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_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(struct mddev *mddev, int nr)
635 struct md_rdev *rdev;
637 rdev_for_each(rdev, mddev)
638 if (rdev->desc_nr == nr)
644 static struct md_rdev *find_rdev_nr_rcu(struct mddev *mddev, int nr)
646 struct md_rdev *rdev;
648 rdev_for_each_rcu(rdev, mddev)
649 if (rdev->desc_nr == nr)
655 static struct md_rdev *find_rdev(struct mddev *mddev, dev_t dev)
657 struct md_rdev *rdev;
659 rdev_for_each(rdev, mddev)
660 if (rdev->bdev->bd_dev == dev)
666 static struct md_rdev *find_rdev_rcu(struct mddev *mddev, dev_t dev)
668 struct md_rdev *rdev;
670 rdev_for_each_rcu(rdev, mddev)
671 if (rdev->bdev->bd_dev == dev)
677 static struct md_personality *find_pers(int level, char *clevel)
679 struct md_personality *pers;
680 list_for_each_entry(pers, &pers_list, list) {
681 if (level != LEVEL_NONE && pers->level == level)
683 if (strcmp(pers->name, clevel)==0)
689 /* return the offset of the super block in 512byte sectors */
690 static inline sector_t calc_dev_sboffset(struct md_rdev *rdev)
692 sector_t num_sectors = i_size_read(rdev->bdev->bd_inode) / 512;
693 return MD_NEW_SIZE_SECTORS(num_sectors);
696 static int alloc_disk_sb(struct md_rdev * rdev)
701 rdev->sb_page = alloc_page(GFP_KERNEL);
702 if (!rdev->sb_page) {
703 printk(KERN_ALERT "md: out of memory.\n");
710 void md_rdev_clear(struct md_rdev *rdev)
713 put_page(rdev->sb_page);
715 rdev->sb_page = NULL;
720 put_page(rdev->bb_page);
721 rdev->bb_page = NULL;
723 kfree(rdev->badblocks.page);
724 rdev->badblocks.page = NULL;
726 EXPORT_SYMBOL_GPL(md_rdev_clear);
728 static void super_written(struct bio *bio, int error)
730 struct md_rdev *rdev = bio->bi_private;
731 struct mddev *mddev = rdev->mddev;
733 if (error || !test_bit(BIO_UPTODATE, &bio->bi_flags)) {
734 printk("md: super_written gets error=%d, uptodate=%d\n",
735 error, test_bit(BIO_UPTODATE, &bio->bi_flags));
736 WARN_ON(test_bit(BIO_UPTODATE, &bio->bi_flags));
737 md_error(mddev, rdev);
740 if (atomic_dec_and_test(&mddev->pending_writes))
741 wake_up(&mddev->sb_wait);
745 void md_super_write(struct mddev *mddev, struct md_rdev *rdev,
746 sector_t sector, int size, struct page *page)
748 /* write first size bytes of page to sector of rdev
749 * Increment mddev->pending_writes before returning
750 * and decrement it on completion, waking up sb_wait
751 * if zero is reached.
752 * If an error occurred, call md_error
754 struct bio *bio = bio_alloc_mddev(GFP_NOIO, 1, mddev);
756 bio->bi_bdev = rdev->meta_bdev ? rdev->meta_bdev : rdev->bdev;
757 bio->bi_sector = sector;
758 bio_add_page(bio, page, size, 0);
759 bio->bi_private = rdev;
760 bio->bi_end_io = super_written;
762 atomic_inc(&mddev->pending_writes);
763 submit_bio(WRITE_FLUSH_FUA, bio);
766 void md_super_wait(struct mddev *mddev)
768 /* wait for all superblock writes that were scheduled to complete */
771 prepare_to_wait(&mddev->sb_wait, &wq, TASK_UNINTERRUPTIBLE);
772 if (atomic_read(&mddev->pending_writes)==0)
776 finish_wait(&mddev->sb_wait, &wq);
779 static void bi_complete(struct bio *bio, int error)
781 complete((struct completion*)bio->bi_private);
784 int sync_page_io(struct md_rdev *rdev, sector_t sector, int size,
785 struct page *page, int rw, bool metadata_op)
787 struct bio *bio = bio_alloc_mddev(GFP_NOIO, 1, rdev->mddev);
788 struct completion event;
793 bio->bi_bdev = (metadata_op && rdev->meta_bdev) ?
794 rdev->meta_bdev : rdev->bdev;
796 bio->bi_sector = sector + rdev->sb_start;
797 else if (rdev->mddev->reshape_position != MaxSector &&
798 (rdev->mddev->reshape_backwards ==
799 (sector >= rdev->mddev->reshape_position)))
800 bio->bi_sector = sector + rdev->new_data_offset;
802 bio->bi_sector = sector + rdev->data_offset;
803 bio_add_page(bio, page, size, 0);
804 init_completion(&event);
805 bio->bi_private = &event;
806 bio->bi_end_io = bi_complete;
808 wait_for_completion(&event);
810 ret = test_bit(BIO_UPTODATE, &bio->bi_flags);
814 EXPORT_SYMBOL_GPL(sync_page_io);
816 static int read_disk_sb(struct md_rdev * rdev, int size)
818 char b[BDEVNAME_SIZE];
819 if (!rdev->sb_page) {
827 if (!sync_page_io(rdev, 0, size, rdev->sb_page, READ, true))
833 printk(KERN_WARNING "md: disabled device %s, could not read superblock.\n",
834 bdevname(rdev->bdev,b));
838 static int uuid_equal(mdp_super_t *sb1, mdp_super_t *sb2)
840 return sb1->set_uuid0 == sb2->set_uuid0 &&
841 sb1->set_uuid1 == sb2->set_uuid1 &&
842 sb1->set_uuid2 == sb2->set_uuid2 &&
843 sb1->set_uuid3 == sb2->set_uuid3;
846 static int sb_equal(mdp_super_t *sb1, mdp_super_t *sb2)
849 mdp_super_t *tmp1, *tmp2;
851 tmp1 = kmalloc(sizeof(*tmp1),GFP_KERNEL);
852 tmp2 = kmalloc(sizeof(*tmp2),GFP_KERNEL);
854 if (!tmp1 || !tmp2) {
856 printk(KERN_INFO "md.c sb_equal(): failed to allocate memory!\n");
864 * nr_disks is not constant
869 ret = (memcmp(tmp1, tmp2, MD_SB_GENERIC_CONSTANT_WORDS * 4) == 0);
877 static u32 md_csum_fold(u32 csum)
879 csum = (csum & 0xffff) + (csum >> 16);
880 return (csum & 0xffff) + (csum >> 16);
883 static unsigned int calc_sb_csum(mdp_super_t * sb)
886 u32 *sb32 = (u32*)sb;
888 unsigned int disk_csum, csum;
890 disk_csum = sb->sb_csum;
893 for (i = 0; i < MD_SB_BYTES/4 ; i++)
895 csum = (newcsum & 0xffffffff) + (newcsum>>32);
899 /* This used to use csum_partial, which was wrong for several
900 * reasons including that different results are returned on
901 * different architectures. It isn't critical that we get exactly
902 * the same return value as before (we always csum_fold before
903 * testing, and that removes any differences). However as we
904 * know that csum_partial always returned a 16bit value on
905 * alphas, do a fold to maximise conformity to previous behaviour.
907 sb->sb_csum = md_csum_fold(disk_csum);
909 sb->sb_csum = disk_csum;
916 * Handle superblock details.
917 * We want to be able to handle multiple superblock formats
918 * so we have a common interface to them all, and an array of
919 * different handlers.
920 * We rely on user-space to write the initial superblock, and support
921 * reading and updating of superblocks.
922 * Interface methods are:
923 * int load_super(struct md_rdev *dev, struct md_rdev *refdev, int minor_version)
924 * loads and validates a superblock on dev.
925 * if refdev != NULL, compare superblocks on both devices
927 * 0 - dev has a superblock that is compatible with refdev
928 * 1 - dev has a superblock that is compatible and newer than refdev
929 * so dev should be used as the refdev in future
930 * -EINVAL superblock incompatible or invalid
931 * -othererror e.g. -EIO
933 * int validate_super(struct mddev *mddev, struct md_rdev *dev)
934 * Verify that dev is acceptable into mddev.
935 * The first time, mddev->raid_disks will be 0, and data from
936 * dev should be merged in. Subsequent calls check that dev
937 * is new enough. Return 0 or -EINVAL
939 * void sync_super(struct mddev *mddev, struct md_rdev *dev)
940 * Update the superblock for rdev with data in mddev
941 * This does not write to disc.
947 struct module *owner;
948 int (*load_super)(struct md_rdev *rdev,
949 struct md_rdev *refdev,
951 int (*validate_super)(struct mddev *mddev,
952 struct md_rdev *rdev);
953 void (*sync_super)(struct mddev *mddev,
954 struct md_rdev *rdev);
955 unsigned long long (*rdev_size_change)(struct md_rdev *rdev,
956 sector_t num_sectors);
957 int (*allow_new_offset)(struct md_rdev *rdev,
958 unsigned long long new_offset);
962 * Check that the given mddev has no bitmap.
964 * This function is called from the run method of all personalities that do not
965 * support bitmaps. It prints an error message and returns non-zero if mddev
966 * has a bitmap. Otherwise, it returns 0.
969 int md_check_no_bitmap(struct mddev *mddev)
971 if (!mddev->bitmap_info.file && !mddev->bitmap_info.offset)
973 printk(KERN_ERR "%s: bitmaps are not supported for %s\n",
974 mdname(mddev), mddev->pers->name);
977 EXPORT_SYMBOL(md_check_no_bitmap);
980 * load_super for 0.90.0
982 static int super_90_load(struct md_rdev *rdev, struct md_rdev *refdev, int minor_version)
984 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
989 * Calculate the position of the superblock (512byte sectors),
990 * it's at the end of the disk.
992 * It also happens to be a multiple of 4Kb.
994 rdev->sb_start = calc_dev_sboffset(rdev);
996 ret = read_disk_sb(rdev, MD_SB_BYTES);
1001 bdevname(rdev->bdev, b);
1002 sb = page_address(rdev->sb_page);
1004 if (sb->md_magic != MD_SB_MAGIC) {
1005 printk(KERN_ERR "md: invalid raid superblock magic on %s\n",
1010 if (sb->major_version != 0 ||
1011 sb->minor_version < 90 ||
1012 sb->minor_version > 91) {
1013 printk(KERN_WARNING "Bad version number %d.%d on %s\n",
1014 sb->major_version, sb->minor_version,
1019 if (sb->raid_disks <= 0)
1022 if (md_csum_fold(calc_sb_csum(sb)) != md_csum_fold(sb->sb_csum)) {
1023 printk(KERN_WARNING "md: invalid superblock checksum on %s\n",
1028 rdev->preferred_minor = sb->md_minor;
1029 rdev->data_offset = 0;
1030 rdev->new_data_offset = 0;
1031 rdev->sb_size = MD_SB_BYTES;
1032 rdev->badblocks.shift = -1;
1034 if (sb->level == LEVEL_MULTIPATH)
1037 rdev->desc_nr = sb->this_disk.number;
1043 mdp_super_t *refsb = page_address(refdev->sb_page);
1044 if (!uuid_equal(refsb, sb)) {
1045 printk(KERN_WARNING "md: %s has different UUID to %s\n",
1046 b, bdevname(refdev->bdev,b2));
1049 if (!sb_equal(refsb, sb)) {
1050 printk(KERN_WARNING "md: %s has same UUID"
1051 " but different superblock to %s\n",
1052 b, bdevname(refdev->bdev, b2));
1056 ev2 = md_event(refsb);
1062 rdev->sectors = rdev->sb_start;
1063 /* Limit to 4TB as metadata cannot record more than that.
1064 * (not needed for Linear and RAID0 as metadata doesn't
1067 if (rdev->sectors >= (2ULL << 32) && sb->level >= 1)
1068 rdev->sectors = (2ULL << 32) - 2;
1070 if (rdev->sectors < ((sector_t)sb->size) * 2 && sb->level >= 1)
1071 /* "this cannot possibly happen" ... */
1079 * validate_super for 0.90.0
1081 static int super_90_validate(struct mddev *mddev, struct md_rdev *rdev)
1084 mdp_super_t *sb = page_address(rdev->sb_page);
1085 __u64 ev1 = md_event(sb);
1087 rdev->raid_disk = -1;
1088 clear_bit(Faulty, &rdev->flags);
1089 clear_bit(In_sync, &rdev->flags);
1090 clear_bit(WriteMostly, &rdev->flags);
1092 if (mddev->raid_disks == 0) {
1093 mddev->major_version = 0;
1094 mddev->minor_version = sb->minor_version;
1095 mddev->patch_version = sb->patch_version;
1096 mddev->external = 0;
1097 mddev->chunk_sectors = sb->chunk_size >> 9;
1098 mddev->ctime = sb->ctime;
1099 mddev->utime = sb->utime;
1100 mddev->level = sb->level;
1101 mddev->clevel[0] = 0;
1102 mddev->layout = sb->layout;
1103 mddev->raid_disks = sb->raid_disks;
1104 mddev->dev_sectors = ((sector_t)sb->size) * 2;
1105 mddev->events = ev1;
1106 mddev->bitmap_info.offset = 0;
1107 mddev->bitmap_info.space = 0;
1108 /* bitmap can use 60 K after the 4K superblocks */
1109 mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9;
1110 mddev->bitmap_info.default_space = 64*2 - (MD_SB_BYTES >> 9);
1111 mddev->reshape_backwards = 0;
1113 if (mddev->minor_version >= 91) {
1114 mddev->reshape_position = sb->reshape_position;
1115 mddev->delta_disks = sb->delta_disks;
1116 mddev->new_level = sb->new_level;
1117 mddev->new_layout = sb->new_layout;
1118 mddev->new_chunk_sectors = sb->new_chunk >> 9;
1119 if (mddev->delta_disks < 0)
1120 mddev->reshape_backwards = 1;
1122 mddev->reshape_position = MaxSector;
1123 mddev->delta_disks = 0;
1124 mddev->new_level = mddev->level;
1125 mddev->new_layout = mddev->layout;
1126 mddev->new_chunk_sectors = mddev->chunk_sectors;
1129 if (sb->state & (1<<MD_SB_CLEAN))
1130 mddev->recovery_cp = MaxSector;
1132 if (sb->events_hi == sb->cp_events_hi &&
1133 sb->events_lo == sb->cp_events_lo) {
1134 mddev->recovery_cp = sb->recovery_cp;
1136 mddev->recovery_cp = 0;
1139 memcpy(mddev->uuid+0, &sb->set_uuid0, 4);
1140 memcpy(mddev->uuid+4, &sb->set_uuid1, 4);
1141 memcpy(mddev->uuid+8, &sb->set_uuid2, 4);
1142 memcpy(mddev->uuid+12,&sb->set_uuid3, 4);
1144 mddev->max_disks = MD_SB_DISKS;
1146 if (sb->state & (1<<MD_SB_BITMAP_PRESENT) &&
1147 mddev->bitmap_info.file == NULL) {
1148 mddev->bitmap_info.offset =
1149 mddev->bitmap_info.default_offset;
1150 mddev->bitmap_info.space =
1151 mddev->bitmap_info.default_space;
1154 } else if (mddev->pers == NULL) {
1155 /* Insist on good event counter while assembling, except
1156 * for spares (which don't need an event count) */
1158 if (sb->disks[rdev->desc_nr].state & (
1159 (1<<MD_DISK_SYNC) | (1 << MD_DISK_ACTIVE)))
1160 if (ev1 < mddev->events)
1162 } else if (mddev->bitmap) {
1163 /* if adding to array with a bitmap, then we can accept an
1164 * older device ... but not too old.
1166 if (ev1 < mddev->bitmap->events_cleared)
1169 if (ev1 < mddev->events)
1170 /* just a hot-add of a new device, leave raid_disk at -1 */
1174 if (mddev->level != LEVEL_MULTIPATH) {
1175 desc = sb->disks + rdev->desc_nr;
1177 if (desc->state & (1<<MD_DISK_FAULTY))
1178 set_bit(Faulty, &rdev->flags);
1179 else if (desc->state & (1<<MD_DISK_SYNC) /* &&
1180 desc->raid_disk < mddev->raid_disks */) {
1181 set_bit(In_sync, &rdev->flags);
1182 rdev->raid_disk = desc->raid_disk;
1183 } else if (desc->state & (1<<MD_DISK_ACTIVE)) {
1184 /* active but not in sync implies recovery up to
1185 * reshape position. We don't know exactly where
1186 * that is, so set to zero for now */
1187 if (mddev->minor_version >= 91) {
1188 rdev->recovery_offset = 0;
1189 rdev->raid_disk = desc->raid_disk;
1192 if (desc->state & (1<<MD_DISK_WRITEMOSTLY))
1193 set_bit(WriteMostly, &rdev->flags);
1194 } else /* MULTIPATH are always insync */
1195 set_bit(In_sync, &rdev->flags);
1200 * sync_super for 0.90.0
1202 static void super_90_sync(struct mddev *mddev, struct md_rdev *rdev)
1205 struct md_rdev *rdev2;
1206 int next_spare = mddev->raid_disks;
1209 /* make rdev->sb match mddev data..
1212 * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
1213 * 3/ any empty disks < next_spare become removed
1215 * disks[0] gets initialised to REMOVED because
1216 * we cannot be sure from other fields if it has
1217 * been initialised or not.
1220 int active=0, working=0,failed=0,spare=0,nr_disks=0;
1222 rdev->sb_size = MD_SB_BYTES;
1224 sb = page_address(rdev->sb_page);
1226 memset(sb, 0, sizeof(*sb));
1228 sb->md_magic = MD_SB_MAGIC;
1229 sb->major_version = mddev->major_version;
1230 sb->patch_version = mddev->patch_version;
1231 sb->gvalid_words = 0; /* ignored */
1232 memcpy(&sb->set_uuid0, mddev->uuid+0, 4);
1233 memcpy(&sb->set_uuid1, mddev->uuid+4, 4);
1234 memcpy(&sb->set_uuid2, mddev->uuid+8, 4);
1235 memcpy(&sb->set_uuid3, mddev->uuid+12,4);
1237 sb->ctime = mddev->ctime;
1238 sb->level = mddev->level;
1239 sb->size = mddev->dev_sectors / 2;
1240 sb->raid_disks = mddev->raid_disks;
1241 sb->md_minor = mddev->md_minor;
1242 sb->not_persistent = 0;
1243 sb->utime = mddev->utime;
1245 sb->events_hi = (mddev->events>>32);
1246 sb->events_lo = (u32)mddev->events;
1248 if (mddev->reshape_position == MaxSector)
1249 sb->minor_version = 90;
1251 sb->minor_version = 91;
1252 sb->reshape_position = mddev->reshape_position;
1253 sb->new_level = mddev->new_level;
1254 sb->delta_disks = mddev->delta_disks;
1255 sb->new_layout = mddev->new_layout;
1256 sb->new_chunk = mddev->new_chunk_sectors << 9;
1258 mddev->minor_version = sb->minor_version;
1261 sb->recovery_cp = mddev->recovery_cp;
1262 sb->cp_events_hi = (mddev->events>>32);
1263 sb->cp_events_lo = (u32)mddev->events;
1264 if (mddev->recovery_cp == MaxSector)
1265 sb->state = (1<< MD_SB_CLEAN);
1267 sb->recovery_cp = 0;
1269 sb->layout = mddev->layout;
1270 sb->chunk_size = mddev->chunk_sectors << 9;
1272 if (mddev->bitmap && mddev->bitmap_info.file == NULL)
1273 sb->state |= (1<<MD_SB_BITMAP_PRESENT);
1275 sb->disks[0].state = (1<<MD_DISK_REMOVED);
1276 rdev_for_each(rdev2, mddev) {
1279 int is_active = test_bit(In_sync, &rdev2->flags);
1281 if (rdev2->raid_disk >= 0 &&
1282 sb->minor_version >= 91)
1283 /* we have nowhere to store the recovery_offset,
1284 * but if it is not below the reshape_position,
1285 * we can piggy-back on that.
1288 if (rdev2->raid_disk < 0 ||
1289 test_bit(Faulty, &rdev2->flags))
1292 desc_nr = rdev2->raid_disk;
1294 desc_nr = next_spare++;
1295 rdev2->desc_nr = desc_nr;
1296 d = &sb->disks[rdev2->desc_nr];
1298 d->number = rdev2->desc_nr;
1299 d->major = MAJOR(rdev2->bdev->bd_dev);
1300 d->minor = MINOR(rdev2->bdev->bd_dev);
1302 d->raid_disk = rdev2->raid_disk;
1304 d->raid_disk = rdev2->desc_nr; /* compatibility */
1305 if (test_bit(Faulty, &rdev2->flags))
1306 d->state = (1<<MD_DISK_FAULTY);
1307 else if (is_active) {
1308 d->state = (1<<MD_DISK_ACTIVE);
1309 if (test_bit(In_sync, &rdev2->flags))
1310 d->state |= (1<<MD_DISK_SYNC);
1318 if (test_bit(WriteMostly, &rdev2->flags))
1319 d->state |= (1<<MD_DISK_WRITEMOSTLY);
1321 /* now set the "removed" and "faulty" bits on any missing devices */
1322 for (i=0 ; i < mddev->raid_disks ; i++) {
1323 mdp_disk_t *d = &sb->disks[i];
1324 if (d->state == 0 && d->number == 0) {
1327 d->state = (1<<MD_DISK_REMOVED);
1328 d->state |= (1<<MD_DISK_FAULTY);
1332 sb->nr_disks = nr_disks;
1333 sb->active_disks = active;
1334 sb->working_disks = working;
1335 sb->failed_disks = failed;
1336 sb->spare_disks = spare;
1338 sb->this_disk = sb->disks[rdev->desc_nr];
1339 sb->sb_csum = calc_sb_csum(sb);
1343 * rdev_size_change for 0.90.0
1345 static unsigned long long
1346 super_90_rdev_size_change(struct md_rdev *rdev, sector_t num_sectors)
1348 if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
1349 return 0; /* component must fit device */
1350 if (rdev->mddev->bitmap_info.offset)
1351 return 0; /* can't move bitmap */
1352 rdev->sb_start = calc_dev_sboffset(rdev);
1353 if (!num_sectors || num_sectors > rdev->sb_start)
1354 num_sectors = rdev->sb_start;
1355 /* Limit to 4TB as metadata cannot record more than that.
1356 * 4TB == 2^32 KB, or 2*2^32 sectors.
1358 if (num_sectors >= (2ULL << 32) && rdev->mddev->level >= 1)
1359 num_sectors = (2ULL << 32) - 2;
1360 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
1362 md_super_wait(rdev->mddev);
1367 super_90_allow_new_offset(struct md_rdev *rdev, unsigned long long new_offset)
1369 /* non-zero offset changes not possible with v0.90 */
1370 return new_offset == 0;
1374 * version 1 superblock
1377 static __le32 calc_sb_1_csum(struct mdp_superblock_1 * sb)
1381 unsigned long long newcsum;
1382 int size = 256 + le32_to_cpu(sb->max_dev)*2;
1383 __le32 *isuper = (__le32*)sb;
1385 disk_csum = sb->sb_csum;
1388 for (; size >= 4; size -= 4)
1389 newcsum += le32_to_cpu(*isuper++);
1392 newcsum += le16_to_cpu(*(__le16*) isuper);
1394 csum = (newcsum & 0xffffffff) + (newcsum >> 32);
1395 sb->sb_csum = disk_csum;
1396 return cpu_to_le32(csum);
1399 static int md_set_badblocks(struct badblocks *bb, sector_t s, int sectors,
1401 static int super_1_load(struct md_rdev *rdev, struct md_rdev *refdev, int minor_version)
1403 struct mdp_superblock_1 *sb;
1407 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
1411 * Calculate the position of the superblock in 512byte sectors.
1412 * It is always aligned to a 4K boundary and
1413 * depeding on minor_version, it can be:
1414 * 0: At least 8K, but less than 12K, from end of device
1415 * 1: At start of device
1416 * 2: 4K from start of device.
1418 switch(minor_version) {
1420 sb_start = i_size_read(rdev->bdev->bd_inode) >> 9;
1422 sb_start &= ~(sector_t)(4*2-1);
1433 rdev->sb_start = sb_start;
1435 /* superblock is rarely larger than 1K, but it can be larger,
1436 * and it is safe to read 4k, so we do that
1438 ret = read_disk_sb(rdev, 4096);
1439 if (ret) return ret;
1442 sb = page_address(rdev->sb_page);
1444 if (sb->magic != cpu_to_le32(MD_SB_MAGIC) ||
1445 sb->major_version != cpu_to_le32(1) ||
1446 le32_to_cpu(sb->max_dev) > (4096-256)/2 ||
1447 le64_to_cpu(sb->super_offset) != rdev->sb_start ||
1448 (le32_to_cpu(sb->feature_map) & ~MD_FEATURE_ALL) != 0)
1451 if (calc_sb_1_csum(sb) != sb->sb_csum) {
1452 printk("md: invalid superblock checksum on %s\n",
1453 bdevname(rdev->bdev,b));
1456 if (le64_to_cpu(sb->data_size) < 10) {
1457 printk("md: data_size too small on %s\n",
1458 bdevname(rdev->bdev,b));
1463 memcmp(sb->pad3, sb->pad3+1, sizeof(sb->pad3) - sizeof(sb->pad3[1])))
1464 /* Some padding is non-zero, might be a new feature */
1467 rdev->preferred_minor = 0xffff;
1468 rdev->data_offset = le64_to_cpu(sb->data_offset);
1469 rdev->new_data_offset = rdev->data_offset;
1470 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE) &&
1471 (le32_to_cpu(sb->feature_map) & MD_FEATURE_NEW_OFFSET))
1472 rdev->new_data_offset += (s32)le32_to_cpu(sb->new_offset);
1473 atomic_set(&rdev->corrected_errors, le32_to_cpu(sb->cnt_corrected_read));
1475 rdev->sb_size = le32_to_cpu(sb->max_dev) * 2 + 256;
1476 bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
1477 if (rdev->sb_size & bmask)
1478 rdev->sb_size = (rdev->sb_size | bmask) + 1;
1481 && rdev->data_offset < sb_start + (rdev->sb_size/512))
1484 && rdev->new_data_offset < sb_start + (rdev->sb_size/512))
1487 if (sb->level == cpu_to_le32(LEVEL_MULTIPATH))
1490 rdev->desc_nr = le32_to_cpu(sb->dev_number);
1492 if (!rdev->bb_page) {
1493 rdev->bb_page = alloc_page(GFP_KERNEL);
1497 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BAD_BLOCKS) &&
1498 rdev->badblocks.count == 0) {
1499 /* need to load the bad block list.
1500 * Currently we limit it to one page.
1506 int sectors = le16_to_cpu(sb->bblog_size);
1507 if (sectors > (PAGE_SIZE / 512))
1509 offset = le32_to_cpu(sb->bblog_offset);
1512 bb_sector = (long long)offset;
1513 if (!sync_page_io(rdev, bb_sector, sectors << 9,
1514 rdev->bb_page, READ, true))
1516 bbp = (u64 *)page_address(rdev->bb_page);
1517 rdev->badblocks.shift = sb->bblog_shift;
1518 for (i = 0 ; i < (sectors << (9-3)) ; i++, bbp++) {
1519 u64 bb = le64_to_cpu(*bbp);
1520 int count = bb & (0x3ff);
1521 u64 sector = bb >> 10;
1522 sector <<= sb->bblog_shift;
1523 count <<= sb->bblog_shift;
1526 if (md_set_badblocks(&rdev->badblocks,
1527 sector, count, 1) == 0)
1530 } else if (sb->bblog_offset != 0)
1531 rdev->badblocks.shift = 0;
1537 struct mdp_superblock_1 *refsb = page_address(refdev->sb_page);
1539 if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 ||
1540 sb->level != refsb->level ||
1541 sb->layout != refsb->layout ||
1542 sb->chunksize != refsb->chunksize) {
1543 printk(KERN_WARNING "md: %s has strangely different"
1544 " superblock to %s\n",
1545 bdevname(rdev->bdev,b),
1546 bdevname(refdev->bdev,b2));
1549 ev1 = le64_to_cpu(sb->events);
1550 ev2 = le64_to_cpu(refsb->events);
1557 if (minor_version) {
1558 sectors = (i_size_read(rdev->bdev->bd_inode) >> 9);
1559 sectors -= rdev->data_offset;
1561 sectors = rdev->sb_start;
1562 if (sectors < le64_to_cpu(sb->data_size))
1564 rdev->sectors = le64_to_cpu(sb->data_size);
1568 static int super_1_validate(struct mddev *mddev, struct md_rdev *rdev)
1570 struct mdp_superblock_1 *sb = page_address(rdev->sb_page);
1571 __u64 ev1 = le64_to_cpu(sb->events);
1573 rdev->raid_disk = -1;
1574 clear_bit(Faulty, &rdev->flags);
1575 clear_bit(In_sync, &rdev->flags);
1576 clear_bit(WriteMostly, &rdev->flags);
1578 if (mddev->raid_disks == 0) {
1579 mddev->major_version = 1;
1580 mddev->patch_version = 0;
1581 mddev->external = 0;
1582 mddev->chunk_sectors = le32_to_cpu(sb->chunksize);
1583 mddev->ctime = le64_to_cpu(sb->ctime) & ((1ULL << 32)-1);
1584 mddev->utime = le64_to_cpu(sb->utime) & ((1ULL << 32)-1);
1585 mddev->level = le32_to_cpu(sb->level);
1586 mddev->clevel[0] = 0;
1587 mddev->layout = le32_to_cpu(sb->layout);
1588 mddev->raid_disks = le32_to_cpu(sb->raid_disks);
1589 mddev->dev_sectors = le64_to_cpu(sb->size);
1590 mddev->events = ev1;
1591 mddev->bitmap_info.offset = 0;
1592 mddev->bitmap_info.space = 0;
1593 /* Default location for bitmap is 1K after superblock
1594 * using 3K - total of 4K
1596 mddev->bitmap_info.default_offset = 1024 >> 9;
1597 mddev->bitmap_info.default_space = (4096-1024) >> 9;
1598 mddev->reshape_backwards = 0;
1600 mddev->recovery_cp = le64_to_cpu(sb->resync_offset);
1601 memcpy(mddev->uuid, sb->set_uuid, 16);
1603 mddev->max_disks = (4096-256)/2;
1605 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) &&
1606 mddev->bitmap_info.file == NULL) {
1607 mddev->bitmap_info.offset =
1608 (__s32)le32_to_cpu(sb->bitmap_offset);
1609 /* Metadata doesn't record how much space is available.
1610 * For 1.0, we assume we can use up to the superblock
1611 * if before, else to 4K beyond superblock.
1612 * For others, assume no change is possible.
1614 if (mddev->minor_version > 0)
1615 mddev->bitmap_info.space = 0;
1616 else if (mddev->bitmap_info.offset > 0)
1617 mddev->bitmap_info.space =
1618 8 - mddev->bitmap_info.offset;
1620 mddev->bitmap_info.space =
1621 -mddev->bitmap_info.offset;
1624 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) {
1625 mddev->reshape_position = le64_to_cpu(sb->reshape_position);
1626 mddev->delta_disks = le32_to_cpu(sb->delta_disks);
1627 mddev->new_level = le32_to_cpu(sb->new_level);
1628 mddev->new_layout = le32_to_cpu(sb->new_layout);
1629 mddev->new_chunk_sectors = le32_to_cpu(sb->new_chunk);
1630 if (mddev->delta_disks < 0 ||
1631 (mddev->delta_disks == 0 &&
1632 (le32_to_cpu(sb->feature_map)
1633 & MD_FEATURE_RESHAPE_BACKWARDS)))
1634 mddev->reshape_backwards = 1;
1636 mddev->reshape_position = MaxSector;
1637 mddev->delta_disks = 0;
1638 mddev->new_level = mddev->level;
1639 mddev->new_layout = mddev->layout;
1640 mddev->new_chunk_sectors = mddev->chunk_sectors;
1643 } else if (mddev->pers == NULL) {
1644 /* Insist of good event counter while assembling, except for
1645 * spares (which don't need an event count) */
1647 if (rdev->desc_nr >= 0 &&
1648 rdev->desc_nr < le32_to_cpu(sb->max_dev) &&
1649 le16_to_cpu(sb->dev_roles[rdev->desc_nr]) < 0xfffe)
1650 if (ev1 < mddev->events)
1652 } else if (mddev->bitmap) {
1653 /* If adding to array with a bitmap, then we can accept an
1654 * older device, but not too old.
1656 if (ev1 < mddev->bitmap->events_cleared)
1659 if (ev1 < mddev->events)
1660 /* just a hot-add of a new device, leave raid_disk at -1 */
1663 if (mddev->level != LEVEL_MULTIPATH) {
1665 if (rdev->desc_nr < 0 ||
1666 rdev->desc_nr >= le32_to_cpu(sb->max_dev)) {
1670 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
1672 case 0xffff: /* spare */
1674 case 0xfffe: /* faulty */
1675 set_bit(Faulty, &rdev->flags);
1678 if ((le32_to_cpu(sb->feature_map) &
1679 MD_FEATURE_RECOVERY_OFFSET))
1680 rdev->recovery_offset = le64_to_cpu(sb->recovery_offset);
1682 set_bit(In_sync, &rdev->flags);
1683 rdev->raid_disk = role;
1686 if (sb->devflags & WriteMostly1)
1687 set_bit(WriteMostly, &rdev->flags);
1688 if (le32_to_cpu(sb->feature_map) & MD_FEATURE_REPLACEMENT)
1689 set_bit(Replacement, &rdev->flags);
1690 } else /* MULTIPATH are always insync */
1691 set_bit(In_sync, &rdev->flags);
1696 static void super_1_sync(struct mddev *mddev, struct md_rdev *rdev)
1698 struct mdp_superblock_1 *sb;
1699 struct md_rdev *rdev2;
1701 /* make rdev->sb match mddev and rdev data. */
1703 sb = page_address(rdev->sb_page);
1705 sb->feature_map = 0;
1707 sb->recovery_offset = cpu_to_le64(0);
1708 memset(sb->pad3, 0, sizeof(sb->pad3));
1710 sb->utime = cpu_to_le64((__u64)mddev->utime);
1711 sb->events = cpu_to_le64(mddev->events);
1713 sb->resync_offset = cpu_to_le64(mddev->recovery_cp);
1715 sb->resync_offset = cpu_to_le64(0);
1717 sb->cnt_corrected_read = cpu_to_le32(atomic_read(&rdev->corrected_errors));
1719 sb->raid_disks = cpu_to_le32(mddev->raid_disks);
1720 sb->size = cpu_to_le64(mddev->dev_sectors);
1721 sb->chunksize = cpu_to_le32(mddev->chunk_sectors);
1722 sb->level = cpu_to_le32(mddev->level);
1723 sb->layout = cpu_to_le32(mddev->layout);
1725 if (test_bit(WriteMostly, &rdev->flags))
1726 sb->devflags |= WriteMostly1;
1728 sb->devflags &= ~WriteMostly1;
1729 sb->data_offset = cpu_to_le64(rdev->data_offset);
1730 sb->data_size = cpu_to_le64(rdev->sectors);
1732 if (mddev->bitmap && mddev->bitmap_info.file == NULL) {
1733 sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_info.offset);
1734 sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET);
1737 if (rdev->raid_disk >= 0 &&
1738 !test_bit(In_sync, &rdev->flags)) {
1740 cpu_to_le32(MD_FEATURE_RECOVERY_OFFSET);
1741 sb->recovery_offset =
1742 cpu_to_le64(rdev->recovery_offset);
1744 if (test_bit(Replacement, &rdev->flags))
1746 cpu_to_le32(MD_FEATURE_REPLACEMENT);
1748 if (mddev->reshape_position != MaxSector) {
1749 sb->feature_map |= cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE);
1750 sb->reshape_position = cpu_to_le64(mddev->reshape_position);
1751 sb->new_layout = cpu_to_le32(mddev->new_layout);
1752 sb->delta_disks = cpu_to_le32(mddev->delta_disks);
1753 sb->new_level = cpu_to_le32(mddev->new_level);
1754 sb->new_chunk = cpu_to_le32(mddev->new_chunk_sectors);
1755 if (mddev->delta_disks == 0 &&
1756 mddev->reshape_backwards)
1758 |= cpu_to_le32(MD_FEATURE_RESHAPE_BACKWARDS);
1759 if (rdev->new_data_offset != rdev->data_offset) {
1761 |= cpu_to_le32(MD_FEATURE_NEW_OFFSET);
1762 sb->new_offset = cpu_to_le32((__u32)(rdev->new_data_offset
1763 - rdev->data_offset));
1767 if (rdev->badblocks.count == 0)
1768 /* Nothing to do for bad blocks*/ ;
1769 else if (sb->bblog_offset == 0)
1770 /* Cannot record bad blocks on this device */
1771 md_error(mddev, rdev);
1773 struct badblocks *bb = &rdev->badblocks;
1774 u64 *bbp = (u64 *)page_address(rdev->bb_page);
1776 sb->feature_map |= cpu_to_le32(MD_FEATURE_BAD_BLOCKS);
1781 seq = read_seqbegin(&bb->lock);
1783 memset(bbp, 0xff, PAGE_SIZE);
1785 for (i = 0 ; i < bb->count ; i++) {
1786 u64 internal_bb = p[i];
1787 u64 store_bb = ((BB_OFFSET(internal_bb) << 10)
1788 | BB_LEN(internal_bb));
1789 bbp[i] = cpu_to_le64(store_bb);
1792 if (read_seqretry(&bb->lock, seq))
1795 bb->sector = (rdev->sb_start +
1796 (int)le32_to_cpu(sb->bblog_offset));
1797 bb->size = le16_to_cpu(sb->bblog_size);
1802 rdev_for_each(rdev2, mddev)
1803 if (rdev2->desc_nr+1 > max_dev)
1804 max_dev = rdev2->desc_nr+1;
1806 if (max_dev > le32_to_cpu(sb->max_dev)) {
1808 sb->max_dev = cpu_to_le32(max_dev);
1809 rdev->sb_size = max_dev * 2 + 256;
1810 bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
1811 if (rdev->sb_size & bmask)
1812 rdev->sb_size = (rdev->sb_size | bmask) + 1;
1814 max_dev = le32_to_cpu(sb->max_dev);
1816 for (i=0; i<max_dev;i++)
1817 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1819 rdev_for_each(rdev2, mddev) {
1821 if (test_bit(Faulty, &rdev2->flags))
1822 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1823 else if (test_bit(In_sync, &rdev2->flags))
1824 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1825 else if (rdev2->raid_disk >= 0)
1826 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1828 sb->dev_roles[i] = cpu_to_le16(0xffff);
1831 sb->sb_csum = calc_sb_1_csum(sb);
1834 static unsigned long long
1835 super_1_rdev_size_change(struct md_rdev *rdev, sector_t num_sectors)
1837 struct mdp_superblock_1 *sb;
1838 sector_t max_sectors;
1839 if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
1840 return 0; /* component must fit device */
1841 if (rdev->data_offset != rdev->new_data_offset)
1842 return 0; /* too confusing */
1843 if (rdev->sb_start < rdev->data_offset) {
1844 /* minor versions 1 and 2; superblock before data */
1845 max_sectors = i_size_read(rdev->bdev->bd_inode) >> 9;
1846 max_sectors -= rdev->data_offset;
1847 if (!num_sectors || num_sectors > max_sectors)
1848 num_sectors = max_sectors;
1849 } else if (rdev->mddev->bitmap_info.offset) {
1850 /* minor version 0 with bitmap we can't move */
1853 /* minor version 0; superblock after data */
1855 sb_start = (i_size_read(rdev->bdev->bd_inode) >> 9) - 8*2;
1856 sb_start &= ~(sector_t)(4*2 - 1);
1857 max_sectors = rdev->sectors + sb_start - rdev->sb_start;
1858 if (!num_sectors || num_sectors > max_sectors)
1859 num_sectors = max_sectors;
1860 rdev->sb_start = sb_start;
1862 sb = page_address(rdev->sb_page);
1863 sb->data_size = cpu_to_le64(num_sectors);
1864 sb->super_offset = rdev->sb_start;
1865 sb->sb_csum = calc_sb_1_csum(sb);
1866 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
1868 md_super_wait(rdev->mddev);
1874 super_1_allow_new_offset(struct md_rdev *rdev,
1875 unsigned long long new_offset)
1877 /* All necessary checks on new >= old have been done */
1878 struct bitmap *bitmap;
1879 if (new_offset >= rdev->data_offset)
1882 /* with 1.0 metadata, there is no metadata to tread on
1883 * so we can always move back */
1884 if (rdev->mddev->minor_version == 0)
1887 /* otherwise we must be sure not to step on
1888 * any metadata, so stay:
1889 * 36K beyond start of superblock
1890 * beyond end of badblocks
1891 * beyond write-intent bitmap
1893 if (rdev->sb_start + (32+4)*2 > new_offset)
1895 bitmap = rdev->mddev->bitmap;
1896 if (bitmap && !rdev->mddev->bitmap_info.file &&
1897 rdev->sb_start + rdev->mddev->bitmap_info.offset +
1898 bitmap->storage.file_pages * (PAGE_SIZE>>9) > new_offset)
1900 if (rdev->badblocks.sector + rdev->badblocks.size > new_offset)
1906 static struct super_type super_types[] = {
1909 .owner = THIS_MODULE,
1910 .load_super = super_90_load,
1911 .validate_super = super_90_validate,
1912 .sync_super = super_90_sync,
1913 .rdev_size_change = super_90_rdev_size_change,
1914 .allow_new_offset = super_90_allow_new_offset,
1918 .owner = THIS_MODULE,
1919 .load_super = super_1_load,
1920 .validate_super = super_1_validate,
1921 .sync_super = super_1_sync,
1922 .rdev_size_change = super_1_rdev_size_change,
1923 .allow_new_offset = super_1_allow_new_offset,
1927 static void sync_super(struct mddev *mddev, struct md_rdev *rdev)
1929 if (mddev->sync_super) {
1930 mddev->sync_super(mddev, rdev);
1934 BUG_ON(mddev->major_version >= ARRAY_SIZE(super_types));
1936 super_types[mddev->major_version].sync_super(mddev, rdev);
1939 static int match_mddev_units(struct mddev *mddev1, struct mddev *mddev2)
1941 struct md_rdev *rdev, *rdev2;
1944 rdev_for_each_rcu(rdev, mddev1)
1945 rdev_for_each_rcu(rdev2, mddev2)
1946 if (rdev->bdev->bd_contains ==
1947 rdev2->bdev->bd_contains) {
1955 static LIST_HEAD(pending_raid_disks);
1958 * Try to register data integrity profile for an mddev
1960 * This is called when an array is started and after a disk has been kicked
1961 * from the array. It only succeeds if all working and active component devices
1962 * are integrity capable with matching profiles.
1964 int md_integrity_register(struct mddev *mddev)
1966 struct md_rdev *rdev, *reference = NULL;
1968 if (list_empty(&mddev->disks))
1969 return 0; /* nothing to do */
1970 if (!mddev->gendisk || blk_get_integrity(mddev->gendisk))
1971 return 0; /* shouldn't register, or already is */
1972 rdev_for_each(rdev, mddev) {
1973 /* skip spares and non-functional disks */
1974 if (test_bit(Faulty, &rdev->flags))
1976 if (rdev->raid_disk < 0)
1979 /* Use the first rdev as the reference */
1983 /* does this rdev's profile match the reference profile? */
1984 if (blk_integrity_compare(reference->bdev->bd_disk,
1985 rdev->bdev->bd_disk) < 0)
1988 if (!reference || !bdev_get_integrity(reference->bdev))
1991 * All component devices are integrity capable and have matching
1992 * profiles, register the common profile for the md device.
1994 if (blk_integrity_register(mddev->gendisk,
1995 bdev_get_integrity(reference->bdev)) != 0) {
1996 printk(KERN_ERR "md: failed to register integrity for %s\n",
2000 printk(KERN_NOTICE "md: data integrity enabled on %s\n", mdname(mddev));
2001 if (bioset_integrity_create(mddev->bio_set, BIO_POOL_SIZE)) {
2002 printk(KERN_ERR "md: failed to create integrity pool for %s\n",
2008 EXPORT_SYMBOL(md_integrity_register);
2010 /* Disable data integrity if non-capable/non-matching disk is being added */
2011 void md_integrity_add_rdev(struct md_rdev *rdev, struct mddev *mddev)
2013 struct blk_integrity *bi_rdev;
2014 struct blk_integrity *bi_mddev;
2016 if (!mddev->gendisk)
2019 bi_rdev = bdev_get_integrity(rdev->bdev);
2020 bi_mddev = blk_get_integrity(mddev->gendisk);
2022 if (!bi_mddev) /* nothing to do */
2024 if (rdev->raid_disk < 0) /* skip spares */
2026 if (bi_rdev && blk_integrity_compare(mddev->gendisk,
2027 rdev->bdev->bd_disk) >= 0)
2029 printk(KERN_NOTICE "disabling data integrity on %s\n", mdname(mddev));
2030 blk_integrity_unregister(mddev->gendisk);
2032 EXPORT_SYMBOL(md_integrity_add_rdev);
2034 static int bind_rdev_to_array(struct md_rdev * rdev, struct mddev * mddev)
2036 char b[BDEVNAME_SIZE];
2046 /* prevent duplicates */
2047 if (find_rdev(mddev, rdev->bdev->bd_dev))
2050 /* make sure rdev->sectors exceeds mddev->dev_sectors */
2051 if (rdev->sectors && (mddev->dev_sectors == 0 ||
2052 rdev->sectors < mddev->dev_sectors)) {
2054 /* Cannot change size, so fail
2055 * If mddev->level <= 0, then we don't care
2056 * about aligning sizes (e.g. linear)
2058 if (mddev->level > 0)
2061 mddev->dev_sectors = rdev->sectors;
2064 /* Verify rdev->desc_nr is unique.
2065 * If it is -1, assign a free number, else
2066 * check number is not in use
2068 if (rdev->desc_nr < 0) {
2070 if (mddev->pers) choice = mddev->raid_disks;
2071 while (find_rdev_nr(mddev, choice))
2073 rdev->desc_nr = choice;
2075 if (find_rdev_nr(mddev, rdev->desc_nr))
2078 if (mddev->max_disks && rdev->desc_nr >= mddev->max_disks) {
2079 printk(KERN_WARNING "md: %s: array is limited to %d devices\n",
2080 mdname(mddev), mddev->max_disks);
2083 bdevname(rdev->bdev,b);
2084 while ( (s=strchr(b, '/')) != NULL)
2087 rdev->mddev = mddev;
2088 printk(KERN_INFO "md: bind<%s>\n", b);
2090 if ((err = kobject_add(&rdev->kobj, &mddev->kobj, "dev-%s", b)))
2093 ko = &part_to_dev(rdev->bdev->bd_part)->kobj;
2094 if (sysfs_create_link(&rdev->kobj, ko, "block"))
2095 /* failure here is OK */;
2096 rdev->sysfs_state = sysfs_get_dirent_safe(rdev->kobj.sd, "state");
2098 list_add_rcu(&rdev->same_set, &mddev->disks);
2099 bd_link_disk_holder(rdev->bdev, mddev->gendisk);
2101 /* May as well allow recovery to be retried once */
2102 mddev->recovery_disabled++;
2107 printk(KERN_WARNING "md: failed to register dev-%s for %s\n",
2112 static void md_delayed_delete(struct work_struct *ws)
2114 struct md_rdev *rdev = container_of(ws, struct md_rdev, del_work);
2115 kobject_del(&rdev->kobj);
2116 kobject_put(&rdev->kobj);
2119 static void unbind_rdev_from_array(struct md_rdev * rdev)
2121 char b[BDEVNAME_SIZE];
2126 bd_unlink_disk_holder(rdev->bdev, rdev->mddev->gendisk);
2127 list_del_rcu(&rdev->same_set);
2128 printk(KERN_INFO "md: unbind<%s>\n", bdevname(rdev->bdev,b));
2130 sysfs_remove_link(&rdev->kobj, "block");
2131 sysfs_put(rdev->sysfs_state);
2132 rdev->sysfs_state = NULL;
2133 rdev->badblocks.count = 0;
2134 /* We need to delay this, otherwise we can deadlock when
2135 * writing to 'remove' to "dev/state". We also need
2136 * to delay it due to rcu usage.
2139 INIT_WORK(&rdev->del_work, md_delayed_delete);
2140 kobject_get(&rdev->kobj);
2141 queue_work(md_misc_wq, &rdev->del_work);
2145 * prevent the device from being mounted, repartitioned or
2146 * otherwise reused by a RAID array (or any other kernel
2147 * subsystem), by bd_claiming the device.
2149 static int lock_rdev(struct md_rdev *rdev, dev_t dev, int shared)
2152 struct block_device *bdev;
2153 char b[BDEVNAME_SIZE];
2155 bdev = blkdev_get_by_dev(dev, FMODE_READ|FMODE_WRITE|FMODE_EXCL,
2156 shared ? (struct md_rdev *)lock_rdev : rdev);
2158 printk(KERN_ERR "md: could not open %s.\n",
2159 __bdevname(dev, b));
2160 return PTR_ERR(bdev);
2166 static void unlock_rdev(struct md_rdev *rdev)
2168 struct block_device *bdev = rdev->bdev;
2172 blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
2175 void md_autodetect_dev(dev_t dev);
2177 static void export_rdev(struct md_rdev * rdev)
2179 char b[BDEVNAME_SIZE];
2180 printk(KERN_INFO "md: export_rdev(%s)\n",
2181 bdevname(rdev->bdev,b));
2184 md_rdev_clear(rdev);
2186 if (test_bit(AutoDetected, &rdev->flags))
2187 md_autodetect_dev(rdev->bdev->bd_dev);
2190 kobject_put(&rdev->kobj);
2193 static void kick_rdev_from_array(struct md_rdev * rdev)
2195 unbind_rdev_from_array(rdev);
2199 static void export_array(struct mddev *mddev)
2201 struct md_rdev *rdev, *tmp;
2203 rdev_for_each_safe(rdev, tmp, mddev) {
2208 kick_rdev_from_array(rdev);
2210 if (!list_empty(&mddev->disks))
2212 mddev->raid_disks = 0;
2213 mddev->major_version = 0;
2216 static void print_desc(mdp_disk_t *desc)
2218 printk(" DISK<N:%d,(%d,%d),R:%d,S:%d>\n", desc->number,
2219 desc->major,desc->minor,desc->raid_disk,desc->state);
2222 static void print_sb_90(mdp_super_t *sb)
2227 "md: SB: (V:%d.%d.%d) ID:<%08x.%08x.%08x.%08x> CT:%08x\n",
2228 sb->major_version, sb->minor_version, sb->patch_version,
2229 sb->set_uuid0, sb->set_uuid1, sb->set_uuid2, sb->set_uuid3,
2231 printk(KERN_INFO "md: L%d S%08d ND:%d RD:%d md%d LO:%d CS:%d\n",
2232 sb->level, sb->size, sb->nr_disks, sb->raid_disks,
2233 sb->md_minor, sb->layout, sb->chunk_size);
2234 printk(KERN_INFO "md: UT:%08x ST:%d AD:%d WD:%d"
2235 " FD:%d SD:%d CSUM:%08x E:%08lx\n",
2236 sb->utime, sb->state, sb->active_disks, sb->working_disks,
2237 sb->failed_disks, sb->spare_disks,
2238 sb->sb_csum, (unsigned long)sb->events_lo);
2241 for (i = 0; i < MD_SB_DISKS; i++) {
2244 desc = sb->disks + i;
2245 if (desc->number || desc->major || desc->minor ||
2246 desc->raid_disk || (desc->state && (desc->state != 4))) {
2247 printk(" D %2d: ", i);
2251 printk(KERN_INFO "md: THIS: ");
2252 print_desc(&sb->this_disk);
2255 static void print_sb_1(struct mdp_superblock_1 *sb)
2259 uuid = sb->set_uuid;
2261 "md: SB: (V:%u) (F:0x%08x) Array-ID:<%pU>\n"
2262 "md: Name: \"%s\" CT:%llu\n",
2263 le32_to_cpu(sb->major_version),
2264 le32_to_cpu(sb->feature_map),
2267 (unsigned long long)le64_to_cpu(sb->ctime)
2268 & MD_SUPERBLOCK_1_TIME_SEC_MASK);
2270 uuid = sb->device_uuid;
2272 "md: L%u SZ%llu RD:%u LO:%u CS:%u DO:%llu DS:%llu SO:%llu"
2274 "md: Dev:%08x UUID: %pU\n"
2275 "md: (F:0x%08x) UT:%llu Events:%llu ResyncOffset:%llu CSUM:0x%08x\n"
2276 "md: (MaxDev:%u) \n",
2277 le32_to_cpu(sb->level),
2278 (unsigned long long)le64_to_cpu(sb->size),
2279 le32_to_cpu(sb->raid_disks),
2280 le32_to_cpu(sb->layout),
2281 le32_to_cpu(sb->chunksize),
2282 (unsigned long long)le64_to_cpu(sb->data_offset),
2283 (unsigned long long)le64_to_cpu(sb->data_size),
2284 (unsigned long long)le64_to_cpu(sb->super_offset),
2285 (unsigned long long)le64_to_cpu(sb->recovery_offset),
2286 le32_to_cpu(sb->dev_number),
2289 (unsigned long long)le64_to_cpu(sb->utime) & MD_SUPERBLOCK_1_TIME_SEC_MASK,
2290 (unsigned long long)le64_to_cpu(sb->events),
2291 (unsigned long long)le64_to_cpu(sb->resync_offset),
2292 le32_to_cpu(sb->sb_csum),
2293 le32_to_cpu(sb->max_dev)
2297 static void print_rdev(struct md_rdev *rdev, int major_version)
2299 char b[BDEVNAME_SIZE];
2300 printk(KERN_INFO "md: rdev %s, Sect:%08llu F:%d S:%d DN:%u\n",
2301 bdevname(rdev->bdev, b), (unsigned long long)rdev->sectors,
2302 test_bit(Faulty, &rdev->flags), test_bit(In_sync, &rdev->flags),
2304 if (rdev->sb_loaded) {
2305 printk(KERN_INFO "md: rdev superblock (MJ:%d):\n", major_version);
2306 switch (major_version) {
2308 print_sb_90(page_address(rdev->sb_page));
2311 print_sb_1(page_address(rdev->sb_page));
2315 printk(KERN_INFO "md: no rdev superblock!\n");
2318 static void md_print_devices(void)
2320 struct list_head *tmp;
2321 struct md_rdev *rdev;
2322 struct mddev *mddev;
2323 char b[BDEVNAME_SIZE];
2326 printk("md: **********************************\n");
2327 printk("md: * <COMPLETE RAID STATE PRINTOUT> *\n");
2328 printk("md: **********************************\n");
2329 for_each_mddev(mddev, tmp) {
2332 bitmap_print_sb(mddev->bitmap);
2334 printk("%s: ", mdname(mddev));
2335 rdev_for_each(rdev, mddev)
2336 printk("<%s>", bdevname(rdev->bdev,b));
2339 rdev_for_each(rdev, mddev)
2340 print_rdev(rdev, mddev->major_version);
2342 printk("md: **********************************\n");
2347 static void sync_sbs(struct mddev * mddev, int nospares)
2349 /* Update each superblock (in-memory image), but
2350 * if we are allowed to, skip spares which already
2351 * have the right event counter, or have one earlier
2352 * (which would mean they aren't being marked as dirty
2353 * with the rest of the array)
2355 struct md_rdev *rdev;
2356 rdev_for_each(rdev, mddev) {
2357 if (rdev->sb_events == mddev->events ||
2359 rdev->raid_disk < 0 &&
2360 rdev->sb_events+1 == mddev->events)) {
2361 /* Don't update this superblock */
2362 rdev->sb_loaded = 2;
2364 sync_super(mddev, rdev);
2365 rdev->sb_loaded = 1;
2370 static void md_update_sb(struct mddev * mddev, int force_change)
2372 struct md_rdev *rdev;
2375 int any_badblocks_changed = 0;
2379 set_bit(MD_CHANGE_DEVS, &mddev->flags);
2383 /* First make sure individual recovery_offsets are correct */
2384 rdev_for_each(rdev, mddev) {
2385 if (rdev->raid_disk >= 0 &&
2386 mddev->delta_disks >= 0 &&
2387 !test_bit(In_sync, &rdev->flags) &&
2388 mddev->curr_resync_completed > rdev->recovery_offset)
2389 rdev->recovery_offset = mddev->curr_resync_completed;
2392 if (!mddev->persistent) {
2393 clear_bit(MD_CHANGE_CLEAN, &mddev->flags);
2394 clear_bit(MD_CHANGE_DEVS, &mddev->flags);
2395 if (!mddev->external) {
2396 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
2397 rdev_for_each(rdev, mddev) {
2398 if (rdev->badblocks.changed) {
2399 rdev->badblocks.changed = 0;
2400 md_ack_all_badblocks(&rdev->badblocks);
2401 md_error(mddev, rdev);
2403 clear_bit(Blocked, &rdev->flags);
2404 clear_bit(BlockedBadBlocks, &rdev->flags);
2405 wake_up(&rdev->blocked_wait);
2408 wake_up(&mddev->sb_wait);
2412 spin_lock_irq(&mddev->write_lock);
2414 mddev->utime = get_seconds();
2416 if (test_and_clear_bit(MD_CHANGE_DEVS, &mddev->flags))
2418 if (test_and_clear_bit(MD_CHANGE_CLEAN, &mddev->flags))
2419 /* just a clean<-> dirty transition, possibly leave spares alone,
2420 * though if events isn't the right even/odd, we will have to do
2426 if (mddev->degraded)
2427 /* If the array is degraded, then skipping spares is both
2428 * dangerous and fairly pointless.
2429 * Dangerous because a device that was removed from the array
2430 * might have a event_count that still looks up-to-date,
2431 * so it can be re-added without a resync.
2432 * Pointless because if there are any spares to skip,
2433 * then a recovery will happen and soon that array won't
2434 * be degraded any more and the spare can go back to sleep then.
2438 sync_req = mddev->in_sync;
2440 /* If this is just a dirty<->clean transition, and the array is clean
2441 * and 'events' is odd, we can roll back to the previous clean state */
2443 && (mddev->in_sync && mddev->recovery_cp == MaxSector)
2444 && mddev->can_decrease_events
2445 && mddev->events != 1) {
2447 mddev->can_decrease_events = 0;
2449 /* otherwise we have to go forward and ... */
2451 mddev->can_decrease_events = nospares;
2454 if (!mddev->events) {
2456 * oops, this 64-bit counter should never wrap.
2457 * Either we are in around ~1 trillion A.C., assuming
2458 * 1 reboot per second, or we have a bug:
2464 rdev_for_each(rdev, mddev) {
2465 if (rdev->badblocks.changed)
2466 any_badblocks_changed++;
2467 if (test_bit(Faulty, &rdev->flags))
2468 set_bit(FaultRecorded, &rdev->flags);
2471 sync_sbs(mddev, nospares);
2472 spin_unlock_irq(&mddev->write_lock);
2474 pr_debug("md: updating %s RAID superblock on device (in sync %d)\n",
2475 mdname(mddev), mddev->in_sync);
2477 bitmap_update_sb(mddev->bitmap);
2478 rdev_for_each(rdev, mddev) {
2479 char b[BDEVNAME_SIZE];
2481 if (rdev->sb_loaded != 1)
2482 continue; /* no noise on spare devices */
2484 if (!test_bit(Faulty, &rdev->flags) &&
2485 rdev->saved_raid_disk == -1) {
2486 md_super_write(mddev,rdev,
2487 rdev->sb_start, rdev->sb_size,
2489 pr_debug("md: (write) %s's sb offset: %llu\n",
2490 bdevname(rdev->bdev, b),
2491 (unsigned long long)rdev->sb_start);
2492 rdev->sb_events = mddev->events;
2493 if (rdev->badblocks.size) {
2494 md_super_write(mddev, rdev,
2495 rdev->badblocks.sector,
2496 rdev->badblocks.size << 9,
2498 rdev->badblocks.size = 0;
2501 } else if (test_bit(Faulty, &rdev->flags))
2502 pr_debug("md: %s (skipping faulty)\n",
2503 bdevname(rdev->bdev, b));
2505 pr_debug("(skipping incremental s/r ");
2507 if (mddev->level == LEVEL_MULTIPATH)
2508 /* only need to write one superblock... */
2511 md_super_wait(mddev);
2512 /* if there was a failure, MD_CHANGE_DEVS was set, and we re-write super */
2514 spin_lock_irq(&mddev->write_lock);
2515 if (mddev->in_sync != sync_req ||
2516 test_bit(MD_CHANGE_DEVS, &mddev->flags)) {
2517 /* have to write it out again */
2518 spin_unlock_irq(&mddev->write_lock);
2521 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
2522 spin_unlock_irq(&mddev->write_lock);
2523 wake_up(&mddev->sb_wait);
2524 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
2525 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
2527 rdev_for_each(rdev, mddev) {
2528 if (test_and_clear_bit(FaultRecorded, &rdev->flags))
2529 clear_bit(Blocked, &rdev->flags);
2531 if (any_badblocks_changed)
2532 md_ack_all_badblocks(&rdev->badblocks);
2533 clear_bit(BlockedBadBlocks, &rdev->flags);
2534 wake_up(&rdev->blocked_wait);
2538 /* words written to sysfs files may, or may not, be \n terminated.
2539 * We want to accept with case. For this we use cmd_match.
2541 static int cmd_match(const char *cmd, const char *str)
2543 /* See if cmd, written into a sysfs file, matches
2544 * str. They must either be the same, or cmd can
2545 * have a trailing newline
2547 while (*cmd && *str && *cmd == *str) {
2558 struct rdev_sysfs_entry {
2559 struct attribute attr;
2560 ssize_t (*show)(struct md_rdev *, char *);
2561 ssize_t (*store)(struct md_rdev *, const char *, size_t);
2565 state_show(struct md_rdev *rdev, char *page)
2570 if (test_bit(Faulty, &rdev->flags) ||
2571 rdev->badblocks.unacked_exist) {
2572 len+= sprintf(page+len, "%sfaulty",sep);
2575 if (test_bit(In_sync, &rdev->flags)) {
2576 len += sprintf(page+len, "%sin_sync",sep);
2579 if (test_bit(WriteMostly, &rdev->flags)) {
2580 len += sprintf(page+len, "%swrite_mostly",sep);
2583 if (test_bit(Blocked, &rdev->flags) ||
2584 (rdev->badblocks.unacked_exist
2585 && !test_bit(Faulty, &rdev->flags))) {
2586 len += sprintf(page+len, "%sblocked", sep);
2589 if (!test_bit(Faulty, &rdev->flags) &&
2590 !test_bit(In_sync, &rdev->flags)) {
2591 len += sprintf(page+len, "%sspare", sep);
2594 if (test_bit(WriteErrorSeen, &rdev->flags)) {
2595 len += sprintf(page+len, "%swrite_error", sep);
2598 if (test_bit(WantReplacement, &rdev->flags)) {
2599 len += sprintf(page+len, "%swant_replacement", sep);
2602 if (test_bit(Replacement, &rdev->flags)) {
2603 len += sprintf(page+len, "%sreplacement", sep);
2607 return len+sprintf(page+len, "\n");
2611 state_store(struct md_rdev *rdev, const char *buf, size_t len)
2614 * faulty - simulates an error
2615 * remove - disconnects the device
2616 * writemostly - sets write_mostly
2617 * -writemostly - clears write_mostly
2618 * blocked - sets the Blocked flags
2619 * -blocked - clears the Blocked and possibly simulates an error
2620 * insync - sets Insync providing device isn't active
2621 * write_error - sets WriteErrorSeen
2622 * -write_error - clears WriteErrorSeen
2625 if (cmd_match(buf, "faulty") && rdev->mddev->pers) {
2626 md_error(rdev->mddev, rdev);
2627 if (test_bit(Faulty, &rdev->flags))
2631 } else if (cmd_match(buf, "remove")) {
2632 if (rdev->raid_disk >= 0)
2635 struct mddev *mddev = rdev->mddev;
2636 kick_rdev_from_array(rdev);
2638 md_update_sb(mddev, 1);
2639 md_new_event(mddev);
2642 } else if (cmd_match(buf, "writemostly")) {
2643 set_bit(WriteMostly, &rdev->flags);
2645 } else if (cmd_match(buf, "-writemostly")) {
2646 clear_bit(WriteMostly, &rdev->flags);
2648 } else if (cmd_match(buf, "blocked")) {
2649 set_bit(Blocked, &rdev->flags);
2651 } else if (cmd_match(buf, "-blocked")) {
2652 if (!test_bit(Faulty, &rdev->flags) &&
2653 rdev->badblocks.unacked_exist) {
2654 /* metadata handler doesn't understand badblocks,
2655 * so we need to fail the device
2657 md_error(rdev->mddev, rdev);
2659 clear_bit(Blocked, &rdev->flags);
2660 clear_bit(BlockedBadBlocks, &rdev->flags);
2661 wake_up(&rdev->blocked_wait);
2662 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2663 md_wakeup_thread(rdev->mddev->thread);
2666 } else if (cmd_match(buf, "insync") && rdev->raid_disk == -1) {
2667 set_bit(In_sync, &rdev->flags);
2669 } else if (cmd_match(buf, "write_error")) {
2670 set_bit(WriteErrorSeen, &rdev->flags);
2672 } else if (cmd_match(buf, "-write_error")) {
2673 clear_bit(WriteErrorSeen, &rdev->flags);
2675 } else if (cmd_match(buf, "want_replacement")) {
2676 /* Any non-spare device that is not a replacement can
2677 * become want_replacement at any time, but we then need to
2678 * check if recovery is needed.
2680 if (rdev->raid_disk >= 0 &&
2681 !test_bit(Replacement, &rdev->flags))
2682 set_bit(WantReplacement, &rdev->flags);
2683 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2684 md_wakeup_thread(rdev->mddev->thread);
2686 } else if (cmd_match(buf, "-want_replacement")) {
2687 /* Clearing 'want_replacement' is always allowed.
2688 * Once replacements starts it is too late though.
2691 clear_bit(WantReplacement, &rdev->flags);
2692 } else if (cmd_match(buf, "replacement")) {
2693 /* Can only set a device as a replacement when array has not
2694 * yet been started. Once running, replacement is automatic
2695 * from spares, or by assigning 'slot'.
2697 if (rdev->mddev->pers)
2700 set_bit(Replacement, &rdev->flags);
2703 } else if (cmd_match(buf, "-replacement")) {
2704 /* Similarly, can only clear Replacement before start */
2705 if (rdev->mddev->pers)
2708 clear_bit(Replacement, &rdev->flags);
2713 sysfs_notify_dirent_safe(rdev->sysfs_state);
2714 return err ? err : len;
2716 static struct rdev_sysfs_entry rdev_state =
2717 __ATTR(state, S_IRUGO|S_IWUSR, state_show, state_store);
2720 errors_show(struct md_rdev *rdev, char *page)
2722 return sprintf(page, "%d\n", atomic_read(&rdev->corrected_errors));
2726 errors_store(struct md_rdev *rdev, const char *buf, size_t len)
2729 unsigned long n = simple_strtoul(buf, &e, 10);
2730 if (*buf && (*e == 0 || *e == '\n')) {
2731 atomic_set(&rdev->corrected_errors, n);
2736 static struct rdev_sysfs_entry rdev_errors =
2737 __ATTR(errors, S_IRUGO|S_IWUSR, errors_show, errors_store);
2740 slot_show(struct md_rdev *rdev, char *page)
2742 if (rdev->raid_disk < 0)
2743 return sprintf(page, "none\n");
2745 return sprintf(page, "%d\n", rdev->raid_disk);
2749 slot_store(struct md_rdev *rdev, const char *buf, size_t len)
2753 int slot = simple_strtoul(buf, &e, 10);
2754 if (strncmp(buf, "none", 4)==0)
2756 else if (e==buf || (*e && *e!= '\n'))
2758 if (rdev->mddev->pers && slot == -1) {
2759 /* Setting 'slot' on an active array requires also
2760 * updating the 'rd%d' link, and communicating
2761 * with the personality with ->hot_*_disk.
2762 * For now we only support removing
2763 * failed/spare devices. This normally happens automatically,
2764 * but not when the metadata is externally managed.
2766 if (rdev->raid_disk == -1)
2768 /* personality does all needed checks */
2769 if (rdev->mddev->pers->hot_remove_disk == NULL)
2771 clear_bit(Blocked, &rdev->flags);
2772 remove_and_add_spares(rdev->mddev, rdev);
2773 if (rdev->raid_disk >= 0)
2775 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2776 md_wakeup_thread(rdev->mddev->thread);
2777 } else if (rdev->mddev->pers) {
2778 /* Activating a spare .. or possibly reactivating
2779 * if we ever get bitmaps working here.
2782 if (rdev->raid_disk != -1)
2785 if (test_bit(MD_RECOVERY_RUNNING, &rdev->mddev->recovery))
2788 if (rdev->mddev->pers->hot_add_disk == NULL)
2791 if (slot >= rdev->mddev->raid_disks &&
2792 slot >= rdev->mddev->raid_disks + rdev->mddev->delta_disks)
2795 rdev->raid_disk = slot;
2796 if (test_bit(In_sync, &rdev->flags))
2797 rdev->saved_raid_disk = slot;
2799 rdev->saved_raid_disk = -1;
2800 clear_bit(In_sync, &rdev->flags);
2801 err = rdev->mddev->pers->
2802 hot_add_disk(rdev->mddev, rdev);
2804 rdev->raid_disk = -1;
2807 sysfs_notify_dirent_safe(rdev->sysfs_state);
2808 if (sysfs_link_rdev(rdev->mddev, rdev))
2809 /* failure here is OK */;
2810 /* don't wakeup anyone, leave that to userspace. */
2812 if (slot >= rdev->mddev->raid_disks &&
2813 slot >= rdev->mddev->raid_disks + rdev->mddev->delta_disks)
2815 rdev->raid_disk = slot;
2816 /* assume it is working */
2817 clear_bit(Faulty, &rdev->flags);
2818 clear_bit(WriteMostly, &rdev->flags);
2819 set_bit(In_sync, &rdev->flags);
2820 sysfs_notify_dirent_safe(rdev->sysfs_state);
2826 static struct rdev_sysfs_entry rdev_slot =
2827 __ATTR(slot, S_IRUGO|S_IWUSR, slot_show, slot_store);
2830 offset_show(struct md_rdev *rdev, char *page)
2832 return sprintf(page, "%llu\n", (unsigned long long)rdev->data_offset);
2836 offset_store(struct md_rdev *rdev, const char *buf, size_t len)
2838 unsigned long long offset;
2839 if (kstrtoull(buf, 10, &offset) < 0)
2841 if (rdev->mddev->pers && rdev->raid_disk >= 0)
2843 if (rdev->sectors && rdev->mddev->external)
2844 /* Must set offset before size, so overlap checks
2847 rdev->data_offset = offset;
2848 rdev->new_data_offset = offset;
2852 static struct rdev_sysfs_entry rdev_offset =
2853 __ATTR(offset, S_IRUGO|S_IWUSR, offset_show, offset_store);
2855 static ssize_t new_offset_show(struct md_rdev *rdev, char *page)
2857 return sprintf(page, "%llu\n",
2858 (unsigned long long)rdev->new_data_offset);
2861 static ssize_t new_offset_store(struct md_rdev *rdev,
2862 const char *buf, size_t len)
2864 unsigned long long new_offset;
2865 struct mddev *mddev = rdev->mddev;
2867 if (kstrtoull(buf, 10, &new_offset) < 0)
2870 if (mddev->sync_thread)
2872 if (new_offset == rdev->data_offset)
2873 /* reset is always permitted */
2875 else if (new_offset > rdev->data_offset) {
2876 /* must not push array size beyond rdev_sectors */
2877 if (new_offset - rdev->data_offset
2878 + mddev->dev_sectors > rdev->sectors)
2881 /* Metadata worries about other space details. */
2883 /* decreasing the offset is inconsistent with a backwards
2886 if (new_offset < rdev->data_offset &&
2887 mddev->reshape_backwards)
2889 /* Increasing offset is inconsistent with forwards
2890 * reshape. reshape_direction should be set to
2891 * 'backwards' first.
2893 if (new_offset > rdev->data_offset &&
2894 !mddev->reshape_backwards)
2897 if (mddev->pers && mddev->persistent &&
2898 !super_types[mddev->major_version]
2899 .allow_new_offset(rdev, new_offset))
2901 rdev->new_data_offset = new_offset;
2902 if (new_offset > rdev->data_offset)
2903 mddev->reshape_backwards = 1;
2904 else if (new_offset < rdev->data_offset)
2905 mddev->reshape_backwards = 0;
2909 static struct rdev_sysfs_entry rdev_new_offset =
2910 __ATTR(new_offset, S_IRUGO|S_IWUSR, new_offset_show, new_offset_store);
2913 rdev_size_show(struct md_rdev *rdev, char *page)
2915 return sprintf(page, "%llu\n", (unsigned long long)rdev->sectors / 2);
2918 static int overlaps(sector_t s1, sector_t l1, sector_t s2, sector_t l2)
2920 /* check if two start/length pairs overlap */
2928 static int strict_blocks_to_sectors(const char *buf, sector_t *sectors)
2930 unsigned long long blocks;
2933 if (kstrtoull(buf, 10, &blocks) < 0)
2936 if (blocks & 1ULL << (8 * sizeof(blocks) - 1))
2937 return -EINVAL; /* sector conversion overflow */
2940 if (new != blocks * 2)
2941 return -EINVAL; /* unsigned long long to sector_t overflow */
2948 rdev_size_store(struct md_rdev *rdev, const char *buf, size_t len)
2950 struct mddev *my_mddev = rdev->mddev;
2951 sector_t oldsectors = rdev->sectors;
2954 if (strict_blocks_to_sectors(buf, §ors) < 0)
2956 if (rdev->data_offset != rdev->new_data_offset)
2957 return -EINVAL; /* too confusing */
2958 if (my_mddev->pers && rdev->raid_disk >= 0) {
2959 if (my_mddev->persistent) {
2960 sectors = super_types[my_mddev->major_version].
2961 rdev_size_change(rdev, sectors);
2964 } else if (!sectors)
2965 sectors = (i_size_read(rdev->bdev->bd_inode) >> 9) -
2967 if (!my_mddev->pers->resize)
2968 /* Cannot change size for RAID0 or Linear etc */
2971 if (sectors < my_mddev->dev_sectors)
2972 return -EINVAL; /* component must fit device */
2974 rdev->sectors = sectors;
2975 if (sectors > oldsectors && my_mddev->external) {
2976 /* need to check that all other rdevs with the same ->bdev
2977 * do not overlap. We need to unlock the mddev to avoid
2978 * a deadlock. We have already changed rdev->sectors, and if
2979 * we have to change it back, we will have the lock again.
2981 struct mddev *mddev;
2983 struct list_head *tmp;
2985 mddev_unlock(my_mddev);
2986 for_each_mddev(mddev, tmp) {
2987 struct md_rdev *rdev2;
2989 mddev_lock_nointr(mddev);
2990 rdev_for_each(rdev2, mddev)
2991 if (rdev->bdev == rdev2->bdev &&
2993 overlaps(rdev->data_offset, rdev->sectors,
2999 mddev_unlock(mddev);
3005 mddev_lock_nointr(my_mddev);
3007 /* Someone else could have slipped in a size
3008 * change here, but doing so is just silly.
3009 * We put oldsectors back because we *know* it is
3010 * safe, and trust userspace not to race with
3013 rdev->sectors = oldsectors;
3020 static struct rdev_sysfs_entry rdev_size =
3021 __ATTR(size, S_IRUGO|S_IWUSR, rdev_size_show, rdev_size_store);
3024 static ssize_t recovery_start_show(struct md_rdev *rdev, char *page)
3026 unsigned long long recovery_start = rdev->recovery_offset;
3028 if (test_bit(In_sync, &rdev->flags) ||
3029 recovery_start == MaxSector)
3030 return sprintf(page, "none\n");
3032 return sprintf(page, "%llu\n", recovery_start);
3035 static ssize_t recovery_start_store(struct md_rdev *rdev, const char *buf, size_t len)
3037 unsigned long long recovery_start;
3039 if (cmd_match(buf, "none"))
3040 recovery_start = MaxSector;
3041 else if (kstrtoull(buf, 10, &recovery_start))
3044 if (rdev->mddev->pers &&
3045 rdev->raid_disk >= 0)
3048 rdev->recovery_offset = recovery_start;
3049 if (recovery_start == MaxSector)
3050 set_bit(In_sync, &rdev->flags);
3052 clear_bit(In_sync, &rdev->flags);
3056 static struct rdev_sysfs_entry rdev_recovery_start =
3057 __ATTR(recovery_start, S_IRUGO|S_IWUSR, recovery_start_show, recovery_start_store);
3061 badblocks_show(struct badblocks *bb, char *page, int unack);
3063 badblocks_store(struct badblocks *bb, const char *page, size_t len, int unack);
3065 static ssize_t bb_show(struct md_rdev *rdev, char *page)
3067 return badblocks_show(&rdev->badblocks, page, 0);
3069 static ssize_t bb_store(struct md_rdev *rdev, const char *page, size_t len)
3071 int rv = badblocks_store(&rdev->badblocks, page, len, 0);
3072 /* Maybe that ack was all we needed */
3073 if (test_and_clear_bit(BlockedBadBlocks, &rdev->flags))
3074 wake_up(&rdev->blocked_wait);
3077 static struct rdev_sysfs_entry rdev_bad_blocks =
3078 __ATTR(bad_blocks, S_IRUGO|S_IWUSR, bb_show, bb_store);
3081 static ssize_t ubb_show(struct md_rdev *rdev, char *page)
3083 return badblocks_show(&rdev->badblocks, page, 1);
3085 static ssize_t ubb_store(struct md_rdev *rdev, const char *page, size_t len)
3087 return badblocks_store(&rdev->badblocks, page, len, 1);
3089 static struct rdev_sysfs_entry rdev_unack_bad_blocks =
3090 __ATTR(unacknowledged_bad_blocks, S_IRUGO|S_IWUSR, ubb_show, ubb_store);
3092 static struct attribute *rdev_default_attrs[] = {
3097 &rdev_new_offset.attr,
3099 &rdev_recovery_start.attr,
3100 &rdev_bad_blocks.attr,
3101 &rdev_unack_bad_blocks.attr,
3105 rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
3107 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
3108 struct md_rdev *rdev = container_of(kobj, struct md_rdev, kobj);
3109 struct mddev *mddev = rdev->mddev;
3115 rv = mddev ? mddev_lock(mddev) : -EBUSY;
3117 if (rdev->mddev == NULL)
3120 rv = entry->show(rdev, page);
3121 mddev_unlock(mddev);
3127 rdev_attr_store(struct kobject *kobj, struct attribute *attr,
3128 const char *page, size_t length)
3130 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
3131 struct md_rdev *rdev = container_of(kobj, struct md_rdev, kobj);
3133 struct mddev *mddev = rdev->mddev;
3137 if (!capable(CAP_SYS_ADMIN))
3139 rv = mddev ? mddev_lock(mddev): -EBUSY;
3141 if (rdev->mddev == NULL)
3144 rv = entry->store(rdev, page, length);
3145 mddev_unlock(mddev);
3150 static void rdev_free(struct kobject *ko)
3152 struct md_rdev *rdev = container_of(ko, struct md_rdev, kobj);
3155 static const struct sysfs_ops rdev_sysfs_ops = {
3156 .show = rdev_attr_show,
3157 .store = rdev_attr_store,
3159 static struct kobj_type rdev_ktype = {
3160 .release = rdev_free,
3161 .sysfs_ops = &rdev_sysfs_ops,
3162 .default_attrs = rdev_default_attrs,
3165 int md_rdev_init(struct md_rdev *rdev)
3168 rdev->saved_raid_disk = -1;
3169 rdev->raid_disk = -1;
3171 rdev->data_offset = 0;
3172 rdev->new_data_offset = 0;
3173 rdev->sb_events = 0;
3174 rdev->last_read_error.tv_sec = 0;
3175 rdev->last_read_error.tv_nsec = 0;
3176 rdev->sb_loaded = 0;
3177 rdev->bb_page = NULL;
3178 atomic_set(&rdev->nr_pending, 0);
3179 atomic_set(&rdev->read_errors, 0);
3180 atomic_set(&rdev->corrected_errors, 0);
3182 INIT_LIST_HEAD(&rdev->same_set);
3183 init_waitqueue_head(&rdev->blocked_wait);
3185 /* Add space to store bad block list.
3186 * This reserves the space even on arrays where it cannot
3187 * be used - I wonder if that matters
3189 rdev->badblocks.count = 0;
3190 rdev->badblocks.shift = -1; /* disabled until explicitly enabled */
3191 rdev->badblocks.page = kmalloc(PAGE_SIZE, GFP_KERNEL);
3192 seqlock_init(&rdev->badblocks.lock);
3193 if (rdev->badblocks.page == NULL)
3198 EXPORT_SYMBOL_GPL(md_rdev_init);
3200 * Import a device. If 'super_format' >= 0, then sanity check the superblock
3202 * mark the device faulty if:
3204 * - the device is nonexistent (zero size)
3205 * - the device has no valid superblock
3207 * a faulty rdev _never_ has rdev->sb set.
3209 static struct md_rdev *md_import_device(dev_t newdev, int super_format, int super_minor)
3211 char b[BDEVNAME_SIZE];
3213 struct md_rdev *rdev;
3216 rdev = kzalloc(sizeof(*rdev), GFP_KERNEL);
3218 printk(KERN_ERR "md: could not alloc mem for new device!\n");
3219 return ERR_PTR(-ENOMEM);
3222 err = md_rdev_init(rdev);
3225 err = alloc_disk_sb(rdev);
3229 err = lock_rdev(rdev, newdev, super_format == -2);
3233 kobject_init(&rdev->kobj, &rdev_ktype);
3235 size = i_size_read(rdev->bdev->bd_inode) >> BLOCK_SIZE_BITS;
3238 "md: %s has zero or unknown size, marking faulty!\n",
3239 bdevname(rdev->bdev,b));
3244 if (super_format >= 0) {
3245 err = super_types[super_format].
3246 load_super(rdev, NULL, super_minor);
3247 if (err == -EINVAL) {
3249 "md: %s does not have a valid v%d.%d "
3250 "superblock, not importing!\n",
3251 bdevname(rdev->bdev,b),
3252 super_format, super_minor);
3257 "md: could not read %s's sb, not importing!\n",
3258 bdevname(rdev->bdev,b));
3268 md_rdev_clear(rdev);
3270 return ERR_PTR(err);
3274 * Check a full RAID array for plausibility
3278 static void analyze_sbs(struct mddev * mddev)
3281 struct md_rdev *rdev, *freshest, *tmp;
3282 char b[BDEVNAME_SIZE];
3285 rdev_for_each_safe(rdev, tmp, mddev)
3286 switch (super_types[mddev->major_version].
3287 load_super(rdev, freshest, mddev->minor_version)) {
3295 "md: fatal superblock inconsistency in %s"
3296 " -- removing from array\n",
3297 bdevname(rdev->bdev,b));
3298 kick_rdev_from_array(rdev);
3302 super_types[mddev->major_version].
3303 validate_super(mddev, freshest);
3306 rdev_for_each_safe(rdev, tmp, mddev) {
3307 if (mddev->max_disks &&
3308 (rdev->desc_nr >= mddev->max_disks ||
3309 i > mddev->max_disks)) {
3311 "md: %s: %s: only %d devices permitted\n",
3312 mdname(mddev), bdevname(rdev->bdev, b),
3314 kick_rdev_from_array(rdev);
3317 if (rdev != freshest)
3318 if (super_types[mddev->major_version].
3319 validate_super(mddev, rdev)) {
3320 printk(KERN_WARNING "md: kicking non-fresh %s"
3322 bdevname(rdev->bdev,b));
3323 kick_rdev_from_array(rdev);
3326 if (mddev->level == LEVEL_MULTIPATH) {
3327 rdev->desc_nr = i++;
3328 rdev->raid_disk = rdev->desc_nr;
3329 set_bit(In_sync, &rdev->flags);
3330 } else if (rdev->raid_disk >= (mddev->raid_disks - min(0, mddev->delta_disks))) {
3331 rdev->raid_disk = -1;
3332 clear_bit(In_sync, &rdev->flags);
3337 /* Read a fixed-point number.
3338 * Numbers in sysfs attributes should be in "standard" units where
3339 * possible, so time should be in seconds.
3340 * However we internally use a a much smaller unit such as
3341 * milliseconds or jiffies.
3342 * This function takes a decimal number with a possible fractional
3343 * component, and produces an integer which is the result of
3344 * multiplying that number by 10^'scale'.
3345 * all without any floating-point arithmetic.
3347 int strict_strtoul_scaled(const char *cp, unsigned long *res, int scale)
3349 unsigned long result = 0;
3351 while (isdigit(*cp) || (*cp == '.' && decimals < 0)) {
3354 else if (decimals < scale) {
3357 result = result * 10 + value;
3369 while (decimals < scale) {
3378 static void md_safemode_timeout(unsigned long data);
3381 safe_delay_show(struct mddev *mddev, char *page)
3383 int msec = (mddev->safemode_delay*1000)/HZ;
3384 return sprintf(page, "%d.%03d\n", msec/1000, msec%1000);
3387 safe_delay_store(struct mddev *mddev, const char *cbuf, size_t len)
3391 if (strict_strtoul_scaled(cbuf, &msec, 3) < 0)
3394 mddev->safemode_delay = 0;
3396 unsigned long old_delay = mddev->safemode_delay;
3397 mddev->safemode_delay = (msec*HZ)/1000;
3398 if (mddev->safemode_delay == 0)
3399 mddev->safemode_delay = 1;
3400 if (mddev->safemode_delay < old_delay || old_delay == 0)
3401 md_safemode_timeout((unsigned long)mddev);
3405 static struct md_sysfs_entry md_safe_delay =
3406 __ATTR(safe_mode_delay, S_IRUGO|S_IWUSR,safe_delay_show, safe_delay_store);
3409 level_show(struct mddev *mddev, char *page)
3411 struct md_personality *p = mddev->pers;
3413 return sprintf(page, "%s\n", p->name);
3414 else if (mddev->clevel[0])
3415 return sprintf(page, "%s\n", mddev->clevel);
3416 else if (mddev->level != LEVEL_NONE)
3417 return sprintf(page, "%d\n", mddev->level);
3423 level_store(struct mddev *mddev, const char *buf, size_t len)
3427 struct md_personality *pers;
3430 struct md_rdev *rdev;
3432 if (mddev->pers == NULL) {
3435 if (len >= sizeof(mddev->clevel))
3437 strncpy(mddev->clevel, buf, len);
3438 if (mddev->clevel[len-1] == '\n')
3440 mddev->clevel[len] = 0;
3441 mddev->level = LEVEL_NONE;
3445 /* request to change the personality. Need to ensure:
3446 * - array is not engaged in resync/recovery/reshape
3447 * - old personality can be suspended
3448 * - new personality will access other array.
3451 if (mddev->sync_thread ||
3452 mddev->reshape_position != MaxSector ||
3453 mddev->sysfs_active)
3456 if (!mddev->pers->quiesce) {
3457 printk(KERN_WARNING "md: %s: %s does not support online personality change\n",
3458 mdname(mddev), mddev->pers->name);
3462 /* Now find the new personality */
3463 if (len == 0 || len >= sizeof(clevel))
3465 strncpy(clevel, buf, len);
3466 if (clevel[len-1] == '\n')
3469 if (kstrtol(clevel, 10, &level))
3472 if (request_module("md-%s", clevel) != 0)
3473 request_module("md-level-%s", clevel);
3474 spin_lock(&pers_lock);
3475 pers = find_pers(level, clevel);
3476 if (!pers || !try_module_get(pers->owner)) {
3477 spin_unlock(&pers_lock);
3478 printk(KERN_WARNING "md: personality %s not loaded\n", clevel);
3481 spin_unlock(&pers_lock);
3483 if (pers == mddev->pers) {
3484 /* Nothing to do! */
3485 module_put(pers->owner);
3488 if (!pers->takeover) {
3489 module_put(pers->owner);
3490 printk(KERN_WARNING "md: %s: %s does not support personality takeover\n",
3491 mdname(mddev), clevel);
3495 rdev_for_each(rdev, mddev)
3496 rdev->new_raid_disk = rdev->raid_disk;
3498 /* ->takeover must set new_* and/or delta_disks
3499 * if it succeeds, and may set them when it fails.
3501 priv = pers->takeover(mddev);
3503 mddev->new_level = mddev->level;
3504 mddev->new_layout = mddev->layout;
3505 mddev->new_chunk_sectors = mddev->chunk_sectors;
3506 mddev->raid_disks -= mddev->delta_disks;
3507 mddev->delta_disks = 0;
3508 mddev->reshape_backwards = 0;
3509 module_put(pers->owner);
3510 printk(KERN_WARNING "md: %s: %s would not accept array\n",
3511 mdname(mddev), clevel);
3512 return PTR_ERR(priv);
3515 /* Looks like we have a winner */
3516 mddev_suspend(mddev);
3517 mddev->pers->stop(mddev);
3519 if (mddev->pers->sync_request == NULL &&
3520 pers->sync_request != NULL) {
3521 /* need to add the md_redundancy_group */
3522 if (sysfs_create_group(&mddev->kobj, &md_redundancy_group))
3524 "md: cannot register extra attributes for %s\n",
3526 mddev->sysfs_action = sysfs_get_dirent(mddev->kobj.sd, "sync_action");
3528 if (mddev->pers->sync_request != NULL &&
3529 pers->sync_request == NULL) {
3530 /* need to remove the md_redundancy_group */
3531 if (mddev->to_remove == NULL)
3532 mddev->to_remove = &md_redundancy_group;
3535 if (mddev->pers->sync_request == NULL &&
3537 /* We are converting from a no-redundancy array
3538 * to a redundancy array and metadata is managed
3539 * externally so we need to be sure that writes
3540 * won't block due to a need to transition
3542 * until external management is started.
3545 mddev->safemode_delay = 0;
3546 mddev->safemode = 0;
3549 rdev_for_each(rdev, mddev) {
3550 if (rdev->raid_disk < 0)
3552 if (rdev->new_raid_disk >= mddev->raid_disks)
3553 rdev->new_raid_disk = -1;
3554 if (rdev->new_raid_disk == rdev->raid_disk)
3556 sysfs_unlink_rdev(mddev, rdev);
3558 rdev_for_each(rdev, mddev) {
3559 if (rdev->raid_disk < 0)
3561 if (rdev->new_raid_disk == rdev->raid_disk)
3563 rdev->raid_disk = rdev->new_raid_disk;
3564 if (rdev->raid_disk < 0)
3565 clear_bit(In_sync, &rdev->flags);
3567 if (sysfs_link_rdev(mddev, rdev))
3568 printk(KERN_WARNING "md: cannot register rd%d"
3569 " for %s after level change\n",
3570 rdev->raid_disk, mdname(mddev));
3574 module_put(mddev->pers->owner);
3576 mddev->private = priv;
3577 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
3578 mddev->level = mddev->new_level;
3579 mddev->layout = mddev->new_layout;
3580 mddev->chunk_sectors = mddev->new_chunk_sectors;
3581 mddev->delta_disks = 0;
3582 mddev->reshape_backwards = 0;
3583 mddev->degraded = 0;
3584 if (mddev->pers->sync_request == NULL) {
3585 /* this is now an array without redundancy, so
3586 * it must always be in_sync
3589 del_timer_sync(&mddev->safemode_timer);
3591 blk_set_stacking_limits(&mddev->queue->limits);
3593 set_bit(MD_CHANGE_DEVS, &mddev->flags);
3594 mddev_resume(mddev);
3595 sysfs_notify(&mddev->kobj, NULL, "level");
3596 md_new_event(mddev);
3600 static struct md_sysfs_entry md_level =
3601 __ATTR(level, S_IRUGO|S_IWUSR, level_show, level_store);
3605 layout_show(struct mddev *mddev, char *page)
3607 /* just a number, not meaningful for all levels */
3608 if (mddev->reshape_position != MaxSector &&
3609 mddev->layout != mddev->new_layout)
3610 return sprintf(page, "%d (%d)\n",
3611 mddev->new_layout, mddev->layout);
3612 return sprintf(page, "%d\n", mddev->layout);
3616 layout_store(struct mddev *mddev, const char *buf, size_t len)
3619 unsigned long n = simple_strtoul(buf, &e, 10);
3621 if (!*buf || (*e && *e != '\n'))
3626 if (mddev->pers->check_reshape == NULL)
3628 mddev->new_layout = n;
3629 err = mddev->pers->check_reshape(mddev);
3631 mddev->new_layout = mddev->layout;
3635 mddev->new_layout = n;
3636 if (mddev->reshape_position == MaxSector)
3641 static struct md_sysfs_entry md_layout =
3642 __ATTR(layout, S_IRUGO|S_IWUSR, layout_show, layout_store);
3646 raid_disks_show(struct mddev *mddev, char *page)
3648 if (mddev->raid_disks == 0)
3650 if (mddev->reshape_position != MaxSector &&
3651 mddev->delta_disks != 0)
3652 return sprintf(page, "%d (%d)\n", mddev->raid_disks,
3653 mddev->raid_disks - mddev->delta_disks);
3654 return sprintf(page, "%d\n", mddev->raid_disks);
3657 static int update_raid_disks(struct mddev *mddev, int raid_disks);
3660 raid_disks_store(struct mddev *mddev, const char *buf, size_t len)
3664 unsigned long n = simple_strtoul(buf, &e, 10);
3666 if (!*buf || (*e && *e != '\n'))
3670 rv = update_raid_disks(mddev, n);
3671 else if (mddev->reshape_position != MaxSector) {
3672 struct md_rdev *rdev;
3673 int olddisks = mddev->raid_disks - mddev->delta_disks;
3675 rdev_for_each(rdev, mddev) {
3677 rdev->data_offset < rdev->new_data_offset)
3680 rdev->data_offset > rdev->new_data_offset)
3683 mddev->delta_disks = n - olddisks;
3684 mddev->raid_disks = n;
3685 mddev->reshape_backwards = (mddev->delta_disks < 0);
3687 mddev->raid_disks = n;
3688 return rv ? rv : len;
3690 static struct md_sysfs_entry md_raid_disks =
3691 __ATTR(raid_disks, S_IRUGO|S_IWUSR, raid_disks_show, raid_disks_store);
3694 chunk_size_show(struct mddev *mddev, char *page)
3696 if (mddev->reshape_position != MaxSector &&
3697 mddev->chunk_sectors != mddev->new_chunk_sectors)
3698 return sprintf(page, "%d (%d)\n",
3699 mddev->new_chunk_sectors << 9,
3700 mddev->chunk_sectors << 9);
3701 return sprintf(page, "%d\n", mddev->chunk_sectors << 9);
3705 chunk_size_store(struct mddev *mddev, const char *buf, size_t len)
3708 unsigned long n = simple_strtoul(buf, &e, 10);
3710 if (!*buf || (*e && *e != '\n'))
3715 if (mddev->pers->check_reshape == NULL)
3717 mddev->new_chunk_sectors = n >> 9;
3718 err = mddev->pers->check_reshape(mddev);
3720 mddev->new_chunk_sectors = mddev->chunk_sectors;
3724 mddev->new_chunk_sectors = n >> 9;
3725 if (mddev->reshape_position == MaxSector)
3726 mddev->chunk_sectors = n >> 9;
3730 static struct md_sysfs_entry md_chunk_size =
3731 __ATTR(chunk_size, S_IRUGO|S_IWUSR, chunk_size_show, chunk_size_store);
3734 resync_start_show(struct mddev *mddev, char *page)
3736 if (mddev->recovery_cp == MaxSector)
3737 return sprintf(page, "none\n");
3738 return sprintf(page, "%llu\n", (unsigned long long)mddev->recovery_cp);
3742 resync_start_store(struct mddev *mddev, const char *buf, size_t len)
3745 unsigned long long n = simple_strtoull(buf, &e, 10);
3747 if (mddev->pers && !test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
3749 if (cmd_match(buf, "none"))
3751 else if (!*buf || (*e && *e != '\n'))
3754 mddev->recovery_cp = n;
3756 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
3759 static struct md_sysfs_entry md_resync_start =
3760 __ATTR(resync_start, S_IRUGO|S_IWUSR, resync_start_show, resync_start_store);
3763 * The array state can be:
3766 * No devices, no size, no level
3767 * Equivalent to STOP_ARRAY ioctl
3769 * May have some settings, but array is not active
3770 * all IO results in error
3771 * When written, doesn't tear down array, but just stops it
3772 * suspended (not supported yet)
3773 * All IO requests will block. The array can be reconfigured.
3774 * Writing this, if accepted, will block until array is quiescent
3776 * no resync can happen. no superblocks get written.
3777 * write requests fail
3779 * like readonly, but behaves like 'clean' on a write request.
3781 * clean - no pending writes, but otherwise active.
3782 * When written to inactive array, starts without resync
3783 * If a write request arrives then
3784 * if metadata is known, mark 'dirty' and switch to 'active'.
3785 * if not known, block and switch to write-pending
3786 * If written to an active array that has pending writes, then fails.
3788 * fully active: IO and resync can be happening.
3789 * When written to inactive array, starts with resync
3792 * clean, but writes are blocked waiting for 'active' to be written.
3795 * like active, but no writes have been seen for a while (100msec).
3798 enum array_state { clear, inactive, suspended, readonly, read_auto, clean, active,
3799 write_pending, active_idle, bad_word};
3800 static char *array_states[] = {
3801 "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active",
3802 "write-pending", "active-idle", NULL };
3804 static int match_word(const char *word, char **list)
3807 for (n=0; list[n]; n++)
3808 if (cmd_match(word, list[n]))
3814 array_state_show(struct mddev *mddev, char *page)
3816 enum array_state st = inactive;
3829 else if (test_bit(MD_CHANGE_PENDING, &mddev->flags))
3831 else if (mddev->safemode)
3837 if (list_empty(&mddev->disks) &&
3838 mddev->raid_disks == 0 &&
3839 mddev->dev_sectors == 0)
3844 return sprintf(page, "%s\n", array_states[st]);
3847 static int do_md_stop(struct mddev * mddev, int ro, struct block_device *bdev);
3848 static int md_set_readonly(struct mddev * mddev, struct block_device *bdev);
3849 static int do_md_run(struct mddev * mddev);
3850 static int restart_array(struct mddev *mddev);
3853 array_state_store(struct mddev *mddev, const char *buf, size_t len)
3856 enum array_state st = match_word(buf, array_states);
3861 /* stopping an active array */
3862 err = do_md_stop(mddev, 0, NULL);
3865 /* stopping an active array */
3867 err = do_md_stop(mddev, 2, NULL);
3869 err = 0; /* already inactive */
3872 break; /* not supported yet */
3875 err = md_set_readonly(mddev, NULL);
3878 set_disk_ro(mddev->gendisk, 1);
3879 err = do_md_run(mddev);
3885 err = md_set_readonly(mddev, NULL);
3886 else if (mddev->ro == 1)
3887 err = restart_array(mddev);
3890 set_disk_ro(mddev->gendisk, 0);
3894 err = do_md_run(mddev);
3899 restart_array(mddev);
3900 spin_lock_irq(&mddev->write_lock);
3901 if (atomic_read(&mddev->writes_pending) == 0) {
3902 if (mddev->in_sync == 0) {
3904 if (mddev->safemode == 1)
3905 mddev->safemode = 0;
3906 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
3911 spin_unlock_irq(&mddev->write_lock);
3917 restart_array(mddev);
3918 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
3919 wake_up(&mddev->sb_wait);
3923 set_disk_ro(mddev->gendisk, 0);
3924 err = do_md_run(mddev);
3929 /* these cannot be set */
3935 if (mddev->hold_active == UNTIL_IOCTL)
3936 mddev->hold_active = 0;
3937 sysfs_notify_dirent_safe(mddev->sysfs_state);
3941 static struct md_sysfs_entry md_array_state =
3942 __ATTR(array_state, S_IRUGO|S_IWUSR, array_state_show, array_state_store);
3945 max_corrected_read_errors_show(struct mddev *mddev, char *page) {
3946 return sprintf(page, "%d\n",
3947 atomic_read(&mddev->max_corr_read_errors));
3951 max_corrected_read_errors_store(struct mddev *mddev, const char *buf, size_t len)
3954 unsigned long n = simple_strtoul(buf, &e, 10);
3956 if (*buf && (*e == 0 || *e == '\n')) {
3957 atomic_set(&mddev->max_corr_read_errors, n);
3963 static struct md_sysfs_entry max_corr_read_errors =
3964 __ATTR(max_read_errors, S_IRUGO|S_IWUSR, max_corrected_read_errors_show,
3965 max_corrected_read_errors_store);
3968 null_show(struct mddev *mddev, char *page)
3974 new_dev_store(struct mddev *mddev, const char *buf, size_t len)
3976 /* buf must be %d:%d\n? giving major and minor numbers */
3977 /* The new device is added to the array.
3978 * If the array has a persistent superblock, we read the
3979 * superblock to initialise info and check validity.
3980 * Otherwise, only checking done is that in bind_rdev_to_array,
3981 * which mainly checks size.
3984 int major = simple_strtoul(buf, &e, 10);
3987 struct md_rdev *rdev;
3990 if (!*buf || *e != ':' || !e[1] || e[1] == '\n')
3992 minor = simple_strtoul(e+1, &e, 10);
3993 if (*e && *e != '\n')
3995 dev = MKDEV(major, minor);
3996 if (major != MAJOR(dev) ||
3997 minor != MINOR(dev))
4001 if (mddev->persistent) {
4002 rdev = md_import_device(dev, mddev->major_version,
4003 mddev->minor_version);
4004 if (!IS_ERR(rdev) && !list_empty(&mddev->disks)) {
4005 struct md_rdev *rdev0
4006 = list_entry(mddev->disks.next,
4007 struct md_rdev, same_set);
4008 err = super_types[mddev->major_version]
4009 .load_super(rdev, rdev0, mddev->minor_version);
4013 } else if (mddev->external)
4014 rdev = md_import_device(dev, -2, -1);
4016 rdev = md_import_device(dev, -1, -1);
4019 return PTR_ERR(rdev);
4020 err = bind_rdev_to_array(rdev, mddev);
4024 return err ? err : len;
4027 static struct md_sysfs_entry md_new_device =
4028 __ATTR(new_dev, S_IWUSR, null_show, new_dev_store);
4031 bitmap_store(struct mddev *mddev, const char *buf, size_t len)
4034 unsigned long chunk, end_chunk;
4038 /* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */
4040 chunk = end_chunk = simple_strtoul(buf, &end, 0);
4041 if (buf == end) break;
4042 if (*end == '-') { /* range */
4044 end_chunk = simple_strtoul(buf, &end, 0);
4045 if (buf == end) break;
4047 if (*end && !isspace(*end)) break;
4048 bitmap_dirty_bits(mddev->bitmap, chunk, end_chunk);
4049 buf = skip_spaces(end);
4051 bitmap_unplug(mddev->bitmap); /* flush the bits to disk */
4056 static struct md_sysfs_entry md_bitmap =
4057 __ATTR(bitmap_set_bits, S_IWUSR, null_show, bitmap_store);
4060 size_show(struct mddev *mddev, char *page)
4062 return sprintf(page, "%llu\n",
4063 (unsigned long long)mddev->dev_sectors / 2);
4066 static int update_size(struct mddev *mddev, sector_t num_sectors);
4069 size_store(struct mddev *mddev, const char *buf, size_t len)
4071 /* If array is inactive, we can reduce the component size, but
4072 * not increase it (except from 0).
4073 * If array is active, we can try an on-line resize
4076 int err = strict_blocks_to_sectors(buf, §ors);
4081 err = update_size(mddev, sectors);
4082 md_update_sb(mddev, 1);
4084 if (mddev->dev_sectors == 0 ||
4085 mddev->dev_sectors > sectors)
4086 mddev->dev_sectors = sectors;
4090 return err ? err : len;
4093 static struct md_sysfs_entry md_size =
4094 __ATTR(component_size, S_IRUGO|S_IWUSR, size_show, size_store);
4097 /* Metadata version.
4099 * 'none' for arrays with no metadata (good luck...)
4100 * 'external' for arrays with externally managed metadata,
4101 * or N.M for internally known formats
4104 metadata_show(struct mddev *mddev, char *page)
4106 if (mddev->persistent)
4107 return sprintf(page, "%d.%d\n",
4108 mddev->major_version, mddev->minor_version);
4109 else if (mddev->external)
4110 return sprintf(page, "external:%s\n", mddev->metadata_type);
4112 return sprintf(page, "none\n");
4116 metadata_store(struct mddev *mddev, const char *buf, size_t len)
4120 /* Changing the details of 'external' metadata is
4121 * always permitted. Otherwise there must be
4122 * no devices attached to the array.
4124 if (mddev->external && strncmp(buf, "external:", 9) == 0)
4126 else if (!list_empty(&mddev->disks))
4129 if (cmd_match(buf, "none")) {
4130 mddev->persistent = 0;
4131 mddev->external = 0;
4132 mddev->major_version = 0;
4133 mddev->minor_version = 90;
4136 if (strncmp(buf, "external:", 9) == 0) {
4137 size_t namelen = len-9;
4138 if (namelen >= sizeof(mddev->metadata_type))
4139 namelen = sizeof(mddev->metadata_type)-1;
4140 strncpy(mddev->metadata_type, buf+9, namelen);
4141 mddev->metadata_type[namelen] = 0;
4142 if (namelen && mddev->metadata_type[namelen-1] == '\n')
4143 mddev->metadata_type[--namelen] = 0;
4144 mddev->persistent = 0;
4145 mddev->external = 1;
4146 mddev->major_version = 0;
4147 mddev->minor_version = 90;
4150 major = simple_strtoul(buf, &e, 10);
4151 if (e==buf || *e != '.')
4154 minor = simple_strtoul(buf, &e, 10);
4155 if (e==buf || (*e && *e != '\n') )
4157 if (major >= ARRAY_SIZE(super_types) || super_types[major].name == NULL)
4159 mddev->major_version = major;
4160 mddev->minor_version = minor;
4161 mddev->persistent = 1;
4162 mddev->external = 0;
4166 static struct md_sysfs_entry md_metadata =
4167 __ATTR(metadata_version, S_IRUGO|S_IWUSR, metadata_show, metadata_store);
4170 action_show(struct mddev *mddev, char *page)
4172 char *type = "idle";
4173 if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
4175 else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
4176 (!mddev->ro && test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))) {
4177 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
4179 else if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
4180 if (!test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
4182 else if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
4186 } else if (test_bit(MD_RECOVERY_RECOVER, &mddev->recovery))
4189 return sprintf(page, "%s\n", type);
4193 action_store(struct mddev *mddev, const char *page, size_t len)
4195 if (!mddev->pers || !mddev->pers->sync_request)
4198 if (cmd_match(page, "frozen"))
4199 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4201 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4203 if (cmd_match(page, "idle") || cmd_match(page, "frozen")) {
4204 if (mddev->sync_thread) {
4205 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
4206 md_reap_sync_thread(mddev);
4208 } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
4209 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
4211 else if (cmd_match(page, "resync"))
4212 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4213 else if (cmd_match(page, "recover")) {
4214 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
4215 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4216 } else if (cmd_match(page, "reshape")) {
4218 if (mddev->pers->start_reshape == NULL)
4220 err = mddev->pers->start_reshape(mddev);
4223 sysfs_notify(&mddev->kobj, NULL, "degraded");
4225 if (cmd_match(page, "check"))
4226 set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
4227 else if (!cmd_match(page, "repair"))
4229 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
4230 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
4232 if (mddev->ro == 2) {
4233 /* A write to sync_action is enough to justify
4234 * canceling read-auto mode
4237 md_wakeup_thread(mddev->sync_thread);
4239 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4240 md_wakeup_thread(mddev->thread);
4241 sysfs_notify_dirent_safe(mddev->sysfs_action);
4245 static struct md_sysfs_entry md_scan_mode =
4246 __ATTR(sync_action, S_IRUGO|S_IWUSR, action_show, action_store);
4249 last_sync_action_show(struct mddev *mddev, char *page)
4251 return sprintf(page, "%s\n", mddev->last_sync_action);
4254 static struct md_sysfs_entry md_last_scan_mode = __ATTR_RO(last_sync_action);
4257 mismatch_cnt_show(struct mddev *mddev, char *page)
4259 return sprintf(page, "%llu\n",
4260 (unsigned long long)
4261 atomic64_read(&mddev->resync_mismatches));
4264 static struct md_sysfs_entry md_mismatches = __ATTR_RO(mismatch_cnt);
4267 sync_min_show(struct mddev *mddev, char *page)
4269 return sprintf(page, "%d (%s)\n", speed_min(mddev),
4270 mddev->sync_speed_min ? "local": "system");
4274 sync_min_store(struct mddev *mddev, const char *buf, size_t len)
4278 if (strncmp(buf, "system", 6)==0) {
4279 mddev->sync_speed_min = 0;
4282 min = simple_strtoul(buf, &e, 10);
4283 if (buf == e || (*e && *e != '\n') || min <= 0)
4285 mddev->sync_speed_min = min;
4289 static struct md_sysfs_entry md_sync_min =
4290 __ATTR(sync_speed_min, S_IRUGO|S_IWUSR, sync_min_show, sync_min_store);
4293 sync_max_show(struct mddev *mddev, char *page)
4295 return sprintf(page, "%d (%s)\n", speed_max(mddev),
4296 mddev->sync_speed_max ? "local": "system");
4300 sync_max_store(struct mddev *mddev, const char *buf, size_t len)
4304 if (strncmp(buf, "system", 6)==0) {
4305 mddev->sync_speed_max = 0;
4308 max = simple_strtoul(buf, &e, 10);
4309 if (buf == e || (*e && *e != '\n') || max <= 0)
4311 mddev->sync_speed_max = max;
4315 static struct md_sysfs_entry md_sync_max =
4316 __ATTR(sync_speed_max, S_IRUGO|S_IWUSR, sync_max_show, sync_max_store);
4319 degraded_show(struct mddev *mddev, char *page)
4321 return sprintf(page, "%d\n", mddev->degraded);
4323 static struct md_sysfs_entry md_degraded = __ATTR_RO(degraded);
4326 sync_force_parallel_show(struct mddev *mddev, char *page)
4328 return sprintf(page, "%d\n", mddev->parallel_resync);
4332 sync_force_parallel_store(struct mddev *mddev, const char *buf, size_t len)
4336 if (kstrtol(buf, 10, &n))
4339 if (n != 0 && n != 1)
4342 mddev->parallel_resync = n;
4344 if (mddev->sync_thread)
4345 wake_up(&resync_wait);
4350 /* force parallel resync, even with shared block devices */
4351 static struct md_sysfs_entry md_sync_force_parallel =
4352 __ATTR(sync_force_parallel, S_IRUGO|S_IWUSR,
4353 sync_force_parallel_show, sync_force_parallel_store);
4356 sync_speed_show(struct mddev *mddev, char *page)
4358 unsigned long resync, dt, db;
4359 if (mddev->curr_resync == 0)
4360 return sprintf(page, "none\n");
4361 resync = mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active);
4362 dt = (jiffies - mddev->resync_mark) / HZ;
4364 db = resync - mddev->resync_mark_cnt;
4365 return sprintf(page, "%lu\n", db/dt/2); /* K/sec */
4368 static struct md_sysfs_entry md_sync_speed = __ATTR_RO(sync_speed);
4371 sync_completed_show(struct mddev *mddev, char *page)
4373 unsigned long long max_sectors, resync;
4375 if (!test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4376 return sprintf(page, "none\n");
4378 if (mddev->curr_resync == 1 ||
4379 mddev->curr_resync == 2)
4380 return sprintf(page, "delayed\n");
4382 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ||
4383 test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
4384 max_sectors = mddev->resync_max_sectors;
4386 max_sectors = mddev->dev_sectors;
4388 resync = mddev->curr_resync_completed;
4389 return sprintf(page, "%llu / %llu\n", resync, max_sectors);
4392 static struct md_sysfs_entry md_sync_completed = __ATTR_RO(sync_completed);
4395 min_sync_show(struct mddev *mddev, char *page)
4397 return sprintf(page, "%llu\n",
4398 (unsigned long long)mddev->resync_min);
4401 min_sync_store(struct mddev *mddev, const char *buf, size_t len)
4403 unsigned long long min;
4404 if (kstrtoull(buf, 10, &min))
4406 if (min > mddev->resync_max)
4408 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4411 /* Must be a multiple of chunk_size */
4412 if (mddev->chunk_sectors) {
4413 sector_t temp = min;
4414 if (sector_div(temp, mddev->chunk_sectors))
4417 mddev->resync_min = min;
4422 static struct md_sysfs_entry md_min_sync =
4423 __ATTR(sync_min, S_IRUGO|S_IWUSR, min_sync_show, min_sync_store);
4426 max_sync_show(struct mddev *mddev, char *page)
4428 if (mddev->resync_max == MaxSector)
4429 return sprintf(page, "max\n");
4431 return sprintf(page, "%llu\n",
4432 (unsigned long long)mddev->resync_max);
4435 max_sync_store(struct mddev *mddev, const char *buf, size_t len)
4437 if (strncmp(buf, "max", 3) == 0)
4438 mddev->resync_max = MaxSector;
4440 unsigned long long max;
4441 if (kstrtoull(buf, 10, &max))
4443 if (max < mddev->resync_min)
4445 if (max < mddev->resync_max &&
4447 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4450 /* Must be a multiple of chunk_size */
4451 if (mddev->chunk_sectors) {
4452 sector_t temp = max;
4453 if (sector_div(temp, mddev->chunk_sectors))
4456 mddev->resync_max = max;
4458 wake_up(&mddev->recovery_wait);
4462 static struct md_sysfs_entry md_max_sync =
4463 __ATTR(sync_max, S_IRUGO|S_IWUSR, max_sync_show, max_sync_store);
4466 suspend_lo_show(struct mddev *mddev, char *page)
4468 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_lo);
4472 suspend_lo_store(struct mddev *mddev, const char *buf, size_t len)
4475 unsigned long long new = simple_strtoull(buf, &e, 10);
4476 unsigned long long old = mddev->suspend_lo;
4478 if (mddev->pers == NULL ||
4479 mddev->pers->quiesce == NULL)
4481 if (buf == e || (*e && *e != '\n'))
4484 mddev->suspend_lo = new;
4486 /* Shrinking suspended region */
4487 mddev->pers->quiesce(mddev, 2);
4489 /* Expanding suspended region - need to wait */
4490 mddev->pers->quiesce(mddev, 1);
4491 mddev->pers->quiesce(mddev, 0);
4495 static struct md_sysfs_entry md_suspend_lo =
4496 __ATTR(suspend_lo, S_IRUGO|S_IWUSR, suspend_lo_show, suspend_lo_store);
4500 suspend_hi_show(struct mddev *mddev, char *page)
4502 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_hi);
4506 suspend_hi_store(struct mddev *mddev, const char *buf, size_t len)
4509 unsigned long long new = simple_strtoull(buf, &e, 10);
4510 unsigned long long old = mddev->suspend_hi;
4512 if (mddev->pers == NULL ||
4513 mddev->pers->quiesce == NULL)
4515 if (buf == e || (*e && *e != '\n'))
4518 mddev->suspend_hi = new;
4520 /* Shrinking suspended region */
4521 mddev->pers->quiesce(mddev, 2);
4523 /* Expanding suspended region - need to wait */
4524 mddev->pers->quiesce(mddev, 1);
4525 mddev->pers->quiesce(mddev, 0);
4529 static struct md_sysfs_entry md_suspend_hi =
4530 __ATTR(suspend_hi, S_IRUGO|S_IWUSR, suspend_hi_show, suspend_hi_store);
4533 reshape_position_show(struct mddev *mddev, char *page)
4535 if (mddev->reshape_position != MaxSector)
4536 return sprintf(page, "%llu\n",
4537 (unsigned long long)mddev->reshape_position);
4538 strcpy(page, "none\n");
4543 reshape_position_store(struct mddev *mddev, const char *buf, size_t len)
4545 struct md_rdev *rdev;
4547 unsigned long long new = simple_strtoull(buf, &e, 10);
4550 if (buf == e || (*e && *e != '\n'))
4552 mddev->reshape_position = new;
4553 mddev->delta_disks = 0;
4554 mddev->reshape_backwards = 0;
4555 mddev->new_level = mddev->level;
4556 mddev->new_layout = mddev->layout;
4557 mddev->new_chunk_sectors = mddev->chunk_sectors;
4558 rdev_for_each(rdev, mddev)
4559 rdev->new_data_offset = rdev->data_offset;
4563 static struct md_sysfs_entry md_reshape_position =
4564 __ATTR(reshape_position, S_IRUGO|S_IWUSR, reshape_position_show,
4565 reshape_position_store);
4568 reshape_direction_show(struct mddev *mddev, char *page)
4570 return sprintf(page, "%s\n",
4571 mddev->reshape_backwards ? "backwards" : "forwards");
4575 reshape_direction_store(struct mddev *mddev, const char *buf, size_t len)
4578 if (cmd_match(buf, "forwards"))
4580 else if (cmd_match(buf, "backwards"))
4584 if (mddev->reshape_backwards == backwards)
4587 /* check if we are allowed to change */
4588 if (mddev->delta_disks)
4591 if (mddev->persistent &&
4592 mddev->major_version == 0)
4595 mddev->reshape_backwards = backwards;
4599 static struct md_sysfs_entry md_reshape_direction =
4600 __ATTR(reshape_direction, S_IRUGO|S_IWUSR, reshape_direction_show,
4601 reshape_direction_store);
4604 array_size_show(struct mddev *mddev, char *page)
4606 if (mddev->external_size)
4607 return sprintf(page, "%llu\n",
4608 (unsigned long long)mddev->array_sectors/2);
4610 return sprintf(page, "default\n");
4614 array_size_store(struct mddev *mddev, const char *buf, size_t len)
4618 if (strncmp(buf, "default", 7) == 0) {
4620 sectors = mddev->pers->size(mddev, 0, 0);
4622 sectors = mddev->array_sectors;
4624 mddev->external_size = 0;
4626 if (strict_blocks_to_sectors(buf, §ors) < 0)
4628 if (mddev->pers && mddev->pers->size(mddev, 0, 0) < sectors)
4631 mddev->external_size = 1;
4634 mddev->array_sectors = sectors;
4636 set_capacity(mddev->gendisk, mddev->array_sectors);
4637 revalidate_disk(mddev->gendisk);
4642 static struct md_sysfs_entry md_array_size =
4643 __ATTR(array_size, S_IRUGO|S_IWUSR, array_size_show,
4646 static struct attribute *md_default_attrs[] = {
4649 &md_raid_disks.attr,
4650 &md_chunk_size.attr,
4652 &md_resync_start.attr,
4654 &md_new_device.attr,
4655 &md_safe_delay.attr,
4656 &md_array_state.attr,
4657 &md_reshape_position.attr,
4658 &md_reshape_direction.attr,
4659 &md_array_size.attr,
4660 &max_corr_read_errors.attr,
4664 static struct attribute *md_redundancy_attrs[] = {
4666 &md_last_scan_mode.attr,
4667 &md_mismatches.attr,
4670 &md_sync_speed.attr,
4671 &md_sync_force_parallel.attr,
4672 &md_sync_completed.attr,
4675 &md_suspend_lo.attr,
4676 &md_suspend_hi.attr,
4681 static struct attribute_group md_redundancy_group = {
4683 .attrs = md_redundancy_attrs,
4688 md_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
4690 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
4691 struct mddev *mddev = container_of(kobj, struct mddev, kobj);
4696 spin_lock(&all_mddevs_lock);
4697 if (list_empty(&mddev->all_mddevs)) {
4698 spin_unlock(&all_mddevs_lock);
4702 spin_unlock(&all_mddevs_lock);
4704 rv = mddev_lock(mddev);
4706 rv = entry->show(mddev, page);
4707 mddev_unlock(mddev);
4714 md_attr_store(struct kobject *kobj, struct attribute *attr,
4715 const char *page, size_t length)
4717 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
4718 struct mddev *mddev = container_of(kobj, struct mddev, kobj);
4723 if (!capable(CAP_SYS_ADMIN))
4725 spin_lock(&all_mddevs_lock);
4726 if (list_empty(&mddev->all_mddevs)) {
4727 spin_unlock(&all_mddevs_lock);
4731 spin_unlock(&all_mddevs_lock);
4732 if (entry->store == new_dev_store)
4733 flush_workqueue(md_misc_wq);
4734 rv = mddev_lock(mddev);
4736 rv = entry->store(mddev, page, length);
4737 mddev_unlock(mddev);
4743 static void md_free(struct kobject *ko)
4745 struct mddev *mddev = container_of(ko, struct mddev, kobj);
4747 if (mddev->sysfs_state)
4748 sysfs_put(mddev->sysfs_state);
4750 if (mddev->gendisk) {
4751 del_gendisk(mddev->gendisk);
4752 put_disk(mddev->gendisk);
4755 blk_cleanup_queue(mddev->queue);
4760 static const struct sysfs_ops md_sysfs_ops = {
4761 .show = md_attr_show,
4762 .store = md_attr_store,
4764 static struct kobj_type md_ktype = {
4766 .sysfs_ops = &md_sysfs_ops,
4767 .default_attrs = md_default_attrs,
4772 static void mddev_delayed_delete(struct work_struct *ws)
4774 struct mddev *mddev = container_of(ws, struct mddev, del_work);
4776 sysfs_remove_group(&mddev->kobj, &md_bitmap_group);
4777 kobject_del(&mddev->kobj);
4778 kobject_put(&mddev->kobj);
4781 static int md_alloc(dev_t dev, char *name)
4783 static DEFINE_MUTEX(disks_mutex);
4784 struct mddev *mddev = mddev_find(dev);
4785 struct gendisk *disk;
4794 partitioned = (MAJOR(mddev->unit) != MD_MAJOR);
4795 shift = partitioned ? MdpMinorShift : 0;
4796 unit = MINOR(mddev->unit) >> shift;
4798 /* wait for any previous instance of this device to be
4799 * completely removed (mddev_delayed_delete).
4801 flush_workqueue(md_misc_wq);
4803 mutex_lock(&disks_mutex);
4809 /* Need to ensure that 'name' is not a duplicate.
4811 struct mddev *mddev2;
4812 spin_lock(&all_mddevs_lock);
4814 list_for_each_entry(mddev2, &all_mddevs, all_mddevs)
4815 if (mddev2->gendisk &&
4816 strcmp(mddev2->gendisk->disk_name, name) == 0) {
4817 spin_unlock(&all_mddevs_lock);
4820 spin_unlock(&all_mddevs_lock);
4824 mddev->queue = blk_alloc_queue(GFP_KERNEL);
4827 mddev->queue->queuedata = mddev;
4829 blk_queue_make_request(mddev->queue, md_make_request);
4830 blk_set_stacking_limits(&mddev->queue->limits);
4832 disk = alloc_disk(1 << shift);
4834 blk_cleanup_queue(mddev->queue);
4835 mddev->queue = NULL;
4838 disk->major = MAJOR(mddev->unit);
4839 disk->first_minor = unit << shift;
4841 strcpy(disk->disk_name, name);
4842 else if (partitioned)
4843 sprintf(disk->disk_name, "md_d%d", unit);
4845 sprintf(disk->disk_name, "md%d", unit);
4846 disk->fops = &md_fops;
4847 disk->private_data = mddev;
4848 disk->queue = mddev->queue;
4849 blk_queue_flush(mddev->queue, REQ_FLUSH | REQ_FUA);
4850 /* Allow extended partitions. This makes the
4851 * 'mdp' device redundant, but we can't really
4854 disk->flags |= GENHD_FL_EXT_DEVT;
4855 mddev->gendisk = disk;
4856 /* As soon as we call add_disk(), another thread could get
4857 * through to md_open, so make sure it doesn't get too far
4859 mutex_lock(&mddev->open_mutex);
4862 error = kobject_init_and_add(&mddev->kobj, &md_ktype,
4863 &disk_to_dev(disk)->kobj, "%s", "md");
4865 /* This isn't possible, but as kobject_init_and_add is marked
4866 * __must_check, we must do something with the result
4868 printk(KERN_WARNING "md: cannot register %s/md - name in use\n",
4872 if (mddev->kobj.sd &&
4873 sysfs_create_group(&mddev->kobj, &md_bitmap_group))
4874 printk(KERN_DEBUG "pointless warning\n");
4875 mutex_unlock(&mddev->open_mutex);
4877 mutex_unlock(&disks_mutex);
4878 if (!error && mddev->kobj.sd) {
4879 kobject_uevent(&mddev->kobj, KOBJ_ADD);
4880 mddev->sysfs_state = sysfs_get_dirent_safe(mddev->kobj.sd, "array_state");
4886 static struct kobject *md_probe(dev_t dev, int *part, void *data)
4888 md_alloc(dev, NULL);
4892 static int add_named_array(const char *val, struct kernel_param *kp)
4894 /* val must be "md_*" where * is not all digits.
4895 * We allocate an array with a large free minor number, and
4896 * set the name to val. val must not already be an active name.
4898 int len = strlen(val);
4899 char buf[DISK_NAME_LEN];
4901 while (len && val[len-1] == '\n')
4903 if (len >= DISK_NAME_LEN)
4905 strlcpy(buf, val, len+1);
4906 if (strncmp(buf, "md_", 3) != 0)
4908 return md_alloc(0, buf);
4911 static void md_safemode_timeout(unsigned long data)
4913 struct mddev *mddev = (struct mddev *) data;
4915 if (!atomic_read(&mddev->writes_pending)) {
4916 mddev->safemode = 1;
4917 if (mddev->external)
4918 sysfs_notify_dirent_safe(mddev->sysfs_state);
4920 md_wakeup_thread(mddev->thread);
4923 static int start_dirty_degraded;
4925 int md_run(struct mddev *mddev)
4928 struct md_rdev *rdev;
4929 struct md_personality *pers;
4931 if (list_empty(&mddev->disks))
4932 /* cannot run an array with no devices.. */
4937 /* Cannot run until previous stop completes properly */
4938 if (mddev->sysfs_active)
4942 * Analyze all RAID superblock(s)
4944 if (!mddev->raid_disks) {
4945 if (!mddev->persistent)
4950 if (mddev->level != LEVEL_NONE)
4951 request_module("md-level-%d", mddev->level);
4952 else if (mddev->clevel[0])
4953 request_module("md-%s", mddev->clevel);
4956 * Drop all container device buffers, from now on
4957 * the only valid external interface is through the md
4960 rdev_for_each(rdev, mddev) {
4961 if (test_bit(Faulty, &rdev->flags))
4963 sync_blockdev(rdev->bdev);
4964 invalidate_bdev(rdev->bdev);
4966 /* perform some consistency tests on the device.
4967 * We don't want the data to overlap the metadata,
4968 * Internal Bitmap issues have been handled elsewhere.
4970 if (rdev->meta_bdev) {
4971 /* Nothing to check */;
4972 } else if (rdev->data_offset < rdev->sb_start) {
4973 if (mddev->dev_sectors &&
4974 rdev->data_offset + mddev->dev_sectors
4976 printk("md: %s: data overlaps metadata\n",
4981 if (rdev->sb_start + rdev->sb_size/512
4982 > rdev->data_offset) {
4983 printk("md: %s: metadata overlaps data\n",
4988 sysfs_notify_dirent_safe(rdev->sysfs_state);
4991 if (mddev->bio_set == NULL)
4992 mddev->bio_set = bioset_create(BIO_POOL_SIZE, 0);
4994 spin_lock(&pers_lock);
4995 pers = find_pers(mddev->level, mddev->clevel);
4996 if (!pers || !try_module_get(pers->owner)) {
4997 spin_unlock(&pers_lock);
4998 if (mddev->level != LEVEL_NONE)
4999 printk(KERN_WARNING "md: personality for level %d is not loaded!\n",
5002 printk(KERN_WARNING "md: personality for level %s is not loaded!\n",
5007 spin_unlock(&pers_lock);
5008 if (mddev->level != pers->level) {
5009 mddev->level = pers->level;
5010 mddev->new_level = pers->level;
5012 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
5014 if (mddev->reshape_position != MaxSector &&
5015 pers->start_reshape == NULL) {
5016 /* This personality cannot handle reshaping... */
5018 module_put(pers->owner);
5022 if (pers->sync_request) {
5023 /* Warn if this is a potentially silly
5026 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
5027 struct md_rdev *rdev2;
5030 rdev_for_each(rdev, mddev)
5031 rdev_for_each(rdev2, mddev) {
5033 rdev->bdev->bd_contains ==
5034 rdev2->bdev->bd_contains) {
5036 "%s: WARNING: %s appears to be"
5037 " on the same physical disk as"
5040 bdevname(rdev->bdev,b),
5041 bdevname(rdev2->bdev,b2));
5048 "True protection against single-disk"
5049 " failure might be compromised.\n");
5052 mddev->recovery = 0;
5053 /* may be over-ridden by personality */
5054 mddev->resync_max_sectors = mddev->dev_sectors;
5056 mddev->ok_start_degraded = start_dirty_degraded;
5058 if (start_readonly && mddev->ro == 0)
5059 mddev->ro = 2; /* read-only, but switch on first write */
5061 err = mddev->pers->run(mddev);
5063 printk(KERN_ERR "md: pers->run() failed ...\n");
5064 else if (mddev->pers->size(mddev, 0, 0) < mddev->array_sectors) {
5065 WARN_ONCE(!mddev->external_size, "%s: default size too small,"
5066 " but 'external_size' not in effect?\n", __func__);
5068 "md: invalid array_size %llu > default size %llu\n",
5069 (unsigned long long)mddev->array_sectors / 2,
5070 (unsigned long long)mddev->pers->size(mddev, 0, 0) / 2);
5072 mddev->pers->stop(mddev);
5074 if (err == 0 && mddev->pers->sync_request &&
5075 (mddev->bitmap_info.file || mddev->bitmap_info.offset)) {
5076 err = bitmap_create(mddev);
5078 printk(KERN_ERR "%s: failed to create bitmap (%d)\n",
5079 mdname(mddev), err);
5080 mddev->pers->stop(mddev);
5084 module_put(mddev->pers->owner);
5086 bitmap_destroy(mddev);
5089 if (mddev->pers->sync_request) {
5090 if (mddev->kobj.sd &&
5091 sysfs_create_group(&mddev->kobj, &md_redundancy_group))
5093 "md: cannot register extra attributes for %s\n",
5095 mddev->sysfs_action = sysfs_get_dirent_safe(mddev->kobj.sd, "sync_action");
5096 } else if (mddev->ro == 2) /* auto-readonly not meaningful */
5099 atomic_set(&mddev->writes_pending,0);
5100 atomic_set(&mddev->max_corr_read_errors,
5101 MD_DEFAULT_MAX_CORRECTED_READ_ERRORS);
5102 mddev->safemode = 0;
5103 mddev->safemode_timer.function = md_safemode_timeout;
5104 mddev->safemode_timer.data = (unsigned long) mddev;
5105 mddev->safemode_delay = (200 * HZ)/1000 +1; /* 200 msec delay */
5109 rdev_for_each(rdev, mddev)
5110 if (rdev->raid_disk >= 0)
5111 if (sysfs_link_rdev(mddev, rdev))
5112 /* failure here is OK */;
5114 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5116 if (mddev->flags & MD_UPDATE_SB_FLAGS)
5117 md_update_sb(mddev, 0);
5119 md_new_event(mddev);
5120 sysfs_notify_dirent_safe(mddev->sysfs_state);
5121 sysfs_notify_dirent_safe(mddev->sysfs_action);
5122 sysfs_notify(&mddev->kobj, NULL, "degraded");
5125 EXPORT_SYMBOL_GPL(md_run);
5127 static int do_md_run(struct mddev *mddev)
5131 err = md_run(mddev);
5134 err = bitmap_load(mddev);
5136 bitmap_destroy(mddev);
5140 md_wakeup_thread(mddev->thread);
5141 md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
5143 set_capacity(mddev->gendisk, mddev->array_sectors);
5144 revalidate_disk(mddev->gendisk);
5146 kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
5151 static int restart_array(struct mddev *mddev)
5153 struct gendisk *disk = mddev->gendisk;
5155 /* Complain if it has no devices */
5156 if (list_empty(&mddev->disks))
5162 mddev->safemode = 0;
5164 set_disk_ro(disk, 0);
5165 printk(KERN_INFO "md: %s switched to read-write mode.\n",
5167 /* Kick recovery or resync if necessary */
5168 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5169 md_wakeup_thread(mddev->thread);
5170 md_wakeup_thread(mddev->sync_thread);
5171 sysfs_notify_dirent_safe(mddev->sysfs_state);
5175 /* similar to deny_write_access, but accounts for our holding a reference
5176 * to the file ourselves */
5177 static int deny_bitmap_write_access(struct file * file)
5179 struct inode *inode = file->f_mapping->host;
5181 spin_lock(&inode->i_lock);
5182 if (atomic_read(&inode->i_writecount) > 1) {
5183 spin_unlock(&inode->i_lock);
5186 atomic_set(&inode->i_writecount, -1);
5187 spin_unlock(&inode->i_lock);
5192 void restore_bitmap_write_access(struct file *file)
5194 struct inode *inode = file->f_mapping->host;
5196 spin_lock(&inode->i_lock);
5197 atomic_set(&inode->i_writecount, 1);
5198 spin_unlock(&inode->i_lock);
5201 static void md_clean(struct mddev *mddev)
5203 mddev->array_sectors = 0;
5204 mddev->external_size = 0;
5205 mddev->dev_sectors = 0;
5206 mddev->raid_disks = 0;
5207 mddev->recovery_cp = 0;
5208 mddev->resync_min = 0;
5209 mddev->resync_max = MaxSector;
5210 mddev->reshape_position = MaxSector;
5211 mddev->external = 0;
5212 mddev->persistent = 0;
5213 mddev->level = LEVEL_NONE;
5214 mddev->clevel[0] = 0;
5217 mddev->metadata_type[0] = 0;
5218 mddev->chunk_sectors = 0;
5219 mddev->ctime = mddev->utime = 0;
5221 mddev->max_disks = 0;
5223 mddev->can_decrease_events = 0;
5224 mddev->delta_disks = 0;
5225 mddev->reshape_backwards = 0;
5226 mddev->new_level = LEVEL_NONE;
5227 mddev->new_layout = 0;
5228 mddev->new_chunk_sectors = 0;
5229 mddev->curr_resync = 0;
5230 atomic64_set(&mddev->resync_mismatches, 0);
5231 mddev->suspend_lo = mddev->suspend_hi = 0;
5232 mddev->sync_speed_min = mddev->sync_speed_max = 0;
5233 mddev->recovery = 0;
5236 mddev->degraded = 0;
5237 mddev->safemode = 0;
5238 mddev->merge_check_needed = 0;
5239 mddev->bitmap_info.offset = 0;
5240 mddev->bitmap_info.default_offset = 0;
5241 mddev->bitmap_info.default_space = 0;
5242 mddev->bitmap_info.chunksize = 0;
5243 mddev->bitmap_info.daemon_sleep = 0;
5244 mddev->bitmap_info.max_write_behind = 0;
5247 static void __md_stop_writes(struct mddev *mddev)
5249 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5250 if (mddev->sync_thread) {
5251 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5252 md_reap_sync_thread(mddev);
5255 del_timer_sync(&mddev->safemode_timer);
5257 bitmap_flush(mddev);
5258 md_super_wait(mddev);
5260 if (mddev->ro == 0 &&
5261 (!mddev->in_sync || (mddev->flags & MD_UPDATE_SB_FLAGS))) {
5262 /* mark array as shutdown cleanly */
5264 md_update_sb(mddev, 1);
5268 void md_stop_writes(struct mddev *mddev)
5270 mddev_lock_nointr(mddev);
5271 __md_stop_writes(mddev);
5272 mddev_unlock(mddev);
5274 EXPORT_SYMBOL_GPL(md_stop_writes);
5276 static void __md_stop(struct mddev *mddev)
5279 mddev->pers->stop(mddev);
5280 if (mddev->pers->sync_request && mddev->to_remove == NULL)
5281 mddev->to_remove = &md_redundancy_group;
5282 module_put(mddev->pers->owner);
5284 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5287 void md_stop(struct mddev *mddev)
5289 /* stop the array and free an attached data structures.
5290 * This is called from dm-raid
5293 bitmap_destroy(mddev);
5295 bioset_free(mddev->bio_set);
5298 EXPORT_SYMBOL_GPL(md_stop);
5300 static int md_set_readonly(struct mddev *mddev, struct block_device *bdev)
5305 if (!test_bit(MD_RECOVERY_FROZEN, &mddev->recovery)) {
5307 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5308 md_wakeup_thread(mddev->thread);
5310 if (mddev->sync_thread) {
5311 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5312 /* Thread might be blocked waiting for metadata update
5313 * which will now never happen */
5314 wake_up_process(mddev->sync_thread->tsk);
5316 mddev_unlock(mddev);
5317 wait_event(resync_wait, mddev->sync_thread == NULL);
5318 mddev_lock_nointr(mddev);
5320 mutex_lock(&mddev->open_mutex);
5321 if (atomic_read(&mddev->openers) > !!bdev ||
5322 mddev->sync_thread ||
5323 (bdev && !test_bit(MD_STILL_CLOSED, &mddev->flags))) {
5324 printk("md: %s still in use.\n",mdname(mddev));
5326 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5327 md_wakeup_thread(mddev->thread);
5333 __md_stop_writes(mddev);
5339 set_disk_ro(mddev->gendisk, 1);
5340 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5341 sysfs_notify_dirent_safe(mddev->sysfs_state);
5345 mutex_unlock(&mddev->open_mutex);
5350 * 0 - completely stop and dis-assemble array
5351 * 2 - stop but do not disassemble array
5353 static int do_md_stop(struct mddev * mddev, int mode,
5354 struct block_device *bdev)
5356 struct gendisk *disk = mddev->gendisk;
5357 struct md_rdev *rdev;
5360 if (!test_bit(MD_RECOVERY_FROZEN, &mddev->recovery)) {
5362 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5363 md_wakeup_thread(mddev->thread);
5365 if (mddev->sync_thread) {
5366 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5367 /* Thread might be blocked waiting for metadata update
5368 * which will now never happen */
5369 wake_up_process(mddev->sync_thread->tsk);
5371 mddev_unlock(mddev);
5372 wait_event(resync_wait, mddev->sync_thread == NULL);
5373 mddev_lock_nointr(mddev);
5375 mutex_lock(&mddev->open_mutex);
5376 if (atomic_read(&mddev->openers) > !!bdev ||
5377 mddev->sysfs_active ||
5378 mddev->sync_thread ||
5379 (bdev && !test_bit(MD_STILL_CLOSED, &mddev->flags))) {
5380 printk("md: %s still in use.\n",mdname(mddev));
5381 mutex_unlock(&mddev->open_mutex);
5383 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5384 md_wakeup_thread(mddev->thread);
5390 set_disk_ro(disk, 0);
5392 __md_stop_writes(mddev);
5394 mddev->queue->merge_bvec_fn = NULL;
5395 mddev->queue->backing_dev_info.congested_fn = NULL;
5397 /* tell userspace to handle 'inactive' */
5398 sysfs_notify_dirent_safe(mddev->sysfs_state);
5400 rdev_for_each(rdev, mddev)
5401 if (rdev->raid_disk >= 0)
5402 sysfs_unlink_rdev(mddev, rdev);
5404 set_capacity(disk, 0);
5405 mutex_unlock(&mddev->open_mutex);
5407 revalidate_disk(disk);
5412 mutex_unlock(&mddev->open_mutex);
5414 * Free resources if final stop
5417 printk(KERN_INFO "md: %s stopped.\n", mdname(mddev));
5419 bitmap_destroy(mddev);
5420 if (mddev->bitmap_info.file) {
5421 restore_bitmap_write_access(mddev->bitmap_info.file);
5422 fput(mddev->bitmap_info.file);
5423 mddev->bitmap_info.file = NULL;
5425 mddev->bitmap_info.offset = 0;
5427 export_array(mddev);
5430 kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
5431 if (mddev->hold_active == UNTIL_STOP)
5432 mddev->hold_active = 0;
5434 blk_integrity_unregister(disk);
5435 md_new_event(mddev);
5436 sysfs_notify_dirent_safe(mddev->sysfs_state);
5441 static void autorun_array(struct mddev *mddev)
5443 struct md_rdev *rdev;
5446 if (list_empty(&mddev->disks))
5449 printk(KERN_INFO "md: running: ");
5451 rdev_for_each(rdev, mddev) {
5452 char b[BDEVNAME_SIZE];
5453 printk("<%s>", bdevname(rdev->bdev,b));
5457 err = do_md_run(mddev);
5459 printk(KERN_WARNING "md: do_md_run() returned %d\n", err);
5460 do_md_stop(mddev, 0, NULL);
5465 * lets try to run arrays based on all disks that have arrived
5466 * until now. (those are in pending_raid_disks)
5468 * the method: pick the first pending disk, collect all disks with
5469 * the same UUID, remove all from the pending list and put them into
5470 * the 'same_array' list. Then order this list based on superblock
5471 * update time (freshest comes first), kick out 'old' disks and
5472 * compare superblocks. If everything's fine then run it.
5474 * If "unit" is allocated, then bump its reference count
5476 static void autorun_devices(int part)
5478 struct md_rdev *rdev0, *rdev, *tmp;
5479 struct mddev *mddev;
5480 char b[BDEVNAME_SIZE];
5482 printk(KERN_INFO "md: autorun ...\n");
5483 while (!list_empty(&pending_raid_disks)) {
5486 LIST_HEAD(candidates);
5487 rdev0 = list_entry(pending_raid_disks.next,
5488 struct md_rdev, same_set);
5490 printk(KERN_INFO "md: considering %s ...\n",
5491 bdevname(rdev0->bdev,b));
5492 INIT_LIST_HEAD(&candidates);
5493 rdev_for_each_list(rdev, tmp, &pending_raid_disks)
5494 if (super_90_load(rdev, rdev0, 0) >= 0) {
5495 printk(KERN_INFO "md: adding %s ...\n",
5496 bdevname(rdev->bdev,b));
5497 list_move(&rdev->same_set, &candidates);
5500 * now we have a set of devices, with all of them having
5501 * mostly sane superblocks. It's time to allocate the
5505 dev = MKDEV(mdp_major,
5506 rdev0->preferred_minor << MdpMinorShift);
5507 unit = MINOR(dev) >> MdpMinorShift;
5509 dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
5512 if (rdev0->preferred_minor != unit) {
5513 printk(KERN_INFO "md: unit number in %s is bad: %d\n",
5514 bdevname(rdev0->bdev, b), rdev0->preferred_minor);
5518 md_probe(dev, NULL, NULL);
5519 mddev = mddev_find(dev);
5520 if (!mddev || !mddev->gendisk) {
5524 "md: cannot allocate memory for md drive.\n");
5527 if (mddev_lock(mddev))
5528 printk(KERN_WARNING "md: %s locked, cannot run\n",
5530 else if (mddev->raid_disks || mddev->major_version
5531 || !list_empty(&mddev->disks)) {
5533 "md: %s already running, cannot run %s\n",
5534 mdname(mddev), bdevname(rdev0->bdev,b));
5535 mddev_unlock(mddev);
5537 printk(KERN_INFO "md: created %s\n", mdname(mddev));
5538 mddev->persistent = 1;
5539 rdev_for_each_list(rdev, tmp, &candidates) {
5540 list_del_init(&rdev->same_set);
5541 if (bind_rdev_to_array(rdev, mddev))
5544 autorun_array(mddev);
5545 mddev_unlock(mddev);
5547 /* on success, candidates will be empty, on error
5550 rdev_for_each_list(rdev, tmp, &candidates) {
5551 list_del_init(&rdev->same_set);
5556 printk(KERN_INFO "md: ... autorun DONE.\n");
5558 #endif /* !MODULE */
5560 static int get_version(void __user * arg)
5564 ver.major = MD_MAJOR_VERSION;
5565 ver.minor = MD_MINOR_VERSION;
5566 ver.patchlevel = MD_PATCHLEVEL_VERSION;
5568 if (copy_to_user(arg, &ver, sizeof(ver)))
5574 static int get_array_info(struct mddev * mddev, void __user * arg)
5576 mdu_array_info_t info;
5577 int nr,working,insync,failed,spare;
5578 struct md_rdev *rdev;
5580 nr = working = insync = failed = spare = 0;
5582 rdev_for_each_rcu(rdev, mddev) {
5584 if (test_bit(Faulty, &rdev->flags))
5588 if (test_bit(In_sync, &rdev->flags))
5596 info.major_version = mddev->major_version;
5597 info.minor_version = mddev->minor_version;
5598 info.patch_version = MD_PATCHLEVEL_VERSION;
5599 info.ctime = mddev->ctime;
5600 info.level = mddev->level;
5601 info.size = mddev->dev_sectors / 2;
5602 if (info.size != mddev->dev_sectors / 2) /* overflow */
5605 info.raid_disks = mddev->raid_disks;
5606 info.md_minor = mddev->md_minor;
5607 info.not_persistent= !mddev->persistent;
5609 info.utime = mddev->utime;
5612 info.state = (1<<MD_SB_CLEAN);
5613 if (mddev->bitmap && mddev->bitmap_info.offset)
5614 info.state = (1<<MD_SB_BITMAP_PRESENT);
5615 info.active_disks = insync;
5616 info.working_disks = working;
5617 info.failed_disks = failed;
5618 info.spare_disks = spare;
5620 info.layout = mddev->layout;
5621 info.chunk_size = mddev->chunk_sectors << 9;
5623 if (copy_to_user(arg, &info, sizeof(info)))
5629 static int get_bitmap_file(struct mddev * mddev, void __user * arg)
5631 mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */
5632 char *ptr, *buf = NULL;
5635 file = kmalloc(sizeof(*file), GFP_NOIO);
5640 /* bitmap disabled, zero the first byte and copy out */
5641 if (!mddev->bitmap || !mddev->bitmap->storage.file) {
5642 file->pathname[0] = '\0';
5646 buf = kmalloc(sizeof(file->pathname), GFP_KERNEL);
5650 ptr = d_path(&mddev->bitmap->storage.file->f_path,
5651 buf, sizeof(file->pathname));
5655 strcpy(file->pathname, ptr);
5659 if (copy_to_user(arg, file, sizeof(*file)))
5667 static int get_disk_info(struct mddev * mddev, void __user * arg)
5669 mdu_disk_info_t info;
5670 struct md_rdev *rdev;
5672 if (copy_from_user(&info, arg, sizeof(info)))
5676 rdev = find_rdev_nr_rcu(mddev, info.number);
5678 info.major = MAJOR(rdev->bdev->bd_dev);
5679 info.minor = MINOR(rdev->bdev->bd_dev);
5680 info.raid_disk = rdev->raid_disk;
5682 if (test_bit(Faulty, &rdev->flags))
5683 info.state |= (1<<MD_DISK_FAULTY);
5684 else if (test_bit(In_sync, &rdev->flags)) {
5685 info.state |= (1<<MD_DISK_ACTIVE);
5686 info.state |= (1<<MD_DISK_SYNC);
5688 if (test_bit(WriteMostly, &rdev->flags))
5689 info.state |= (1<<MD_DISK_WRITEMOSTLY);
5691 info.major = info.minor = 0;
5692 info.raid_disk = -1;
5693 info.state = (1<<MD_DISK_REMOVED);
5697 if (copy_to_user(arg, &info, sizeof(info)))
5703 static int add_new_disk(struct mddev * mddev, mdu_disk_info_t *info)
5705 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
5706 struct md_rdev *rdev;
5707 dev_t dev = MKDEV(info->major,info->minor);
5709 if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
5712 if (!mddev->raid_disks) {
5714 /* expecting a device which has a superblock */
5715 rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
5718 "md: md_import_device returned %ld\n",
5720 return PTR_ERR(rdev);
5722 if (!list_empty(&mddev->disks)) {
5723 struct md_rdev *rdev0
5724 = list_entry(mddev->disks.next,
5725 struct md_rdev, same_set);
5726 err = super_types[mddev->major_version]
5727 .load_super(rdev, rdev0, mddev->minor_version);
5730 "md: %s has different UUID to %s\n",
5731 bdevname(rdev->bdev,b),
5732 bdevname(rdev0->bdev,b2));
5737 err = bind_rdev_to_array(rdev, mddev);
5744 * add_new_disk can be used once the array is assembled
5745 * to add "hot spares". They must already have a superblock
5750 if (!mddev->pers->hot_add_disk) {
5752 "%s: personality does not support diskops!\n",
5756 if (mddev->persistent)
5757 rdev = md_import_device(dev, mddev->major_version,
5758 mddev->minor_version);
5760 rdev = md_import_device(dev, -1, -1);
5763 "md: md_import_device returned %ld\n",
5765 return PTR_ERR(rdev);
5767 /* set saved_raid_disk if appropriate */
5768 if (!mddev->persistent) {
5769 if (info->state & (1<<MD_DISK_SYNC) &&
5770 info->raid_disk < mddev->raid_disks) {
5771 rdev->raid_disk = info->raid_disk;
5772 set_bit(In_sync, &rdev->flags);
5774 rdev->raid_disk = -1;
5776 super_types[mddev->major_version].
5777 validate_super(mddev, rdev);
5778 if ((info->state & (1<<MD_DISK_SYNC)) &&
5779 rdev->raid_disk != info->raid_disk) {
5780 /* This was a hot-add request, but events doesn't
5781 * match, so reject it.
5787 if (test_bit(In_sync, &rdev->flags))
5788 rdev->saved_raid_disk = rdev->raid_disk;
5790 rdev->saved_raid_disk = -1;
5792 clear_bit(In_sync, &rdev->flags); /* just to be sure */
5793 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
5794 set_bit(WriteMostly, &rdev->flags);
5796 clear_bit(WriteMostly, &rdev->flags);
5798 rdev->raid_disk = -1;
5799 err = bind_rdev_to_array(rdev, mddev);
5800 if (!err && !mddev->pers->hot_remove_disk) {
5801 /* If there is hot_add_disk but no hot_remove_disk
5802 * then added disks for geometry changes,
5803 * and should be added immediately.
5805 super_types[mddev->major_version].
5806 validate_super(mddev, rdev);
5807 err = mddev->pers->hot_add_disk(mddev, rdev);
5809 unbind_rdev_from_array(rdev);
5814 sysfs_notify_dirent_safe(rdev->sysfs_state);
5816 set_bit(MD_CHANGE_DEVS, &mddev->flags);
5817 if (mddev->degraded)
5818 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
5819 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5821 md_new_event(mddev);
5822 md_wakeup_thread(mddev->thread);
5826 /* otherwise, add_new_disk is only allowed
5827 * for major_version==0 superblocks
5829 if (mddev->major_version != 0) {
5830 printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n",
5835 if (!(info->state & (1<<MD_DISK_FAULTY))) {
5837 rdev = md_import_device(dev, -1, 0);
5840 "md: error, md_import_device() returned %ld\n",
5842 return PTR_ERR(rdev);
5844 rdev->desc_nr = info->number;
5845 if (info->raid_disk < mddev->raid_disks)
5846 rdev->raid_disk = info->raid_disk;
5848 rdev->raid_disk = -1;
5850 if (rdev->raid_disk < mddev->raid_disks)
5851 if (info->state & (1<<MD_DISK_SYNC))
5852 set_bit(In_sync, &rdev->flags);
5854 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
5855 set_bit(WriteMostly, &rdev->flags);
5857 if (!mddev->persistent) {
5858 printk(KERN_INFO "md: nonpersistent superblock ...\n");
5859 rdev->sb_start = i_size_read(rdev->bdev->bd_inode) / 512;
5861 rdev->sb_start = calc_dev_sboffset(rdev);
5862 rdev->sectors = rdev->sb_start;
5864 err = bind_rdev_to_array(rdev, mddev);
5874 static int hot_remove_disk(struct mddev * mddev, dev_t dev)
5876 char b[BDEVNAME_SIZE];
5877 struct md_rdev *rdev;
5879 rdev = find_rdev(mddev, dev);
5883 clear_bit(Blocked, &rdev->flags);
5884 remove_and_add_spares(mddev, rdev);
5886 if (rdev->raid_disk >= 0)
5889 kick_rdev_from_array(rdev);
5890 md_update_sb(mddev, 1);
5891 md_new_event(mddev);
5895 printk(KERN_WARNING "md: cannot remove active disk %s from %s ...\n",
5896 bdevname(rdev->bdev,b), mdname(mddev));
5900 static int hot_add_disk(struct mddev * mddev, dev_t dev)
5902 char b[BDEVNAME_SIZE];
5904 struct md_rdev *rdev;
5909 if (mddev->major_version != 0) {
5910 printk(KERN_WARNING "%s: HOT_ADD may only be used with"
5911 " version-0 superblocks.\n",
5915 if (!mddev->pers->hot_add_disk) {
5917 "%s: personality does not support diskops!\n",
5922 rdev = md_import_device(dev, -1, 0);
5925 "md: error, md_import_device() returned %ld\n",
5930 if (mddev->persistent)
5931 rdev->sb_start = calc_dev_sboffset(rdev);
5933 rdev->sb_start = i_size_read(rdev->bdev->bd_inode) / 512;
5935 rdev->sectors = rdev->sb_start;
5937 if (test_bit(Faulty, &rdev->flags)) {
5939 "md: can not hot-add faulty %s disk to %s!\n",
5940 bdevname(rdev->bdev,b), mdname(mddev));
5944 clear_bit(In_sync, &rdev->flags);
5946 rdev->saved_raid_disk = -1;
5947 err = bind_rdev_to_array(rdev, mddev);
5952 * The rest should better be atomic, we can have disk failures
5953 * noticed in interrupt contexts ...
5956 rdev->raid_disk = -1;
5958 md_update_sb(mddev, 1);
5961 * Kick recovery, maybe this spare has to be added to the
5962 * array immediately.
5964 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5965 md_wakeup_thread(mddev->thread);
5966 md_new_event(mddev);
5974 static int set_bitmap_file(struct mddev *mddev, int fd)
5979 if (!mddev->pers->quiesce)
5981 if (mddev->recovery || mddev->sync_thread)
5983 /* we should be able to change the bitmap.. */
5989 return -EEXIST; /* cannot add when bitmap is present */
5990 mddev->bitmap_info.file = fget(fd);
5992 if (mddev->bitmap_info.file == NULL) {
5993 printk(KERN_ERR "%s: error: failed to get bitmap file\n",
5998 err = deny_bitmap_write_access(mddev->bitmap_info.file);
6000 printk(KERN_ERR "%s: error: bitmap file is already in use\n",
6002 fput(mddev->bitmap_info.file);
6003 mddev->bitmap_info.file = NULL;
6006 mddev->bitmap_info.offset = 0; /* file overrides offset */
6007 } else if (mddev->bitmap == NULL)
6008 return -ENOENT; /* cannot remove what isn't there */
6011 mddev->pers->quiesce(mddev, 1);
6013 err = bitmap_create(mddev);
6015 err = bitmap_load(mddev);
6017 if (fd < 0 || err) {
6018 bitmap_destroy(mddev);
6019 fd = -1; /* make sure to put the file */
6021 mddev->pers->quiesce(mddev, 0);
6024 if (mddev->bitmap_info.file) {
6025 restore_bitmap_write_access(mddev->bitmap_info.file);
6026 fput(mddev->bitmap_info.file);
6028 mddev->bitmap_info.file = NULL;
6035 * set_array_info is used two different ways
6036 * The original usage is when creating a new array.
6037 * In this usage, raid_disks is > 0 and it together with
6038 * level, size, not_persistent,layout,chunksize determine the
6039 * shape of the array.
6040 * This will always create an array with a type-0.90.0 superblock.
6041 * The newer usage is when assembling an array.
6042 * In this case raid_disks will be 0, and the major_version field is
6043 * use to determine which style super-blocks are to be found on the devices.
6044 * The minor and patch _version numbers are also kept incase the
6045 * super_block handler wishes to interpret them.
6047 static int set_array_info(struct mddev * mddev, mdu_array_info_t *info)
6050 if (info->raid_disks == 0) {
6051 /* just setting version number for superblock loading */
6052 if (info->major_version < 0 ||
6053 info->major_version >= ARRAY_SIZE(super_types) ||
6054 super_types[info->major_version].name == NULL) {
6055 /* maybe try to auto-load a module? */
6057 "md: superblock version %d not known\n",
6058 info->major_version);
6061 mddev->major_version = info->major_version;
6062 mddev->minor_version = info->minor_version;
6063 mddev->patch_version = info->patch_version;
6064 mddev->persistent = !info->not_persistent;
6065 /* ensure mddev_put doesn't delete this now that there
6066 * is some minimal configuration.
6068 mddev->ctime = get_seconds();
6071 mddev->major_version = MD_MAJOR_VERSION;
6072 mddev->minor_version = MD_MINOR_VERSION;
6073 mddev->patch_version = MD_PATCHLEVEL_VERSION;
6074 mddev->ctime = get_seconds();
6076 mddev->level = info->level;
6077 mddev->clevel[0] = 0;
6078 mddev->dev_sectors = 2 * (sector_t)info->size;
6079 mddev->raid_disks = info->raid_disks;
6080 /* don't set md_minor, it is determined by which /dev/md* was
6083 if (info->state & (1<<MD_SB_CLEAN))
6084 mddev->recovery_cp = MaxSector;
6086 mddev->recovery_cp = 0;
6087 mddev->persistent = ! info->not_persistent;
6088 mddev->external = 0;
6090 mddev->layout = info->layout;
6091 mddev->chunk_sectors = info->chunk_size >> 9;
6093 mddev->max_disks = MD_SB_DISKS;
6095 if (mddev->persistent)
6097 set_bit(MD_CHANGE_DEVS, &mddev->flags);
6099 mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9;
6100 mddev->bitmap_info.default_space = 64*2 - (MD_SB_BYTES >> 9);
6101 mddev->bitmap_info.offset = 0;
6103 mddev->reshape_position = MaxSector;
6106 * Generate a 128 bit UUID
6108 get_random_bytes(mddev->uuid, 16);
6110 mddev->new_level = mddev->level;
6111 mddev->new_chunk_sectors = mddev->chunk_sectors;
6112 mddev->new_layout = mddev->layout;
6113 mddev->delta_disks = 0;
6114 mddev->reshape_backwards = 0;
6119 void md_set_array_sectors(struct mddev *mddev, sector_t array_sectors)
6121 WARN(!mddev_is_locked(mddev), "%s: unlocked mddev!\n", __func__);
6123 if (mddev->external_size)
6126 mddev->array_sectors = array_sectors;
6128 EXPORT_SYMBOL(md_set_array_sectors);
6130 static int update_size(struct mddev *mddev, sector_t num_sectors)
6132 struct md_rdev *rdev;
6134 int fit = (num_sectors == 0);
6136 if (mddev->pers->resize == NULL)
6138 /* The "num_sectors" is the number of sectors of each device that
6139 * is used. This can only make sense for arrays with redundancy.
6140 * linear and raid0 always use whatever space is available. We can only
6141 * consider changing this number if no resync or reconstruction is
6142 * happening, and if the new size is acceptable. It must fit before the
6143 * sb_start or, if that is <data_offset, it must fit before the size
6144 * of each device. If num_sectors is zero, we find the largest size
6147 if (mddev->sync_thread)
6150 rdev_for_each(rdev, mddev) {
6151 sector_t avail = rdev->sectors;
6153 if (fit && (num_sectors == 0 || num_sectors > avail))
6154 num_sectors = avail;
6155 if (avail < num_sectors)
6158 rv = mddev->pers->resize(mddev, num_sectors);
6160 revalidate_disk(mddev->gendisk);
6164 static int update_raid_disks(struct mddev *mddev, int raid_disks)
6167 struct md_rdev *rdev;
6168 /* change the number of raid disks */
6169 if (mddev->pers->check_reshape == NULL)
6171 if (raid_disks <= 0 ||
6172 (mddev->max_disks && raid_disks >= mddev->max_disks))
6174 if (mddev->sync_thread || mddev->reshape_position != MaxSector)
6177 rdev_for_each(rdev, mddev) {
6178 if (mddev->raid_disks < raid_disks &&
6179 rdev->data_offset < rdev->new_data_offset)
6181 if (mddev->raid_disks > raid_disks &&
6182 rdev->data_offset > rdev->new_data_offset)
6186 mddev->delta_disks = raid_disks - mddev->raid_disks;
6187 if (mddev->delta_disks < 0)
6188 mddev->reshape_backwards = 1;
6189 else if (mddev->delta_disks > 0)
6190 mddev->reshape_backwards = 0;
6192 rv = mddev->pers->check_reshape(mddev);
6194 mddev->delta_disks = 0;
6195 mddev->reshape_backwards = 0;
6202 * update_array_info is used to change the configuration of an
6204 * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
6205 * fields in the info are checked against the array.
6206 * Any differences that cannot be handled will cause an error.
6207 * Normally, only one change can be managed at a time.
6209 static int update_array_info(struct mddev *mddev, mdu_array_info_t *info)
6215 /* calculate expected state,ignoring low bits */
6216 if (mddev->bitmap && mddev->bitmap_info.offset)
6217 state |= (1 << MD_SB_BITMAP_PRESENT);
6219 if (mddev->major_version != info->major_version ||
6220 mddev->minor_version != info->minor_version ||
6221 /* mddev->patch_version != info->patch_version || */
6222 mddev->ctime != info->ctime ||
6223 mddev->level != info->level ||
6224 /* mddev->layout != info->layout || */
6225 !mddev->persistent != info->not_persistent||
6226 mddev->chunk_sectors != info->chunk_size >> 9 ||
6227 /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
6228 ((state^info->state) & 0xfffffe00)
6231 /* Check there is only one change */
6232 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
6234 if (mddev->raid_disks != info->raid_disks)
6236 if (mddev->layout != info->layout)
6238 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT))
6245 if (mddev->layout != info->layout) {
6247 * we don't need to do anything at the md level, the
6248 * personality will take care of it all.
6250 if (mddev->pers->check_reshape == NULL)
6253 mddev->new_layout = info->layout;
6254 rv = mddev->pers->check_reshape(mddev);
6256 mddev->new_layout = mddev->layout;
6260 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
6261 rv = update_size(mddev, (sector_t)info->size * 2);
6263 if (mddev->raid_disks != info->raid_disks)
6264 rv = update_raid_disks(mddev, info->raid_disks);
6266 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) {
6267 if (mddev->pers->quiesce == NULL)
6269 if (mddev->recovery || mddev->sync_thread)
6271 if (info->state & (1<<MD_SB_BITMAP_PRESENT)) {
6272 /* add the bitmap */
6275 if (mddev->bitmap_info.default_offset == 0)
6277 mddev->bitmap_info.offset =
6278 mddev->bitmap_info.default_offset;
6279 mddev->bitmap_info.space =
6280 mddev->bitmap_info.default_space;
6281 mddev->pers->quiesce(mddev, 1);
6282 rv = bitmap_create(mddev);
6284 rv = bitmap_load(mddev);
6286 bitmap_destroy(mddev);
6287 mddev->pers->quiesce(mddev, 0);
6289 /* remove the bitmap */
6292 if (mddev->bitmap->storage.file)
6294 mddev->pers->quiesce(mddev, 1);
6295 bitmap_destroy(mddev);
6296 mddev->pers->quiesce(mddev, 0);
6297 mddev->bitmap_info.offset = 0;
6300 md_update_sb(mddev, 1);
6304 static int set_disk_faulty(struct mddev *mddev, dev_t dev)
6306 struct md_rdev *rdev;
6309 if (mddev->pers == NULL)
6313 rdev = find_rdev_rcu(mddev, dev);
6317 md_error(mddev, rdev);
6318 if (!test_bit(Faulty, &rdev->flags))
6326 * We have a problem here : there is no easy way to give a CHS
6327 * virtual geometry. We currently pretend that we have a 2 heads
6328 * 4 sectors (with a BIG number of cylinders...). This drives
6329 * dosfs just mad... ;-)
6331 static int md_getgeo(struct block_device *bdev, struct hd_geometry *geo)
6333 struct mddev *mddev = bdev->bd_disk->private_data;
6337 geo->cylinders = mddev->array_sectors / 8;
6341 static int md_ioctl(struct block_device *bdev, fmode_t mode,
6342 unsigned int cmd, unsigned long arg)
6345 void __user *argp = (void __user *)arg;
6346 struct mddev *mddev = NULL;
6351 case GET_ARRAY_INFO:
6355 if (!capable(CAP_SYS_ADMIN))
6360 * Commands dealing with the RAID driver but not any
6365 err = get_version(argp);
6368 case PRINT_RAID_DEBUG:
6376 autostart_arrays(arg);
6383 * Commands creating/starting a new array:
6386 mddev = bdev->bd_disk->private_data;
6393 /* Some actions do not requires the mutex */
6395 case GET_ARRAY_INFO:
6396 if (!mddev->raid_disks && !mddev->external)
6399 err = get_array_info(mddev, argp);
6403 if (!mddev->raid_disks && !mddev->external)
6406 err = get_disk_info(mddev, argp);
6409 case SET_DISK_FAULTY:
6410 err = set_disk_faulty(mddev, new_decode_dev(arg));
6414 if (cmd == ADD_NEW_DISK)
6415 /* need to ensure md_delayed_delete() has completed */
6416 flush_workqueue(md_misc_wq);
6418 if (cmd == HOT_REMOVE_DISK)
6419 /* need to ensure recovery thread has run */
6420 wait_event_interruptible_timeout(mddev->sb_wait,
6421 !test_bit(MD_RECOVERY_NEEDED,
6423 msecs_to_jiffies(5000));
6424 if (cmd == STOP_ARRAY || cmd == STOP_ARRAY_RO) {
6425 /* Need to flush page cache, and ensure no-one else opens
6428 mutex_lock(&mddev->open_mutex);
6429 if (atomic_read(&mddev->openers) > 1) {
6430 mutex_unlock(&mddev->open_mutex);
6434 set_bit(MD_STILL_CLOSED, &mddev->flags);
6435 mutex_unlock(&mddev->open_mutex);
6436 sync_blockdev(bdev);
6438 err = mddev_lock(mddev);
6441 "md: ioctl lock interrupted, reason %d, cmd %d\n",
6446 if (cmd == SET_ARRAY_INFO) {
6447 mdu_array_info_t info;
6449 memset(&info, 0, sizeof(info));
6450 else if (copy_from_user(&info, argp, sizeof(info))) {
6455 err = update_array_info(mddev, &info);
6457 printk(KERN_WARNING "md: couldn't update"
6458 " array info. %d\n", err);
6463 if (!list_empty(&mddev->disks)) {
6465 "md: array %s already has disks!\n",
6470 if (mddev->raid_disks) {
6472 "md: array %s already initialised!\n",
6477 err = set_array_info(mddev, &info);
6479 printk(KERN_WARNING "md: couldn't set"
6480 " array info. %d\n", err);
6487 * Commands querying/configuring an existing array:
6489 /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
6490 * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */
6491 if ((!mddev->raid_disks && !mddev->external)
6492 && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY
6493 && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE
6494 && cmd != GET_BITMAP_FILE) {
6500 * Commands even a read-only array can execute:
6503 case GET_BITMAP_FILE:
6504 err = get_bitmap_file(mddev, argp);
6507 case RESTART_ARRAY_RW:
6508 err = restart_array(mddev);
6512 err = do_md_stop(mddev, 0, bdev);
6516 err = md_set_readonly(mddev, bdev);
6519 case HOT_REMOVE_DISK:
6520 err = hot_remove_disk(mddev, new_decode_dev(arg));
6524 /* We can support ADD_NEW_DISK on read-only arrays
6525 * on if we are re-adding a preexisting device.
6526 * So require mddev->pers and MD_DISK_SYNC.
6529 mdu_disk_info_t info;
6530 if (copy_from_user(&info, argp, sizeof(info)))
6532 else if (!(info.state & (1<<MD_DISK_SYNC)))
6533 /* Need to clear read-only for this */
6536 err = add_new_disk(mddev, &info);
6542 if (get_user(ro, (int __user *)(arg))) {
6548 /* if the bdev is going readonly the value of mddev->ro
6549 * does not matter, no writes are coming
6554 /* are we are already prepared for writes? */
6558 /* transitioning to readauto need only happen for
6559 * arrays that call md_write_start
6562 err = restart_array(mddev);
6565 set_disk_ro(mddev->gendisk, 0);
6572 * The remaining ioctls are changing the state of the
6573 * superblock, so we do not allow them on read-only arrays.
6574 * However non-MD ioctls (e.g. get-size) will still come through
6575 * here and hit the 'default' below, so only disallow
6576 * 'md' ioctls, and switch to rw mode if started auto-readonly.
6578 if (_IOC_TYPE(cmd) == MD_MAJOR && mddev->ro && mddev->pers) {
6579 if (mddev->ro == 2) {
6581 sysfs_notify_dirent_safe(mddev->sysfs_state);
6582 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6583 /* mddev_unlock will wake thread */
6584 /* If a device failed while we were read-only, we
6585 * need to make sure the metadata is updated now.
6587 if (test_bit(MD_CHANGE_DEVS, &mddev->flags)) {
6588 mddev_unlock(mddev);
6589 wait_event(mddev->sb_wait,
6590 !test_bit(MD_CHANGE_DEVS, &mddev->flags) &&
6591 !test_bit(MD_CHANGE_PENDING, &mddev->flags));
6592 mddev_lock_nointr(mddev);
6603 mdu_disk_info_t info;
6604 if (copy_from_user(&info, argp, sizeof(info)))
6607 err = add_new_disk(mddev, &info);
6612 err = hot_add_disk(mddev, new_decode_dev(arg));
6616 err = do_md_run(mddev);
6619 case SET_BITMAP_FILE:
6620 err = set_bitmap_file(mddev, (int)arg);
6630 if (mddev->hold_active == UNTIL_IOCTL &&
6632 mddev->hold_active = 0;
6633 mddev_unlock(mddev);
6642 #ifdef CONFIG_COMPAT
6643 static int md_compat_ioctl(struct block_device *bdev, fmode_t mode,
6644 unsigned int cmd, unsigned long arg)
6647 case HOT_REMOVE_DISK:
6649 case SET_DISK_FAULTY:
6650 case SET_BITMAP_FILE:
6651 /* These take in integer arg, do not convert */
6654 arg = (unsigned long)compat_ptr(arg);
6658 return md_ioctl(bdev, mode, cmd, arg);
6660 #endif /* CONFIG_COMPAT */
6662 static int md_open(struct block_device *bdev, fmode_t mode)
6665 * Succeed if we can lock the mddev, which confirms that
6666 * it isn't being stopped right now.
6668 struct mddev *mddev = mddev_find(bdev->bd_dev);
6674 if (mddev->gendisk != bdev->bd_disk) {
6675 /* we are racing with mddev_put which is discarding this
6679 /* Wait until bdev->bd_disk is definitely gone */
6680 flush_workqueue(md_misc_wq);
6681 /* Then retry the open from the top */
6682 return -ERESTARTSYS;
6684 BUG_ON(mddev != bdev->bd_disk->private_data);
6686 if ((err = mutex_lock_interruptible(&mddev->open_mutex)))
6690 atomic_inc(&mddev->openers);
6691 clear_bit(MD_STILL_CLOSED, &mddev->flags);
6692 mutex_unlock(&mddev->open_mutex);
6694 check_disk_change(bdev);
6699 static void md_release(struct gendisk *disk, fmode_t mode)
6701 struct mddev *mddev = disk->private_data;
6704 atomic_dec(&mddev->openers);
6708 static int md_media_changed(struct gendisk *disk)
6710 struct mddev *mddev = disk->private_data;
6712 return mddev->changed;
6715 static int md_revalidate(struct gendisk *disk)
6717 struct mddev *mddev = disk->private_data;
6722 static const struct block_device_operations md_fops =
6724 .owner = THIS_MODULE,
6726 .release = md_release,
6728 #ifdef CONFIG_COMPAT
6729 .compat_ioctl = md_compat_ioctl,
6731 .getgeo = md_getgeo,
6732 .media_changed = md_media_changed,
6733 .revalidate_disk= md_revalidate,
6736 static int md_thread(void * arg)
6738 struct md_thread *thread = arg;
6741 * md_thread is a 'system-thread', it's priority should be very
6742 * high. We avoid resource deadlocks individually in each
6743 * raid personality. (RAID5 does preallocation) We also use RR and
6744 * the very same RT priority as kswapd, thus we will never get
6745 * into a priority inversion deadlock.
6747 * we definitely have to have equal or higher priority than
6748 * bdflush, otherwise bdflush will deadlock if there are too
6749 * many dirty RAID5 blocks.
6752 allow_signal(SIGKILL);
6753 while (!kthread_should_stop()) {
6755 /* We need to wait INTERRUPTIBLE so that
6756 * we don't add to the load-average.
6757 * That means we need to be sure no signals are
6760 if (signal_pending(current))
6761 flush_signals(current);
6763 wait_event_interruptible_timeout
6765 test_bit(THREAD_WAKEUP, &thread->flags)
6766 || kthread_should_stop(),
6769 clear_bit(THREAD_WAKEUP, &thread->flags);
6770 if (!kthread_should_stop())
6771 thread->run(thread);
6777 void md_wakeup_thread(struct md_thread *thread)
6780 pr_debug("md: waking up MD thread %s.\n", thread->tsk->comm);
6781 set_bit(THREAD_WAKEUP, &thread->flags);
6782 wake_up(&thread->wqueue);
6786 struct md_thread *md_register_thread(void (*run) (struct md_thread *),
6787 struct mddev *mddev, const char *name)
6789 struct md_thread *thread;
6791 thread = kzalloc(sizeof(struct md_thread), GFP_KERNEL);
6795 init_waitqueue_head(&thread->wqueue);
6798 thread->mddev = mddev;
6799 thread->timeout = MAX_SCHEDULE_TIMEOUT;
6800 thread->tsk = kthread_run(md_thread, thread,
6802 mdname(thread->mddev),
6804 if (IS_ERR(thread->tsk)) {
6811 void md_unregister_thread(struct md_thread **threadp)
6813 struct md_thread *thread = *threadp;
6816 pr_debug("interrupting MD-thread pid %d\n", task_pid_nr(thread->tsk));
6817 /* Locking ensures that mddev_unlock does not wake_up a
6818 * non-existent thread
6820 spin_lock(&pers_lock);
6822 spin_unlock(&pers_lock);
6824 kthread_stop(thread->tsk);
6828 void md_error(struct mddev *mddev, struct md_rdev *rdev)
6835 if (!rdev || test_bit(Faulty, &rdev->flags))
6838 if (!mddev->pers || !mddev->pers->error_handler)
6840 mddev->pers->error_handler(mddev,rdev);
6841 if (mddev->degraded)
6842 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
6843 sysfs_notify_dirent_safe(rdev->sysfs_state);
6844 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6845 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6846 md_wakeup_thread(mddev->thread);
6847 if (mddev->event_work.func)
6848 queue_work(md_misc_wq, &mddev->event_work);
6849 md_new_event_inintr(mddev);
6852 /* seq_file implementation /proc/mdstat */
6854 static void status_unused(struct seq_file *seq)
6857 struct md_rdev *rdev;
6859 seq_printf(seq, "unused devices: ");
6861 list_for_each_entry(rdev, &pending_raid_disks, same_set) {
6862 char b[BDEVNAME_SIZE];
6864 seq_printf(seq, "%s ",
6865 bdevname(rdev->bdev,b));
6868 seq_printf(seq, "<none>");
6870 seq_printf(seq, "\n");
6874 static void status_resync(struct seq_file *seq, struct mddev * mddev)
6876 sector_t max_sectors, resync, res;
6877 unsigned long dt, db;
6880 unsigned int per_milli;
6882 if (mddev->curr_resync <= 3)
6885 resync = mddev->curr_resync
6886 - atomic_read(&mddev->recovery_active);
6888 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ||
6889 test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
6890 max_sectors = mddev->resync_max_sectors;
6892 max_sectors = mddev->dev_sectors;
6895 * Should not happen.
6901 /* Pick 'scale' such that (resync>>scale)*1000 will fit
6902 * in a sector_t, and (max_sectors>>scale) will fit in a
6903 * u32, as those are the requirements for sector_div.
6904 * Thus 'scale' must be at least 10
6907 if (sizeof(sector_t) > sizeof(unsigned long)) {
6908 while ( max_sectors/2 > (1ULL<<(scale+32)))
6911 res = (resync>>scale)*1000;
6912 sector_div(res, (u32)((max_sectors>>scale)+1));
6916 int i, x = per_milli/50, y = 20-x;
6917 seq_printf(seq, "[");
6918 for (i = 0; i < x; i++)
6919 seq_printf(seq, "=");
6920 seq_printf(seq, ">");
6921 for (i = 0; i < y; i++)
6922 seq_printf(seq, ".");
6923 seq_printf(seq, "] ");
6925 seq_printf(seq, " %s =%3u.%u%% (%llu/%llu)",
6926 (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)?
6928 (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)?
6930 (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
6931 "resync" : "recovery"))),
6932 per_milli/10, per_milli % 10,
6933 (unsigned long long) resync/2,
6934 (unsigned long long) max_sectors/2);
6937 * dt: time from mark until now
6938 * db: blocks written from mark until now
6939 * rt: remaining time
6941 * rt is a sector_t, so could be 32bit or 64bit.
6942 * So we divide before multiply in case it is 32bit and close
6944 * We scale the divisor (db) by 32 to avoid losing precision
6945 * near the end of resync when the number of remaining sectors
6947 * We then divide rt by 32 after multiplying by db to compensate.
6948 * The '+1' avoids division by zero if db is very small.
6950 dt = ((jiffies - mddev->resync_mark) / HZ);
6952 db = (mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active))
6953 - mddev->resync_mark_cnt;
6955 rt = max_sectors - resync; /* number of remaining sectors */
6956 sector_div(rt, db/32+1);
6960 seq_printf(seq, " finish=%lu.%lumin", (unsigned long)rt / 60,
6961 ((unsigned long)rt % 60)/6);
6963 seq_printf(seq, " speed=%ldK/sec", db/2/dt);
6966 static void *md_seq_start(struct seq_file *seq, loff_t *pos)
6968 struct list_head *tmp;
6970 struct mddev *mddev;
6978 spin_lock(&all_mddevs_lock);
6979 list_for_each(tmp,&all_mddevs)
6981 mddev = list_entry(tmp, struct mddev, all_mddevs);
6983 spin_unlock(&all_mddevs_lock);
6986 spin_unlock(&all_mddevs_lock);
6988 return (void*)2;/* tail */
6992 static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
6994 struct list_head *tmp;
6995 struct mddev *next_mddev, *mddev = v;
7001 spin_lock(&all_mddevs_lock);
7003 tmp = all_mddevs.next;
7005 tmp = mddev->all_mddevs.next;
7006 if (tmp != &all_mddevs)
7007 next_mddev = mddev_get(list_entry(tmp,struct mddev,all_mddevs));
7009 next_mddev = (void*)2;
7012 spin_unlock(&all_mddevs_lock);
7020 static void md_seq_stop(struct seq_file *seq, void *v)
7022 struct mddev *mddev = v;
7024 if (mddev && v != (void*)1 && v != (void*)2)
7028 static int md_seq_show(struct seq_file *seq, void *v)
7030 struct mddev *mddev = v;
7032 struct md_rdev *rdev;
7034 if (v == (void*)1) {
7035 struct md_personality *pers;
7036 seq_printf(seq, "Personalities : ");
7037 spin_lock(&pers_lock);
7038 list_for_each_entry(pers, &pers_list, list)
7039 seq_printf(seq, "[%s] ", pers->name);
7041 spin_unlock(&pers_lock);
7042 seq_printf(seq, "\n");
7043 seq->poll_event = atomic_read(&md_event_count);
7046 if (v == (void*)2) {
7051 if (mddev_lock(mddev) < 0)
7054 if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
7055 seq_printf(seq, "%s : %sactive", mdname(mddev),
7056 mddev->pers ? "" : "in");
7059 seq_printf(seq, " (read-only)");
7061 seq_printf(seq, " (auto-read-only)");
7062 seq_printf(seq, " %s", mddev->pers->name);
7066 rdev_for_each(rdev, mddev) {
7067 char b[BDEVNAME_SIZE];
7068 seq_printf(seq, " %s[%d]",
7069 bdevname(rdev->bdev,b), rdev->desc_nr);
7070 if (test_bit(WriteMostly, &rdev->flags))
7071 seq_printf(seq, "(W)");
7072 if (test_bit(Faulty, &rdev->flags)) {
7073 seq_printf(seq, "(F)");
7076 if (rdev->raid_disk < 0)
7077 seq_printf(seq, "(S)"); /* spare */
7078 if (test_bit(Replacement, &rdev->flags))
7079 seq_printf(seq, "(R)");
7080 sectors += rdev->sectors;
7083 if (!list_empty(&mddev->disks)) {
7085 seq_printf(seq, "\n %llu blocks",
7086 (unsigned long long)
7087 mddev->array_sectors / 2);
7089 seq_printf(seq, "\n %llu blocks",
7090 (unsigned long long)sectors / 2);
7092 if (mddev->persistent) {
7093 if (mddev->major_version != 0 ||
7094 mddev->minor_version != 90) {
7095 seq_printf(seq," super %d.%d",
7096 mddev->major_version,
7097 mddev->minor_version);
7099 } else if (mddev->external)
7100 seq_printf(seq, " super external:%s",
7101 mddev->metadata_type);
7103 seq_printf(seq, " super non-persistent");
7106 mddev->pers->status(seq, mddev);
7107 seq_printf(seq, "\n ");
7108 if (mddev->pers->sync_request) {
7109 if (mddev->curr_resync > 2) {
7110 status_resync(seq, mddev);
7111 seq_printf(seq, "\n ");
7112 } else if (mddev->curr_resync >= 1)
7113 seq_printf(seq, "\tresync=DELAYED\n ");
7114 else if (mddev->recovery_cp < MaxSector)
7115 seq_printf(seq, "\tresync=PENDING\n ");
7118 seq_printf(seq, "\n ");
7120 bitmap_status(seq, mddev->bitmap);
7122 seq_printf(seq, "\n");
7124 mddev_unlock(mddev);
7129 static const struct seq_operations md_seq_ops = {
7130 .start = md_seq_start,
7131 .next = md_seq_next,
7132 .stop = md_seq_stop,
7133 .show = md_seq_show,
7136 static int md_seq_open(struct inode *inode, struct file *file)
7138 struct seq_file *seq;
7141 error = seq_open(file, &md_seq_ops);
7145 seq = file->private_data;
7146 seq->poll_event = atomic_read(&md_event_count);
7150 static unsigned int mdstat_poll(struct file *filp, poll_table *wait)
7152 struct seq_file *seq = filp->private_data;
7155 poll_wait(filp, &md_event_waiters, wait);
7157 /* always allow read */
7158 mask = POLLIN | POLLRDNORM;
7160 if (seq->poll_event != atomic_read(&md_event_count))
7161 mask |= POLLERR | POLLPRI;
7165 static const struct file_operations md_seq_fops = {
7166 .owner = THIS_MODULE,
7167 .open = md_seq_open,
7169 .llseek = seq_lseek,
7170 .release = seq_release_private,
7171 .poll = mdstat_poll,
7174 int register_md_personality(struct md_personality *p)
7176 spin_lock(&pers_lock);
7177 list_add_tail(&p->list, &pers_list);
7178 printk(KERN_INFO "md: %s personality registered for level %d\n", p->name, p->level);
7179 spin_unlock(&pers_lock);
7183 int unregister_md_personality(struct md_personality *p)
7185 printk(KERN_INFO "md: %s personality unregistered\n", p->name);
7186 spin_lock(&pers_lock);
7187 list_del_init(&p->list);
7188 spin_unlock(&pers_lock);
7192 static int is_mddev_idle(struct mddev *mddev, int init)
7194 struct md_rdev * rdev;
7200 rdev_for_each_rcu(rdev, mddev) {
7201 struct gendisk *disk = rdev->bdev->bd_contains->bd_disk;
7202 curr_events = (int)part_stat_read(&disk->part0, sectors[0]) +
7203 (int)part_stat_read(&disk->part0, sectors[1]) -
7204 atomic_read(&disk->sync_io);
7205 /* sync IO will cause sync_io to increase before the disk_stats
7206 * as sync_io is counted when a request starts, and
7207 * disk_stats is counted when it completes.
7208 * So resync activity will cause curr_events to be smaller than
7209 * when there was no such activity.
7210 * non-sync IO will cause disk_stat to increase without
7211 * increasing sync_io so curr_events will (eventually)
7212 * be larger than it was before. Once it becomes
7213 * substantially larger, the test below will cause
7214 * the array to appear non-idle, and resync will slow
7216 * If there is a lot of outstanding resync activity when
7217 * we set last_event to curr_events, then all that activity
7218 * completing might cause the array to appear non-idle
7219 * and resync will be slowed down even though there might
7220 * not have been non-resync activity. This will only
7221 * happen once though. 'last_events' will soon reflect
7222 * the state where there is little or no outstanding
7223 * resync requests, and further resync activity will
7224 * always make curr_events less than last_events.
7227 if (init || curr_events - rdev->last_events > 64) {
7228 rdev->last_events = curr_events;
7236 void md_done_sync(struct mddev *mddev, int blocks, int ok)
7238 /* another "blocks" (512byte) blocks have been synced */
7239 atomic_sub(blocks, &mddev->recovery_active);
7240 wake_up(&mddev->recovery_wait);
7242 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
7243 set_bit(MD_RECOVERY_ERROR, &mddev->recovery);
7244 md_wakeup_thread(mddev->thread);
7245 // stop recovery, signal do_sync ....
7250 /* md_write_start(mddev, bi)
7251 * If we need to update some array metadata (e.g. 'active' flag
7252 * in superblock) before writing, schedule a superblock update
7253 * and wait for it to complete.
7255 void md_write_start(struct mddev *mddev, struct bio *bi)
7258 if (bio_data_dir(bi) != WRITE)
7261 BUG_ON(mddev->ro == 1);
7262 if (mddev->ro == 2) {
7263 /* need to switch to read/write */
7265 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7266 md_wakeup_thread(mddev->thread);
7267 md_wakeup_thread(mddev->sync_thread);
7270 atomic_inc(&mddev->writes_pending);
7271 if (mddev->safemode == 1)
7272 mddev->safemode = 0;
7273 if (mddev->in_sync) {
7274 spin_lock_irq(&mddev->write_lock);
7275 if (mddev->in_sync) {
7277 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
7278 set_bit(MD_CHANGE_PENDING, &mddev->flags);
7279 md_wakeup_thread(mddev->thread);
7282 spin_unlock_irq(&mddev->write_lock);
7285 sysfs_notify_dirent_safe(mddev->sysfs_state);
7286 wait_event(mddev->sb_wait,
7287 !test_bit(MD_CHANGE_PENDING, &mddev->flags));
7290 void md_write_end(struct mddev *mddev)
7292 if (atomic_dec_and_test(&mddev->writes_pending)) {
7293 if (mddev->safemode == 2)
7294 md_wakeup_thread(mddev->thread);
7295 else if (mddev->safemode_delay)
7296 mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay);
7300 /* md_allow_write(mddev)
7301 * Calling this ensures that the array is marked 'active' so that writes
7302 * may proceed without blocking. It is important to call this before
7303 * attempting a GFP_KERNEL allocation while holding the mddev lock.
7304 * Must be called with mddev_lock held.
7306 * In the ->external case MD_CHANGE_CLEAN can not be cleared until mddev->lock
7307 * is dropped, so return -EAGAIN after notifying userspace.
7309 int md_allow_write(struct mddev *mddev)
7315 if (!mddev->pers->sync_request)
7318 spin_lock_irq(&mddev->write_lock);
7319 if (mddev->in_sync) {
7321 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
7322 set_bit(MD_CHANGE_PENDING, &mddev->flags);
7323 if (mddev->safemode_delay &&
7324 mddev->safemode == 0)
7325 mddev->safemode = 1;
7326 spin_unlock_irq(&mddev->write_lock);
7327 md_update_sb(mddev, 0);
7328 sysfs_notify_dirent_safe(mddev->sysfs_state);
7330 spin_unlock_irq(&mddev->write_lock);
7332 if (test_bit(MD_CHANGE_PENDING, &mddev->flags))
7337 EXPORT_SYMBOL_GPL(md_allow_write);
7339 #define SYNC_MARKS 10
7340 #define SYNC_MARK_STEP (3*HZ)
7341 #define UPDATE_FREQUENCY (5*60*HZ)
7342 void md_do_sync(struct md_thread *thread)
7344 struct mddev *mddev = thread->mddev;
7345 struct mddev *mddev2;
7346 unsigned int currspeed = 0,
7348 sector_t max_sectors,j, io_sectors;
7349 unsigned long mark[SYNC_MARKS];
7350 unsigned long update_time;
7351 sector_t mark_cnt[SYNC_MARKS];
7353 struct list_head *tmp;
7354 sector_t last_check;
7356 struct md_rdev *rdev;
7357 char *desc, *action = NULL;
7358 struct blk_plug plug;
7360 /* just incase thread restarts... */
7361 if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
7363 if (mddev->ro) /* never try to sync a read-only array */
7366 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
7367 if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)) {
7368 desc = "data-check";
7370 } else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
7371 desc = "requested-resync";
7375 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
7380 mddev->last_sync_action = action ?: desc;
7382 /* we overload curr_resync somewhat here.
7383 * 0 == not engaged in resync at all
7384 * 2 == checking that there is no conflict with another sync
7385 * 1 == like 2, but have yielded to allow conflicting resync to
7387 * other == active in resync - this many blocks
7389 * Before starting a resync we must have set curr_resync to
7390 * 2, and then checked that every "conflicting" array has curr_resync
7391 * less than ours. When we find one that is the same or higher
7392 * we wait on resync_wait. To avoid deadlock, we reduce curr_resync
7393 * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
7394 * This will mean we have to start checking from the beginning again.
7399 mddev->curr_resync = 2;
7402 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
7404 for_each_mddev(mddev2, tmp) {
7405 if (mddev2 == mddev)
7407 if (!mddev->parallel_resync
7408 && mddev2->curr_resync
7409 && match_mddev_units(mddev, mddev2)) {
7411 if (mddev < mddev2 && mddev->curr_resync == 2) {
7412 /* arbitrarily yield */
7413 mddev->curr_resync = 1;
7414 wake_up(&resync_wait);
7416 if (mddev > mddev2 && mddev->curr_resync == 1)
7417 /* no need to wait here, we can wait the next
7418 * time 'round when curr_resync == 2
7421 /* We need to wait 'interruptible' so as not to
7422 * contribute to the load average, and not to
7423 * be caught by 'softlockup'
7425 prepare_to_wait(&resync_wait, &wq, TASK_INTERRUPTIBLE);
7426 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
7427 mddev2->curr_resync >= mddev->curr_resync) {
7428 printk(KERN_INFO "md: delaying %s of %s"
7429 " until %s has finished (they"
7430 " share one or more physical units)\n",
7431 desc, mdname(mddev), mdname(mddev2));
7433 if (signal_pending(current))
7434 flush_signals(current);
7436 finish_wait(&resync_wait, &wq);
7439 finish_wait(&resync_wait, &wq);
7442 } while (mddev->curr_resync < 2);
7445 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
7446 /* resync follows the size requested by the personality,
7447 * which defaults to physical size, but can be virtual size
7449 max_sectors = mddev->resync_max_sectors;
7450 atomic64_set(&mddev->resync_mismatches, 0);
7451 /* we don't use the checkpoint if there's a bitmap */
7452 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
7453 j = mddev->resync_min;
7454 else if (!mddev->bitmap)
7455 j = mddev->recovery_cp;
7457 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
7458 max_sectors = mddev->resync_max_sectors;
7460 /* recovery follows the physical size of devices */
7461 max_sectors = mddev->dev_sectors;
7464 rdev_for_each_rcu(rdev, mddev)
7465 if (rdev->raid_disk >= 0 &&
7466 !test_bit(Faulty, &rdev->flags) &&
7467 !test_bit(In_sync, &rdev->flags) &&
7468 rdev->recovery_offset < j)
7469 j = rdev->recovery_offset;
7473 printk(KERN_INFO "md: %s of RAID array %s\n", desc, mdname(mddev));
7474 printk(KERN_INFO "md: minimum _guaranteed_ speed:"
7475 " %d KB/sec/disk.\n", speed_min(mddev));
7476 printk(KERN_INFO "md: using maximum available idle IO bandwidth "
7477 "(but not more than %d KB/sec) for %s.\n",
7478 speed_max(mddev), desc);
7480 is_mddev_idle(mddev, 1); /* this initializes IO event counters */
7483 for (m = 0; m < SYNC_MARKS; m++) {
7485 mark_cnt[m] = io_sectors;
7488 mddev->resync_mark = mark[last_mark];
7489 mddev->resync_mark_cnt = mark_cnt[last_mark];
7492 * Tune reconstruction:
7494 window = 32*(PAGE_SIZE/512);
7495 printk(KERN_INFO "md: using %dk window, over a total of %lluk.\n",
7496 window/2, (unsigned long long)max_sectors/2);
7498 atomic_set(&mddev->recovery_active, 0);
7503 "md: resuming %s of %s from checkpoint.\n",
7504 desc, mdname(mddev));
7505 mddev->curr_resync = j;
7507 mddev->curr_resync = 3; /* no longer delayed */
7508 mddev->curr_resync_completed = j;
7509 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
7510 md_new_event(mddev);
7511 update_time = jiffies;
7513 blk_start_plug(&plug);
7514 while (j < max_sectors) {
7519 if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
7520 ((mddev->curr_resync > mddev->curr_resync_completed &&
7521 (mddev->curr_resync - mddev->curr_resync_completed)
7522 > (max_sectors >> 4)) ||
7523 time_after_eq(jiffies, update_time + UPDATE_FREQUENCY) ||
7524 (j - mddev->curr_resync_completed)*2
7525 >= mddev->resync_max - mddev->curr_resync_completed
7527 /* time to update curr_resync_completed */
7528 wait_event(mddev->recovery_wait,
7529 atomic_read(&mddev->recovery_active) == 0);
7530 mddev->curr_resync_completed = j;
7531 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) &&
7532 j > mddev->recovery_cp)
7533 mddev->recovery_cp = j;
7534 update_time = jiffies;
7535 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
7536 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
7539 while (j >= mddev->resync_max &&
7540 !test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
7541 /* As this condition is controlled by user-space,
7542 * we can block indefinitely, so use '_interruptible'
7543 * to avoid triggering warnings.
7545 flush_signals(current); /* just in case */
7546 wait_event_interruptible(mddev->recovery_wait,
7547 mddev->resync_max > j
7548 || test_bit(MD_RECOVERY_INTR,
7552 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
7555 sectors = mddev->pers->sync_request(mddev, j, &skipped,
7556 currspeed < speed_min(mddev));
7558 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
7562 if (!skipped) { /* actual IO requested */
7563 io_sectors += sectors;
7564 atomic_add(sectors, &mddev->recovery_active);
7567 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
7572 mddev->curr_resync = j;
7573 mddev->curr_mark_cnt = io_sectors;
7574 if (last_check == 0)
7575 /* this is the earliest that rebuild will be
7576 * visible in /proc/mdstat
7578 md_new_event(mddev);
7580 if (last_check + window > io_sectors || j == max_sectors)
7583 last_check = io_sectors;
7585 if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) {
7587 int next = (last_mark+1) % SYNC_MARKS;
7589 mddev->resync_mark = mark[next];
7590 mddev->resync_mark_cnt = mark_cnt[next];
7591 mark[next] = jiffies;
7592 mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active);
7596 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
7600 * this loop exits only if either when we are slower than
7601 * the 'hard' speed limit, or the system was IO-idle for
7603 * the system might be non-idle CPU-wise, but we only care
7604 * about not overloading the IO subsystem. (things like an
7605 * e2fsck being done on the RAID array should execute fast)
7609 currspeed = ((unsigned long)(io_sectors-mddev->resync_mark_cnt))/2
7610 /((jiffies-mddev->resync_mark)/HZ +1) +1;
7612 if (currspeed > speed_min(mddev)) {
7613 if ((currspeed > speed_max(mddev)) ||
7614 !is_mddev_idle(mddev, 0)) {
7620 printk(KERN_INFO "md: %s: %s %s.\n",mdname(mddev), desc,
7621 test_bit(MD_RECOVERY_INTR, &mddev->recovery)
7622 ? "interrupted" : "done");
7624 * this also signals 'finished resyncing' to md_stop
7626 blk_finish_plug(&plug);
7627 wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
7629 /* tell personality that we are finished */
7630 mddev->pers->sync_request(mddev, max_sectors, &skipped, 1);
7632 if (!test_bit(MD_RECOVERY_CHECK, &mddev->recovery) &&
7633 mddev->curr_resync > 2) {
7634 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
7635 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
7636 if (mddev->curr_resync >= mddev->recovery_cp) {
7638 "md: checkpointing %s of %s.\n",
7639 desc, mdname(mddev));
7640 if (test_bit(MD_RECOVERY_ERROR,
7642 mddev->recovery_cp =
7643 mddev->curr_resync_completed;
7645 mddev->recovery_cp =
7649 mddev->recovery_cp = MaxSector;
7651 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery))
7652 mddev->curr_resync = MaxSector;
7654 rdev_for_each_rcu(rdev, mddev)
7655 if (rdev->raid_disk >= 0 &&
7656 mddev->delta_disks >= 0 &&
7657 !test_bit(Faulty, &rdev->flags) &&
7658 !test_bit(In_sync, &rdev->flags) &&
7659 rdev->recovery_offset < mddev->curr_resync)
7660 rdev->recovery_offset = mddev->curr_resync;
7665 set_bit(MD_CHANGE_DEVS, &mddev->flags);
7667 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
7668 /* We completed so min/max setting can be forgotten if used. */
7669 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
7670 mddev->resync_min = 0;
7671 mddev->resync_max = MaxSector;
7672 } else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
7673 mddev->resync_min = mddev->curr_resync_completed;
7674 mddev->curr_resync = 0;
7675 wake_up(&resync_wait);
7676 set_bit(MD_RECOVERY_DONE, &mddev->recovery);
7677 md_wakeup_thread(mddev->thread);
7680 EXPORT_SYMBOL_GPL(md_do_sync);
7682 static int remove_and_add_spares(struct mddev *mddev,
7683 struct md_rdev *this)
7685 struct md_rdev *rdev;
7689 rdev_for_each(rdev, mddev)
7690 if ((this == NULL || rdev == this) &&
7691 rdev->raid_disk >= 0 &&
7692 !test_bit(Blocked, &rdev->flags) &&
7693 (test_bit(Faulty, &rdev->flags) ||
7694 ! test_bit(In_sync, &rdev->flags)) &&
7695 atomic_read(&rdev->nr_pending)==0) {
7696 if (mddev->pers->hot_remove_disk(
7697 mddev, rdev) == 0) {
7698 sysfs_unlink_rdev(mddev, rdev);
7699 rdev->raid_disk = -1;
7703 if (removed && mddev->kobj.sd)
7704 sysfs_notify(&mddev->kobj, NULL, "degraded");
7709 rdev_for_each(rdev, mddev) {
7710 if (rdev->raid_disk >= 0 &&
7711 !test_bit(In_sync, &rdev->flags) &&
7712 !test_bit(Faulty, &rdev->flags))
7714 if (rdev->raid_disk >= 0)
7716 if (test_bit(Faulty, &rdev->flags))
7719 rdev->saved_raid_disk < 0)
7722 rdev->recovery_offset = 0;
7724 hot_add_disk(mddev, rdev) == 0) {
7725 if (sysfs_link_rdev(mddev, rdev))
7726 /* failure here is OK */;
7728 md_new_event(mddev);
7729 set_bit(MD_CHANGE_DEVS, &mddev->flags);
7734 set_bit(MD_CHANGE_DEVS, &mddev->flags);
7739 * This routine is regularly called by all per-raid-array threads to
7740 * deal with generic issues like resync and super-block update.
7741 * Raid personalities that don't have a thread (linear/raid0) do not
7742 * need this as they never do any recovery or update the superblock.
7744 * It does not do any resync itself, but rather "forks" off other threads
7745 * to do that as needed.
7746 * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
7747 * "->recovery" and create a thread at ->sync_thread.
7748 * When the thread finishes it sets MD_RECOVERY_DONE
7749 * and wakeups up this thread which will reap the thread and finish up.
7750 * This thread also removes any faulty devices (with nr_pending == 0).
7752 * The overall approach is:
7753 * 1/ if the superblock needs updating, update it.
7754 * 2/ If a recovery thread is running, don't do anything else.
7755 * 3/ If recovery has finished, clean up, possibly marking spares active.
7756 * 4/ If there are any faulty devices, remove them.
7757 * 5/ If array is degraded, try to add spares devices
7758 * 6/ If array has spares or is not in-sync, start a resync thread.
7760 void md_check_recovery(struct mddev *mddev)
7762 if (mddev->suspended)
7766 bitmap_daemon_work(mddev);
7768 if (signal_pending(current)) {
7769 if (mddev->pers->sync_request && !mddev->external) {
7770 printk(KERN_INFO "md: %s in immediate safe mode\n",
7772 mddev->safemode = 2;
7774 flush_signals(current);
7777 if (mddev->ro && !test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
7780 (mddev->flags & MD_UPDATE_SB_FLAGS & ~ (1<<MD_CHANGE_PENDING)) ||
7781 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
7782 test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
7783 (mddev->external == 0 && mddev->safemode == 1) ||
7784 (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending)
7785 && !mddev->in_sync && mddev->recovery_cp == MaxSector)
7789 if (mddev_trylock(mddev)) {
7793 /* On a read-only array we can:
7794 * - remove failed devices
7795 * - add already-in_sync devices if the array itself
7797 * As we only add devices that are already in-sync,
7798 * we can activate the spares immediately.
7800 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7801 remove_and_add_spares(mddev, NULL);
7802 mddev->pers->spare_active(mddev);
7806 if (!mddev->external) {
7808 spin_lock_irq(&mddev->write_lock);
7809 if (mddev->safemode &&
7810 !atomic_read(&mddev->writes_pending) &&
7812 mddev->recovery_cp == MaxSector) {
7815 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
7817 if (mddev->safemode == 1)
7818 mddev->safemode = 0;
7819 spin_unlock_irq(&mddev->write_lock);
7821 sysfs_notify_dirent_safe(mddev->sysfs_state);
7824 if (mddev->flags & MD_UPDATE_SB_FLAGS)
7825 md_update_sb(mddev, 0);
7827 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
7828 !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
7829 /* resync/recovery still happening */
7830 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7833 if (mddev->sync_thread) {
7834 md_reap_sync_thread(mddev);
7837 /* Set RUNNING before clearing NEEDED to avoid
7838 * any transients in the value of "sync_action".
7840 mddev->curr_resync_completed = 0;
7841 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
7842 /* Clear some bits that don't mean anything, but
7845 clear_bit(MD_RECOVERY_INTR, &mddev->recovery);
7846 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
7848 if (!test_and_clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
7849 test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
7851 /* no recovery is running.
7852 * remove any failed drives, then
7853 * add spares if possible.
7854 * Spares are also removed and re-added, to allow
7855 * the personality to fail the re-add.
7858 if (mddev->reshape_position != MaxSector) {
7859 if (mddev->pers->check_reshape == NULL ||
7860 mddev->pers->check_reshape(mddev) != 0)
7861 /* Cannot proceed */
7863 set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
7864 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
7865 } else if ((spares = remove_and_add_spares(mddev, NULL))) {
7866 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
7867 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
7868 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
7869 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
7870 } else if (mddev->recovery_cp < MaxSector) {
7871 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
7872 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
7873 } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
7874 /* nothing to be done ... */
7877 if (mddev->pers->sync_request) {
7879 /* We are adding a device or devices to an array
7880 * which has the bitmap stored on all devices.
7881 * So make sure all bitmap pages get written
7883 bitmap_write_all(mddev->bitmap);
7885 mddev->sync_thread = md_register_thread(md_do_sync,
7888 if (!mddev->sync_thread) {
7889 printk(KERN_ERR "%s: could not start resync"
7892 /* leave the spares where they are, it shouldn't hurt */
7893 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
7894 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
7895 clear_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
7896 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
7897 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
7899 md_wakeup_thread(mddev->sync_thread);
7900 sysfs_notify_dirent_safe(mddev->sysfs_action);
7901 md_new_event(mddev);
7904 wake_up(&mddev->sb_wait);
7906 if (!mddev->sync_thread) {
7907 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
7908 if (test_and_clear_bit(MD_RECOVERY_RECOVER,
7910 if (mddev->sysfs_action)
7911 sysfs_notify_dirent_safe(mddev->sysfs_action);
7913 mddev_unlock(mddev);
7917 void md_reap_sync_thread(struct mddev *mddev)
7919 struct md_rdev *rdev;
7921 /* resync has finished, collect result */
7922 md_unregister_thread(&mddev->sync_thread);
7923 wake_up(&resync_wait);
7924 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
7925 !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
7927 /* activate any spares */
7928 if (mddev->pers->spare_active(mddev)) {
7929 sysfs_notify(&mddev->kobj, NULL,
7931 set_bit(MD_CHANGE_DEVS, &mddev->flags);
7934 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
7935 mddev->pers->finish_reshape)
7936 mddev->pers->finish_reshape(mddev);
7938 /* If array is no-longer degraded, then any saved_raid_disk
7939 * information must be scrapped. Also if any device is now
7940 * In_sync we must scrape the saved_raid_disk for that device
7941 * do the superblock for an incrementally recovered device
7944 rdev_for_each(rdev, mddev)
7945 if (!mddev->degraded ||
7946 test_bit(In_sync, &rdev->flags))
7947 rdev->saved_raid_disk = -1;
7949 md_update_sb(mddev, 1);
7950 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
7951 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
7952 clear_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
7953 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
7954 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
7955 /* flag recovery needed just to double check */
7956 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7957 sysfs_notify_dirent_safe(mddev->sysfs_action);
7958 md_new_event(mddev);
7959 if (mddev->event_work.func)
7960 queue_work(md_misc_wq, &mddev->event_work);
7963 void md_wait_for_blocked_rdev(struct md_rdev *rdev, struct mddev *mddev)
7965 sysfs_notify_dirent_safe(rdev->sysfs_state);
7966 wait_event_timeout(rdev->blocked_wait,
7967 !test_bit(Blocked, &rdev->flags) &&
7968 !test_bit(BlockedBadBlocks, &rdev->flags),
7969 msecs_to_jiffies(5000));
7970 rdev_dec_pending(rdev, mddev);
7972 EXPORT_SYMBOL(md_wait_for_blocked_rdev);
7974 void md_finish_reshape(struct mddev *mddev)
7976 /* called be personality module when reshape completes. */
7977 struct md_rdev *rdev;
7979 rdev_for_each(rdev, mddev) {
7980 if (rdev->data_offset > rdev->new_data_offset)
7981 rdev->sectors += rdev->data_offset - rdev->new_data_offset;
7983 rdev->sectors -= rdev->new_data_offset - rdev->data_offset;
7984 rdev->data_offset = rdev->new_data_offset;
7987 EXPORT_SYMBOL(md_finish_reshape);
7989 /* Bad block management.
7990 * We can record which blocks on each device are 'bad' and so just
7991 * fail those blocks, or that stripe, rather than the whole device.
7992 * Entries in the bad-block table are 64bits wide. This comprises:
7993 * Length of bad-range, in sectors: 0-511 for lengths 1-512
7994 * Start of bad-range, sector offset, 54 bits (allows 8 exbibytes)
7995 * A 'shift' can be set so that larger blocks are tracked and
7996 * consequently larger devices can be covered.
7997 * 'Acknowledged' flag - 1 bit. - the most significant bit.
7999 * Locking of the bad-block table uses a seqlock so md_is_badblock
8000 * might need to retry if it is very unlucky.
8001 * We will sometimes want to check for bad blocks in a bi_end_io function,
8002 * so we use the write_seqlock_irq variant.
8004 * When looking for a bad block we specify a range and want to
8005 * know if any block in the range is bad. So we binary-search
8006 * to the last range that starts at-or-before the given endpoint,
8007 * (or "before the sector after the target range")
8008 * then see if it ends after the given start.
8010 * 0 if there are no known bad blocks in the range
8011 * 1 if there are known bad block which are all acknowledged
8012 * -1 if there are bad blocks which have not yet been acknowledged in metadata.
8013 * plus the start/length of the first bad section we overlap.
8015 int md_is_badblock(struct badblocks *bb, sector_t s, int sectors,
8016 sector_t *first_bad, int *bad_sectors)
8022 sector_t target = s + sectors;
8025 if (bb->shift > 0) {
8026 /* round the start down, and the end up */
8028 target += (1<<bb->shift) - 1;
8029 target >>= bb->shift;
8030 sectors = target - s;
8032 /* 'target' is now the first block after the bad range */
8035 seq = read_seqbegin(&bb->lock);
8040 /* Binary search between lo and hi for 'target'
8041 * i.e. for the last range that starts before 'target'
8043 /* INVARIANT: ranges before 'lo' and at-or-after 'hi'
8044 * are known not to be the last range before target.
8045 * VARIANT: hi-lo is the number of possible
8046 * ranges, and decreases until it reaches 1
8048 while (hi - lo > 1) {
8049 int mid = (lo + hi) / 2;
8050 sector_t a = BB_OFFSET(p[mid]);
8052 /* This could still be the one, earlier ranges
8056 /* This and later ranges are definitely out. */
8059 /* 'lo' might be the last that started before target, but 'hi' isn't */
8061 /* need to check all range that end after 's' to see if
8062 * any are unacknowledged.
8065 BB_OFFSET(p[lo]) + BB_LEN(p[lo]) > s) {
8066 if (BB_OFFSET(p[lo]) < target) {
8067 /* starts before the end, and finishes after
8068 * the start, so they must overlap
8070 if (rv != -1 && BB_ACK(p[lo]))
8074 *first_bad = BB_OFFSET(p[lo]);
8075 *bad_sectors = BB_LEN(p[lo]);
8081 if (read_seqretry(&bb->lock, seq))
8086 EXPORT_SYMBOL_GPL(md_is_badblock);
8089 * Add a range of bad blocks to the table.
8090 * This might extend the table, or might contract it
8091 * if two adjacent ranges can be merged.
8092 * We binary-search to find the 'insertion' point, then
8093 * decide how best to handle it.
8095 static int md_set_badblocks(struct badblocks *bb, sector_t s, int sectors,
8101 unsigned long flags;
8104 /* badblocks are disabled */
8108 /* round the start down, and the end up */
8109 sector_t next = s + sectors;
8111 next += (1<<bb->shift) - 1;
8116 write_seqlock_irqsave(&bb->lock, flags);
8121 /* Find the last range that starts at-or-before 's' */
8122 while (hi - lo > 1) {
8123 int mid = (lo + hi) / 2;
8124 sector_t a = BB_OFFSET(p[mid]);
8130 if (hi > lo && BB_OFFSET(p[lo]) > s)
8134 /* we found a range that might merge with the start
8137 sector_t a = BB_OFFSET(p[lo]);
8138 sector_t e = a + BB_LEN(p[lo]);
8139 int ack = BB_ACK(p[lo]);
8141 /* Yes, we can merge with a previous range */
8142 if (s == a && s + sectors >= e)
8143 /* new range covers old */
8146 ack = ack && acknowledged;
8148 if (e < s + sectors)
8150 if (e - a <= BB_MAX_LEN) {
8151 p[lo] = BB_MAKE(a, e-a, ack);
8154 /* does not all fit in one range,
8155 * make p[lo] maximal
8157 if (BB_LEN(p[lo]) != BB_MAX_LEN)
8158 p[lo] = BB_MAKE(a, BB_MAX_LEN, ack);
8164 if (sectors && hi < bb->count) {
8165 /* 'hi' points to the first range that starts after 's'.
8166 * Maybe we can merge with the start of that range */
8167 sector_t a = BB_OFFSET(p[hi]);
8168 sector_t e = a + BB_LEN(p[hi]);
8169 int ack = BB_ACK(p[hi]);
8170 if (a <= s + sectors) {
8171 /* merging is possible */
8172 if (e <= s + sectors) {
8177 ack = ack && acknowledged;
8180 if (e - a <= BB_MAX_LEN) {
8181 p[hi] = BB_MAKE(a, e-a, ack);
8184 p[hi] = BB_MAKE(a, BB_MAX_LEN, ack);
8192 if (sectors == 0 && hi < bb->count) {
8193 /* we might be able to combine lo and hi */
8194 /* Note: 's' is at the end of 'lo' */
8195 sector_t a = BB_OFFSET(p[hi]);
8196 int lolen = BB_LEN(p[lo]);
8197 int hilen = BB_LEN(p[hi]);
8198 int newlen = lolen + hilen - (s - a);
8199 if (s >= a && newlen < BB_MAX_LEN) {
8200 /* yes, we can combine them */
8201 int ack = BB_ACK(p[lo]) && BB_ACK(p[hi]);
8202 p[lo] = BB_MAKE(BB_OFFSET(p[lo]), newlen, ack);
8203 memmove(p + hi, p + hi + 1,
8204 (bb->count - hi - 1) * 8);
8209 /* didn't merge (it all).
8210 * Need to add a range just before 'hi' */
8211 if (bb->count >= MD_MAX_BADBLOCKS) {
8212 /* No room for more */
8216 int this_sectors = sectors;
8217 memmove(p + hi + 1, p + hi,
8218 (bb->count - hi) * 8);
8221 if (this_sectors > BB_MAX_LEN)
8222 this_sectors = BB_MAX_LEN;
8223 p[hi] = BB_MAKE(s, this_sectors, acknowledged);
8224 sectors -= this_sectors;
8231 bb->unacked_exist = 1;
8232 write_sequnlock_irqrestore(&bb->lock, flags);
8237 int rdev_set_badblocks(struct md_rdev *rdev, sector_t s, int sectors,
8242 s += rdev->new_data_offset;
8244 s += rdev->data_offset;
8245 rv = md_set_badblocks(&rdev->badblocks,
8248 /* Make sure they get written out promptly */
8249 sysfs_notify_dirent_safe(rdev->sysfs_state);
8250 set_bit(MD_CHANGE_CLEAN, &rdev->mddev->flags);
8251 md_wakeup_thread(rdev->mddev->thread);
8255 EXPORT_SYMBOL_GPL(rdev_set_badblocks);
8258 * Remove a range of bad blocks from the table.
8259 * This may involve extending the table if we spilt a region,
8260 * but it must not fail. So if the table becomes full, we just
8261 * drop the remove request.
8263 static int md_clear_badblocks(struct badblocks *bb, sector_t s, int sectors)
8267 sector_t target = s + sectors;
8270 if (bb->shift > 0) {
8271 /* When clearing we round the start up and the end down.
8272 * This should not matter as the shift should align with
8273 * the block size and no rounding should ever be needed.
8274 * However it is better the think a block is bad when it
8275 * isn't than to think a block is not bad when it is.
8277 s += (1<<bb->shift) - 1;
8279 target >>= bb->shift;
8280 sectors = target - s;
8283 write_seqlock_irq(&bb->lock);
8288 /* Find the last range that starts before 'target' */
8289 while (hi - lo > 1) {
8290 int mid = (lo + hi) / 2;
8291 sector_t a = BB_OFFSET(p[mid]);
8298 /* p[lo] is the last range that could overlap the
8299 * current range. Earlier ranges could also overlap,
8300 * but only this one can overlap the end of the range.
8302 if (BB_OFFSET(p[lo]) + BB_LEN(p[lo]) > target) {
8303 /* Partial overlap, leave the tail of this range */
8304 int ack = BB_ACK(p[lo]);
8305 sector_t a = BB_OFFSET(p[lo]);
8306 sector_t end = a + BB_LEN(p[lo]);
8309 /* we need to split this range */
8310 if (bb->count >= MD_MAX_BADBLOCKS) {
8314 memmove(p+lo+1, p+lo, (bb->count - lo) * 8);
8316 p[lo] = BB_MAKE(a, s-a, ack);
8319 p[lo] = BB_MAKE(target, end - target, ack);
8320 /* there is no longer an overlap */
8325 BB_OFFSET(p[lo]) + BB_LEN(p[lo]) > s) {
8326 /* This range does overlap */
8327 if (BB_OFFSET(p[lo]) < s) {
8328 /* Keep the early parts of this range. */
8329 int ack = BB_ACK(p[lo]);
8330 sector_t start = BB_OFFSET(p[lo]);
8331 p[lo] = BB_MAKE(start, s - start, ack);
8332 /* now low doesn't overlap, so.. */
8337 /* 'lo' is strictly before, 'hi' is strictly after,
8338 * anything between needs to be discarded
8341 memmove(p+lo+1, p+hi, (bb->count - hi) * 8);
8342 bb->count -= (hi - lo - 1);
8348 write_sequnlock_irq(&bb->lock);
8352 int rdev_clear_badblocks(struct md_rdev *rdev, sector_t s, int sectors,
8356 s += rdev->new_data_offset;
8358 s += rdev->data_offset;
8359 return md_clear_badblocks(&rdev->badblocks,
8362 EXPORT_SYMBOL_GPL(rdev_clear_badblocks);
8365 * Acknowledge all bad blocks in a list.
8366 * This only succeeds if ->changed is clear. It is used by
8367 * in-kernel metadata updates
8369 void md_ack_all_badblocks(struct badblocks *bb)
8371 if (bb->page == NULL || bb->changed)
8372 /* no point even trying */
8374 write_seqlock_irq(&bb->lock);
8376 if (bb->changed == 0 && bb->unacked_exist) {
8379 for (i = 0; i < bb->count ; i++) {
8380 if (!BB_ACK(p[i])) {
8381 sector_t start = BB_OFFSET(p[i]);
8382 int len = BB_LEN(p[i]);
8383 p[i] = BB_MAKE(start, len, 1);
8386 bb->unacked_exist = 0;
8388 write_sequnlock_irq(&bb->lock);
8390 EXPORT_SYMBOL_GPL(md_ack_all_badblocks);
8392 /* sysfs access to bad-blocks list.
8393 * We present two files.
8394 * 'bad-blocks' lists sector numbers and lengths of ranges that
8395 * are recorded as bad. The list is truncated to fit within
8396 * the one-page limit of sysfs.
8397 * Writing "sector length" to this file adds an acknowledged
8399 * 'unacknowledged-bad-blocks' lists bad blocks that have not yet
8400 * been acknowledged. Writing to this file adds bad blocks
8401 * without acknowledging them. This is largely for testing.
8405 badblocks_show(struct badblocks *bb, char *page, int unack)
8416 seq = read_seqbegin(&bb->lock);
8421 while (len < PAGE_SIZE && i < bb->count) {
8422 sector_t s = BB_OFFSET(p[i]);
8423 unsigned int length = BB_LEN(p[i]);
8424 int ack = BB_ACK(p[i]);
8430 len += snprintf(page+len, PAGE_SIZE-len, "%llu %u\n",
8431 (unsigned long long)s << bb->shift,
8432 length << bb->shift);
8434 if (unack && len == 0)
8435 bb->unacked_exist = 0;
8437 if (read_seqretry(&bb->lock, seq))
8446 badblocks_store(struct badblocks *bb, const char *page, size_t len, int unack)
8448 unsigned long long sector;
8452 /* Allow clearing via sysfs *only* for testing/debugging.
8453 * Normally only a successful write may clear a badblock
8456 if (page[0] == '-') {
8460 #endif /* DO_DEBUG */
8462 switch (sscanf(page, "%llu %d%c", §or, &length, &newline)) {
8464 if (newline != '\n')
8476 md_clear_badblocks(bb, sector, length);
8479 #endif /* DO_DEBUG */
8480 if (md_set_badblocks(bb, sector, length, !unack))
8486 static int md_notify_reboot(struct notifier_block *this,
8487 unsigned long code, void *x)
8489 struct list_head *tmp;
8490 struct mddev *mddev;
8493 for_each_mddev(mddev, tmp) {
8494 if (mddev_trylock(mddev)) {
8496 __md_stop_writes(mddev);
8497 mddev->safemode = 2;
8498 mddev_unlock(mddev);
8503 * certain more exotic SCSI devices are known to be
8504 * volatile wrt too early system reboots. While the
8505 * right place to handle this issue is the given
8506 * driver, we do want to have a safe RAID driver ...
8514 static struct notifier_block md_notifier = {
8515 .notifier_call = md_notify_reboot,
8517 .priority = INT_MAX, /* before any real devices */
8520 static void md_geninit(void)
8522 pr_debug("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
8524 proc_create("mdstat", S_IRUGO, NULL, &md_seq_fops);
8527 static int __init md_init(void)
8531 md_wq = alloc_workqueue("md", WQ_MEM_RECLAIM, 0);
8535 md_misc_wq = alloc_workqueue("md_misc", 0, 0);
8539 if ((ret = register_blkdev(MD_MAJOR, "md")) < 0)
8542 if ((ret = register_blkdev(0, "mdp")) < 0)
8546 blk_register_region(MKDEV(MD_MAJOR, 0), 1UL<<MINORBITS, THIS_MODULE,
8547 md_probe, NULL, NULL);
8548 blk_register_region(MKDEV(mdp_major, 0), 1UL<<MINORBITS, THIS_MODULE,
8549 md_probe, NULL, NULL);
8551 register_reboot_notifier(&md_notifier);
8552 raid_table_header = register_sysctl_table(raid_root_table);
8558 unregister_blkdev(MD_MAJOR, "md");
8560 destroy_workqueue(md_misc_wq);
8562 destroy_workqueue(md_wq);
8570 * Searches all registered partitions for autorun RAID arrays
8574 static LIST_HEAD(all_detected_devices);
8575 struct detected_devices_node {
8576 struct list_head list;
8580 void md_autodetect_dev(dev_t dev)
8582 struct detected_devices_node *node_detected_dev;
8584 node_detected_dev = kzalloc(sizeof(*node_detected_dev), GFP_KERNEL);
8585 if (node_detected_dev) {
8586 node_detected_dev->dev = dev;
8587 list_add_tail(&node_detected_dev->list, &all_detected_devices);
8589 printk(KERN_CRIT "md: md_autodetect_dev: kzalloc failed"
8590 ", skipping dev(%d,%d)\n", MAJOR(dev), MINOR(dev));
8595 static void autostart_arrays(int part)
8597 struct md_rdev *rdev;
8598 struct detected_devices_node *node_detected_dev;
8600 int i_scanned, i_passed;
8605 printk(KERN_INFO "md: Autodetecting RAID arrays.\n");
8607 while (!list_empty(&all_detected_devices) && i_scanned < INT_MAX) {
8609 node_detected_dev = list_entry(all_detected_devices.next,
8610 struct detected_devices_node, list);
8611 list_del(&node_detected_dev->list);
8612 dev = node_detected_dev->dev;
8613 kfree(node_detected_dev);
8614 rdev = md_import_device(dev,0, 90);
8618 if (test_bit(Faulty, &rdev->flags)) {
8622 set_bit(AutoDetected, &rdev->flags);
8623 list_add(&rdev->same_set, &pending_raid_disks);
8627 printk(KERN_INFO "md: Scanned %d and added %d devices.\n",
8628 i_scanned, i_passed);
8630 autorun_devices(part);
8633 #endif /* !MODULE */
8635 static __exit void md_exit(void)
8637 struct mddev *mddev;
8638 struct list_head *tmp;
8640 blk_unregister_region(MKDEV(MD_MAJOR,0), 1U << MINORBITS);
8641 blk_unregister_region(MKDEV(mdp_major,0), 1U << MINORBITS);
8643 unregister_blkdev(MD_MAJOR,"md");
8644 unregister_blkdev(mdp_major, "mdp");
8645 unregister_reboot_notifier(&md_notifier);
8646 unregister_sysctl_table(raid_table_header);
8647 remove_proc_entry("mdstat", NULL);
8648 for_each_mddev(mddev, tmp) {
8649 export_array(mddev);
8650 mddev->hold_active = 0;
8652 destroy_workqueue(md_misc_wq);
8653 destroy_workqueue(md_wq);
8656 subsys_initcall(md_init);
8657 module_exit(md_exit)
8659 static int get_ro(char *buffer, struct kernel_param *kp)
8661 return sprintf(buffer, "%d", start_readonly);
8663 static int set_ro(const char *val, struct kernel_param *kp)
8666 int num = simple_strtoul(val, &e, 10);
8667 if (*val && (*e == '\0' || *e == '\n')) {
8668 start_readonly = num;
8674 module_param_call(start_ro, set_ro, get_ro, NULL, S_IRUSR|S_IWUSR);
8675 module_param(start_dirty_degraded, int, S_IRUGO|S_IWUSR);
8677 module_param_call(new_array, add_named_array, NULL, NULL, S_IWUSR);
8679 EXPORT_SYMBOL(register_md_personality);
8680 EXPORT_SYMBOL(unregister_md_personality);
8681 EXPORT_SYMBOL(md_error);
8682 EXPORT_SYMBOL(md_done_sync);
8683 EXPORT_SYMBOL(md_write_start);
8684 EXPORT_SYMBOL(md_write_end);
8685 EXPORT_SYMBOL(md_register_thread);
8686 EXPORT_SYMBOL(md_unregister_thread);
8687 EXPORT_SYMBOL(md_wakeup_thread);
8688 EXPORT_SYMBOL(md_check_recovery);
8689 EXPORT_SYMBOL(md_reap_sync_thread);
8690 MODULE_LICENSE("GPL");
8691 MODULE_DESCRIPTION("MD RAID framework");
8693 MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR);