2 md.c : Multiple Devices driver for Linux
3 Copyright (C) 1998, 1999, 2000 Ingo Molnar
5 completely rewritten, based on the MD driver code from Marc Zyngier
9 - RAID-1/RAID-5 extensions by Miguel de Icaza, Gadi Oxman, Ingo Molnar
10 - RAID-6 extensions by H. Peter Anvin <hpa@zytor.com>
11 - boot support for linear and striped mode by Harald Hoyer <HarryH@Royal.Net>
12 - kerneld support by Boris Tobotras <boris@xtalk.msk.su>
13 - kmod support by: Cyrus Durgin
14 - RAID0 bugfixes: Mark Anthony Lisher <markal@iname.com>
15 - Devfs support by Richard Gooch <rgooch@atnf.csiro.au>
17 - lots of fixes and improvements to the RAID1/RAID5 and generic
18 RAID code (such as request based resynchronization):
20 Neil Brown <neilb@cse.unsw.edu.au>.
22 - persistent bitmap code
23 Copyright (C) 2003-2004, Paul Clements, SteelEye Technology, Inc.
25 This program is free software; you can redistribute it and/or modify
26 it under the terms of the GNU General Public License as published by
27 the Free Software Foundation; either version 2, or (at your option)
30 You should have received a copy of the GNU General Public License
31 (for example /usr/src/linux/COPYING); if not, write to the Free
32 Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
35 #include <linux/kthread.h>
36 #include <linux/blkdev.h>
37 #include <linux/sysctl.h>
38 #include <linux/seq_file.h>
39 #include <linux/mutex.h>
40 #include <linux/buffer_head.h> /* for invalidate_bdev */
41 #include <linux/poll.h>
42 #include <linux/ctype.h>
43 #include <linux/string.h>
44 #include <linux/hdreg.h>
45 #include <linux/proc_fs.h>
46 #include <linux/random.h>
47 #include <linux/module.h>
48 #include <linux/reboot.h>
49 #include <linux/file.h>
50 #include <linux/compat.h>
51 #include <linux/delay.h>
52 #include <linux/raid/md_p.h>
53 #include <linux/raid/md_u.h>
54 #include <linux/slab.h>
59 static void autostart_arrays(int part);
62 /* pers_list is a list of registered personalities protected
64 * pers_lock does extra service to protect accesses to
65 * mddev->thread when the mutex cannot be held.
67 static LIST_HEAD(pers_list);
68 static DEFINE_SPINLOCK(pers_lock);
70 static void md_print_devices(void);
72 static DECLARE_WAIT_QUEUE_HEAD(resync_wait);
73 static struct workqueue_struct *md_wq;
74 static struct workqueue_struct *md_misc_wq;
76 #define MD_BUG(x...) { printk("md: bug in file %s, line %d\n", __FILE__, __LINE__); md_print_devices(); }
79 * Default number of read corrections we'll attempt on an rdev
80 * before ejecting it from the array. We divide the read error
81 * count by 2 for every hour elapsed between read errors.
83 #define MD_DEFAULT_MAX_CORRECTED_READ_ERRORS 20
85 * Current RAID-1,4,5 parallel reconstruction 'guaranteed speed limit'
86 * is 1000 KB/sec, so the extra system load does not show up that much.
87 * Increase it if you want to have more _guaranteed_ speed. Note that
88 * the RAID driver will use the maximum available bandwidth if the IO
89 * subsystem is idle. There is also an 'absolute maximum' reconstruction
90 * speed limit - in case reconstruction slows down your system despite
93 * you can change it via /proc/sys/dev/raid/speed_limit_min and _max.
94 * or /sys/block/mdX/md/sync_speed_{min,max}
97 static int sysctl_speed_limit_min = 1000;
98 static int sysctl_speed_limit_max = 200000;
99 static inline int speed_min(struct mddev *mddev)
101 return mddev->sync_speed_min ?
102 mddev->sync_speed_min : sysctl_speed_limit_min;
105 static inline int speed_max(struct mddev *mddev)
107 return mddev->sync_speed_max ?
108 mddev->sync_speed_max : sysctl_speed_limit_max;
111 static struct ctl_table_header *raid_table_header;
113 static ctl_table raid_table[] = {
115 .procname = "speed_limit_min",
116 .data = &sysctl_speed_limit_min,
117 .maxlen = sizeof(int),
118 .mode = S_IRUGO|S_IWUSR,
119 .proc_handler = proc_dointvec,
122 .procname = "speed_limit_max",
123 .data = &sysctl_speed_limit_max,
124 .maxlen = sizeof(int),
125 .mode = S_IRUGO|S_IWUSR,
126 .proc_handler = proc_dointvec,
131 static ctl_table raid_dir_table[] = {
135 .mode = S_IRUGO|S_IXUGO,
141 static ctl_table raid_root_table[] = {
146 .child = raid_dir_table,
151 static const struct block_device_operations md_fops;
153 static int start_readonly;
156 * like bio_clone, but with a local bio set
159 static void mddev_bio_destructor(struct bio *bio)
161 struct mddev *mddev, **mddevp;
166 bio_free(bio, mddev->bio_set);
169 struct bio *bio_alloc_mddev(gfp_t gfp_mask, int nr_iovecs,
173 struct mddev **mddevp;
175 if (!mddev || !mddev->bio_set)
176 return bio_alloc(gfp_mask, nr_iovecs);
178 b = bio_alloc_bioset(gfp_mask, nr_iovecs,
184 b->bi_destructor = mddev_bio_destructor;
187 EXPORT_SYMBOL_GPL(bio_alloc_mddev);
189 struct bio *bio_clone_mddev(struct bio *bio, gfp_t gfp_mask,
193 struct mddev **mddevp;
195 if (!mddev || !mddev->bio_set)
196 return bio_clone(bio, gfp_mask);
198 b = bio_alloc_bioset(gfp_mask, bio->bi_max_vecs,
204 b->bi_destructor = mddev_bio_destructor;
206 if (bio_integrity(bio)) {
209 ret = bio_integrity_clone(b, bio, gfp_mask, mddev->bio_set);
219 EXPORT_SYMBOL_GPL(bio_clone_mddev);
221 void md_trim_bio(struct bio *bio, int offset, int size)
223 /* 'bio' is a cloned bio which we need to trim to match
224 * the given offset and size.
225 * This requires adjusting bi_sector, bi_size, and bi_io_vec
228 struct bio_vec *bvec;
232 if (offset == 0 && size == bio->bi_size)
235 bio->bi_sector += offset;
238 clear_bit(BIO_SEG_VALID, &bio->bi_flags);
240 while (bio->bi_idx < bio->bi_vcnt &&
241 bio->bi_io_vec[bio->bi_idx].bv_len <= offset) {
242 /* remove this whole bio_vec */
243 offset -= bio->bi_io_vec[bio->bi_idx].bv_len;
246 if (bio->bi_idx < bio->bi_vcnt) {
247 bio->bi_io_vec[bio->bi_idx].bv_offset += offset;
248 bio->bi_io_vec[bio->bi_idx].bv_len -= offset;
250 /* avoid any complications with bi_idx being non-zero*/
252 memmove(bio->bi_io_vec, bio->bi_io_vec+bio->bi_idx,
253 (bio->bi_vcnt - bio->bi_idx) * sizeof(struct bio_vec));
254 bio->bi_vcnt -= bio->bi_idx;
257 /* Make sure vcnt and last bv are not too big */
258 bio_for_each_segment(bvec, bio, i) {
259 if (sofar + bvec->bv_len > size)
260 bvec->bv_len = size - sofar;
261 if (bvec->bv_len == 0) {
265 sofar += bvec->bv_len;
268 EXPORT_SYMBOL_GPL(md_trim_bio);
271 * We have a system wide 'event count' that is incremented
272 * on any 'interesting' event, and readers of /proc/mdstat
273 * can use 'poll' or 'select' to find out when the event
277 * start array, stop array, error, add device, remove device,
278 * start build, activate spare
280 static DECLARE_WAIT_QUEUE_HEAD(md_event_waiters);
281 static atomic_t md_event_count;
282 void md_new_event(struct mddev *mddev)
284 atomic_inc(&md_event_count);
285 wake_up(&md_event_waiters);
287 EXPORT_SYMBOL_GPL(md_new_event);
289 /* Alternate version that can be called from interrupts
290 * when calling sysfs_notify isn't needed.
292 static void md_new_event_inintr(struct mddev *mddev)
294 atomic_inc(&md_event_count);
295 wake_up(&md_event_waiters);
299 * Enables to iterate over all existing md arrays
300 * all_mddevs_lock protects this list.
302 static LIST_HEAD(all_mddevs);
303 static DEFINE_SPINLOCK(all_mddevs_lock);
307 * iterates through all used mddevs in the system.
308 * We take care to grab the all_mddevs_lock whenever navigating
309 * the list, and to always hold a refcount when unlocked.
310 * Any code which breaks out of this loop while own
311 * a reference to the current mddev and must mddev_put it.
313 #define for_each_mddev(_mddev,_tmp) \
315 for (({ spin_lock(&all_mddevs_lock); \
316 _tmp = all_mddevs.next; \
318 ({ if (_tmp != &all_mddevs) \
319 mddev_get(list_entry(_tmp, struct mddev, all_mddevs));\
320 spin_unlock(&all_mddevs_lock); \
321 if (_mddev) mddev_put(_mddev); \
322 _mddev = list_entry(_tmp, struct mddev, all_mddevs); \
323 _tmp != &all_mddevs;}); \
324 ({ spin_lock(&all_mddevs_lock); \
325 _tmp = _tmp->next;}) \
329 /* Rather than calling directly into the personality make_request function,
330 * IO requests come here first so that we can check if the device is
331 * being suspended pending a reconfiguration.
332 * We hold a refcount over the call to ->make_request. By the time that
333 * call has finished, the bio has been linked into some internal structure
334 * and so is visible to ->quiesce(), so we don't need the refcount any more.
336 static void md_make_request(struct request_queue *q, struct bio *bio)
338 const int rw = bio_data_dir(bio);
339 struct mddev *mddev = q->queuedata;
341 unsigned int sectors;
343 if (mddev == NULL || mddev->pers == NULL
348 smp_rmb(); /* Ensure implications of 'active' are visible */
350 if (mddev->suspended) {
353 prepare_to_wait(&mddev->sb_wait, &__wait,
354 TASK_UNINTERRUPTIBLE);
355 if (!mddev->suspended)
361 finish_wait(&mddev->sb_wait, &__wait);
363 atomic_inc(&mddev->active_io);
367 * save the sectors now since our bio can
368 * go away inside make_request
370 sectors = bio_sectors(bio);
371 mddev->pers->make_request(mddev, bio);
373 cpu = part_stat_lock();
374 part_stat_inc(cpu, &mddev->gendisk->part0, ios[rw]);
375 part_stat_add(cpu, &mddev->gendisk->part0, sectors[rw], sectors);
378 if (atomic_dec_and_test(&mddev->active_io) && mddev->suspended)
379 wake_up(&mddev->sb_wait);
382 /* mddev_suspend makes sure no new requests are submitted
383 * to the device, and that any requests that have been submitted
384 * are completely handled.
385 * Once ->stop is called and completes, the module will be completely
388 void mddev_suspend(struct mddev *mddev)
390 BUG_ON(mddev->suspended);
391 mddev->suspended = 1;
393 wait_event(mddev->sb_wait, atomic_read(&mddev->active_io) == 0);
394 mddev->pers->quiesce(mddev, 1);
396 del_timer_sync(&mddev->safemode_timer);
398 EXPORT_SYMBOL_GPL(mddev_suspend);
400 void mddev_resume(struct mddev *mddev)
402 mddev->suspended = 0;
403 wake_up(&mddev->sb_wait);
404 mddev->pers->quiesce(mddev, 0);
406 md_wakeup_thread(mddev->thread);
407 md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
409 EXPORT_SYMBOL_GPL(mddev_resume);
411 int mddev_congested(struct mddev *mddev, int bits)
413 return mddev->suspended;
415 EXPORT_SYMBOL(mddev_congested);
418 * Generic flush handling for md
421 static void md_end_flush(struct bio *bio, int err)
423 struct md_rdev *rdev = bio->bi_private;
424 struct mddev *mddev = rdev->mddev;
426 rdev_dec_pending(rdev, mddev);
428 if (atomic_dec_and_test(&mddev->flush_pending)) {
429 /* The pre-request flush has finished */
430 queue_work(md_wq, &mddev->flush_work);
435 static void md_submit_flush_data(struct work_struct *ws);
437 static void submit_flushes(struct work_struct *ws)
439 struct mddev *mddev = container_of(ws, struct mddev, flush_work);
440 struct md_rdev *rdev;
442 INIT_WORK(&mddev->flush_work, md_submit_flush_data);
443 atomic_set(&mddev->flush_pending, 1);
445 list_for_each_entry_rcu(rdev, &mddev->disks, same_set)
446 if (rdev->raid_disk >= 0 &&
447 !test_bit(Faulty, &rdev->flags)) {
448 /* Take two references, one is dropped
449 * when request finishes, one after
450 * we reclaim rcu_read_lock
453 atomic_inc(&rdev->nr_pending);
454 atomic_inc(&rdev->nr_pending);
456 bi = bio_alloc_mddev(GFP_NOIO, 0, mddev);
457 bi->bi_end_io = md_end_flush;
458 bi->bi_private = rdev;
459 bi->bi_bdev = rdev->bdev;
460 atomic_inc(&mddev->flush_pending);
461 submit_bio(WRITE_FLUSH, bi);
463 rdev_dec_pending(rdev, mddev);
466 if (atomic_dec_and_test(&mddev->flush_pending))
467 queue_work(md_wq, &mddev->flush_work);
470 static void md_submit_flush_data(struct work_struct *ws)
472 struct mddev *mddev = container_of(ws, struct mddev, flush_work);
473 struct bio *bio = mddev->flush_bio;
475 if (bio->bi_size == 0)
476 /* an empty barrier - all done */
479 bio->bi_rw &= ~REQ_FLUSH;
480 mddev->pers->make_request(mddev, bio);
483 mddev->flush_bio = NULL;
484 wake_up(&mddev->sb_wait);
487 void md_flush_request(struct mddev *mddev, struct bio *bio)
489 spin_lock_irq(&mddev->write_lock);
490 wait_event_lock_irq(mddev->sb_wait,
492 mddev->write_lock, /*nothing*/);
493 mddev->flush_bio = bio;
494 spin_unlock_irq(&mddev->write_lock);
496 INIT_WORK(&mddev->flush_work, submit_flushes);
497 queue_work(md_wq, &mddev->flush_work);
499 EXPORT_SYMBOL(md_flush_request);
501 /* Support for plugging.
502 * This mirrors the plugging support in request_queue, but does not
503 * require having a whole queue or request structures.
504 * We allocate an md_plug_cb for each md device and each thread it gets
505 * plugged on. This links tot the private plug_handle structure in the
506 * personality data where we keep a count of the number of outstanding
507 * plugs so other code can see if a plug is active.
510 struct blk_plug_cb cb;
514 static void plugger_unplug(struct blk_plug_cb *cb)
516 struct md_plug_cb *mdcb = container_of(cb, struct md_plug_cb, cb);
517 if (atomic_dec_and_test(&mdcb->mddev->plug_cnt))
518 md_wakeup_thread(mdcb->mddev->thread);
522 /* Check that an unplug wakeup will come shortly.
523 * If not, wakeup the md thread immediately
525 int mddev_check_plugged(struct mddev *mddev)
527 struct blk_plug *plug = current->plug;
528 struct md_plug_cb *mdcb;
533 list_for_each_entry(mdcb, &plug->cb_list, cb.list) {
534 if (mdcb->cb.callback == plugger_unplug &&
535 mdcb->mddev == mddev) {
536 /* Already on the list, move to top */
537 if (mdcb != list_first_entry(&plug->cb_list,
540 list_move(&mdcb->cb.list, &plug->cb_list);
544 /* Not currently on the callback list */
545 mdcb = kmalloc(sizeof(*mdcb), GFP_ATOMIC);
550 mdcb->cb.callback = plugger_unplug;
551 atomic_inc(&mddev->plug_cnt);
552 list_add(&mdcb->cb.list, &plug->cb_list);
555 EXPORT_SYMBOL_GPL(mddev_check_plugged);
557 static inline struct mddev *mddev_get(struct mddev *mddev)
559 atomic_inc(&mddev->active);
563 static void mddev_delayed_delete(struct work_struct *ws);
565 static void mddev_put(struct mddev *mddev)
567 struct bio_set *bs = NULL;
569 if (!atomic_dec_and_lock(&mddev->active, &all_mddevs_lock))
571 if (!mddev->raid_disks && list_empty(&mddev->disks) &&
572 mddev->ctime == 0 && !mddev->hold_active) {
573 /* Array is not configured at all, and not held active,
575 list_del_init(&mddev->all_mddevs);
577 mddev->bio_set = NULL;
578 if (mddev->gendisk) {
579 /* We did a probe so need to clean up. Call
580 * queue_work inside the spinlock so that
581 * flush_workqueue() after mddev_find will
582 * succeed in waiting for the work to be done.
584 INIT_WORK(&mddev->del_work, mddev_delayed_delete);
585 queue_work(md_misc_wq, &mddev->del_work);
589 spin_unlock(&all_mddevs_lock);
594 void mddev_init(struct mddev *mddev)
596 mutex_init(&mddev->open_mutex);
597 mutex_init(&mddev->reconfig_mutex);
598 mutex_init(&mddev->bitmap_info.mutex);
599 INIT_LIST_HEAD(&mddev->disks);
600 INIT_LIST_HEAD(&mddev->all_mddevs);
601 init_timer(&mddev->safemode_timer);
602 atomic_set(&mddev->active, 1);
603 atomic_set(&mddev->openers, 0);
604 atomic_set(&mddev->active_io, 0);
605 atomic_set(&mddev->plug_cnt, 0);
606 spin_lock_init(&mddev->write_lock);
607 atomic_set(&mddev->flush_pending, 0);
608 init_waitqueue_head(&mddev->sb_wait);
609 init_waitqueue_head(&mddev->recovery_wait);
610 mddev->reshape_position = MaxSector;
611 mddev->resync_min = 0;
612 mddev->resync_max = MaxSector;
613 mddev->level = LEVEL_NONE;
615 EXPORT_SYMBOL_GPL(mddev_init);
617 static struct mddev * mddev_find(dev_t unit)
619 struct mddev *mddev, *new = NULL;
621 if (unit && MAJOR(unit) != MD_MAJOR)
622 unit &= ~((1<<MdpMinorShift)-1);
625 spin_lock(&all_mddevs_lock);
628 list_for_each_entry(mddev, &all_mddevs, all_mddevs)
629 if (mddev->unit == unit) {
631 spin_unlock(&all_mddevs_lock);
637 list_add(&new->all_mddevs, &all_mddevs);
638 spin_unlock(&all_mddevs_lock);
639 new->hold_active = UNTIL_IOCTL;
643 /* find an unused unit number */
644 static int next_minor = 512;
645 int start = next_minor;
649 dev = MKDEV(MD_MAJOR, next_minor);
651 if (next_minor > MINORMASK)
653 if (next_minor == start) {
654 /* Oh dear, all in use. */
655 spin_unlock(&all_mddevs_lock);
661 list_for_each_entry(mddev, &all_mddevs, all_mddevs)
662 if (mddev->unit == dev) {
668 new->md_minor = MINOR(dev);
669 new->hold_active = UNTIL_STOP;
670 list_add(&new->all_mddevs, &all_mddevs);
671 spin_unlock(&all_mddevs_lock);
674 spin_unlock(&all_mddevs_lock);
676 new = kzalloc(sizeof(*new), GFP_KERNEL);
681 if (MAJOR(unit) == MD_MAJOR)
682 new->md_minor = MINOR(unit);
684 new->md_minor = MINOR(unit) >> MdpMinorShift;
691 static inline int mddev_lock(struct mddev * mddev)
693 return mutex_lock_interruptible(&mddev->reconfig_mutex);
696 static inline int mddev_is_locked(struct mddev *mddev)
698 return mutex_is_locked(&mddev->reconfig_mutex);
701 static inline int mddev_trylock(struct mddev * mddev)
703 return mutex_trylock(&mddev->reconfig_mutex);
706 static struct attribute_group md_redundancy_group;
708 static void mddev_unlock(struct mddev * mddev)
710 if (mddev->to_remove) {
711 /* These cannot be removed under reconfig_mutex as
712 * an access to the files will try to take reconfig_mutex
713 * while holding the file unremovable, which leads to
715 * So hold set sysfs_active while the remove in happeing,
716 * and anything else which might set ->to_remove or my
717 * otherwise change the sysfs namespace will fail with
718 * -EBUSY if sysfs_active is still set.
719 * We set sysfs_active under reconfig_mutex and elsewhere
720 * test it under the same mutex to ensure its correct value
723 struct attribute_group *to_remove = mddev->to_remove;
724 mddev->to_remove = NULL;
725 mddev->sysfs_active = 1;
726 mutex_unlock(&mddev->reconfig_mutex);
728 if (mddev->kobj.sd) {
729 if (to_remove != &md_redundancy_group)
730 sysfs_remove_group(&mddev->kobj, to_remove);
731 if (mddev->pers == NULL ||
732 mddev->pers->sync_request == NULL) {
733 sysfs_remove_group(&mddev->kobj, &md_redundancy_group);
734 if (mddev->sysfs_action)
735 sysfs_put(mddev->sysfs_action);
736 mddev->sysfs_action = NULL;
739 mddev->sysfs_active = 0;
741 mutex_unlock(&mddev->reconfig_mutex);
743 /* As we've dropped the mutex we need a spinlock to
744 * make sure the thread doesn't disappear
746 spin_lock(&pers_lock);
747 md_wakeup_thread(mddev->thread);
748 spin_unlock(&pers_lock);
751 static struct md_rdev * find_rdev_nr(struct mddev *mddev, int nr)
753 struct md_rdev *rdev;
755 list_for_each_entry(rdev, &mddev->disks, same_set)
756 if (rdev->desc_nr == nr)
762 static struct md_rdev * find_rdev(struct mddev * mddev, dev_t dev)
764 struct md_rdev *rdev;
766 list_for_each_entry(rdev, &mddev->disks, same_set)
767 if (rdev->bdev->bd_dev == dev)
773 static struct md_personality *find_pers(int level, char *clevel)
775 struct md_personality *pers;
776 list_for_each_entry(pers, &pers_list, list) {
777 if (level != LEVEL_NONE && pers->level == level)
779 if (strcmp(pers->name, clevel)==0)
785 /* return the offset of the super block in 512byte sectors */
786 static inline sector_t calc_dev_sboffset(struct md_rdev *rdev)
788 sector_t num_sectors = i_size_read(rdev->bdev->bd_inode) / 512;
789 return MD_NEW_SIZE_SECTORS(num_sectors);
792 static int alloc_disk_sb(struct md_rdev * rdev)
797 rdev->sb_page = alloc_page(GFP_KERNEL);
798 if (!rdev->sb_page) {
799 printk(KERN_ALERT "md: out of memory.\n");
806 static void free_disk_sb(struct md_rdev * rdev)
809 put_page(rdev->sb_page);
811 rdev->sb_page = NULL;
816 put_page(rdev->bb_page);
817 rdev->bb_page = NULL;
822 static void super_written(struct bio *bio, int error)
824 struct md_rdev *rdev = bio->bi_private;
825 struct mddev *mddev = rdev->mddev;
827 if (error || !test_bit(BIO_UPTODATE, &bio->bi_flags)) {
828 printk("md: super_written gets error=%d, uptodate=%d\n",
829 error, test_bit(BIO_UPTODATE, &bio->bi_flags));
830 WARN_ON(test_bit(BIO_UPTODATE, &bio->bi_flags));
831 md_error(mddev, rdev);
834 if (atomic_dec_and_test(&mddev->pending_writes))
835 wake_up(&mddev->sb_wait);
839 void md_super_write(struct mddev *mddev, struct md_rdev *rdev,
840 sector_t sector, int size, struct page *page)
842 /* write first size bytes of page to sector of rdev
843 * Increment mddev->pending_writes before returning
844 * and decrement it on completion, waking up sb_wait
845 * if zero is reached.
846 * If an error occurred, call md_error
848 struct bio *bio = bio_alloc_mddev(GFP_NOIO, 1, mddev);
850 bio->bi_bdev = rdev->meta_bdev ? rdev->meta_bdev : rdev->bdev;
851 bio->bi_sector = sector;
852 bio_add_page(bio, page, size, 0);
853 bio->bi_private = rdev;
854 bio->bi_end_io = super_written;
856 atomic_inc(&mddev->pending_writes);
857 submit_bio(WRITE_FLUSH_FUA, bio);
860 void md_super_wait(struct mddev *mddev)
862 /* wait for all superblock writes that were scheduled to complete */
865 prepare_to_wait(&mddev->sb_wait, &wq, TASK_UNINTERRUPTIBLE);
866 if (atomic_read(&mddev->pending_writes)==0)
870 finish_wait(&mddev->sb_wait, &wq);
873 static void bi_complete(struct bio *bio, int error)
875 complete((struct completion*)bio->bi_private);
878 int sync_page_io(struct md_rdev *rdev, sector_t sector, int size,
879 struct page *page, int rw, bool metadata_op)
881 struct bio *bio = bio_alloc_mddev(GFP_NOIO, 1, rdev->mddev);
882 struct completion event;
887 bio->bi_bdev = (metadata_op && rdev->meta_bdev) ?
888 rdev->meta_bdev : rdev->bdev;
890 bio->bi_sector = sector + rdev->sb_start;
892 bio->bi_sector = sector + rdev->data_offset;
893 bio_add_page(bio, page, size, 0);
894 init_completion(&event);
895 bio->bi_private = &event;
896 bio->bi_end_io = bi_complete;
898 wait_for_completion(&event);
900 ret = test_bit(BIO_UPTODATE, &bio->bi_flags);
904 EXPORT_SYMBOL_GPL(sync_page_io);
906 static int read_disk_sb(struct md_rdev * rdev, int size)
908 char b[BDEVNAME_SIZE];
909 if (!rdev->sb_page) {
917 if (!sync_page_io(rdev, 0, size, rdev->sb_page, READ, true))
923 printk(KERN_WARNING "md: disabled device %s, could not read superblock.\n",
924 bdevname(rdev->bdev,b));
928 static int uuid_equal(mdp_super_t *sb1, mdp_super_t *sb2)
930 return sb1->set_uuid0 == sb2->set_uuid0 &&
931 sb1->set_uuid1 == sb2->set_uuid1 &&
932 sb1->set_uuid2 == sb2->set_uuid2 &&
933 sb1->set_uuid3 == sb2->set_uuid3;
936 static int sb_equal(mdp_super_t *sb1, mdp_super_t *sb2)
939 mdp_super_t *tmp1, *tmp2;
941 tmp1 = kmalloc(sizeof(*tmp1),GFP_KERNEL);
942 tmp2 = kmalloc(sizeof(*tmp2),GFP_KERNEL);
944 if (!tmp1 || !tmp2) {
946 printk(KERN_INFO "md.c sb_equal(): failed to allocate memory!\n");
954 * nr_disks is not constant
959 ret = (memcmp(tmp1, tmp2, MD_SB_GENERIC_CONSTANT_WORDS * 4) == 0);
967 static u32 md_csum_fold(u32 csum)
969 csum = (csum & 0xffff) + (csum >> 16);
970 return (csum & 0xffff) + (csum >> 16);
973 static unsigned int calc_sb_csum(mdp_super_t * sb)
976 u32 *sb32 = (u32*)sb;
978 unsigned int disk_csum, csum;
980 disk_csum = sb->sb_csum;
983 for (i = 0; i < MD_SB_BYTES/4 ; i++)
985 csum = (newcsum & 0xffffffff) + (newcsum>>32);
989 /* This used to use csum_partial, which was wrong for several
990 * reasons including that different results are returned on
991 * different architectures. It isn't critical that we get exactly
992 * the same return value as before (we always csum_fold before
993 * testing, and that removes any differences). However as we
994 * know that csum_partial always returned a 16bit value on
995 * alphas, do a fold to maximise conformity to previous behaviour.
997 sb->sb_csum = md_csum_fold(disk_csum);
999 sb->sb_csum = disk_csum;
1006 * Handle superblock details.
1007 * We want to be able to handle multiple superblock formats
1008 * so we have a common interface to them all, and an array of
1009 * different handlers.
1010 * We rely on user-space to write the initial superblock, and support
1011 * reading and updating of superblocks.
1012 * Interface methods are:
1013 * int load_super(struct md_rdev *dev, struct md_rdev *refdev, int minor_version)
1014 * loads and validates a superblock on dev.
1015 * if refdev != NULL, compare superblocks on both devices
1017 * 0 - dev has a superblock that is compatible with refdev
1018 * 1 - dev has a superblock that is compatible and newer than refdev
1019 * so dev should be used as the refdev in future
1020 * -EINVAL superblock incompatible or invalid
1021 * -othererror e.g. -EIO
1023 * int validate_super(struct mddev *mddev, struct md_rdev *dev)
1024 * Verify that dev is acceptable into mddev.
1025 * The first time, mddev->raid_disks will be 0, and data from
1026 * dev should be merged in. Subsequent calls check that dev
1027 * is new enough. Return 0 or -EINVAL
1029 * void sync_super(struct mddev *mddev, struct md_rdev *dev)
1030 * Update the superblock for rdev with data in mddev
1031 * This does not write to disc.
1037 struct module *owner;
1038 int (*load_super)(struct md_rdev *rdev, struct md_rdev *refdev,
1040 int (*validate_super)(struct mddev *mddev, struct md_rdev *rdev);
1041 void (*sync_super)(struct mddev *mddev, struct md_rdev *rdev);
1042 unsigned long long (*rdev_size_change)(struct md_rdev *rdev,
1043 sector_t num_sectors);
1047 * Check that the given mddev has no bitmap.
1049 * This function is called from the run method of all personalities that do not
1050 * support bitmaps. It prints an error message and returns non-zero if mddev
1051 * has a bitmap. Otherwise, it returns 0.
1054 int md_check_no_bitmap(struct mddev *mddev)
1056 if (!mddev->bitmap_info.file && !mddev->bitmap_info.offset)
1058 printk(KERN_ERR "%s: bitmaps are not supported for %s\n",
1059 mdname(mddev), mddev->pers->name);
1062 EXPORT_SYMBOL(md_check_no_bitmap);
1065 * load_super for 0.90.0
1067 static int super_90_load(struct md_rdev *rdev, struct md_rdev *refdev, int minor_version)
1069 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
1074 * Calculate the position of the superblock (512byte sectors),
1075 * it's at the end of the disk.
1077 * It also happens to be a multiple of 4Kb.
1079 rdev->sb_start = calc_dev_sboffset(rdev);
1081 ret = read_disk_sb(rdev, MD_SB_BYTES);
1082 if (ret) return ret;
1086 bdevname(rdev->bdev, b);
1087 sb = page_address(rdev->sb_page);
1089 if (sb->md_magic != MD_SB_MAGIC) {
1090 printk(KERN_ERR "md: invalid raid superblock magic on %s\n",
1095 if (sb->major_version != 0 ||
1096 sb->minor_version < 90 ||
1097 sb->minor_version > 91) {
1098 printk(KERN_WARNING "Bad version number %d.%d on %s\n",
1099 sb->major_version, sb->minor_version,
1104 if (sb->raid_disks <= 0)
1107 if (md_csum_fold(calc_sb_csum(sb)) != md_csum_fold(sb->sb_csum)) {
1108 printk(KERN_WARNING "md: invalid superblock checksum on %s\n",
1113 rdev->preferred_minor = sb->md_minor;
1114 rdev->data_offset = 0;
1115 rdev->sb_size = MD_SB_BYTES;
1116 rdev->badblocks.shift = -1;
1118 if (sb->level == LEVEL_MULTIPATH)
1121 rdev->desc_nr = sb->this_disk.number;
1127 mdp_super_t *refsb = page_address(refdev->sb_page);
1128 if (!uuid_equal(refsb, sb)) {
1129 printk(KERN_WARNING "md: %s has different UUID to %s\n",
1130 b, bdevname(refdev->bdev,b2));
1133 if (!sb_equal(refsb, sb)) {
1134 printk(KERN_WARNING "md: %s has same UUID"
1135 " but different superblock to %s\n",
1136 b, bdevname(refdev->bdev, b2));
1140 ev2 = md_event(refsb);
1146 rdev->sectors = rdev->sb_start;
1147 /* Limit to 4TB as metadata cannot record more than that.
1148 * (not needed for Linear and RAID0 as metadata doesn't
1151 if (rdev->sectors >= (2ULL << 32) && sb->level >= 1)
1152 rdev->sectors = (2ULL << 32) - 2;
1154 if (rdev->sectors < ((sector_t)sb->size) * 2 && sb->level >= 1)
1155 /* "this cannot possibly happen" ... */
1163 * validate_super for 0.90.0
1165 static int super_90_validate(struct mddev *mddev, struct md_rdev *rdev)
1168 mdp_super_t *sb = page_address(rdev->sb_page);
1169 __u64 ev1 = md_event(sb);
1171 rdev->raid_disk = -1;
1172 clear_bit(Faulty, &rdev->flags);
1173 clear_bit(In_sync, &rdev->flags);
1174 clear_bit(WriteMostly, &rdev->flags);
1176 if (mddev->raid_disks == 0) {
1177 mddev->major_version = 0;
1178 mddev->minor_version = sb->minor_version;
1179 mddev->patch_version = sb->patch_version;
1180 mddev->external = 0;
1181 mddev->chunk_sectors = sb->chunk_size >> 9;
1182 mddev->ctime = sb->ctime;
1183 mddev->utime = sb->utime;
1184 mddev->level = sb->level;
1185 mddev->clevel[0] = 0;
1186 mddev->layout = sb->layout;
1187 mddev->raid_disks = sb->raid_disks;
1188 mddev->dev_sectors = ((sector_t)sb->size) * 2;
1189 mddev->events = ev1;
1190 mddev->bitmap_info.offset = 0;
1191 mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9;
1193 if (mddev->minor_version >= 91) {
1194 mddev->reshape_position = sb->reshape_position;
1195 mddev->delta_disks = sb->delta_disks;
1196 mddev->new_level = sb->new_level;
1197 mddev->new_layout = sb->new_layout;
1198 mddev->new_chunk_sectors = sb->new_chunk >> 9;
1200 mddev->reshape_position = MaxSector;
1201 mddev->delta_disks = 0;
1202 mddev->new_level = mddev->level;
1203 mddev->new_layout = mddev->layout;
1204 mddev->new_chunk_sectors = mddev->chunk_sectors;
1207 if (sb->state & (1<<MD_SB_CLEAN))
1208 mddev->recovery_cp = MaxSector;
1210 if (sb->events_hi == sb->cp_events_hi &&
1211 sb->events_lo == sb->cp_events_lo) {
1212 mddev->recovery_cp = sb->recovery_cp;
1214 mddev->recovery_cp = 0;
1217 memcpy(mddev->uuid+0, &sb->set_uuid0, 4);
1218 memcpy(mddev->uuid+4, &sb->set_uuid1, 4);
1219 memcpy(mddev->uuid+8, &sb->set_uuid2, 4);
1220 memcpy(mddev->uuid+12,&sb->set_uuid3, 4);
1222 mddev->max_disks = MD_SB_DISKS;
1224 if (sb->state & (1<<MD_SB_BITMAP_PRESENT) &&
1225 mddev->bitmap_info.file == NULL)
1226 mddev->bitmap_info.offset =
1227 mddev->bitmap_info.default_offset;
1229 } else if (mddev->pers == NULL) {
1230 /* Insist on good event counter while assembling, except
1231 * for spares (which don't need an event count) */
1233 if (sb->disks[rdev->desc_nr].state & (
1234 (1<<MD_DISK_SYNC) | (1 << MD_DISK_ACTIVE)))
1235 if (ev1 < mddev->events)
1237 } else if (mddev->bitmap) {
1238 /* if adding to array with a bitmap, then we can accept an
1239 * older device ... but not too old.
1241 if (ev1 < mddev->bitmap->events_cleared)
1244 if (ev1 < mddev->events)
1245 /* just a hot-add of a new device, leave raid_disk at -1 */
1249 if (mddev->level != LEVEL_MULTIPATH) {
1250 desc = sb->disks + rdev->desc_nr;
1252 if (desc->state & (1<<MD_DISK_FAULTY))
1253 set_bit(Faulty, &rdev->flags);
1254 else if (desc->state & (1<<MD_DISK_SYNC) /* &&
1255 desc->raid_disk < mddev->raid_disks */) {
1256 set_bit(In_sync, &rdev->flags);
1257 rdev->raid_disk = desc->raid_disk;
1258 } else if (desc->state & (1<<MD_DISK_ACTIVE)) {
1259 /* active but not in sync implies recovery up to
1260 * reshape position. We don't know exactly where
1261 * that is, so set to zero for now */
1262 if (mddev->minor_version >= 91) {
1263 rdev->recovery_offset = 0;
1264 rdev->raid_disk = desc->raid_disk;
1267 if (desc->state & (1<<MD_DISK_WRITEMOSTLY))
1268 set_bit(WriteMostly, &rdev->flags);
1269 } else /* MULTIPATH are always insync */
1270 set_bit(In_sync, &rdev->flags);
1275 * sync_super for 0.90.0
1277 static void super_90_sync(struct mddev *mddev, struct md_rdev *rdev)
1280 struct md_rdev *rdev2;
1281 int next_spare = mddev->raid_disks;
1284 /* make rdev->sb match mddev data..
1287 * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
1288 * 3/ any empty disks < next_spare become removed
1290 * disks[0] gets initialised to REMOVED because
1291 * we cannot be sure from other fields if it has
1292 * been initialised or not.
1295 int active=0, working=0,failed=0,spare=0,nr_disks=0;
1297 rdev->sb_size = MD_SB_BYTES;
1299 sb = page_address(rdev->sb_page);
1301 memset(sb, 0, sizeof(*sb));
1303 sb->md_magic = MD_SB_MAGIC;
1304 sb->major_version = mddev->major_version;
1305 sb->patch_version = mddev->patch_version;
1306 sb->gvalid_words = 0; /* ignored */
1307 memcpy(&sb->set_uuid0, mddev->uuid+0, 4);
1308 memcpy(&sb->set_uuid1, mddev->uuid+4, 4);
1309 memcpy(&sb->set_uuid2, mddev->uuid+8, 4);
1310 memcpy(&sb->set_uuid3, mddev->uuid+12,4);
1312 sb->ctime = mddev->ctime;
1313 sb->level = mddev->level;
1314 sb->size = mddev->dev_sectors / 2;
1315 sb->raid_disks = mddev->raid_disks;
1316 sb->md_minor = mddev->md_minor;
1317 sb->not_persistent = 0;
1318 sb->utime = mddev->utime;
1320 sb->events_hi = (mddev->events>>32);
1321 sb->events_lo = (u32)mddev->events;
1323 if (mddev->reshape_position == MaxSector)
1324 sb->minor_version = 90;
1326 sb->minor_version = 91;
1327 sb->reshape_position = mddev->reshape_position;
1328 sb->new_level = mddev->new_level;
1329 sb->delta_disks = mddev->delta_disks;
1330 sb->new_layout = mddev->new_layout;
1331 sb->new_chunk = mddev->new_chunk_sectors << 9;
1333 mddev->minor_version = sb->minor_version;
1336 sb->recovery_cp = mddev->recovery_cp;
1337 sb->cp_events_hi = (mddev->events>>32);
1338 sb->cp_events_lo = (u32)mddev->events;
1339 if (mddev->recovery_cp == MaxSector)
1340 sb->state = (1<< MD_SB_CLEAN);
1342 sb->recovery_cp = 0;
1344 sb->layout = mddev->layout;
1345 sb->chunk_size = mddev->chunk_sectors << 9;
1347 if (mddev->bitmap && mddev->bitmap_info.file == NULL)
1348 sb->state |= (1<<MD_SB_BITMAP_PRESENT);
1350 sb->disks[0].state = (1<<MD_DISK_REMOVED);
1351 list_for_each_entry(rdev2, &mddev->disks, same_set) {
1354 int is_active = test_bit(In_sync, &rdev2->flags);
1356 if (rdev2->raid_disk >= 0 &&
1357 sb->minor_version >= 91)
1358 /* we have nowhere to store the recovery_offset,
1359 * but if it is not below the reshape_position,
1360 * we can piggy-back on that.
1363 if (rdev2->raid_disk < 0 ||
1364 test_bit(Faulty, &rdev2->flags))
1367 desc_nr = rdev2->raid_disk;
1369 desc_nr = next_spare++;
1370 rdev2->desc_nr = desc_nr;
1371 d = &sb->disks[rdev2->desc_nr];
1373 d->number = rdev2->desc_nr;
1374 d->major = MAJOR(rdev2->bdev->bd_dev);
1375 d->minor = MINOR(rdev2->bdev->bd_dev);
1377 d->raid_disk = rdev2->raid_disk;
1379 d->raid_disk = rdev2->desc_nr; /* compatibility */
1380 if (test_bit(Faulty, &rdev2->flags))
1381 d->state = (1<<MD_DISK_FAULTY);
1382 else if (is_active) {
1383 d->state = (1<<MD_DISK_ACTIVE);
1384 if (test_bit(In_sync, &rdev2->flags))
1385 d->state |= (1<<MD_DISK_SYNC);
1393 if (test_bit(WriteMostly, &rdev2->flags))
1394 d->state |= (1<<MD_DISK_WRITEMOSTLY);
1396 /* now set the "removed" and "faulty" bits on any missing devices */
1397 for (i=0 ; i < mddev->raid_disks ; i++) {
1398 mdp_disk_t *d = &sb->disks[i];
1399 if (d->state == 0 && d->number == 0) {
1402 d->state = (1<<MD_DISK_REMOVED);
1403 d->state |= (1<<MD_DISK_FAULTY);
1407 sb->nr_disks = nr_disks;
1408 sb->active_disks = active;
1409 sb->working_disks = working;
1410 sb->failed_disks = failed;
1411 sb->spare_disks = spare;
1413 sb->this_disk = sb->disks[rdev->desc_nr];
1414 sb->sb_csum = calc_sb_csum(sb);
1418 * rdev_size_change for 0.90.0
1420 static unsigned long long
1421 super_90_rdev_size_change(struct md_rdev *rdev, sector_t num_sectors)
1423 if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
1424 return 0; /* component must fit device */
1425 if (rdev->mddev->bitmap_info.offset)
1426 return 0; /* can't move bitmap */
1427 rdev->sb_start = calc_dev_sboffset(rdev);
1428 if (!num_sectors || num_sectors > rdev->sb_start)
1429 num_sectors = rdev->sb_start;
1430 /* Limit to 4TB as metadata cannot record more than that.
1431 * 4TB == 2^32 KB, or 2*2^32 sectors.
1433 if (num_sectors >= (2ULL << 32) && rdev->mddev->level >= 1)
1434 num_sectors = (2ULL << 32) - 2;
1435 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
1437 md_super_wait(rdev->mddev);
1443 * version 1 superblock
1446 static __le32 calc_sb_1_csum(struct mdp_superblock_1 * sb)
1450 unsigned long long newcsum;
1451 int size = 256 + le32_to_cpu(sb->max_dev)*2;
1452 __le32 *isuper = (__le32*)sb;
1455 disk_csum = sb->sb_csum;
1458 for (i=0; size>=4; size -= 4 )
1459 newcsum += le32_to_cpu(*isuper++);
1462 newcsum += le16_to_cpu(*(__le16*) isuper);
1464 csum = (newcsum & 0xffffffff) + (newcsum >> 32);
1465 sb->sb_csum = disk_csum;
1466 return cpu_to_le32(csum);
1469 static int md_set_badblocks(struct badblocks *bb, sector_t s, int sectors,
1471 static int super_1_load(struct md_rdev *rdev, struct md_rdev *refdev, int minor_version)
1473 struct mdp_superblock_1 *sb;
1476 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
1480 * Calculate the position of the superblock in 512byte sectors.
1481 * It is always aligned to a 4K boundary and
1482 * depeding on minor_version, it can be:
1483 * 0: At least 8K, but less than 12K, from end of device
1484 * 1: At start of device
1485 * 2: 4K from start of device.
1487 switch(minor_version) {
1489 sb_start = i_size_read(rdev->bdev->bd_inode) >> 9;
1491 sb_start &= ~(sector_t)(4*2-1);
1502 rdev->sb_start = sb_start;
1504 /* superblock is rarely larger than 1K, but it can be larger,
1505 * and it is safe to read 4k, so we do that
1507 ret = read_disk_sb(rdev, 4096);
1508 if (ret) return ret;
1511 sb = page_address(rdev->sb_page);
1513 if (sb->magic != cpu_to_le32(MD_SB_MAGIC) ||
1514 sb->major_version != cpu_to_le32(1) ||
1515 le32_to_cpu(sb->max_dev) > (4096-256)/2 ||
1516 le64_to_cpu(sb->super_offset) != rdev->sb_start ||
1517 (le32_to_cpu(sb->feature_map) & ~MD_FEATURE_ALL) != 0)
1520 if (calc_sb_1_csum(sb) != sb->sb_csum) {
1521 printk("md: invalid superblock checksum on %s\n",
1522 bdevname(rdev->bdev,b));
1525 if (le64_to_cpu(sb->data_size) < 10) {
1526 printk("md: data_size too small on %s\n",
1527 bdevname(rdev->bdev,b));
1531 rdev->preferred_minor = 0xffff;
1532 rdev->data_offset = le64_to_cpu(sb->data_offset);
1533 atomic_set(&rdev->corrected_errors, le32_to_cpu(sb->cnt_corrected_read));
1535 rdev->sb_size = le32_to_cpu(sb->max_dev) * 2 + 256;
1536 bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
1537 if (rdev->sb_size & bmask)
1538 rdev->sb_size = (rdev->sb_size | bmask) + 1;
1541 && rdev->data_offset < sb_start + (rdev->sb_size/512))
1544 if (sb->level == cpu_to_le32(LEVEL_MULTIPATH))
1547 rdev->desc_nr = le32_to_cpu(sb->dev_number);
1549 if (!rdev->bb_page) {
1550 rdev->bb_page = alloc_page(GFP_KERNEL);
1554 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BAD_BLOCKS) &&
1555 rdev->badblocks.count == 0) {
1556 /* need to load the bad block list.
1557 * Currently we limit it to one page.
1563 int sectors = le16_to_cpu(sb->bblog_size);
1564 if (sectors > (PAGE_SIZE / 512))
1566 offset = le32_to_cpu(sb->bblog_offset);
1569 bb_sector = (long long)offset;
1570 if (!sync_page_io(rdev, bb_sector, sectors << 9,
1571 rdev->bb_page, READ, true))
1573 bbp = (u64 *)page_address(rdev->bb_page);
1574 rdev->badblocks.shift = sb->bblog_shift;
1575 for (i = 0 ; i < (sectors << (9-3)) ; i++, bbp++) {
1576 u64 bb = le64_to_cpu(*bbp);
1577 int count = bb & (0x3ff);
1578 u64 sector = bb >> 10;
1579 sector <<= sb->bblog_shift;
1580 count <<= sb->bblog_shift;
1583 if (md_set_badblocks(&rdev->badblocks,
1584 sector, count, 1) == 0)
1587 } else if (sb->bblog_offset == 0)
1588 rdev->badblocks.shift = -1;
1594 struct mdp_superblock_1 *refsb = page_address(refdev->sb_page);
1596 if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 ||
1597 sb->level != refsb->level ||
1598 sb->layout != refsb->layout ||
1599 sb->chunksize != refsb->chunksize) {
1600 printk(KERN_WARNING "md: %s has strangely different"
1601 " superblock to %s\n",
1602 bdevname(rdev->bdev,b),
1603 bdevname(refdev->bdev,b2));
1606 ev1 = le64_to_cpu(sb->events);
1607 ev2 = le64_to_cpu(refsb->events);
1615 rdev->sectors = (i_size_read(rdev->bdev->bd_inode) >> 9) -
1616 le64_to_cpu(sb->data_offset);
1618 rdev->sectors = rdev->sb_start;
1619 if (rdev->sectors < le64_to_cpu(sb->data_size))
1621 rdev->sectors = le64_to_cpu(sb->data_size);
1622 if (le64_to_cpu(sb->size) > rdev->sectors)
1627 static int super_1_validate(struct mddev *mddev, struct md_rdev *rdev)
1629 struct mdp_superblock_1 *sb = page_address(rdev->sb_page);
1630 __u64 ev1 = le64_to_cpu(sb->events);
1632 rdev->raid_disk = -1;
1633 clear_bit(Faulty, &rdev->flags);
1634 clear_bit(In_sync, &rdev->flags);
1635 clear_bit(WriteMostly, &rdev->flags);
1637 if (mddev->raid_disks == 0) {
1638 mddev->major_version = 1;
1639 mddev->patch_version = 0;
1640 mddev->external = 0;
1641 mddev->chunk_sectors = le32_to_cpu(sb->chunksize);
1642 mddev->ctime = le64_to_cpu(sb->ctime) & ((1ULL << 32)-1);
1643 mddev->utime = le64_to_cpu(sb->utime) & ((1ULL << 32)-1);
1644 mddev->level = le32_to_cpu(sb->level);
1645 mddev->clevel[0] = 0;
1646 mddev->layout = le32_to_cpu(sb->layout);
1647 mddev->raid_disks = le32_to_cpu(sb->raid_disks);
1648 mddev->dev_sectors = le64_to_cpu(sb->size);
1649 mddev->events = ev1;
1650 mddev->bitmap_info.offset = 0;
1651 mddev->bitmap_info.default_offset = 1024 >> 9;
1653 mddev->recovery_cp = le64_to_cpu(sb->resync_offset);
1654 memcpy(mddev->uuid, sb->set_uuid, 16);
1656 mddev->max_disks = (4096-256)/2;
1658 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) &&
1659 mddev->bitmap_info.file == NULL )
1660 mddev->bitmap_info.offset =
1661 (__s32)le32_to_cpu(sb->bitmap_offset);
1663 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) {
1664 mddev->reshape_position = le64_to_cpu(sb->reshape_position);
1665 mddev->delta_disks = le32_to_cpu(sb->delta_disks);
1666 mddev->new_level = le32_to_cpu(sb->new_level);
1667 mddev->new_layout = le32_to_cpu(sb->new_layout);
1668 mddev->new_chunk_sectors = le32_to_cpu(sb->new_chunk);
1670 mddev->reshape_position = MaxSector;
1671 mddev->delta_disks = 0;
1672 mddev->new_level = mddev->level;
1673 mddev->new_layout = mddev->layout;
1674 mddev->new_chunk_sectors = mddev->chunk_sectors;
1677 } else if (mddev->pers == NULL) {
1678 /* Insist of good event counter while assembling, except for
1679 * spares (which don't need an event count) */
1681 if (rdev->desc_nr >= 0 &&
1682 rdev->desc_nr < le32_to_cpu(sb->max_dev) &&
1683 le16_to_cpu(sb->dev_roles[rdev->desc_nr]) < 0xfffe)
1684 if (ev1 < mddev->events)
1686 } else if (mddev->bitmap) {
1687 /* If adding to array with a bitmap, then we can accept an
1688 * older device, but not too old.
1690 if (ev1 < mddev->bitmap->events_cleared)
1693 if (ev1 < mddev->events)
1694 /* just a hot-add of a new device, leave raid_disk at -1 */
1697 if (mddev->level != LEVEL_MULTIPATH) {
1699 if (rdev->desc_nr < 0 ||
1700 rdev->desc_nr >= le32_to_cpu(sb->max_dev)) {
1704 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
1706 case 0xffff: /* spare */
1708 case 0xfffe: /* faulty */
1709 set_bit(Faulty, &rdev->flags);
1712 if ((le32_to_cpu(sb->feature_map) &
1713 MD_FEATURE_RECOVERY_OFFSET))
1714 rdev->recovery_offset = le64_to_cpu(sb->recovery_offset);
1716 set_bit(In_sync, &rdev->flags);
1717 rdev->raid_disk = role;
1720 if (sb->devflags & WriteMostly1)
1721 set_bit(WriteMostly, &rdev->flags);
1722 } else /* MULTIPATH are always insync */
1723 set_bit(In_sync, &rdev->flags);
1728 static void super_1_sync(struct mddev *mddev, struct md_rdev *rdev)
1730 struct mdp_superblock_1 *sb;
1731 struct md_rdev *rdev2;
1733 /* make rdev->sb match mddev and rdev data. */
1735 sb = page_address(rdev->sb_page);
1737 sb->feature_map = 0;
1739 sb->recovery_offset = cpu_to_le64(0);
1740 memset(sb->pad1, 0, sizeof(sb->pad1));
1741 memset(sb->pad3, 0, sizeof(sb->pad3));
1743 sb->utime = cpu_to_le64((__u64)mddev->utime);
1744 sb->events = cpu_to_le64(mddev->events);
1746 sb->resync_offset = cpu_to_le64(mddev->recovery_cp);
1748 sb->resync_offset = cpu_to_le64(0);
1750 sb->cnt_corrected_read = cpu_to_le32(atomic_read(&rdev->corrected_errors));
1752 sb->raid_disks = cpu_to_le32(mddev->raid_disks);
1753 sb->size = cpu_to_le64(mddev->dev_sectors);
1754 sb->chunksize = cpu_to_le32(mddev->chunk_sectors);
1755 sb->level = cpu_to_le32(mddev->level);
1756 sb->layout = cpu_to_le32(mddev->layout);
1758 if (test_bit(WriteMostly, &rdev->flags))
1759 sb->devflags |= WriteMostly1;
1761 sb->devflags &= ~WriteMostly1;
1763 if (mddev->bitmap && mddev->bitmap_info.file == NULL) {
1764 sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_info.offset);
1765 sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET);
1768 if (rdev->raid_disk >= 0 &&
1769 !test_bit(In_sync, &rdev->flags)) {
1771 cpu_to_le32(MD_FEATURE_RECOVERY_OFFSET);
1772 sb->recovery_offset =
1773 cpu_to_le64(rdev->recovery_offset);
1776 if (mddev->reshape_position != MaxSector) {
1777 sb->feature_map |= cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE);
1778 sb->reshape_position = cpu_to_le64(mddev->reshape_position);
1779 sb->new_layout = cpu_to_le32(mddev->new_layout);
1780 sb->delta_disks = cpu_to_le32(mddev->delta_disks);
1781 sb->new_level = cpu_to_le32(mddev->new_level);
1782 sb->new_chunk = cpu_to_le32(mddev->new_chunk_sectors);
1785 if (rdev->badblocks.count == 0)
1786 /* Nothing to do for bad blocks*/ ;
1787 else if (sb->bblog_offset == 0)
1788 /* Cannot record bad blocks on this device */
1789 md_error(mddev, rdev);
1791 struct badblocks *bb = &rdev->badblocks;
1792 u64 *bbp = (u64 *)page_address(rdev->bb_page);
1794 sb->feature_map |= cpu_to_le32(MD_FEATURE_BAD_BLOCKS);
1799 seq = read_seqbegin(&bb->lock);
1801 memset(bbp, 0xff, PAGE_SIZE);
1803 for (i = 0 ; i < bb->count ; i++) {
1804 u64 internal_bb = *p++;
1805 u64 store_bb = ((BB_OFFSET(internal_bb) << 10)
1806 | BB_LEN(internal_bb));
1807 *bbp++ = cpu_to_le64(store_bb);
1810 if (read_seqretry(&bb->lock, seq))
1813 bb->sector = (rdev->sb_start +
1814 (int)le32_to_cpu(sb->bblog_offset));
1815 bb->size = le16_to_cpu(sb->bblog_size);
1820 list_for_each_entry(rdev2, &mddev->disks, same_set)
1821 if (rdev2->desc_nr+1 > max_dev)
1822 max_dev = rdev2->desc_nr+1;
1824 if (max_dev > le32_to_cpu(sb->max_dev)) {
1826 sb->max_dev = cpu_to_le32(max_dev);
1827 rdev->sb_size = max_dev * 2 + 256;
1828 bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
1829 if (rdev->sb_size & bmask)
1830 rdev->sb_size = (rdev->sb_size | bmask) + 1;
1832 max_dev = le32_to_cpu(sb->max_dev);
1834 for (i=0; i<max_dev;i++)
1835 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1837 list_for_each_entry(rdev2, &mddev->disks, same_set) {
1839 if (test_bit(Faulty, &rdev2->flags))
1840 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1841 else if (test_bit(In_sync, &rdev2->flags))
1842 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1843 else if (rdev2->raid_disk >= 0)
1844 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1846 sb->dev_roles[i] = cpu_to_le16(0xffff);
1849 sb->sb_csum = calc_sb_1_csum(sb);
1852 static unsigned long long
1853 super_1_rdev_size_change(struct md_rdev *rdev, sector_t num_sectors)
1855 struct mdp_superblock_1 *sb;
1856 sector_t max_sectors;
1857 if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
1858 return 0; /* component must fit device */
1859 if (rdev->sb_start < rdev->data_offset) {
1860 /* minor versions 1 and 2; superblock before data */
1861 max_sectors = i_size_read(rdev->bdev->bd_inode) >> 9;
1862 max_sectors -= rdev->data_offset;
1863 if (!num_sectors || num_sectors > max_sectors)
1864 num_sectors = max_sectors;
1865 } else if (rdev->mddev->bitmap_info.offset) {
1866 /* minor version 0 with bitmap we can't move */
1869 /* minor version 0; superblock after data */
1871 sb_start = (i_size_read(rdev->bdev->bd_inode) >> 9) - 8*2;
1872 sb_start &= ~(sector_t)(4*2 - 1);
1873 max_sectors = rdev->sectors + sb_start - rdev->sb_start;
1874 if (!num_sectors || num_sectors > max_sectors)
1875 num_sectors = max_sectors;
1876 rdev->sb_start = sb_start;
1878 sb = page_address(rdev->sb_page);
1879 sb->data_size = cpu_to_le64(num_sectors);
1880 sb->super_offset = rdev->sb_start;
1881 sb->sb_csum = calc_sb_1_csum(sb);
1882 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
1884 md_super_wait(rdev->mddev);
1888 static struct super_type super_types[] = {
1891 .owner = THIS_MODULE,
1892 .load_super = super_90_load,
1893 .validate_super = super_90_validate,
1894 .sync_super = super_90_sync,
1895 .rdev_size_change = super_90_rdev_size_change,
1899 .owner = THIS_MODULE,
1900 .load_super = super_1_load,
1901 .validate_super = super_1_validate,
1902 .sync_super = super_1_sync,
1903 .rdev_size_change = super_1_rdev_size_change,
1907 static void sync_super(struct mddev *mddev, struct md_rdev *rdev)
1909 if (mddev->sync_super) {
1910 mddev->sync_super(mddev, rdev);
1914 BUG_ON(mddev->major_version >= ARRAY_SIZE(super_types));
1916 super_types[mddev->major_version].sync_super(mddev, rdev);
1919 static int match_mddev_units(struct mddev *mddev1, struct mddev *mddev2)
1921 struct md_rdev *rdev, *rdev2;
1924 rdev_for_each_rcu(rdev, mddev1)
1925 rdev_for_each_rcu(rdev2, mddev2)
1926 if (rdev->bdev->bd_contains ==
1927 rdev2->bdev->bd_contains) {
1935 static LIST_HEAD(pending_raid_disks);
1938 * Try to register data integrity profile for an mddev
1940 * This is called when an array is started and after a disk has been kicked
1941 * from the array. It only succeeds if all working and active component devices
1942 * are integrity capable with matching profiles.
1944 int md_integrity_register(struct mddev *mddev)
1946 struct md_rdev *rdev, *reference = NULL;
1948 if (list_empty(&mddev->disks))
1949 return 0; /* nothing to do */
1950 if (!mddev->gendisk || blk_get_integrity(mddev->gendisk))
1951 return 0; /* shouldn't register, or already is */
1952 list_for_each_entry(rdev, &mddev->disks, same_set) {
1953 /* skip spares and non-functional disks */
1954 if (test_bit(Faulty, &rdev->flags))
1956 if (rdev->raid_disk < 0)
1959 /* Use the first rdev as the reference */
1963 /* does this rdev's profile match the reference profile? */
1964 if (blk_integrity_compare(reference->bdev->bd_disk,
1965 rdev->bdev->bd_disk) < 0)
1968 if (!reference || !bdev_get_integrity(reference->bdev))
1971 * All component devices are integrity capable and have matching
1972 * profiles, register the common profile for the md device.
1974 if (blk_integrity_register(mddev->gendisk,
1975 bdev_get_integrity(reference->bdev)) != 0) {
1976 printk(KERN_ERR "md: failed to register integrity for %s\n",
1980 printk(KERN_NOTICE "md: data integrity enabled on %s\n", mdname(mddev));
1981 if (bioset_integrity_create(mddev->bio_set, BIO_POOL_SIZE)) {
1982 printk(KERN_ERR "md: failed to create integrity pool for %s\n",
1988 EXPORT_SYMBOL(md_integrity_register);
1990 /* Disable data integrity if non-capable/non-matching disk is being added */
1991 void md_integrity_add_rdev(struct md_rdev *rdev, struct mddev *mddev)
1993 struct blk_integrity *bi_rdev = bdev_get_integrity(rdev->bdev);
1994 struct blk_integrity *bi_mddev = blk_get_integrity(mddev->gendisk);
1996 if (!bi_mddev) /* nothing to do */
1998 if (rdev->raid_disk < 0) /* skip spares */
2000 if (bi_rdev && blk_integrity_compare(mddev->gendisk,
2001 rdev->bdev->bd_disk) >= 0)
2003 printk(KERN_NOTICE "disabling data integrity on %s\n", mdname(mddev));
2004 blk_integrity_unregister(mddev->gendisk);
2006 EXPORT_SYMBOL(md_integrity_add_rdev);
2008 static int bind_rdev_to_array(struct md_rdev * rdev, struct mddev * mddev)
2010 char b[BDEVNAME_SIZE];
2020 /* prevent duplicates */
2021 if (find_rdev(mddev, rdev->bdev->bd_dev))
2024 /* make sure rdev->sectors exceeds mddev->dev_sectors */
2025 if (rdev->sectors && (mddev->dev_sectors == 0 ||
2026 rdev->sectors < mddev->dev_sectors)) {
2028 /* Cannot change size, so fail
2029 * If mddev->level <= 0, then we don't care
2030 * about aligning sizes (e.g. linear)
2032 if (mddev->level > 0)
2035 mddev->dev_sectors = rdev->sectors;
2038 /* Verify rdev->desc_nr is unique.
2039 * If it is -1, assign a free number, else
2040 * check number is not in use
2042 if (rdev->desc_nr < 0) {
2044 if (mddev->pers) choice = mddev->raid_disks;
2045 while (find_rdev_nr(mddev, choice))
2047 rdev->desc_nr = choice;
2049 if (find_rdev_nr(mddev, rdev->desc_nr))
2052 if (mddev->max_disks && rdev->desc_nr >= mddev->max_disks) {
2053 printk(KERN_WARNING "md: %s: array is limited to %d devices\n",
2054 mdname(mddev), mddev->max_disks);
2057 bdevname(rdev->bdev,b);
2058 while ( (s=strchr(b, '/')) != NULL)
2061 rdev->mddev = mddev;
2062 printk(KERN_INFO "md: bind<%s>\n", b);
2064 if ((err = kobject_add(&rdev->kobj, &mddev->kobj, "dev-%s", b)))
2067 ko = &part_to_dev(rdev->bdev->bd_part)->kobj;
2068 if (sysfs_create_link(&rdev->kobj, ko, "block"))
2069 /* failure here is OK */;
2070 rdev->sysfs_state = sysfs_get_dirent_safe(rdev->kobj.sd, "state");
2072 list_add_rcu(&rdev->same_set, &mddev->disks);
2073 bd_link_disk_holder(rdev->bdev, mddev->gendisk);
2075 /* May as well allow recovery to be retried once */
2076 mddev->recovery_disabled++;
2081 printk(KERN_WARNING "md: failed to register dev-%s for %s\n",
2086 static void md_delayed_delete(struct work_struct *ws)
2088 struct md_rdev *rdev = container_of(ws, struct md_rdev, del_work);
2089 kobject_del(&rdev->kobj);
2090 kobject_put(&rdev->kobj);
2093 static void unbind_rdev_from_array(struct md_rdev * rdev)
2095 char b[BDEVNAME_SIZE];
2100 bd_unlink_disk_holder(rdev->bdev, rdev->mddev->gendisk);
2101 list_del_rcu(&rdev->same_set);
2102 printk(KERN_INFO "md: unbind<%s>\n", bdevname(rdev->bdev,b));
2104 sysfs_remove_link(&rdev->kobj, "block");
2105 sysfs_put(rdev->sysfs_state);
2106 rdev->sysfs_state = NULL;
2107 kfree(rdev->badblocks.page);
2108 rdev->badblocks.count = 0;
2109 rdev->badblocks.page = NULL;
2110 /* We need to delay this, otherwise we can deadlock when
2111 * writing to 'remove' to "dev/state". We also need
2112 * to delay it due to rcu usage.
2115 INIT_WORK(&rdev->del_work, md_delayed_delete);
2116 kobject_get(&rdev->kobj);
2117 queue_work(md_misc_wq, &rdev->del_work);
2121 * prevent the device from being mounted, repartitioned or
2122 * otherwise reused by a RAID array (or any other kernel
2123 * subsystem), by bd_claiming the device.
2125 static int lock_rdev(struct md_rdev *rdev, dev_t dev, int shared)
2128 struct block_device *bdev;
2129 char b[BDEVNAME_SIZE];
2131 bdev = blkdev_get_by_dev(dev, FMODE_READ|FMODE_WRITE|FMODE_EXCL,
2132 shared ? (struct md_rdev *)lock_rdev : rdev);
2134 printk(KERN_ERR "md: could not open %s.\n",
2135 __bdevname(dev, b));
2136 return PTR_ERR(bdev);
2142 static void unlock_rdev(struct md_rdev *rdev)
2144 struct block_device *bdev = rdev->bdev;
2148 blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
2151 void md_autodetect_dev(dev_t dev);
2153 static void export_rdev(struct md_rdev * rdev)
2155 char b[BDEVNAME_SIZE];
2156 printk(KERN_INFO "md: export_rdev(%s)\n",
2157 bdevname(rdev->bdev,b));
2162 if (test_bit(AutoDetected, &rdev->flags))
2163 md_autodetect_dev(rdev->bdev->bd_dev);
2166 kobject_put(&rdev->kobj);
2169 static void kick_rdev_from_array(struct md_rdev * rdev)
2171 unbind_rdev_from_array(rdev);
2175 static void export_array(struct mddev *mddev)
2177 struct md_rdev *rdev, *tmp;
2179 rdev_for_each(rdev, tmp, mddev) {
2184 kick_rdev_from_array(rdev);
2186 if (!list_empty(&mddev->disks))
2188 mddev->raid_disks = 0;
2189 mddev->major_version = 0;
2192 static void print_desc(mdp_disk_t *desc)
2194 printk(" DISK<N:%d,(%d,%d),R:%d,S:%d>\n", desc->number,
2195 desc->major,desc->minor,desc->raid_disk,desc->state);
2198 static void print_sb_90(mdp_super_t *sb)
2203 "md: SB: (V:%d.%d.%d) ID:<%08x.%08x.%08x.%08x> CT:%08x\n",
2204 sb->major_version, sb->minor_version, sb->patch_version,
2205 sb->set_uuid0, sb->set_uuid1, sb->set_uuid2, sb->set_uuid3,
2207 printk(KERN_INFO "md: L%d S%08d ND:%d RD:%d md%d LO:%d CS:%d\n",
2208 sb->level, sb->size, sb->nr_disks, sb->raid_disks,
2209 sb->md_minor, sb->layout, sb->chunk_size);
2210 printk(KERN_INFO "md: UT:%08x ST:%d AD:%d WD:%d"
2211 " FD:%d SD:%d CSUM:%08x E:%08lx\n",
2212 sb->utime, sb->state, sb->active_disks, sb->working_disks,
2213 sb->failed_disks, sb->spare_disks,
2214 sb->sb_csum, (unsigned long)sb->events_lo);
2217 for (i = 0; i < MD_SB_DISKS; i++) {
2220 desc = sb->disks + i;
2221 if (desc->number || desc->major || desc->minor ||
2222 desc->raid_disk || (desc->state && (desc->state != 4))) {
2223 printk(" D %2d: ", i);
2227 printk(KERN_INFO "md: THIS: ");
2228 print_desc(&sb->this_disk);
2231 static void print_sb_1(struct mdp_superblock_1 *sb)
2235 uuid = sb->set_uuid;
2237 "md: SB: (V:%u) (F:0x%08x) Array-ID:<%pU>\n"
2238 "md: Name: \"%s\" CT:%llu\n",
2239 le32_to_cpu(sb->major_version),
2240 le32_to_cpu(sb->feature_map),
2243 (unsigned long long)le64_to_cpu(sb->ctime)
2244 & MD_SUPERBLOCK_1_TIME_SEC_MASK);
2246 uuid = sb->device_uuid;
2248 "md: L%u SZ%llu RD:%u LO:%u CS:%u DO:%llu DS:%llu SO:%llu"
2250 "md: Dev:%08x UUID: %pU\n"
2251 "md: (F:0x%08x) UT:%llu Events:%llu ResyncOffset:%llu CSUM:0x%08x\n"
2252 "md: (MaxDev:%u) \n",
2253 le32_to_cpu(sb->level),
2254 (unsigned long long)le64_to_cpu(sb->size),
2255 le32_to_cpu(sb->raid_disks),
2256 le32_to_cpu(sb->layout),
2257 le32_to_cpu(sb->chunksize),
2258 (unsigned long long)le64_to_cpu(sb->data_offset),
2259 (unsigned long long)le64_to_cpu(sb->data_size),
2260 (unsigned long long)le64_to_cpu(sb->super_offset),
2261 (unsigned long long)le64_to_cpu(sb->recovery_offset),
2262 le32_to_cpu(sb->dev_number),
2265 (unsigned long long)le64_to_cpu(sb->utime) & MD_SUPERBLOCK_1_TIME_SEC_MASK,
2266 (unsigned long long)le64_to_cpu(sb->events),
2267 (unsigned long long)le64_to_cpu(sb->resync_offset),
2268 le32_to_cpu(sb->sb_csum),
2269 le32_to_cpu(sb->max_dev)
2273 static void print_rdev(struct md_rdev *rdev, int major_version)
2275 char b[BDEVNAME_SIZE];
2276 printk(KERN_INFO "md: rdev %s, Sect:%08llu F:%d S:%d DN:%u\n",
2277 bdevname(rdev->bdev, b), (unsigned long long)rdev->sectors,
2278 test_bit(Faulty, &rdev->flags), test_bit(In_sync, &rdev->flags),
2280 if (rdev->sb_loaded) {
2281 printk(KERN_INFO "md: rdev superblock (MJ:%d):\n", major_version);
2282 switch (major_version) {
2284 print_sb_90(page_address(rdev->sb_page));
2287 print_sb_1(page_address(rdev->sb_page));
2291 printk(KERN_INFO "md: no rdev superblock!\n");
2294 static void md_print_devices(void)
2296 struct list_head *tmp;
2297 struct md_rdev *rdev;
2298 struct mddev *mddev;
2299 char b[BDEVNAME_SIZE];
2302 printk("md: **********************************\n");
2303 printk("md: * <COMPLETE RAID STATE PRINTOUT> *\n");
2304 printk("md: **********************************\n");
2305 for_each_mddev(mddev, tmp) {
2308 bitmap_print_sb(mddev->bitmap);
2310 printk("%s: ", mdname(mddev));
2311 list_for_each_entry(rdev, &mddev->disks, same_set)
2312 printk("<%s>", bdevname(rdev->bdev,b));
2315 list_for_each_entry(rdev, &mddev->disks, same_set)
2316 print_rdev(rdev, mddev->major_version);
2318 printk("md: **********************************\n");
2323 static void sync_sbs(struct mddev * mddev, int nospares)
2325 /* Update each superblock (in-memory image), but
2326 * if we are allowed to, skip spares which already
2327 * have the right event counter, or have one earlier
2328 * (which would mean they aren't being marked as dirty
2329 * with the rest of the array)
2331 struct md_rdev *rdev;
2332 list_for_each_entry(rdev, &mddev->disks, same_set) {
2333 if (rdev->sb_events == mddev->events ||
2335 rdev->raid_disk < 0 &&
2336 rdev->sb_events+1 == mddev->events)) {
2337 /* Don't update this superblock */
2338 rdev->sb_loaded = 2;
2340 sync_super(mddev, rdev);
2341 rdev->sb_loaded = 1;
2346 static void md_update_sb(struct mddev * mddev, int force_change)
2348 struct md_rdev *rdev;
2351 int any_badblocks_changed = 0;
2354 /* First make sure individual recovery_offsets are correct */
2355 list_for_each_entry(rdev, &mddev->disks, same_set) {
2356 if (rdev->raid_disk >= 0 &&
2357 mddev->delta_disks >= 0 &&
2358 !test_bit(In_sync, &rdev->flags) &&
2359 mddev->curr_resync_completed > rdev->recovery_offset)
2360 rdev->recovery_offset = mddev->curr_resync_completed;
2363 if (!mddev->persistent) {
2364 clear_bit(MD_CHANGE_CLEAN, &mddev->flags);
2365 clear_bit(MD_CHANGE_DEVS, &mddev->flags);
2366 if (!mddev->external) {
2367 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
2368 list_for_each_entry(rdev, &mddev->disks, same_set) {
2369 if (rdev->badblocks.changed) {
2370 rdev->badblocks.changed = 0;
2371 md_ack_all_badblocks(&rdev->badblocks);
2372 md_error(mddev, rdev);
2374 clear_bit(Blocked, &rdev->flags);
2375 clear_bit(BlockedBadBlocks, &rdev->flags);
2376 wake_up(&rdev->blocked_wait);
2379 wake_up(&mddev->sb_wait);
2383 spin_lock_irq(&mddev->write_lock);
2385 mddev->utime = get_seconds();
2387 if (test_and_clear_bit(MD_CHANGE_DEVS, &mddev->flags))
2389 if (test_and_clear_bit(MD_CHANGE_CLEAN, &mddev->flags))
2390 /* just a clean<-> dirty transition, possibly leave spares alone,
2391 * though if events isn't the right even/odd, we will have to do
2397 if (mddev->degraded)
2398 /* If the array is degraded, then skipping spares is both
2399 * dangerous and fairly pointless.
2400 * Dangerous because a device that was removed from the array
2401 * might have a event_count that still looks up-to-date,
2402 * so it can be re-added without a resync.
2403 * Pointless because if there are any spares to skip,
2404 * then a recovery will happen and soon that array won't
2405 * be degraded any more and the spare can go back to sleep then.
2409 sync_req = mddev->in_sync;
2411 /* If this is just a dirty<->clean transition, and the array is clean
2412 * and 'events' is odd, we can roll back to the previous clean state */
2414 && (mddev->in_sync && mddev->recovery_cp == MaxSector)
2415 && mddev->can_decrease_events
2416 && mddev->events != 1) {
2418 mddev->can_decrease_events = 0;
2420 /* otherwise we have to go forward and ... */
2422 mddev->can_decrease_events = nospares;
2425 if (!mddev->events) {
2427 * oops, this 64-bit counter should never wrap.
2428 * Either we are in around ~1 trillion A.C., assuming
2429 * 1 reboot per second, or we have a bug:
2435 list_for_each_entry(rdev, &mddev->disks, same_set) {
2436 if (rdev->badblocks.changed)
2437 any_badblocks_changed++;
2438 if (test_bit(Faulty, &rdev->flags))
2439 set_bit(FaultRecorded, &rdev->flags);
2442 sync_sbs(mddev, nospares);
2443 spin_unlock_irq(&mddev->write_lock);
2445 pr_debug("md: updating %s RAID superblock on device (in sync %d)\n",
2446 mdname(mddev), mddev->in_sync);
2448 bitmap_update_sb(mddev->bitmap);
2449 list_for_each_entry(rdev, &mddev->disks, same_set) {
2450 char b[BDEVNAME_SIZE];
2452 if (rdev->sb_loaded != 1)
2453 continue; /* no noise on spare devices */
2455 if (!test_bit(Faulty, &rdev->flags) &&
2456 rdev->saved_raid_disk == -1) {
2457 md_super_write(mddev,rdev,
2458 rdev->sb_start, rdev->sb_size,
2460 pr_debug("md: (write) %s's sb offset: %llu\n",
2461 bdevname(rdev->bdev, b),
2462 (unsigned long long)rdev->sb_start);
2463 rdev->sb_events = mddev->events;
2464 if (rdev->badblocks.size) {
2465 md_super_write(mddev, rdev,
2466 rdev->badblocks.sector,
2467 rdev->badblocks.size << 9,
2469 rdev->badblocks.size = 0;
2472 } else if (test_bit(Faulty, &rdev->flags))
2473 pr_debug("md: %s (skipping faulty)\n",
2474 bdevname(rdev->bdev, b));
2476 pr_debug("(skipping incremental s/r ");
2478 if (mddev->level == LEVEL_MULTIPATH)
2479 /* only need to write one superblock... */
2482 md_super_wait(mddev);
2483 /* if there was a failure, MD_CHANGE_DEVS was set, and we re-write super */
2485 spin_lock_irq(&mddev->write_lock);
2486 if (mddev->in_sync != sync_req ||
2487 test_bit(MD_CHANGE_DEVS, &mddev->flags)) {
2488 /* have to write it out again */
2489 spin_unlock_irq(&mddev->write_lock);
2492 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
2493 spin_unlock_irq(&mddev->write_lock);
2494 wake_up(&mddev->sb_wait);
2495 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
2496 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
2498 list_for_each_entry(rdev, &mddev->disks, same_set) {
2499 if (test_and_clear_bit(FaultRecorded, &rdev->flags))
2500 clear_bit(Blocked, &rdev->flags);
2502 if (any_badblocks_changed)
2503 md_ack_all_badblocks(&rdev->badblocks);
2504 clear_bit(BlockedBadBlocks, &rdev->flags);
2505 wake_up(&rdev->blocked_wait);
2509 /* words written to sysfs files may, or may not, be \n terminated.
2510 * We want to accept with case. For this we use cmd_match.
2512 static int cmd_match(const char *cmd, const char *str)
2514 /* See if cmd, written into a sysfs file, matches
2515 * str. They must either be the same, or cmd can
2516 * have a trailing newline
2518 while (*cmd && *str && *cmd == *str) {
2529 struct rdev_sysfs_entry {
2530 struct attribute attr;
2531 ssize_t (*show)(struct md_rdev *, char *);
2532 ssize_t (*store)(struct md_rdev *, const char *, size_t);
2536 state_show(struct md_rdev *rdev, char *page)
2541 if (test_bit(Faulty, &rdev->flags) ||
2542 rdev->badblocks.unacked_exist) {
2543 len+= sprintf(page+len, "%sfaulty",sep);
2546 if (test_bit(In_sync, &rdev->flags)) {
2547 len += sprintf(page+len, "%sin_sync",sep);
2550 if (test_bit(WriteMostly, &rdev->flags)) {
2551 len += sprintf(page+len, "%swrite_mostly",sep);
2554 if (test_bit(Blocked, &rdev->flags) ||
2555 (rdev->badblocks.unacked_exist
2556 && !test_bit(Faulty, &rdev->flags))) {
2557 len += sprintf(page+len, "%sblocked", sep);
2560 if (!test_bit(Faulty, &rdev->flags) &&
2561 !test_bit(In_sync, &rdev->flags)) {
2562 len += sprintf(page+len, "%sspare", sep);
2565 if (test_bit(WriteErrorSeen, &rdev->flags)) {
2566 len += sprintf(page+len, "%swrite_error", sep);
2569 return len+sprintf(page+len, "\n");
2573 state_store(struct md_rdev *rdev, const char *buf, size_t len)
2576 * faulty - simulates an error
2577 * remove - disconnects the device
2578 * writemostly - sets write_mostly
2579 * -writemostly - clears write_mostly
2580 * blocked - sets the Blocked flags
2581 * -blocked - clears the Blocked and possibly simulates an error
2582 * insync - sets Insync providing device isn't active
2583 * write_error - sets WriteErrorSeen
2584 * -write_error - clears WriteErrorSeen
2587 if (cmd_match(buf, "faulty") && rdev->mddev->pers) {
2588 md_error(rdev->mddev, rdev);
2589 if (test_bit(Faulty, &rdev->flags))
2593 } else if (cmd_match(buf, "remove")) {
2594 if (rdev->raid_disk >= 0)
2597 struct mddev *mddev = rdev->mddev;
2598 kick_rdev_from_array(rdev);
2600 md_update_sb(mddev, 1);
2601 md_new_event(mddev);
2604 } else if (cmd_match(buf, "writemostly")) {
2605 set_bit(WriteMostly, &rdev->flags);
2607 } else if (cmd_match(buf, "-writemostly")) {
2608 clear_bit(WriteMostly, &rdev->flags);
2610 } else if (cmd_match(buf, "blocked")) {
2611 set_bit(Blocked, &rdev->flags);
2613 } else if (cmd_match(buf, "-blocked")) {
2614 if (!test_bit(Faulty, &rdev->flags) &&
2615 rdev->badblocks.unacked_exist) {
2616 /* metadata handler doesn't understand badblocks,
2617 * so we need to fail the device
2619 md_error(rdev->mddev, rdev);
2621 clear_bit(Blocked, &rdev->flags);
2622 clear_bit(BlockedBadBlocks, &rdev->flags);
2623 wake_up(&rdev->blocked_wait);
2624 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2625 md_wakeup_thread(rdev->mddev->thread);
2628 } else if (cmd_match(buf, "insync") && rdev->raid_disk == -1) {
2629 set_bit(In_sync, &rdev->flags);
2631 } else if (cmd_match(buf, "write_error")) {
2632 set_bit(WriteErrorSeen, &rdev->flags);
2634 } else if (cmd_match(buf, "-write_error")) {
2635 clear_bit(WriteErrorSeen, &rdev->flags);
2639 sysfs_notify_dirent_safe(rdev->sysfs_state);
2640 return err ? err : len;
2642 static struct rdev_sysfs_entry rdev_state =
2643 __ATTR(state, S_IRUGO|S_IWUSR, state_show, state_store);
2646 errors_show(struct md_rdev *rdev, char *page)
2648 return sprintf(page, "%d\n", atomic_read(&rdev->corrected_errors));
2652 errors_store(struct md_rdev *rdev, const char *buf, size_t len)
2655 unsigned long n = simple_strtoul(buf, &e, 10);
2656 if (*buf && (*e == 0 || *e == '\n')) {
2657 atomic_set(&rdev->corrected_errors, n);
2662 static struct rdev_sysfs_entry rdev_errors =
2663 __ATTR(errors, S_IRUGO|S_IWUSR, errors_show, errors_store);
2666 slot_show(struct md_rdev *rdev, char *page)
2668 if (rdev->raid_disk < 0)
2669 return sprintf(page, "none\n");
2671 return sprintf(page, "%d\n", rdev->raid_disk);
2675 slot_store(struct md_rdev *rdev, const char *buf, size_t len)
2679 int slot = simple_strtoul(buf, &e, 10);
2680 if (strncmp(buf, "none", 4)==0)
2682 else if (e==buf || (*e && *e!= '\n'))
2684 if (rdev->mddev->pers && slot == -1) {
2685 /* Setting 'slot' on an active array requires also
2686 * updating the 'rd%d' link, and communicating
2687 * with the personality with ->hot_*_disk.
2688 * For now we only support removing
2689 * failed/spare devices. This normally happens automatically,
2690 * but not when the metadata is externally managed.
2692 if (rdev->raid_disk == -1)
2694 /* personality does all needed checks */
2695 if (rdev->mddev->pers->hot_remove_disk == NULL)
2697 err = rdev->mddev->pers->
2698 hot_remove_disk(rdev->mddev, rdev->raid_disk);
2701 sysfs_unlink_rdev(rdev->mddev, rdev);
2702 rdev->raid_disk = -1;
2703 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2704 md_wakeup_thread(rdev->mddev->thread);
2705 } else if (rdev->mddev->pers) {
2706 struct md_rdev *rdev2;
2707 /* Activating a spare .. or possibly reactivating
2708 * if we ever get bitmaps working here.
2711 if (rdev->raid_disk != -1)
2714 if (test_bit(MD_RECOVERY_RUNNING, &rdev->mddev->recovery))
2717 if (rdev->mddev->pers->hot_add_disk == NULL)
2720 list_for_each_entry(rdev2, &rdev->mddev->disks, same_set)
2721 if (rdev2->raid_disk == slot)
2724 if (slot >= rdev->mddev->raid_disks &&
2725 slot >= rdev->mddev->raid_disks + rdev->mddev->delta_disks)
2728 rdev->raid_disk = slot;
2729 if (test_bit(In_sync, &rdev->flags))
2730 rdev->saved_raid_disk = slot;
2732 rdev->saved_raid_disk = -1;
2733 clear_bit(In_sync, &rdev->flags);
2734 err = rdev->mddev->pers->
2735 hot_add_disk(rdev->mddev, rdev);
2737 rdev->raid_disk = -1;
2740 sysfs_notify_dirent_safe(rdev->sysfs_state);
2741 if (sysfs_link_rdev(rdev->mddev, rdev))
2742 /* failure here is OK */;
2743 /* don't wakeup anyone, leave that to userspace. */
2745 if (slot >= rdev->mddev->raid_disks &&
2746 slot >= rdev->mddev->raid_disks + rdev->mddev->delta_disks)
2748 rdev->raid_disk = slot;
2749 /* assume it is working */
2750 clear_bit(Faulty, &rdev->flags);
2751 clear_bit(WriteMostly, &rdev->flags);
2752 set_bit(In_sync, &rdev->flags);
2753 sysfs_notify_dirent_safe(rdev->sysfs_state);
2759 static struct rdev_sysfs_entry rdev_slot =
2760 __ATTR(slot, S_IRUGO|S_IWUSR, slot_show, slot_store);
2763 offset_show(struct md_rdev *rdev, char *page)
2765 return sprintf(page, "%llu\n", (unsigned long long)rdev->data_offset);
2769 offset_store(struct md_rdev *rdev, const char *buf, size_t len)
2772 unsigned long long offset = simple_strtoull(buf, &e, 10);
2773 if (e==buf || (*e && *e != '\n'))
2775 if (rdev->mddev->pers && rdev->raid_disk >= 0)
2777 if (rdev->sectors && rdev->mddev->external)
2778 /* Must set offset before size, so overlap checks
2781 rdev->data_offset = offset;
2785 static struct rdev_sysfs_entry rdev_offset =
2786 __ATTR(offset, S_IRUGO|S_IWUSR, offset_show, offset_store);
2789 rdev_size_show(struct md_rdev *rdev, char *page)
2791 return sprintf(page, "%llu\n", (unsigned long long)rdev->sectors / 2);
2794 static int overlaps(sector_t s1, sector_t l1, sector_t s2, sector_t l2)
2796 /* check if two start/length pairs overlap */
2804 static int strict_blocks_to_sectors(const char *buf, sector_t *sectors)
2806 unsigned long long blocks;
2809 if (strict_strtoull(buf, 10, &blocks) < 0)
2812 if (blocks & 1ULL << (8 * sizeof(blocks) - 1))
2813 return -EINVAL; /* sector conversion overflow */
2816 if (new != blocks * 2)
2817 return -EINVAL; /* unsigned long long to sector_t overflow */
2824 rdev_size_store(struct md_rdev *rdev, const char *buf, size_t len)
2826 struct mddev *my_mddev = rdev->mddev;
2827 sector_t oldsectors = rdev->sectors;
2830 if (strict_blocks_to_sectors(buf, §ors) < 0)
2832 if (my_mddev->pers && rdev->raid_disk >= 0) {
2833 if (my_mddev->persistent) {
2834 sectors = super_types[my_mddev->major_version].
2835 rdev_size_change(rdev, sectors);
2838 } else if (!sectors)
2839 sectors = (i_size_read(rdev->bdev->bd_inode) >> 9) -
2842 if (sectors < my_mddev->dev_sectors)
2843 return -EINVAL; /* component must fit device */
2845 rdev->sectors = sectors;
2846 if (sectors > oldsectors && my_mddev->external) {
2847 /* need to check that all other rdevs with the same ->bdev
2848 * do not overlap. We need to unlock the mddev to avoid
2849 * a deadlock. We have already changed rdev->sectors, and if
2850 * we have to change it back, we will have the lock again.
2852 struct mddev *mddev;
2854 struct list_head *tmp;
2856 mddev_unlock(my_mddev);
2857 for_each_mddev(mddev, tmp) {
2858 struct md_rdev *rdev2;
2861 list_for_each_entry(rdev2, &mddev->disks, same_set)
2862 if (rdev->bdev == rdev2->bdev &&
2864 overlaps(rdev->data_offset, rdev->sectors,
2870 mddev_unlock(mddev);
2876 mddev_lock(my_mddev);
2878 /* Someone else could have slipped in a size
2879 * change here, but doing so is just silly.
2880 * We put oldsectors back because we *know* it is
2881 * safe, and trust userspace not to race with
2884 rdev->sectors = oldsectors;
2891 static struct rdev_sysfs_entry rdev_size =
2892 __ATTR(size, S_IRUGO|S_IWUSR, rdev_size_show, rdev_size_store);
2895 static ssize_t recovery_start_show(struct md_rdev *rdev, char *page)
2897 unsigned long long recovery_start = rdev->recovery_offset;
2899 if (test_bit(In_sync, &rdev->flags) ||
2900 recovery_start == MaxSector)
2901 return sprintf(page, "none\n");
2903 return sprintf(page, "%llu\n", recovery_start);
2906 static ssize_t recovery_start_store(struct md_rdev *rdev, const char *buf, size_t len)
2908 unsigned long long recovery_start;
2910 if (cmd_match(buf, "none"))
2911 recovery_start = MaxSector;
2912 else if (strict_strtoull(buf, 10, &recovery_start))
2915 if (rdev->mddev->pers &&
2916 rdev->raid_disk >= 0)
2919 rdev->recovery_offset = recovery_start;
2920 if (recovery_start == MaxSector)
2921 set_bit(In_sync, &rdev->flags);
2923 clear_bit(In_sync, &rdev->flags);
2927 static struct rdev_sysfs_entry rdev_recovery_start =
2928 __ATTR(recovery_start, S_IRUGO|S_IWUSR, recovery_start_show, recovery_start_store);
2932 badblocks_show(struct badblocks *bb, char *page, int unack);
2934 badblocks_store(struct badblocks *bb, const char *page, size_t len, int unack);
2936 static ssize_t bb_show(struct md_rdev *rdev, char *page)
2938 return badblocks_show(&rdev->badblocks, page, 0);
2940 static ssize_t bb_store(struct md_rdev *rdev, const char *page, size_t len)
2942 int rv = badblocks_store(&rdev->badblocks, page, len, 0);
2943 /* Maybe that ack was all we needed */
2944 if (test_and_clear_bit(BlockedBadBlocks, &rdev->flags))
2945 wake_up(&rdev->blocked_wait);
2948 static struct rdev_sysfs_entry rdev_bad_blocks =
2949 __ATTR(bad_blocks, S_IRUGO|S_IWUSR, bb_show, bb_store);
2952 static ssize_t ubb_show(struct md_rdev *rdev, char *page)
2954 return badblocks_show(&rdev->badblocks, page, 1);
2956 static ssize_t ubb_store(struct md_rdev *rdev, const char *page, size_t len)
2958 return badblocks_store(&rdev->badblocks, page, len, 1);
2960 static struct rdev_sysfs_entry rdev_unack_bad_blocks =
2961 __ATTR(unacknowledged_bad_blocks, S_IRUGO|S_IWUSR, ubb_show, ubb_store);
2963 static struct attribute *rdev_default_attrs[] = {
2969 &rdev_recovery_start.attr,
2970 &rdev_bad_blocks.attr,
2971 &rdev_unack_bad_blocks.attr,
2975 rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
2977 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
2978 struct md_rdev *rdev = container_of(kobj, struct md_rdev, kobj);
2979 struct mddev *mddev = rdev->mddev;
2985 rv = mddev ? mddev_lock(mddev) : -EBUSY;
2987 if (rdev->mddev == NULL)
2990 rv = entry->show(rdev, page);
2991 mddev_unlock(mddev);
2997 rdev_attr_store(struct kobject *kobj, struct attribute *attr,
2998 const char *page, size_t length)
3000 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
3001 struct md_rdev *rdev = container_of(kobj, struct md_rdev, kobj);
3003 struct mddev *mddev = rdev->mddev;
3007 if (!capable(CAP_SYS_ADMIN))
3009 rv = mddev ? mddev_lock(mddev): -EBUSY;
3011 if (rdev->mddev == NULL)
3014 rv = entry->store(rdev, page, length);
3015 mddev_unlock(mddev);
3020 static void rdev_free(struct kobject *ko)
3022 struct md_rdev *rdev = container_of(ko, struct md_rdev, kobj);
3025 static const struct sysfs_ops rdev_sysfs_ops = {
3026 .show = rdev_attr_show,
3027 .store = rdev_attr_store,
3029 static struct kobj_type rdev_ktype = {
3030 .release = rdev_free,
3031 .sysfs_ops = &rdev_sysfs_ops,
3032 .default_attrs = rdev_default_attrs,
3035 int md_rdev_init(struct md_rdev *rdev)
3038 rdev->saved_raid_disk = -1;
3039 rdev->raid_disk = -1;
3041 rdev->data_offset = 0;
3042 rdev->sb_events = 0;
3043 rdev->last_read_error.tv_sec = 0;
3044 rdev->last_read_error.tv_nsec = 0;
3045 rdev->sb_loaded = 0;
3046 rdev->bb_page = NULL;
3047 atomic_set(&rdev->nr_pending, 0);
3048 atomic_set(&rdev->read_errors, 0);
3049 atomic_set(&rdev->corrected_errors, 0);
3051 INIT_LIST_HEAD(&rdev->same_set);
3052 init_waitqueue_head(&rdev->blocked_wait);
3054 /* Add space to store bad block list.
3055 * This reserves the space even on arrays where it cannot
3056 * be used - I wonder if that matters
3058 rdev->badblocks.count = 0;
3059 rdev->badblocks.shift = 0;
3060 rdev->badblocks.page = kmalloc(PAGE_SIZE, GFP_KERNEL);
3061 seqlock_init(&rdev->badblocks.lock);
3062 if (rdev->badblocks.page == NULL)
3067 EXPORT_SYMBOL_GPL(md_rdev_init);
3069 * Import a device. If 'super_format' >= 0, then sanity check the superblock
3071 * mark the device faulty if:
3073 * - the device is nonexistent (zero size)
3074 * - the device has no valid superblock
3076 * a faulty rdev _never_ has rdev->sb set.
3078 static struct md_rdev *md_import_device(dev_t newdev, int super_format, int super_minor)
3080 char b[BDEVNAME_SIZE];
3082 struct md_rdev *rdev;
3085 rdev = kzalloc(sizeof(*rdev), GFP_KERNEL);
3087 printk(KERN_ERR "md: could not alloc mem for new device!\n");
3088 return ERR_PTR(-ENOMEM);
3091 err = md_rdev_init(rdev);
3094 err = alloc_disk_sb(rdev);
3098 err = lock_rdev(rdev, newdev, super_format == -2);
3102 kobject_init(&rdev->kobj, &rdev_ktype);
3104 size = i_size_read(rdev->bdev->bd_inode) >> BLOCK_SIZE_BITS;
3107 "md: %s has zero or unknown size, marking faulty!\n",
3108 bdevname(rdev->bdev,b));
3113 if (super_format >= 0) {
3114 err = super_types[super_format].
3115 load_super(rdev, NULL, super_minor);
3116 if (err == -EINVAL) {
3118 "md: %s does not have a valid v%d.%d "
3119 "superblock, not importing!\n",
3120 bdevname(rdev->bdev,b),
3121 super_format, super_minor);
3126 "md: could not read %s's sb, not importing!\n",
3127 bdevname(rdev->bdev,b));
3131 if (super_format == -1)
3132 /* hot-add for 0.90, or non-persistent: so no badblocks */
3133 rdev->badblocks.shift = -1;
3141 kfree(rdev->badblocks.page);
3143 return ERR_PTR(err);
3147 * Check a full RAID array for plausibility
3151 static void analyze_sbs(struct mddev * mddev)
3154 struct md_rdev *rdev, *freshest, *tmp;
3155 char b[BDEVNAME_SIZE];
3158 rdev_for_each(rdev, tmp, mddev)
3159 switch (super_types[mddev->major_version].
3160 load_super(rdev, freshest, mddev->minor_version)) {
3168 "md: fatal superblock inconsistency in %s"
3169 " -- removing from array\n",
3170 bdevname(rdev->bdev,b));
3171 kick_rdev_from_array(rdev);
3175 super_types[mddev->major_version].
3176 validate_super(mddev, freshest);
3179 rdev_for_each(rdev, tmp, mddev) {
3180 if (mddev->max_disks &&
3181 (rdev->desc_nr >= mddev->max_disks ||
3182 i > mddev->max_disks)) {
3184 "md: %s: %s: only %d devices permitted\n",
3185 mdname(mddev), bdevname(rdev->bdev, b),
3187 kick_rdev_from_array(rdev);
3190 if (rdev != freshest)
3191 if (super_types[mddev->major_version].
3192 validate_super(mddev, rdev)) {
3193 printk(KERN_WARNING "md: kicking non-fresh %s"
3195 bdevname(rdev->bdev,b));
3196 kick_rdev_from_array(rdev);
3199 if (mddev->level == LEVEL_MULTIPATH) {
3200 rdev->desc_nr = i++;
3201 rdev->raid_disk = rdev->desc_nr;
3202 set_bit(In_sync, &rdev->flags);
3203 } else if (rdev->raid_disk >= (mddev->raid_disks - min(0, mddev->delta_disks))) {
3204 rdev->raid_disk = -1;
3205 clear_bit(In_sync, &rdev->flags);
3210 /* Read a fixed-point number.
3211 * Numbers in sysfs attributes should be in "standard" units where
3212 * possible, so time should be in seconds.
3213 * However we internally use a a much smaller unit such as
3214 * milliseconds or jiffies.
3215 * This function takes a decimal number with a possible fractional
3216 * component, and produces an integer which is the result of
3217 * multiplying that number by 10^'scale'.
3218 * all without any floating-point arithmetic.
3220 int strict_strtoul_scaled(const char *cp, unsigned long *res, int scale)
3222 unsigned long result = 0;
3224 while (isdigit(*cp) || (*cp == '.' && decimals < 0)) {
3227 else if (decimals < scale) {
3230 result = result * 10 + value;
3242 while (decimals < scale) {
3251 static void md_safemode_timeout(unsigned long data);
3254 safe_delay_show(struct mddev *mddev, char *page)
3256 int msec = (mddev->safemode_delay*1000)/HZ;
3257 return sprintf(page, "%d.%03d\n", msec/1000, msec%1000);
3260 safe_delay_store(struct mddev *mddev, const char *cbuf, size_t len)
3264 if (strict_strtoul_scaled(cbuf, &msec, 3) < 0)
3267 mddev->safemode_delay = 0;
3269 unsigned long old_delay = mddev->safemode_delay;
3270 mddev->safemode_delay = (msec*HZ)/1000;
3271 if (mddev->safemode_delay == 0)
3272 mddev->safemode_delay = 1;
3273 if (mddev->safemode_delay < old_delay)
3274 md_safemode_timeout((unsigned long)mddev);
3278 static struct md_sysfs_entry md_safe_delay =
3279 __ATTR(safe_mode_delay, S_IRUGO|S_IWUSR,safe_delay_show, safe_delay_store);
3282 level_show(struct mddev *mddev, char *page)
3284 struct md_personality *p = mddev->pers;
3286 return sprintf(page, "%s\n", p->name);
3287 else if (mddev->clevel[0])
3288 return sprintf(page, "%s\n", mddev->clevel);
3289 else if (mddev->level != LEVEL_NONE)
3290 return sprintf(page, "%d\n", mddev->level);
3296 level_store(struct mddev *mddev, const char *buf, size_t len)
3300 struct md_personality *pers;
3303 struct md_rdev *rdev;
3305 if (mddev->pers == NULL) {
3308 if (len >= sizeof(mddev->clevel))
3310 strncpy(mddev->clevel, buf, len);
3311 if (mddev->clevel[len-1] == '\n')
3313 mddev->clevel[len] = 0;
3314 mddev->level = LEVEL_NONE;
3318 /* request to change the personality. Need to ensure:
3319 * - array is not engaged in resync/recovery/reshape
3320 * - old personality can be suspended
3321 * - new personality will access other array.
3324 if (mddev->sync_thread ||
3325 mddev->reshape_position != MaxSector ||
3326 mddev->sysfs_active)
3329 if (!mddev->pers->quiesce) {
3330 printk(KERN_WARNING "md: %s: %s does not support online personality change\n",
3331 mdname(mddev), mddev->pers->name);
3335 /* Now find the new personality */
3336 if (len == 0 || len >= sizeof(clevel))
3338 strncpy(clevel, buf, len);
3339 if (clevel[len-1] == '\n')
3342 if (strict_strtol(clevel, 10, &level))
3345 if (request_module("md-%s", clevel) != 0)
3346 request_module("md-level-%s", clevel);
3347 spin_lock(&pers_lock);
3348 pers = find_pers(level, clevel);
3349 if (!pers || !try_module_get(pers->owner)) {
3350 spin_unlock(&pers_lock);
3351 printk(KERN_WARNING "md: personality %s not loaded\n", clevel);
3354 spin_unlock(&pers_lock);
3356 if (pers == mddev->pers) {
3357 /* Nothing to do! */
3358 module_put(pers->owner);
3361 if (!pers->takeover) {
3362 module_put(pers->owner);
3363 printk(KERN_WARNING "md: %s: %s does not support personality takeover\n",
3364 mdname(mddev), clevel);
3368 list_for_each_entry(rdev, &mddev->disks, same_set)
3369 rdev->new_raid_disk = rdev->raid_disk;
3371 /* ->takeover must set new_* and/or delta_disks
3372 * if it succeeds, and may set them when it fails.
3374 priv = pers->takeover(mddev);
3376 mddev->new_level = mddev->level;
3377 mddev->new_layout = mddev->layout;
3378 mddev->new_chunk_sectors = mddev->chunk_sectors;
3379 mddev->raid_disks -= mddev->delta_disks;
3380 mddev->delta_disks = 0;
3381 module_put(pers->owner);
3382 printk(KERN_WARNING "md: %s: %s would not accept array\n",
3383 mdname(mddev), clevel);
3384 return PTR_ERR(priv);
3387 /* Looks like we have a winner */
3388 mddev_suspend(mddev);
3389 mddev->pers->stop(mddev);
3391 if (mddev->pers->sync_request == NULL &&
3392 pers->sync_request != NULL) {
3393 /* need to add the md_redundancy_group */
3394 if (sysfs_create_group(&mddev->kobj, &md_redundancy_group))
3396 "md: cannot register extra attributes for %s\n",
3398 mddev->sysfs_action = sysfs_get_dirent(mddev->kobj.sd, NULL, "sync_action");
3400 if (mddev->pers->sync_request != NULL &&
3401 pers->sync_request == NULL) {
3402 /* need to remove the md_redundancy_group */
3403 if (mddev->to_remove == NULL)
3404 mddev->to_remove = &md_redundancy_group;
3407 if (mddev->pers->sync_request == NULL &&
3409 /* We are converting from a no-redundancy array
3410 * to a redundancy array and metadata is managed
3411 * externally so we need to be sure that writes
3412 * won't block due to a need to transition
3414 * until external management is started.
3417 mddev->safemode_delay = 0;
3418 mddev->safemode = 0;
3421 list_for_each_entry(rdev, &mddev->disks, same_set) {
3422 if (rdev->raid_disk < 0)
3424 if (rdev->new_raid_disk >= mddev->raid_disks)
3425 rdev->new_raid_disk = -1;
3426 if (rdev->new_raid_disk == rdev->raid_disk)
3428 sysfs_unlink_rdev(mddev, rdev);
3430 list_for_each_entry(rdev, &mddev->disks, same_set) {
3431 if (rdev->raid_disk < 0)
3433 if (rdev->new_raid_disk == rdev->raid_disk)
3435 rdev->raid_disk = rdev->new_raid_disk;
3436 if (rdev->raid_disk < 0)
3437 clear_bit(In_sync, &rdev->flags);
3439 if (sysfs_link_rdev(mddev, rdev))
3440 printk(KERN_WARNING "md: cannot register rd%d"
3441 " for %s after level change\n",
3442 rdev->raid_disk, mdname(mddev));
3446 module_put(mddev->pers->owner);
3448 mddev->private = priv;
3449 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
3450 mddev->level = mddev->new_level;
3451 mddev->layout = mddev->new_layout;
3452 mddev->chunk_sectors = mddev->new_chunk_sectors;
3453 mddev->delta_disks = 0;
3454 mddev->degraded = 0;
3455 if (mddev->pers->sync_request == NULL) {
3456 /* this is now an array without redundancy, so
3457 * it must always be in_sync
3460 del_timer_sync(&mddev->safemode_timer);
3463 mddev_resume(mddev);
3464 set_bit(MD_CHANGE_DEVS, &mddev->flags);
3465 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3466 md_wakeup_thread(mddev->thread);
3467 sysfs_notify(&mddev->kobj, NULL, "level");
3468 md_new_event(mddev);
3472 static struct md_sysfs_entry md_level =
3473 __ATTR(level, S_IRUGO|S_IWUSR, level_show, level_store);
3477 layout_show(struct mddev *mddev, char *page)
3479 /* just a number, not meaningful for all levels */
3480 if (mddev->reshape_position != MaxSector &&
3481 mddev->layout != mddev->new_layout)
3482 return sprintf(page, "%d (%d)\n",
3483 mddev->new_layout, mddev->layout);
3484 return sprintf(page, "%d\n", mddev->layout);
3488 layout_store(struct mddev *mddev, const char *buf, size_t len)
3491 unsigned long n = simple_strtoul(buf, &e, 10);
3493 if (!*buf || (*e && *e != '\n'))
3498 if (mddev->pers->check_reshape == NULL)
3500 mddev->new_layout = n;
3501 err = mddev->pers->check_reshape(mddev);
3503 mddev->new_layout = mddev->layout;
3507 mddev->new_layout = n;
3508 if (mddev->reshape_position == MaxSector)
3513 static struct md_sysfs_entry md_layout =
3514 __ATTR(layout, S_IRUGO|S_IWUSR, layout_show, layout_store);
3518 raid_disks_show(struct mddev *mddev, char *page)
3520 if (mddev->raid_disks == 0)
3522 if (mddev->reshape_position != MaxSector &&
3523 mddev->delta_disks != 0)
3524 return sprintf(page, "%d (%d)\n", mddev->raid_disks,
3525 mddev->raid_disks - mddev->delta_disks);
3526 return sprintf(page, "%d\n", mddev->raid_disks);
3529 static int update_raid_disks(struct mddev *mddev, int raid_disks);
3532 raid_disks_store(struct mddev *mddev, const char *buf, size_t len)
3536 unsigned long n = simple_strtoul(buf, &e, 10);
3538 if (!*buf || (*e && *e != '\n'))
3542 rv = update_raid_disks(mddev, n);
3543 else if (mddev->reshape_position != MaxSector) {
3544 int olddisks = mddev->raid_disks - mddev->delta_disks;
3545 mddev->delta_disks = n - olddisks;
3546 mddev->raid_disks = n;
3548 mddev->raid_disks = n;
3549 return rv ? rv : len;
3551 static struct md_sysfs_entry md_raid_disks =
3552 __ATTR(raid_disks, S_IRUGO|S_IWUSR, raid_disks_show, raid_disks_store);
3555 chunk_size_show(struct mddev *mddev, char *page)
3557 if (mddev->reshape_position != MaxSector &&
3558 mddev->chunk_sectors != mddev->new_chunk_sectors)
3559 return sprintf(page, "%d (%d)\n",
3560 mddev->new_chunk_sectors << 9,
3561 mddev->chunk_sectors << 9);
3562 return sprintf(page, "%d\n", mddev->chunk_sectors << 9);
3566 chunk_size_store(struct mddev *mddev, const char *buf, size_t len)
3569 unsigned long n = simple_strtoul(buf, &e, 10);
3571 if (!*buf || (*e && *e != '\n'))
3576 if (mddev->pers->check_reshape == NULL)
3578 mddev->new_chunk_sectors = n >> 9;
3579 err = mddev->pers->check_reshape(mddev);
3581 mddev->new_chunk_sectors = mddev->chunk_sectors;
3585 mddev->new_chunk_sectors = n >> 9;
3586 if (mddev->reshape_position == MaxSector)
3587 mddev->chunk_sectors = n >> 9;
3591 static struct md_sysfs_entry md_chunk_size =
3592 __ATTR(chunk_size, S_IRUGO|S_IWUSR, chunk_size_show, chunk_size_store);
3595 resync_start_show(struct mddev *mddev, char *page)
3597 if (mddev->recovery_cp == MaxSector)
3598 return sprintf(page, "none\n");
3599 return sprintf(page, "%llu\n", (unsigned long long)mddev->recovery_cp);
3603 resync_start_store(struct mddev *mddev, const char *buf, size_t len)
3606 unsigned long long n = simple_strtoull(buf, &e, 10);
3608 if (mddev->pers && !test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
3610 if (cmd_match(buf, "none"))
3612 else if (!*buf || (*e && *e != '\n'))
3615 mddev->recovery_cp = n;
3618 static struct md_sysfs_entry md_resync_start =
3619 __ATTR(resync_start, S_IRUGO|S_IWUSR, resync_start_show, resync_start_store);
3622 * The array state can be:
3625 * No devices, no size, no level
3626 * Equivalent to STOP_ARRAY ioctl
3628 * May have some settings, but array is not active
3629 * all IO results in error
3630 * When written, doesn't tear down array, but just stops it
3631 * suspended (not supported yet)
3632 * All IO requests will block. The array can be reconfigured.
3633 * Writing this, if accepted, will block until array is quiescent
3635 * no resync can happen. no superblocks get written.
3636 * write requests fail
3638 * like readonly, but behaves like 'clean' on a write request.
3640 * clean - no pending writes, but otherwise active.
3641 * When written to inactive array, starts without resync
3642 * If a write request arrives then
3643 * if metadata is known, mark 'dirty' and switch to 'active'.
3644 * if not known, block and switch to write-pending
3645 * If written to an active array that has pending writes, then fails.
3647 * fully active: IO and resync can be happening.
3648 * When written to inactive array, starts with resync
3651 * clean, but writes are blocked waiting for 'active' to be written.
3654 * like active, but no writes have been seen for a while (100msec).
3657 enum array_state { clear, inactive, suspended, readonly, read_auto, clean, active,
3658 write_pending, active_idle, bad_word};
3659 static char *array_states[] = {
3660 "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active",
3661 "write-pending", "active-idle", NULL };
3663 static int match_word(const char *word, char **list)
3666 for (n=0; list[n]; n++)
3667 if (cmd_match(word, list[n]))
3673 array_state_show(struct mddev *mddev, char *page)
3675 enum array_state st = inactive;
3688 else if (test_bit(MD_CHANGE_PENDING, &mddev->flags))
3690 else if (mddev->safemode)
3696 if (list_empty(&mddev->disks) &&
3697 mddev->raid_disks == 0 &&
3698 mddev->dev_sectors == 0)
3703 return sprintf(page, "%s\n", array_states[st]);
3706 static int do_md_stop(struct mddev * mddev, int ro, struct block_device *bdev);
3707 static int md_set_readonly(struct mddev * mddev, struct block_device *bdev);
3708 static int do_md_run(struct mddev * mddev);
3709 static int restart_array(struct mddev *mddev);
3712 array_state_store(struct mddev *mddev, const char *buf, size_t len)
3715 enum array_state st = match_word(buf, array_states);
3720 /* stopping an active array */
3721 if (atomic_read(&mddev->openers) > 0)
3723 err = do_md_stop(mddev, 0, NULL);
3726 /* stopping an active array */
3728 if (atomic_read(&mddev->openers) > 0)
3730 err = do_md_stop(mddev, 2, NULL);
3732 err = 0; /* already inactive */
3735 break; /* not supported yet */
3738 err = md_set_readonly(mddev, NULL);
3741 set_disk_ro(mddev->gendisk, 1);
3742 err = do_md_run(mddev);
3748 err = md_set_readonly(mddev, NULL);
3749 else if (mddev->ro == 1)
3750 err = restart_array(mddev);
3753 set_disk_ro(mddev->gendisk, 0);
3757 err = do_md_run(mddev);
3762 restart_array(mddev);
3763 spin_lock_irq(&mddev->write_lock);
3764 if (atomic_read(&mddev->writes_pending) == 0) {
3765 if (mddev->in_sync == 0) {
3767 if (mddev->safemode == 1)
3768 mddev->safemode = 0;
3769 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
3774 spin_unlock_irq(&mddev->write_lock);
3780 restart_array(mddev);
3781 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
3782 wake_up(&mddev->sb_wait);
3786 set_disk_ro(mddev->gendisk, 0);
3787 err = do_md_run(mddev);
3792 /* these cannot be set */
3798 if (mddev->hold_active == UNTIL_IOCTL)
3799 mddev->hold_active = 0;
3800 sysfs_notify_dirent_safe(mddev->sysfs_state);
3804 static struct md_sysfs_entry md_array_state =
3805 __ATTR(array_state, S_IRUGO|S_IWUSR, array_state_show, array_state_store);
3808 max_corrected_read_errors_show(struct mddev *mddev, char *page) {
3809 return sprintf(page, "%d\n",
3810 atomic_read(&mddev->max_corr_read_errors));
3814 max_corrected_read_errors_store(struct mddev *mddev, const char *buf, size_t len)
3817 unsigned long n = simple_strtoul(buf, &e, 10);
3819 if (*buf && (*e == 0 || *e == '\n')) {
3820 atomic_set(&mddev->max_corr_read_errors, n);
3826 static struct md_sysfs_entry max_corr_read_errors =
3827 __ATTR(max_read_errors, S_IRUGO|S_IWUSR, max_corrected_read_errors_show,
3828 max_corrected_read_errors_store);
3831 null_show(struct mddev *mddev, char *page)
3837 new_dev_store(struct mddev *mddev, const char *buf, size_t len)
3839 /* buf must be %d:%d\n? giving major and minor numbers */
3840 /* The new device is added to the array.
3841 * If the array has a persistent superblock, we read the
3842 * superblock to initialise info and check validity.
3843 * Otherwise, only checking done is that in bind_rdev_to_array,
3844 * which mainly checks size.
3847 int major = simple_strtoul(buf, &e, 10);
3850 struct md_rdev *rdev;
3853 if (!*buf || *e != ':' || !e[1] || e[1] == '\n')
3855 minor = simple_strtoul(e+1, &e, 10);
3856 if (*e && *e != '\n')
3858 dev = MKDEV(major, minor);
3859 if (major != MAJOR(dev) ||
3860 minor != MINOR(dev))
3864 if (mddev->persistent) {
3865 rdev = md_import_device(dev, mddev->major_version,
3866 mddev->minor_version);
3867 if (!IS_ERR(rdev) && !list_empty(&mddev->disks)) {
3868 struct md_rdev *rdev0
3869 = list_entry(mddev->disks.next,
3870 struct md_rdev, same_set);
3871 err = super_types[mddev->major_version]
3872 .load_super(rdev, rdev0, mddev->minor_version);
3876 } else if (mddev->external)
3877 rdev = md_import_device(dev, -2, -1);
3879 rdev = md_import_device(dev, -1, -1);
3882 return PTR_ERR(rdev);
3883 err = bind_rdev_to_array(rdev, mddev);
3887 return err ? err : len;
3890 static struct md_sysfs_entry md_new_device =
3891 __ATTR(new_dev, S_IWUSR, null_show, new_dev_store);
3894 bitmap_store(struct mddev *mddev, const char *buf, size_t len)
3897 unsigned long chunk, end_chunk;
3901 /* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */
3903 chunk = end_chunk = simple_strtoul(buf, &end, 0);
3904 if (buf == end) break;
3905 if (*end == '-') { /* range */
3907 end_chunk = simple_strtoul(buf, &end, 0);
3908 if (buf == end) break;
3910 if (*end && !isspace(*end)) break;
3911 bitmap_dirty_bits(mddev->bitmap, chunk, end_chunk);
3912 buf = skip_spaces(end);
3914 bitmap_unplug(mddev->bitmap); /* flush the bits to disk */
3919 static struct md_sysfs_entry md_bitmap =
3920 __ATTR(bitmap_set_bits, S_IWUSR, null_show, bitmap_store);
3923 size_show(struct mddev *mddev, char *page)
3925 return sprintf(page, "%llu\n",
3926 (unsigned long long)mddev->dev_sectors / 2);
3929 static int update_size(struct mddev *mddev, sector_t num_sectors);
3932 size_store(struct mddev *mddev, const char *buf, size_t len)
3934 /* If array is inactive, we can reduce the component size, but
3935 * not increase it (except from 0).
3936 * If array is active, we can try an on-line resize
3939 int err = strict_blocks_to_sectors(buf, §ors);
3944 err = update_size(mddev, sectors);
3945 md_update_sb(mddev, 1);
3947 if (mddev->dev_sectors == 0 ||
3948 mddev->dev_sectors > sectors)
3949 mddev->dev_sectors = sectors;
3953 return err ? err : len;
3956 static struct md_sysfs_entry md_size =
3957 __ATTR(component_size, S_IRUGO|S_IWUSR, size_show, size_store);
3962 * 'none' for arrays with no metadata (good luck...)
3963 * 'external' for arrays with externally managed metadata,
3964 * or N.M for internally known formats
3967 metadata_show(struct mddev *mddev, char *page)
3969 if (mddev->persistent)
3970 return sprintf(page, "%d.%d\n",
3971 mddev->major_version, mddev->minor_version);
3972 else if (mddev->external)
3973 return sprintf(page, "external:%s\n", mddev->metadata_type);
3975 return sprintf(page, "none\n");
3979 metadata_store(struct mddev *mddev, const char *buf, size_t len)
3983 /* Changing the details of 'external' metadata is
3984 * always permitted. Otherwise there must be
3985 * no devices attached to the array.
3987 if (mddev->external && strncmp(buf, "external:", 9) == 0)
3989 else if (!list_empty(&mddev->disks))
3992 if (cmd_match(buf, "none")) {
3993 mddev->persistent = 0;
3994 mddev->external = 0;
3995 mddev->major_version = 0;
3996 mddev->minor_version = 90;
3999 if (strncmp(buf, "external:", 9) == 0) {
4000 size_t namelen = len-9;
4001 if (namelen >= sizeof(mddev->metadata_type))
4002 namelen = sizeof(mddev->metadata_type)-1;
4003 strncpy(mddev->metadata_type, buf+9, namelen);
4004 mddev->metadata_type[namelen] = 0;
4005 if (namelen && mddev->metadata_type[namelen-1] == '\n')
4006 mddev->metadata_type[--namelen] = 0;
4007 mddev->persistent = 0;
4008 mddev->external = 1;
4009 mddev->major_version = 0;
4010 mddev->minor_version = 90;
4013 major = simple_strtoul(buf, &e, 10);
4014 if (e==buf || *e != '.')
4017 minor = simple_strtoul(buf, &e, 10);
4018 if (e==buf || (*e && *e != '\n') )
4020 if (major >= ARRAY_SIZE(super_types) || super_types[major].name == NULL)
4022 mddev->major_version = major;
4023 mddev->minor_version = minor;
4024 mddev->persistent = 1;
4025 mddev->external = 0;
4029 static struct md_sysfs_entry md_metadata =
4030 __ATTR(metadata_version, S_IRUGO|S_IWUSR, metadata_show, metadata_store);
4033 action_show(struct mddev *mddev, char *page)
4035 char *type = "idle";
4036 if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
4038 else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
4039 (!mddev->ro && test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))) {
4040 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
4042 else if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
4043 if (!test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
4045 else if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
4049 } else if (test_bit(MD_RECOVERY_RECOVER, &mddev->recovery))
4052 return sprintf(page, "%s\n", type);
4055 static void reap_sync_thread(struct mddev *mddev);
4058 action_store(struct mddev *mddev, const char *page, size_t len)
4060 if (!mddev->pers || !mddev->pers->sync_request)
4063 if (cmd_match(page, "frozen"))
4064 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4066 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4068 if (cmd_match(page, "idle") || cmd_match(page, "frozen")) {
4069 if (mddev->sync_thread) {
4070 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
4071 reap_sync_thread(mddev);
4073 } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
4074 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
4076 else if (cmd_match(page, "resync"))
4077 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4078 else if (cmd_match(page, "recover")) {
4079 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
4080 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4081 } else if (cmd_match(page, "reshape")) {
4083 if (mddev->pers->start_reshape == NULL)
4085 err = mddev->pers->start_reshape(mddev);
4088 sysfs_notify(&mddev->kobj, NULL, "degraded");
4090 if (cmd_match(page, "check"))
4091 set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
4092 else if (!cmd_match(page, "repair"))
4094 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
4095 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
4097 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4098 md_wakeup_thread(mddev->thread);
4099 sysfs_notify_dirent_safe(mddev->sysfs_action);
4104 mismatch_cnt_show(struct mddev *mddev, char *page)
4106 return sprintf(page, "%llu\n",
4107 (unsigned long long) mddev->resync_mismatches);
4110 static struct md_sysfs_entry md_scan_mode =
4111 __ATTR(sync_action, S_IRUGO|S_IWUSR, action_show, action_store);
4114 static struct md_sysfs_entry md_mismatches = __ATTR_RO(mismatch_cnt);
4117 sync_min_show(struct mddev *mddev, char *page)
4119 return sprintf(page, "%d (%s)\n", speed_min(mddev),
4120 mddev->sync_speed_min ? "local": "system");
4124 sync_min_store(struct mddev *mddev, const char *buf, size_t len)
4128 if (strncmp(buf, "system", 6)==0) {
4129 mddev->sync_speed_min = 0;
4132 min = simple_strtoul(buf, &e, 10);
4133 if (buf == e || (*e && *e != '\n') || min <= 0)
4135 mddev->sync_speed_min = min;
4139 static struct md_sysfs_entry md_sync_min =
4140 __ATTR(sync_speed_min, S_IRUGO|S_IWUSR, sync_min_show, sync_min_store);
4143 sync_max_show(struct mddev *mddev, char *page)
4145 return sprintf(page, "%d (%s)\n", speed_max(mddev),
4146 mddev->sync_speed_max ? "local": "system");
4150 sync_max_store(struct mddev *mddev, const char *buf, size_t len)
4154 if (strncmp(buf, "system", 6)==0) {
4155 mddev->sync_speed_max = 0;
4158 max = simple_strtoul(buf, &e, 10);
4159 if (buf == e || (*e && *e != '\n') || max <= 0)
4161 mddev->sync_speed_max = max;
4165 static struct md_sysfs_entry md_sync_max =
4166 __ATTR(sync_speed_max, S_IRUGO|S_IWUSR, sync_max_show, sync_max_store);
4169 degraded_show(struct mddev *mddev, char *page)
4171 return sprintf(page, "%d\n", mddev->degraded);
4173 static struct md_sysfs_entry md_degraded = __ATTR_RO(degraded);
4176 sync_force_parallel_show(struct mddev *mddev, char *page)
4178 return sprintf(page, "%d\n", mddev->parallel_resync);
4182 sync_force_parallel_store(struct mddev *mddev, const char *buf, size_t len)
4186 if (strict_strtol(buf, 10, &n))
4189 if (n != 0 && n != 1)
4192 mddev->parallel_resync = n;
4194 if (mddev->sync_thread)
4195 wake_up(&resync_wait);
4200 /* force parallel resync, even with shared block devices */
4201 static struct md_sysfs_entry md_sync_force_parallel =
4202 __ATTR(sync_force_parallel, S_IRUGO|S_IWUSR,
4203 sync_force_parallel_show, sync_force_parallel_store);
4206 sync_speed_show(struct mddev *mddev, char *page)
4208 unsigned long resync, dt, db;
4209 if (mddev->curr_resync == 0)
4210 return sprintf(page, "none\n");
4211 resync = mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active);
4212 dt = (jiffies - mddev->resync_mark) / HZ;
4214 db = resync - mddev->resync_mark_cnt;
4215 return sprintf(page, "%lu\n", db/dt/2); /* K/sec */
4218 static struct md_sysfs_entry md_sync_speed = __ATTR_RO(sync_speed);
4221 sync_completed_show(struct mddev *mddev, char *page)
4223 unsigned long long max_sectors, resync;
4225 if (!test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4226 return sprintf(page, "none\n");
4228 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
4229 max_sectors = mddev->resync_max_sectors;
4231 max_sectors = mddev->dev_sectors;
4233 resync = mddev->curr_resync_completed;
4234 return sprintf(page, "%llu / %llu\n", resync, max_sectors);
4237 static struct md_sysfs_entry md_sync_completed = __ATTR_RO(sync_completed);
4240 min_sync_show(struct mddev *mddev, char *page)
4242 return sprintf(page, "%llu\n",
4243 (unsigned long long)mddev->resync_min);
4246 min_sync_store(struct mddev *mddev, const char *buf, size_t len)
4248 unsigned long long min;
4249 if (strict_strtoull(buf, 10, &min))
4251 if (min > mddev->resync_max)
4253 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4256 /* Must be a multiple of chunk_size */
4257 if (mddev->chunk_sectors) {
4258 sector_t temp = min;
4259 if (sector_div(temp, mddev->chunk_sectors))
4262 mddev->resync_min = min;
4267 static struct md_sysfs_entry md_min_sync =
4268 __ATTR(sync_min, S_IRUGO|S_IWUSR, min_sync_show, min_sync_store);
4271 max_sync_show(struct mddev *mddev, char *page)
4273 if (mddev->resync_max == MaxSector)
4274 return sprintf(page, "max\n");
4276 return sprintf(page, "%llu\n",
4277 (unsigned long long)mddev->resync_max);
4280 max_sync_store(struct mddev *mddev, const char *buf, size_t len)
4282 if (strncmp(buf, "max", 3) == 0)
4283 mddev->resync_max = MaxSector;
4285 unsigned long long max;
4286 if (strict_strtoull(buf, 10, &max))
4288 if (max < mddev->resync_min)
4290 if (max < mddev->resync_max &&
4292 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4295 /* Must be a multiple of chunk_size */
4296 if (mddev->chunk_sectors) {
4297 sector_t temp = max;
4298 if (sector_div(temp, mddev->chunk_sectors))
4301 mddev->resync_max = max;
4303 wake_up(&mddev->recovery_wait);
4307 static struct md_sysfs_entry md_max_sync =
4308 __ATTR(sync_max, S_IRUGO|S_IWUSR, max_sync_show, max_sync_store);
4311 suspend_lo_show(struct mddev *mddev, char *page)
4313 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_lo);
4317 suspend_lo_store(struct mddev *mddev, const char *buf, size_t len)
4320 unsigned long long new = simple_strtoull(buf, &e, 10);
4321 unsigned long long old = mddev->suspend_lo;
4323 if (mddev->pers == NULL ||
4324 mddev->pers->quiesce == NULL)
4326 if (buf == e || (*e && *e != '\n'))
4329 mddev->suspend_lo = new;
4331 /* Shrinking suspended region */
4332 mddev->pers->quiesce(mddev, 2);
4334 /* Expanding suspended region - need to wait */
4335 mddev->pers->quiesce(mddev, 1);
4336 mddev->pers->quiesce(mddev, 0);
4340 static struct md_sysfs_entry md_suspend_lo =
4341 __ATTR(suspend_lo, S_IRUGO|S_IWUSR, suspend_lo_show, suspend_lo_store);
4345 suspend_hi_show(struct mddev *mddev, char *page)
4347 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_hi);
4351 suspend_hi_store(struct mddev *mddev, const char *buf, size_t len)
4354 unsigned long long new = simple_strtoull(buf, &e, 10);
4355 unsigned long long old = mddev->suspend_hi;
4357 if (mddev->pers == NULL ||
4358 mddev->pers->quiesce == NULL)
4360 if (buf == e || (*e && *e != '\n'))
4363 mddev->suspend_hi = new;
4365 /* Shrinking suspended region */
4366 mddev->pers->quiesce(mddev, 2);
4368 /* Expanding suspended region - need to wait */
4369 mddev->pers->quiesce(mddev, 1);
4370 mddev->pers->quiesce(mddev, 0);
4374 static struct md_sysfs_entry md_suspend_hi =
4375 __ATTR(suspend_hi, S_IRUGO|S_IWUSR, suspend_hi_show, suspend_hi_store);
4378 reshape_position_show(struct mddev *mddev, char *page)
4380 if (mddev->reshape_position != MaxSector)
4381 return sprintf(page, "%llu\n",
4382 (unsigned long long)mddev->reshape_position);
4383 strcpy(page, "none\n");
4388 reshape_position_store(struct mddev *mddev, const char *buf, size_t len)
4391 unsigned long long new = simple_strtoull(buf, &e, 10);
4394 if (buf == e || (*e && *e != '\n'))
4396 mddev->reshape_position = new;
4397 mddev->delta_disks = 0;
4398 mddev->new_level = mddev->level;
4399 mddev->new_layout = mddev->layout;
4400 mddev->new_chunk_sectors = mddev->chunk_sectors;
4404 static struct md_sysfs_entry md_reshape_position =
4405 __ATTR(reshape_position, S_IRUGO|S_IWUSR, reshape_position_show,
4406 reshape_position_store);
4409 array_size_show(struct mddev *mddev, char *page)
4411 if (mddev->external_size)
4412 return sprintf(page, "%llu\n",
4413 (unsigned long long)mddev->array_sectors/2);
4415 return sprintf(page, "default\n");
4419 array_size_store(struct mddev *mddev, const char *buf, size_t len)
4423 if (strncmp(buf, "default", 7) == 0) {
4425 sectors = mddev->pers->size(mddev, 0, 0);
4427 sectors = mddev->array_sectors;
4429 mddev->external_size = 0;
4431 if (strict_blocks_to_sectors(buf, §ors) < 0)
4433 if (mddev->pers && mddev->pers->size(mddev, 0, 0) < sectors)
4436 mddev->external_size = 1;
4439 mddev->array_sectors = sectors;
4441 set_capacity(mddev->gendisk, mddev->array_sectors);
4442 revalidate_disk(mddev->gendisk);
4447 static struct md_sysfs_entry md_array_size =
4448 __ATTR(array_size, S_IRUGO|S_IWUSR, array_size_show,
4451 static struct attribute *md_default_attrs[] = {
4454 &md_raid_disks.attr,
4455 &md_chunk_size.attr,
4457 &md_resync_start.attr,
4459 &md_new_device.attr,
4460 &md_safe_delay.attr,
4461 &md_array_state.attr,
4462 &md_reshape_position.attr,
4463 &md_array_size.attr,
4464 &max_corr_read_errors.attr,
4468 static struct attribute *md_redundancy_attrs[] = {
4470 &md_mismatches.attr,
4473 &md_sync_speed.attr,
4474 &md_sync_force_parallel.attr,
4475 &md_sync_completed.attr,
4478 &md_suspend_lo.attr,
4479 &md_suspend_hi.attr,
4484 static struct attribute_group md_redundancy_group = {
4486 .attrs = md_redundancy_attrs,
4491 md_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
4493 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
4494 struct mddev *mddev = container_of(kobj, struct mddev, kobj);
4499 spin_lock(&all_mddevs_lock);
4500 if (list_empty(&mddev->all_mddevs)) {
4501 spin_unlock(&all_mddevs_lock);
4505 spin_unlock(&all_mddevs_lock);
4507 rv = mddev_lock(mddev);
4509 rv = entry->show(mddev, page);
4510 mddev_unlock(mddev);
4517 md_attr_store(struct kobject *kobj, struct attribute *attr,
4518 const char *page, size_t length)
4520 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
4521 struct mddev *mddev = container_of(kobj, struct mddev, kobj);
4526 if (!capable(CAP_SYS_ADMIN))
4528 spin_lock(&all_mddevs_lock);
4529 if (list_empty(&mddev->all_mddevs)) {
4530 spin_unlock(&all_mddevs_lock);
4534 spin_unlock(&all_mddevs_lock);
4535 rv = mddev_lock(mddev);
4537 rv = entry->store(mddev, page, length);
4538 mddev_unlock(mddev);
4544 static void md_free(struct kobject *ko)
4546 struct mddev *mddev = container_of(ko, struct mddev, kobj);
4548 if (mddev->sysfs_state)
4549 sysfs_put(mddev->sysfs_state);
4551 if (mddev->gendisk) {
4552 del_gendisk(mddev->gendisk);
4553 put_disk(mddev->gendisk);
4556 blk_cleanup_queue(mddev->queue);
4561 static const struct sysfs_ops md_sysfs_ops = {
4562 .show = md_attr_show,
4563 .store = md_attr_store,
4565 static struct kobj_type md_ktype = {
4567 .sysfs_ops = &md_sysfs_ops,
4568 .default_attrs = md_default_attrs,
4573 static void mddev_delayed_delete(struct work_struct *ws)
4575 struct mddev *mddev = container_of(ws, struct mddev, del_work);
4577 sysfs_remove_group(&mddev->kobj, &md_bitmap_group);
4578 kobject_del(&mddev->kobj);
4579 kobject_put(&mddev->kobj);
4582 static int md_alloc(dev_t dev, char *name)
4584 static DEFINE_MUTEX(disks_mutex);
4585 struct mddev *mddev = mddev_find(dev);
4586 struct gendisk *disk;
4595 partitioned = (MAJOR(mddev->unit) != MD_MAJOR);
4596 shift = partitioned ? MdpMinorShift : 0;
4597 unit = MINOR(mddev->unit) >> shift;
4599 /* wait for any previous instance of this device to be
4600 * completely removed (mddev_delayed_delete).
4602 flush_workqueue(md_misc_wq);
4604 mutex_lock(&disks_mutex);
4610 /* Need to ensure that 'name' is not a duplicate.
4612 struct mddev *mddev2;
4613 spin_lock(&all_mddevs_lock);
4615 list_for_each_entry(mddev2, &all_mddevs, all_mddevs)
4616 if (mddev2->gendisk &&
4617 strcmp(mddev2->gendisk->disk_name, name) == 0) {
4618 spin_unlock(&all_mddevs_lock);
4621 spin_unlock(&all_mddevs_lock);
4625 mddev->queue = blk_alloc_queue(GFP_KERNEL);
4628 mddev->queue->queuedata = mddev;
4630 blk_queue_make_request(mddev->queue, md_make_request);
4632 disk = alloc_disk(1 << shift);
4634 blk_cleanup_queue(mddev->queue);
4635 mddev->queue = NULL;
4638 disk->major = MAJOR(mddev->unit);
4639 disk->first_minor = unit << shift;
4641 strcpy(disk->disk_name, name);
4642 else if (partitioned)
4643 sprintf(disk->disk_name, "md_d%d", unit);
4645 sprintf(disk->disk_name, "md%d", unit);
4646 disk->fops = &md_fops;
4647 disk->private_data = mddev;
4648 disk->queue = mddev->queue;
4649 blk_queue_flush(mddev->queue, REQ_FLUSH | REQ_FUA);
4650 /* Allow extended partitions. This makes the
4651 * 'mdp' device redundant, but we can't really
4654 disk->flags |= GENHD_FL_EXT_DEVT;
4655 mddev->gendisk = disk;
4656 /* As soon as we call add_disk(), another thread could get
4657 * through to md_open, so make sure it doesn't get too far
4659 mutex_lock(&mddev->open_mutex);
4662 error = kobject_init_and_add(&mddev->kobj, &md_ktype,
4663 &disk_to_dev(disk)->kobj, "%s", "md");
4665 /* This isn't possible, but as kobject_init_and_add is marked
4666 * __must_check, we must do something with the result
4668 printk(KERN_WARNING "md: cannot register %s/md - name in use\n",
4672 if (mddev->kobj.sd &&
4673 sysfs_create_group(&mddev->kobj, &md_bitmap_group))
4674 printk(KERN_DEBUG "pointless warning\n");
4675 mutex_unlock(&mddev->open_mutex);
4677 mutex_unlock(&disks_mutex);
4678 if (!error && mddev->kobj.sd) {
4679 kobject_uevent(&mddev->kobj, KOBJ_ADD);
4680 mddev->sysfs_state = sysfs_get_dirent_safe(mddev->kobj.sd, "array_state");
4686 static struct kobject *md_probe(dev_t dev, int *part, void *data)
4688 md_alloc(dev, NULL);
4692 static int add_named_array(const char *val, struct kernel_param *kp)
4694 /* val must be "md_*" where * is not all digits.
4695 * We allocate an array with a large free minor number, and
4696 * set the name to val. val must not already be an active name.
4698 int len = strlen(val);
4699 char buf[DISK_NAME_LEN];
4701 while (len && val[len-1] == '\n')
4703 if (len >= DISK_NAME_LEN)
4705 strlcpy(buf, val, len+1);
4706 if (strncmp(buf, "md_", 3) != 0)
4708 return md_alloc(0, buf);
4711 static void md_safemode_timeout(unsigned long data)
4713 struct mddev *mddev = (struct mddev *) data;
4715 if (!atomic_read(&mddev->writes_pending)) {
4716 mddev->safemode = 1;
4717 if (mddev->external)
4718 sysfs_notify_dirent_safe(mddev->sysfs_state);
4720 md_wakeup_thread(mddev->thread);
4723 static int start_dirty_degraded;
4725 int md_run(struct mddev *mddev)
4728 struct md_rdev *rdev;
4729 struct md_personality *pers;
4731 if (list_empty(&mddev->disks))
4732 /* cannot run an array with no devices.. */
4737 /* Cannot run until previous stop completes properly */
4738 if (mddev->sysfs_active)
4742 * Analyze all RAID superblock(s)
4744 if (!mddev->raid_disks) {
4745 if (!mddev->persistent)
4750 if (mddev->level != LEVEL_NONE)
4751 request_module("md-level-%d", mddev->level);
4752 else if (mddev->clevel[0])
4753 request_module("md-%s", mddev->clevel);
4756 * Drop all container device buffers, from now on
4757 * the only valid external interface is through the md
4760 list_for_each_entry(rdev, &mddev->disks, same_set) {
4761 if (test_bit(Faulty, &rdev->flags))
4763 sync_blockdev(rdev->bdev);
4764 invalidate_bdev(rdev->bdev);
4766 /* perform some consistency tests on the device.
4767 * We don't want the data to overlap the metadata,
4768 * Internal Bitmap issues have been handled elsewhere.
4770 if (rdev->meta_bdev) {
4771 /* Nothing to check */;
4772 } else if (rdev->data_offset < rdev->sb_start) {
4773 if (mddev->dev_sectors &&
4774 rdev->data_offset + mddev->dev_sectors
4776 printk("md: %s: data overlaps metadata\n",
4781 if (rdev->sb_start + rdev->sb_size/512
4782 > rdev->data_offset) {
4783 printk("md: %s: metadata overlaps data\n",
4788 sysfs_notify_dirent_safe(rdev->sysfs_state);
4791 if (mddev->bio_set == NULL)
4792 mddev->bio_set = bioset_create(BIO_POOL_SIZE,
4793 sizeof(struct mddev *));
4795 spin_lock(&pers_lock);
4796 pers = find_pers(mddev->level, mddev->clevel);
4797 if (!pers || !try_module_get(pers->owner)) {
4798 spin_unlock(&pers_lock);
4799 if (mddev->level != LEVEL_NONE)
4800 printk(KERN_WARNING "md: personality for level %d is not loaded!\n",
4803 printk(KERN_WARNING "md: personality for level %s is not loaded!\n",
4808 spin_unlock(&pers_lock);
4809 if (mddev->level != pers->level) {
4810 mddev->level = pers->level;
4811 mddev->new_level = pers->level;
4813 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
4815 if (mddev->reshape_position != MaxSector &&
4816 pers->start_reshape == NULL) {
4817 /* This personality cannot handle reshaping... */
4819 module_put(pers->owner);
4823 if (pers->sync_request) {
4824 /* Warn if this is a potentially silly
4827 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
4828 struct md_rdev *rdev2;
4831 list_for_each_entry(rdev, &mddev->disks, same_set)
4832 list_for_each_entry(rdev2, &mddev->disks, same_set) {
4834 rdev->bdev->bd_contains ==
4835 rdev2->bdev->bd_contains) {
4837 "%s: WARNING: %s appears to be"
4838 " on the same physical disk as"
4841 bdevname(rdev->bdev,b),
4842 bdevname(rdev2->bdev,b2));
4849 "True protection against single-disk"
4850 " failure might be compromised.\n");
4853 mddev->recovery = 0;
4854 /* may be over-ridden by personality */
4855 mddev->resync_max_sectors = mddev->dev_sectors;
4857 mddev->ok_start_degraded = start_dirty_degraded;
4859 if (start_readonly && mddev->ro == 0)
4860 mddev->ro = 2; /* read-only, but switch on first write */
4862 err = mddev->pers->run(mddev);
4864 printk(KERN_ERR "md: pers->run() failed ...\n");
4865 else if (mddev->pers->size(mddev, 0, 0) < mddev->array_sectors) {
4866 WARN_ONCE(!mddev->external_size, "%s: default size too small,"
4867 " but 'external_size' not in effect?\n", __func__);
4869 "md: invalid array_size %llu > default size %llu\n",
4870 (unsigned long long)mddev->array_sectors / 2,
4871 (unsigned long long)mddev->pers->size(mddev, 0, 0) / 2);
4873 mddev->pers->stop(mddev);
4875 if (err == 0 && mddev->pers->sync_request) {
4876 err = bitmap_create(mddev);
4878 printk(KERN_ERR "%s: failed to create bitmap (%d)\n",
4879 mdname(mddev), err);
4880 mddev->pers->stop(mddev);
4884 module_put(mddev->pers->owner);
4886 bitmap_destroy(mddev);
4889 if (mddev->pers->sync_request) {
4890 if (mddev->kobj.sd &&
4891 sysfs_create_group(&mddev->kobj, &md_redundancy_group))
4893 "md: cannot register extra attributes for %s\n",
4895 mddev->sysfs_action = sysfs_get_dirent_safe(mddev->kobj.sd, "sync_action");
4896 } else if (mddev->ro == 2) /* auto-readonly not meaningful */
4899 atomic_set(&mddev->writes_pending,0);
4900 atomic_set(&mddev->max_corr_read_errors,
4901 MD_DEFAULT_MAX_CORRECTED_READ_ERRORS);
4902 mddev->safemode = 0;
4903 mddev->safemode_timer.function = md_safemode_timeout;
4904 mddev->safemode_timer.data = (unsigned long) mddev;
4905 mddev->safemode_delay = (200 * HZ)/1000 +1; /* 200 msec delay */
4909 list_for_each_entry(rdev, &mddev->disks, same_set)
4910 if (rdev->raid_disk >= 0)
4911 if (sysfs_link_rdev(mddev, rdev))
4912 /* failure here is OK */;
4914 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4917 md_update_sb(mddev, 0);
4919 md_new_event(mddev);
4920 sysfs_notify_dirent_safe(mddev->sysfs_state);
4921 sysfs_notify_dirent_safe(mddev->sysfs_action);
4922 sysfs_notify(&mddev->kobj, NULL, "degraded");
4925 EXPORT_SYMBOL_GPL(md_run);
4927 static int do_md_run(struct mddev *mddev)
4931 err = md_run(mddev);
4934 err = bitmap_load(mddev);
4936 bitmap_destroy(mddev);
4940 md_wakeup_thread(mddev->thread);
4941 md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
4943 set_capacity(mddev->gendisk, mddev->array_sectors);
4944 revalidate_disk(mddev->gendisk);
4946 kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
4951 static int restart_array(struct mddev *mddev)
4953 struct gendisk *disk = mddev->gendisk;
4955 /* Complain if it has no devices */
4956 if (list_empty(&mddev->disks))
4962 mddev->safemode = 0;
4964 set_disk_ro(disk, 0);
4965 printk(KERN_INFO "md: %s switched to read-write mode.\n",
4967 /* Kick recovery or resync if necessary */
4968 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4969 md_wakeup_thread(mddev->thread);
4970 md_wakeup_thread(mddev->sync_thread);
4971 sysfs_notify_dirent_safe(mddev->sysfs_state);
4975 /* similar to deny_write_access, but accounts for our holding a reference
4976 * to the file ourselves */
4977 static int deny_bitmap_write_access(struct file * file)
4979 struct inode *inode = file->f_mapping->host;
4981 spin_lock(&inode->i_lock);
4982 if (atomic_read(&inode->i_writecount) > 1) {
4983 spin_unlock(&inode->i_lock);
4986 atomic_set(&inode->i_writecount, -1);
4987 spin_unlock(&inode->i_lock);
4992 void restore_bitmap_write_access(struct file *file)
4994 struct inode *inode = file->f_mapping->host;
4996 spin_lock(&inode->i_lock);
4997 atomic_set(&inode->i_writecount, 1);
4998 spin_unlock(&inode->i_lock);
5001 static void md_clean(struct mddev *mddev)
5003 mddev->array_sectors = 0;
5004 mddev->external_size = 0;
5005 mddev->dev_sectors = 0;
5006 mddev->raid_disks = 0;
5007 mddev->recovery_cp = 0;
5008 mddev->resync_min = 0;
5009 mddev->resync_max = MaxSector;
5010 mddev->reshape_position = MaxSector;
5011 mddev->external = 0;
5012 mddev->persistent = 0;
5013 mddev->level = LEVEL_NONE;
5014 mddev->clevel[0] = 0;
5017 mddev->metadata_type[0] = 0;
5018 mddev->chunk_sectors = 0;
5019 mddev->ctime = mddev->utime = 0;
5021 mddev->max_disks = 0;
5023 mddev->can_decrease_events = 0;
5024 mddev->delta_disks = 0;
5025 mddev->new_level = LEVEL_NONE;
5026 mddev->new_layout = 0;
5027 mddev->new_chunk_sectors = 0;
5028 mddev->curr_resync = 0;
5029 mddev->resync_mismatches = 0;
5030 mddev->suspend_lo = mddev->suspend_hi = 0;
5031 mddev->sync_speed_min = mddev->sync_speed_max = 0;
5032 mddev->recovery = 0;
5035 mddev->degraded = 0;
5036 mddev->safemode = 0;
5037 mddev->bitmap_info.offset = 0;
5038 mddev->bitmap_info.default_offset = 0;
5039 mddev->bitmap_info.chunksize = 0;
5040 mddev->bitmap_info.daemon_sleep = 0;
5041 mddev->bitmap_info.max_write_behind = 0;
5044 static void __md_stop_writes(struct mddev *mddev)
5046 if (mddev->sync_thread) {
5047 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5048 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5049 reap_sync_thread(mddev);
5052 del_timer_sync(&mddev->safemode_timer);
5054 bitmap_flush(mddev);
5055 md_super_wait(mddev);
5057 if (!mddev->in_sync || mddev->flags) {
5058 /* mark array as shutdown cleanly */
5060 md_update_sb(mddev, 1);
5064 void md_stop_writes(struct mddev *mddev)
5067 __md_stop_writes(mddev);
5068 mddev_unlock(mddev);
5070 EXPORT_SYMBOL_GPL(md_stop_writes);
5072 void md_stop(struct mddev *mddev)
5075 mddev->pers->stop(mddev);
5076 if (mddev->pers->sync_request && mddev->to_remove == NULL)
5077 mddev->to_remove = &md_redundancy_group;
5078 module_put(mddev->pers->owner);
5080 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5082 EXPORT_SYMBOL_GPL(md_stop);
5084 static int md_set_readonly(struct mddev *mddev, struct block_device *bdev)
5087 mutex_lock(&mddev->open_mutex);
5088 if (atomic_read(&mddev->openers) > !!bdev) {
5089 printk("md: %s still in use.\n",mdname(mddev));
5094 sync_blockdev(bdev);
5096 __md_stop_writes(mddev);
5102 set_disk_ro(mddev->gendisk, 1);
5103 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5104 sysfs_notify_dirent_safe(mddev->sysfs_state);
5108 mutex_unlock(&mddev->open_mutex);
5113 * 0 - completely stop and dis-assemble array
5114 * 2 - stop but do not disassemble array
5116 static int do_md_stop(struct mddev * mddev, int mode,
5117 struct block_device *bdev)
5119 struct gendisk *disk = mddev->gendisk;
5120 struct md_rdev *rdev;
5122 mutex_lock(&mddev->open_mutex);
5123 if (atomic_read(&mddev->openers) > !!bdev ||
5124 mddev->sysfs_active) {
5125 printk("md: %s still in use.\n",mdname(mddev));
5126 mutex_unlock(&mddev->open_mutex);
5130 /* It is possible IO was issued on some other
5131 * open file which was closed before we took ->open_mutex.
5132 * As that was not the last close __blkdev_put will not
5133 * have called sync_blockdev, so we must.
5135 sync_blockdev(bdev);
5139 set_disk_ro(disk, 0);
5141 __md_stop_writes(mddev);
5143 mddev->queue->merge_bvec_fn = NULL;
5144 mddev->queue->backing_dev_info.congested_fn = NULL;
5146 /* tell userspace to handle 'inactive' */
5147 sysfs_notify_dirent_safe(mddev->sysfs_state);
5149 list_for_each_entry(rdev, &mddev->disks, same_set)
5150 if (rdev->raid_disk >= 0)
5151 sysfs_unlink_rdev(mddev, rdev);
5153 set_capacity(disk, 0);
5154 mutex_unlock(&mddev->open_mutex);
5156 revalidate_disk(disk);
5161 mutex_unlock(&mddev->open_mutex);
5163 * Free resources if final stop
5166 printk(KERN_INFO "md: %s stopped.\n", mdname(mddev));
5168 bitmap_destroy(mddev);
5169 if (mddev->bitmap_info.file) {
5170 restore_bitmap_write_access(mddev->bitmap_info.file);
5171 fput(mddev->bitmap_info.file);
5172 mddev->bitmap_info.file = NULL;
5174 mddev->bitmap_info.offset = 0;
5176 export_array(mddev);
5179 kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
5180 if (mddev->hold_active == UNTIL_STOP)
5181 mddev->hold_active = 0;
5183 blk_integrity_unregister(disk);
5184 md_new_event(mddev);
5185 sysfs_notify_dirent_safe(mddev->sysfs_state);
5190 static void autorun_array(struct mddev *mddev)
5192 struct md_rdev *rdev;
5195 if (list_empty(&mddev->disks))
5198 printk(KERN_INFO "md: running: ");
5200 list_for_each_entry(rdev, &mddev->disks, same_set) {
5201 char b[BDEVNAME_SIZE];
5202 printk("<%s>", bdevname(rdev->bdev,b));
5206 err = do_md_run(mddev);
5208 printk(KERN_WARNING "md: do_md_run() returned %d\n", err);
5209 do_md_stop(mddev, 0, NULL);
5214 * lets try to run arrays based on all disks that have arrived
5215 * until now. (those are in pending_raid_disks)
5217 * the method: pick the first pending disk, collect all disks with
5218 * the same UUID, remove all from the pending list and put them into
5219 * the 'same_array' list. Then order this list based on superblock
5220 * update time (freshest comes first), kick out 'old' disks and
5221 * compare superblocks. If everything's fine then run it.
5223 * If "unit" is allocated, then bump its reference count
5225 static void autorun_devices(int part)
5227 struct md_rdev *rdev0, *rdev, *tmp;
5228 struct mddev *mddev;
5229 char b[BDEVNAME_SIZE];
5231 printk(KERN_INFO "md: autorun ...\n");
5232 while (!list_empty(&pending_raid_disks)) {
5235 LIST_HEAD(candidates);
5236 rdev0 = list_entry(pending_raid_disks.next,
5237 struct md_rdev, same_set);
5239 printk(KERN_INFO "md: considering %s ...\n",
5240 bdevname(rdev0->bdev,b));
5241 INIT_LIST_HEAD(&candidates);
5242 rdev_for_each_list(rdev, tmp, &pending_raid_disks)
5243 if (super_90_load(rdev, rdev0, 0) >= 0) {
5244 printk(KERN_INFO "md: adding %s ...\n",
5245 bdevname(rdev->bdev,b));
5246 list_move(&rdev->same_set, &candidates);
5249 * now we have a set of devices, with all of them having
5250 * mostly sane superblocks. It's time to allocate the
5254 dev = MKDEV(mdp_major,
5255 rdev0->preferred_minor << MdpMinorShift);
5256 unit = MINOR(dev) >> MdpMinorShift;
5258 dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
5261 if (rdev0->preferred_minor != unit) {
5262 printk(KERN_INFO "md: unit number in %s is bad: %d\n",
5263 bdevname(rdev0->bdev, b), rdev0->preferred_minor);
5267 md_probe(dev, NULL, NULL);
5268 mddev = mddev_find(dev);
5269 if (!mddev || !mddev->gendisk) {
5273 "md: cannot allocate memory for md drive.\n");
5276 if (mddev_lock(mddev))
5277 printk(KERN_WARNING "md: %s locked, cannot run\n",
5279 else if (mddev->raid_disks || mddev->major_version
5280 || !list_empty(&mddev->disks)) {
5282 "md: %s already running, cannot run %s\n",
5283 mdname(mddev), bdevname(rdev0->bdev,b));
5284 mddev_unlock(mddev);
5286 printk(KERN_INFO "md: created %s\n", mdname(mddev));
5287 mddev->persistent = 1;
5288 rdev_for_each_list(rdev, tmp, &candidates) {
5289 list_del_init(&rdev->same_set);
5290 if (bind_rdev_to_array(rdev, mddev))
5293 autorun_array(mddev);
5294 mddev_unlock(mddev);
5296 /* on success, candidates will be empty, on error
5299 rdev_for_each_list(rdev, tmp, &candidates) {
5300 list_del_init(&rdev->same_set);
5305 printk(KERN_INFO "md: ... autorun DONE.\n");
5307 #endif /* !MODULE */
5309 static int get_version(void __user * arg)
5313 ver.major = MD_MAJOR_VERSION;
5314 ver.minor = MD_MINOR_VERSION;
5315 ver.patchlevel = MD_PATCHLEVEL_VERSION;
5317 if (copy_to_user(arg, &ver, sizeof(ver)))
5323 static int get_array_info(struct mddev * mddev, void __user * arg)
5325 mdu_array_info_t info;
5326 int nr,working,insync,failed,spare;
5327 struct md_rdev *rdev;
5329 nr=working=insync=failed=spare=0;
5330 list_for_each_entry(rdev, &mddev->disks, same_set) {
5332 if (test_bit(Faulty, &rdev->flags))
5336 if (test_bit(In_sync, &rdev->flags))
5343 info.major_version = mddev->major_version;
5344 info.minor_version = mddev->minor_version;
5345 info.patch_version = MD_PATCHLEVEL_VERSION;
5346 info.ctime = mddev->ctime;
5347 info.level = mddev->level;
5348 info.size = mddev->dev_sectors / 2;
5349 if (info.size != mddev->dev_sectors / 2) /* overflow */
5352 info.raid_disks = mddev->raid_disks;
5353 info.md_minor = mddev->md_minor;
5354 info.not_persistent= !mddev->persistent;
5356 info.utime = mddev->utime;
5359 info.state = (1<<MD_SB_CLEAN);
5360 if (mddev->bitmap && mddev->bitmap_info.offset)
5361 info.state = (1<<MD_SB_BITMAP_PRESENT);
5362 info.active_disks = insync;
5363 info.working_disks = working;
5364 info.failed_disks = failed;
5365 info.spare_disks = spare;
5367 info.layout = mddev->layout;
5368 info.chunk_size = mddev->chunk_sectors << 9;
5370 if (copy_to_user(arg, &info, sizeof(info)))
5376 static int get_bitmap_file(struct mddev * mddev, void __user * arg)
5378 mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */
5379 char *ptr, *buf = NULL;
5382 if (md_allow_write(mddev))
5383 file = kmalloc(sizeof(*file), GFP_NOIO);
5385 file = kmalloc(sizeof(*file), GFP_KERNEL);
5390 /* bitmap disabled, zero the first byte and copy out */
5391 if (!mddev->bitmap || !mddev->bitmap->file) {
5392 file->pathname[0] = '\0';
5396 buf = kmalloc(sizeof(file->pathname), GFP_KERNEL);
5400 ptr = d_path(&mddev->bitmap->file->f_path, buf, sizeof(file->pathname));
5404 strcpy(file->pathname, ptr);
5408 if (copy_to_user(arg, file, sizeof(*file)))
5416 static int get_disk_info(struct mddev * mddev, void __user * arg)
5418 mdu_disk_info_t info;
5419 struct md_rdev *rdev;
5421 if (copy_from_user(&info, arg, sizeof(info)))
5424 rdev = find_rdev_nr(mddev, info.number);
5426 info.major = MAJOR(rdev->bdev->bd_dev);
5427 info.minor = MINOR(rdev->bdev->bd_dev);
5428 info.raid_disk = rdev->raid_disk;
5430 if (test_bit(Faulty, &rdev->flags))
5431 info.state |= (1<<MD_DISK_FAULTY);
5432 else if (test_bit(In_sync, &rdev->flags)) {
5433 info.state |= (1<<MD_DISK_ACTIVE);
5434 info.state |= (1<<MD_DISK_SYNC);
5436 if (test_bit(WriteMostly, &rdev->flags))
5437 info.state |= (1<<MD_DISK_WRITEMOSTLY);
5439 info.major = info.minor = 0;
5440 info.raid_disk = -1;
5441 info.state = (1<<MD_DISK_REMOVED);
5444 if (copy_to_user(arg, &info, sizeof(info)))
5450 static int add_new_disk(struct mddev * mddev, mdu_disk_info_t *info)
5452 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
5453 struct md_rdev *rdev;
5454 dev_t dev = MKDEV(info->major,info->minor);
5456 if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
5459 if (!mddev->raid_disks) {
5461 /* expecting a device which has a superblock */
5462 rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
5465 "md: md_import_device returned %ld\n",
5467 return PTR_ERR(rdev);
5469 if (!list_empty(&mddev->disks)) {
5470 struct md_rdev *rdev0
5471 = list_entry(mddev->disks.next,
5472 struct md_rdev, same_set);
5473 err = super_types[mddev->major_version]
5474 .load_super(rdev, rdev0, mddev->minor_version);
5477 "md: %s has different UUID to %s\n",
5478 bdevname(rdev->bdev,b),
5479 bdevname(rdev0->bdev,b2));
5484 err = bind_rdev_to_array(rdev, mddev);
5491 * add_new_disk can be used once the array is assembled
5492 * to add "hot spares". They must already have a superblock
5497 if (!mddev->pers->hot_add_disk) {
5499 "%s: personality does not support diskops!\n",
5503 if (mddev->persistent)
5504 rdev = md_import_device(dev, mddev->major_version,
5505 mddev->minor_version);
5507 rdev = md_import_device(dev, -1, -1);
5510 "md: md_import_device returned %ld\n",
5512 return PTR_ERR(rdev);
5514 /* set saved_raid_disk if appropriate */
5515 if (!mddev->persistent) {
5516 if (info->state & (1<<MD_DISK_SYNC) &&
5517 info->raid_disk < mddev->raid_disks) {
5518 rdev->raid_disk = info->raid_disk;
5519 set_bit(In_sync, &rdev->flags);
5521 rdev->raid_disk = -1;
5523 super_types[mddev->major_version].
5524 validate_super(mddev, rdev);
5525 if ((info->state & (1<<MD_DISK_SYNC)) &&
5526 (!test_bit(In_sync, &rdev->flags) ||
5527 rdev->raid_disk != info->raid_disk)) {
5528 /* This was a hot-add request, but events doesn't
5529 * match, so reject it.
5535 if (test_bit(In_sync, &rdev->flags))
5536 rdev->saved_raid_disk = rdev->raid_disk;
5538 rdev->saved_raid_disk = -1;
5540 clear_bit(In_sync, &rdev->flags); /* just to be sure */
5541 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
5542 set_bit(WriteMostly, &rdev->flags);
5544 clear_bit(WriteMostly, &rdev->flags);
5546 rdev->raid_disk = -1;
5547 err = bind_rdev_to_array(rdev, mddev);
5548 if (!err && !mddev->pers->hot_remove_disk) {
5549 /* If there is hot_add_disk but no hot_remove_disk
5550 * then added disks for geometry changes,
5551 * and should be added immediately.
5553 super_types[mddev->major_version].
5554 validate_super(mddev, rdev);
5555 err = mddev->pers->hot_add_disk(mddev, rdev);
5557 unbind_rdev_from_array(rdev);
5562 sysfs_notify_dirent_safe(rdev->sysfs_state);
5564 md_update_sb(mddev, 1);
5565 if (mddev->degraded)
5566 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
5567 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5569 md_new_event(mddev);
5570 md_wakeup_thread(mddev->thread);
5574 /* otherwise, add_new_disk is only allowed
5575 * for major_version==0 superblocks
5577 if (mddev->major_version != 0) {
5578 printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n",
5583 if (!(info->state & (1<<MD_DISK_FAULTY))) {
5585 rdev = md_import_device(dev, -1, 0);
5588 "md: error, md_import_device() returned %ld\n",
5590 return PTR_ERR(rdev);
5592 rdev->desc_nr = info->number;
5593 if (info->raid_disk < mddev->raid_disks)
5594 rdev->raid_disk = info->raid_disk;
5596 rdev->raid_disk = -1;
5598 if (rdev->raid_disk < mddev->raid_disks)
5599 if (info->state & (1<<MD_DISK_SYNC))
5600 set_bit(In_sync, &rdev->flags);
5602 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
5603 set_bit(WriteMostly, &rdev->flags);
5605 if (!mddev->persistent) {
5606 printk(KERN_INFO "md: nonpersistent superblock ...\n");
5607 rdev->sb_start = i_size_read(rdev->bdev->bd_inode) / 512;
5609 rdev->sb_start = calc_dev_sboffset(rdev);
5610 rdev->sectors = rdev->sb_start;
5612 err = bind_rdev_to_array(rdev, mddev);
5622 static int hot_remove_disk(struct mddev * mddev, dev_t dev)
5624 char b[BDEVNAME_SIZE];
5625 struct md_rdev *rdev;
5627 rdev = find_rdev(mddev, dev);
5631 if (rdev->raid_disk >= 0)
5634 kick_rdev_from_array(rdev);
5635 md_update_sb(mddev, 1);
5636 md_new_event(mddev);
5640 printk(KERN_WARNING "md: cannot remove active disk %s from %s ...\n",
5641 bdevname(rdev->bdev,b), mdname(mddev));
5645 static int hot_add_disk(struct mddev * mddev, dev_t dev)
5647 char b[BDEVNAME_SIZE];
5649 struct md_rdev *rdev;
5654 if (mddev->major_version != 0) {
5655 printk(KERN_WARNING "%s: HOT_ADD may only be used with"
5656 " version-0 superblocks.\n",
5660 if (!mddev->pers->hot_add_disk) {
5662 "%s: personality does not support diskops!\n",
5667 rdev = md_import_device(dev, -1, 0);
5670 "md: error, md_import_device() returned %ld\n",
5675 if (mddev->persistent)
5676 rdev->sb_start = calc_dev_sboffset(rdev);
5678 rdev->sb_start = i_size_read(rdev->bdev->bd_inode) / 512;
5680 rdev->sectors = rdev->sb_start;
5682 if (test_bit(Faulty, &rdev->flags)) {
5684 "md: can not hot-add faulty %s disk to %s!\n",
5685 bdevname(rdev->bdev,b), mdname(mddev));
5689 clear_bit(In_sync, &rdev->flags);
5691 rdev->saved_raid_disk = -1;
5692 err = bind_rdev_to_array(rdev, mddev);
5697 * The rest should better be atomic, we can have disk failures
5698 * noticed in interrupt contexts ...
5701 rdev->raid_disk = -1;
5703 md_update_sb(mddev, 1);
5706 * Kick recovery, maybe this spare has to be added to the
5707 * array immediately.
5709 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5710 md_wakeup_thread(mddev->thread);
5711 md_new_event(mddev);
5719 static int set_bitmap_file(struct mddev *mddev, int fd)
5724 if (!mddev->pers->quiesce)
5726 if (mddev->recovery || mddev->sync_thread)
5728 /* we should be able to change the bitmap.. */
5734 return -EEXIST; /* cannot add when bitmap is present */
5735 mddev->bitmap_info.file = fget(fd);
5737 if (mddev->bitmap_info.file == NULL) {
5738 printk(KERN_ERR "%s: error: failed to get bitmap file\n",
5743 err = deny_bitmap_write_access(mddev->bitmap_info.file);
5745 printk(KERN_ERR "%s: error: bitmap file is already in use\n",
5747 fput(mddev->bitmap_info.file);
5748 mddev->bitmap_info.file = NULL;
5751 mddev->bitmap_info.offset = 0; /* file overrides offset */
5752 } else if (mddev->bitmap == NULL)
5753 return -ENOENT; /* cannot remove what isn't there */
5756 mddev->pers->quiesce(mddev, 1);
5758 err = bitmap_create(mddev);
5760 err = bitmap_load(mddev);
5762 if (fd < 0 || err) {
5763 bitmap_destroy(mddev);
5764 fd = -1; /* make sure to put the file */
5766 mddev->pers->quiesce(mddev, 0);
5769 if (mddev->bitmap_info.file) {
5770 restore_bitmap_write_access(mddev->bitmap_info.file);
5771 fput(mddev->bitmap_info.file);
5773 mddev->bitmap_info.file = NULL;
5780 * set_array_info is used two different ways
5781 * The original usage is when creating a new array.
5782 * In this usage, raid_disks is > 0 and it together with
5783 * level, size, not_persistent,layout,chunksize determine the
5784 * shape of the array.
5785 * This will always create an array with a type-0.90.0 superblock.
5786 * The newer usage is when assembling an array.
5787 * In this case raid_disks will be 0, and the major_version field is
5788 * use to determine which style super-blocks are to be found on the devices.
5789 * The minor and patch _version numbers are also kept incase the
5790 * super_block handler wishes to interpret them.
5792 static int set_array_info(struct mddev * mddev, mdu_array_info_t *info)
5795 if (info->raid_disks == 0) {
5796 /* just setting version number for superblock loading */
5797 if (info->major_version < 0 ||
5798 info->major_version >= ARRAY_SIZE(super_types) ||
5799 super_types[info->major_version].name == NULL) {
5800 /* maybe try to auto-load a module? */
5802 "md: superblock version %d not known\n",
5803 info->major_version);
5806 mddev->major_version = info->major_version;
5807 mddev->minor_version = info->minor_version;
5808 mddev->patch_version = info->patch_version;
5809 mddev->persistent = !info->not_persistent;
5810 /* ensure mddev_put doesn't delete this now that there
5811 * is some minimal configuration.
5813 mddev->ctime = get_seconds();
5816 mddev->major_version = MD_MAJOR_VERSION;
5817 mddev->minor_version = MD_MINOR_VERSION;
5818 mddev->patch_version = MD_PATCHLEVEL_VERSION;
5819 mddev->ctime = get_seconds();
5821 mddev->level = info->level;
5822 mddev->clevel[0] = 0;
5823 mddev->dev_sectors = 2 * (sector_t)info->size;
5824 mddev->raid_disks = info->raid_disks;
5825 /* don't set md_minor, it is determined by which /dev/md* was
5828 if (info->state & (1<<MD_SB_CLEAN))
5829 mddev->recovery_cp = MaxSector;
5831 mddev->recovery_cp = 0;
5832 mddev->persistent = ! info->not_persistent;
5833 mddev->external = 0;
5835 mddev->layout = info->layout;
5836 mddev->chunk_sectors = info->chunk_size >> 9;
5838 mddev->max_disks = MD_SB_DISKS;
5840 if (mddev->persistent)
5842 set_bit(MD_CHANGE_DEVS, &mddev->flags);
5844 mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9;
5845 mddev->bitmap_info.offset = 0;
5847 mddev->reshape_position = MaxSector;
5850 * Generate a 128 bit UUID
5852 get_random_bytes(mddev->uuid, 16);
5854 mddev->new_level = mddev->level;
5855 mddev->new_chunk_sectors = mddev->chunk_sectors;
5856 mddev->new_layout = mddev->layout;
5857 mddev->delta_disks = 0;
5862 void md_set_array_sectors(struct mddev *mddev, sector_t array_sectors)
5864 WARN(!mddev_is_locked(mddev), "%s: unlocked mddev!\n", __func__);
5866 if (mddev->external_size)
5869 mddev->array_sectors = array_sectors;
5871 EXPORT_SYMBOL(md_set_array_sectors);
5873 static int update_size(struct mddev *mddev, sector_t num_sectors)
5875 struct md_rdev *rdev;
5877 int fit = (num_sectors == 0);
5879 if (mddev->pers->resize == NULL)
5881 /* The "num_sectors" is the number of sectors of each device that
5882 * is used. This can only make sense for arrays with redundancy.
5883 * linear and raid0 always use whatever space is available. We can only
5884 * consider changing this number if no resync or reconstruction is
5885 * happening, and if the new size is acceptable. It must fit before the
5886 * sb_start or, if that is <data_offset, it must fit before the size
5887 * of each device. If num_sectors is zero, we find the largest size
5890 if (mddev->sync_thread)
5893 /* Sorry, cannot grow a bitmap yet, just remove it,
5897 list_for_each_entry(rdev, &mddev->disks, same_set) {
5898 sector_t avail = rdev->sectors;
5900 if (fit && (num_sectors == 0 || num_sectors > avail))
5901 num_sectors = avail;
5902 if (avail < num_sectors)
5905 rv = mddev->pers->resize(mddev, num_sectors);
5907 revalidate_disk(mddev->gendisk);
5911 static int update_raid_disks(struct mddev *mddev, int raid_disks)
5914 /* change the number of raid disks */
5915 if (mddev->pers->check_reshape == NULL)
5917 if (raid_disks <= 0 ||
5918 (mddev->max_disks && raid_disks >= mddev->max_disks))
5920 if (mddev->sync_thread || mddev->reshape_position != MaxSector)
5922 mddev->delta_disks = raid_disks - mddev->raid_disks;
5924 rv = mddev->pers->check_reshape(mddev);
5926 mddev->delta_disks = 0;
5932 * update_array_info is used to change the configuration of an
5934 * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
5935 * fields in the info are checked against the array.
5936 * Any differences that cannot be handled will cause an error.
5937 * Normally, only one change can be managed at a time.
5939 static int update_array_info(struct mddev *mddev, mdu_array_info_t *info)
5945 /* calculate expected state,ignoring low bits */
5946 if (mddev->bitmap && mddev->bitmap_info.offset)
5947 state |= (1 << MD_SB_BITMAP_PRESENT);
5949 if (mddev->major_version != info->major_version ||
5950 mddev->minor_version != info->minor_version ||
5951 /* mddev->patch_version != info->patch_version || */
5952 mddev->ctime != info->ctime ||
5953 mddev->level != info->level ||
5954 /* mddev->layout != info->layout || */
5955 !mddev->persistent != info->not_persistent||
5956 mddev->chunk_sectors != info->chunk_size >> 9 ||
5957 /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
5958 ((state^info->state) & 0xfffffe00)
5961 /* Check there is only one change */
5962 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
5964 if (mddev->raid_disks != info->raid_disks)
5966 if (mddev->layout != info->layout)
5968 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT))
5975 if (mddev->layout != info->layout) {
5977 * we don't need to do anything at the md level, the
5978 * personality will take care of it all.
5980 if (mddev->pers->check_reshape == NULL)
5983 mddev->new_layout = info->layout;
5984 rv = mddev->pers->check_reshape(mddev);
5986 mddev->new_layout = mddev->layout;
5990 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
5991 rv = update_size(mddev, (sector_t)info->size * 2);
5993 if (mddev->raid_disks != info->raid_disks)
5994 rv = update_raid_disks(mddev, info->raid_disks);
5996 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) {
5997 if (mddev->pers->quiesce == NULL)
5999 if (mddev->recovery || mddev->sync_thread)
6001 if (info->state & (1<<MD_SB_BITMAP_PRESENT)) {
6002 /* add the bitmap */
6005 if (mddev->bitmap_info.default_offset == 0)
6007 mddev->bitmap_info.offset =
6008 mddev->bitmap_info.default_offset;
6009 mddev->pers->quiesce(mddev, 1);
6010 rv = bitmap_create(mddev);
6012 rv = bitmap_load(mddev);
6014 bitmap_destroy(mddev);
6015 mddev->pers->quiesce(mddev, 0);
6017 /* remove the bitmap */
6020 if (mddev->bitmap->file)
6022 mddev->pers->quiesce(mddev, 1);
6023 bitmap_destroy(mddev);
6024 mddev->pers->quiesce(mddev, 0);
6025 mddev->bitmap_info.offset = 0;
6028 md_update_sb(mddev, 1);
6032 static int set_disk_faulty(struct mddev *mddev, dev_t dev)
6034 struct md_rdev *rdev;
6036 if (mddev->pers == NULL)
6039 rdev = find_rdev(mddev, dev);
6043 md_error(mddev, rdev);
6044 if (!test_bit(Faulty, &rdev->flags))
6050 * We have a problem here : there is no easy way to give a CHS
6051 * virtual geometry. We currently pretend that we have a 2 heads
6052 * 4 sectors (with a BIG number of cylinders...). This drives
6053 * dosfs just mad... ;-)
6055 static int md_getgeo(struct block_device *bdev, struct hd_geometry *geo)
6057 struct mddev *mddev = bdev->bd_disk->private_data;
6061 geo->cylinders = mddev->array_sectors / 8;
6065 static int md_ioctl(struct block_device *bdev, fmode_t mode,
6066 unsigned int cmd, unsigned long arg)
6069 void __user *argp = (void __user *)arg;
6070 struct mddev *mddev = NULL;
6073 if (!capable(CAP_SYS_ADMIN))
6077 * Commands dealing with the RAID driver but not any
6083 err = get_version(argp);
6086 case PRINT_RAID_DEBUG:
6094 autostart_arrays(arg);
6101 * Commands creating/starting a new array:
6104 mddev = bdev->bd_disk->private_data;
6111 err = mddev_lock(mddev);
6114 "md: ioctl lock interrupted, reason %d, cmd %d\n",
6121 case SET_ARRAY_INFO:
6123 mdu_array_info_t info;
6125 memset(&info, 0, sizeof(info));
6126 else if (copy_from_user(&info, argp, sizeof(info))) {
6131 err = update_array_info(mddev, &info);
6133 printk(KERN_WARNING "md: couldn't update"
6134 " array info. %d\n", err);
6139 if (!list_empty(&mddev->disks)) {
6141 "md: array %s already has disks!\n",
6146 if (mddev->raid_disks) {
6148 "md: array %s already initialised!\n",
6153 err = set_array_info(mddev, &info);
6155 printk(KERN_WARNING "md: couldn't set"
6156 " array info. %d\n", err);
6166 * Commands querying/configuring an existing array:
6168 /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
6169 * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */
6170 if ((!mddev->raid_disks && !mddev->external)
6171 && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY
6172 && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE
6173 && cmd != GET_BITMAP_FILE) {
6179 * Commands even a read-only array can execute:
6183 case GET_ARRAY_INFO:
6184 err = get_array_info(mddev, argp);
6187 case GET_BITMAP_FILE:
6188 err = get_bitmap_file(mddev, argp);
6192 err = get_disk_info(mddev, argp);
6195 case RESTART_ARRAY_RW:
6196 err = restart_array(mddev);
6200 err = do_md_stop(mddev, 0, bdev);
6204 err = md_set_readonly(mddev, bdev);
6208 if (get_user(ro, (int __user *)(arg))) {
6214 /* if the bdev is going readonly the value of mddev->ro
6215 * does not matter, no writes are coming
6220 /* are we are already prepared for writes? */
6224 /* transitioning to readauto need only happen for
6225 * arrays that call md_write_start
6228 err = restart_array(mddev);
6231 set_disk_ro(mddev->gendisk, 0);
6238 * The remaining ioctls are changing the state of the
6239 * superblock, so we do not allow them on read-only arrays.
6240 * However non-MD ioctls (e.g. get-size) will still come through
6241 * here and hit the 'default' below, so only disallow
6242 * 'md' ioctls, and switch to rw mode if started auto-readonly.
6244 if (_IOC_TYPE(cmd) == MD_MAJOR && mddev->ro && mddev->pers) {
6245 if (mddev->ro == 2) {
6247 sysfs_notify_dirent_safe(mddev->sysfs_state);
6248 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6249 md_wakeup_thread(mddev->thread);
6260 mdu_disk_info_t info;
6261 if (copy_from_user(&info, argp, sizeof(info)))
6264 err = add_new_disk(mddev, &info);
6268 case HOT_REMOVE_DISK:
6269 err = hot_remove_disk(mddev, new_decode_dev(arg));
6273 err = hot_add_disk(mddev, new_decode_dev(arg));
6276 case SET_DISK_FAULTY:
6277 err = set_disk_faulty(mddev, new_decode_dev(arg));
6281 err = do_md_run(mddev);
6284 case SET_BITMAP_FILE:
6285 err = set_bitmap_file(mddev, (int)arg);
6295 if (mddev->hold_active == UNTIL_IOCTL &&
6297 mddev->hold_active = 0;
6298 mddev_unlock(mddev);
6307 #ifdef CONFIG_COMPAT
6308 static int md_compat_ioctl(struct block_device *bdev, fmode_t mode,
6309 unsigned int cmd, unsigned long arg)
6312 case HOT_REMOVE_DISK:
6314 case SET_DISK_FAULTY:
6315 case SET_BITMAP_FILE:
6316 /* These take in integer arg, do not convert */
6319 arg = (unsigned long)compat_ptr(arg);
6323 return md_ioctl(bdev, mode, cmd, arg);
6325 #endif /* CONFIG_COMPAT */
6327 static int md_open(struct block_device *bdev, fmode_t mode)
6330 * Succeed if we can lock the mddev, which confirms that
6331 * it isn't being stopped right now.
6333 struct mddev *mddev = mddev_find(bdev->bd_dev);
6336 if (mddev->gendisk != bdev->bd_disk) {
6337 /* we are racing with mddev_put which is discarding this
6341 /* Wait until bdev->bd_disk is definitely gone */
6342 flush_workqueue(md_misc_wq);
6343 /* Then retry the open from the top */
6344 return -ERESTARTSYS;
6346 BUG_ON(mddev != bdev->bd_disk->private_data);
6348 if ((err = mutex_lock_interruptible(&mddev->open_mutex)))
6352 atomic_inc(&mddev->openers);
6353 mutex_unlock(&mddev->open_mutex);
6355 check_disk_change(bdev);
6360 static int md_release(struct gendisk *disk, fmode_t mode)
6362 struct mddev *mddev = disk->private_data;
6365 atomic_dec(&mddev->openers);
6371 static int md_media_changed(struct gendisk *disk)
6373 struct mddev *mddev = disk->private_data;
6375 return mddev->changed;
6378 static int md_revalidate(struct gendisk *disk)
6380 struct mddev *mddev = disk->private_data;
6385 static const struct block_device_operations md_fops =
6387 .owner = THIS_MODULE,
6389 .release = md_release,
6391 #ifdef CONFIG_COMPAT
6392 .compat_ioctl = md_compat_ioctl,
6394 .getgeo = md_getgeo,
6395 .media_changed = md_media_changed,
6396 .revalidate_disk= md_revalidate,
6399 static int md_thread(void * arg)
6401 struct md_thread *thread = arg;
6404 * md_thread is a 'system-thread', it's priority should be very
6405 * high. We avoid resource deadlocks individually in each
6406 * raid personality. (RAID5 does preallocation) We also use RR and
6407 * the very same RT priority as kswapd, thus we will never get
6408 * into a priority inversion deadlock.
6410 * we definitely have to have equal or higher priority than
6411 * bdflush, otherwise bdflush will deadlock if there are too
6412 * many dirty RAID5 blocks.
6415 allow_signal(SIGKILL);
6416 while (!kthread_should_stop()) {
6418 /* We need to wait INTERRUPTIBLE so that
6419 * we don't add to the load-average.
6420 * That means we need to be sure no signals are
6423 if (signal_pending(current))
6424 flush_signals(current);
6426 wait_event_interruptible_timeout
6428 test_bit(THREAD_WAKEUP, &thread->flags)
6429 || kthread_should_stop(),
6432 clear_bit(THREAD_WAKEUP, &thread->flags);
6433 if (!kthread_should_stop())
6434 thread->run(thread->mddev);
6440 void md_wakeup_thread(struct md_thread *thread)
6443 pr_debug("md: waking up MD thread %s.\n", thread->tsk->comm);
6444 set_bit(THREAD_WAKEUP, &thread->flags);
6445 wake_up(&thread->wqueue);
6449 struct md_thread *md_register_thread(void (*run) (struct mddev *), struct mddev *mddev,
6452 struct md_thread *thread;
6454 thread = kzalloc(sizeof(struct md_thread), GFP_KERNEL);
6458 init_waitqueue_head(&thread->wqueue);
6461 thread->mddev = mddev;
6462 thread->timeout = MAX_SCHEDULE_TIMEOUT;
6463 thread->tsk = kthread_run(md_thread, thread,
6465 mdname(thread->mddev),
6466 name ?: mddev->pers->name);
6467 if (IS_ERR(thread->tsk)) {
6474 void md_unregister_thread(struct md_thread **threadp)
6476 struct md_thread *thread = *threadp;
6479 pr_debug("interrupting MD-thread pid %d\n", task_pid_nr(thread->tsk));
6480 /* Locking ensures that mddev_unlock does not wake_up a
6481 * non-existent thread
6483 spin_lock(&pers_lock);
6485 spin_unlock(&pers_lock);
6487 kthread_stop(thread->tsk);
6491 void md_error(struct mddev *mddev, struct md_rdev *rdev)
6498 if (!rdev || test_bit(Faulty, &rdev->flags))
6501 if (!mddev->pers || !mddev->pers->error_handler)
6503 mddev->pers->error_handler(mddev,rdev);
6504 if (mddev->degraded)
6505 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
6506 sysfs_notify_dirent_safe(rdev->sysfs_state);
6507 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6508 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6509 md_wakeup_thread(mddev->thread);
6510 if (mddev->event_work.func)
6511 queue_work(md_misc_wq, &mddev->event_work);
6512 md_new_event_inintr(mddev);
6515 /* seq_file implementation /proc/mdstat */
6517 static void status_unused(struct seq_file *seq)
6520 struct md_rdev *rdev;
6522 seq_printf(seq, "unused devices: ");
6524 list_for_each_entry(rdev, &pending_raid_disks, same_set) {
6525 char b[BDEVNAME_SIZE];
6527 seq_printf(seq, "%s ",
6528 bdevname(rdev->bdev,b));
6531 seq_printf(seq, "<none>");
6533 seq_printf(seq, "\n");
6537 static void status_resync(struct seq_file *seq, struct mddev * mddev)
6539 sector_t max_sectors, resync, res;
6540 unsigned long dt, db;
6543 unsigned int per_milli;
6545 resync = mddev->curr_resync - atomic_read(&mddev->recovery_active);
6547 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
6548 max_sectors = mddev->resync_max_sectors;
6550 max_sectors = mddev->dev_sectors;
6553 * Should not happen.
6559 /* Pick 'scale' such that (resync>>scale)*1000 will fit
6560 * in a sector_t, and (max_sectors>>scale) will fit in a
6561 * u32, as those are the requirements for sector_div.
6562 * Thus 'scale' must be at least 10
6565 if (sizeof(sector_t) > sizeof(unsigned long)) {
6566 while ( max_sectors/2 > (1ULL<<(scale+32)))
6569 res = (resync>>scale)*1000;
6570 sector_div(res, (u32)((max_sectors>>scale)+1));
6574 int i, x = per_milli/50, y = 20-x;
6575 seq_printf(seq, "[");
6576 for (i = 0; i < x; i++)
6577 seq_printf(seq, "=");
6578 seq_printf(seq, ">");
6579 for (i = 0; i < y; i++)
6580 seq_printf(seq, ".");
6581 seq_printf(seq, "] ");
6583 seq_printf(seq, " %s =%3u.%u%% (%llu/%llu)",
6584 (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)?
6586 (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)?
6588 (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
6589 "resync" : "recovery"))),
6590 per_milli/10, per_milli % 10,
6591 (unsigned long long) resync/2,
6592 (unsigned long long) max_sectors/2);
6595 * dt: time from mark until now
6596 * db: blocks written from mark until now
6597 * rt: remaining time
6599 * rt is a sector_t, so could be 32bit or 64bit.
6600 * So we divide before multiply in case it is 32bit and close
6602 * We scale the divisor (db) by 32 to avoid losing precision
6603 * near the end of resync when the number of remaining sectors
6605 * We then divide rt by 32 after multiplying by db to compensate.
6606 * The '+1' avoids division by zero if db is very small.
6608 dt = ((jiffies - mddev->resync_mark) / HZ);
6610 db = (mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active))
6611 - mddev->resync_mark_cnt;
6613 rt = max_sectors - resync; /* number of remaining sectors */
6614 sector_div(rt, db/32+1);
6618 seq_printf(seq, " finish=%lu.%lumin", (unsigned long)rt / 60,
6619 ((unsigned long)rt % 60)/6);
6621 seq_printf(seq, " speed=%ldK/sec", db/2/dt);
6624 static void *md_seq_start(struct seq_file *seq, loff_t *pos)
6626 struct list_head *tmp;
6628 struct mddev *mddev;
6636 spin_lock(&all_mddevs_lock);
6637 list_for_each(tmp,&all_mddevs)
6639 mddev = list_entry(tmp, struct mddev, all_mddevs);
6641 spin_unlock(&all_mddevs_lock);
6644 spin_unlock(&all_mddevs_lock);
6646 return (void*)2;/* tail */
6650 static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
6652 struct list_head *tmp;
6653 struct mddev *next_mddev, *mddev = v;
6659 spin_lock(&all_mddevs_lock);
6661 tmp = all_mddevs.next;
6663 tmp = mddev->all_mddevs.next;
6664 if (tmp != &all_mddevs)
6665 next_mddev = mddev_get(list_entry(tmp,struct mddev,all_mddevs));
6667 next_mddev = (void*)2;
6670 spin_unlock(&all_mddevs_lock);
6678 static void md_seq_stop(struct seq_file *seq, void *v)
6680 struct mddev *mddev = v;
6682 if (mddev && v != (void*)1 && v != (void*)2)
6686 static int md_seq_show(struct seq_file *seq, void *v)
6688 struct mddev *mddev = v;
6690 struct md_rdev *rdev;
6691 struct bitmap *bitmap;
6693 if (v == (void*)1) {
6694 struct md_personality *pers;
6695 seq_printf(seq, "Personalities : ");
6696 spin_lock(&pers_lock);
6697 list_for_each_entry(pers, &pers_list, list)
6698 seq_printf(seq, "[%s] ", pers->name);
6700 spin_unlock(&pers_lock);
6701 seq_printf(seq, "\n");
6702 seq->poll_event = atomic_read(&md_event_count);
6705 if (v == (void*)2) {
6710 if (mddev_lock(mddev) < 0)
6713 if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
6714 seq_printf(seq, "%s : %sactive", mdname(mddev),
6715 mddev->pers ? "" : "in");
6718 seq_printf(seq, " (read-only)");
6720 seq_printf(seq, " (auto-read-only)");
6721 seq_printf(seq, " %s", mddev->pers->name);
6725 list_for_each_entry(rdev, &mddev->disks, same_set) {
6726 char b[BDEVNAME_SIZE];
6727 seq_printf(seq, " %s[%d]",
6728 bdevname(rdev->bdev,b), rdev->desc_nr);
6729 if (test_bit(WriteMostly, &rdev->flags))
6730 seq_printf(seq, "(W)");
6731 if (test_bit(Faulty, &rdev->flags)) {
6732 seq_printf(seq, "(F)");
6734 } else if (rdev->raid_disk < 0)
6735 seq_printf(seq, "(S)"); /* spare */
6736 sectors += rdev->sectors;
6739 if (!list_empty(&mddev->disks)) {
6741 seq_printf(seq, "\n %llu blocks",
6742 (unsigned long long)
6743 mddev->array_sectors / 2);
6745 seq_printf(seq, "\n %llu blocks",
6746 (unsigned long long)sectors / 2);
6748 if (mddev->persistent) {
6749 if (mddev->major_version != 0 ||
6750 mddev->minor_version != 90) {
6751 seq_printf(seq," super %d.%d",
6752 mddev->major_version,
6753 mddev->minor_version);
6755 } else if (mddev->external)
6756 seq_printf(seq, " super external:%s",
6757 mddev->metadata_type);
6759 seq_printf(seq, " super non-persistent");
6762 mddev->pers->status(seq, mddev);
6763 seq_printf(seq, "\n ");
6764 if (mddev->pers->sync_request) {
6765 if (mddev->curr_resync > 2) {
6766 status_resync(seq, mddev);
6767 seq_printf(seq, "\n ");
6768 } else if (mddev->curr_resync == 1 || mddev->curr_resync == 2)
6769 seq_printf(seq, "\tresync=DELAYED\n ");
6770 else if (mddev->recovery_cp < MaxSector)
6771 seq_printf(seq, "\tresync=PENDING\n ");
6774 seq_printf(seq, "\n ");
6776 if ((bitmap = mddev->bitmap)) {
6777 unsigned long chunk_kb;
6778 unsigned long flags;
6779 spin_lock_irqsave(&bitmap->lock, flags);
6780 chunk_kb = mddev->bitmap_info.chunksize >> 10;
6781 seq_printf(seq, "bitmap: %lu/%lu pages [%luKB], "
6783 bitmap->pages - bitmap->missing_pages,
6785 (bitmap->pages - bitmap->missing_pages)
6786 << (PAGE_SHIFT - 10),
6787 chunk_kb ? chunk_kb : mddev->bitmap_info.chunksize,
6788 chunk_kb ? "KB" : "B");
6790 seq_printf(seq, ", file: ");
6791 seq_path(seq, &bitmap->file->f_path, " \t\n");
6794 seq_printf(seq, "\n");
6795 spin_unlock_irqrestore(&bitmap->lock, flags);
6798 seq_printf(seq, "\n");
6800 mddev_unlock(mddev);
6805 static const struct seq_operations md_seq_ops = {
6806 .start = md_seq_start,
6807 .next = md_seq_next,
6808 .stop = md_seq_stop,
6809 .show = md_seq_show,
6812 static int md_seq_open(struct inode *inode, struct file *file)
6814 struct seq_file *seq;
6817 error = seq_open(file, &md_seq_ops);
6821 seq = file->private_data;
6822 seq->poll_event = atomic_read(&md_event_count);
6826 static unsigned int mdstat_poll(struct file *filp, poll_table *wait)
6828 struct seq_file *seq = filp->private_data;
6831 poll_wait(filp, &md_event_waiters, wait);
6833 /* always allow read */
6834 mask = POLLIN | POLLRDNORM;
6836 if (seq->poll_event != atomic_read(&md_event_count))
6837 mask |= POLLERR | POLLPRI;
6841 static const struct file_operations md_seq_fops = {
6842 .owner = THIS_MODULE,
6843 .open = md_seq_open,
6845 .llseek = seq_lseek,
6846 .release = seq_release_private,
6847 .poll = mdstat_poll,
6850 int register_md_personality(struct md_personality *p)
6852 spin_lock(&pers_lock);
6853 list_add_tail(&p->list, &pers_list);
6854 printk(KERN_INFO "md: %s personality registered for level %d\n", p->name, p->level);
6855 spin_unlock(&pers_lock);
6859 int unregister_md_personality(struct md_personality *p)
6861 printk(KERN_INFO "md: %s personality unregistered\n", p->name);
6862 spin_lock(&pers_lock);
6863 list_del_init(&p->list);
6864 spin_unlock(&pers_lock);
6868 static int is_mddev_idle(struct mddev *mddev, int init)
6870 struct md_rdev * rdev;
6876 rdev_for_each_rcu(rdev, mddev) {
6877 struct gendisk *disk = rdev->bdev->bd_contains->bd_disk;
6878 curr_events = (int)part_stat_read(&disk->part0, sectors[0]) +
6879 (int)part_stat_read(&disk->part0, sectors[1]) -
6880 atomic_read(&disk->sync_io);
6881 /* sync IO will cause sync_io to increase before the disk_stats
6882 * as sync_io is counted when a request starts, and
6883 * disk_stats is counted when it completes.
6884 * So resync activity will cause curr_events to be smaller than
6885 * when there was no such activity.
6886 * non-sync IO will cause disk_stat to increase without
6887 * increasing sync_io so curr_events will (eventually)
6888 * be larger than it was before. Once it becomes
6889 * substantially larger, the test below will cause
6890 * the array to appear non-idle, and resync will slow
6892 * If there is a lot of outstanding resync activity when
6893 * we set last_event to curr_events, then all that activity
6894 * completing might cause the array to appear non-idle
6895 * and resync will be slowed down even though there might
6896 * not have been non-resync activity. This will only
6897 * happen once though. 'last_events' will soon reflect
6898 * the state where there is little or no outstanding
6899 * resync requests, and further resync activity will
6900 * always make curr_events less than last_events.
6903 if (init || curr_events - rdev->last_events > 64) {
6904 rdev->last_events = curr_events;
6912 void md_done_sync(struct mddev *mddev, int blocks, int ok)
6914 /* another "blocks" (512byte) blocks have been synced */
6915 atomic_sub(blocks, &mddev->recovery_active);
6916 wake_up(&mddev->recovery_wait);
6918 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6919 md_wakeup_thread(mddev->thread);
6920 // stop recovery, signal do_sync ....
6925 /* md_write_start(mddev, bi)
6926 * If we need to update some array metadata (e.g. 'active' flag
6927 * in superblock) before writing, schedule a superblock update
6928 * and wait for it to complete.
6930 void md_write_start(struct mddev *mddev, struct bio *bi)
6933 if (bio_data_dir(bi) != WRITE)
6936 BUG_ON(mddev->ro == 1);
6937 if (mddev->ro == 2) {
6938 /* need to switch to read/write */
6940 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6941 md_wakeup_thread(mddev->thread);
6942 md_wakeup_thread(mddev->sync_thread);
6945 atomic_inc(&mddev->writes_pending);
6946 if (mddev->safemode == 1)
6947 mddev->safemode = 0;
6948 if (mddev->in_sync) {
6949 spin_lock_irq(&mddev->write_lock);
6950 if (mddev->in_sync) {
6952 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
6953 set_bit(MD_CHANGE_PENDING, &mddev->flags);
6954 md_wakeup_thread(mddev->thread);
6957 spin_unlock_irq(&mddev->write_lock);
6960 sysfs_notify_dirent_safe(mddev->sysfs_state);
6961 wait_event(mddev->sb_wait,
6962 !test_bit(MD_CHANGE_PENDING, &mddev->flags));
6965 void md_write_end(struct mddev *mddev)
6967 if (atomic_dec_and_test(&mddev->writes_pending)) {
6968 if (mddev->safemode == 2)
6969 md_wakeup_thread(mddev->thread);
6970 else if (mddev->safemode_delay)
6971 mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay);
6975 /* md_allow_write(mddev)
6976 * Calling this ensures that the array is marked 'active' so that writes
6977 * may proceed without blocking. It is important to call this before
6978 * attempting a GFP_KERNEL allocation while holding the mddev lock.
6979 * Must be called with mddev_lock held.
6981 * In the ->external case MD_CHANGE_CLEAN can not be cleared until mddev->lock
6982 * is dropped, so return -EAGAIN after notifying userspace.
6984 int md_allow_write(struct mddev *mddev)
6990 if (!mddev->pers->sync_request)
6993 spin_lock_irq(&mddev->write_lock);
6994 if (mddev->in_sync) {
6996 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
6997 set_bit(MD_CHANGE_PENDING, &mddev->flags);
6998 if (mddev->safemode_delay &&
6999 mddev->safemode == 0)
7000 mddev->safemode = 1;
7001 spin_unlock_irq(&mddev->write_lock);
7002 md_update_sb(mddev, 0);
7003 sysfs_notify_dirent_safe(mddev->sysfs_state);
7005 spin_unlock_irq(&mddev->write_lock);
7007 if (test_bit(MD_CHANGE_PENDING, &mddev->flags))
7012 EXPORT_SYMBOL_GPL(md_allow_write);
7014 #define SYNC_MARKS 10
7015 #define SYNC_MARK_STEP (3*HZ)
7016 void md_do_sync(struct mddev *mddev)
7018 struct mddev *mddev2;
7019 unsigned int currspeed = 0,
7021 sector_t max_sectors,j, io_sectors;
7022 unsigned long mark[SYNC_MARKS];
7023 sector_t mark_cnt[SYNC_MARKS];
7025 struct list_head *tmp;
7026 sector_t last_check;
7028 struct md_rdev *rdev;
7031 /* just incase thread restarts... */
7032 if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
7034 if (mddev->ro) /* never try to sync a read-only array */
7037 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
7038 if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
7039 desc = "data-check";
7040 else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
7041 desc = "requested-resync";
7044 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
7049 /* we overload curr_resync somewhat here.
7050 * 0 == not engaged in resync at all
7051 * 2 == checking that there is no conflict with another sync
7052 * 1 == like 2, but have yielded to allow conflicting resync to
7054 * other == active in resync - this many blocks
7056 * Before starting a resync we must have set curr_resync to
7057 * 2, and then checked that every "conflicting" array has curr_resync
7058 * less than ours. When we find one that is the same or higher
7059 * we wait on resync_wait. To avoid deadlock, we reduce curr_resync
7060 * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
7061 * This will mean we have to start checking from the beginning again.
7066 mddev->curr_resync = 2;
7069 if (kthread_should_stop())
7070 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
7072 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
7074 for_each_mddev(mddev2, tmp) {
7075 if (mddev2 == mddev)
7077 if (!mddev->parallel_resync
7078 && mddev2->curr_resync
7079 && match_mddev_units(mddev, mddev2)) {
7081 if (mddev < mddev2 && mddev->curr_resync == 2) {
7082 /* arbitrarily yield */
7083 mddev->curr_resync = 1;
7084 wake_up(&resync_wait);
7086 if (mddev > mddev2 && mddev->curr_resync == 1)
7087 /* no need to wait here, we can wait the next
7088 * time 'round when curr_resync == 2
7091 /* We need to wait 'interruptible' so as not to
7092 * contribute to the load average, and not to
7093 * be caught by 'softlockup'
7095 prepare_to_wait(&resync_wait, &wq, TASK_INTERRUPTIBLE);
7096 if (!kthread_should_stop() &&
7097 mddev2->curr_resync >= mddev->curr_resync) {
7098 printk(KERN_INFO "md: delaying %s of %s"
7099 " until %s has finished (they"
7100 " share one or more physical units)\n",
7101 desc, mdname(mddev), mdname(mddev2));
7103 if (signal_pending(current))
7104 flush_signals(current);
7106 finish_wait(&resync_wait, &wq);
7109 finish_wait(&resync_wait, &wq);
7112 } while (mddev->curr_resync < 2);
7115 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
7116 /* resync follows the size requested by the personality,
7117 * which defaults to physical size, but can be virtual size
7119 max_sectors = mddev->resync_max_sectors;
7120 mddev->resync_mismatches = 0;
7121 /* we don't use the checkpoint if there's a bitmap */
7122 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
7123 j = mddev->resync_min;
7124 else if (!mddev->bitmap)
7125 j = mddev->recovery_cp;
7127 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
7128 max_sectors = mddev->dev_sectors;
7130 /* recovery follows the physical size of devices */
7131 max_sectors = mddev->dev_sectors;
7134 list_for_each_entry_rcu(rdev, &mddev->disks, same_set)
7135 if (rdev->raid_disk >= 0 &&
7136 !test_bit(Faulty, &rdev->flags) &&
7137 !test_bit(In_sync, &rdev->flags) &&
7138 rdev->recovery_offset < j)
7139 j = rdev->recovery_offset;
7143 printk(KERN_INFO "md: %s of RAID array %s\n", desc, mdname(mddev));
7144 printk(KERN_INFO "md: minimum _guaranteed_ speed:"
7145 " %d KB/sec/disk.\n", speed_min(mddev));
7146 printk(KERN_INFO "md: using maximum available idle IO bandwidth "
7147 "(but not more than %d KB/sec) for %s.\n",
7148 speed_max(mddev), desc);
7150 is_mddev_idle(mddev, 1); /* this initializes IO event counters */
7153 for (m = 0; m < SYNC_MARKS; m++) {
7155 mark_cnt[m] = io_sectors;
7158 mddev->resync_mark = mark[last_mark];
7159 mddev->resync_mark_cnt = mark_cnt[last_mark];
7162 * Tune reconstruction:
7164 window = 32*(PAGE_SIZE/512);
7165 printk(KERN_INFO "md: using %dk window, over a total of %lluk.\n",
7166 window/2, (unsigned long long)max_sectors/2);
7168 atomic_set(&mddev->recovery_active, 0);
7173 "md: resuming %s of %s from checkpoint.\n",
7174 desc, mdname(mddev));
7175 mddev->curr_resync = j;
7177 mddev->curr_resync_completed = j;
7179 while (j < max_sectors) {
7184 if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
7185 ((mddev->curr_resync > mddev->curr_resync_completed &&
7186 (mddev->curr_resync - mddev->curr_resync_completed)
7187 > (max_sectors >> 4)) ||
7188 (j - mddev->curr_resync_completed)*2
7189 >= mddev->resync_max - mddev->curr_resync_completed
7191 /* time to update curr_resync_completed */
7192 wait_event(mddev->recovery_wait,
7193 atomic_read(&mddev->recovery_active) == 0);
7194 mddev->curr_resync_completed = j;
7195 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
7196 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
7199 while (j >= mddev->resync_max && !kthread_should_stop()) {
7200 /* As this condition is controlled by user-space,
7201 * we can block indefinitely, so use '_interruptible'
7202 * to avoid triggering warnings.
7204 flush_signals(current); /* just in case */
7205 wait_event_interruptible(mddev->recovery_wait,
7206 mddev->resync_max > j
7207 || kthread_should_stop());
7210 if (kthread_should_stop())
7213 sectors = mddev->pers->sync_request(mddev, j, &skipped,
7214 currspeed < speed_min(mddev));
7216 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
7220 if (!skipped) { /* actual IO requested */
7221 io_sectors += sectors;
7222 atomic_add(sectors, &mddev->recovery_active);
7225 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
7229 if (j>1) mddev->curr_resync = j;
7230 mddev->curr_mark_cnt = io_sectors;
7231 if (last_check == 0)
7232 /* this is the earliest that rebuild will be
7233 * visible in /proc/mdstat
7235 md_new_event(mddev);
7237 if (last_check + window > io_sectors || j == max_sectors)
7240 last_check = io_sectors;
7242 if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) {
7244 int next = (last_mark+1) % SYNC_MARKS;
7246 mddev->resync_mark = mark[next];
7247 mddev->resync_mark_cnt = mark_cnt[next];
7248 mark[next] = jiffies;
7249 mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active);
7254 if (kthread_should_stop())
7259 * this loop exits only if either when we are slower than
7260 * the 'hard' speed limit, or the system was IO-idle for
7262 * the system might be non-idle CPU-wise, but we only care
7263 * about not overloading the IO subsystem. (things like an
7264 * e2fsck being done on the RAID array should execute fast)
7268 currspeed = ((unsigned long)(io_sectors-mddev->resync_mark_cnt))/2
7269 /((jiffies-mddev->resync_mark)/HZ +1) +1;
7271 if (currspeed > speed_min(mddev)) {
7272 if ((currspeed > speed_max(mddev)) ||
7273 !is_mddev_idle(mddev, 0)) {
7279 printk(KERN_INFO "md: %s: %s done.\n",mdname(mddev), desc);
7281 * this also signals 'finished resyncing' to md_stop
7284 wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
7286 /* tell personality that we are finished */
7287 mddev->pers->sync_request(mddev, max_sectors, &skipped, 1);
7289 if (!test_bit(MD_RECOVERY_CHECK, &mddev->recovery) &&
7290 mddev->curr_resync > 2) {
7291 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
7292 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
7293 if (mddev->curr_resync >= mddev->recovery_cp) {
7295 "md: checkpointing %s of %s.\n",
7296 desc, mdname(mddev));
7297 mddev->recovery_cp = mddev->curr_resync;
7300 mddev->recovery_cp = MaxSector;
7302 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery))
7303 mddev->curr_resync = MaxSector;
7305 list_for_each_entry_rcu(rdev, &mddev->disks, same_set)
7306 if (rdev->raid_disk >= 0 &&
7307 mddev->delta_disks >= 0 &&
7308 !test_bit(Faulty, &rdev->flags) &&
7309 !test_bit(In_sync, &rdev->flags) &&
7310 rdev->recovery_offset < mddev->curr_resync)
7311 rdev->recovery_offset = mddev->curr_resync;
7315 set_bit(MD_CHANGE_DEVS, &mddev->flags);
7318 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
7319 /* We completed so min/max setting can be forgotten if used. */
7320 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
7321 mddev->resync_min = 0;
7322 mddev->resync_max = MaxSector;
7323 } else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
7324 mddev->resync_min = mddev->curr_resync_completed;
7325 mddev->curr_resync = 0;
7326 wake_up(&resync_wait);
7327 set_bit(MD_RECOVERY_DONE, &mddev->recovery);
7328 md_wakeup_thread(mddev->thread);
7333 * got a signal, exit.
7336 "md: md_do_sync() got signal ... exiting\n");
7337 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
7341 EXPORT_SYMBOL_GPL(md_do_sync);
7343 static int remove_and_add_spares(struct mddev *mddev)
7345 struct md_rdev *rdev;
7348 mddev->curr_resync_completed = 0;
7350 list_for_each_entry(rdev, &mddev->disks, same_set)
7351 if (rdev->raid_disk >= 0 &&
7352 !test_bit(Blocked, &rdev->flags) &&
7353 (test_bit(Faulty, &rdev->flags) ||
7354 ! test_bit(In_sync, &rdev->flags)) &&
7355 atomic_read(&rdev->nr_pending)==0) {
7356 if (mddev->pers->hot_remove_disk(
7357 mddev, rdev->raid_disk)==0) {
7358 sysfs_unlink_rdev(mddev, rdev);
7359 rdev->raid_disk = -1;
7363 if (mddev->degraded) {
7364 list_for_each_entry(rdev, &mddev->disks, same_set) {
7365 if (rdev->raid_disk >= 0 &&
7366 !test_bit(In_sync, &rdev->flags) &&
7367 !test_bit(Faulty, &rdev->flags))
7369 if (rdev->raid_disk < 0
7370 && !test_bit(Faulty, &rdev->flags)) {
7371 rdev->recovery_offset = 0;
7373 hot_add_disk(mddev, rdev) == 0) {
7374 if (sysfs_link_rdev(mddev, rdev))
7375 /* failure here is OK */;
7377 md_new_event(mddev);
7378 set_bit(MD_CHANGE_DEVS, &mddev->flags);
7386 static void reap_sync_thread(struct mddev *mddev)
7388 struct md_rdev *rdev;
7390 /* resync has finished, collect result */
7391 md_unregister_thread(&mddev->sync_thread);
7392 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
7393 !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
7395 /* activate any spares */
7396 if (mddev->pers->spare_active(mddev))
7397 sysfs_notify(&mddev->kobj, NULL,
7400 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
7401 mddev->pers->finish_reshape)
7402 mddev->pers->finish_reshape(mddev);
7404 /* If array is no-longer degraded, then any saved_raid_disk
7405 * information must be scrapped. Also if any device is now
7406 * In_sync we must scrape the saved_raid_disk for that device
7407 * do the superblock for an incrementally recovered device
7410 list_for_each_entry(rdev, &mddev->disks, same_set)
7411 if (!mddev->degraded ||
7412 test_bit(In_sync, &rdev->flags))
7413 rdev->saved_raid_disk = -1;
7415 md_update_sb(mddev, 1);
7416 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
7417 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
7418 clear_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
7419 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
7420 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
7421 /* flag recovery needed just to double check */
7422 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7423 sysfs_notify_dirent_safe(mddev->sysfs_action);
7424 md_new_event(mddev);
7425 if (mddev->event_work.func)
7426 queue_work(md_misc_wq, &mddev->event_work);
7430 * This routine is regularly called by all per-raid-array threads to
7431 * deal with generic issues like resync and super-block update.
7432 * Raid personalities that don't have a thread (linear/raid0) do not
7433 * need this as they never do any recovery or update the superblock.
7435 * It does not do any resync itself, but rather "forks" off other threads
7436 * to do that as needed.
7437 * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
7438 * "->recovery" and create a thread at ->sync_thread.
7439 * When the thread finishes it sets MD_RECOVERY_DONE
7440 * and wakeups up this thread which will reap the thread and finish up.
7441 * This thread also removes any faulty devices (with nr_pending == 0).
7443 * The overall approach is:
7444 * 1/ if the superblock needs updating, update it.
7445 * 2/ If a recovery thread is running, don't do anything else.
7446 * 3/ If recovery has finished, clean up, possibly marking spares active.
7447 * 4/ If there are any faulty devices, remove them.
7448 * 5/ If array is degraded, try to add spares devices
7449 * 6/ If array has spares or is not in-sync, start a resync thread.
7451 void md_check_recovery(struct mddev *mddev)
7453 if (mddev->suspended)
7457 bitmap_daemon_work(mddev);
7459 if (signal_pending(current)) {
7460 if (mddev->pers->sync_request && !mddev->external) {
7461 printk(KERN_INFO "md: %s in immediate safe mode\n",
7463 mddev->safemode = 2;
7465 flush_signals(current);
7468 if (mddev->ro && !test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
7471 (mddev->flags & ~ (1<<MD_CHANGE_PENDING)) ||
7472 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
7473 test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
7474 (mddev->external == 0 && mddev->safemode == 1) ||
7475 (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending)
7476 && !mddev->in_sync && mddev->recovery_cp == MaxSector)
7480 if (mddev_trylock(mddev)) {
7484 /* Only thing we do on a ro array is remove
7487 struct md_rdev *rdev;
7488 list_for_each_entry(rdev, &mddev->disks, same_set)
7489 if (rdev->raid_disk >= 0 &&
7490 !test_bit(Blocked, &rdev->flags) &&
7491 test_bit(Faulty, &rdev->flags) &&
7492 atomic_read(&rdev->nr_pending)==0) {
7493 if (mddev->pers->hot_remove_disk(
7494 mddev, rdev->raid_disk)==0) {
7495 sysfs_unlink_rdev(mddev, rdev);
7496 rdev->raid_disk = -1;
7499 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7503 if (!mddev->external) {
7505 spin_lock_irq(&mddev->write_lock);
7506 if (mddev->safemode &&
7507 !atomic_read(&mddev->writes_pending) &&
7509 mddev->recovery_cp == MaxSector) {
7512 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
7514 if (mddev->safemode == 1)
7515 mddev->safemode = 0;
7516 spin_unlock_irq(&mddev->write_lock);
7518 sysfs_notify_dirent_safe(mddev->sysfs_state);
7522 md_update_sb(mddev, 0);
7524 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
7525 !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
7526 /* resync/recovery still happening */
7527 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7530 if (mddev->sync_thread) {
7531 reap_sync_thread(mddev);
7534 /* Set RUNNING before clearing NEEDED to avoid
7535 * any transients in the value of "sync_action".
7537 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
7538 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7539 /* Clear some bits that don't mean anything, but
7542 clear_bit(MD_RECOVERY_INTR, &mddev->recovery);
7543 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
7545 if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
7547 /* no recovery is running.
7548 * remove any failed drives, then
7549 * add spares if possible.
7550 * Spare are also removed and re-added, to allow
7551 * the personality to fail the re-add.
7554 if (mddev->reshape_position != MaxSector) {
7555 if (mddev->pers->check_reshape == NULL ||
7556 mddev->pers->check_reshape(mddev) != 0)
7557 /* Cannot proceed */
7559 set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
7560 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
7561 } else if ((spares = remove_and_add_spares(mddev))) {
7562 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
7563 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
7564 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
7565 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
7566 } else if (mddev->recovery_cp < MaxSector) {
7567 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
7568 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
7569 } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
7570 /* nothing to be done ... */
7573 if (mddev->pers->sync_request) {
7574 if (spares && mddev->bitmap && ! mddev->bitmap->file) {
7575 /* We are adding a device or devices to an array
7576 * which has the bitmap stored on all devices.
7577 * So make sure all bitmap pages get written
7579 bitmap_write_all(mddev->bitmap);
7581 mddev->sync_thread = md_register_thread(md_do_sync,
7584 if (!mddev->sync_thread) {
7585 printk(KERN_ERR "%s: could not start resync"
7588 /* leave the spares where they are, it shouldn't hurt */
7589 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
7590 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
7591 clear_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
7592 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
7593 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
7595 md_wakeup_thread(mddev->sync_thread);
7596 sysfs_notify_dirent_safe(mddev->sysfs_action);
7597 md_new_event(mddev);
7600 if (!mddev->sync_thread) {
7601 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
7602 if (test_and_clear_bit(MD_RECOVERY_RECOVER,
7604 if (mddev->sysfs_action)
7605 sysfs_notify_dirent_safe(mddev->sysfs_action);
7607 mddev_unlock(mddev);
7611 void md_wait_for_blocked_rdev(struct md_rdev *rdev, struct mddev *mddev)
7613 sysfs_notify_dirent_safe(rdev->sysfs_state);
7614 wait_event_timeout(rdev->blocked_wait,
7615 !test_bit(Blocked, &rdev->flags) &&
7616 !test_bit(BlockedBadBlocks, &rdev->flags),
7617 msecs_to_jiffies(5000));
7618 rdev_dec_pending(rdev, mddev);
7620 EXPORT_SYMBOL(md_wait_for_blocked_rdev);
7623 /* Bad block management.
7624 * We can record which blocks on each device are 'bad' and so just
7625 * fail those blocks, or that stripe, rather than the whole device.
7626 * Entries in the bad-block table are 64bits wide. This comprises:
7627 * Length of bad-range, in sectors: 0-511 for lengths 1-512
7628 * Start of bad-range, sector offset, 54 bits (allows 8 exbibytes)
7629 * A 'shift' can be set so that larger blocks are tracked and
7630 * consequently larger devices can be covered.
7631 * 'Acknowledged' flag - 1 bit. - the most significant bit.
7633 * Locking of the bad-block table uses a seqlock so md_is_badblock
7634 * might need to retry if it is very unlucky.
7635 * We will sometimes want to check for bad blocks in a bi_end_io function,
7636 * so we use the write_seqlock_irq variant.
7638 * When looking for a bad block we specify a range and want to
7639 * know if any block in the range is bad. So we binary-search
7640 * to the last range that starts at-or-before the given endpoint,
7641 * (or "before the sector after the target range")
7642 * then see if it ends after the given start.
7644 * 0 if there are no known bad blocks in the range
7645 * 1 if there are known bad block which are all acknowledged
7646 * -1 if there are bad blocks which have not yet been acknowledged in metadata.
7647 * plus the start/length of the first bad section we overlap.
7649 int md_is_badblock(struct badblocks *bb, sector_t s, int sectors,
7650 sector_t *first_bad, int *bad_sectors)
7656 sector_t target = s + sectors;
7659 if (bb->shift > 0) {
7660 /* round the start down, and the end up */
7662 target += (1<<bb->shift) - 1;
7663 target >>= bb->shift;
7664 sectors = target - s;
7666 /* 'target' is now the first block after the bad range */
7669 seq = read_seqbegin(&bb->lock);
7673 /* Binary search between lo and hi for 'target'
7674 * i.e. for the last range that starts before 'target'
7676 /* INVARIANT: ranges before 'lo' and at-or-after 'hi'
7677 * are known not to be the last range before target.
7678 * VARIANT: hi-lo is the number of possible
7679 * ranges, and decreases until it reaches 1
7681 while (hi - lo > 1) {
7682 int mid = (lo + hi) / 2;
7683 sector_t a = BB_OFFSET(p[mid]);
7685 /* This could still be the one, earlier ranges
7689 /* This and later ranges are definitely out. */
7692 /* 'lo' might be the last that started before target, but 'hi' isn't */
7694 /* need to check all range that end after 's' to see if
7695 * any are unacknowledged.
7698 BB_OFFSET(p[lo]) + BB_LEN(p[lo]) > s) {
7699 if (BB_OFFSET(p[lo]) < target) {
7700 /* starts before the end, and finishes after
7701 * the start, so they must overlap
7703 if (rv != -1 && BB_ACK(p[lo]))
7707 *first_bad = BB_OFFSET(p[lo]);
7708 *bad_sectors = BB_LEN(p[lo]);
7714 if (read_seqretry(&bb->lock, seq))
7719 EXPORT_SYMBOL_GPL(md_is_badblock);
7722 * Add a range of bad blocks to the table.
7723 * This might extend the table, or might contract it
7724 * if two adjacent ranges can be merged.
7725 * We binary-search to find the 'insertion' point, then
7726 * decide how best to handle it.
7728 static int md_set_badblocks(struct badblocks *bb, sector_t s, int sectors,
7736 /* badblocks are disabled */
7740 /* round the start down, and the end up */
7741 sector_t next = s + sectors;
7743 next += (1<<bb->shift) - 1;
7748 write_seqlock_irq(&bb->lock);
7753 /* Find the last range that starts at-or-before 's' */
7754 while (hi - lo > 1) {
7755 int mid = (lo + hi) / 2;
7756 sector_t a = BB_OFFSET(p[mid]);
7762 if (hi > lo && BB_OFFSET(p[lo]) > s)
7766 /* we found a range that might merge with the start
7769 sector_t a = BB_OFFSET(p[lo]);
7770 sector_t e = a + BB_LEN(p[lo]);
7771 int ack = BB_ACK(p[lo]);
7773 /* Yes, we can merge with a previous range */
7774 if (s == a && s + sectors >= e)
7775 /* new range covers old */
7778 ack = ack && acknowledged;
7780 if (e < s + sectors)
7782 if (e - a <= BB_MAX_LEN) {
7783 p[lo] = BB_MAKE(a, e-a, ack);
7786 /* does not all fit in one range,
7787 * make p[lo] maximal
7789 if (BB_LEN(p[lo]) != BB_MAX_LEN)
7790 p[lo] = BB_MAKE(a, BB_MAX_LEN, ack);
7796 if (sectors && hi < bb->count) {
7797 /* 'hi' points to the first range that starts after 's'.
7798 * Maybe we can merge with the start of that range */
7799 sector_t a = BB_OFFSET(p[hi]);
7800 sector_t e = a + BB_LEN(p[hi]);
7801 int ack = BB_ACK(p[hi]);
7802 if (a <= s + sectors) {
7803 /* merging is possible */
7804 if (e <= s + sectors) {
7809 ack = ack && acknowledged;
7812 if (e - a <= BB_MAX_LEN) {
7813 p[hi] = BB_MAKE(a, e-a, ack);
7816 p[hi] = BB_MAKE(a, BB_MAX_LEN, ack);
7824 if (sectors == 0 && hi < bb->count) {
7825 /* we might be able to combine lo and hi */
7826 /* Note: 's' is at the end of 'lo' */
7827 sector_t a = BB_OFFSET(p[hi]);
7828 int lolen = BB_LEN(p[lo]);
7829 int hilen = BB_LEN(p[hi]);
7830 int newlen = lolen + hilen - (s - a);
7831 if (s >= a && newlen < BB_MAX_LEN) {
7832 /* yes, we can combine them */
7833 int ack = BB_ACK(p[lo]) && BB_ACK(p[hi]);
7834 p[lo] = BB_MAKE(BB_OFFSET(p[lo]), newlen, ack);
7835 memmove(p + hi, p + hi + 1,
7836 (bb->count - hi - 1) * 8);
7841 /* didn't merge (it all).
7842 * Need to add a range just before 'hi' */
7843 if (bb->count >= MD_MAX_BADBLOCKS) {
7844 /* No room for more */
7848 int this_sectors = sectors;
7849 memmove(p + hi + 1, p + hi,
7850 (bb->count - hi) * 8);
7853 if (this_sectors > BB_MAX_LEN)
7854 this_sectors = BB_MAX_LEN;
7855 p[hi] = BB_MAKE(s, this_sectors, acknowledged);
7856 sectors -= this_sectors;
7863 bb->unacked_exist = 1;
7864 write_sequnlock_irq(&bb->lock);
7869 int rdev_set_badblocks(struct md_rdev *rdev, sector_t s, int sectors,
7872 int rv = md_set_badblocks(&rdev->badblocks,
7873 s + rdev->data_offset, sectors, acknowledged);
7875 /* Make sure they get written out promptly */
7876 sysfs_notify_dirent_safe(rdev->sysfs_state);
7877 set_bit(MD_CHANGE_CLEAN, &rdev->mddev->flags);
7878 md_wakeup_thread(rdev->mddev->thread);
7882 EXPORT_SYMBOL_GPL(rdev_set_badblocks);
7885 * Remove a range of bad blocks from the table.
7886 * This may involve extending the table if we spilt a region,
7887 * but it must not fail. So if the table becomes full, we just
7888 * drop the remove request.
7890 static int md_clear_badblocks(struct badblocks *bb, sector_t s, int sectors)
7894 sector_t target = s + sectors;
7897 if (bb->shift > 0) {
7898 /* When clearing we round the start up and the end down.
7899 * This should not matter as the shift should align with
7900 * the block size and no rounding should ever be needed.
7901 * However it is better the think a block is bad when it
7902 * isn't than to think a block is not bad when it is.
7904 s += (1<<bb->shift) - 1;
7906 target >>= bb->shift;
7907 sectors = target - s;
7910 write_seqlock_irq(&bb->lock);
7915 /* Find the last range that starts before 'target' */
7916 while (hi - lo > 1) {
7917 int mid = (lo + hi) / 2;
7918 sector_t a = BB_OFFSET(p[mid]);
7925 /* p[lo] is the last range that could overlap the
7926 * current range. Earlier ranges could also overlap,
7927 * but only this one can overlap the end of the range.
7929 if (BB_OFFSET(p[lo]) + BB_LEN(p[lo]) > target) {
7930 /* Partial overlap, leave the tail of this range */
7931 int ack = BB_ACK(p[lo]);
7932 sector_t a = BB_OFFSET(p[lo]);
7933 sector_t end = a + BB_LEN(p[lo]);
7936 /* we need to split this range */
7937 if (bb->count >= MD_MAX_BADBLOCKS) {
7941 memmove(p+lo+1, p+lo, (bb->count - lo) * 8);
7943 p[lo] = BB_MAKE(a, s-a, ack);
7946 p[lo] = BB_MAKE(target, end - target, ack);
7947 /* there is no longer an overlap */
7952 BB_OFFSET(p[lo]) + BB_LEN(p[lo]) > s) {
7953 /* This range does overlap */
7954 if (BB_OFFSET(p[lo]) < s) {
7955 /* Keep the early parts of this range. */
7956 int ack = BB_ACK(p[lo]);
7957 sector_t start = BB_OFFSET(p[lo]);
7958 p[lo] = BB_MAKE(start, s - start, ack);
7959 /* now low doesn't overlap, so.. */
7964 /* 'lo' is strictly before, 'hi' is strictly after,
7965 * anything between needs to be discarded
7968 memmove(p+lo+1, p+hi, (bb->count - hi) * 8);
7969 bb->count -= (hi - lo - 1);
7975 write_sequnlock_irq(&bb->lock);
7979 int rdev_clear_badblocks(struct md_rdev *rdev, sector_t s, int sectors)
7981 return md_clear_badblocks(&rdev->badblocks,
7982 s + rdev->data_offset,
7985 EXPORT_SYMBOL_GPL(rdev_clear_badblocks);
7988 * Acknowledge all bad blocks in a list.
7989 * This only succeeds if ->changed is clear. It is used by
7990 * in-kernel metadata updates
7992 void md_ack_all_badblocks(struct badblocks *bb)
7994 if (bb->page == NULL || bb->changed)
7995 /* no point even trying */
7997 write_seqlock_irq(&bb->lock);
7999 if (bb->changed == 0) {
8002 for (i = 0; i < bb->count ; i++) {
8003 if (!BB_ACK(p[i])) {
8004 sector_t start = BB_OFFSET(p[i]);
8005 int len = BB_LEN(p[i]);
8006 p[i] = BB_MAKE(start, len, 1);
8009 bb->unacked_exist = 0;
8011 write_sequnlock_irq(&bb->lock);
8013 EXPORT_SYMBOL_GPL(md_ack_all_badblocks);
8015 /* sysfs access to bad-blocks list.
8016 * We present two files.
8017 * 'bad-blocks' lists sector numbers and lengths of ranges that
8018 * are recorded as bad. The list is truncated to fit within
8019 * the one-page limit of sysfs.
8020 * Writing "sector length" to this file adds an acknowledged
8022 * 'unacknowledged-bad-blocks' lists bad blocks that have not yet
8023 * been acknowledged. Writing to this file adds bad blocks
8024 * without acknowledging them. This is largely for testing.
8028 badblocks_show(struct badblocks *bb, char *page, int unack)
8039 seq = read_seqbegin(&bb->lock);
8044 while (len < PAGE_SIZE && i < bb->count) {
8045 sector_t s = BB_OFFSET(p[i]);
8046 unsigned int length = BB_LEN(p[i]);
8047 int ack = BB_ACK(p[i]);
8053 len += snprintf(page+len, PAGE_SIZE-len, "%llu %u\n",
8054 (unsigned long long)s << bb->shift,
8055 length << bb->shift);
8057 if (unack && len == 0)
8058 bb->unacked_exist = 0;
8060 if (read_seqretry(&bb->lock, seq))
8069 badblocks_store(struct badblocks *bb, const char *page, size_t len, int unack)
8071 unsigned long long sector;
8075 /* Allow clearing via sysfs *only* for testing/debugging.
8076 * Normally only a successful write may clear a badblock
8079 if (page[0] == '-') {
8083 #endif /* DO_DEBUG */
8085 switch (sscanf(page, "%llu %d%c", §or, &length, &newline)) {
8087 if (newline != '\n')
8099 md_clear_badblocks(bb, sector, length);
8102 #endif /* DO_DEBUG */
8103 if (md_set_badblocks(bb, sector, length, !unack))
8109 static int md_notify_reboot(struct notifier_block *this,
8110 unsigned long code, void *x)
8112 struct list_head *tmp;
8113 struct mddev *mddev;
8116 for_each_mddev(mddev, tmp) {
8117 if (mddev_trylock(mddev)) {
8119 __md_stop_writes(mddev);
8120 mddev->safemode = 2;
8121 mddev_unlock(mddev);
8126 * certain more exotic SCSI devices are known to be
8127 * volatile wrt too early system reboots. While the
8128 * right place to handle this issue is the given
8129 * driver, we do want to have a safe RAID driver ...
8137 static struct notifier_block md_notifier = {
8138 .notifier_call = md_notify_reboot,
8140 .priority = INT_MAX, /* before any real devices */
8143 static void md_geninit(void)
8145 pr_debug("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
8147 proc_create("mdstat", S_IRUGO, NULL, &md_seq_fops);
8150 static int __init md_init(void)
8154 md_wq = alloc_workqueue("md", WQ_MEM_RECLAIM, 0);
8158 md_misc_wq = alloc_workqueue("md_misc", 0, 0);
8162 if ((ret = register_blkdev(MD_MAJOR, "md")) < 0)
8165 if ((ret = register_blkdev(0, "mdp")) < 0)
8169 blk_register_region(MKDEV(MD_MAJOR, 0), 1UL<<MINORBITS, THIS_MODULE,
8170 md_probe, NULL, NULL);
8171 blk_register_region(MKDEV(mdp_major, 0), 1UL<<MINORBITS, THIS_MODULE,
8172 md_probe, NULL, NULL);
8174 register_reboot_notifier(&md_notifier);
8175 raid_table_header = register_sysctl_table(raid_root_table);
8181 unregister_blkdev(MD_MAJOR, "md");
8183 destroy_workqueue(md_misc_wq);
8185 destroy_workqueue(md_wq);
8193 * Searches all registered partitions for autorun RAID arrays
8197 static LIST_HEAD(all_detected_devices);
8198 struct detected_devices_node {
8199 struct list_head list;
8203 void md_autodetect_dev(dev_t dev)
8205 struct detected_devices_node *node_detected_dev;
8207 node_detected_dev = kzalloc(sizeof(*node_detected_dev), GFP_KERNEL);
8208 if (node_detected_dev) {
8209 node_detected_dev->dev = dev;
8210 list_add_tail(&node_detected_dev->list, &all_detected_devices);
8212 printk(KERN_CRIT "md: md_autodetect_dev: kzalloc failed"
8213 ", skipping dev(%d,%d)\n", MAJOR(dev), MINOR(dev));
8218 static void autostart_arrays(int part)
8220 struct md_rdev *rdev;
8221 struct detected_devices_node *node_detected_dev;
8223 int i_scanned, i_passed;
8228 printk(KERN_INFO "md: Autodetecting RAID arrays.\n");
8230 while (!list_empty(&all_detected_devices) && i_scanned < INT_MAX) {
8232 node_detected_dev = list_entry(all_detected_devices.next,
8233 struct detected_devices_node, list);
8234 list_del(&node_detected_dev->list);
8235 dev = node_detected_dev->dev;
8236 kfree(node_detected_dev);
8237 rdev = md_import_device(dev,0, 90);
8241 if (test_bit(Faulty, &rdev->flags)) {
8245 set_bit(AutoDetected, &rdev->flags);
8246 list_add(&rdev->same_set, &pending_raid_disks);
8250 printk(KERN_INFO "md: Scanned %d and added %d devices.\n",
8251 i_scanned, i_passed);
8253 autorun_devices(part);
8256 #endif /* !MODULE */
8258 static __exit void md_exit(void)
8260 struct mddev *mddev;
8261 struct list_head *tmp;
8263 blk_unregister_region(MKDEV(MD_MAJOR,0), 1U << MINORBITS);
8264 blk_unregister_region(MKDEV(mdp_major,0), 1U << MINORBITS);
8266 unregister_blkdev(MD_MAJOR,"md");
8267 unregister_blkdev(mdp_major, "mdp");
8268 unregister_reboot_notifier(&md_notifier);
8269 unregister_sysctl_table(raid_table_header);
8270 remove_proc_entry("mdstat", NULL);
8271 for_each_mddev(mddev, tmp) {
8272 export_array(mddev);
8273 mddev->hold_active = 0;
8275 destroy_workqueue(md_misc_wq);
8276 destroy_workqueue(md_wq);
8279 subsys_initcall(md_init);
8280 module_exit(md_exit)
8282 static int get_ro(char *buffer, struct kernel_param *kp)
8284 return sprintf(buffer, "%d", start_readonly);
8286 static int set_ro(const char *val, struct kernel_param *kp)
8289 int num = simple_strtoul(val, &e, 10);
8290 if (*val && (*e == '\0' || *e == '\n')) {
8291 start_readonly = num;
8297 module_param_call(start_ro, set_ro, get_ro, NULL, S_IRUSR|S_IWUSR);
8298 module_param(start_dirty_degraded, int, S_IRUGO|S_IWUSR);
8300 module_param_call(new_array, add_named_array, NULL, NULL, S_IWUSR);
8302 EXPORT_SYMBOL(register_md_personality);
8303 EXPORT_SYMBOL(unregister_md_personality);
8304 EXPORT_SYMBOL(md_error);
8305 EXPORT_SYMBOL(md_done_sync);
8306 EXPORT_SYMBOL(md_write_start);
8307 EXPORT_SYMBOL(md_write_end);
8308 EXPORT_SYMBOL(md_register_thread);
8309 EXPORT_SYMBOL(md_unregister_thread);
8310 EXPORT_SYMBOL(md_wakeup_thread);
8311 EXPORT_SYMBOL(md_check_recovery);
8312 MODULE_LICENSE("GPL");
8313 MODULE_DESCRIPTION("MD RAID framework");
8315 MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR);