdev2->raid_disk;
1278 d->raid_disk = rdev2->desc_nr; /* compatibility */
1279 if (test_bit(Faulty, &rdev2->flags))
1280 d->state = (1<<MD_DISK_FAULTY);
1281 else if (is_active) {
1282 d->state = (1<<MD_DISK_ACTIVE);
1283 if (test_bit(In_sync, &rdev2->flags))
1284 d->state |= (1<<MD_DISK_SYNC);
1292 if (test_bit(WriteMostly, &rdev2->flags))
1293 d->state |= (1<<MD_DISK_WRITEMOSTLY);
1295 /* now set the "removed" and "faulty" bits on any missing devices */
1296 for (i=0 ; i < mddev->raid_disks ; i++) {
1297 mdp_disk_t *d = &sb->disks[i];
1298 if (d->state == 0 && d->number == 0) {
1301 d->state = (1<<MD_DISK_REMOVED);
1302 d->state |= (1<<MD_DISK_FAULTY);
1306 sb->nr_disks = nr_disks;
1307 sb->active_disks = active;
1308 sb->working_disks = working;
1309 sb->failed_disks = failed;
1310 sb->spare_disks = spare;
1312 sb->this_disk = sb->disks[rdev->desc_nr];
1313 sb->sb_csum = calc_sb_csum(sb);
1317 * rdev_size_change for 0.90.0
1319 static unsigned long long
1320 super_90_rdev_size_change(struct md_rdev *rdev, sector_t num_sectors)
1322 if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
1323 return 0; /* component must fit device */
1324 if (rdev->mddev->bitmap_info.offset)
1325 return 0; /* can't move bitmap */
1326 rdev->sb_start = calc_dev_sboffset(rdev);
1327 if (!num_sectors || num_sectors > rdev->sb_start)
1328 num_sectors = rdev->sb_start;
1329 /* Limit to 4TB as metadata cannot record more than that.
1330 * 4TB == 2^32 KB, or 2*2^32 sectors.
1332 if (num_sectors >= (2ULL << 32) && rdev->mddev->level >= 1)
1333 num_sectors = (2ULL << 32) - 2;
1334 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
1336 md_super_wait(rdev->mddev);
1341 super_90_allow_new_offset(struct md_rdev *rdev, unsigned long long new_offset)
1343 /* non-zero offset changes not possible with v0.90 */
1344 return new_offset == 0;
1348 * version 1 superblock
1351 static __le32 calc_sb_1_csum(struct mdp_superblock_1 * sb)
1355 unsigned long long newcsum;
1356 int size = 256 + le32_to_cpu(sb->max_dev)*2;
1357 __le32 *isuper = (__le32*)sb;
1359 disk_csum = sb->sb_csum;
1362 for (; size >= 4; size -= 4)
1363 newcsum += le32_to_cpu(*isuper++);
1366 newcsum += le16_to_cpu(*(__le16*) isuper);
1368 csum = (newcsum & 0xffffffff) + (newcsum >> 32);
1369 sb->sb_csum = disk_csum;
1370 return cpu_to_le32(csum);
1373 static int md_set_badblocks(struct badblocks *bb, sector_t s, int sectors,
1375 static int super_1_load(struct md_rdev *rdev, struct md_rdev *refdev, int minor_version)
1377 struct mdp_superblock_1 *sb;
1381 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
1385 * Calculate the position of the superblock in 512byte sectors.
1386 * It is always aligned to a 4K boundary and
1387 * depeding on minor_version, it can be:
1388 * 0: At least 8K, but less than 12K, from end of device
1389 * 1: At start of device
1390 * 2: 4K from start of device.
1392 switch(minor_version) {
1394 sb_start = i_size_read(rdev->bdev->bd_inode) >> 9;
1396 sb_start &= ~(sector_t)(4*2-1);
1407 rdev->sb_start = sb_start;
1409 /* superblock is rarely larger than 1K, but it can be larger,
1410 * and it is safe to read 4k, so we do that
1412 ret = read_disk_sb(rdev, 4096);
1413 if (ret) return ret;
1416 sb = page_address(rdev->sb_page);
1418 if (sb->magic != cpu_to_le32(MD_SB_MAGIC) ||
1419 sb->major_version != cpu_to_le32(1) ||
1420 le32_to_cpu(sb->max_dev) > (4096-256)/2 ||
1421 le64_to_cpu(sb->super_offset) != rdev->sb_start ||
1422 (le32_to_cpu(sb->feature_map) & ~MD_FEATURE_ALL) != 0)
1425 if (calc_sb_1_csum(sb) != sb->sb_csum) {
1426 printk("md: invalid superblock checksum on %s\n",
1427 bdevname(rdev->bdev,b));
1430 if (le64_to_cpu(sb->data_size) < 10) {
1431 printk("md: data_size too small on %s\n",
1432 bdevname(rdev->bdev,b));
1437 memcmp(sb->pad3, sb->pad3+1, sizeof(sb->pad3) - sizeof(sb->pad3[1])))
1438 /* Some padding is non-zero, might be a new feature */
1441 rdev->preferred_minor = 0xffff;
1442 rdev->data_offset = le64_to_cpu(sb->data_offset);
1443 rdev->new_data_offset = rdev->data_offset;
1444 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE) &&
1445 (le32_to_cpu(sb->feature_map) & MD_FEATURE_NEW_OFFSET))
1446 rdev->new_data_offset += (s32)le32_to_cpu(sb->new_offset);
1447 atomic_set(&rdev->corrected_errors, le32_to_cpu(sb->cnt_corrected_read));
1449 rdev->sb_size = le32_to_cpu(sb->max_dev) * 2 + 256;
1450 bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
1451 if (rdev->sb_size & bmask)
1452 rdev->sb_size = (rdev->sb_size | bmask) + 1;
1455 && rdev->data_offset < sb_start + (rdev->sb_size/512))
1458 && rdev->new_data_offset < sb_start + (rdev->sb_size/512))
1461 if (sb->level == cpu_to_le32(LEVEL_MULTIPATH))
1464 rdev->desc_nr = le32_to_cpu(sb->dev_number);
1466 if (!rdev->bb_page) {
1467 rdev->bb_page = alloc_page(GFP_KERNEL);
1471 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BAD_BLOCKS) &&
1472 rdev->badblocks.count == 0) {
1473 /* need to load the bad block list.
1474 * Currently we limit it to one page.
1480 int sectors = le16_to_cpu(sb->bblog_size);
1481 if (sectors > (PAGE_SIZE / 512))
1483 offset = le32_to_cpu(sb->bblog_offset);
1486 bb_sector = (long long)offset;
1487 if (!sync_page_io(rdev, bb_sector, sectors << 9,
1488 rdev->bb_page, READ, true))
1490 bbp = (u64 *)page_address(rdev->bb_page);
1491 rdev->badblocks.shift = sb->bblog_shift;
1492 for (i = 0 ; i < (sectors << (9-3)) ; i++, bbp++) {
1493 u64 bb = le64_to_cpu(*bbp);
1494 int count = bb & (0x3ff);
1495 u64 sector = bb >> 10;
1496 sector <<= sb->bblog_shift;
1497 count <<= sb->bblog_shift;
1500 if (md_set_badblocks(&rdev->badblocks,
1501 sector, count, 1) == 0)
1504 } else if (sb->bblog_offset != 0)
1505 rdev->badblocks.shift = 0;
1511 struct mdp_superblock_1 *refsb = page_address(refdev->sb_page);
1513 if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 ||
1514 sb->level != refsb->level ||
1515 sb->layout != refsb->layout ||
1516 sb->chunksize != refsb->chunksize) {
1517 printk(KERN_WARNING "md: %s has strangely different"
1518 " superblock to %s\n",
1519 bdevname(rdev->bdev,b),
1520 bdevname(refdev->bdev,b2));
1523 ev1 = le64_to_cpu(sb->events);
1524 ev2 = le64_to_cpu(refsb->events);
1531 if (minor_version) {
1532 sectors = (i_size_read(rdev->bdev->bd_inode) >> 9);
1533 sectors -= rdev->data_offset;
1535 sectors = rdev->sb_start;
1536 if (sectors < le64_to_cpu(sb->data_size))
1538 rdev->sectors = le64_to_cpu(sb->data_size);
1542 static int super_1_validate(struct mddev *mddev, struct md_rdev *rdev)
1544 struct mdp_superblock_1 *sb = page_address(rdev->sb_page);
1545 __u64 ev1 = le64_to_cpu(sb->events);
1547 rdev->raid_disk = -1;
1548 clear_bit(Faulty, &rdev->flags);
1549 clear_bit(In_sync, &rdev->flags);
1550 clear_bit(Bitmap_sync, &rdev->flags);
1551 clear_bit(WriteMostly, &rdev->flags);
1553 if (mddev->raid_disks == 0) {
1554 mddev->major_version = 1;
1555 mddev->patch_version = 0;
1556 mddev->external = 0;
1557 mddev->chunk_sectors = le32_to_cpu(sb->chunksize);
1558 mddev->ctime = le64_to_cpu(sb->ctime) & ((1ULL << 32)-1);
1559 mddev->utime = le64_to_cpu(sb->utime) & ((1ULL << 32)-1);
1560 mddev->level = le32_to_cpu(sb->level);
1561 mddev->clevel[0] = 0;
1562 mddev->layout = le32_to_cpu(sb->layout);
1563 mddev->raid_disks = le32_to_cpu(sb->raid_disks);
1564 mddev->dev_sectors = le64_to_cpu(sb->size);
1565 mddev->events = ev1;
1566 mddev->bitmap_info.offset = 0;
1567 mddev->bitmap_info.space = 0;
1568 /* Default location for bitmap is 1K after superblock
1569 * using 3K - total of 4K
1571 mddev->bitmap_info.default_offset = 1024 >> 9;
1572 mddev->bitmap_info.default_space = (4096-1024) >> 9;
1573 mddev->reshape_backwards = 0;
1575 mddev->recovery_cp = le64_to_cpu(sb->resync_offset);
1576 memcpy(mddev->uuid, sb->set_uuid, 16);
1578 mddev->max_disks = (4096-256)/2;
1580 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) &&
1581 mddev->bitmap_info.file == NULL) {
1582 mddev->bitmap_info.offset =
1583 (__s32)le32_to_cpu(sb->bitmap_offset);
1584 /* Metadata doesn't record how much space is available.
1585 * For 1.0, we assume we can use up to the superblock
1586 * if before, else to 4K beyond superblock.
1587 * For others, assume no change is possible.
1589 if (mddev->minor_version > 0)
1590 mddev->bitmap_info.space = 0;
1591 else if (mddev->bitmap_info.offset > 0)
1592 mddev->bitmap_info.space =
1593 8 - mddev->bitmap_info.offset;
1595 mddev->bitmap_info.space =
1596 -mddev->bitmap_info.offset;
1599 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) {
1600 mddev->reshape_position = le64_to_cpu(sb->reshape_position);
1601 mddev->delta_disks = le32_to_cpu(sb->delta_disks);
1602 mddev->new_level = le32_to_cpu(sb->new_level);
1603 mddev->new_layout = le32_to_cpu(sb->new_layout);
1604 mddev->new_chunk_sectors = le32_to_cpu(sb->new_chunk);
1605 if (mddev->delta_disks < 0 ||
1606 (mddev->delta_disks == 0 &&
1607 (le32_to_cpu(sb->feature_map)
1608 & MD_FEATURE_RESHAPE_BACKWARDS)))
1609 mddev->reshape_backwards = 1;
1611 mddev->reshape_position = MaxSector;
1612 mddev->delta_disks = 0;
1613 mddev->new_level = mddev->level;
1614 mddev->new_layout = mddev->layout;
1615 mddev->new_chunk_sectors = mddev->chunk_sectors;
1618 } else if (mddev->pers == NULL) {
1619 /* Insist of good event counter while assembling, except for
1620 * spares (which don't need an event count) */
1622 if (rdev->desc_nr >= 0 &&
1623 rdev->desc_nr < le32_to_cpu(sb->max_dev) &&
1624 le16_to_cpu(sb->dev_roles[rdev->desc_nr]) < 0xfffe)
1625 if (ev1 < mddev->events)
1627 } else if (mddev->bitmap) {
1628 /* If adding to array with a bitmap, then we can accept an
1629 * older device, but not too old.
1631 if (ev1 < mddev->bitmap->events_cleared)
1633 if (ev1 < mddev->events)
1634 set_bit(Bitmap_sync, &rdev->flags);
1636 if (ev1 < mddev->events)
1637 /* just a hot-add of a new device, leave raid_disk at -1 */
1640 if (mddev->level != LEVEL_MULTIPATH) {
1642 if (rdev->desc_nr < 0 ||
1643 rdev->desc_nr >= le32_to_cpu(sb->max_dev)) {
1647 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
1649 case 0xffff: /* spare */
1651 case 0xfffe: /* faulty */
1652 set_bit(Faulty, &rdev->flags);
1655 rdev->saved_raid_disk = role;
1656 if ((le32_to_cpu(sb->feature_map) &
1657 MD_FEATURE_RECOVERY_OFFSET)) {
1658 rdev->recovery_offset = le64_to_cpu(sb->recovery_offset);
1659 if (!(le32_to_cpu(sb->feature_map) &
1660 MD_FEATURE_RECOVERY_BITMAP))
1661 rdev->saved_raid_disk = -1;
1663 set_bit(In_sync, &rdev->flags);
1664 rdev->raid_disk = role;
1667 if (sb->devflags & WriteMostly1)
1668 set_bit(WriteMostly, &rdev->flags);
1669 if (le32_to_cpu(sb->feature_map) & MD_FEATURE_REPLACEMENT)
1670 set_bit(Replacement, &rdev->flags);
1671 } else /* MULTIPATH are always insync */
1672 set_bit(In_sync, &rdev->flags);
1677 static void super_1_sync(struct mddev *mddev, struct md_rdev *rdev)
1679 struct mdp_superblock_1 *sb;
1680 struct md_rdev *rdev2;
1682 /* make rdev->sb match mddev and rdev data. */
1684 sb = page_address(rdev->sb_page);
1686 sb->feature_map = 0;
1688 sb->recovery_offset = cpu_to_le64(0);
1689 memset(sb->pad3, 0, sizeof(sb->pad3));
1691 sb->utime = cpu_to_le64((__u64)mddev->utime);
1692 sb->events = cpu_to_le64(mddev->events);
1694 sb->resync_offset = cpu_to_le64(mddev->recovery_cp);
1696 sb->resync_offset = cpu_to_le64(0);
1698 sb->cnt_corrected_read = cpu_to_le32(atomic_read(&rdev->corrected_errors));
1700 sb->raid_disks = cpu_to_le32(mddev->raid_disks);
1701 sb->size = cpu_to_le64(mddev->dev_sectors);
1702 sb->chunksize = cpu_to_le32(mddev->chunk_sectors);
1703 sb->level = cpu_to_le32(mddev->level);
1704 sb->layout = cpu_to_le32(mddev->layout);
1706 if (test_bit(WriteMostly, &rdev->flags))
1707 sb->devflags |= WriteMostly1;
1709 sb->devflags &= ~WriteMostly1;
1710 sb->data_offset = cpu_to_le64(rdev->data_offset);
1711 sb->data_size = cpu_to_le64(rdev->sectors);
1713 if (mddev->bitmap && mddev->bitmap_info.file == NULL) {
1714 sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_info.offset);
1715 sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET);
1718 if (rdev->raid_disk >= 0 &&
1719 !test_bit(In_sync, &rdev->flags)) {
1721 cpu_to_le32(MD_FEATURE_RECOVERY_OFFSET);
1722 sb->recovery_offset =
1723 cpu_to_le64(rdev->recovery_offset);
1724 if (rdev->saved_raid_disk >= 0 && mddev->bitmap)
1726 cpu_to_le32(MD_FEATURE_RECOVERY_BITMAP);
1728 if (test_bit(Replacement, &rdev->flags))
1730 cpu_to_le32(MD_FEATURE_REPLACEMENT);
1732 if (mddev->reshape_position != MaxSector) {
1733 sb->feature_map |= cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE);
1734 sb->reshape_position = cpu_to_le64(mddev->reshape_position);
1735 sb->new_layout = cpu_to_le32(mddev->new_layout);
1736 sb->delta_disks = cpu_to_le32(mddev->delta_disks);
1737 sb->new_level = cpu_to_le32(mddev->new_level);
1738 sb->new_chunk = cpu_to_le32(mddev->new_chunk_sectors);
1739 if (mddev->delta_disks == 0 &&
1740 mddev->reshape_backwards)
1742 |= cpu_to_le32(MD_FEATURE_RESHAPE_BACKWARDS);
1743 if (rdev->new_data_offset != rdev->data_offset) {
1745 |= cpu_to_le32(MD_FEATURE_NEW_OFFSET);
1746 sb->new_offset = cpu_to_le32((__u32)(rdev->new_data_offset
1747 - rdev->data_offset));
1751 if (rdev->badblocks.count == 0)
1752 /* Nothing to do for bad blocks*/ ;
1753 else if (sb->bblog_offset == 0)
1754 /* Cannot record bad blocks on this device */
1755 md_error(mddev, rdev);
1757 struct badblocks *bb = &rdev->badblocks;
1758 u64 *bbp = (u64 *)page_address(rdev->bb_page);
1760 sb->feature_map |= cpu_to_le32(MD_FEATURE_BAD_BLOCKS);
1765 seq = read_seqbegin(&bb->lock);
1767 memset(bbp, 0xff, PAGE_SIZE);
1769 for (i = 0 ; i < bb->count ; i++) {
1770 u64 internal_bb = p[i];
1771 u64 store_bb = ((BB_OFFSET(internal_bb) << 10)
1772 | BB_LEN(internal_bb));
1773 bbp[i] = cpu_to_le64(store_bb);
1776 if (read_seqretry(&bb->lock, seq))
1779 bb->sector = (rdev->sb_start +
1780 (int)le32_to_cpu(sb->bblog_offset));
1781 bb->size = le16_to_cpu(sb->bblog_size);
1786 rdev_for_each(rdev2, mddev)
1787 if (rdev2->desc_nr+1 > max_dev)
1788 max_dev = rdev2->desc_nr+1;
1790 if (max_dev > le32_to_cpu(sb->max_dev)) {
1792 sb->max_dev = cpu_to_le32(max_dev);
1793 rdev->sb_size = max_dev * 2 + 256;
1794 bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
1795 if (rdev->sb_size & bmask)
1796 rdev->sb_size = (rdev->sb_size | bmask) + 1;
1798 max_dev = le32_to_cpu(sb->max_dev);
1800 for (i=0; i<max_dev;i++)
1801 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1803 rdev_for_each(rdev2, mddev) {
1805 if (test_bit(Faulty, &rdev2->flags))
1806 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1807 else if (test_bit(In_sync, &rdev2->flags))
1808 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1809 else if (rdev2->raid_disk >= 0)
1810 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1812 sb->dev_roles[i] = cpu_to_le16(0xffff);
1815 sb->sb_csum = calc_sb_1_csum(sb);
1818 static unsigned long long
1819 super_1_rdev_size_change(struct md_rdev *rdev, sector_t num_sectors)
1821 struct mdp_superblock_1 *sb;
1822 sector_t max_sectors;
1823 if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
1824 return 0; /* component must fit device */
1825 if (rdev->data_offset != rdev->new_data_offset)
1826 return 0; /* too confusing */
1827 if (rdev->sb_start < rdev->data_offset) {
1828 /* minor versions 1 and 2; superblock before data */
1829 max_sectors = i_size_read(rdev->bdev->bd_inode) >> 9;
1830 max_sectors -= rdev->data_offset;
1831 if (!num_sectors || num_sectors > max_sectors)
1832 num_sectors = max_sectors;
1833 } else if (rdev->mddev->bitmap_info.offset) {
1834 /* minor version 0 with bitmap we can't move */
1837 /* minor version 0; superblock after data */
1839 sb_start = (i_size_read(rdev->bdev->bd_inode) >> 9) - 8*2;
1840 sb_start &= ~(sector_t)(4*2 - 1);
1841 max_sectors = rdev->sectors + sb_start - rdev->sb_start;
1842 if (!num_sectors || num_sectors > max_sectors)
1843 num_sectors = max_sectors;
1844 rdev->sb_start = sb_start;
1846 sb = page_address(rdev->sb_page);
1847 sb->data_size = cpu_to_le64(num_sectors);
1848 sb->super_offset = rdev->sb_start;
1849 sb->sb_csum = calc_sb_1_csum(sb);
1850 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
1852 md_super_wait(rdev->mddev);
1858 super_1_allow_new_offset(struct md_rdev *rdev,
1859 unsigned long long new_offset)
1861 /* All necessary checks on new >= old have been done */
1862 struct bitmap *bitmap;
1863 if (new_offset >= rdev->data_offset)
1866 /* with 1.0 metadata, there is no metadata to tread on
1867 * so we can always move back */
1868 if (rdev->mddev->minor_version == 0)
1871 /* otherwise we must be sure not to step on
1872 * any metadata, so stay:
1873 * 36K beyond start of superblock
1874 * beyond end of badblocks
1875 * beyond write-intent bitmap
1877 if (rdev->sb_start + (32+4)*2 > new_offset)
1879 bitmap = rdev->mddev->bitmap;
1880 if (bitmap && !rdev->mddev->bitmap_info.file &&
1881 rdev->sb_start + rdev->mddev->bitmap_info.offset +
1882 bitmap->storage.file_pages * (PAGE_SIZE>>9) > new_offset)
1884 if (rdev->badblocks.sector + rdev->badblocks.size > new_offset)
1890 static struct super_type super_types[] = {
1893 .owner = THIS_MODULE,
1894 .load_super = super_90_load,
1895 .validate_super = super_90_validate,
1896 .sync_super = super_90_sync,
1897 .rdev_size_change = super_90_rdev_size_change,
1898 .allow_new_offset = super_90_allow_new_offset,
1902 .owner = THIS_MODULE,
1903 .load_super = super_1_load,
1904 .validate_super = super_1_validate,
1905 .sync_super = super_1_sync,
1906 .rdev_size_change = super_1_rdev_size_change,
1907 .allow_new_offset = super_1_allow_new_offset,
1911 static void sync_super(struct mddev *mddev, struct md_rdev *rdev)
1913 if (mddev->sync_super) {
1914 mddev->sync_super(mddev, rdev);
1918 BUG_ON(mddev->major_version >= ARRAY_SIZE(super_types));
1920 super_types[mddev->major_version].sync_super(mddev, rdev);
1923 static int match_mddev_units(struct mddev *mddev1, struct mddev *mddev2)
1925 struct md_rdev *rdev, *rdev2;
1928 rdev_for_each_rcu(rdev, mddev1)
1929 rdev_for_each_rcu(rdev2, mddev2)
1930 if (rdev->bdev->bd_contains ==
1931 rdev2->bdev->bd_contains) {
1939 static LIST_HEAD(pending_raid_disks);
1942 * Try to register data integrity profile for an mddev
1944 * This is called when an array is started and after a disk has been kicked
1945 * from the array. It only succeeds if all working and active component devices
1946 * are integrity capable with matching profiles.
1948 int md_integrity_register(struct mddev *mddev)
1950 struct md_rdev *rdev, *reference = NULL;
1952 if (list_empty(&mddev->disks))
1953 return 0; /* nothing to do */
1954 if (!mddev->gendisk || blk_get_integrity(mddev->gendisk))
1955 return 0; /* shouldn't register, or already is */
1956 rdev_for_each(rdev, mddev) {
1957 /* skip spares and non-functional disks */
1958 if (test_bit(Faulty, &rdev->flags))
1960 if (rdev->raid_disk < 0)
1963 /* Use the first rdev as the reference */
1967 /* does this rdev's profile match the reference profile? */
1968 if (blk_integrity_compare(reference->bdev->bd_disk,
1969 rdev->bdev->bd_disk) < 0)
1972 if (!reference || !bdev_get_integrity(reference->bdev))
1975 * All component devices are integrity capable and have matching
1976 * profiles, register the common profile for the md device.
1978 if (blk_integrity_register(mddev->gendisk,
1979 bdev_get_integrity(reference->bdev)) != 0) {
1980 printk(KERN_ERR "md: failed to register integrity for %s\n",
1984 printk(KERN_NOTICE "md: data integrity enabled on %s\n", mdname(mddev));
1985 if (bioset_integrity_create(mddev->bio_set, BIO_POOL_SIZE)) {
1986 printk(KERN_ERR "md: failed to create integrity pool for %s\n",
1992 EXPORT_SYMBOL(md_integrity_register);
1994 /* Disable data integrity if non-capable/non-matching disk is being added */
1995 void md_integrity_add_rdev(struct md_rdev *rdev, struct mddev *mddev)
1997 struct blk_integrity *bi_rdev;
1998 struct blk_integrity *bi_mddev;
2000 if (!mddev->gendisk)
2003 bi_rdev = bdev_get_integrity(rdev->bdev);
2004 bi_mddev = blk_get_integrity(mddev->gendisk);
2006 if (!bi_mddev) /* nothing to do */
2008 if (rdev->raid_disk < 0) /* skip spares */
2010 if (bi_rdev && blk_integrity_compare(mddev->gendisk,
2011 rdev->bdev->bd_disk) >= 0)
2013 printk(KERN_NOTICE "disabling data integrity on %s\n", mdname(mddev));
2014 blk_integrity_unregister(mddev->gendisk);
2016 EXPORT_SYMBOL(md_integrity_add_rdev);
2018 static int bind_rdev_to_array(struct md_rdev * rdev, struct mddev * mddev)
2020 char b[BDEVNAME_SIZE];
2030 /* prevent duplicates */
2031 if (find_rdev(mddev, rdev->bdev->bd_dev))
2034 /* make sure rdev->sectors exceeds mddev->dev_sectors */
2035 if (rdev->sectors && (mddev->dev_sectors == 0 ||
2036 rdev->sectors < mddev->dev_sectors)) {
2038 /* Cannot change size, so fail
2039 * If mddev->level <= 0, then we don't care
2040 * about aligning sizes (e.g. linear)
2042 if (mddev->level > 0)
2045 mddev->dev_sectors = rdev->sectors;
2048 /* Verify rdev->desc_nr is unique.
2049 * If it is -1, assign a free number, else
2050 * check number is not in use
2053 if (rdev->desc_nr < 0) {
2056 choice = mddev->raid_disks;
2057 while (find_rdev_nr_rcu(mddev, choice))
2059 rdev->desc_nr = choice;
2061 if (find_rdev_nr_rcu(mddev, rdev->desc_nr)) {
2067 if (mddev->max_disks && rdev->desc_nr >= mddev->max_disks) {
2068 printk(KERN_WARNING "md: %s: array is limited to %d devices\n",
2069 mdname(mddev), mddev->max_disks);
2072 bdevname(rdev->bdev,b);
2073 while ( (s=strchr(b, '/')) != NULL)
2076 rdev->mddev = mddev;
2077 printk(KERN_INFO "md: bind<%s>\n", b);
2079 if ((err = kobject_add(&rdev->kobj, &mddev->kobj, "dev-%s", b)))
2082 ko = &part_to_dev(rdev->bdev->bd_part)->kobj;
2083 if (sysfs_create_link(&rdev->kobj, ko, "block"))
2084 /* failure here is OK */;
2085 rdev->sysfs_state = sysfs_get_dirent_safe(rdev->kobj.sd, "state");
2087 list_add_rcu(&rdev->same_set, &mddev->disks);
2088 bd_link_disk_holder(rdev->bdev, mddev->gendisk);
2090 /* May as well allow recovery to be retried once */
2091 mddev->recovery_disabled++;
2096 printk(KERN_WARNING "md: failed to register dev-%s for %s\n",
2101 static void md_delayed_delete(struct work_struct *ws)
2103 struct md_rdev *rdev = container_of(ws, struct md_rdev, del_work);
2104 kobject_del(&rdev->kobj);
2105 kobject_put(&rdev->kobj);
2108 static void unbind_rdev_from_array(struct md_rdev * rdev)
2110 char b[BDEVNAME_SIZE];
2115 bd_unlink_disk_holder(rdev->bdev, rdev->mddev->gendisk);
2116 list_del_rcu(&rdev->same_set);
2117 printk(KERN_INFO "md: unbind<%s>\n", bdevname(rdev->bdev,b));
2119 sysfs_remove_link(&rdev->kobj, "block");
2120 sysfs_put(rdev->sysfs_state);
2121 rdev->sysfs_state = NULL;
2122 rdev->badblocks.count = 0;
2123 /* We need to delay this, otherwise we can deadlock when
2124 * writing to 'remove' to "dev/state". We also need
2125 * to delay it due to rcu usage.
2128 INIT_WORK(&rdev->del_work, md_delayed_delete);
2129 kobject_get(&rdev->kobj);
2130 queue_work(md_misc_wq, &rdev->del_work);
2134 * prevent the device from being mounted, repartitioned or
2135 * otherwise reused by a RAID array (or any other kernel
2136 * subsystem), by bd_claiming the device.
2138 static int lock_rdev(struct md_rdev *rdev, dev_t dev, int shared)
2141 struct block_device *bdev;
2142 char b[BDEVNAME_SIZE];
2144 bdev = blkdev_get_by_dev(dev, FMODE_READ|FMODE_WRITE|FMODE_EXCL,
2145 shared ? (struct md_rdev *)lock_rdev : rdev);
2147 printk(KERN_ERR "md: could not open %s.\n",
2148 __bdevname(dev, b));
2149 return PTR_ERR(bdev);
2155 static void unlock_rdev(struct md_rdev *rdev)
2157 struct block_device *bdev = rdev->bdev;
2161 blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
2164 void md_autodetect_dev(dev_t dev);
2166 static void export_rdev(struct md_rdev * rdev)
2168 char b[BDEVNAME_SIZE];
2169 printk(KERN_INFO "md: export_rdev(%s)\n",
2170 bdevname(rdev->bdev,b));
2173 md_rdev_clear(rdev);
2175 if (test_bit(AutoDetected, &rdev->flags))
2176 md_autodetect_dev(rdev->bdev->bd_dev);
2179 kobject_put(&rdev->kobj);
2182 static void kick_rdev_from_array(struct md_rdev * rdev)
2184 unbind_rdev_from_array(rdev);
2188 static void export_array(struct mddev *mddev)
2190 struct md_rdev *rdev;
2192 while (!list_empty(&mddev->disks)) {
2193 rdev = list_first_entry(&mddev->disks, struct md_rdev,
2195 kick_rdev_from_array(rdev);
2197 mddev->raid_disks = 0;
2198 mddev->major_version = 0;
2201 static void print_desc(mdp_disk_t *desc)
2203 printk(" DISK<N:%d,(%d,%d),R:%d,S:%d>\n", desc->number,
2204 desc->major,desc->minor,desc->raid_disk,desc->state);
2207 static void print_sb_90(mdp_super_t *sb)
2212 "md: SB: (V:%d.%d.%d) ID:<%08x.%08x.%08x.%08x> CT:%08x\n",
2213 sb->major_version, sb->minor_version, sb->patch_version,
2214 sb->set_uuid0, sb->set_uuid1, sb->set_uuid2, sb->set_uuid3,
2216 printk(KERN_INFO "md: L%d S%08d ND:%d RD:%d md%d LO:%d CS:%d\n",
2217 sb->level, sb->size, sb->nr_disks, sb->raid_disks,
2218 sb->md_minor, sb->layout, sb->chunk_size);
2219 printk(KERN_INFO "md: UT:%08x ST:%d AD:%d WD:%d"
2220 " FD:%d SD:%d CSUM:%08x E:%08lx\n",
2221 sb->utime, sb->state, sb->active_disks, sb->working_disks,
2222 sb->failed_disks, sb->spare_disks,
2223 sb->sb_csum, (unsigned long)sb->events_lo);
2226 for (i = 0; i < MD_SB_DISKS; i++) {
2229 desc = sb->disks + i;
2230 if (desc->number || desc->major || desc->minor ||
2231 desc->raid_disk || (desc->state && (desc->state != 4))) {
2232 printk(" D %2d: ", i);
2236 printk(KERN_INFO "md: THIS: ");
2237 print_desc(&sb->this_disk);
2240 static void print_sb_1(struct mdp_superblock_1 *sb)
2244 uuid = sb->set_uuid;
2246 "md: SB: (V:%u) (F:0x%08x) Array-ID:<%pU>\n"
2247 "md: Name: \"%s\" CT:%llu\n",
2248 le32_to_cpu(sb->major_version),
2249 le32_to_cpu(sb->feature_map),
2252 (unsigned long long)le64_to_cpu(sb->ctime)
2253 & MD_SUPERBLOCK_1_TIME_SEC_MASK);
2255 uuid = sb->device_uuid;
2257 "md: L%u SZ%llu RD:%u LO:%u CS:%u DO:%llu DS:%llu SO:%llu"
2259 "md: Dev:%08x UUID: %pU\n"
2260 "md: (F:0x%08x) UT:%llu Events:%llu ResyncOffset:%llu CSUM:0x%08x\n"
2261 "md: (MaxDev:%u) \n",
2262 le32_to_cpu(sb->level),
2263 (unsigned long long)le64_to_cpu(sb->size),
2264 le32_to_cpu(sb->raid_disks),
2265 le32_to_cpu(sb->layout),
2266 le32_to_cpu(sb->chunksize),
2267 (unsigned long long)le64_to_cpu(sb->data_offset),
2268 (unsigned long long)le64_to_cpu(sb->data_size),
2269 (unsigned long long)le64_to_cpu(sb->super_offset),
2270 (unsigned long long)le64_to_cpu(sb->recovery_offset),
2271 le32_to_cpu(sb->dev_number),
2274 (unsigned long long)le64_to_cpu(sb->utime) & MD_SUPERBLOCK_1_TIME_SEC_MASK,
2275 (unsigned long long)le64_to_cpu(sb->events),
2276 (unsigned long long)le64_to_cpu(sb->resync_offset),
2277 le32_to_cpu(sb->sb_csum),
2278 le32_to_cpu(sb->max_dev)
2282 static void print_rdev(struct md_rdev *rdev, int major_version)
2284 char b[BDEVNAME_SIZE];
2285 printk(KERN_INFO "md: rdev %s, Sect:%08llu F:%d S:%d DN:%u\n",
2286 bdevname(rdev->bdev, b), (unsigned long long)rdev->sectors,
2287 test_bit(Faulty, &rdev->flags), test_bit(In_sync, &rdev->flags),
2289 if (rdev->sb_loaded) {
2290 printk(KERN_INFO "md: rdev superblock (MJ:%d):\n", major_version);
2291 switch (major_version) {
2293 print_sb_90(page_address(rdev->sb_page));
2296 print_sb_1(page_address(rdev->sb_page));
2300 printk(KERN_INFO "md: no rdev superblock!\n");
2303 static void md_print_devices(void)
2305 struct list_head *tmp;
2306 struct md_rdev *rdev;
2307 struct mddev *mddev;
2308 char b[BDEVNAME_SIZE];
2311 printk("md: **********************************\n");
2312 printk("md: * <COMPLETE RAID STATE PRINTOUT> *\n");
2313 printk("md: **********************************\n");
2314 for_each_mddev(mddev, tmp) {
2317 bitmap_print_sb(mddev->bitmap);
2319 printk("%s: ", mdname(mddev));
2320 rdev_for_each(rdev, mddev)
2321 printk("<%s>", bdevname(rdev->bdev,b));
2324 rdev_for_each(rdev, mddev)
2325 print_rdev(rdev, mddev->major_version);
2327 printk("md: **********************************\n");
2332 static void sync_sbs(struct mddev * mddev, int nospares)
2334 /* Update each superblock (in-memory image), but
2335 * if we are allowed to, skip spares which already
2336 * have the right event counter, or have one earlier
2337 * (which would mean they aren't being marked as dirty
2338 * with the rest of the array)
2340 struct md_rdev *rdev;
2341 rdev_for_each(rdev, mddev) {
2342 if (rdev->sb_events == mddev->events ||
2344 rdev->raid_disk < 0 &&
2345 rdev->sb_events+1 == mddev->events)) {
2346 /* Don't update this superblock */
2347 rdev->sb_loaded = 2;
2349 sync_super(mddev, rdev);
2350 rdev->sb_loaded = 1;
2355 static void md_update_sb(struct mddev * mddev, int force_change)
2357 struct md_rdev *rdev;
2360 int any_badblocks_changed = 0;
2364 set_bit(MD_CHANGE_DEVS, &mddev->flags);
2368 /* First make sure individual recovery_offsets are correct */
2369 rdev_for_each(rdev, mddev) {
2370 if (rdev->raid_disk >= 0 &&
2371 mddev->delta_disks >= 0 &&
2372 !test_bit(In_sync, &rdev->flags) &&
2373 mddev->curr_resync_completed > rdev->recovery_offset)
2374 rdev->recovery_offset = mddev->curr_resync_completed;
2377 if (!mddev->persistent) {
2378 clear_bit(MD_CHANGE_CLEAN, &mddev->flags);
2379 clear_bit(MD_CHANGE_DEVS, &mddev->flags);
2380 if (!mddev->external) {
2381 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
2382 rdev_for_each(rdev, mddev) {
2383 if (rdev->badblocks.changed) {
2384 rdev->badblocks.changed = 0;
2385 md_ack_all_badblocks(&rdev->badblocks);
2386 md_error(mddev, rdev);
2388 clear_bit(Blocked, &rdev->flags);
2389 clear_bit(BlockedBadBlocks, &rdev->flags);
2390 wake_up(&rdev->blocked_wait);
2393 wake_up(&mddev->sb_wait);
2397 spin_lock_irq(&mddev->write_lock);
2399 mddev->utime = get_seconds();
2401 if (test_and_clear_bit(MD_CHANGE_DEVS, &mddev->flags))
2403 if (test_and_clear_bit(MD_CHANGE_CLEAN, &mddev->flags))
2404 /* just a clean<-> dirty transition, possibly leave spares alone,
2405 * though if events isn't the right even/odd, we will have to do
2411 if (mddev->degraded)
2412 /* If the array is degraded, then skipping spares is both
2413 * dangerous and fairly pointless.
2414 * Dangerous because a device that was removed from the array
2415 * might have a event_count that still looks up-to-date,
2416 * so it can be re-added without a resync.
2417 * Pointless because if there are any spares to skip,
2418 * then a recovery will happen and soon that array won't
2419 * be degraded any more and the spare can go back to sleep then.
2423 sync_req = mddev->in_sync;
2425 /* If this is just a dirty<->clean transition, and the array is clean
2426 * and 'events' is odd, we can roll back to the previous clean state */
2428 && (mddev->in_sync && mddev->recovery_cp == MaxSector)
2429 && mddev->can_decrease_events
2430 && mddev->events != 1) {
2432 mddev->can_decrease_events = 0;
2434 /* otherwise we have to go forward and ... */
2436 mddev->can_decrease_events = nospares;
2439 if (!mddev->events) {
2441 * oops, this 64-bit counter should never wrap.
2442 * Either we are in around ~1 trillion A.C., assuming
2443 * 1 reboot per second, or we have a bug:
2449 rdev_for_each(rdev, mddev) {
2450 if (rdev->badblocks.changed)
2451 any_badblocks_changed++;
2452 if (test_bit(Faulty, &rdev->flags))
2453 set_bit(FaultRecorded, &rdev->flags);
2456 sync_sbs(mddev, nospares);
2457 spin_unlock_irq(&mddev->write_lock);
2459 pr_debug("md: updating %s RAID superblock on device (in sync %d)\n",
2460 mdname(mddev), mddev->in_sync);
2462 bitmap_update_sb(mddev->bitmap);
2463 rdev_for_each(rdev, mddev) {
2464 char b[BDEVNAME_SIZE];
2466 if (rdev->sb_loaded != 1)
2467 continue; /* no noise on spare devices */
2469 if (!test_bit(Faulty, &rdev->flags)) {
2470 md_super_write(mddev,rdev,
2471 rdev->sb_start, rdev->sb_size,
2473 pr_debug("md: (write) %s's sb offset: %llu\n",
2474 bdevname(rdev->bdev, b),
2475 (unsigned long long)rdev->sb_start);
2476 rdev->sb_events = mddev->events;
2477 if (rdev->badblocks.size) {
2478 md_super_write(mddev, rdev,
2479 rdev->badblocks.sector,
2480 rdev->badblocks.size << 9,
2482 rdev->badblocks.size = 0;
2486 pr_debug("md: %s (skipping faulty)\n",
2487 bdevname(rdev->bdev, b));
2489 if (mddev->level == LEVEL_MULTIPATH)
2490 /* only need to write one superblock... */
2493 md_super_wait(mddev);
2494 /* if there was a failure, MD_CHANGE_DEVS was set, and we re-write super */
2496 spin_lock_irq(&mddev->write_lock);
2497 if (mddev->in_sync != sync_req ||
2498 test_bit(MD_CHANGE_DEVS, &mddev->flags)) {
2499 /* have to write it out again */
2500 spin_unlock_irq(&mddev->write_lock);
2503 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
2504 spin_unlock_irq(&mddev->write_lock);
2505 wake_up(&mddev->sb_wait);
2506 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
2507 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
2509 rdev_for_each(rdev, mddev) {
2510 if (test_and_clear_bit(FaultRecorded, &rdev->flags))
2511 clear_bit(Blocked, &rdev->flags);
2513 if (any_badblocks_changed)
2514 md_ack_all_badblocks(&rdev->badblocks);
2515 clear_bit(BlockedBadBlocks, &rdev->flags);
2516 wake_up(&rdev->blocked_wait);
2520 /* words written to sysfs files may, or may not, be \n terminated.
2521 * We want to accept with case. For this we use cmd_match.
2523 static int cmd_match(const char *cmd, const char *str)
2525 /* See if cmd, written into a sysfs file, matches
2526 * str. They must either be the same, or cmd can
2527 * have a trailing newline
2529 while (*cmd && *str && *cmd == *str) {
2540 struct rdev_sysfs_entry {
2541 struct attribute attr;
2542 ssize_t (*show)(struct md_rdev *, char *);
2543 ssize_t (*store)(struct md_rdev *, const char *, size_t);
2547 state_show(struct md_rdev *rdev, char *page)
2552 if (test_bit(Faulty, &rdev->flags) ||
2553 rdev->badblocks.unacked_exist) {
2554 len+= sprintf(page+len, "%sfaulty",sep);
2557 if (test_bit(In_sync, &rdev->flags)) {
2558 len += sprintf(page+len, "%sin_sync",sep);
2561 if (test_bit(WriteMostly, &rdev->flags)) {
2562 len += sprintf(page+len, "%swrite_mostly",sep);
2565 if (test_bit(Blocked, &rdev->flags) ||
2566 (rdev->badblocks.unacked_exist
2567 && !test_bit(Faulty, &rdev->flags))) {
2568 len += sprintf(page+len, "%sblocked", sep);
2571 if (!test_bit(Faulty, &rdev->flags) &&
2572 !test_bit(In_sync, &rdev->flags)) {
2573 len += sprintf(page+len, "%sspare", sep);
2576 if (test_bit(WriteErrorSeen, &rdev->flags)) {
2577 len += sprintf(page+len, "%swrite_error", sep);
2580 if (test_bit(WantReplacement, &rdev->flags)) {
2581 len += sprintf(page+len, "%swant_replacement", sep);
2584 if (test_bit(Replacement, &rdev->flags)) {
2585 len += sprintf(page+len, "%sreplacement", sep);
2589 return len+sprintf(page+len, "\n");
2593 state_store(struct md_rdev *rdev, const char *buf, size_t len)
2596 * faulty - simulates an error
2597 * remove - disconnects the device
2598 * writemostly - sets write_mostly
2599 * -writemostly - clears write_mostly
2600 * blocked - sets the Blocked flags
2601 * -blocked - clears the Blocked and possibly simulates an error
2602 * insync - sets Insync providing device isn't active
2603 * -insync - clear Insync for a device with a slot assigned,
2604 * so that it gets rebuilt based on bitmap
2605 * write_error - sets WriteErrorSeen
2606 * -write_error - clears WriteErrorSeen
2609 if (cmd_match(buf, "faulty") && rdev->mddev->pers) {
2610 md_error(rdev->mddev, rdev);
2611 if (test_bit(Faulty, &rdev->flags))
2615 } else if (cmd_match(buf, "remove")) {
2616 if (rdev->raid_disk >= 0)
2619 struct mddev *mddev = rdev->mddev;
2620 kick_rdev_from_array(rdev);
2622 md_update_sb(mddev, 1);
2623 md_new_event(mddev);
2626 } else if (cmd_match(buf, "writemostly")) {
2627 set_bit(WriteMostly, &rdev->flags);
2629 } else if (cmd_match(buf, "-writemostly")) {
2630 clear_bit(WriteMostly, &rdev->flags);
2632 } else if (cmd_match(buf, "blocked")) {
2633 set_bit(Blocked, &rdev->flags);
2635 } else if (cmd_match(buf, "-blocked")) {
2636 if (!test_bit(Faulty, &rdev->flags) &&
2637 rdev->badblocks.unacked_exist) {
2638 /* metadata handler doesn't understand badblocks,
2639 * so we need to fail the device
2641 md_error(rdev->mddev, rdev);
2643 clear_bit(Blocked, &rdev->flags);
2644 clear_bit(BlockedBadBlocks, &rdev->flags);
2645 wake_up(&rdev->blocked_wait);
2646 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2647 md_wakeup_thread(rdev->mddev->thread);
2650 } else if (cmd_match(buf, "insync") && rdev->raid_disk == -1) {
2651 set_bit(In_sync, &rdev->flags);
2653 } else if (cmd_match(buf, "-insync") && rdev->raid_disk >= 0) {
2654 clear_bit(In_sync, &rdev->flags);
2655 rdev->saved_raid_disk = rdev->raid_disk;
2656 rdev->raid_disk = -1;
2658 } else if (cmd_match(buf, "write_error")) {
2659 set_bit(WriteErrorSeen, &rdev->flags);
2661 } else if (cmd_match(buf, "-write_error")) {
2662 clear_bit(WriteErrorSeen, &rdev->flags);
2664 } else if (cmd_match(buf, "want_replacement")) {
2665 /* Any non-spare device that is not a replacement can
2666 * become want_replacement at any time, but we then need to
2667 * check if recovery is needed.
2669 if (rdev->raid_disk >= 0 &&
2670 !test_bit(Replacement, &rdev->flags))
2671 set_bit(WantReplacement, &rdev->flags);
2672 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2673 md_wakeup_thread(rdev->mddev->thread);
2675 } else if (cmd_match(buf, "-want_replacement")) {
2676 /* Clearing 'want_replacement' is always allowed.
2677 * Once replacements starts it is too late though.
2680 clear_bit(WantReplacement, &rdev->flags);
2681 } else if (cmd_match(buf, "replacement")) {
2682 /* Can only set a device as a replacement when array has not
2683 * yet been started. Once running, replacement is automatic
2684 * from spares, or by assigning 'slot'.
2686 if (rdev->mddev->pers)
2689 set_bit(Replacement, &rdev->flags);
2692 } else if (cmd_match(buf, "-replacement")) {
2693 /* Similarly, can only clear Replacement before start */
2694 if (rdev->mddev->pers)
2697 clear_bit(Replacement, &rdev->flags);
2702 sysfs_notify_dirent_safe(rdev->sysfs_state);
2703 return err ? err : len;
2705 static struct rdev_sysfs_entry rdev_state =
2706 __ATTR(state, S_IRUGO|S_IWUSR, state_show, state_store);
2709 errors_show(struct md_rdev *rdev, char *page)
2711 return sprintf(page, "%d\n", atomic_read(&rdev->corrected_errors));
2715 errors_store(struct md_rdev *rdev, const char *buf, size_t len)
2718 unsigned long n = simple_strtoul(buf, &e, 10);
2719 if (*buf && (*e == 0 || *e == '\n')) {
2720 atomic_set(&rdev->corrected_errors, n);
2725 static struct rdev_sysfs_entry rdev_errors =
2726 __ATTR(errors, S_IRUGO|S_IWUSR, errors_show, errors_store);
2729 slot_show(struct md_rdev *rdev, char *page)
2731 if (rdev->raid_disk < 0)
2732 return sprintf(page, "none\n");
2734 return sprintf(page, "%d\n", rdev->raid_disk);
2738 slot_store(struct md_rdev *rdev, const char *buf, size_t len)
2742 int slot = simple_strtoul(buf, &e, 10);
2743 if (strncmp(buf, "none", 4)==0)
2745 else if (e==buf || (*e && *e!= '\n'))
2747 if (rdev->mddev->pers && slot == -1) {
2748 /* Setting 'slot' on an active array requires also
2749 * updating the 'rd%d' link, and communicating
2750 * with the personality with ->hot_*_disk.
2751 * For now we only support removing
2752 * failed/spare devices. This normally happens automatically,
2753 * but not when the metadata is externally managed.
2755 if (rdev->raid_disk == -1)
2757 /* personality does all needed checks */
2758 if (rdev->mddev->pers->hot_remove_disk == NULL)
2760 clear_bit(Blocked, &rdev->flags);
2761 remove_and_add_spares(rdev->mddev, rdev);
2762 if (rdev->raid_disk >= 0)
2764 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2765 md_wakeup_thread(rdev->mddev->thread);
2766 } else if (rdev->mddev->pers) {
2767 /* Activating a spare .. or possibly reactivating
2768 * if we ever get bitmaps working here.
2771 if (rdev->raid_disk != -1)
2774 if (test_bit(MD_RECOVERY_RUNNING, &rdev->mddev->recovery))
2777 if (rdev->mddev->pers->hot_add_disk == NULL)
2780 if (slot >= rdev->mddev->raid_disks &&
2781 slot >= rdev->mddev->raid_disks + rdev->mddev->delta_disks)
2784 rdev->raid_disk = slot;
2785 if (test_bit(In_sync, &rdev->flags))
2786 rdev->saved_raid_disk = slot;
2788 rdev->saved_raid_disk = -1;
2789 clear_bit(In_sync, &rdev->flags);
2790 clear_bit(Bitmap_sync, &rdev->flags);
2791 err = rdev->mddev->pers->
2792 hot_add_disk(rdev->mddev, rdev);
2794 rdev->raid_disk = -1;
2797 sysfs_notify_dirent_safe(rdev->sysfs_state);
2798 if (sysfs_link_rdev(rdev->mddev, rdev))
2799 /* failure here is OK */;
2800 /* don't wakeup anyone, leave that to userspace. */
2802 if (slot >= rdev->mddev->raid_disks &&
2803 slot >= rdev->mddev->raid_disks + rdev->mddev->delta_disks)
2805 rdev->raid_disk = slot;
2806 /* assume it is working */
2807 clear_bit(Faulty, &rdev->flags);
2808 clear_bit(WriteMostly, &rdev->flags);
2809 set_bit(In_sync, &rdev->flags);
2810 sysfs_notify_dirent_safe(rdev->sysfs_state);
2816 static struct rdev_sysfs_entry rdev_slot =
2817 __ATTR(slot, S_IRUGO|S_IWUSR, slot_show, slot_store);
2820 offset_show(struct md_rdev *rdev, char *page)
2822 return sprintf(page, "%llu\n", (unsigned long long)rdev->data_offset);
2826 offset_store(struct md_rdev *rdev, const char *buf, size_t len)
2828 unsigned long long offset;
2829 if (kstrtoull(buf, 10, &offset) < 0)
2831 if (rdev->mddev->pers && rdev->raid_disk >= 0)
2833 if (rdev->sectors && rdev->mddev->external)
2834 /* Must set offset before size, so overlap checks
2837 rdev->data_offset = offset;
2838 rdev->new_data_offset = offset;
2842 static struct rdev_sysfs_entry rdev_offset =
2843 __ATTR(offset, S_IRUGO|S_IWUSR, offset_show, offset_store);
2845 static ssize_t new_offset_show(struct md_rdev *rdev, char *page)
2847 return sprintf(page, "%llu\n",
2848 (unsigned long long)rdev->new_data_offset);
2851 static ssize_t new_offset_store(struct md_rdev *rdev,
2852 const char *buf, size_t len)
2854 unsigned long long new_offset;
2855 struct mddev *mddev = rdev->mddev;
2857 if (kstrtoull(buf, 10, &new_offset) < 0)
2860 if (mddev->sync_thread)
2862 if (new_offset == rdev->data_offset)
2863 /* reset is always permitted */
2865 else if (new_offset > rdev->data_offset) {
2866 /* must not push array size beyond rdev_sectors */
2867 if (new_offset - rdev->data_offset
2868 + mddev->dev_sectors > rdev->sectors)
2871 /* Metadata worries about other space details. */
2873 /* decreasing the offset is inconsistent with a backwards
2876 if (new_offset < rdev->data_offset &&
2877 mddev->reshape_backwards)
2879 /* Increasing offset is inconsistent with forwards
2880 * reshape. reshape_direction should be set to
2881 * 'backwards' first.
2883 if (new_offset > rdev->data_offset &&
2884 !mddev->reshape_backwards)
2887 if (mddev->pers && mddev->persistent &&
2888 !super_types[mddev->major_version]
2889 .allow_new_offset(rdev, new_offset))
2891 rdev->new_data_offset = new_offset;
2892 if (new_offset > rdev->data_offset)
2893 mddev->reshape_backwards = 1;
2894 else if (new_offset < rdev->data_offset)
2895 mddev->reshape_backwards = 0;
2899 static struct rdev_sysfs_entry rdev_new_offset =
2900 __ATTR(new_offset, S_IRUGO|S_IWUSR, new_offset_show, new_offset_store);
2903 rdev_size_show(struct md_rdev *rdev, char *page)
2905 return sprintf(page, "%llu\n", (unsigned long long)rdev->sectors / 2);
2908 static int overlaps(sector_t s1, sector_t l1, sector_t s2, sector_t l2)
2910 /* check if two start/length pairs overlap */
2918 static int strict_blocks_to_sectors(const char *buf, sector_t *sectors)
2920 unsigned long long blocks;
2923 if (kstrtoull(buf, 10, &blocks) < 0)
2926 if (blocks & 1ULL << (8 * sizeof(blocks) - 1))
2927 return -EINVAL; /* sector conversion overflow */
2930 if (new != blocks * 2)
2931 return -EINVAL; /* unsigned long long to sector_t overflow */
2938 rdev_size_store(struct md_rdev *rdev, const char *buf, size_t len)
2940 struct mddev *my_mddev = rdev->mddev;
2941 sector_t oldsectors = rdev->sectors;
2944 if (strict_blocks_to_sectors(buf, §ors) < 0)
2946 if (rdev->data_offset != rdev->new_data_offset)
2947 return -EINVAL; /* too confusing */
2948 if (my_mddev->pers && rdev->raid_disk >= 0) {
2949 if (my_mddev->persistent) {
2950 sectors = super_types[my_mddev->major_version].
2951 rdev_size_change(rdev, sectors);
2954 } else if (!sectors)
2955 sectors = (i_size_read(rdev->bdev->bd_inode) >> 9) -
2957 if (!my_mddev->pers->resize)
2958 /* Cannot change size for RAID0 or Linear etc */
2961 if (sectors < my_mddev->dev_sectors)
2962 return -EINVAL; /* component must fit device */
2964 rdev->sectors = sectors;
2965 if (sectors > oldsectors && my_mddev->external) {
2966 /* need to check that all other rdevs with the same ->bdev
2967 * do not overlap. We need to unlock the mddev to avoid
2968 * a deadlock. We have already changed rdev->sectors, and if
2969 * we have to change it back, we will have the lock again.
2971 struct mddev *mddev;
2973 struct list_head *tmp;
2975 mddev_unlock(my_mddev);
2976 for_each_mddev(mddev, tmp) {
2977 struct md_rdev *rdev2;
2979 mddev_lock_nointr(mddev);
2980 rdev_for_each(rdev2, mddev)
2981 if (rdev->bdev == rdev2->bdev &&
2983 overlaps(rdev->data_offset, rdev->sectors,
2989 mddev_unlock(mddev);
2995 mddev_lock_nointr(my_mddev);
2997 /* Someone else could have slipped in a size
2998 * change here, but doing so is just silly.
2999 * We put oldsectors back because we *know* it is
3000 * safe, and trust userspace not to race with
3003 rdev->sectors = oldsectors;
3010 static struct rdev_sysfs_entry rdev_size =
3011 __ATTR(size, S_IRUGO|S_IWUSR, rdev_size_show, rdev_size_store);
3014 static ssize_t recovery_start_show(struct md_rdev *rdev, char *page)
3016 unsigned long long recovery_start = rdev->recovery_offset;
3018 if (test_bit(In_sync, &rdev->flags) ||
3019 recovery_start == MaxSector)
3020 return sprintf(page, "none\n");
3022 return sprintf(page, "%llu\n", recovery_start);
3025 static ssize_t recovery_start_store(struct md_rdev *rdev, const char *buf, size_t len)
3027 unsigned long long recovery_start;
3029 if (cmd_match(buf, "none"))
3030 recovery_start = MaxSector;
3031 else if (kstrtoull(buf, 10, &recovery_start))
3034 if (rdev->mddev->pers &&
3035 rdev->raid_disk >= 0)
3038 rdev->recovery_offset = recovery_start;
3039 if (recovery_start == MaxSector)
3040 set_bit(In_sync, &rdev->flags);
3042 clear_bit(In_sync, &rdev->flags);
3046 static struct rdev_sysfs_entry rdev_recovery_start =
3047 __ATTR(recovery_start, S_IRUGO|S_IWUSR, recovery_start_show, recovery_start_store);
3051 badblocks_show(struct badblocks *bb, char *page, int unack);
3053 badblocks_store(struct badblocks *bb, const char *page, size_t len, int unack);
3055 static ssize_t bb_show(struct md_rdev *rdev, char *page)
3057 return badblocks_show(&rdev->badblocks, page, 0);
3059 static ssize_t bb_store(struct md_rdev *rdev, const char *page, size_t len)
3061 int rv = badblocks_store(&rdev->badblocks, page, len, 0);
3062 /* Maybe that ack was all we needed */
3063 if (test_and_clear_bit(BlockedBadBlocks, &rdev->flags))
3064 wake_up(&rdev->blocked_wait);
3067 static struct rdev_sysfs_entry rdev_bad_blocks =
3068 __ATTR(bad_blocks, S_IRUGO|S_IWUSR, bb_show, bb_store);
3071 static ssize_t ubb_show(struct md_rdev *rdev, char *page)
3073 return badblocks_show(&rdev->badblocks, page, 1);
3075 static ssize_t ubb_store(struct md_rdev *rdev, const char *page, size_t len)
3077 return badblocks_store(&rdev->badblocks, page, len, 1);
3079 static struct rdev_sysfs_entry rdev_unack_bad_blocks =
3080 __ATTR(unacknowledged_bad_blocks, S_IRUGO|S_IWUSR, ubb_show, ubb_store);
3082 static struct attribute *rdev_default_attrs[] = {
3087 &rdev_new_offset.attr,
3089 &rdev_recovery_start.attr,
3090 &rdev_bad_blocks.attr,
3091 &rdev_unack_bad_blocks.attr,
3095 rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
3097 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
3098 struct md_rdev *rdev = container_of(kobj, struct md_rdev, kobj);
3099 struct mddev *mddev = rdev->mddev;
3105 rv = mddev ? mddev_lock(mddev) : -EBUSY;
3107 if (rdev->mddev == NULL)
3110 rv = entry->show(rdev, page);
3111 mddev_unlock(mddev);
3117 rdev_attr_store(struct kobject *kobj, struct attribute *attr,
3118 const char *page, size_t length)
3120 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
3121 struct md_rdev *rdev = container_of(kobj, struct md_rdev, kobj);
3123 struct mddev *mddev = rdev->mddev;
3127 if (!capable(CAP_SYS_ADMIN))
3129 rv = mddev ? mddev_lock(mddev): -EBUSY;
3131 if (rdev->mddev == NULL)
3134 rv = entry->store(rdev, page, length);
3135 mddev_unlock(mddev);
3140 static void rdev_free(struct kobject *ko)
3142 struct md_rdev *rdev = container_of(ko, struct md_rdev, kobj);
3145 static const struct sysfs_ops rdev_sysfs_ops = {
3146 .show = rdev_attr_show,
3147 .store = rdev_attr_store,
3149 static struct kobj_type rdev_ktype = {
3150 .release = rdev_free,
3151 .sysfs_ops = &rdev_sysfs_ops,
3152 .default_attrs = rdev_default_attrs,
3155 int md_rdev_init(struct md_rdev *rdev)
3158 rdev->saved_raid_disk = -1;
3159 rdev->raid_disk = -1;
3161 rdev->data_offset = 0;
3162 rdev->new_data_offset = 0;
3163 rdev->sb_events = 0;
3164 rdev->last_read_error.tv_sec = 0;
3165 rdev->last_read_error.tv_nsec = 0;
3166 rdev->sb_loaded = 0;
3167 rdev->bb_page = NULL;
3168 atomic_set(&rdev->nr_pending, 0);
3169 atomic_set(&rdev->read_errors, 0);
3170 atomic_set(&rdev->corrected_errors, 0);
3172 INIT_LIST_HEAD(&rdev->same_set);
3173 init_waitqueue_head(&rdev->blocked_wait);
3175 /* Add space to store bad block list.
3176 * This reserves the space even on arrays where it cannot
3177 * be used - I wonder if that matters
3179 rdev->badblocks.count = 0;
3180 rdev->badblocks.shift = -1; /* disabled until explicitly enabled */
3181 rdev->badblocks.page = kmalloc(PAGE_SIZE, GFP_KERNEL);
3182 seqlock_init(&rdev->badblocks.lock);
3183 if (rdev->badblocks.page == NULL)
3188 EXPORT_SYMBOL_GPL(md_rdev_init);
3190 * Import a device. If 'super_format' >= 0, then sanity check the superblock
3192 * mark the device faulty if:
3194 * - the device is nonexistent (zero size)
3195 * - the device has no valid superblock
3197 * a faulty rdev _never_ has rdev->sb set.
3199 static struct md_rdev *md_import_device(dev_t newdev, int super_format, int super_minor)
3201 char b[BDEVNAME_SIZE];
3203 struct md_rdev *rdev;
3206 rdev = kzalloc(sizeof(*rdev), GFP_KERNEL);
3208 printk(KERN_ERR "md: could not alloc mem for new device!\n");
3209 return ERR_PTR(-ENOMEM);
3212 err = md_rdev_init(rdev);
3215 err = alloc_disk_sb(rdev);
3219 err = lock_rdev(rdev, newdev, super_format == -2);
3223 kobject_init(&rdev->kobj, &rdev_ktype);
3225 size = i_size_read(rdev->bdev->bd_inode) >> BLOCK_SIZE_BITS;
3228 "md: %s has zero or unknown size, marking faulty!\n",
3229 bdevname(rdev->bdev,b));
3234 if (super_format >= 0) {
3235 err = super_types[super_format].
3236 load_super(rdev, NULL, super_minor);
3237 if (err == -EINVAL) {
3239 "md: %s does not have a valid v%d.%d "
3240 "superblock, not importing!\n",
3241 bdevname(rdev->bdev,b),
3242 super_format, super_minor);
3247 "md: could not read %s's sb, not importing!\n",
3248 bdevname(rdev->bdev,b));
3258 md_rdev_clear(rdev);
3260 return ERR_PTR(err);
3264 * Check a full RAID array for plausibility
3268 static void analyze_sbs(struct mddev * mddev)
3271 struct md_rdev *rdev, *freshest, *tmp;
3272 char b[BDEVNAME_SIZE];
3275 rdev_for_each_safe(rdev, tmp, mddev)
3276 switch (super_types[mddev->major_version].
3277 load_super(rdev, freshest, mddev->minor_version)) {
3285 "md: fatal superblock inconsistency in %s"
3286 " -- removing from array\n",
3287 bdevname(rdev->bdev,b));
3288 kick_rdev_from_array(rdev);
3292 super_types[mddev->major_version].
3293 validate_super(mddev, freshest);
3296 rdev_for_each_safe(rdev, tmp, mddev) {
3297 if (mddev->max_disks &&
3298 (rdev->desc_nr >= mddev->max_disks ||
3299 i > mddev->max_disks)) {
3301 "md: %s: %s: only %d devices permitted\n",
3302 mdname(mddev), bdevname(rdev->bdev, b),
3304 kick_rdev_from_array(rdev);
3307 if (rdev != freshest)
3308 if (super_types[mddev->major_version].
3309 validate_super(mddev, rdev)) {
3310 printk(KERN_WARNING "md: kicking non-fresh %s"
3312 bdevname(rdev->bdev,b));
3313 kick_rdev_from_array(rdev);
3316 if (mddev->level == LEVEL_MULTIPATH) {
3317 rdev->desc_nr = i++;
3318 rdev->raid_disk = rdev->desc_nr;
3319 set_bit(In_sync, &rdev->flags);
3320 } else if (rdev->raid_disk >= (mddev->raid_disks - min(0, mddev->delta_disks))) {
3321 rdev->raid_disk = -1;
3322 clear_bit(In_sync, &rdev->flags);
3327 /* Read a fixed-point number.
3328 * Numbers in sysfs attributes should be in "standard" units where
3329 * possible, so time should be in seconds.
3330 * However we internally use a a much smaller unit such as
3331 * milliseconds or jiffies.
3332 * This function takes a decimal number with a possible fractional
3333 * component, and produces an integer which is the result of
3334 * multiplying that number by 10^'scale'.
3335 * all without any floating-point arithmetic.
3337 int strict_strtoul_scaled(const char *cp, unsigned long *res, int scale)
3339 unsigned long result = 0;
3341 while (isdigit(*cp) || (*cp == '.' && decimals < 0)) {
3344 else if (decimals < scale) {
3347 result = result * 10 + value;
3359 while (decimals < scale) {
3368 static void md_safemode_timeout(unsigned long data);
3371 safe_delay_show(struct mddev *mddev, char *page)
3373 int msec = (mddev->safemode_delay*1000)/HZ;
3374 return sprintf(page, "%d.%03d\n", msec/1000, msec%1000);
3377 safe_delay_store(struct mddev *mddev, const char *cbuf, size_t len)
3381 if (strict_strtoul_scaled(cbuf, &msec, 3) < 0)
3384 mddev->safemode_delay = 0;
3386 unsigned long old_delay = mddev->safemode_delay;
3387 mddev->safemode_delay = (msec*HZ)/1000;
3388 if (mddev->safemode_delay == 0)
3389 mddev->safemode_delay = 1;
3390 if (mddev->safemode_delay < old_delay || old_delay == 0)
3391 md_safemode_timeout((unsigned long)mddev);
3395 static struct md_sysfs_entry md_safe_delay =
3396 __ATTR(safe_mode_delay, S_IRUGO|S_IWUSR,safe_delay_show, safe_delay_store);
3399 level_show(struct mddev *mddev, char *page)
3401 struct md_personality *p = mddev->pers;
3403 return sprintf(page, "%s\n", p->name);
3404 else if (mddev->clevel[0])
3405 return sprintf(page, "%s\n", mddev->clevel);
3406 else if (mddev->level != LEVEL_NONE)
3407 return sprintf(page, "%d\n", mddev->level);
3413 level_store(struct mddev *mddev, const char *buf, size_t len)
3417 struct md_personality *pers;
3420 struct md_rdev *rdev;
3422 if (mddev->pers == NULL) {
3425 if (len >= sizeof(mddev->clevel))
3427 strncpy(mddev->clevel, buf, len);
3428 if (mddev->clevel[len-1] == '\n')
3430 mddev->clevel[len] = 0;
3431 mddev->level = LEVEL_NONE;
3437 /* request to change the personality. Need to ensure:
3438 * - array is not engaged in resync/recovery/reshape
3439 * - old personality can be suspended
3440 * - new personality will access other array.
3443 if (mddev->sync_thread ||
3444 mddev->reshape_position != MaxSector ||
3445 mddev->sysfs_active)
3448 if (!mddev->pers->quiesce) {
3449 printk(KERN_WARNING "md: %s: %s does not support online personality change\n",
3450 mdname(mddev), mddev->pers->name);
3454 /* Now find the new personality */
3455 if (len == 0 || len >= sizeof(clevel))
3457 strncpy(clevel, buf, len);
3458 if (clevel[len-1] == '\n')
3461 if (kstrtol(clevel, 10, &level))
3464 if (request_module("md-%s", clevel) != 0)
3465 request_module("md-level-%s", clevel);
3466 spin_lock(&pers_lock);
3467 pers = find_pers(level, clevel);
3468 if (!pers || !try_module_get(pers->owner)) {
3469 spin_unlock(&pers_lock);
3470 printk(KERN_WARNING "md: personality %s not loaded\n", clevel);
3473 spin_unlock(&pers_lock);
3475 if (pers == mddev->pers) {
3476 /* Nothing to do! */
3477 module_put(pers->owner);
3480 if (!pers->takeover) {
3481 module_put(pers->owner);
3482 printk(KERN_WARNING "md: %s: %s does not support personality takeover\n",
3483 mdname(mddev), clevel);
3487 rdev_for_each(rdev, mddev)
3488 rdev->new_raid_disk = rdev->raid_disk;
3490 /* ->takeover must set new_* and/or delta_disks
3491 * if it succeeds, and may set them when it fails.
3493 priv = pers->takeover(mddev);
3495 mddev->new_level = mddev->level;
3496 mddev->new_layout = mddev->layout;
3497 mddev->new_chunk_sectors = mddev->chunk_sectors;
3498 mddev->raid_disks -= mddev->delta_disks;
3499 mddev->delta_disks = 0;
3500 mddev->reshape_backwards = 0;
3501 module_put(pers->owner);
3502 printk(KERN_WARNING "md: %s: %s would not accept array\n",
3503 mdname(mddev), clevel);
3504 return PTR_ERR(priv);
3507 /* Looks like we have a winner */
3508 mddev_suspend(mddev);
3509 mddev->pers->stop(mddev);
3511 if (mddev->pers->sync_request == NULL &&
3512 pers->sync_request != NULL) {
3513 /* need to add the md_redundancy_group */
3514 if (sysfs_create_group(&mddev->kobj, &md_redundancy_group))
3516 "md: cannot register extra attributes for %s\n",
3518 mddev->sysfs_action = sysfs_get_dirent(mddev->kobj.sd, "sync_action");
3520 if (mddev->pers->sync_request != NULL &&
3521 pers->sync_request == NULL) {
3522 /* need to remove the md_redundancy_group */
3523 if (mddev->to_remove == NULL)
3524 mddev->to_remove = &md_redundancy_group;
3527 if (mddev->pers->sync_request == NULL &&
3529 /* We are converting from a no-redundancy array
3530 * to a redundancy array and metadata is managed
3531 * externally so we need to be sure that writes
3532 * won't block due to a need to transition
3534 * until external management is started.
3537 mddev->safemode_delay = 0;
3538 mddev->safemode = 0;
3541 rdev_for_each(rdev, mddev) {
3542 if (rdev->raid_disk < 0)
3544 if (rdev->new_raid_disk >= mddev->raid_disks)
3545 rdev->new_raid_disk = -1;
3546 if (rdev->new_raid_disk == rdev->raid_disk)
3548 sysfs_unlink_rdev(mddev, rdev);
3550 rdev_for_each(rdev, mddev) {
3551 if (rdev->raid_disk < 0)
3553 if (rdev->new_raid_disk == rdev->raid_disk)
3555 rdev->raid_disk = rdev->new_raid_disk;
3556 if (rdev->raid_disk < 0)
3557 clear_bit(In_sync, &rdev->flags);
3559 if (sysfs_link_rdev(mddev, rdev))
3560 printk(KERN_WARNING "md: cannot register rd%d"
3561 " for %s after level change\n",
3562 rdev->raid_disk, mdname(mddev));
3566 module_put(mddev->pers->owner);
3568 mddev->private = priv;
3569 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
3570 mddev->level = mddev->new_level;
3571 mddev->layout = mddev->new_layout;
3572 mddev->chunk_sectors = mddev->new_chunk_sectors;
3573 mddev->delta_disks = 0;
3574 mddev->reshape_backwards = 0;
3575 mddev->degraded = 0;
3576 if (mddev->pers->sync_request == NULL) {
3577 /* this is now an array without redundancy, so
3578 * it must always be in_sync
3581 del_timer_sync(&mddev->safemode_timer);
3583 blk_set_stacking_limits(&mddev->queue->limits);
3585 set_bit(MD_CHANGE_DEVS, &mddev->flags);
3586 mddev_resume(mddev);
3588 md_update_sb(mddev, 1);
3589 sysfs_notify(&mddev->kobj, NULL, "level");
3590 md_new_event(mddev);
3594 static struct md_sysfs_entry md_level =
3595 __ATTR(level, S_IRUGO|S_IWUSR, level_show, level_store);
3599 layout_show(struct mddev *mddev, char *page)
3601 /* just a number, not meaningful for all levels */
3602 if (mddev->reshape_position != MaxSector &&
3603 mddev->layout != mddev->new_layout)
3604 return sprintf(page, "%d (%d)\n",
3605 mddev->new_layout, mddev->layout);
3606 return sprintf(page, "%d\n", mddev->layout);
3610 layout_store(struct mddev *mddev, const char *buf, size_t len)
3613 unsigned long n = simple_strtoul(buf, &e, 10);
3615 if (!*buf || (*e && *e != '\n'))
3620 if (mddev->pers->check_reshape == NULL)
3624 mddev->new_layout = n;
3625 err = mddev->pers->check_reshape(mddev);
3627 mddev->new_layout = mddev->layout;
3631 mddev->new_layout = n;
3632 if (mddev->reshape_position == MaxSector)
3637 static struct md_sysfs_entry md_layout =
3638 __ATTR(layout, S_IRUGO|S_IWUSR, layout_show, layout_store);
3642 raid_disks_show(struct mddev *mddev, char *page)
3644 if (mddev->raid_disks == 0)
3646 if (mddev->reshape_position != MaxSector &&
3647 mddev->delta_disks != 0)
3648 return sprintf(page, "%d (%d)\n", mddev->raid_disks,
3649 mddev->raid_disks - mddev->delta_disks);
3650 return sprintf(page, "%d\n", mddev->raid_disks);
3653 static int update_raid_disks(struct mddev *mddev, int raid_disks);
3656 raid_disks_store(struct mddev *mddev, const char *buf, size_t len)
3660 unsigned long n = simple_strtoul(buf, &e, 10);
3662 if (!*buf || (*e && *e != '\n'))
3666 rv = update_raid_disks(mddev, n);
3667 else if (mddev->reshape_position != MaxSector) {
3668 struct md_rdev *rdev;
3669 int olddisks = mddev->raid_disks - mddev->delta_disks;
3671 rdev_for_each(rdev, mddev) {
3673 rdev->data_offset < rdev->new_data_offset)
3676 rdev->data_offset > rdev->new_data_offset)
3679 mddev->delta_disks = n - olddisks;
3680 mddev->raid_disks = n;
3681 mddev->reshape_backwards = (mddev->delta_disks < 0);
3683 mddev->raid_disks = n;
3684 return rv ? rv : len;
3686 static struct md_sysfs_entry md_raid_disks =
3687 __ATTR(raid_disks, S_IRUGO|S_IWUSR, raid_disks_show, raid_disks_store);
3690 chunk_size_show(struct mddev *mddev, char *page)
3692 if (mddev->reshape_position != MaxSector &&
3693 mddev->chunk_sectors != mddev->new_chunk_sectors)
3694 return sprintf(page, "%d (%d)\n",
3695 mddev->new_chunk_sectors << 9,
3696 mddev->chunk_sectors << 9);
3697 return sprintf(page, "%d\n", mddev->chunk_sectors << 9);
3701 chunk_size_store(struct mddev *mddev, const char *buf, size_t len)
3704 unsigned long n = simple_strtoul(buf, &e, 10);
3706 if (!*buf || (*e && *e != '\n'))
3711 if (mddev->pers->check_reshape == NULL)
3715 mddev->new_chunk_sectors = n >> 9;
3716 err = mddev->pers->check_reshape(mddev);
3718 mddev->new_chunk_sectors = mddev->chunk_sectors;
3722 mddev->new_chunk_sectors = n >> 9;
3723 if (mddev->reshape_position == MaxSector)
3724 mddev->chunk_sectors = n >> 9;
3728 static struct md_sysfs_entry md_chunk_size =
3729 __ATTR(chunk_size, S_IRUGO|S_IWUSR, chunk_size_show, chunk_size_store);
3732 resync_start_show(struct mddev *mddev, char *page)
3734 if (mddev->recovery_cp == MaxSector)
3735 return sprintf(page, "none\n");
3736 return sprintf(page, "%llu\n", (unsigned long long)mddev->recovery_cp);
3740 resync_start_store(struct mddev *mddev, const char *buf, size_t len)
3743 unsigned long long n = simple_strtoull(buf, &e, 10);
3745 if (mddev->pers && !test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
3747 if (cmd_match(buf, "none"))
3749 else if (!*buf || (*e && *e != '\n'))
3752 mddev->recovery_cp = n;
3754 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
3757 static struct md_sysfs_entry md_resync_start =
3758 __ATTR(resync_start, S_IRUGO|S_IWUSR, resync_start_show, resync_start_store);
3761 * The array state can be:
3764 * No devices, no size, no level
3765 * Equivalent to STOP_ARRAY ioctl
3767 * May have some settings, but array is not active
3768 * all IO results in error
3769 * When written, doesn't tear down array, but just stops it
3770 * suspended (not supported yet)
3771 * All IO requests will block. The array can be reconfigured.
3772 * Writing this, if accepted, will block until array is quiescent
3774 * no resync can happen. no superblocks get written.
3775 * write requests fail
3777 * like readonly, but behaves like 'clean' on a write request.
3779 * clean - no pending writes, but otherwise active.
3780 * When written to inactive array, starts without resync
3781 * If a write request arrives then
3782 * if metadata is known, mark 'dirty' and switch to 'active'.
3783 * if not known, block and switch to write-pending
3784 * If written to an active array that has pending writes, then fails.
3786 * fully active: IO and resync can be happening.
3787 * When written to inactive array, starts with resync
3790 * clean, but writes are blocked waiting for 'active' to be written.
3793 * like active, but no writes have been seen for a while (100msec).
3796 enum array_state { clear, inactive, suspended, readonly, read_auto, clean, active,
3797 write_pending, active_idle, bad_word};
3798 static char *array_states[] = {
3799 "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active",
3800 "write-pending", "active-idle", NULL };
3802 static int match_word(const char *word, char **list)
3805 for (n=0; list[n]; n++)
3806 if (cmd_match(word, list[n]))
3812 array_state_show(struct mddev *mddev, char *page)
3814 enum array_state st = inactive;
3827 else if (test_bit(MD_CHANGE_PENDING, &mddev->flags))
3829 else if (mddev->safemode)
3835 if (list_empty(&mddev->disks) &&
3836 mddev->raid_disks == 0 &&
3837 mddev->dev_sectors == 0)
3842 return sprintf(page, "%s\n", array_states[st]);
3845 static int do_md_stop(struct mddev * mddev, int ro, struct block_device *bdev);
3846 static int md_set_readonly(struct mddev * mddev, struct block_device *bdev);
3847 static int do_md_run(struct mddev * mddev);
3848 static int restart_array(struct mddev *mddev);
3851 array_state_store(struct mddev *mddev, const char *buf, size_t len)
3854 enum array_state st = match_word(buf, array_states);
3859 /* stopping an active array */
3860 err = do_md_stop(mddev, 0, NULL);
3863 /* stopping an active array */
3865 err = do_md_stop(mddev, 2, NULL);
3867 err = 0; /* already inactive */
3870 break; /* not supported yet */
3873 err = md_set_readonly(mddev, NULL);
3876 set_disk_ro(mddev->gendisk, 1);
3877 err = do_md_run(mddev);
3883 err = md_set_readonly(mddev, NULL);
3884 else if (mddev->ro == 1)
3885 err = restart_array(mddev);
3888 set_disk_ro(mddev->gendisk, 0);
3892 err = do_md_run(mddev);
3897 restart_array(mddev);
3898 spin_lock_irq(&mddev->write_lock);
3899 if (atomic_read(&mddev->writes_pending) == 0) {
3900 if (mddev->in_sync == 0) {
3902 if (mddev->safemode == 1)
3903 mddev->safemode = 0;
3904 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
3909 spin_unlock_irq(&mddev->write_lock);
3915 restart_array(mddev);
3916 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
3917 wake_up(&mddev->sb_wait);
3921 set_disk_ro(mddev->gendisk, 0);
3922 err = do_md_run(mddev);
3927 /* these cannot be set */
3933 if (mddev->hold_active == UNTIL_IOCTL)
3934 mddev->hold_active = 0;
3935 sysfs_notify_dirent_safe(mddev->sysfs_state);
3939 static struct md_sysfs_entry md_array_state =
3940 __ATTR(array_state, S_IRUGO|S_IWUSR, array_state_show, array_state_store);
3943 max_corrected_read_errors_show(struct mddev *mddev, char *page) {
3944 return sprintf(page, "%d\n",
3945 atomic_read(&mddev->max_corr_read_errors));
3949 max_corrected_read_errors_store(struct mddev *mddev, const char *buf, size_t len)
3952 unsigned long n = simple_strtoul(buf, &e, 10);
3954 if (*buf && (*e == 0 || *e == '\n')) {
3955 atomic_set(&mddev->max_corr_read_errors, n);
3961 static struct md_sysfs_entry max_corr_read_errors =
3962 __ATTR(max_read_errors, S_IRUGO|S_IWUSR, max_corrected_read_errors_show,
3963 max_corrected_read_errors_store);
3966 null_show(struct mddev *mddev, char *page)
3972 new_dev_store(struct mddev *mddev, const char *buf, size_t len)
3974 /* buf must be %d:%d\n? giving major and minor numbers */
3975 /* The new device is added to the array.
3976 * If the array has a persistent superblock, we read the
3977 * superblock to initialise info and check validity.
3978 * Otherwise, only checking done is that in bind_rdev_to_array,
3979 * which mainly checks size.
3982 int major = simple_strtoul(buf, &e, 10);
3985 struct md_rdev *rdev;
3988 if (!*buf || *e != ':' || !e[1] || e[1] == '\n')
3990 minor = simple_strtoul(e+1, &e, 10);
3991 if (*e && *e != '\n')
3993 dev = MKDEV(major, minor);
3994 if (major != MAJOR(dev) ||
3995 minor != MINOR(dev))
3999 if (mddev->persistent) {
4000 rdev = md_import_device(dev, mddev->major_version,
4001 mddev->minor_version);
4002 if (!IS_ERR(rdev) && !list_empty(&mddev->disks)) {
4003 struct md_rdev *rdev0
4004 = list_entry(mddev->disks.next,
4005 struct md_rdev, same_set);
4006 err = super_types[mddev->major_version]
4007 .load_super(rdev, rdev0, mddev->minor_version);
4011 } else if (mddev->external)
4012 rdev = md_import_device(dev, -2, -1);
4014 rdev = md_import_device(dev, -1, -1);
4017 return PTR_ERR(rdev);
4018 err = bind_rdev_to_array(rdev, mddev);
4022 return err ? err : len;
4025 static struct md_sysfs_entry md_new_device =
4026 __ATTR(new_dev, S_IWUSR, null_show, new_dev_store);
4029 bitmap_store(struct mddev *mddev, const char *buf, size_t len)
4032 unsigned long chunk, end_chunk;
4036 /* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */
4038 chunk = end_chunk = simple_strtoul(buf, &end, 0);
4039 if (buf == end) break;
4040 if (*end == '-') { /* range */
4042 end_chunk = simple_strtoul(buf, &end, 0);
4043 if (buf == end) break;
4045 if (*end && !isspace(*end)) break;
4046 bitmap_dirty_bits(mddev->bitmap, chunk, end_chunk);
4047 buf = skip_spaces(end);
4049 bitmap_unplug(mddev->bitmap); /* flush the bits to disk */
4054 static struct md_sysfs_entry md_bitmap =
4055 __ATTR(bitmap_set_bits, S_IWUSR, null_show, bitmap_store);
4058 size_show(struct mddev *mddev, char *page)
4060 return sprintf(page, "%llu\n",
4061 (unsigned long long)mddev->dev_sectors / 2);
4064 static int update_size(struct mddev *mddev, sector_t num_sectors);
4067 size_store(struct mddev *mddev, const char *buf, size_t len)
4069 /* If array is inactive, we can reduce the component size, but
4070 * not increase it (except from 0).
4071 * If array is active, we can try an on-line resize
4074 int err = strict_blocks_to_sectors(buf, §ors);
4079 err = update_size(mddev, sectors);
4080 md_update_sb(mddev, 1);
4082 if (mddev->dev_sectors == 0 ||
4083 mddev->dev_sectors > sectors)
4084 mddev->dev_sectors = sectors;
4088 return err ? err : len;
4091 static struct md_sysfs_entry md_size =
4092 __ATTR(component_size, S_IRUGO|S_IWUSR, size_show, size_store);
4095 /* Metadata version.
4097 * 'none' for arrays with no metadata (good luck...)
4098 * 'external' for arrays with externally managed metadata,
4099 * or N.M for internally known formats
4102 metadata_show(struct mddev *mddev, char *page)
4104 if (mddev->persistent)
4105 return sprintf(page, "%d.%d\n",
4106 mddev->major_version, mddev->minor_version);
4107 else if (mddev->external)
4108 return sprintf(page, "external:%s\n", mddev->metadata_type);
4110 return sprintf(page, "none\n");
4114 metadata_store(struct mddev *mddev, const char *buf, size_t len)
4118 /* Changing the details of 'external' metadata is
4119 * always permitted. Otherwise there must be
4120 * no devices attached to the array.
4122 if (mddev->external && strncmp(buf, "external:", 9) == 0)
4124 else if (!list_empty(&mddev->disks))
4127 if (cmd_match(buf, "none")) {
4128 mddev->persistent = 0;
4129 mddev->external = 0;
4130 mddev->major_version = 0;
4131 mddev->minor_version = 90;
4134 if (strncmp(buf, "external:", 9) == 0) {
4135 size_t namelen = len-9;
4136 if (namelen >= sizeof(mddev->metadata_type))
4137 namelen = sizeof(mddev->metadata_type)-1;
4138 strncpy(mddev->metadata_type, buf+9, namelen);
4139 mddev->metadata_type[namelen] = 0;
4140 if (namelen && mddev->metadata_type[namelen-1] == '\n')
4141 mddev->metadata_type[--namelen] = 0;
4142 mddev->persistent = 0;
4143 mddev->external = 1;
4144 mddev->major_version = 0;
4145 mddev->minor_version = 90;
4148 major = simple_strtoul(buf, &e, 10);
4149 if (e==buf || *e != '.')
4152 minor = simple_strtoul(buf, &e, 10);
4153 if (e==buf || (*e && *e != '\n') )
4155 if (major >= ARRAY_SIZE(super_types) || super_types[major].name == NULL)
4157 mddev->major_version = major;
4158 mddev->minor_version = minor;
4159 mddev->persistent = 1;
4160 mddev->external = 0;
4164 static struct md_sysfs_entry md_metadata =
4165 __ATTR(metadata_version, S_IRUGO|S_IWUSR, metadata_show, metadata_store);
4168 action_show(struct mddev *mddev, char *page)
4170 char *type = "idle";
4171 if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
4173 else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
4174 (!mddev->ro && test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))) {
4175 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
4177 else if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
4178 if (!test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
4180 else if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
4184 } else if (test_bit(MD_RECOVERY_RECOVER, &mddev->recovery))
4187 return sprintf(page, "%s\n", type);
4191 action_store(struct mddev *mddev, const char *page, size_t len)
4193 if (!mddev->pers || !mddev->pers->sync_request)
4196 if (cmd_match(page, "frozen"))
4197 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4199 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4201 if (cmd_match(page, "idle") || cmd_match(page, "frozen")) {
4202 if (mddev->sync_thread) {
4203 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
4204 md_reap_sync_thread(mddev);
4206 } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
4207 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
4209 else if (cmd_match(page, "resync"))
4210 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4211 else if (cmd_match(page, "recover")) {
4212 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
4213 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4214 } else if (cmd_match(page, "reshape")) {
4216 if (mddev->pers->start_reshape == NULL)
4218 err = mddev->pers->start_reshape(mddev);
4221 sysfs_notify(&mddev->kobj, NULL, "degraded");
4223 if (cmd_match(page, "check"))
4224 set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
4225 else if (!cmd_match(page, "repair"))
4227 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
4228 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
4230 if (mddev->ro == 2) {
4231 /* A write to sync_action is enough to justify
4232 * canceling read-auto mode
4235 md_wakeup_thread(mddev->sync_thread);
4237 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4238 md_wakeup_thread(mddev->thread);
4239 sysfs_notify_dirent_safe(mddev->sysfs_action);
4243 static struct md_sysfs_entry md_scan_mode =
4244 __ATTR(sync_action, S_IRUGO|S_IWUSR, action_show, action_store);
4247 last_sync_action_show(struct mddev *mddev, char *page)
4249 return sprintf(page, "%s\n", mddev->last_sync_action);
4252 static struct md_sysfs_entry md_last_scan_mode = __ATTR_RO(last_sync_action);
4255 mismatch_cnt_show(struct mddev *mddev, char *page)
4257 return sprintf(page, "%llu\n",
4258 (unsigned long long)
4259 atomic64_read(&mddev->resync_mismatches));
4262 static struct md_sysfs_entry md_mismatches = __ATTR_RO(mismatch_cnt);
4265 sync_min_show(struct mddev *mddev, char *page)
4267 return sprintf(page, "%d (%s)\n", speed_min(mddev),
4268 mddev->sync_speed_min ? "local": "system");
4272 sync_min_store(struct mddev *mddev, const char *buf, size_t len)
4276 if (strncmp(buf, "system", 6)==0) {
4277 mddev->sync_speed_min = 0;
4280 min = simple_strtoul(buf, &e, 10);
4281 if (buf == e || (*e && *e != '\n') || min <= 0)
4283 mddev->sync_speed_min = min;
4287 static struct md_sysfs_entry md_sync_min =
4288 __ATTR(sync_speed_min, S_IRUGO|S_IWUSR, sync_min_show, sync_min_store);
4291 sync_max_show(struct mddev *mddev, char *page)
4293 return sprintf(page, "%d (%s)\n", speed_max(mddev),
4294 mddev->sync_speed_max ? "local": "system");
4298 sync_max_store(struct mddev *mddev, const char *buf, size_t len)
4302 if (strncmp(buf, "system", 6)==0) {
4303 mddev->sync_speed_max = 0;
4306 max = simple_strtoul(buf, &e, 10);
4307 if (buf == e || (*e && *e != '\n') || max <= 0)
4309 mddev->sync_speed_max = max;
4313 static struct md_sysfs_entry md_sync_max =
4314 __ATTR(sync_speed_max, S_IRUGO|S_IWUSR, sync_max_show, sync_max_store);
4317 degraded_show(struct mddev *mddev, char *page)
4319 return sprintf(page, "%d\n", mddev->degraded);
4321 static struct md_sysfs_entry md_degraded = __ATTR_RO(degraded);
4324 sync_force_parallel_show(struct mddev *mddev, char *page)
4326 return sprintf(page, "%d\n", mddev->parallel_resync);
4330 sync_force_parallel_store(struct mddev *mddev, const char *buf, size_t len)
4334 if (kstrtol(buf, 10, &n))
4337 if (n != 0 && n != 1)
4340 mddev->parallel_resync = n;
4342 if (mddev->sync_thread)
4343 wake_up(&resync_wait);
4348 /* force parallel resync, even with shared block devices */
4349 static struct md_sysfs_entry md_sync_force_parallel =
4350 __ATTR(sync_force_parallel, S_IRUGO|S_IWUSR,
4351 sync_force_parallel_show, sync_force_parallel_store);
4354 sync_speed_show(struct mddev *mddev, char *page)
4356 unsigned long resync, dt, db;
4357 if (mddev->curr_resync == 0)
4358 return sprintf(page, "none\n");
4359 resync = mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active);
4360 dt = (jiffies - mddev->resync_mark) / HZ;
4362 db = resync - mddev->resync_mark_cnt;
4363 return sprintf(page, "%lu\n", db/dt/2); /* K/sec */
4366 static struct md_sysfs_entry md_sync_speed = __ATTR_RO(sync_speed);
4369 sync_completed_show(struct mddev *mddev, char *page)
4371 unsigned long long max_sectors, resync;
4373 if (!test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4374 return sprintf(page, "none\n");
4376 if (mddev->curr_resync == 1 ||
4377 mddev->curr_resync == 2)
4378 return sprintf(page, "delayed\n");
4380 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ||
4381 test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
4382 max_sectors = mddev->resync_max_sectors;
4384 max_sectors = mddev->dev_sectors;
4386 resync = mddev->curr_resync_completed;
4387 return sprintf(page, "%llu / %llu\n", resync, max_sectors);
4390 static struct md_sysfs_entry md_sync_completed = __ATTR_RO(sync_completed);
4393 min_sync_show(struct mddev *mddev, char *page)
4395 return sprintf(page, "%llu\n",
4396 (unsigned long long)mddev->resync_min);
4399 min_sync_store(struct mddev *mddev, const char *buf, size_t len)
4401 unsigned long long min;
4402 if (kstrtoull(buf, 10, &min))
4404 if (min > mddev->resync_max)
4406 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4409 /* Must be a multiple of chunk_size */
4410 if (mddev->chunk_sectors) {
4411 sector_t temp = min;
4412 if (sector_div(temp, mddev->chunk_sectors))
4415 mddev->resync_min = min;
4420 static struct md_sysfs_entry md_min_sync =
4421 __ATTR(sync_min, S_IRUGO|S_IWUSR, min_sync_show, min_sync_store);
4424 max_sync_show(struct mddev *mddev, char *page)
4426 if (mddev->resync_max == MaxSector)
4427 return sprintf(page, "max\n");
4429 return sprintf(page, "%llu\n",
4430 (unsigned long long)mddev->resync_max);
4433 max_sync_store(struct mddev *mddev, const char *buf, size_t len)
4435 if (strncmp(buf, "max", 3) == 0)
4436 mddev->resync_max = MaxSector;
4438 unsigned long long max;
4439 if (kstrtoull(buf, 10, &max))
4441 if (max < mddev->resync_min)
4443 if (max < mddev->resync_max &&
4445 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4448 /* Must be a multiple of chunk_size */
4449 if (mddev->chunk_sectors) {
4450 sector_t temp = max;
4451 if (sector_div(temp, mddev->chunk_sectors))
4454 mddev->resync_max = max;
4456 wake_up(&mddev->recovery_wait);
4460 static struct md_sysfs_entry md_max_sync =
4461 __ATTR(sync_max, S_IRUGO|S_IWUSR, max_sync_show, max_sync_store);
4464 suspend_lo_show(struct mddev *mddev, char *page)
4466 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_lo);
4470 suspend_lo_store(struct mddev *mddev, const char *buf, size_t len)
4473 unsigned long long new = simple_strtoull(buf, &e, 10);
4474 unsigned long long old = mddev->suspend_lo;
4476 if (mddev->pers == NULL ||
4477 mddev->pers->quiesce == NULL)
4479 if (buf == e || (*e && *e != '\n'))
4482 mddev->suspend_lo = new;
4484 /* Shrinking suspended region */
4485 mddev->pers->quiesce(mddev, 2);
4487 /* Expanding suspended region - need to wait */
4488 mddev->pers->quiesce(mddev, 1);
4489 mddev->pers->quiesce(mddev, 0);
4493 static struct md_sysfs_entry md_suspend_lo =
4494 __ATTR(suspend_lo, S_IRUGO|S_IWUSR, suspend_lo_show, suspend_lo_store);
4498 suspend_hi_show(struct mddev *mddev, char *page)
4500 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_hi);
4504 suspend_hi_store(struct mddev *mddev, const char *buf, size_t len)
4507 unsigned long long new = simple_strtoull(buf, &e, 10);
4508 unsigned long long old = mddev->suspend_hi;
4510 if (mddev->pers == NULL ||
4511 mddev->pers->quiesce == NULL)
4513 if (buf == e || (*e && *e != '\n'))
4516 mddev->suspend_hi = new;
4518 /* Shrinking suspended region */
4519 mddev->pers->quiesce(mddev, 2);
4521 /* Expanding suspended region - need to wait */
4522 mddev->pers->quiesce(mddev, 1);
4523 mddev->pers->quiesce(mddev, 0);
4527 static struct md_sysfs_entry md_suspend_hi =
4528 __ATTR(suspend_hi, S_IRUGO|S_IWUSR, suspend_hi_show, suspend_hi_store);
4531 reshape_position_show(struct mddev *mddev, char *page)
4533 if (mddev->reshape_position != MaxSector)
4534 return sprintf(page, "%llu\n",
4535 (unsigned long long)mddev->reshape_position);
4536 strcpy(page, "none\n");
4541 reshape_position_store(struct mddev *mddev, const char *buf, size_t len)
4543 struct md_rdev *rdev;
4545 unsigned long long new = simple_strtoull(buf, &e, 10);
4548 if (buf == e || (*e && *e != '\n'))
4550 mddev->reshape_position = new;
4551 mddev->delta_disks = 0;
4552 mddev->reshape_backwards = 0;
4553 mddev->new_level = mddev->level;
4554 mddev->new_layout = mddev->layout;
4555 mddev->new_chunk_sectors = mddev->chunk_sectors;
4556 rdev_for_each(rdev, mddev)
4557 rdev->new_data_offset = rdev->data_offset;
4561 static struct md_sysfs_entry md_reshape_position =
4562 __ATTR(reshape_position, S_IRUGO|S_IWUSR, reshape_position_show,
4563 reshape_position_store);
4566 reshape_direction_show(struct mddev *mddev, char *page)
4568 return sprintf(page, "%s\n",
4569 mddev->reshape_backwards ? "backwards" : "forwards");
4573 reshape_direction_store(struct mddev *mddev, const char *buf, size_t len)
4576 if (cmd_match(buf, "forwards"))
4578 else if (cmd_match(buf, "backwards"))
4582 if (mddev->reshape_backwards == backwards)
4585 /* check if we are allowed to change */
4586 if (mddev->delta_disks)
4589 if (mddev->persistent &&
4590 mddev->major_version == 0)
4593 mddev->reshape_backwards = backwards;
4597 static struct md_sysfs_entry md_reshape_direction =
4598 __ATTR(reshape_direction, S_IRUGO|S_IWUSR, reshape_direction_show,
4599 reshape_direction_store);
4602 array_size_show(struct mddev *mddev, char *page)
4604 if (mddev->external_size)
4605 return sprintf(page, "%llu\n",
4606 (unsigned long long)mddev->array_sectors/2);
4608 return sprintf(page, "default\n");
4612 array_size_store(struct mddev *mddev, const char *buf, size_t len)
4616 if (strncmp(buf, "default", 7) == 0) {
4618 sectors = mddev->pers->size(mddev, 0, 0);
4620 sectors = mddev->array_sectors;
4622 mddev->external_size = 0;
4624 if (strict_blocks_to_sectors(buf, §ors) < 0)
4626 if (mddev->pers && mddev->pers->size(mddev, 0, 0) < sectors)
4629 mddev->external_size = 1;
4632 mddev->array_sectors = sectors;
4634 set_capacity(mddev->gendisk, mddev->array_sectors);
4635 revalidate_disk(mddev->gendisk);
4640 static struct md_sysfs_entry md_array_size =
4641 __ATTR(array_size, S_IRUGO|S_IWUSR, array_size_show,
4644 static struct attribute *md_default_attrs[] = {
4647 &md_raid_disks.attr,
4648 &md_chunk_size.attr,
4650 &md_resync_start.attr,
4652 &md_new_device.attr,
4653 &md_safe_delay.attr,
4654 &md_array_state.attr,
4655 &md_reshape_position.attr,
4656 &md_reshape_direction.attr,
4657 &md_array_size.attr,
4658 &max_corr_read_errors.attr,
4662 static struct attribute *md_redundancy_attrs[] = {
4664 &md_last_scan_mode.attr,
4665 &md_mismatches.attr,
4668 &md_sync_speed.attr,
4669 &md_sync_force_parallel.attr,
4670 &md_sync_completed.attr,
4673 &md_suspend_lo.attr,
4674 &md_suspend_hi.attr,
4679 static struct attribute_group md_redundancy_group = {
4681 .attrs = md_redundancy_attrs,
4686 md_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
4688 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
4689 struct mddev *mddev = container_of(kobj, struct mddev, kobj);
4694 spin_lock(&all_mddevs_lock);
4695 if (list_empty(&mddev->all_mddevs)) {
4696 spin_unlock(&all_mddevs_lock);
4700 spin_unlock(&all_mddevs_lock);
4702 rv = mddev_lock(mddev);
4704 rv = entry->show(mddev, page);
4705 mddev_unlock(mddev);
4712 md_attr_store(struct kobject *kobj, struct attribute *attr,
4713 const char *page, size_t length)
4715 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
4716 struct mddev *mddev = container_of(kobj, struct mddev, kobj);
4721 if (!capable(CAP_SYS_ADMIN))
4723 spin_lock(&all_mddevs_lock);
4724 if (list_empty(&mddev->all_mddevs)) {
4725 spin_unlock(&all_mddevs_lock);
4729 spin_unlock(&all_mddevs_lock);
4730 if (entry->store == new_dev_store)
4731 flush_workqueue(md_misc_wq);
4732 rv = mddev_lock(mddev);
4734 rv = entry->store(mddev, page, length);
4735 mddev_unlock(mddev);
4741 static void md_free(struct kobject *ko)
4743 struct mddev *mddev = container_of(ko, struct mddev, kobj);
4745 if (mddev->sysfs_state)
4746 sysfs_put(mddev->sysfs_state);
4748 if (mddev->gendisk) {
4749 del_gendisk(mddev->gendisk);
4750 put_disk(mddev->gendisk);
4753 blk_cleanup_queue(mddev->queue);
4758 static const struct sysfs_ops md_sysfs_ops = {
4759 .show = md_attr_show,
4760 .store = md_attr_store,
4762 static struct kobj_type md_ktype = {
4764 .sysfs_ops = &md_sysfs_ops,
4765 .default_attrs = md_default_attrs,
4770 static void mddev_delayed_delete(struct work_struct *ws)
4772 struct mddev *mddev = container_of(ws, struct mddev, del_work);
4774 sysfs_remove_group(&mddev->kobj, &md_bitmap_group);
4775 kobject_del(&mddev->kobj);
4776 kobject_put(&mddev->kobj);
4779 static int md_alloc(dev_t dev, char *name)
4781 static DEFINE_MUTEX(disks_mutex);
4782 struct mddev *mddev = mddev_find(dev);
4783 struct gendisk *disk;
4792 partitioned = (MAJOR(mddev->unit) != MD_MAJOR);
4793 shift = partitioned ? MdpMinorShift : 0;
4794 unit = MINOR(mddev->unit) >> shift;
4796 /* wait for any previous instance of this device to be
4797 * completely removed (mddev_delayed_delete).
4799 flush_workqueue(md_misc_wq);
4801 mutex_lock(&disks_mutex);
4807 /* Need to ensure that 'name' is not a duplicate.
4809 struct mddev *mddev2;
4810 spin_lock(&all_mddevs_lock);
4812 list_for_each_entry(mddev2, &all_mddevs, all_mddevs)
4813 if (mddev2->gendisk &&
4814 strcmp(mddev2->gendisk->disk_name, name) == 0) {
4815 spin_unlock(&all_mddevs_lock);
4818 spin_unlock(&all_mddevs_lock);
4822 mddev->queue = blk_alloc_queue(GFP_KERNEL);
4825 mddev->queue->queuedata = mddev;
4827 blk_queue_make_request(mddev->queue, md_make_request);
4828 blk_set_stacking_limits(&mddev->queue->limits);
4830 disk = alloc_disk(1 << shift);
4832 blk_cleanup_queue(mddev->queue);
4833 mddev->queue = NULL;
4836 disk->major = MAJOR(mddev->unit);
4837 disk->first_minor = unit << shift;
4839 strcpy(disk->disk_name, name);
4840 else if (partitioned)
4841 sprintf(disk->disk_name, "md_d%d", unit);
4843 sprintf(disk->disk_name, "md%d", unit);
4844 disk->fops = &md_fops;
4845 disk->private_data = mddev;
4846 disk->queue = mddev->queue;
4847 blk_queue_flush(mddev->queue, REQ_FLUSH | REQ_FUA);
4848 /* Allow extended partitions. This makes the
4849 * 'mdp' device redundant, but we can't really
4852 disk->flags |= GENHD_FL_EXT_DEVT;
4853 mddev->gendisk = disk;
4854 /* As soon as we call add_disk(), another thread could get
4855 * through to md_open, so make sure it doesn't get too far
4857 mutex_lock(&mddev->open_mutex);
4860 error = kobject_init_and_add(&mddev->kobj, &md_ktype,
4861 &disk_to_dev(disk)->kobj, "%s", "md");
4863 /* This isn't possible, but as kobject_init_and_add is marked
4864 * __must_check, we must do something with the result
4866 printk(KERN_WARNING "md: cannot register %s/md - name in use\n",
4870 if (mddev->kobj.sd &&
4871 sysfs_create_group(&mddev->kobj, &md_bitmap_group))
4872 printk(KERN_DEBUG "pointless warning\n");
4873 mutex_unlock(&mddev->open_mutex);
4875 mutex_unlock(&disks_mutex);
4876 if (!error && mddev->kobj.sd) {
4877 kobject_uevent(&mddev->kobj, KOBJ_ADD);
4878 mddev->sysfs_state = sysfs_get_dirent_safe(mddev->kobj.sd, "array_state");
4884 static struct kobject *md_probe(dev_t dev, int *part, void *data)
4886 md_alloc(dev, NULL);
4890 static int add_named_array(const char *val, struct kernel_param *kp)
4892 /* val must be "md_*" where * is not all digits.
4893 * We allocate an array with a large free minor number, and
4894 * set the name to val. val must not already be an active name.
4896 int len = strlen(val);
4897 char buf[DISK_NAME_LEN];
4899 while (len && val[len-1] == '\n')
4901 if (len >= DISK_NAME_LEN)
4903 strlcpy(buf, val, len+1);
4904 if (strncmp(buf, "md_", 3) != 0)
4906 return md_alloc(0, buf);
4909 static void md_safemode_timeout(unsigned long data)
4911 struct mddev *mddev = (struct mddev *) data;
4913 if (!atomic_read(&mddev->writes_pending)) {
4914 mddev->safemode = 1;
4915 if (mddev->external)
4916 sysfs_notify_dirent_safe(mddev->sysfs_state);
4918 md_wakeup_thread(mddev->thread);
4921 static int start_dirty_degraded;
4923 int md_run(struct mddev *mddev)
4926 struct md_rdev *rdev;
4927 struct md_personality *pers;
4929 if (list_empty(&mddev->disks))
4930 /* cannot run an array with no devices.. */
4935 /* Cannot run until previous stop completes properly */
4936 if (mddev->sysfs_active)
4940 * Analyze all RAID superblock(s)
4942 if (!mddev->raid_disks) {
4943 if (!mddev->persistent)
4948 if (mddev->level != LEVEL_NONE)
4949 request_module("md-level-%d", mddev->level);
4950 else if (mddev->clevel[0])
4951 request_module("md-%s", mddev->clevel);
4954 * Drop all container device buffers, from now on
4955 * the only valid external interface is through the md
4958 rdev_for_each(rdev, mddev) {
4959 if (test_bit(Faulty, &rdev->flags))
4961 sync_blockdev(rdev->bdev);
4962 invalidate_bdev(rdev->bdev);
4964 /* perform some consistency tests on the device.
4965 * We don't want the data to overlap the metadata,
4966 * Internal Bitmap issues have been handled elsewhere.
4968 if (rdev->meta_bdev) {
4969 /* Nothing to check */;
4970 } else if (rdev->data_offset < rdev->sb_start) {
4971 if (mddev->dev_sectors &&
4972 rdev->data_offset + mddev->dev_sectors
4974 printk("md: %s: data overlaps metadata\n",
4979 if (rdev->sb_start + rdev->sb_size/512
4980 > rdev->data_offset) {
4981 printk("md: %s: metadata overlaps data\n",
4986 sysfs_notify_dirent_safe(rdev->sysfs_state);
4989 if (mddev->bio_set == NULL)
4990 mddev->bio_set = bioset_create(BIO_POOL_SIZE, 0);
4992 spin_lock(&pers_lock);
4993 pers = find_pers(mddev->level, mddev->clevel);
4994 if (!pers || !try_module_get(pers->owner)) {
4995 spin_unlock(&pers_lock);
4996 if (mddev->level != LEVEL_NONE)
4997 printk(KERN_WARNING "md: personality for level %d is not loaded!\n",
5000 printk(KERN_WARNING "md: personality for level %s is not loaded!\n",
5005 spin_unlock(&pers_lock);
5006 if (mddev->level != pers->level) {
5007 mddev->level = pers->level;
5008 mddev->new_level = pers->level;
5010 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
5012 if (mddev->reshape_position != MaxSector &&
5013 pers->start_reshape == NULL) {
5014 /* This personality cannot handle reshaping... */
5016 module_put(pers->owner);
5020 if (pers->sync_request) {
5021 /* Warn if this is a potentially silly
5024 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
5025 struct md_rdev *rdev2;
5028 rdev_for_each(rdev, mddev)
5029 rdev_for_each(rdev2, mddev) {
5031 rdev->bdev->bd_contains ==
5032 rdev2->bdev->bd_contains) {
5034 "%s: WARNING: %s appears to be"
5035 " on the same physical disk as"
5038 bdevname(rdev->bdev,b),
5039 bdevname(rdev2->bdev,b2));
5046 "True protection against single-disk"
5047 " failure might be compromised.\n");
5050 mddev->recovery = 0;
5051 /* may be over-ridden by personality */
5052 mddev->resync_max_sectors = mddev->dev_sectors;
5054 mddev->ok_start_degraded = start_dirty_degraded;
5056 if (start_readonly && mddev->ro == 0)
5057 mddev->ro = 2; /* read-only, but switch on first write */
5059 err = mddev->pers->run(mddev);
5061 printk(KERN_ERR "md: pers->run() failed ...\n");
5062 else if (mddev->pers->size(mddev, 0, 0) < mddev->array_sectors) {
5063 WARN_ONCE(!mddev->external_size, "%s: default size too small,"
5064 " but 'external_size' not in effect?\n", __func__);
5066 "md: invalid array_size %llu > default size %llu\n",
5067 (unsigned long long)mddev->array_sectors / 2,
5068 (unsigned long long)mddev->pers->size(mddev, 0, 0) / 2);
5070 mddev->pers->stop(mddev);
5072 if (err == 0 && mddev->pers->sync_request &&
5073 (mddev->bitmap_info.file || mddev->bitmap_info.offset)) {
5074 err = bitmap_create(mddev);
5076 printk(KERN_ERR "%s: failed to create bitmap (%d)\n",
5077 mdname(mddev), err);
5078 mddev->pers->stop(mddev);
5082 module_put(mddev->pers->owner);
5084 bitmap_destroy(mddev);
5087 if (mddev->pers->sync_request) {
5088 if (mddev->kobj.sd &&
5089 sysfs_create_group(&mddev->kobj, &md_redundancy_group))
5091 "md: cannot register extra attributes for %s\n",
5093 mddev->sysfs_action = sysfs_get_dirent_safe(mddev->kobj.sd, "sync_action");
5094 } else if (mddev->ro == 2) /* auto-readonly not meaningful */
5097 atomic_set(&mddev->writes_pending,0);
5098 atomic_set(&mddev->max_corr_read_errors,
5099 MD_DEFAULT_MAX_CORRECTED_READ_ERRORS);
5100 mddev->safemode = 0;
5101 mddev->safemode_timer.function = md_safemode_timeout;
5102 mddev->safemode_timer.data = (unsigned long) mddev;
5103 mddev->safemode_delay = (200 * HZ)/1000 +1; /* 200 msec delay */
5107 rdev_for_each(rdev, mddev)
5108 if (rdev->raid_disk >= 0)
5109 if (sysfs_link_rdev(mddev, rdev))
5110 /* failure here is OK */;
5112 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5114 if (mddev->flags & MD_UPDATE_SB_FLAGS)
5115 md_update_sb(mddev, 0);
5117 md_new_event(mddev);
5118 sysfs_notify_dirent_safe(mddev->sysfs_state);
5119 sysfs_notify_dirent_safe(mddev->sysfs_action);
5120 sysfs_notify(&mddev->kobj, NULL, "degraded");
5123 EXPORT_SYMBOL_GPL(md_run);
5125 static int do_md_run(struct mddev *mddev)
5129 err = md_run(mddev);
5132 err = bitmap_load(mddev);
5134 bitmap_destroy(mddev);
5138 md_wakeup_thread(mddev->thread);
5139 md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
5141 set_capacity(mddev->gendisk, mddev->array_sectors);
5142 revalidate_disk(mddev->gendisk);
5144 kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
5149 static int restart_array(struct mddev *mddev)
5151 struct gendisk *disk = mddev->gendisk;
5153 /* Complain if it has no devices */
5154 if (list_empty(&mddev->disks))
5160 mddev->safemode = 0;
5162 set_disk_ro(disk, 0);
5163 printk(KERN_INFO "md: %s switched to read-write mode.\n",
5165 /* Kick recovery or resync if necessary */
5166 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5167 md_wakeup_thread(mddev->thread);
5168 md_wakeup_thread(mddev->sync_thread);
5169 sysfs_notify_dirent_safe(mddev->sysfs_state);
5173 static void md_clean(struct mddev *mddev)
5175 mddev->array_sectors = 0;
5176 mddev->external_size = 0;
5177 mddev->dev_sectors = 0;
5178 mddev->raid_disks = 0;
5179 mddev->recovery_cp = 0;
5180 mddev->resync_min = 0;
5181 mddev->resync_max = MaxSector;
5182 mddev->reshape_position = MaxSector;
5183 mddev->external = 0;
5184 mddev->persistent = 0;
5185 mddev->level = LEVEL_NONE;
5186 mddev->clevel[0] = 0;
5189 mddev->metadata_type[0] = 0;
5190 mddev->chunk_sectors = 0;
5191 mddev->ctime = mddev->utime = 0;
5193 mddev->max_disks = 0;
5195 mddev->can_decrease_events = 0;
5196 mddev->delta_disks = 0;
5197 mddev->reshape_backwards = 0;
5198 mddev->new_level = LEVEL_NONE;
5199 mddev->new_layout = 0;
5200 mddev->new_chunk_sectors = 0;
5201 mddev->curr_resync = 0;
5202 atomic64_set(&mddev->resync_mismatches, 0);
5203 mddev->suspend_lo = mddev->suspend_hi = 0;
5204 mddev->sync_speed_min = mddev->sync_speed_max = 0;
5205 mddev->recovery = 0;
5208 mddev->degraded = 0;
5209 mddev->safemode = 0;
5210 mddev->merge_check_needed = 0;
5211 mddev->bitmap_info.offset = 0;
5212 mddev->bitmap_info.default_offset = 0;
5213 mddev->bitmap_info.default_space = 0;
5214 mddev->bitmap_info.chunksize = 0;
5215 mddev->bitmap_info.daemon_sleep = 0;
5216 mddev->bitmap_info.max_write_behind = 0;
5219 static void __md_stop_writes(struct mddev *mddev)
5221 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5222 if (mddev->sync_thread) {
5223 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5224 md_reap_sync_thread(mddev);
5227 del_timer_sync(&mddev->safemode_timer);
5229 bitmap_flush(mddev);
5230 md_super_wait(mddev);
5232 if (mddev->ro == 0 &&
5233 (!mddev->in_sync || (mddev->flags & MD_UPDATE_SB_FLAGS))) {
5234 /* mark array as shutdown cleanly */
5236 md_update_sb(mddev, 1);
5240 void md_stop_writes(struct mddev *mddev)
5242 mddev_lock_nointr(mddev);
5243 __md_stop_writes(mddev);
5244 mddev_unlock(mddev);
5246 EXPORT_SYMBOL_GPL(md_stop_writes);
5248 static void __md_stop(struct mddev *mddev)
5251 mddev->pers->stop(mddev);
5252 if (mddev->pers->sync_request && mddev->to_remove == NULL)
5253 mddev->to_remove = &md_redundancy_group;
5254 module_put(mddev->pers->owner);
5256 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5259 void md_stop(struct mddev *mddev)
5261 /* stop the array and free an attached data structures.
5262 * This is called from dm-raid
5265 bitmap_destroy(mddev);
5267 bioset_free(mddev->bio_set);
5270 EXPORT_SYMBOL_GPL(md_stop);
5272 static int md_set_readonly(struct mddev *mddev, struct block_device *bdev)
5277 if (!test_bit(MD_RECOVERY_FROZEN, &mddev->recovery)) {
5279 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5280 md_wakeup_thread(mddev->thread);
5282 if (mddev->sync_thread) {
5283 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5284 /* Thread might be blocked waiting for metadata update
5285 * which will now never happen */
5286 wake_up_process(mddev->sync_thread->tsk);
5288 mddev_unlock(mddev);
5289 wait_event(resync_wait, mddev->sync_thread == NULL);
5290 mddev_lock_nointr(mddev);
5292 mutex_lock(&mddev->open_mutex);
5293 if ((mddev->pers && atomic_read(&mddev->openers) > !!bdev) ||
5294 mddev->sync_thread ||
5295 (bdev && !test_bit(MD_STILL_CLOSED, &mddev->flags))) {
5296 printk("md: %s still in use.\n",mdname(mddev));
5298 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5299 md_wakeup_thread(mddev->thread);
5305 __md_stop_writes(mddev);
5311 set_disk_ro(mddev->gendisk, 1);
5312 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5313 sysfs_notify_dirent_safe(mddev->sysfs_state);
5317 mutex_unlock(&mddev->open_mutex);
5322 * 0 - completely stop and dis-assemble array
5323 * 2 - stop but do not disassemble array
5325 static int do_md_stop(struct mddev * mddev, int mode,
5326 struct block_device *bdev)
5328 struct gendisk *disk = mddev->gendisk;
5329 struct md_rdev *rdev;
5332 if (!test_bit(MD_RECOVERY_FROZEN, &mddev->recovery)) {
5334 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5335 md_wakeup_thread(mddev->thread);
5337 if (mddev->sync_thread) {
5338 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5339 /* Thread might be blocked waiting for metadata update
5340 * which will now never happen */
5341 wake_up_process(mddev->sync_thread->tsk);
5343 mddev_unlock(mddev);
5344 wait_event(resync_wait, mddev->sync_thread == NULL);
5345 mddev_lock_nointr(mddev);
5347 mutex_lock(&mddev->open_mutex);
5348 if ((mddev->pers && atomic_read(&mddev->openers) > !!bdev) ||
5349 mddev->sysfs_active ||
5350 mddev->sync_thread ||
5351 (bdev && !test_bit(MD_STILL_CLOSED, &mddev->flags))) {
5352 printk("md: %s still in use.\n",mdname(mddev));
5353 mutex_unlock(&mddev->open_mutex);
5355 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5356 md_wakeup_thread(mddev->thread);
5362 set_disk_ro(disk, 0);
5364 __md_stop_writes(mddev);
5366 mddev->queue->merge_bvec_fn = NULL;
5367 mddev->queue->backing_dev_info.congested_fn = NULL;
5369 /* tell userspace to handle 'inactive' */
5370 sysfs_notify_dirent_safe(mddev->sysfs_state);
5372 rdev_for_each(rdev, mddev)
5373 if (rdev->raid_disk >= 0)
5374 sysfs_unlink_rdev(mddev, rdev);
5376 set_capacity(disk, 0);
5377 mutex_unlock(&mddev->open_mutex);
5379 revalidate_disk(disk);
5384 mutex_unlock(&mddev->open_mutex);
5386 * Free resources if final stop
5389 printk(KERN_INFO "md: %s stopped.\n", mdname(mddev));
5391 bitmap_destroy(mddev);
5392 if (mddev->bitmap_info.file) {
5393 fput(mddev->bitmap_info.file);
5394 mddev->bitmap_info.file = NULL;
5396 mddev->bitmap_info.offset = 0;
5398 export_array(mddev);
5401 kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
5402 if (mddev->hold_active == UNTIL_STOP)
5403 mddev->hold_active = 0;
5405 blk_integrity_unregister(disk);
5406 md_new_event(mddev);
5407 sysfs_notify_dirent_safe(mddev->sysfs_state);
5412 static void autorun_array(struct mddev *mddev)
5414 struct md_rdev *rdev;
5417 if (list_empty(&mddev->disks))
5420 printk(KERN_INFO "md: running: ");
5422 rdev_for_each(rdev, mddev) {
5423 char b[BDEVNAME_SIZE];
5424 printk("<%s>", bdevname(rdev->bdev,b));
5428 err = do_md_run(mddev);
5430 printk(KERN_WARNING "md: do_md_run() returned %d\n", err);
5431 do_md_stop(mddev, 0, NULL);
5436 * lets try to run arrays based on all disks that have arrived
5437 * until now. (those are in pending_raid_disks)
5439 * the method: pick the first pending disk, collect all disks with
5440 * the same UUID, remove all from the pending list and put them into
5441 * the 'same_array' list. Then order this list based on superblock
5442 * update time (freshest comes first), kick out 'old' disks and
5443 * compare superblocks. If everything's fine then run it.
5445 * If "unit" is allocated, then bump its reference count
5447 static void autorun_devices(int part)
5449 struct md_rdev *rdev0, *rdev, *tmp;
5450 struct mddev *mddev;
5451 char b[BDEVNAME_SIZE];
5453 printk(KERN_INFO "md: autorun ...\n");
5454 while (!list_empty(&pending_raid_disks)) {
5457 LIST_HEAD(candidates);
5458 rdev0 = list_entry(pending_raid_disks.next,
5459 struct md_rdev, same_set);
5461 printk(KERN_INFO "md: considering %s ...\n",
5462 bdevname(rdev0->bdev,b));
5463 INIT_LIST_HEAD(&candidates);
5464 rdev_for_each_list(rdev, tmp, &pending_raid_disks)
5465 if (super_90_load(rdev, rdev0, 0) >= 0) {
5466 printk(KERN_INFO "md: adding %s ...\n",
5467 bdevname(rdev->bdev,b));
5468 list_move(&rdev->same_set, &candidates);
5471 * now we have a set of devices, with all of them having
5472 * mostly sane superblocks. It's time to allocate the
5476 dev = MKDEV(mdp_major,
5477 rdev0->preferred_minor << MdpMinorShift);
5478 unit = MINOR(dev) >> MdpMinorShift;
5480 dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
5483 if (rdev0->preferred_minor != unit) {
5484 printk(KERN_INFO "md: unit number in %s is bad: %d\n",
5485 bdevname(rdev0->bdev, b), rdev0->preferred_minor);
5489 md_probe(dev, NULL, NULL);
5490 mddev = mddev_find(dev);
5491 if (!mddev || !mddev->gendisk) {
5495 "md: cannot allocate memory for md drive.\n");
5498 if (mddev_lock(mddev))
5499 printk(KERN_WARNING "md: %s locked, cannot run\n",
5501 else if (mddev->raid_disks || mddev->major_version
5502 || !list_empty(&mddev->disks)) {
5504 "md: %s already running, cannot run %s\n",
5505 mdname(mddev), bdevname(rdev0->bdev,b));
5506 mddev_unlock(mddev);
5508 printk(KERN_INFO "md: created %s\n", mdname(mddev));
5509 mddev->persistent = 1;
5510 rdev_for_each_list(rdev, tmp, &candidates) {
5511 list_del_init(&rdev->same_set);
5512 if (bind_rdev_to_array(rdev, mddev))
5515 autorun_array(mddev);
5516 mddev_unlock(mddev);
5518 /* on success, candidates will be empty, on error
5521 rdev_for_each_list(rdev, tmp, &candidates) {
5522 list_del_init(&rdev->same_set);
5527 printk(KERN_INFO "md: ... autorun DONE.\n");
5529 #endif /* !MODULE */
5531 static int get_version(void __user * arg)
5535 ver.major = MD_MAJOR_VERSION;
5536 ver.minor = MD_MINOR_VERSION;
5537 ver.patchlevel = MD_PATCHLEVEL_VERSION;
5539 if (copy_to_user(arg, &ver, sizeof(ver)))
5545 static int get_array_info(struct mddev * mddev, void __user * arg)
5547 mdu_array_info_t info;
5548 int nr,working,insync,failed,spare;
5549 struct md_rdev *rdev;
5551 nr = working = insync = failed = spare = 0;
5553 rdev_for_each_rcu(rdev, mddev) {
5555 if (test_bit(Faulty, &rdev->flags))
5559 if (test_bit(In_sync, &rdev->flags))
5567 info.major_version = mddev->major_version;
5568 info.minor_version = mddev->minor_version;
5569 info.patch_version = MD_PATCHLEVEL_VERSION;
5570 info.ctime = mddev->ctime;
5571 info.level = mddev->level;
5572 info.size = mddev->dev_sectors / 2;
5573 if (info.size != mddev->dev_sectors / 2) /* overflow */
5576 info.raid_disks = mddev->raid_disks;
5577 info.md_minor = mddev->md_minor;
5578 info.not_persistent= !mddev->persistent;
5580 info.utime = mddev->utime;
5583 info.state = (1<<MD_SB_CLEAN);
5584 if (mddev->bitmap && mddev->bitmap_info.offset)
5585 info.state |= (1<<MD_SB_BITMAP_PRESENT);
5586 info.active_disks = insync;
5587 info.working_disks = working;
5588 info.failed_disks = failed;
5589 info.spare_disks = spare;
5591 info.layout = mddev->layout;
5592 info.chunk_size = mddev->chunk_sectors << 9;
5594 if (copy_to_user(arg, &info, sizeof(info)))
5600 static int get_bitmap_file(struct mddev * mddev, void __user * arg)
5602 mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */
5603 char *ptr, *buf = NULL;
5606 file = kmalloc(sizeof(*file), GFP_NOIO);
5611 /* bitmap disabled, zero the first byte and copy out */
5612 if (!mddev->bitmap || !mddev->bitmap->storage.file) {
5613 file->pathname[0] = '\0';
5617 buf = kmalloc(sizeof(file->pathname), GFP_KERNEL);
5621 ptr = d_path(&mddev->bitmap->storage.file->f_path,
5622 buf, sizeof(file->pathname));
5626 strcpy(file->pathname, ptr);
5630 if (copy_to_user(arg, file, sizeof(*file)))
5638 static int get_disk_info(struct mddev * mddev, void __user * arg)
5640 mdu_disk_info_t info;
5641 struct md_rdev *rdev;
5643 if (copy_from_user(&info, arg, sizeof(info)))
5647 rdev = find_rdev_nr_rcu(mddev, info.number);
5649 info.major = MAJOR(rdev->bdev->bd_dev);
5650 info.minor = MINOR(rdev->bdev->bd_dev);
5651 info.raid_disk = rdev->raid_disk;
5653 if (test_bit(Faulty, &rdev->flags))
5654 info.state |= (1<<MD_DISK_FAULTY);
5655 else if (test_bit(In_sync, &rdev->flags)) {
5656 info.state |= (1<<MD_DISK_ACTIVE);
5657 info.state |= (1<<MD_DISK_SYNC);
5659 if (test_bit(WriteMostly, &rdev->flags))
5660 info.state |= (1<<MD_DISK_WRITEMOSTLY);
5662 info.major = info.minor = 0;
5663 info.raid_disk = -1;
5664 info.state = (1<<MD_DISK_REMOVED);
5668 if (copy_to_user(arg, &info, sizeof(info)))
5674 static int add_new_disk(struct mddev * mddev, mdu_disk_info_t *info)
5676 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
5677 struct md_rdev *rdev;
5678 dev_t dev = MKDEV(info->major,info->minor);
5680 if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
5683 if (!mddev->raid_disks) {
5685 /* expecting a device which has a superblock */
5686 rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
5689 "md: md_import_device returned %ld\n",
5691 return PTR_ERR(rdev);
5693 if (!list_empty(&mddev->disks)) {
5694 struct md_rdev *rdev0
5695 = list_entry(mddev->disks.next,
5696 struct md_rdev, same_set);
5697 err = super_types[mddev->major_version]
5698 .load_super(rdev, rdev0, mddev->minor_version);
5701 "md: %s has different UUID to %s\n",
5702 bdevname(rdev->bdev,b),
5703 bdevname(rdev0->bdev,b2));
5708 err = bind_rdev_to_array(rdev, mddev);
5715 * add_new_disk can be used once the array is assembled
5716 * to add "hot spares". They must already have a superblock
5721 if (!mddev->pers->hot_add_disk) {
5723 "%s: personality does not support diskops!\n",
5727 if (mddev->persistent)
5728 rdev = md_import_device(dev, mddev->major_version,
5729 mddev->minor_version);
5731 rdev = md_import_device(dev, -1, -1);
5734 "md: md_import_device returned %ld\n",
5736 return PTR_ERR(rdev);
5738 /* set saved_raid_disk if appropriate */
5739 if (!mddev->persistent) {
5740 if (info->state & (1<<MD_DISK_SYNC) &&
5741 info->raid_disk < mddev->raid_disks) {
5742 rdev->raid_disk = info->raid_disk;
5743 set_bit(In_sync, &rdev->flags);
5744 clear_bit(Bitmap_sync, &rdev->flags);
5746 rdev->raid_disk = -1;
5747 rdev->saved_raid_disk = rdev->raid_disk;
5749 super_types[mddev->major_version].
5750 validate_super(mddev, rdev);
5751 if ((info->state & (1<<MD_DISK_SYNC)) &&
5752 rdev->raid_disk != info->raid_disk) {
5753 /* This was a hot-add request, but events doesn't
5754 * match, so reject it.
5760 clear_bit(In_sync, &rdev->flags); /* just to be sure */
5761 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
5762 set_bit(WriteMostly, &rdev->flags);
5764 clear_bit(WriteMostly, &rdev->flags);
5766 rdev->raid_disk = -1;
5767 err = bind_rdev_to_array(rdev, mddev);
5768 if (!err && !mddev->pers->hot_remove_disk) {
5769 /* If there is hot_add_disk but no hot_remove_disk
5770 * then added disks for geometry changes,
5771 * and should be added immediately.
5773 super_types[mddev->major_version].
5774 validate_super(mddev, rdev);
5775 err = mddev->pers->hot_add_disk(mddev, rdev);
5777 unbind_rdev_from_array(rdev);
5782 sysfs_notify_dirent_safe(rdev->sysfs_state);
5784 set_bit(MD_CHANGE_DEVS, &mddev->flags);
5785 if (mddev->degraded)
5786 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
5787 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5789 md_new_event(mddev);
5790 md_wakeup_thread(mddev->thread);
5794 /* otherwise, add_new_disk is only allowed
5795 * for major_version==0 superblocks
5797 if (mddev->major_version != 0) {
5798 printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n",
5803 if (!(info->state & (1<<MD_DISK_FAULTY))) {
5805 rdev = md_import_device(dev, -1, 0);
5808 "md: error, md_import_device() returned %ld\n",
5810 return PTR_ERR(rdev);
5812 rdev->desc_nr = info->number;
5813 if (info->raid_disk < mddev->raid_disks)
5814 rdev->raid_disk = info->raid_disk;
5816 rdev->raid_disk = -1;
5818 if (rdev->raid_disk < mddev->raid_disks)
5819 if (info->state & (1<<MD_DISK_SYNC))
5820 set_bit(In_sync, &rdev->flags);
5822 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
5823 set_bit(WriteMostly, &rdev->flags);
5825 if (!mddev->persistent) {
5826 printk(KERN_INFO "md: nonpersistent superblock ...\n");
5827 rdev->sb_start = i_size_read(rdev->bdev->bd_inode) / 512;
5829 rdev->sb_start = calc_dev_sboffset(rdev);
5830 rdev->sectors = rdev->sb_start;
5832 err = bind_rdev_to_array(rdev, mddev);
5842 static int hot_remove_disk(struct mddev * mddev, dev_t dev)
5844 char b[BDEVNAME_SIZE];
5845 struct md_rdev *rdev;
5847 rdev = find_rdev(mddev, dev);
5851 clear_bit(Blocked, &rdev->flags);
5852 remove_and_add_spares(mddev, rdev);
5854 if (rdev->raid_disk >= 0)
5857 kick_rdev_from_array(rdev);
5858 md_update_sb(mddev, 1);
5859 md_new_event(mddev);
5863 printk(KERN_WARNING "md: cannot remove active disk %s from %s ...\n",
5864 bdevname(rdev->bdev,b), mdname(mddev));
5868 static int hot_add_disk(struct mddev * mddev, dev_t dev)
5870 char b[BDEVNAME_SIZE];
5872 struct md_rdev *rdev;
5877 if (mddev->major_version != 0) {
5878 printk(KERN_WARNING "%s: HOT_ADD may only be used with"
5879 " version-0 superblocks.\n",
5883 if (!mddev->pers->hot_add_disk) {
5885 "%s: personality does not support diskops!\n",
5890 rdev = md_import_device(dev, -1, 0);
5893 "md: error, md_import_device() returned %ld\n",
5898 if (mddev->persistent)
5899 rdev->sb_start = calc_dev_sboffset(rdev);
5901 rdev->sb_start = i_size_read(rdev->bdev->bd_inode) / 512;
5903 rdev->sectors = rdev->sb_start;
5905 if (test_bit(Faulty, &rdev->flags)) {
5907 "md: can not hot-add faulty %s disk to %s!\n",
5908 bdevname(rdev->bdev,b), mdname(mddev));
5912 clear_bit(In_sync, &rdev->flags);
5914 rdev->saved_raid_disk = -1;
5915 err = bind_rdev_to_array(rdev, mddev);
5920 * The rest should better be atomic, we can have disk failures
5921 * noticed in interrupt contexts ...
5924 rdev->raid_disk = -1;
5926 md_update_sb(mddev, 1);
5929 * Kick recovery, maybe this spare has to be added to the
5930 * array immediately.
5932 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5933 md_wakeup_thread(mddev->thread);
5934 md_new_event(mddev);
5942 static int set_bitmap_file(struct mddev *mddev, int fd)
5947 if (!mddev->pers->quiesce || !mddev->thread)
5949 if (mddev->recovery || mddev->sync_thread)
5951 /* we should be able to change the bitmap.. */
5956 struct inode *inode;
5958 return -EEXIST; /* cannot add when bitmap is present */
5959 mddev->bitmap_info.file = fget(fd);
5961 if (mddev->bitmap_info.file == NULL) {
5962 printk(KERN_ERR "%s: error: failed to get bitmap file\n",
5967 inode = mddev->bitmap_info.file->f_mapping->host;
5968 if (!S_ISREG(inode->i_mode)) {
5969 printk(KERN_ERR "%s: error: bitmap file must be a regular file\n",
5972 } else if (!(mddev->bitmap_info.file->f_mode & FMODE_WRITE)) {
5973 printk(KERN_ERR "%s: error: bitmap file must open for write\n",
5976 } else if (atomic_read(&inode->i_writecount) != 1) {
5977 printk(KERN_ERR "%s: error: bitmap file is already in use\n",
5982 fput(mddev->bitmap_info.file);
5983 mddev->bitmap_info.file = NULL;
5986 mddev->bitmap_info.offset = 0; /* file overrides offset */
5987 } else if (mddev->bitmap == NULL)
5988 return -ENOENT; /* cannot remove what isn't there */
5991 mddev->pers->quiesce(mddev, 1);
5993 err = bitmap_create(mddev);
5995 err = bitmap_load(mddev);
5997 if (fd < 0 || err) {
5998 bitmap_destroy(mddev);
5999 fd = -1; /* make sure to put the file */
6001 mddev->pers->quiesce(mddev, 0);
6004 if (mddev->bitmap_info.file)
6005 fput(mddev->bitmap_info.file);
6006 mddev->bitmap_info.file = NULL;
6013 * set_array_info is used two different ways
6014 * The original usage is when creating a new array.
6015 * In this usage, raid_disks is > 0 and it together with
6016 * level, size, not_persistent,layout,chunksize determine the
6017 * shape of the array.
6018 * This will always create an array with a type-0.90.0 superblock.
6019 * The newer usage is when assembling an array.
6020 * In this case raid_disks will be 0, and the major_version field is
6021 * use to determine which style super-blocks are to be found on the devices.
6022 * The minor and patch _version numbers are also kept incase the
6023 * super_block handler wishes to interpret them.
6025 static int set_array_info(struct mddev * mddev, mdu_array_info_t *info)
6028 if (info->raid_disks == 0) {
6029 /* just setting version number for superblock loading */
6030 if (info->major_version < 0 ||
6031 info->major_version >= ARRAY_SIZE(super_types) ||
6032 super_types[info->major_version].name == NULL) {
6033 /* maybe try to auto-load a module? */
6035 "md: superblock version %d not known\n",
6036 info->major_version);
6039 mddev->major_version = info->major_version;
6040 mddev->minor_version = info->minor_version;
6041 mddev->patch_version = info->patch_version;
6042 mddev->persistent = !info->not_persistent;
6043 /* ensure mddev_put doesn't delete this now that there
6044 * is some minimal configuration.
6046 mddev->ctime = get_seconds();
6049 mddev->major_version = MD_MAJOR_VERSION;
6050 mddev->minor_version = MD_MINOR_VERSION;
6051 mddev->patch_version = MD_PATCHLEVEL_VERSION;
6052 mddev->ctime = get_seconds();
6054 mddev->level = info->level;
6055 mddev->clevel[0] = 0;
6056 mddev->dev_sectors = 2 * (sector_t)info->size;
6057 mddev->raid_disks = info->raid_disks;
6058 /* don't set md_minor, it is determined by which /dev/md* was
6061 if (info->state & (1<<MD_SB_CLEAN))
6062 mddev->recovery_cp = MaxSector;
6064 mddev->recovery_cp = 0;
6065 mddev->persistent = ! info->not_persistent;
6066 mddev->external = 0;
6068 mddev->layout = info->layout;
6069 mddev->chunk_sectors = info->chunk_size >> 9;
6071 mddev->max_disks = MD_SB_DISKS;
6073 if (mddev->persistent)
6075 set_bit(MD_CHANGE_DEVS, &mddev->flags);
6077 mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9;
6078 mddev->bitmap_info.default_space = 64*2 - (MD_SB_BYTES >> 9);
6079 mddev->bitmap_info.offset = 0;
6081 mddev->reshape_position = MaxSector;
6084 * Generate a 128 bit UUID
6086 get_random_bytes(mddev->uuid, 16);
6088 mddev->new_level = mddev->level;
6089 mddev->new_chunk_sectors = mddev->chunk_sectors;
6090 mddev->new_layout = mddev->layout;
6091 mddev->delta_disks = 0;
6092 mddev->reshape_backwards = 0;
6097 void md_set_array_sectors(struct mddev *mddev, sector_t array_sectors)
6099 WARN(!mddev_is_locked(mddev), "%s: unlocked mddev!\n", __func__);
6101 if (mddev->external_size)
6104 mddev->array_sectors = array_sectors;
6106 EXPORT_SYMBOL(md_set_array_sectors);
6108 static int update_size(struct mddev *mddev, sector_t num_sectors)
6110 struct md_rdev *rdev;
6112 int fit = (num_sectors == 0);
6114 if (mddev->pers->resize == NULL)
6116 /* The "num_sectors" is the number of sectors of each device that
6117 * is used. This can only make sense for arrays with redundancy.
6118 * linear and raid0 always use whatever space is available. We can only
6119 * consider changing this number if no resync or reconstruction is
6120 * happening, and if the new size is acceptable. It must fit before the
6121 * sb_start or, if that is <data_offset, it must fit before the size
6122 * of each device. If num_sectors is zero, we find the largest size
6125 if (mddev->sync_thread)
6130 rdev_for_each(rdev, mddev) {
6131 sector_t avail = rdev->sectors;
6133 if (fit && (num_sectors == 0 || num_sectors > avail))
6134 num_sectors = avail;
6135 if (avail < num_sectors)
6138 rv = mddev->pers->resize(mddev, num_sectors);
6140 revalidate_disk(mddev->gendisk);
6144 static int update_raid_disks(struct mddev *mddev, int raid_disks)
6147 struct md_rdev *rdev;
6148 /* change the number of raid disks */
6149 if (mddev->pers->check_reshape == NULL)
6153 if (raid_disks <= 0 ||
6154 (mddev->max_disks && raid_disks >= mddev->max_disks))
6156 if (mddev->sync_thread || mddev->reshape_position != MaxSector)
6159 rdev_for_each(rdev, mddev) {
6160 if (mddev->raid_disks < raid_disks &&
6161 rdev->data_offset < rdev->new_data_offset)
6163 if (mddev->raid_disks > raid_disks &&
6164 rdev->data_offset > rdev->new_data_offset)
6168 mddev->delta_disks = raid_disks - mddev->raid_disks;
6169 if (mddev->delta_disks < 0)
6170 mddev->reshape_backwards = 1;
6171 else if (mddev->delta_disks > 0)
6172 mddev->reshape_backwards = 0;
6174 rv = mddev->pers->check_reshape(mddev);
6176 mddev->delta_disks = 0;
6177 mddev->reshape_backwards = 0;
6184 * update_array_info is used to change the configuration of an
6186 * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
6187 * fields in the info are checked against the array.
6188 * Any differences that cannot be handled will cause an error.
6189 * Normally, only one change can be managed at a time.
6191 static int update_array_info(struct mddev *mddev, mdu_array_info_t *info)
6197 /* calculate expected state,ignoring low bits */
6198 if (mddev->bitmap && mddev->bitmap_info.offset)
6199 state |= (1 << MD_SB_BITMAP_PRESENT);
6201 if (mddev->major_version != info->major_version ||
6202 mddev->minor_version != info->minor_version ||
6203 /* mddev->patch_version != info->patch_version || */
6204 mddev->ctime != info->ctime ||
6205 mddev->level != info->level ||
6206 /* mddev->layout != info->layout || */
6207 !mddev->persistent != info->not_persistent||
6208 mddev->chunk_sectors != info->chunk_size >> 9 ||
6209 /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
6210 ((state^info->state) & 0xfffffe00)
6213 /* Check there is only one change */
6214 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
6216 if (mddev->raid_disks != info->raid_disks)
6218 if (mddev->layout != info->layout)
6220 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT))
6227 if (mddev->layout != info->layout) {
6229 * we don't need to do anything at the md level, the
6230 * personality will take care of it all.
6232 if (mddev->pers->check_reshape == NULL)
6235 mddev->new_layout = info->layout;
6236 rv = mddev->pers->check_reshape(mddev);
6238 mddev->new_layout = mddev->layout;
6242 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
6243 rv = update_size(mddev, (sector_t)info->size * 2);
6245 if (mddev->raid_disks != info->raid_disks)
6246 rv = update_raid_disks(mddev, info->raid_disks);
6248 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) {
6249 if (mddev->pers->quiesce == NULL || mddev->thread == NULL)
6251 if (mddev->recovery || mddev->sync_thread)
6253 if (info->state & (1<<MD_SB_BITMAP_PRESENT)) {
6254 /* add the bitmap */
6257 if (mddev->bitmap_info.default_offset == 0)
6259 mddev->bitmap_info.offset =
6260 mddev->bitmap_info.default_offset;
6261 mddev->bitmap_info.space =
6262 mddev->bitmap_info.default_space;
6263 mddev->pers->quiesce(mddev, 1);
6264 rv = bitmap_create(mddev);
6266 rv = bitmap_load(mddev);
6268 bitmap_destroy(mddev);
6269 mddev->pers->quiesce(mddev, 0);
6271 /* remove the bitmap */
6274 if (mddev->bitmap->storage.file)
6276 mddev->pers->quiesce(mddev, 1);
6277 bitmap_destroy(mddev);
6278 mddev->pers->quiesce(mddev, 0);
6279 mddev->bitmap_info.offset = 0;
6282 md_update_sb(mddev, 1);
6286 static int set_disk_faulty(struct mddev *mddev, dev_t dev)
6288 struct md_rdev *rdev;
6291 if (mddev->pers == NULL)
6295 rdev = find_rdev_rcu(mddev, dev);
6299 md_error(mddev, rdev);
6300 if (!test_bit(Faulty, &rdev->flags))
6308 * We have a problem here : there is no easy way to give a CHS
6309 * virtual geometry. We currently pretend that we have a 2 heads
6310 * 4 sectors (with a BIG number of cylinders...). This drives
6311 * dosfs just mad... ;-)
6313 static int md_getgeo(struct block_device *bdev, struct hd_geometry *geo)
6315 struct mddev *mddev = bdev->bd_disk->private_data;
6319 geo->cylinders = mddev->array_sectors / 8;
6323 static inline bool md_ioctl_valid(unsigned int cmd)
6328 case GET_ARRAY_INFO:
6329 case GET_BITMAP_FILE:
6332 case HOT_REMOVE_DISK:
6333 case PRINT_RAID_DEBUG:
6336 case RESTART_ARRAY_RW:
6338 case SET_ARRAY_INFO:
6339 case SET_BITMAP_FILE:
6340 case SET_DISK_FAULTY:
6349 static int md_ioctl(struct block_device *bdev, fmode_t mode,
6350 unsigned int cmd, unsigned long arg)
6353 void __user *argp = (void __user *)arg;
6354 struct mddev *mddev = NULL;
6357 if (!md_ioctl_valid(cmd))
6362 case GET_ARRAY_INFO:
6366 if (!capable(CAP_SYS_ADMIN))
6371 * Commands dealing with the RAID driver but not any
6376 err = get_version(argp);
6379 case PRINT_RAID_DEBUG:
6387 autostart_arrays(arg);
6394 * Commands creating/starting a new array:
6397 mddev = bdev->bd_disk->private_data;
6404 /* Some actions do not requires the mutex */
6406 case GET_ARRAY_INFO:
6407 if (!mddev->raid_disks && !mddev->external)
6410 err = get_array_info(mddev, argp);
6414 if (!mddev->raid_disks && !mddev->external)
6417 err = get_disk_info(mddev, argp);
6420 case SET_DISK_FAULTY:
6421 err = set_disk_faulty(mddev, new_decode_dev(arg));
6425 if (cmd == ADD_NEW_DISK)
6426 /* need to ensure md_delayed_delete() has completed */
6427 flush_workqueue(md_misc_wq);
6429 if (cmd == HOT_REMOVE_DISK)
6430 /* need to ensure recovery thread has run */
6431 wait_event_interruptible_timeout(mddev->sb_wait,
6432 !test_bit(MD_RECOVERY_NEEDED,
6434 msecs_to_jiffies(5000));
6435 if (cmd == STOP_ARRAY || cmd == STOP_ARRAY_RO) {
6436 /* Need to flush page cache, and ensure no-one else opens
6439 mutex_lock(&mddev->open_mutex);
6440 if (mddev->pers && atomic_read(&mddev->openers) > 1) {
6441 mutex_unlock(&mddev->open_mutex);
6445 set_bit(MD_STILL_CLOSED, &mddev->flags);
6446 mutex_unlock(&mddev->open_mutex);
6447 sync_blockdev(bdev);
6449 err = mddev_lock(mddev);
6452 "md: ioctl lock interrupted, reason %d, cmd %d\n",
6457 if (cmd == SET_ARRAY_INFO) {
6458 mdu_array_info_t info;
6460 memset(&info, 0, sizeof(info));
6461 else if (copy_from_user(&info, argp, sizeof(info))) {
6466 err = update_array_info(mddev, &info);
6468 printk(KERN_WARNING "md: couldn't update"
6469 " array info. %d\n", err);
6474 if (!list_empty(&mddev->disks)) {
6476 "md: array %s already has disks!\n",
6481 if (mddev->raid_disks) {
6483 "md: array %s already initialised!\n",
6488 err = set_array_info(mddev, &info);
6490 printk(KERN_WARNING "md: couldn't set"
6491 " array info. %d\n", err);
6498 * Commands querying/configuring an existing array:
6500 /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
6501 * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */
6502 if ((!mddev->raid_disks && !mddev->external)
6503 && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY
6504 && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE
6505 && cmd != GET_BITMAP_FILE) {
6511 * Commands even a read-only array can execute:
6514 case GET_BITMAP_FILE:
6515 err = get_bitmap_file(mddev, argp);
6518 case RESTART_ARRAY_RW:
6519 err = restart_array(mddev);
6523 err = do_md_stop(mddev, 0, bdev);
6527 err = md_set_readonly(mddev, bdev);
6530 case HOT_REMOVE_DISK:
6531 err = hot_remove_disk(mddev, new_decode_dev(arg));
6535 /* We can support ADD_NEW_DISK on read-only arrays
6536 * on if we are re-adding a preexisting device.
6537 * So require mddev->pers and MD_DISK_SYNC.
6540 mdu_disk_info_t info;
6541 if (copy_from_user(&info, argp, sizeof(info)))
6543 else if (!(info.state & (1<<MD_DISK_SYNC)))
6544 /* Need to clear read-only for this */
6547 err = add_new_disk(mddev, &info);
6553 if (get_user(ro, (int __user *)(arg))) {
6559 /* if the bdev is going readonly the value of mddev->ro
6560 * does not matter, no writes are coming
6565 /* are we are already prepared for writes? */
6569 /* transitioning to readauto need only happen for
6570 * arrays that call md_write_start
6573 err = restart_array(mddev);
6576 set_disk_ro(mddev->gendisk, 0);
6583 * The remaining ioctls are changing the state of the
6584 * superblock, so we do not allow them on read-only arrays.
6585 * However non-MD ioctls (e.g. get-size) will still come through
6586 * here and hit the 'default' below, so only disallow
6587 * 'md' ioctls, and switch to rw mode if started auto-readonly.
6589 if (_IOC_TYPE(cmd) == MD_MAJOR && mddev->ro && mddev->pers) {
6590 if (mddev->ro == 2) {
6592 sysfs_notify_dirent_safe(mddev->sysfs_state);
6593 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6594 /* mddev_unlock will wake thread */
6595 /* If a device failed while we were read-only, we
6596 * need to make sure the metadata is updated now.
6598 if (test_bit(MD_CHANGE_DEVS, &mddev->flags)) {
6599 mddev_unlock(mddev);
6600 wait_event(mddev->sb_wait,
6601 !test_bit(MD_CHANGE_DEVS, &mddev->flags) &&
6602 !test_bit(MD_CHANGE_PENDING, &mddev->flags));
6603 mddev_lock_nointr(mddev);
6614 mdu_disk_info_t info;
6615 if (copy_from_user(&info, argp, sizeof(info)))
6618 err = add_new_disk(mddev, &info);
6623 err = hot_add_disk(mddev, new_decode_dev(arg));
6627 err = do_md_run(mddev);
6630 case SET_BITMAP_FILE:
6631 err = set_bitmap_file(mddev, (int)arg);
6641 if (mddev->hold_active == UNTIL_IOCTL &&
6643 mddev->hold_active = 0;
6644 mddev_unlock(mddev);
6653 #ifdef CONFIG_COMPAT
6654 static int md_compat_ioctl(struct block_device *bdev, fmode_t mode,
6655 unsigned int cmd, unsigned long arg)
6658 case HOT_REMOVE_DISK:
6660 case SET_DISK_FAULTY:
6661 case SET_BITMAP_FILE:
6662 /* These take in integer arg, do not convert */
6665 arg = (unsigned long)compat_ptr(arg);
6669 return md_ioctl(bdev, mode, cmd, arg);
6671 #endif /* CONFIG_COMPAT */
6673 static int md_open(struct block_device *bdev, fmode_t mode)
6676 * Succeed if we can lock the mddev, which confirms that
6677 * it isn't being stopped right now.
6679 struct mddev *mddev = mddev_find(bdev->bd_dev);
6685 if (mddev->gendisk != bdev->bd_disk) {
6686 /* we are racing with mddev_put which is discarding this
6690 /* Wait until bdev->bd_disk is definitely gone */
6691 flush_workqueue(md_misc_wq);
6692 /* Then retry the open from the top */
6693 return -ERESTARTSYS;
6695 BUG_ON(mddev != bdev->bd_disk->private_data);
6697 if ((err = mutex_lock_interruptible(&mddev->open_mutex)))
6701 atomic_inc(&mddev->openers);
6702 clear_bit(MD_STILL_CLOSED, &mddev->flags);
6703 mutex_unlock(&mddev->open_mutex);
6705 check_disk_change(bdev);
6710 static void md_release(struct gendisk *disk, fmode_t mode)
6712 struct mddev *mddev = disk->private_data;
6715 atomic_dec(&mddev->openers);
6719 static int md_media_changed(struct gendisk *disk)
6721 struct mddev *mddev = disk->private_data;
6723 return mddev->changed;
6726 static int md_revalidate(struct gendisk *disk)
6728 struct mddev *mddev = disk->private_data;
6733 static const struct block_device_operations md_fops =
6735 .owner = THIS_MODULE,
6737 .release = md_release,
6739 #ifdef CONFIG_COMPAT
6740 .compat_ioctl = md_compat_ioctl,
6742 .getgeo = md_getgeo,
6743 .media_changed = md_media_changed,
6744 .revalidate_disk= md_revalidate,
6747 static int md_thread(void * arg)
6749 struct md_thread *thread = arg;
6752 * md_thread is a 'system-thread', it's priority should be very
6753 * high. We avoid resource deadlocks individually in each
6754 * raid personality. (RAID5 does preallocation) We also use RR and
6755 * the very same RT priority as kswapd, thus we will never get
6756 * into a priority inversion deadlock.
6758 * we definitely have to have equal or higher priority than
6759 * bdflush, otherwise bdflush will deadlock if there are too
6760 * many dirty RAID5 blocks.
6763 allow_signal(SIGKILL);
6764 while (!kthread_should_stop()) {
6766 /* We need to wait INTERRUPTIBLE so that
6767 * we don't add to the load-average.
6768 * That means we need to be sure no signals are
6771 if (signal_pending(current))
6772 flush_signals(current);
6774 wait_event_interruptible_timeout
6776 test_bit(THREAD_WAKEUP, &thread->flags)
6777 || kthread_should_stop(),
6780 clear_bit(THREAD_WAKEUP, &thread->flags);
6781 if (!kthread_should_stop())
6782 thread->run(thread);
6788 void md_wakeup_thread(struct md_thread *thread)
6791 pr_debug("md: waking up MD thread %s.\n", thread->tsk->comm);
6792 set_bit(THREAD_WAKEUP, &thread->flags);
6793 wake_up(&thread->wqueue);
6797 struct md_thread *md_register_thread(void (*run) (struct md_thread *),
6798 struct mddev *mddev, const char *name)
6800 struct md_thread *thread;
6802 thread = kzalloc(sizeof(struct md_thread), GFP_KERNEL);
6806 init_waitqueue_head(&thread->wqueue);
6809 thread->mddev = mddev;
6810 thread->timeout = MAX_SCHEDULE_TIMEOUT;
6811 thread->tsk = kthread_run(md_thread, thread,
6813 mdname(thread->mddev),
6815 if (IS_ERR(thread->tsk)) {
6822 void md_unregister_thread(struct md_thread **threadp)
6824 struct md_thread *thread = *threadp;
6827 pr_debug("interrupting MD-thread pid %d\n", task_pid_nr(thread->tsk));
6828 /* Locking ensures that mddev_unlock does not wake_up a
6829 * non-existent thread
6831 spin_lock(&pers_lock);
6833 spin_unlock(&pers_lock);
6835 kthread_stop(thread->tsk);
6839 void md_error(struct mddev *mddev, struct md_rdev *rdev)
6846 if (!rdev || test_bit(Faulty, &rdev->flags))
6849 if (!mddev->pers || !mddev->pers->error_handler)
6851 mddev->pers->error_handler(mddev,rdev);
6852 if (mddev->degraded)
6853 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
6854 sysfs_notify_dirent_safe(rdev->sysfs_state);
6855 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6856 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6857 md_wakeup_thread(mddev->thread);
6858 if (mddev->event_work.func)
6859 queue_work(md_misc_wq, &mddev->event_work);
6860 md_new_event_inintr(mddev);
6863 /* seq_file implementation /proc/mdstat */
6865 static void status_unused(struct seq_file *seq)
6868 struct md_rdev *rdev;
6870 seq_printf(seq, "unused devices: ");
6872 list_for_each_entry(rdev, &pending_raid_disks, same_set) {
6873 char b[BDEVNAME_SIZE];
6875 seq_printf(seq, "%s ",
6876 bdevname(rdev->bdev,b));
6879 seq_printf(seq, "<none>");
6881 seq_printf(seq, "\n");
6885 static void status_resync(struct seq_file *seq, struct mddev * mddev)
6887 sector_t max_sectors, resync, res;
6888 unsigned long dt, db;
6891 unsigned int per_milli;
6893 if (mddev->curr_resync <= 3)
6896 resync = mddev->curr_resync
6897 - atomic_read(&mddev->recovery_active);
6899 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ||
6900 test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
6901 max_sectors = mddev->resync_max_sectors;
6903 max_sectors = mddev->dev_sectors;
6906 * Should not happen.
6912 /* Pick 'scale' such that (resync>>scale)*1000 will fit
6913 * in a sector_t, and (max_sectors>>scale) will fit in a
6914 * u32, as those are the requirements for sector_div.
6915 * Thus 'scale' must be at least 10
6918 if (sizeof(sector_t) > sizeof(unsigned long)) {
6919 while ( max_sectors/2 > (1ULL<<(scale+32)))
6922 res = (resync>>scale)*1000;
6923 sector_div(res, (u32)((max_sectors>>scale)+1));
6927 int i, x = per_milli/50, y = 20-x;
6928 seq_printf(seq, "[");
6929 for (i = 0; i < x; i++)
6930 seq_printf(seq, "=");
6931 seq_printf(seq, ">");
6932 for (i = 0; i < y; i++)
6933 seq_printf(seq, ".");
6934 seq_printf(seq, "] ");
6936 seq_printf(seq, " %s =%3u.%u%% (%llu/%llu)",
6937 (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)?
6939 (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)?
6941 (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
6942 "resync" : "recovery"))),
6943 per_milli/10, per_milli % 10,
6944 (unsigned long long) resync/2,
6945 (unsigned long long) max_sectors/2);
6948 * dt: time from mark until now
6949 * db: blocks written from mark until now
6950 * rt: remaining time
6952 * rt is a sector_t, so could be 32bit or 64bit.
6953 * So we divide before multiply in case it is 32bit and close
6955 * We scale the divisor (db) by 32 to avoid losing precision
6956 * near the end of resync when the number of remaining sectors
6958 * We then divide rt by 32 after multiplying by db to compensate.
6959 * The '+1' avoids division by zero if db is very small.
6961 dt = ((jiffies - mddev->resync_mark) / HZ);
6963 db = (mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active))
6964 - mddev->resync_mark_cnt;
6966 rt = max_sectors - resync; /* number of remaining sectors */
6967 sector_div(rt, db/32+1);
6971 seq_printf(seq, " finish=%lu.%lumin", (unsigned long)rt / 60,
6972 ((unsigned long)rt % 60)/6);
6974 seq_printf(seq, " speed=%ldK/sec", db/2/dt);
6977 static void *md_seq_start(struct seq_file *seq, loff_t *pos)
6979 struct list_head *tmp;
6981 struct mddev *mddev;
6989 spin_lock(&all_mddevs_lock);
6990 list_for_each(tmp,&all_mddevs)
6992 mddev = list_entry(tmp, struct mddev, all_mddevs);
6994 spin_unlock(&all_mddevs_lock);
6997 spin_unlock(&all_mddevs_lock);
6999 return (void*)2;/* tail */
7003 static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
7005 struct list_head *tmp;
7006 struct mddev *next_mddev, *mddev = v;
7012 spin_lock(&all_mddevs_lock);
7014 tmp = all_mddevs.next;
7016 tmp = mddev->all_mddevs.next;
7017 if (tmp != &all_mddevs)
7018 next_mddev = mddev_get(list_entry(tmp,struct mddev,all_mddevs));
7020 next_mddev = (void*)2;
7023 spin_unlock(&all_mddevs_lock);
7031 static void md_seq_stop(struct seq_file *seq, void *v)
7033 struct mddev *mddev = v;
7035 if (mddev && v != (void*)1 && v != (void*)2)
7039 static int md_seq_show(struct seq_file *seq, void *v)
7041 struct mddev *mddev = v;
7043 struct md_rdev *rdev;
7045 if (v == (void*)1) {
7046 struct md_personality *pers;
7047 seq_printf(seq, "Personalities : ");
7048 spin_lock(&pers_lock);
7049 list_for_each_entry(pers, &pers_list, list)
7050 seq_printf(seq, "[%s] ", pers->name);
7052 spin_unlock(&pers_lock);
7053 seq_printf(seq, "\n");
7054 seq->poll_event = atomic_read(&md_event_count);
7057 if (v == (void*)2) {
7062 if (mddev_lock(mddev) < 0)
7065 if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
7066 seq_printf(seq, "%s : %sactive", mdname(mddev),
7067 mddev->pers ? "" : "in");
7070 seq_printf(seq, " (read-only)");
7072 seq_printf(seq, " (auto-read-only)");
7073 seq_printf(seq, " %s", mddev->pers->name);
7077 rdev_for_each(rdev, mddev) {
7078 char b[BDEVNAME_SIZE];
7079 seq_printf(seq, " %s[%d]",
7080 bdevname(rdev->bdev,b), rdev->desc_nr);
7081 if (test_bit(WriteMostly, &rdev->flags))
7082 seq_printf(seq, "(W)");
7083 if (test_bit(Faulty, &rdev->flags)) {
7084 seq_printf(seq, "(F)");
7087 if (rdev->raid_disk < 0)
7088 seq_printf(seq, "(S)"); /* spare */
7089 if (test_bit(Replacement, &rdev->flags))
7090 seq_printf(seq, "(R)");
7091 sectors += rdev->sectors;
7094 if (!list_empty(&mddev->disks)) {
7096 seq_printf(seq, "\n %llu blocks",
7097 (unsigned long long)
7098 mddev->array_sectors / 2);
7100 seq_printf(seq, "\n %llu blocks",
7101 (unsigned long long)sectors / 2);
7103 if (mddev->persistent) {
7104 if (mddev->major_version != 0 ||
7105 mddev->minor_version != 90) {
7106 seq_printf(seq," super %d.%d",
7107 mddev->major_version,
7108 mddev->minor_version);
7110 } else if (mddev->external)
7111 seq_printf(seq, " super external:%s",
7112 mddev->metadata_type);
7114 seq_printf(seq, " super non-persistent");
7117 mddev->pers->status(seq, mddev);
7118 seq_printf(seq, "\n ");
7119 if (mddev->pers->sync_request) {
7120 if (mddev->curr_resync > 2) {
7121 status_resync(seq, mddev);
7122 seq_printf(seq, "\n ");
7123 } else if (mddev->curr_resync >= 1)
7124 seq_printf(seq, "\tresync=DELAYED\n ");
7125 else if (mddev->recovery_cp < MaxSector)
7126 seq_printf(seq, "\tresync=PENDING\n ");
7129 seq_printf(seq, "\n ");
7131 bitmap_status(seq, mddev->bitmap);
7133 seq_printf(seq, "\n");
7135 mddev_unlock(mddev);
7140 static const struct seq_operations md_seq_ops = {
7141 .start = md_seq_start,
7142 .next = md_seq_next,
7143 .stop = md_seq_stop,
7144 .show = md_seq_show,
7147 static int md_seq_open(struct inode *inode, struct file *file)
7149 struct seq_file *seq;
7152 error = seq_open(file, &md_seq_ops);
7156 seq = file->private_data;
7157 seq->poll_event = atomic_read(&md_event_count);
7161 static int md_unloading;
7162 static unsigned int mdstat_poll(struct file *filp, poll_table *wait)
7164 struct seq_file *seq = filp->private_data;
7168 return POLLIN|POLLRDNORM|POLLERR|POLLPRI;;
7169 poll_wait(filp, &md_event_waiters, wait);
7171 /* always allow read */
7172 mask = POLLIN | POLLRDNORM;
7174 if (seq->poll_event != atomic_read(&md_event_count))
7175 mask |= POLLERR | POLLPRI;
7179 static const struct file_operations md_seq_fops = {
7180 .owner = THIS_MODULE,
7181 .open = md_seq_open,
7183 .llseek = seq_lseek,
7184 .release = seq_release_private,
7185 .poll = mdstat_poll,
7188 int register_md_personality(struct md_personality *p)
7190 printk(KERN_INFO "md: %s personality registered for level %d\n",
7192 spin_lock(&pers_lock);
7193 list_add_tail(&p->list, &pers_list);
7194 spin_unlock(&pers_lock);
7198 int unregister_md_personality(struct md_personality *p)
7200 printk(KERN_INFO "md: %s personality unregistered\n", p->name);
7201 spin_lock(&pers_lock);
7202 list_del_init(&p->list);
7203 spin_unlock(&pers_lock);
7207 static int is_mddev_idle(struct mddev *mddev, int init)
7209 struct md_rdev * rdev;
7215 rdev_for_each_rcu(rdev, mddev) {
7216 struct gendisk *disk = rdev->bdev->bd_contains->bd_disk;
7217 curr_events = (int)part_stat_read(&disk->part0, sectors[0]) +
7218 (int)part_stat_read(&disk->part0, sectors[1]) -
7219 atomic_read(&disk->sync_io);
7220 /* sync IO will cause sync_io to increase before the disk_stats
7221 * as sync_io is counted when a request starts, and
7222 * disk_stats is counted when it completes.
7223 * So resync activity will cause curr_events to be smaller than
7224 * when there was no such activity.
7225 * non-sync IO will cause disk_stat to increase without
7226 * increasing sync_io so curr_events will (eventually)
7227 * be larger than it was before. Once it becomes
7228 * substantially larger, the test below will cause
7229 * the array to appear non-idle, and resync will slow
7231 * If there is a lot of outstanding resync activity when
7232 * we set last_event to curr_events, then all that activity
7233 * completing might cause the array to appear non-idle
7234 * and resync will be slowed down even though there might
7235 * not have been non-resync activity. This will only
7236 * happen once though. 'last_events' will soon reflect
7237 * the state where there is little or no outstanding
7238 * resync requests, and further resync activity will
7239 * always make curr_events less than last_events.
7242 if (init || curr_events - rdev->last_events > 64) {
7243 rdev->last_events = curr_events;
7251 void md_done_sync(struct mddev *mddev, int blocks, int ok)
7253 /* another "blocks" (512byte) blocks have been synced */
7254 atomic_sub(blocks, &mddev->recovery_active);
7255 wake_up(&mddev->recovery_wait);
7257 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
7258 set_bit(MD_RECOVERY_ERROR, &mddev->recovery);
7259 md_wakeup_thread(mddev->thread);
7260 // stop recovery, signal do_sync ....
7265 /* md_write_start(mddev, bi)
7266 * If we need to update some array metadata (e.g. 'active' flag
7267 * in superblock) before writing, schedule a superblock update
7268 * and wait for it to complete.
7270 void md_write_start(struct mddev *mddev, struct bio *bi)
7273 if (bio_data_dir(bi) != WRITE)
7276 BUG_ON(mddev->ro == 1);
7277 if (mddev->ro == 2) {
7278 /* need to switch to read/write */
7280 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7281 md_wakeup_thread(mddev->thread);
7282 md_wakeup_thread(mddev->sync_thread);
7285 atomic_inc(&mddev->writes_pending);
7286 if (mddev->safemode == 1)
7287 mddev->safemode = 0;
7288 if (mddev->in_sync) {
7289 spin_lock_irq(&mddev->write_lock);
7290 if (mddev->in_sync) {
7292 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
7293 set_bit(MD_CHANGE_PENDING, &mddev->flags);
7294 md_wakeup_thread(mddev->thread);
7297 spin_unlock_irq(&mddev->write_lock);
7300 sysfs_notify_dirent_safe(mddev->sysfs_state);
7301 wait_event(mddev->sb_wait,
7302 !test_bit(MD_CHANGE_PENDING, &mddev->flags));
7305 void md_write_end(struct mddev *mddev)
7307 if (atomic_dec_and_test(&mddev->writes_pending)) {
7308 if (mddev->safemode == 2)
7309 md_wakeup_thread(mddev->thread);
7310 else if (mddev->safemode_delay)
7311 mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay);
7315 /* md_allow_write(mddev)
7316 * Calling this ensures that the array is marked 'active' so that writes
7317 * may proceed without blocking. It is important to call this before
7318 * attempting a GFP_KERNEL allocation while holding the mddev lock.
7319 * Must be called with mddev_lock held.
7321 * In the ->external case MD_CHANGE_CLEAN can not be cleared until mddev->lock
7322 * is dropped, so return -EAGAIN after notifying userspace.
7324 int md_allow_write(struct mddev *mddev)
7330 if (!mddev->pers->sync_request)
7333 spin_lock_irq(&mddev->write_lock);
7334 if (mddev->in_sync) {
7336 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
7337 set_bit(MD_CHANGE_PENDING, &mddev->flags);
7338 if (mddev->safemode_delay &&
7339 mddev->safemode == 0)
7340 mddev->safemode = 1;
7341 spin_unlock_irq(&mddev->write_lock);
7342 md_update_sb(mddev, 0);
7343 sysfs_notify_dirent_safe(mddev->sysfs_state);
7345 spin_unlock_irq(&mddev->write_lock);
7347 if (test_bit(MD_CHANGE_PENDING, &mddev->flags))
7352 EXPORT_SYMBOL_GPL(md_allow_write);
7354 #define SYNC_MARKS 10
7355 #define SYNC_MARK_STEP (3*HZ)
7356 #define UPDATE_FREQUENCY (5*60*HZ)
7357 void md_do_sync(struct md_thread *thread)
7359 struct mddev *mddev = thread->mddev;
7360 struct mddev *mddev2;
7361 unsigned int currspeed = 0,
7363 sector_t max_sectors,j, io_sectors, recovery_done;
7364 unsigned long mark[SYNC_MARKS];
7365 unsigned long update_time;
7366 sector_t mark_cnt[SYNC_MARKS];
7368 struct list_head *tmp;
7369 sector_t last_check;
7371 struct md_rdev *rdev;
7372 char *desc, *action = NULL;
7373 struct blk_plug plug;
7375 /* just incase thread restarts... */
7376 if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
7378 if (mddev->ro) {/* never try to sync a read-only array */
7379 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
7383 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
7384 if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)) {
7385 desc = "data-check";
7387 } else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
7388 desc = "requested-resync";
7392 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
7397 mddev->last_sync_action = action ?: desc;
7399 /* we overload curr_resync somewhat here.
7400 * 0 == not engaged in resync at all
7401 * 2 == checking that there is no conflict with another sync
7402 * 1 == like 2, but have yielded to allow conflicting resync to
7404 * other == active in resync - this many blocks
7406 * Before starting a resync we must have set curr_resync to
7407 * 2, and then checked that every "conflicting" array has curr_resync
7408 * less than ours. When we find one that is the same or higher
7409 * we wait on resync_wait. To avoid deadlock, we reduce curr_resync
7410 * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
7411 * This will mean we have to start checking from the beginning again.
7416 mddev->curr_resync = 2;
7419 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
7421 for_each_mddev(mddev2, tmp) {
7422 if (mddev2 == mddev)
7424 if (!mddev->parallel_resync
7425 && mddev2->curr_resync
7426 && match_mddev_units(mddev, mddev2)) {
7428 if (mddev < mddev2 && mddev->curr_resync == 2) {
7429 /* arbitrarily yield */
7430 mddev->curr_resync = 1;
7431 wake_up(&resync_wait);
7433 if (mddev > mddev2 && mddev->curr_resync == 1)
7434 /* no need to wait here, we can wait the next
7435 * time 'round when curr_resync == 2
7438 /* We need to wait 'interruptible' so as not to
7439 * contribute to the load average, and not to
7440 * be caught by 'softlockup'
7442 prepare_to_wait(&resync_wait, &wq, TASK_INTERRUPTIBLE);
7443 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
7444 mddev2->curr_resync >= mddev->curr_resync) {
7445 printk(KERN_INFO "md: delaying %s of %s"
7446 " until %s has finished (they"
7447 " share one or more physical units)\n",
7448 desc, mdname(mddev), mdname(mddev2));
7450 if (signal_pending(current))
7451 flush_signals(current);
7453 finish_wait(&resync_wait, &wq);
7456 finish_wait(&resync_wait, &wq);
7459 } while (mddev->curr_resync < 2);
7462 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
7463 /* resync follows the size requested by the personality,
7464 * which defaults to physical size, but can be virtual size
7466 max_sectors = mddev->resync_max_sectors;
7467 atomic64_set(&mddev->resync_mismatches, 0);
7468 /* we don't use the checkpoint if there's a bitmap */
7469 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
7470 j = mddev->resync_min;
7471 else if (!mddev->bitmap)
7472 j = mddev->recovery_cp;
7474 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
7475 max_sectors = mddev->resync_max_sectors;
7477 /* recovery follows the physical size of devices */
7478 max_sectors = mddev->dev_sectors;
7481 rdev_for_each_rcu(rdev, mddev)
7482 if (rdev->raid_disk >= 0 &&
7483 !test_bit(Faulty, &rdev->flags) &&
7484 !test_bit(In_sync, &rdev->flags) &&
7485 rdev->recovery_offset < j)
7486 j = rdev->recovery_offset;
7489 /* If there is a bitmap, we need to make sure all
7490 * writes that started before we added a spare
7491 * complete before we start doing a recovery.
7492 * Otherwise the write might complete and (via
7493 * bitmap_endwrite) set a bit in the bitmap after the
7494 * recovery has checked that bit and skipped that
7497 if (mddev->bitmap) {
7498 mddev->pers->quiesce(mddev, 1);
7499 mddev->pers->quiesce(mddev, 0);
7503 printk(KERN_INFO "md: %s of RAID array %s\n", desc, mdname(mddev));
7504 printk(KERN_INFO "md: minimum _guaranteed_ speed:"
7505 " %d KB/sec/disk.\n", speed_min(mddev));
7506 printk(KERN_INFO "md: using maximum available idle IO bandwidth "
7507 "(but not more than %d KB/sec) for %s.\n",
7508 speed_max(mddev), desc);
7510 is_mddev_idle(mddev, 1); /* this initializes IO event counters */
7513 for (m = 0; m < SYNC_MARKS; m++) {
7515 mark_cnt[m] = io_sectors;
7518 mddev->resync_mark = mark[last_mark];
7519 mddev->resync_mark_cnt = mark_cnt[last_mark];
7522 * Tune reconstruction:
7524 window = 32*(PAGE_SIZE/512);
7525 printk(KERN_INFO "md: using %dk window, over a total of %lluk.\n",
7526 window/2, (unsigned long long)max_sectors/2);
7528 atomic_set(&mddev->recovery_active, 0);
7533 "md: resuming %s of %s from checkpoint.\n",
7534 desc, mdname(mddev));
7535 mddev->curr_resync = j;
7537 mddev->curr_resync = 3; /* no longer delayed */
7538 mddev->curr_resync_completed = j;
7539 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
7540 md_new_event(mddev);
7541 update_time = jiffies;
7543 blk_start_plug(&plug);
7544 while (j < max_sectors) {
7549 if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
7550 ((mddev->curr_resync > mddev->curr_resync_completed &&
7551 (mddev->curr_resync - mddev->curr_resync_completed)
7552 > (max_sectors >> 4)) ||
7553 time_after_eq(jiffies, update_time + UPDATE_FREQUENCY) ||
7554 (j - mddev->curr_resync_completed)*2
7555 >= mddev->resync_max - mddev->curr_resync_completed
7557 /* time to update curr_resync_completed */
7558 wait_event(mddev->recovery_wait,
7559 atomic_read(&mddev->recovery_active) == 0);
7560 mddev->curr_resync_completed = j;
7561 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) &&
7562 j > mddev->recovery_cp)
7563 mddev->recovery_cp = j;
7564 update_time = jiffies;
7565 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
7566 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
7569 while (j >= mddev->resync_max &&
7570 !test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
7571 /* As this condition is controlled by user-space,
7572 * we can block indefinitely, so use '_interruptible'
7573 * to avoid triggering warnings.
7575 flush_signals(current); /* just in case */
7576 wait_event_interruptible(mddev->recovery_wait,
7577 mddev->resync_max > j
7578 || test_bit(MD_RECOVERY_INTR,
7582 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
7585 sectors = mddev->pers->sync_request(mddev, j, &skipped,
7586 currspeed < speed_min(mddev));
7588 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
7592 if (!skipped) { /* actual IO requested */
7593 io_sectors += sectors;
7594 atomic_add(sectors, &mddev->recovery_active);
7597 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
7602 mddev->curr_resync = j;
7603 mddev->curr_mark_cnt = io_sectors;
7604 if (last_check == 0)
7605 /* this is the earliest that rebuild will be
7606 * visible in /proc/mdstat
7608 md_new_event(mddev);
7610 if (last_check + window > io_sectors || j == max_sectors)
7613 last_check = io_sectors;
7615 if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) {
7617 int next = (last_mark+1) % SYNC_MARKS;
7619 mddev->resync_mark = mark[next];
7620 mddev->resync_mark_cnt = mark_cnt[next];
7621 mark[next] = jiffies;
7622 mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active);
7626 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
7630 * this loop exits only if either when we are slower than
7631 * the 'hard' speed limit, or the system was IO-idle for
7633 * the system might be non-idle CPU-wise, but we only care
7634 * about not overloading the IO subsystem. (things like an
7635 * e2fsck being done on the RAID array should execute fast)
7639 recovery_done = io_sectors - atomic_read(&mddev->recovery_active);
7640 currspeed = ((unsigned long)(recovery_done - mddev->resync_mark_cnt))/2
7641 /((jiffies-mddev->resync_mark)/HZ +1) +1;
7643 if (currspeed > speed_min(mddev)) {
7644 if ((currspeed > speed_max(mddev)) ||
7645 !is_mddev_idle(mddev, 0)) {
7651 printk(KERN_INFO "md: %s: %s %s.\n",mdname(mddev), desc,
7652 test_bit(MD_RECOVERY_INTR, &mddev->recovery)
7653 ? "interrupted" : "done");
7655 * this also signals 'finished resyncing' to md_stop
7657 blk_finish_plug(&plug);
7658 wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
7660 /* tell personality that we are finished */
7661 mddev->pers->sync_request(mddev, max_sectors, &skipped, 1);
7663 if (!test_bit(MD_RECOVERY_CHECK, &mddev->recovery) &&
7664 mddev->curr_resync > 2) {
7665 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
7666 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
7667 if (mddev->curr_resync >= mddev->recovery_cp) {
7669 "md: checkpointing %s of %s.\n",
7670 desc, mdname(mddev));
7671 if (test_bit(MD_RECOVERY_ERROR,
7673 mddev->recovery_cp =
7674 mddev->curr_resync_completed;
7676 mddev->recovery_cp =
7680 mddev->recovery_cp = MaxSector;
7682 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery))
7683 mddev->curr_resync = MaxSector;
7685 rdev_for_each_rcu(rdev, mddev)
7686 if (rdev->raid_disk >= 0 &&
7687 mddev->delta_disks >= 0 &&
7688 !test_bit(Faulty, &rdev->flags) &&
7689 !test_bit(In_sync, &rdev->flags) &&
7690 rdev->recovery_offset < mddev->curr_resync)
7691 rdev->recovery_offset = mddev->curr_resync;
7696 set_bit(MD_CHANGE_DEVS, &mddev->flags);
7698 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
7699 /* We completed so min/max setting can be forgotten if used. */
7700 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
7701 mddev->resync_min = 0;
7702 mddev->resync_max = MaxSector;
7703 } else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
7704 mddev->resync_min = mddev->curr_resync_completed;
7705 mddev->curr_resync = 0;
7706 wake_up(&resync_wait);
7707 set_bit(MD_RECOVERY_DONE, &mddev->recovery);
7708 md_wakeup_thread(mddev->thread);
7711 EXPORT_SYMBOL_GPL(md_do_sync);
7713 static int remove_and_add_spares(struct mddev *mddev,
7714 struct md_rdev *this)
7716 struct md_rdev *rdev;
7720 rdev_for_each(rdev, mddev)
7721 if ((this == NULL || rdev == this) &&
7722 rdev->raid_disk >= 0 &&
7723 !test_bit(Blocked, &rdev->flags) &&
7724 (test_bit(Faulty, &rdev->flags) ||
7725 ! test_bit(In_sync, &rdev->flags)) &&
7726 atomic_read(&rdev->nr_pending)==0) {
7727 if (mddev->pers->hot_remove_disk(
7728 mddev, rdev) == 0) {
7729 sysfs_unlink_rdev(mddev, rdev);
7730 rdev->raid_disk = -1;
7734 if (removed && mddev->kobj.sd)
7735 sysfs_notify(&mddev->kobj, NULL, "degraded");
7740 rdev_for_each(rdev, mddev) {
7741 if (rdev->raid_disk >= 0 &&
7742 !test_bit(In_sync, &rdev->flags) &&
7743 !test_bit(Faulty, &rdev->flags))
7745 if (rdev->raid_disk >= 0)
7747 if (test_bit(Faulty, &rdev->flags))
7750 ! (rdev->saved_raid_disk >= 0 &&
7751 !test_bit(Bitmap_sync, &rdev->flags)))
7754 if (rdev->saved_raid_disk < 0)
7755 rdev->recovery_offset = 0;
7757 hot_add_disk(mddev, rdev) == 0) {
7758 if (sysfs_link_rdev(mddev, rdev))
7759 /* failure here is OK */;
7761 md_new_event(mddev);
7762 set_bit(MD_CHANGE_DEVS, &mddev->flags);
7767 set_bit(MD_CHANGE_DEVS, &mddev->flags);
7772 * This routine is regularly called by all per-raid-array threads to
7773 * deal with generic issues like resync and super-block update.
7774 * Raid personalities that don't have a thread (linear/raid0) do not
7775 * need this as they never do any recovery or update the superblock.
7777 * It does not do any resync itself, but rather "forks" off other threads
7778 * to do that as needed.
7779 * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
7780 * "->recovery" and create a thread at ->sync_thread.
7781 * When the thread finishes it sets MD_RECOVERY_DONE
7782 * and wakeups up this thread which will reap the thread and finish up.
7783 * This thread also removes any faulty devices (with nr_pending == 0).
7785 * The overall approach is:
7786 * 1/ if the superblock needs updating, update it.
7787 * 2/ If a recovery thread is running, don't do anything else.
7788 * 3/ If recovery has finished, clean up, possibly marking spares active.
7789 * 4/ If there are any faulty devices, remove them.
7790 * 5/ If array is degraded, try to add spares devices
7791 * 6/ If array has spares or is not in-sync, start a resync thread.
7793 void md_check_recovery(struct mddev *mddev)
7795 if (mddev->suspended)
7799 bitmap_daemon_work(mddev);
7801 if (signal_pending(current)) {
7802 if (mddev->pers->sync_request && !mddev->external) {
7803 printk(KERN_INFO "md: %s in immediate safe mode\n",
7805 mddev->safemode = 2;
7807 flush_signals(current);
7810 if (mddev->ro && !test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
7813 (mddev->flags & MD_UPDATE_SB_FLAGS & ~ (1<<MD_CHANGE_PENDING)) ||
7814 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
7815 test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
7816 (mddev->external == 0 && mddev->safemode == 1) ||
7817 (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending)
7818 && !mddev->in_sync && mddev->recovery_cp == MaxSector)
7822 if (mddev_trylock(mddev)) {
7826 /* On a read-only array we can:
7827 * - remove failed devices
7828 * - add already-in_sync devices if the array itself
7830 * As we only add devices that are already in-sync,
7831 * we can activate the spares immediately.
7833 remove_and_add_spares(mddev, NULL);
7834 /* There is no thread, but we need to call
7835 * ->spare_active and clear saved_raid_disk
7837 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
7838 md_reap_sync_thread(mddev);
7839 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7843 if (!mddev->external) {
7845 spin_lock_irq(&mddev->write_lock);
7846 if (mddev->safemode &&
7847 !atomic_read(&mddev->writes_pending) &&
7849 mddev->recovery_cp == MaxSector) {
7852 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
7854 if (mddev->safemode == 1)
7855 mddev->safemode = 0;
7856 spin_unlock_irq(&mddev->write_lock);
7858 sysfs_notify_dirent_safe(mddev->sysfs_state);
7861 if (mddev->flags & MD_UPDATE_SB_FLAGS)
7862 md_update_sb(mddev, 0);
7864 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
7865 !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
7866 /* resync/recovery still happening */
7867 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7870 if (mddev->sync_thread) {
7871 md_reap_sync_thread(mddev);
7874 /* Set RUNNING before clearing NEEDED to avoid
7875 * any transients in the value of "sync_action".
7877 mddev->curr_resync_completed = 0;
7878 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
7879 /* Clear some bits that don't mean anything, but
7882 clear_bit(MD_RECOVERY_INTR, &mddev->recovery);
7883 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
7885 if (!test_and_clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
7886 test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
7888 /* no recovery is running.
7889 * remove any failed drives, then
7890 * add spares if possible.
7891 * Spares are also removed and re-added, to allow
7892 * the personality to fail the re-add.
7895 if (mddev->reshape_position != MaxSector) {
7896 if (mddev->pers->check_reshape == NULL ||
7897 mddev->pers->check_reshape(mddev) != 0)
7898 /* Cannot proceed */
7900 set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
7901 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
7902 } else if ((spares = remove_and_add_spares(mddev, NULL))) {
7903 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
7904 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
7905 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
7906 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
7907 } else if (mddev->recovery_cp < MaxSector) {
7908 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
7909 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
7910 } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
7911 /* nothing to be done ... */
7914 if (mddev->pers->sync_request) {
7916 /* We are adding a device or devices to an array
7917 * which has the bitmap stored on all devices.
7918 * So make sure all bitmap pages get written
7920 bitmap_write_all(mddev->bitmap);
7922 mddev->sync_thread = md_register_thread(md_do_sync,
7925 if (!mddev->sync_thread) {
7926 printk(KERN_ERR "%s: could not start resync"
7929 /* leave the spares where they are, it shouldn't hurt */
7930 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
7931 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
7932 clear_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
7933 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
7934 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
7936 md_wakeup_thread(mddev->sync_thread);
7937 sysfs_notify_dirent_safe(mddev->sysfs_action);
7938 md_new_event(mddev);
7941 wake_up(&mddev->sb_wait);
7943 if (!mddev->sync_thread) {
7944 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
7945 if (test_and_clear_bit(MD_RECOVERY_RECOVER,
7947 if (mddev->sysfs_action)
7948 sysfs_notify_dirent_safe(mddev->sysfs_action);
7950 mddev_unlock(mddev);
7954 void md_reap_sync_thread(struct mddev *mddev)
7956 struct md_rdev *rdev;
7958 /* resync has finished, collect result */
7959 md_unregister_thread(&mddev->sync_thread);
7960 wake_up(&resync_wait);
7961 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
7962 !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
7964 /* activate any spares */
7965 if (mddev->pers->spare_active(mddev)) {
7966 sysfs_notify(&mddev->kobj, NULL,
7968 set_bit(MD_CHANGE_DEVS, &mddev->flags);
7971 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
7972 mddev->pers->finish_reshape)
7973 mddev->pers->finish_reshape(mddev);
7975 /* If array is no-longer degraded, then any saved_raid_disk
7976 * information must be scrapped.
7978 if (!mddev->degraded)
7979 rdev_for_each(rdev, mddev)
7980 rdev->saved_raid_disk = -1;
7982 md_update_sb(mddev, 1);
7983 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
7984 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
7985 clear_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
7986 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
7987 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
7988 /* flag recovery needed just to double check */
7989 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7990 sysfs_notify_dirent_safe(mddev->sysfs_action);
7991 md_new_event(mddev);
7992 if (mddev->event_work.func)
7993 queue_work(md_misc_wq, &mddev->event_work);
7996 void md_wait_for_blocked_rdev(struct md_rdev *rdev, struct mddev *mddev)
7998 sysfs_notify_dirent_safe(rdev->sysfs_state);
7999 wait_event_timeout(rdev->blocked_wait,
8000 !test_bit(Blocked, &rdev->flags) &&
8001 !test_bit(BlockedBadBlocks, &rdev->flags),
8002 msecs_to_jiffies(5000));
8003 rdev_dec_pending(rdev, mddev);
8005 EXPORT_SYMBOL(md_wait_for_blocked_rdev);
8007 void md_finish_reshape(struct mddev *mddev)
8009 /* called be personality module when reshape completes. */
8010 struct md_rdev *rdev;
8012 rdev_for_each(rdev, mddev) {
8013 if (rdev->data_offset > rdev->new_data_offset)
8014 rdev->sectors += rdev->data_offset - rdev->new_data_offset;
8016 rdev->sectors -= rdev->new_data_offset - rdev->data_offset;
8017 rdev->data_offset = rdev->new_data_offset;
8020 EXPORT_SYMBOL(md_finish_reshape);
8022 /* Bad block management.
8023 * We can record which blocks on each device are 'bad' and so just
8024 * fail those blocks, or that stripe, rather than the whole device.
8025 * Entries in the bad-block table are 64bits wide. This comprises:
8026 * Length of bad-range, in sectors: 0-511 for lengths 1-512
8027 * Start of bad-range, sector offset, 54 bits (allows 8 exbibytes)
8028 * A 'shift' can be set so that larger blocks are tracked and
8029 * consequently larger devices can be covered.
8030 * 'Acknowledged' flag - 1 bit. - the most significant bit.
8032 * Locking of the bad-block table uses a seqlock so md_is_badblock
8033 * might need to retry if it is very unlucky.
8034 * We will sometimes want to check for bad blocks in a bi_end_io function,
8035 * so we use the write_seqlock_irq variant.
8037 * When looking for a bad block we specify a range and want to
8038 * know if any block in the range is bad. So we binary-search
8039 * to the last range that starts at-or-before the given endpoint,
8040 * (or "before the sector after the target range")
8041 * then see if it ends after the given start.
8043 * 0 if there are no known bad blocks in the range
8044 * 1 if there are known bad block which are all acknowledged
8045 * -1 if there are bad blocks which have not yet been acknowledged in metadata.
8046 * plus the start/length of the first bad section we overlap.
8048 int md_is_badblock(struct badblocks *bb, sector_t s, int sectors,
8049 sector_t *first_bad, int *bad_sectors)
8055 sector_t target = s + sectors;
8058 if (bb->shift > 0) {
8059 /* round the start down, and the end up */
8061 target += (1<<bb->shift) - 1;
8062 target >>= bb->shift;
8063 sectors = target - s;
8065 /* 'target' is now the first block after the bad range */
8068 seq = read_seqbegin(&bb->lock);
8073 /* Binary search between lo and hi for 'target'
8074 * i.e. for the last range that starts before 'target'
8076 /* INVARIANT: ranges before 'lo' and at-or-after 'hi'
8077 * are known not to be the last range before target.
8078 * VARIANT: hi-lo is the number of possible
8079 * ranges, and decreases until it reaches 1
8081 while (hi - lo > 1) {
8082 int mid = (lo + hi) / 2;
8083 sector_t a = BB_OFFSET(p[mid]);
8085 /* This could still be the one, earlier ranges
8089 /* This and later ranges are definitely out. */
8092 /* 'lo' might be the last that started before target, but 'hi' isn't */
8094 /* need to check all range that end after 's' to see if
8095 * any are unacknowledged.
8098 BB_OFFSET(p[lo]) + BB_LEN(p[lo]) > s) {
8099 if (BB_OFFSET(p[lo]) < target) {
8100 /* starts before the end, and finishes after
8101 * the start, so they must overlap
8103 if (rv != -1 && BB_ACK(p[lo]))
8107 *first_bad = BB_OFFSET(p[lo]);
8108 *bad_sectors = BB_LEN(p[lo]);
8114 if (read_seqretry(&bb->lock, seq))
8119 EXPORT_SYMBOL_GPL(md_is_badblock);
8122 * Add a range of bad blocks to the table.
8123 * This might extend the table, or might contract it
8124 * if two adjacent ranges can be merged.
8125 * We binary-search to find the 'insertion' point, then
8126 * decide how best to handle it.
8128 static int md_set_badblocks(struct badblocks *bb, sector_t s, int sectors,
8134 unsigned long flags;
8137 /* badblocks are disabled */
8141 /* round the start down, and the end up */
8142 sector_t next = s + sectors;
8144 next += (1<<bb->shift) - 1;
8149 write_seqlock_irqsave(&bb->lock, flags);
8154 /* Find the last range that starts at-or-before 's' */
8155 while (hi - lo > 1) {
8156 int mid = (lo + hi) / 2;
8157 sector_t a = BB_OFFSET(p[mid]);
8163 if (hi > lo && BB_OFFSET(p[lo]) > s)
8167 /* we found a range that might merge with the start
8170 sector_t a = BB_OFFSET(p[lo]);
8171 sector_t e = a + BB_LEN(p[lo]);
8172 int ack = BB_ACK(p[lo]);
8174 /* Yes, we can merge with a previous range */
8175 if (s == a && s + sectors >= e)
8176 /* new range covers old */
8179 ack = ack && acknowledged;
8181 if (e < s + sectors)
8183 if (e - a <= BB_MAX_LEN) {
8184 p[lo] = BB_MAKE(a, e-a, ack);
8187 /* does not all fit in one range,
8188 * make p[lo] maximal
8190 if (BB_LEN(p[lo]) != BB_MAX_LEN)
8191 p[lo] = BB_MAKE(a, BB_MAX_LEN, ack);
8197 if (sectors && hi < bb->count) {
8198 /* 'hi' points to the first range that starts after 's'.
8199 * Maybe we can merge with the start of that range */
8200 sector_t a = BB_OFFSET(p[hi]);
8201 sector_t e = a + BB_LEN(p[hi]);
8202 int ack = BB_ACK(p[hi]);
8203 if (a <= s + sectors) {
8204 /* merging is possible */
8205 if (e <= s + sectors) {
8210 ack = ack && acknowledged;
8213 if (e - a <= BB_MAX_LEN) {
8214 p[hi] = BB_MAKE(a, e-a, ack);
8217 p[hi] = BB_MAKE(a, BB_MAX_LEN, ack);
8225 if (sectors == 0 && hi < bb->count) {
8226 /* we might be able to combine lo and hi */
8227 /* Note: 's' is at the end of 'lo' */
8228 sector_t a = BB_OFFSET(p[hi]);
8229 int lolen = BB_LEN(p[lo]);
8230 int hilen = BB_LEN(p[hi]);
8231 int newlen = lolen + hilen - (s - a);
8232 if (s >= a && newlen < BB_MAX_LEN) {
8233 /* yes, we can combine them */
8234 int ack = BB_ACK(p[lo]) && BB_ACK(p[hi]);
8235 p[lo] = BB_MAKE(BB_OFFSET(p[lo]), newlen, ack);
8236 memmove(p + hi, p + hi + 1,
8237 (bb->count - hi - 1) * 8);
8242 /* didn't merge (it all).
8243 * Need to add a range just before 'hi' */
8244 if (bb->count >= MD_MAX_BADBLOCKS) {
8245 /* No room for more */
8249 int this_sectors = sectors;
8250 memmove(p + hi + 1, p + hi,
8251 (bb->count - hi) * 8);
8254 if (this_sectors > BB_MAX_LEN)
8255 this_sectors = BB_MAX_LEN;
8256 p[hi] = BB_MAKE(s, this_sectors, acknowledged);
8257 sectors -= this_sectors;
8264 bb->unacked_exist = 1;
8265 write_sequnlock_irqrestore(&bb->lock, flags);
8270 int rdev_set_badblocks(struct md_rdev *rdev, sector_t s, int sectors,
8275 s += rdev->new_data_offset;
8277 s += rdev->data_offset;
8278 rv = md_set_badblocks(&rdev->badblocks,
8281 /* Make sure they get written out promptly */
8282 sysfs_notify_dirent_safe(rdev->sysfs_state);
8283 set_bit(MD_CHANGE_CLEAN, &rdev->mddev->flags);
8284 md_wakeup_thread(rdev->mddev->thread);
8288 EXPORT_SYMBOL_GPL(rdev_set_badblocks);
8291 * Remove a range of bad blocks from the table.
8292 * This may involve extending the table if we spilt a region,
8293 * but it must not fail. So if the table becomes full, we just
8294 * drop the remove request.
8296 static int md_clear_badblocks(struct badblocks *bb, sector_t s, int sectors)
8300 sector_t target = s + sectors;
8303 if (bb->shift > 0) {
8304 /* When clearing we round the start up and the end down.
8305 * This should not matter as the shift should align with
8306 * the block size and no rounding should ever be needed.
8307 * However it is better the think a block is bad when it
8308 * isn't than to think a block is not bad when it is.
8310 s += (1<<bb->shift) - 1;
8312 target >>= bb->shift;
8313 sectors = target - s;
8316 write_seqlock_irq(&bb->lock);
8321 /* Find the last range that starts before 'target' */
8322 while (hi - lo > 1) {
8323 int mid = (lo + hi) / 2;
8324 sector_t a = BB_OFFSET(p[mid]);
8331 /* p[lo] is the last range that could overlap the
8332 * current range. Earlier ranges could also overlap,
8333 * but only this one can overlap the end of the range.
8335 if (BB_OFFSET(p[lo]) + BB_LEN(p[lo]) > target) {
8336 /* Partial overlap, leave the tail of this range */
8337 int ack = BB_ACK(p[lo]);
8338 sector_t a = BB_OFFSET(p[lo]);
8339 sector_t end = a + BB_LEN(p[lo]);
8342 /* we need to split this range */
8343 if (bb->count >= MD_MAX_BADBLOCKS) {
8347 memmove(p+lo+1, p+lo, (bb->count - lo) * 8);
8349 p[lo] = BB_MAKE(a, s-a, ack);
8352 p[lo] = BB_MAKE(target, end - target, ack);
8353 /* there is no longer an overlap */
8358 BB_OFFSET(p[lo]) + BB_LEN(p[lo]) > s) {
8359 /* This range does overlap */
8360 if (BB_OFFSET(p[lo]) < s) {
8361 /* Keep the early parts of this range. */
8362 int ack = BB_ACK(p[lo]);
8363 sector_t start = BB_OFFSET(p[lo]);
8364 p[lo] = BB_MAKE(start, s - start, ack);
8365 /* now low doesn't overlap, so.. */
8370 /* 'lo' is strictly before, 'hi' is strictly after,
8371 * anything between needs to be discarded
8374 memmove(p+lo+1, p+hi, (bb->count - hi) * 8);
8375 bb->count -= (hi - lo - 1);
8381 write_sequnlock_irq(&bb->lock);
8385 int rdev_clear_badblocks(struct md_rdev *rdev, sector_t s, int sectors,
8389 s += rdev->new_data_offset;
8391 s += rdev->data_offset;
8392 return md_clear_badblocks(&rdev->badblocks,
8395 EXPORT_SYMBOL_GPL(rdev_clear_badblocks);
8398 * Acknowledge all bad blocks in a list.
8399 * This only succeeds if ->changed is clear. It is used by
8400 * in-kernel metadata updates
8402 void md_ack_all_badblocks(struct badblocks *bb)
8404 if (bb->page == NULL || bb->changed)
8405 /* no point even trying */
8407 write_seqlock_irq(&bb->lock);
8409 if (bb->changed == 0 && bb->unacked_exist) {
8412 for (i = 0; i < bb->count ; i++) {
8413 if (!BB_ACK(p[i])) {
8414 sector_t start = BB_OFFSET(p[i]);
8415 int len = BB_LEN(p[i]);
8416 p[i] = BB_MAKE(start, len, 1);
8419 bb->unacked_exist = 0;
8421 write_sequnlock_irq(&bb->lock);
8423 EXPORT_SYMBOL_GPL(md_ack_all_badblocks);
8425 /* sysfs access to bad-blocks list.
8426 * We present two files.
8427 * 'bad-blocks' lists sector numbers and lengths of ranges that
8428 * are recorded as bad. The list is truncated to fit within
8429 * the one-page limit of sysfs.
8430 * Writing "sector length" to this file adds an acknowledged
8432 * 'unacknowledged-bad-blocks' lists bad blocks that have not yet
8433 * been acknowledged. Writing to this file adds bad blocks
8434 * without acknowledging them. This is largely for testing.
8438 badblocks_show(struct badblocks *bb, char *page, int unack)
8449 seq = read_seqbegin(&bb->lock);
8454 while (len < PAGE_SIZE && i < bb->count) {
8455 sector_t s = BB_OFFSET(p[i]);
8456 unsigned int length = BB_LEN(p[i]);
8457 int ack = BB_ACK(p[i]);
8463 len += snprintf(page+len, PAGE_SIZE-len, "%llu %u\n",
8464 (unsigned long long)s << bb->shift,
8465 length << bb->shift);
8467 if (unack && len == 0)
8468 bb->unacked_exist = 0;
8470 if (read_seqretry(&bb->lock, seq))
8479 badblocks_store(struct badblocks *bb, const char *page, size_t len, int unack)
8481 unsigned long long sector;
8485 /* Allow clearing via sysfs *only* for testing/debugging.
8486 * Normally only a successful write may clear a badblock
8489 if (page[0] == '-') {
8493 #endif /* DO_DEBUG */
8495 switch (sscanf(page, "%llu %d%c", §or, &length, &newline)) {
8497 if (newline != '\n')
8509 md_clear_badblocks(bb, sector, length);
8512 #endif /* DO_DEBUG */
8513 if (md_set_badblocks(bb, sector, length, !unack))
8519 static int md_notify_reboot(struct notifier_block *this,
8520 unsigned long code, void *x)
8522 struct list_head *tmp;
8523 struct mddev *mddev;
8526 for_each_mddev(mddev, tmp) {
8527 if (mddev_trylock(mddev)) {
8529 __md_stop_writes(mddev);
8530 if (mddev->persistent)
8531 mddev->safemode = 2;
8532 mddev_unlock(mddev);
8537 * certain more exotic SCSI devices are known to be
8538 * volatile wrt too early system reboots. While the
8539 * right place to handle this issue is the given
8540 * driver, we do want to have a safe RAID driver ...
8548 static struct notifier_block md_notifier = {
8549 .notifier_call = md_notify_reboot,
8551 .priority = INT_MAX, /* before any real devices */
8554 static void md_geninit(void)
8556 pr_debug("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
8558 proc_create("mdstat", S_IRUGO, NULL, &md_seq_fops);
8561 static int __init md_init(void)
8565 md_wq = alloc_workqueue("md", WQ_MEM_RECLAIM, 0);
8569 md_misc_wq = alloc_workqueue("md_misc", 0, 0);
8573 if ((ret = register_blkdev(MD_MAJOR, "md")) < 0)
8576 if ((ret = register_blkdev(0, "mdp")) < 0)
8580 blk_register_region(MKDEV(MD_MAJOR, 0), 512, THIS_MODULE,
8581 md_probe, NULL, NULL);
8582 blk_register_region(MKDEV(mdp_major, 0), 1UL<<MINORBITS, THIS_MODULE,
8583 md_probe, NULL, NULL);
8585 register_reboot_notifier(&md_notifier);
8586 raid_table_header = register_sysctl_table(raid_root_table);
8592 unregister_blkdev(MD_MAJOR, "md");
8594 destroy_workqueue(md_misc_wq);
8596 destroy_workqueue(md_wq);
8604 * Searches all registered partitions for autorun RAID arrays
8608 static LIST_HEAD(all_detected_devices);
8609 struct detected_devices_node {
8610 struct list_head list;
8614 void md_autodetect_dev(dev_t dev)
8616 struct detected_devices_node *node_detected_dev;
8618 node_detected_dev = kzalloc(sizeof(*node_detected_dev), GFP_KERNEL);
8619 if (node_detected_dev) {
8620 node_detected_dev->dev = dev;
8621 list_add_tail(&node_detected_dev->list, &all_detected_devices);
8623 printk(KERN_CRIT "md: md_autodetect_dev: kzalloc failed"
8624 ", skipping dev(%d,%d)\n", MAJOR(dev), MINOR(dev));
8629 static void autostart_arrays(int part)
8631 struct md_rdev *rdev;
8632 struct detected_devices_node *node_detected_dev;
8634 int i_scanned, i_passed;
8639 printk(KERN_INFO "md: Autodetecting RAID arrays.\n");
8641 while (!list_empty(&all_detected_devices) && i_scanned < INT_MAX) {
8643 node_detected_dev = list_entry(all_detected_devices.next,
8644 struct detected_devices_node, list);
8645 list_del(&node_detected_dev->list);
8646 dev = node_detected_dev->dev;
8647 kfree(node_detected_dev);
8648 rdev = md_import_device(dev,0, 90);
8652 if (test_bit(Faulty, &rdev->flags)) {
8656 set_bit(AutoDetected, &rdev->flags);
8657 list_add(&rdev->same_set, &pending_raid_disks);
8661 printk(KERN_INFO "md: Scanned %d and added %d devices.\n",
8662 i_scanned, i_passed);
8664 autorun_devices(part);
8667 #endif /* !MODULE */
8669 static __exit void md_exit(void)
8671 struct mddev *mddev;
8672 struct list_head *tmp;
8675 blk_unregister_region(MKDEV(MD_MAJOR,0), 512);
8676 blk_unregister_region(MKDEV(mdp_major,0), 1U << MINORBITS);
8678 unregister_blkdev(MD_MAJOR,"md");
8679 unregister_blkdev(mdp_major, "mdp");
8680 unregister_reboot_notifier(&md_notifier);
8681 unregister_sysctl_table(raid_table_header);
8683 /* We cannot unload the modules while some process is
8684 * waiting for us in select() or poll() - wake them up
8687 while (waitqueue_active(&md_event_waiters)) {
8688 /* not safe to leave yet */
8689 wake_up(&md_event_waiters);
8693 remove_proc_entry("mdstat", NULL);
8695 for_each_mddev(mddev, tmp) {
8696 export_array(mddev);
8697 mddev->hold_active = 0;
8699 destroy_workqueue(md_misc_wq);
8700 destroy_workqueue(md_wq);
8703 subsys_initcall(md_init);
8704 module_exit(md_exit)
8706 static int get_ro(char *buffer, struct kernel_param *kp)
8708 return sprintf(buffer, "%d", start_readonly);
8710 static int set_ro(const char *val, struct kernel_param *kp)
8713 int num = simple_strtoul(val, &e, 10);
8714 if (*val && (*e == '\0' || *e == '\n')) {
8715 start_readonly = num;
8721 module_param_call(start_ro, set_ro, get_ro, NULL, S_IRUSR|S_IWUSR);
8722 module_param(start_dirty_degraded, int, S_IRUGO|S_IWUSR);
8724 module_param_call(new_array, add_named_array, NULL, NULL, S_IWUSR);
8726 EXPORT_SYMBOL(register_md_personality);
8727 EXPORT_SYMBOL(unregister_md_personality);
8728 EXPORT_SYMBOL(md_error);
8729 EXPORT_SYMBOL(md_done_sync);
8730 EXPORT_SYMBOL(md_write_start);
8731 EXPORT_SYMBOL(md_write_end);
8732 EXPORT_SYMBOL(md_register_thread);
8733 EXPORT_SYMBOL(md_unregister_thread);
8734 EXPORT_SYMBOL(md_wakeup_thread);
8735 EXPORT_SYMBOL(md_check_recovery);
8736 EXPORT_SYMBOL(md_reap_sync_thread);
8737 MODULE_LICENSE("GPL");
8738 MODULE_DESCRIPTION("MD RAID framework");
8740 MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR);