2 md.c : Multiple Devices driver for Linux
3 Copyright (C) 1998, 1999, 2000 Ingo Molnar
5 completely rewritten, based on the MD driver code from Marc Zyngier
9 - RAID-1/RAID-5 extensions by Miguel de Icaza, Gadi Oxman, Ingo Molnar
10 - RAID-6 extensions by H. Peter Anvin <hpa@zytor.com>
11 - boot support for linear and striped mode by Harald Hoyer <HarryH@Royal.Net>
12 - kerneld support by Boris Tobotras <boris@xtalk.msk.su>
13 - kmod support by: Cyrus Durgin
14 - RAID0 bugfixes: Mark Anthony Lisher <markal@iname.com>
15 - Devfs support by Richard Gooch <rgooch@atnf.csiro.au>
17 - lots of fixes and improvements to the RAID1/RAID5 and generic
18 RAID code (such as request based resynchronization):
20 Neil Brown <neilb@cse.unsw.edu.au>.
22 - persistent bitmap code
23 Copyright (C) 2003-2004, Paul Clements, SteelEye Technology, Inc.
25 This program is free software; you can redistribute it and/or modify
26 it under the terms of the GNU General Public License as published by
27 the Free Software Foundation; either version 2, or (at your option)
30 You should have received a copy of the GNU General Public License
31 (for example /usr/src/linux/COPYING); if not, write to the Free
32 Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
35 #include <linux/kthread.h>
36 #include <linux/blkdev.h>
37 #include <linux/sysctl.h>
38 #include <linux/seq_file.h>
40 #include <linux/poll.h>
41 #include <linux/ctype.h>
42 #include <linux/string.h>
43 #include <linux/hdreg.h>
44 #include <linux/proc_fs.h>
45 #include <linux/random.h>
46 #include <linux/module.h>
47 #include <linux/reboot.h>
48 #include <linux/file.h>
49 #include <linux/compat.h>
50 #include <linux/delay.h>
51 #include <linux/raid/md_p.h>
52 #include <linux/raid/md_u.h>
53 #include <linux/slab.h>
58 static void autostart_arrays(int part);
61 /* pers_list is a list of registered personalities protected
63 * pers_lock does extra service to protect accesses to
64 * mddev->thread when the mutex cannot be held.
66 static LIST_HEAD(pers_list);
67 static DEFINE_SPINLOCK(pers_lock);
69 static void md_print_devices(void);
71 static DECLARE_WAIT_QUEUE_HEAD(resync_wait);
72 static struct workqueue_struct *md_wq;
73 static struct workqueue_struct *md_misc_wq;
75 #define MD_BUG(x...) { printk("md: bug in file %s, line %d\n", __FILE__, __LINE__); md_print_devices(); }
78 * Default number of read corrections we'll attempt on an rdev
79 * before ejecting it from the array. We divide the read error
80 * count by 2 for every hour elapsed between read errors.
82 #define MD_DEFAULT_MAX_CORRECTED_READ_ERRORS 20
84 * Current RAID-1,4,5 parallel reconstruction 'guaranteed speed limit'
85 * is 1000 KB/sec, so the extra system load does not show up that much.
86 * Increase it if you want to have more _guaranteed_ speed. Note that
87 * the RAID driver will use the maximum available bandwidth if the IO
88 * subsystem is idle. There is also an 'absolute maximum' reconstruction
89 * speed limit - in case reconstruction slows down your system despite
92 * you can change it via /proc/sys/dev/raid/speed_limit_min and _max.
93 * or /sys/block/mdX/md/sync_speed_{min,max}
96 static int sysctl_speed_limit_min = 1000;
97 static int sysctl_speed_limit_max = 200000;
98 static inline int speed_min(struct mddev *mddev)
100 return mddev->sync_speed_min ?
101 mddev->sync_speed_min : sysctl_speed_limit_min;
104 static inline int speed_max(struct mddev *mddev)
106 return mddev->sync_speed_max ?
107 mddev->sync_speed_max : sysctl_speed_limit_max;
110 static struct ctl_table_header *raid_table_header;
112 static ctl_table raid_table[] = {
114 .procname = "speed_limit_min",
115 .data = &sysctl_speed_limit_min,
116 .maxlen = sizeof(int),
117 .mode = S_IRUGO|S_IWUSR,
118 .proc_handler = proc_dointvec,
121 .procname = "speed_limit_max",
122 .data = &sysctl_speed_limit_max,
123 .maxlen = sizeof(int),
124 .mode = S_IRUGO|S_IWUSR,
125 .proc_handler = proc_dointvec,
130 static ctl_table raid_dir_table[] = {
134 .mode = S_IRUGO|S_IXUGO,
140 static ctl_table raid_root_table[] = {
145 .child = raid_dir_table,
150 static const struct block_device_operations md_fops;
152 static int start_readonly;
155 * like bio_clone, but with a local bio set
158 static void mddev_bio_destructor(struct bio *bio)
160 struct mddev *mddev, **mddevp;
165 bio_free(bio, mddev->bio_set);
168 struct bio *bio_alloc_mddev(gfp_t gfp_mask, int nr_iovecs,
172 struct mddev **mddevp;
174 if (!mddev || !mddev->bio_set)
175 return bio_alloc(gfp_mask, nr_iovecs);
177 b = bio_alloc_bioset(gfp_mask, nr_iovecs,
183 b->bi_destructor = mddev_bio_destructor;
186 EXPORT_SYMBOL_GPL(bio_alloc_mddev);
188 struct bio *bio_clone_mddev(struct bio *bio, gfp_t gfp_mask,
192 struct mddev **mddevp;
194 if (!mddev || !mddev->bio_set)
195 return bio_clone(bio, gfp_mask);
197 b = bio_alloc_bioset(gfp_mask, bio->bi_max_vecs,
203 b->bi_destructor = mddev_bio_destructor;
205 if (bio_integrity(bio)) {
208 ret = bio_integrity_clone(b, bio, gfp_mask, mddev->bio_set);
218 EXPORT_SYMBOL_GPL(bio_clone_mddev);
220 void md_trim_bio(struct bio *bio, int offset, int size)
222 /* 'bio' is a cloned bio which we need to trim to match
223 * the given offset and size.
224 * This requires adjusting bi_sector, bi_size, and bi_io_vec
227 struct bio_vec *bvec;
231 if (offset == 0 && size == bio->bi_size)
234 bio->bi_sector += offset;
237 clear_bit(BIO_SEG_VALID, &bio->bi_flags);
239 while (bio->bi_idx < bio->bi_vcnt &&
240 bio->bi_io_vec[bio->bi_idx].bv_len <= offset) {
241 /* remove this whole bio_vec */
242 offset -= bio->bi_io_vec[bio->bi_idx].bv_len;
245 if (bio->bi_idx < bio->bi_vcnt) {
246 bio->bi_io_vec[bio->bi_idx].bv_offset += offset;
247 bio->bi_io_vec[bio->bi_idx].bv_len -= offset;
249 /* avoid any complications with bi_idx being non-zero*/
251 memmove(bio->bi_io_vec, bio->bi_io_vec+bio->bi_idx,
252 (bio->bi_vcnt - bio->bi_idx) * sizeof(struct bio_vec));
253 bio->bi_vcnt -= bio->bi_idx;
256 /* Make sure vcnt and last bv are not too big */
257 bio_for_each_segment(bvec, bio, i) {
258 if (sofar + bvec->bv_len > size)
259 bvec->bv_len = size - sofar;
260 if (bvec->bv_len == 0) {
264 sofar += bvec->bv_len;
267 EXPORT_SYMBOL_GPL(md_trim_bio);
270 * We have a system wide 'event count' that is incremented
271 * on any 'interesting' event, and readers of /proc/mdstat
272 * can use 'poll' or 'select' to find out when the event
276 * start array, stop array, error, add device, remove device,
277 * start build, activate spare
279 static DECLARE_WAIT_QUEUE_HEAD(md_event_waiters);
280 static atomic_t md_event_count;
281 void md_new_event(struct mddev *mddev)
283 atomic_inc(&md_event_count);
284 wake_up(&md_event_waiters);
286 EXPORT_SYMBOL_GPL(md_new_event);
288 /* Alternate version that can be called from interrupts
289 * when calling sysfs_notify isn't needed.
291 static void md_new_event_inintr(struct mddev *mddev)
293 atomic_inc(&md_event_count);
294 wake_up(&md_event_waiters);
298 * Enables to iterate over all existing md arrays
299 * all_mddevs_lock protects this list.
301 static LIST_HEAD(all_mddevs);
302 static DEFINE_SPINLOCK(all_mddevs_lock);
306 * iterates through all used mddevs in the system.
307 * We take care to grab the all_mddevs_lock whenever navigating
308 * the list, and to always hold a refcount when unlocked.
309 * Any code which breaks out of this loop while own
310 * a reference to the current mddev and must mddev_put it.
312 #define for_each_mddev(_mddev,_tmp) \
314 for (({ spin_lock(&all_mddevs_lock); \
315 _tmp = all_mddevs.next; \
317 ({ if (_tmp != &all_mddevs) \
318 mddev_get(list_entry(_tmp, struct mddev, all_mddevs));\
319 spin_unlock(&all_mddevs_lock); \
320 if (_mddev) mddev_put(_mddev); \
321 _mddev = list_entry(_tmp, struct mddev, all_mddevs); \
322 _tmp != &all_mddevs;}); \
323 ({ spin_lock(&all_mddevs_lock); \
324 _tmp = _tmp->next;}) \
328 /* Rather than calling directly into the personality make_request function,
329 * IO requests come here first so that we can check if the device is
330 * being suspended pending a reconfiguration.
331 * We hold a refcount over the call to ->make_request. By the time that
332 * call has finished, the bio has been linked into some internal structure
333 * and so is visible to ->quiesce(), so we don't need the refcount any more.
335 static void md_make_request(struct request_queue *q, struct bio *bio)
337 const int rw = bio_data_dir(bio);
338 struct mddev *mddev = q->queuedata;
340 unsigned int sectors;
342 if (mddev == NULL || mddev->pers == NULL
347 smp_rmb(); /* Ensure implications of 'active' are visible */
349 if (mddev->suspended) {
352 prepare_to_wait(&mddev->sb_wait, &__wait,
353 TASK_UNINTERRUPTIBLE);
354 if (!mddev->suspended)
360 finish_wait(&mddev->sb_wait, &__wait);
362 atomic_inc(&mddev->active_io);
366 * save the sectors now since our bio can
367 * go away inside make_request
369 sectors = bio_sectors(bio);
370 mddev->pers->make_request(mddev, bio);
372 cpu = part_stat_lock();
373 part_stat_inc(cpu, &mddev->gendisk->part0, ios[rw]);
374 part_stat_add(cpu, &mddev->gendisk->part0, sectors[rw], sectors);
377 if (atomic_dec_and_test(&mddev->active_io) && mddev->suspended)
378 wake_up(&mddev->sb_wait);
381 /* mddev_suspend makes sure no new requests are submitted
382 * to the device, and that any requests that have been submitted
383 * are completely handled.
384 * Once ->stop is called and completes, the module will be completely
387 void mddev_suspend(struct mddev *mddev)
389 BUG_ON(mddev->suspended);
390 mddev->suspended = 1;
392 wait_event(mddev->sb_wait, atomic_read(&mddev->active_io) == 0);
393 mddev->pers->quiesce(mddev, 1);
395 del_timer_sync(&mddev->safemode_timer);
397 EXPORT_SYMBOL_GPL(mddev_suspend);
399 void mddev_resume(struct mddev *mddev)
401 mddev->suspended = 0;
402 wake_up(&mddev->sb_wait);
403 mddev->pers->quiesce(mddev, 0);
405 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
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 rdev_for_each_rcu(rdev, mddev)
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->reshape_backwards = 0;
612 mddev->resync_min = 0;
613 mddev->resync_max = MaxSector;
614 mddev->level = LEVEL_NONE;
616 EXPORT_SYMBOL_GPL(mddev_init);
618 static struct mddev * mddev_find(dev_t unit)
620 struct mddev *mddev, *new = NULL;
622 if (unit && MAJOR(unit) != MD_MAJOR)
623 unit &= ~((1<<MdpMinorShift)-1);
626 spin_lock(&all_mddevs_lock);
629 list_for_each_entry(mddev, &all_mddevs, all_mddevs)
630 if (mddev->unit == unit) {
632 spin_unlock(&all_mddevs_lock);
638 list_add(&new->all_mddevs, &all_mddevs);
639 spin_unlock(&all_mddevs_lock);
640 new->hold_active = UNTIL_IOCTL;
644 /* find an unused unit number */
645 static int next_minor = 512;
646 int start = next_minor;
650 dev = MKDEV(MD_MAJOR, next_minor);
652 if (next_minor > MINORMASK)
654 if (next_minor == start) {
655 /* Oh dear, all in use. */
656 spin_unlock(&all_mddevs_lock);
662 list_for_each_entry(mddev, &all_mddevs, all_mddevs)
663 if (mddev->unit == dev) {
669 new->md_minor = MINOR(dev);
670 new->hold_active = UNTIL_STOP;
671 list_add(&new->all_mddevs, &all_mddevs);
672 spin_unlock(&all_mddevs_lock);
675 spin_unlock(&all_mddevs_lock);
677 new = kzalloc(sizeof(*new), GFP_KERNEL);
682 if (MAJOR(unit) == MD_MAJOR)
683 new->md_minor = MINOR(unit);
685 new->md_minor = MINOR(unit) >> MdpMinorShift;
692 static inline int mddev_lock(struct mddev * mddev)
694 return mutex_lock_interruptible(&mddev->reconfig_mutex);
697 static inline int mddev_is_locked(struct mddev *mddev)
699 return mutex_is_locked(&mddev->reconfig_mutex);
702 static inline int mddev_trylock(struct mddev * mddev)
704 return mutex_trylock(&mddev->reconfig_mutex);
707 static struct attribute_group md_redundancy_group;
709 static void mddev_unlock(struct mddev * mddev)
711 if (mddev->to_remove) {
712 /* These cannot be removed under reconfig_mutex as
713 * an access to the files will try to take reconfig_mutex
714 * while holding the file unremovable, which leads to
716 * So hold set sysfs_active while the remove in happeing,
717 * and anything else which might set ->to_remove or my
718 * otherwise change the sysfs namespace will fail with
719 * -EBUSY if sysfs_active is still set.
720 * We set sysfs_active under reconfig_mutex and elsewhere
721 * test it under the same mutex to ensure its correct value
724 struct attribute_group *to_remove = mddev->to_remove;
725 mddev->to_remove = NULL;
726 mddev->sysfs_active = 1;
727 mutex_unlock(&mddev->reconfig_mutex);
729 if (mddev->kobj.sd) {
730 if (to_remove != &md_redundancy_group)
731 sysfs_remove_group(&mddev->kobj, to_remove);
732 if (mddev->pers == NULL ||
733 mddev->pers->sync_request == NULL) {
734 sysfs_remove_group(&mddev->kobj, &md_redundancy_group);
735 if (mddev->sysfs_action)
736 sysfs_put(mddev->sysfs_action);
737 mddev->sysfs_action = NULL;
740 mddev->sysfs_active = 0;
742 mutex_unlock(&mddev->reconfig_mutex);
744 /* As we've dropped the mutex we need a spinlock to
745 * make sure the thread doesn't disappear
747 spin_lock(&pers_lock);
748 md_wakeup_thread(mddev->thread);
749 spin_unlock(&pers_lock);
752 static struct md_rdev * find_rdev_nr(struct mddev *mddev, int nr)
754 struct md_rdev *rdev;
756 rdev_for_each(rdev, mddev)
757 if (rdev->desc_nr == nr)
763 static struct md_rdev * find_rdev(struct mddev * mddev, dev_t dev)
765 struct md_rdev *rdev;
767 rdev_for_each(rdev, mddev)
768 if (rdev->bdev->bd_dev == dev)
774 static struct md_personality *find_pers(int level, char *clevel)
776 struct md_personality *pers;
777 list_for_each_entry(pers, &pers_list, list) {
778 if (level != LEVEL_NONE && pers->level == level)
780 if (strcmp(pers->name, clevel)==0)
786 /* return the offset of the super block in 512byte sectors */
787 static inline sector_t calc_dev_sboffset(struct md_rdev *rdev)
789 sector_t num_sectors = i_size_read(rdev->bdev->bd_inode) / 512;
790 return MD_NEW_SIZE_SECTORS(num_sectors);
793 static int alloc_disk_sb(struct md_rdev * rdev)
798 rdev->sb_page = alloc_page(GFP_KERNEL);
799 if (!rdev->sb_page) {
800 printk(KERN_ALERT "md: out of memory.\n");
807 void md_rdev_clear(struct md_rdev *rdev)
810 put_page(rdev->sb_page);
812 rdev->sb_page = NULL;
817 put_page(rdev->bb_page);
818 rdev->bb_page = NULL;
820 kfree(rdev->badblocks.page);
821 rdev->badblocks.page = NULL;
823 EXPORT_SYMBOL_GPL(md_rdev_clear);
825 static void super_written(struct bio *bio, int error)
827 struct md_rdev *rdev = bio->bi_private;
828 struct mddev *mddev = rdev->mddev;
830 if (error || !test_bit(BIO_UPTODATE, &bio->bi_flags)) {
831 printk("md: super_written gets error=%d, uptodate=%d\n",
832 error, test_bit(BIO_UPTODATE, &bio->bi_flags));
833 WARN_ON(test_bit(BIO_UPTODATE, &bio->bi_flags));
834 md_error(mddev, rdev);
837 if (atomic_dec_and_test(&mddev->pending_writes))
838 wake_up(&mddev->sb_wait);
842 void md_super_write(struct mddev *mddev, struct md_rdev *rdev,
843 sector_t sector, int size, struct page *page)
845 /* write first size bytes of page to sector of rdev
846 * Increment mddev->pending_writes before returning
847 * and decrement it on completion, waking up sb_wait
848 * if zero is reached.
849 * If an error occurred, call md_error
851 struct bio *bio = bio_alloc_mddev(GFP_NOIO, 1, mddev);
853 bio->bi_bdev = rdev->meta_bdev ? rdev->meta_bdev : rdev->bdev;
854 bio->bi_sector = sector;
855 bio_add_page(bio, page, size, 0);
856 bio->bi_private = rdev;
857 bio->bi_end_io = super_written;
859 atomic_inc(&mddev->pending_writes);
860 submit_bio(WRITE_FLUSH_FUA, bio);
863 void md_super_wait(struct mddev *mddev)
865 /* wait for all superblock writes that were scheduled to complete */
868 prepare_to_wait(&mddev->sb_wait, &wq, TASK_UNINTERRUPTIBLE);
869 if (atomic_read(&mddev->pending_writes)==0)
873 finish_wait(&mddev->sb_wait, &wq);
876 static void bi_complete(struct bio *bio, int error)
878 complete((struct completion*)bio->bi_private);
881 int sync_page_io(struct md_rdev *rdev, sector_t sector, int size,
882 struct page *page, int rw, bool metadata_op)
884 struct bio *bio = bio_alloc_mddev(GFP_NOIO, 1, rdev->mddev);
885 struct completion event;
890 bio->bi_bdev = (metadata_op && rdev->meta_bdev) ?
891 rdev->meta_bdev : rdev->bdev;
893 bio->bi_sector = sector + rdev->sb_start;
894 else if (rdev->mddev->reshape_position != MaxSector &&
895 (rdev->mddev->reshape_backwards ==
896 (sector >= rdev->mddev->reshape_position)))
897 bio->bi_sector = sector + rdev->new_data_offset;
899 bio->bi_sector = sector + rdev->data_offset;
900 bio_add_page(bio, page, size, 0);
901 init_completion(&event);
902 bio->bi_private = &event;
903 bio->bi_end_io = bi_complete;
905 wait_for_completion(&event);
907 ret = test_bit(BIO_UPTODATE, &bio->bi_flags);
911 EXPORT_SYMBOL_GPL(sync_page_io);
913 static int read_disk_sb(struct md_rdev * rdev, int size)
915 char b[BDEVNAME_SIZE];
916 if (!rdev->sb_page) {
924 if (!sync_page_io(rdev, 0, size, rdev->sb_page, READ, true))
930 printk(KERN_WARNING "md: disabled device %s, could not read superblock.\n",
931 bdevname(rdev->bdev,b));
935 static int uuid_equal(mdp_super_t *sb1, mdp_super_t *sb2)
937 return sb1->set_uuid0 == sb2->set_uuid0 &&
938 sb1->set_uuid1 == sb2->set_uuid1 &&
939 sb1->set_uuid2 == sb2->set_uuid2 &&
940 sb1->set_uuid3 == sb2->set_uuid3;
943 static int sb_equal(mdp_super_t *sb1, mdp_super_t *sb2)
946 mdp_super_t *tmp1, *tmp2;
948 tmp1 = kmalloc(sizeof(*tmp1),GFP_KERNEL);
949 tmp2 = kmalloc(sizeof(*tmp2),GFP_KERNEL);
951 if (!tmp1 || !tmp2) {
953 printk(KERN_INFO "md.c sb_equal(): failed to allocate memory!\n");
961 * nr_disks is not constant
966 ret = (memcmp(tmp1, tmp2, MD_SB_GENERIC_CONSTANT_WORDS * 4) == 0);
974 static u32 md_csum_fold(u32 csum)
976 csum = (csum & 0xffff) + (csum >> 16);
977 return (csum & 0xffff) + (csum >> 16);
980 static unsigned int calc_sb_csum(mdp_super_t * sb)
983 u32 *sb32 = (u32*)sb;
985 unsigned int disk_csum, csum;
987 disk_csum = sb->sb_csum;
990 for (i = 0; i < MD_SB_BYTES/4 ; i++)
992 csum = (newcsum & 0xffffffff) + (newcsum>>32);
996 /* This used to use csum_partial, which was wrong for several
997 * reasons including that different results are returned on
998 * different architectures. It isn't critical that we get exactly
999 * the same return value as before (we always csum_fold before
1000 * testing, and that removes any differences). However as we
1001 * know that csum_partial always returned a 16bit value on
1002 * alphas, do a fold to maximise conformity to previous behaviour.
1004 sb->sb_csum = md_csum_fold(disk_csum);
1006 sb->sb_csum = disk_csum;
1013 * Handle superblock details.
1014 * We want to be able to handle multiple superblock formats
1015 * so we have a common interface to them all, and an array of
1016 * different handlers.
1017 * We rely on user-space to write the initial superblock, and support
1018 * reading and updating of superblocks.
1019 * Interface methods are:
1020 * int load_super(struct md_rdev *dev, struct md_rdev *refdev, int minor_version)
1021 * loads and validates a superblock on dev.
1022 * if refdev != NULL, compare superblocks on both devices
1024 * 0 - dev has a superblock that is compatible with refdev
1025 * 1 - dev has a superblock that is compatible and newer than refdev
1026 * so dev should be used as the refdev in future
1027 * -EINVAL superblock incompatible or invalid
1028 * -othererror e.g. -EIO
1030 * int validate_super(struct mddev *mddev, struct md_rdev *dev)
1031 * Verify that dev is acceptable into mddev.
1032 * The first time, mddev->raid_disks will be 0, and data from
1033 * dev should be merged in. Subsequent calls check that dev
1034 * is new enough. Return 0 or -EINVAL
1036 * void sync_super(struct mddev *mddev, struct md_rdev *dev)
1037 * Update the superblock for rdev with data in mddev
1038 * This does not write to disc.
1044 struct module *owner;
1045 int (*load_super)(struct md_rdev *rdev,
1046 struct md_rdev *refdev,
1048 int (*validate_super)(struct mddev *mddev,
1049 struct md_rdev *rdev);
1050 void (*sync_super)(struct mddev *mddev,
1051 struct md_rdev *rdev);
1052 unsigned long long (*rdev_size_change)(struct md_rdev *rdev,
1053 sector_t num_sectors);
1054 int (*allow_new_offset)(struct md_rdev *rdev,
1055 unsigned long long new_offset);
1059 * Check that the given mddev has no bitmap.
1061 * This function is called from the run method of all personalities that do not
1062 * support bitmaps. It prints an error message and returns non-zero if mddev
1063 * has a bitmap. Otherwise, it returns 0.
1066 int md_check_no_bitmap(struct mddev *mddev)
1068 if (!mddev->bitmap_info.file && !mddev->bitmap_info.offset)
1070 printk(KERN_ERR "%s: bitmaps are not supported for %s\n",
1071 mdname(mddev), mddev->pers->name);
1074 EXPORT_SYMBOL(md_check_no_bitmap);
1077 * load_super for 0.90.0
1079 static int super_90_load(struct md_rdev *rdev, struct md_rdev *refdev, int minor_version)
1081 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
1086 * Calculate the position of the superblock (512byte sectors),
1087 * it's at the end of the disk.
1089 * It also happens to be a multiple of 4Kb.
1091 rdev->sb_start = calc_dev_sboffset(rdev);
1093 ret = read_disk_sb(rdev, MD_SB_BYTES);
1094 if (ret) return ret;
1098 bdevname(rdev->bdev, b);
1099 sb = page_address(rdev->sb_page);
1101 if (sb->md_magic != MD_SB_MAGIC) {
1102 printk(KERN_ERR "md: invalid raid superblock magic on %s\n",
1107 if (sb->major_version != 0 ||
1108 sb->minor_version < 90 ||
1109 sb->minor_version > 91) {
1110 printk(KERN_WARNING "Bad version number %d.%d on %s\n",
1111 sb->major_version, sb->minor_version,
1116 if (sb->raid_disks <= 0)
1119 if (md_csum_fold(calc_sb_csum(sb)) != md_csum_fold(sb->sb_csum)) {
1120 printk(KERN_WARNING "md: invalid superblock checksum on %s\n",
1125 rdev->preferred_minor = sb->md_minor;
1126 rdev->data_offset = 0;
1127 rdev->new_data_offset = 0;
1128 rdev->sb_size = MD_SB_BYTES;
1129 rdev->badblocks.shift = -1;
1131 if (sb->level == LEVEL_MULTIPATH)
1134 rdev->desc_nr = sb->this_disk.number;
1140 mdp_super_t *refsb = page_address(refdev->sb_page);
1141 if (!uuid_equal(refsb, sb)) {
1142 printk(KERN_WARNING "md: %s has different UUID to %s\n",
1143 b, bdevname(refdev->bdev,b2));
1146 if (!sb_equal(refsb, sb)) {
1147 printk(KERN_WARNING "md: %s has same UUID"
1148 " but different superblock to %s\n",
1149 b, bdevname(refdev->bdev, b2));
1153 ev2 = md_event(refsb);
1159 rdev->sectors = rdev->sb_start;
1160 /* Limit to 4TB as metadata cannot record more than that */
1161 if (rdev->sectors >= (2ULL << 32))
1162 rdev->sectors = (2ULL << 32) - 2;
1164 if (rdev->sectors < ((sector_t)sb->size) * 2 && sb->level >= 1)
1165 /* "this cannot possibly happen" ... */
1173 * validate_super for 0.90.0
1175 static int super_90_validate(struct mddev *mddev, struct md_rdev *rdev)
1178 mdp_super_t *sb = page_address(rdev->sb_page);
1179 __u64 ev1 = md_event(sb);
1181 rdev->raid_disk = -1;
1182 clear_bit(Faulty, &rdev->flags);
1183 clear_bit(In_sync, &rdev->flags);
1184 clear_bit(WriteMostly, &rdev->flags);
1186 if (mddev->raid_disks == 0) {
1187 mddev->major_version = 0;
1188 mddev->minor_version = sb->minor_version;
1189 mddev->patch_version = sb->patch_version;
1190 mddev->external = 0;
1191 mddev->chunk_sectors = sb->chunk_size >> 9;
1192 mddev->ctime = sb->ctime;
1193 mddev->utime = sb->utime;
1194 mddev->level = sb->level;
1195 mddev->clevel[0] = 0;
1196 mddev->layout = sb->layout;
1197 mddev->raid_disks = sb->raid_disks;
1198 mddev->dev_sectors = ((sector_t)sb->size) * 2;
1199 mddev->events = ev1;
1200 mddev->bitmap_info.offset = 0;
1201 mddev->bitmap_info.space = 0;
1202 /* bitmap can use 60 K after the 4K superblocks */
1203 mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9;
1204 mddev->bitmap_info.default_space = 64*2 - (MD_SB_BYTES >> 9);
1205 mddev->reshape_backwards = 0;
1207 if (mddev->minor_version >= 91) {
1208 mddev->reshape_position = sb->reshape_position;
1209 mddev->delta_disks = sb->delta_disks;
1210 mddev->new_level = sb->new_level;
1211 mddev->new_layout = sb->new_layout;
1212 mddev->new_chunk_sectors = sb->new_chunk >> 9;
1213 if (mddev->delta_disks < 0)
1214 mddev->reshape_backwards = 1;
1216 mddev->reshape_position = MaxSector;
1217 mddev->delta_disks = 0;
1218 mddev->new_level = mddev->level;
1219 mddev->new_layout = mddev->layout;
1220 mddev->new_chunk_sectors = mddev->chunk_sectors;
1223 if (sb->state & (1<<MD_SB_CLEAN))
1224 mddev->recovery_cp = MaxSector;
1226 if (sb->events_hi == sb->cp_events_hi &&
1227 sb->events_lo == sb->cp_events_lo) {
1228 mddev->recovery_cp = sb->recovery_cp;
1230 mddev->recovery_cp = 0;
1233 memcpy(mddev->uuid+0, &sb->set_uuid0, 4);
1234 memcpy(mddev->uuid+4, &sb->set_uuid1, 4);
1235 memcpy(mddev->uuid+8, &sb->set_uuid2, 4);
1236 memcpy(mddev->uuid+12,&sb->set_uuid3, 4);
1238 mddev->max_disks = MD_SB_DISKS;
1240 if (sb->state & (1<<MD_SB_BITMAP_PRESENT) &&
1241 mddev->bitmap_info.file == NULL) {
1242 mddev->bitmap_info.offset =
1243 mddev->bitmap_info.default_offset;
1244 mddev->bitmap_info.space =
1245 mddev->bitmap_info.space;
1248 } else if (mddev->pers == NULL) {
1249 /* Insist on good event counter while assembling, except
1250 * for spares (which don't need an event count) */
1252 if (sb->disks[rdev->desc_nr].state & (
1253 (1<<MD_DISK_SYNC) | (1 << MD_DISK_ACTIVE)))
1254 if (ev1 < mddev->events)
1256 } else if (mddev->bitmap) {
1257 /* if adding to array with a bitmap, then we can accept an
1258 * older device ... but not too old.
1260 if (ev1 < mddev->bitmap->events_cleared)
1263 if (ev1 < mddev->events)
1264 /* just a hot-add of a new device, leave raid_disk at -1 */
1268 if (mddev->level != LEVEL_MULTIPATH) {
1269 desc = sb->disks + rdev->desc_nr;
1271 if (desc->state & (1<<MD_DISK_FAULTY))
1272 set_bit(Faulty, &rdev->flags);
1273 else if (desc->state & (1<<MD_DISK_SYNC) /* &&
1274 desc->raid_disk < mddev->raid_disks */) {
1275 set_bit(In_sync, &rdev->flags);
1276 rdev->raid_disk = desc->raid_disk;
1277 } else if (desc->state & (1<<MD_DISK_ACTIVE)) {
1278 /* active but not in sync implies recovery up to
1279 * reshape position. We don't know exactly where
1280 * that is, so set to zero for now */
1281 if (mddev->minor_version >= 91) {
1282 rdev->recovery_offset = 0;
1283 rdev->raid_disk = desc->raid_disk;
1286 if (desc->state & (1<<MD_DISK_WRITEMOSTLY))
1287 set_bit(WriteMostly, &rdev->flags);
1288 } else /* MULTIPATH are always insync */
1289 set_bit(In_sync, &rdev->flags);
1294 * sync_super for 0.90.0
1296 static void super_90_sync(struct mddev *mddev, struct md_rdev *rdev)
1299 struct md_rdev *rdev2;
1300 int next_spare = mddev->raid_disks;
1303 /* make rdev->sb match mddev data..
1306 * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
1307 * 3/ any empty disks < next_spare become removed
1309 * disks[0] gets initialised to REMOVED because
1310 * we cannot be sure from other fields if it has
1311 * been initialised or not.
1314 int active=0, working=0,failed=0,spare=0,nr_disks=0;
1316 rdev->sb_size = MD_SB_BYTES;
1318 sb = page_address(rdev->sb_page);
1320 memset(sb, 0, sizeof(*sb));
1322 sb->md_magic = MD_SB_MAGIC;
1323 sb->major_version = mddev->major_version;
1324 sb->patch_version = mddev->patch_version;
1325 sb->gvalid_words = 0; /* ignored */
1326 memcpy(&sb->set_uuid0, mddev->uuid+0, 4);
1327 memcpy(&sb->set_uuid1, mddev->uuid+4, 4);
1328 memcpy(&sb->set_uuid2, mddev->uuid+8, 4);
1329 memcpy(&sb->set_uuid3, mddev->uuid+12,4);
1331 sb->ctime = mddev->ctime;
1332 sb->level = mddev->level;
1333 sb->size = mddev->dev_sectors / 2;
1334 sb->raid_disks = mddev->raid_disks;
1335 sb->md_minor = mddev->md_minor;
1336 sb->not_persistent = 0;
1337 sb->utime = mddev->utime;
1339 sb->events_hi = (mddev->events>>32);
1340 sb->events_lo = (u32)mddev->events;
1342 if (mddev->reshape_position == MaxSector)
1343 sb->minor_version = 90;
1345 sb->minor_version = 91;
1346 sb->reshape_position = mddev->reshape_position;
1347 sb->new_level = mddev->new_level;
1348 sb->delta_disks = mddev->delta_disks;
1349 sb->new_layout = mddev->new_layout;
1350 sb->new_chunk = mddev->new_chunk_sectors << 9;
1352 mddev->minor_version = sb->minor_version;
1355 sb->recovery_cp = mddev->recovery_cp;
1356 sb->cp_events_hi = (mddev->events>>32);
1357 sb->cp_events_lo = (u32)mddev->events;
1358 if (mddev->recovery_cp == MaxSector)
1359 sb->state = (1<< MD_SB_CLEAN);
1361 sb->recovery_cp = 0;
1363 sb->layout = mddev->layout;
1364 sb->chunk_size = mddev->chunk_sectors << 9;
1366 if (mddev->bitmap && mddev->bitmap_info.file == NULL)
1367 sb->state |= (1<<MD_SB_BITMAP_PRESENT);
1369 sb->disks[0].state = (1<<MD_DISK_REMOVED);
1370 rdev_for_each(rdev2, mddev) {
1373 int is_active = test_bit(In_sync, &rdev2->flags);
1375 if (rdev2->raid_disk >= 0 &&
1376 sb->minor_version >= 91)
1377 /* we have nowhere to store the recovery_offset,
1378 * but if it is not below the reshape_position,
1379 * we can piggy-back on that.
1382 if (rdev2->raid_disk < 0 ||
1383 test_bit(Faulty, &rdev2->flags))
1386 desc_nr = rdev2->raid_disk;
1388 desc_nr = next_spare++;
1389 rdev2->desc_nr = desc_nr;
1390 d = &sb->disks[rdev2->desc_nr];
1392 d->number = rdev2->desc_nr;
1393 d->major = MAJOR(rdev2->bdev->bd_dev);
1394 d->minor = MINOR(rdev2->bdev->bd_dev);
1396 d->raid_disk = rdev2->raid_disk;
1398 d->raid_disk = rdev2->desc_nr; /* compatibility */
1399 if (test_bit(Faulty, &rdev2->flags))
1400 d->state = (1<<MD_DISK_FAULTY);
1401 else if (is_active) {
1402 d->state = (1<<MD_DISK_ACTIVE);
1403 if (test_bit(In_sync, &rdev2->flags))
1404 d->state |= (1<<MD_DISK_SYNC);
1412 if (test_bit(WriteMostly, &rdev2->flags))
1413 d->state |= (1<<MD_DISK_WRITEMOSTLY);
1415 /* now set the "removed" and "faulty" bits on any missing devices */
1416 for (i=0 ; i < mddev->raid_disks ; i++) {
1417 mdp_disk_t *d = &sb->disks[i];
1418 if (d->state == 0 && d->number == 0) {
1421 d->state = (1<<MD_DISK_REMOVED);
1422 d->state |= (1<<MD_DISK_FAULTY);
1426 sb->nr_disks = nr_disks;
1427 sb->active_disks = active;
1428 sb->working_disks = working;
1429 sb->failed_disks = failed;
1430 sb->spare_disks = spare;
1432 sb->this_disk = sb->disks[rdev->desc_nr];
1433 sb->sb_csum = calc_sb_csum(sb);
1437 * rdev_size_change for 0.90.0
1439 static unsigned long long
1440 super_90_rdev_size_change(struct md_rdev *rdev, sector_t num_sectors)
1442 if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
1443 return 0; /* component must fit device */
1444 if (rdev->mddev->bitmap_info.offset)
1445 return 0; /* can't move bitmap */
1446 rdev->sb_start = calc_dev_sboffset(rdev);
1447 if (!num_sectors || num_sectors > rdev->sb_start)
1448 num_sectors = rdev->sb_start;
1449 /* Limit to 4TB as metadata cannot record more than that.
1450 * 4TB == 2^32 KB, or 2*2^32 sectors.
1452 if (num_sectors >= (2ULL << 32))
1453 num_sectors = (2ULL << 32) - 2;
1454 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
1456 md_super_wait(rdev->mddev);
1461 super_90_allow_new_offset(struct md_rdev *rdev, unsigned long long new_offset)
1463 /* non-zero offset changes not possible with v0.90 */
1464 return new_offset == 0;
1468 * version 1 superblock
1471 static __le32 calc_sb_1_csum(struct mdp_superblock_1 * sb)
1475 unsigned long long newcsum;
1476 int size = 256 + le32_to_cpu(sb->max_dev)*2;
1477 __le32 *isuper = (__le32*)sb;
1480 disk_csum = sb->sb_csum;
1483 for (i=0; size>=4; size -= 4 )
1484 newcsum += le32_to_cpu(*isuper++);
1487 newcsum += le16_to_cpu(*(__le16*) isuper);
1489 csum = (newcsum & 0xffffffff) + (newcsum >> 32);
1490 sb->sb_csum = disk_csum;
1491 return cpu_to_le32(csum);
1494 static int md_set_badblocks(struct badblocks *bb, sector_t s, int sectors,
1496 static int super_1_load(struct md_rdev *rdev, struct md_rdev *refdev, int minor_version)
1498 struct mdp_superblock_1 *sb;
1502 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
1506 * Calculate the position of the superblock in 512byte sectors.
1507 * It is always aligned to a 4K boundary and
1508 * depeding on minor_version, it can be:
1509 * 0: At least 8K, but less than 12K, from end of device
1510 * 1: At start of device
1511 * 2: 4K from start of device.
1513 switch(minor_version) {
1515 sb_start = i_size_read(rdev->bdev->bd_inode) >> 9;
1517 sb_start &= ~(sector_t)(4*2-1);
1528 rdev->sb_start = sb_start;
1530 /* superblock is rarely larger than 1K, but it can be larger,
1531 * and it is safe to read 4k, so we do that
1533 ret = read_disk_sb(rdev, 4096);
1534 if (ret) return ret;
1537 sb = page_address(rdev->sb_page);
1539 if (sb->magic != cpu_to_le32(MD_SB_MAGIC) ||
1540 sb->major_version != cpu_to_le32(1) ||
1541 le32_to_cpu(sb->max_dev) > (4096-256)/2 ||
1542 le64_to_cpu(sb->super_offset) != rdev->sb_start ||
1543 (le32_to_cpu(sb->feature_map) & ~MD_FEATURE_ALL) != 0)
1546 if (calc_sb_1_csum(sb) != sb->sb_csum) {
1547 printk("md: invalid superblock checksum on %s\n",
1548 bdevname(rdev->bdev,b));
1551 if (le64_to_cpu(sb->data_size) < 10) {
1552 printk("md: data_size too small on %s\n",
1553 bdevname(rdev->bdev,b));
1558 memcmp(sb->pad3, sb->pad3+1, sizeof(sb->pad3) - sizeof(sb->pad3[1])))
1559 /* Some padding is non-zero, might be a new feature */
1562 rdev->preferred_minor = 0xffff;
1563 rdev->data_offset = le64_to_cpu(sb->data_offset);
1564 rdev->new_data_offset = rdev->data_offset;
1565 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE) &&
1566 (le32_to_cpu(sb->feature_map) & MD_FEATURE_NEW_OFFSET))
1567 rdev->new_data_offset += (s32)le32_to_cpu(sb->new_offset);
1568 atomic_set(&rdev->corrected_errors, le32_to_cpu(sb->cnt_corrected_read));
1570 rdev->sb_size = le32_to_cpu(sb->max_dev) * 2 + 256;
1571 bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
1572 if (rdev->sb_size & bmask)
1573 rdev->sb_size = (rdev->sb_size | bmask) + 1;
1576 && rdev->data_offset < sb_start + (rdev->sb_size/512))
1579 && rdev->new_data_offset < sb_start + (rdev->sb_size/512))
1582 if (sb->level == cpu_to_le32(LEVEL_MULTIPATH))
1585 rdev->desc_nr = le32_to_cpu(sb->dev_number);
1587 if (!rdev->bb_page) {
1588 rdev->bb_page = alloc_page(GFP_KERNEL);
1592 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BAD_BLOCKS) &&
1593 rdev->badblocks.count == 0) {
1594 /* need to load the bad block list.
1595 * Currently we limit it to one page.
1601 int sectors = le16_to_cpu(sb->bblog_size);
1602 if (sectors > (PAGE_SIZE / 512))
1604 offset = le32_to_cpu(sb->bblog_offset);
1607 bb_sector = (long long)offset;
1608 if (!sync_page_io(rdev, bb_sector, sectors << 9,
1609 rdev->bb_page, READ, true))
1611 bbp = (u64 *)page_address(rdev->bb_page);
1612 rdev->badblocks.shift = sb->bblog_shift;
1613 for (i = 0 ; i < (sectors << (9-3)) ; i++, bbp++) {
1614 u64 bb = le64_to_cpu(*bbp);
1615 int count = bb & (0x3ff);
1616 u64 sector = bb >> 10;
1617 sector <<= sb->bblog_shift;
1618 count <<= sb->bblog_shift;
1621 if (md_set_badblocks(&rdev->badblocks,
1622 sector, count, 1) == 0)
1625 } else if (sb->bblog_offset == 0)
1626 rdev->badblocks.shift = -1;
1632 struct mdp_superblock_1 *refsb = page_address(refdev->sb_page);
1634 if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 ||
1635 sb->level != refsb->level ||
1636 sb->layout != refsb->layout ||
1637 sb->chunksize != refsb->chunksize) {
1638 printk(KERN_WARNING "md: %s has strangely different"
1639 " superblock to %s\n",
1640 bdevname(rdev->bdev,b),
1641 bdevname(refdev->bdev,b2));
1644 ev1 = le64_to_cpu(sb->events);
1645 ev2 = le64_to_cpu(refsb->events);
1652 if (minor_version) {
1653 sectors = (i_size_read(rdev->bdev->bd_inode) >> 9);
1654 sectors -= rdev->data_offset;
1656 sectors = rdev->sb_start;
1657 if (sectors < le64_to_cpu(sb->data_size))
1659 rdev->sectors = le64_to_cpu(sb->data_size);
1663 static int super_1_validate(struct mddev *mddev, struct md_rdev *rdev)
1665 struct mdp_superblock_1 *sb = page_address(rdev->sb_page);
1666 __u64 ev1 = le64_to_cpu(sb->events);
1668 rdev->raid_disk = -1;
1669 clear_bit(Faulty, &rdev->flags);
1670 clear_bit(In_sync, &rdev->flags);
1671 clear_bit(WriteMostly, &rdev->flags);
1673 if (mddev->raid_disks == 0) {
1674 mddev->major_version = 1;
1675 mddev->patch_version = 0;
1676 mddev->external = 0;
1677 mddev->chunk_sectors = le32_to_cpu(sb->chunksize);
1678 mddev->ctime = le64_to_cpu(sb->ctime) & ((1ULL << 32)-1);
1679 mddev->utime = le64_to_cpu(sb->utime) & ((1ULL << 32)-1);
1680 mddev->level = le32_to_cpu(sb->level);
1681 mddev->clevel[0] = 0;
1682 mddev->layout = le32_to_cpu(sb->layout);
1683 mddev->raid_disks = le32_to_cpu(sb->raid_disks);
1684 mddev->dev_sectors = le64_to_cpu(sb->size);
1685 mddev->events = ev1;
1686 mddev->bitmap_info.offset = 0;
1687 mddev->bitmap_info.space = 0;
1688 /* Default location for bitmap is 1K after superblock
1689 * using 3K - total of 4K
1691 mddev->bitmap_info.default_offset = 1024 >> 9;
1692 mddev->bitmap_info.default_space = (4096-1024) >> 9;
1693 mddev->reshape_backwards = 0;
1695 mddev->recovery_cp = le64_to_cpu(sb->resync_offset);
1696 memcpy(mddev->uuid, sb->set_uuid, 16);
1698 mddev->max_disks = (4096-256)/2;
1700 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) &&
1701 mddev->bitmap_info.file == NULL) {
1702 mddev->bitmap_info.offset =
1703 (__s32)le32_to_cpu(sb->bitmap_offset);
1704 /* Metadata doesn't record how much space is available.
1705 * For 1.0, we assume we can use up to the superblock
1706 * if before, else to 4K beyond superblock.
1707 * For others, assume no change is possible.
1709 if (mddev->minor_version > 0)
1710 mddev->bitmap_info.space = 0;
1711 else if (mddev->bitmap_info.offset > 0)
1712 mddev->bitmap_info.space =
1713 8 - mddev->bitmap_info.offset;
1715 mddev->bitmap_info.space =
1716 -mddev->bitmap_info.offset;
1719 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) {
1720 mddev->reshape_position = le64_to_cpu(sb->reshape_position);
1721 mddev->delta_disks = le32_to_cpu(sb->delta_disks);
1722 mddev->new_level = le32_to_cpu(sb->new_level);
1723 mddev->new_layout = le32_to_cpu(sb->new_layout);
1724 mddev->new_chunk_sectors = le32_to_cpu(sb->new_chunk);
1725 if (mddev->delta_disks < 0 ||
1726 (mddev->delta_disks == 0 &&
1727 (le32_to_cpu(sb->feature_map)
1728 & MD_FEATURE_RESHAPE_BACKWARDS)))
1729 mddev->reshape_backwards = 1;
1731 mddev->reshape_position = MaxSector;
1732 mddev->delta_disks = 0;
1733 mddev->new_level = mddev->level;
1734 mddev->new_layout = mddev->layout;
1735 mddev->new_chunk_sectors = mddev->chunk_sectors;
1738 } else if (mddev->pers == NULL) {
1739 /* Insist of good event counter while assembling, except for
1740 * spares (which don't need an event count) */
1742 if (rdev->desc_nr >= 0 &&
1743 rdev->desc_nr < le32_to_cpu(sb->max_dev) &&
1744 le16_to_cpu(sb->dev_roles[rdev->desc_nr]) < 0xfffe)
1745 if (ev1 < mddev->events)
1747 } else if (mddev->bitmap) {
1748 /* If adding to array with a bitmap, then we can accept an
1749 * older device, but not too old.
1751 if (ev1 < mddev->bitmap->events_cleared)
1754 if (ev1 < mddev->events)
1755 /* just a hot-add of a new device, leave raid_disk at -1 */
1758 if (mddev->level != LEVEL_MULTIPATH) {
1760 if (rdev->desc_nr < 0 ||
1761 rdev->desc_nr >= le32_to_cpu(sb->max_dev)) {
1765 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
1767 case 0xffff: /* spare */
1769 case 0xfffe: /* faulty */
1770 set_bit(Faulty, &rdev->flags);
1773 if ((le32_to_cpu(sb->feature_map) &
1774 MD_FEATURE_RECOVERY_OFFSET))
1775 rdev->recovery_offset = le64_to_cpu(sb->recovery_offset);
1777 set_bit(In_sync, &rdev->flags);
1778 rdev->raid_disk = role;
1781 if (sb->devflags & WriteMostly1)
1782 set_bit(WriteMostly, &rdev->flags);
1783 if (le32_to_cpu(sb->feature_map) & MD_FEATURE_REPLACEMENT)
1784 set_bit(Replacement, &rdev->flags);
1785 } else /* MULTIPATH are always insync */
1786 set_bit(In_sync, &rdev->flags);
1791 static void super_1_sync(struct mddev *mddev, struct md_rdev *rdev)
1793 struct mdp_superblock_1 *sb;
1794 struct md_rdev *rdev2;
1796 /* make rdev->sb match mddev and rdev data. */
1798 sb = page_address(rdev->sb_page);
1800 sb->feature_map = 0;
1802 sb->recovery_offset = cpu_to_le64(0);
1803 memset(sb->pad3, 0, sizeof(sb->pad3));
1805 sb->utime = cpu_to_le64((__u64)mddev->utime);
1806 sb->events = cpu_to_le64(mddev->events);
1808 sb->resync_offset = cpu_to_le64(mddev->recovery_cp);
1810 sb->resync_offset = cpu_to_le64(0);
1812 sb->cnt_corrected_read = cpu_to_le32(atomic_read(&rdev->corrected_errors));
1814 sb->raid_disks = cpu_to_le32(mddev->raid_disks);
1815 sb->size = cpu_to_le64(mddev->dev_sectors);
1816 sb->chunksize = cpu_to_le32(mddev->chunk_sectors);
1817 sb->level = cpu_to_le32(mddev->level);
1818 sb->layout = cpu_to_le32(mddev->layout);
1820 if (test_bit(WriteMostly, &rdev->flags))
1821 sb->devflags |= WriteMostly1;
1823 sb->devflags &= ~WriteMostly1;
1824 sb->data_offset = cpu_to_le64(rdev->data_offset);
1825 sb->data_size = cpu_to_le64(rdev->sectors);
1827 if (mddev->bitmap && mddev->bitmap_info.file == NULL) {
1828 sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_info.offset);
1829 sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET);
1832 if (rdev->raid_disk >= 0 &&
1833 !test_bit(In_sync, &rdev->flags)) {
1835 cpu_to_le32(MD_FEATURE_RECOVERY_OFFSET);
1836 sb->recovery_offset =
1837 cpu_to_le64(rdev->recovery_offset);
1839 if (test_bit(Replacement, &rdev->flags))
1841 cpu_to_le32(MD_FEATURE_REPLACEMENT);
1843 if (mddev->reshape_position != MaxSector) {
1844 sb->feature_map |= cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE);
1845 sb->reshape_position = cpu_to_le64(mddev->reshape_position);
1846 sb->new_layout = cpu_to_le32(mddev->new_layout);
1847 sb->delta_disks = cpu_to_le32(mddev->delta_disks);
1848 sb->new_level = cpu_to_le32(mddev->new_level);
1849 sb->new_chunk = cpu_to_le32(mddev->new_chunk_sectors);
1850 if (mddev->delta_disks == 0 &&
1851 mddev->reshape_backwards)
1853 |= cpu_to_le32(MD_FEATURE_RESHAPE_BACKWARDS);
1854 if (rdev->new_data_offset != rdev->data_offset) {
1856 |= cpu_to_le32(MD_FEATURE_NEW_OFFSET);
1857 sb->new_offset = cpu_to_le32((__u32)(rdev->new_data_offset
1858 - rdev->data_offset));
1862 if (rdev->badblocks.count == 0)
1863 /* Nothing to do for bad blocks*/ ;
1864 else if (sb->bblog_offset == 0)
1865 /* Cannot record bad blocks on this device */
1866 md_error(mddev, rdev);
1868 struct badblocks *bb = &rdev->badblocks;
1869 u64 *bbp = (u64 *)page_address(rdev->bb_page);
1871 sb->feature_map |= cpu_to_le32(MD_FEATURE_BAD_BLOCKS);
1876 seq = read_seqbegin(&bb->lock);
1878 memset(bbp, 0xff, PAGE_SIZE);
1880 for (i = 0 ; i < bb->count ; i++) {
1881 u64 internal_bb = *p++;
1882 u64 store_bb = ((BB_OFFSET(internal_bb) << 10)
1883 | BB_LEN(internal_bb));
1884 *bbp++ = cpu_to_le64(store_bb);
1887 if (read_seqretry(&bb->lock, seq))
1890 bb->sector = (rdev->sb_start +
1891 (int)le32_to_cpu(sb->bblog_offset));
1892 bb->size = le16_to_cpu(sb->bblog_size);
1897 rdev_for_each(rdev2, mddev)
1898 if (rdev2->desc_nr+1 > max_dev)
1899 max_dev = rdev2->desc_nr+1;
1901 if (max_dev > le32_to_cpu(sb->max_dev)) {
1903 sb->max_dev = cpu_to_le32(max_dev);
1904 rdev->sb_size = max_dev * 2 + 256;
1905 bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
1906 if (rdev->sb_size & bmask)
1907 rdev->sb_size = (rdev->sb_size | bmask) + 1;
1909 max_dev = le32_to_cpu(sb->max_dev);
1911 for (i=0; i<max_dev;i++)
1912 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1914 rdev_for_each(rdev2, mddev) {
1916 if (test_bit(Faulty, &rdev2->flags))
1917 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1918 else if (test_bit(In_sync, &rdev2->flags))
1919 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1920 else if (rdev2->raid_disk >= 0)
1921 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1923 sb->dev_roles[i] = cpu_to_le16(0xffff);
1926 sb->sb_csum = calc_sb_1_csum(sb);
1929 static unsigned long long
1930 super_1_rdev_size_change(struct md_rdev *rdev, sector_t num_sectors)
1932 struct mdp_superblock_1 *sb;
1933 sector_t max_sectors;
1934 if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
1935 return 0; /* component must fit device */
1936 if (rdev->data_offset != rdev->new_data_offset)
1937 return 0; /* too confusing */
1938 if (rdev->sb_start < rdev->data_offset) {
1939 /* minor versions 1 and 2; superblock before data */
1940 max_sectors = i_size_read(rdev->bdev->bd_inode) >> 9;
1941 max_sectors -= rdev->data_offset;
1942 if (!num_sectors || num_sectors > max_sectors)
1943 num_sectors = max_sectors;
1944 } else if (rdev->mddev->bitmap_info.offset) {
1945 /* minor version 0 with bitmap we can't move */
1948 /* minor version 0; superblock after data */
1950 sb_start = (i_size_read(rdev->bdev->bd_inode) >> 9) - 8*2;
1951 sb_start &= ~(sector_t)(4*2 - 1);
1952 max_sectors = rdev->sectors + sb_start - rdev->sb_start;
1953 if (!num_sectors || num_sectors > max_sectors)
1954 num_sectors = max_sectors;
1955 rdev->sb_start = sb_start;
1957 sb = page_address(rdev->sb_page);
1958 sb->data_size = cpu_to_le64(num_sectors);
1959 sb->super_offset = rdev->sb_start;
1960 sb->sb_csum = calc_sb_1_csum(sb);
1961 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
1963 md_super_wait(rdev->mddev);
1969 super_1_allow_new_offset(struct md_rdev *rdev,
1970 unsigned long long new_offset)
1972 /* All necessary checks on new >= old have been done */
1973 struct bitmap *bitmap;
1974 if (new_offset >= rdev->data_offset)
1977 /* with 1.0 metadata, there is no metadata to tread on
1978 * so we can always move back */
1979 if (rdev->mddev->minor_version == 0)
1982 /* otherwise we must be sure not to step on
1983 * any metadata, so stay:
1984 * 36K beyond start of superblock
1985 * beyond end of badblocks
1986 * beyond write-intent bitmap
1988 if (rdev->sb_start + (32+4)*2 > new_offset)
1990 bitmap = rdev->mddev->bitmap;
1991 if (bitmap && !rdev->mddev->bitmap_info.file &&
1992 rdev->sb_start + rdev->mddev->bitmap_info.offset +
1993 bitmap->storage.file_pages * (PAGE_SIZE>>9) > new_offset)
1995 if (rdev->badblocks.sector + rdev->badblocks.size > new_offset)
2001 static struct super_type super_types[] = {
2004 .owner = THIS_MODULE,
2005 .load_super = super_90_load,
2006 .validate_super = super_90_validate,
2007 .sync_super = super_90_sync,
2008 .rdev_size_change = super_90_rdev_size_change,
2009 .allow_new_offset = super_90_allow_new_offset,
2013 .owner = THIS_MODULE,
2014 .load_super = super_1_load,
2015 .validate_super = super_1_validate,
2016 .sync_super = super_1_sync,
2017 .rdev_size_change = super_1_rdev_size_change,
2018 .allow_new_offset = super_1_allow_new_offset,
2022 static void sync_super(struct mddev *mddev, struct md_rdev *rdev)
2024 if (mddev->sync_super) {
2025 mddev->sync_super(mddev, rdev);
2029 BUG_ON(mddev->major_version >= ARRAY_SIZE(super_types));
2031 super_types[mddev->major_version].sync_super(mddev, rdev);
2034 static int match_mddev_units(struct mddev *mddev1, struct mddev *mddev2)
2036 struct md_rdev *rdev, *rdev2;
2039 rdev_for_each_rcu(rdev, mddev1)
2040 rdev_for_each_rcu(rdev2, mddev2)
2041 if (rdev->bdev->bd_contains ==
2042 rdev2->bdev->bd_contains) {
2050 static LIST_HEAD(pending_raid_disks);
2053 * Try to register data integrity profile for an mddev
2055 * This is called when an array is started and after a disk has been kicked
2056 * from the array. It only succeeds if all working and active component devices
2057 * are integrity capable with matching profiles.
2059 int md_integrity_register(struct mddev *mddev)
2061 struct md_rdev *rdev, *reference = NULL;
2063 if (list_empty(&mddev->disks))
2064 return 0; /* nothing to do */
2065 if (!mddev->gendisk || blk_get_integrity(mddev->gendisk))
2066 return 0; /* shouldn't register, or already is */
2067 rdev_for_each(rdev, mddev) {
2068 /* skip spares and non-functional disks */
2069 if (test_bit(Faulty, &rdev->flags))
2071 if (rdev->raid_disk < 0)
2074 /* Use the first rdev as the reference */
2078 /* does this rdev's profile match the reference profile? */
2079 if (blk_integrity_compare(reference->bdev->bd_disk,
2080 rdev->bdev->bd_disk) < 0)
2083 if (!reference || !bdev_get_integrity(reference->bdev))
2086 * All component devices are integrity capable and have matching
2087 * profiles, register the common profile for the md device.
2089 if (blk_integrity_register(mddev->gendisk,
2090 bdev_get_integrity(reference->bdev)) != 0) {
2091 printk(KERN_ERR "md: failed to register integrity for %s\n",
2095 printk(KERN_NOTICE "md: data integrity enabled on %s\n", mdname(mddev));
2096 if (bioset_integrity_create(mddev->bio_set, BIO_POOL_SIZE)) {
2097 printk(KERN_ERR "md: failed to create integrity pool for %s\n",
2103 EXPORT_SYMBOL(md_integrity_register);
2105 /* Disable data integrity if non-capable/non-matching disk is being added */
2106 void md_integrity_add_rdev(struct md_rdev *rdev, struct mddev *mddev)
2108 struct blk_integrity *bi_rdev = bdev_get_integrity(rdev->bdev);
2109 struct blk_integrity *bi_mddev = blk_get_integrity(mddev->gendisk);
2111 if (!bi_mddev) /* nothing to do */
2113 if (rdev->raid_disk < 0) /* skip spares */
2115 if (bi_rdev && blk_integrity_compare(mddev->gendisk,
2116 rdev->bdev->bd_disk) >= 0)
2118 printk(KERN_NOTICE "disabling data integrity on %s\n", mdname(mddev));
2119 blk_integrity_unregister(mddev->gendisk);
2121 EXPORT_SYMBOL(md_integrity_add_rdev);
2123 static int bind_rdev_to_array(struct md_rdev * rdev, struct mddev * mddev)
2125 char b[BDEVNAME_SIZE];
2135 /* prevent duplicates */
2136 if (find_rdev(mddev, rdev->bdev->bd_dev))
2139 /* make sure rdev->sectors exceeds mddev->dev_sectors */
2140 if (rdev->sectors && (mddev->dev_sectors == 0 ||
2141 rdev->sectors < mddev->dev_sectors)) {
2143 /* Cannot change size, so fail
2144 * If mddev->level <= 0, then we don't care
2145 * about aligning sizes (e.g. linear)
2147 if (mddev->level > 0)
2150 mddev->dev_sectors = rdev->sectors;
2153 /* Verify rdev->desc_nr is unique.
2154 * If it is -1, assign a free number, else
2155 * check number is not in use
2157 if (rdev->desc_nr < 0) {
2159 if (mddev->pers) choice = mddev->raid_disks;
2160 while (find_rdev_nr(mddev, choice))
2162 rdev->desc_nr = choice;
2164 if (find_rdev_nr(mddev, rdev->desc_nr))
2167 if (mddev->max_disks && rdev->desc_nr >= mddev->max_disks) {
2168 printk(KERN_WARNING "md: %s: array is limited to %d devices\n",
2169 mdname(mddev), mddev->max_disks);
2172 bdevname(rdev->bdev,b);
2173 while ( (s=strchr(b, '/')) != NULL)
2176 rdev->mddev = mddev;
2177 printk(KERN_INFO "md: bind<%s>\n", b);
2179 if ((err = kobject_add(&rdev->kobj, &mddev->kobj, "dev-%s", b)))
2182 ko = &part_to_dev(rdev->bdev->bd_part)->kobj;
2183 if (sysfs_create_link(&rdev->kobj, ko, "block"))
2184 /* failure here is OK */;
2185 rdev->sysfs_state = sysfs_get_dirent_safe(rdev->kobj.sd, "state");
2187 list_add_rcu(&rdev->same_set, &mddev->disks);
2188 bd_link_disk_holder(rdev->bdev, mddev->gendisk);
2190 /* May as well allow recovery to be retried once */
2191 mddev->recovery_disabled++;
2196 printk(KERN_WARNING "md: failed to register dev-%s for %s\n",
2201 static void md_delayed_delete(struct work_struct *ws)
2203 struct md_rdev *rdev = container_of(ws, struct md_rdev, del_work);
2204 kobject_del(&rdev->kobj);
2205 kobject_put(&rdev->kobj);
2208 static void unbind_rdev_from_array(struct md_rdev * rdev)
2210 char b[BDEVNAME_SIZE];
2215 bd_unlink_disk_holder(rdev->bdev, rdev->mddev->gendisk);
2216 list_del_rcu(&rdev->same_set);
2217 printk(KERN_INFO "md: unbind<%s>\n", bdevname(rdev->bdev,b));
2219 sysfs_remove_link(&rdev->kobj, "block");
2220 sysfs_put(rdev->sysfs_state);
2221 rdev->sysfs_state = NULL;
2222 rdev->badblocks.count = 0;
2223 /* We need to delay this, otherwise we can deadlock when
2224 * writing to 'remove' to "dev/state". We also need
2225 * to delay it due to rcu usage.
2228 INIT_WORK(&rdev->del_work, md_delayed_delete);
2229 kobject_get(&rdev->kobj);
2230 queue_work(md_misc_wq, &rdev->del_work);
2234 * prevent the device from being mounted, repartitioned or
2235 * otherwise reused by a RAID array (or any other kernel
2236 * subsystem), by bd_claiming the device.
2238 static int lock_rdev(struct md_rdev *rdev, dev_t dev, int shared)
2241 struct block_device *bdev;
2242 char b[BDEVNAME_SIZE];
2244 bdev = blkdev_get_by_dev(dev, FMODE_READ|FMODE_WRITE|FMODE_EXCL,
2245 shared ? (struct md_rdev *)lock_rdev : rdev);
2247 printk(KERN_ERR "md: could not open %s.\n",
2248 __bdevname(dev, b));
2249 return PTR_ERR(bdev);
2255 static void unlock_rdev(struct md_rdev *rdev)
2257 struct block_device *bdev = rdev->bdev;
2261 blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
2264 void md_autodetect_dev(dev_t dev);
2266 static void export_rdev(struct md_rdev * rdev)
2268 char b[BDEVNAME_SIZE];
2269 printk(KERN_INFO "md: export_rdev(%s)\n",
2270 bdevname(rdev->bdev,b));
2273 md_rdev_clear(rdev);
2275 if (test_bit(AutoDetected, &rdev->flags))
2276 md_autodetect_dev(rdev->bdev->bd_dev);
2279 kobject_put(&rdev->kobj);
2282 static void kick_rdev_from_array(struct md_rdev * rdev)
2284 unbind_rdev_from_array(rdev);
2288 static void export_array(struct mddev *mddev)
2290 struct md_rdev *rdev, *tmp;
2292 rdev_for_each_safe(rdev, tmp, mddev) {
2297 kick_rdev_from_array(rdev);
2299 if (!list_empty(&mddev->disks))
2301 mddev->raid_disks = 0;
2302 mddev->major_version = 0;
2305 static void print_desc(mdp_disk_t *desc)
2307 printk(" DISK<N:%d,(%d,%d),R:%d,S:%d>\n", desc->number,
2308 desc->major,desc->minor,desc->raid_disk,desc->state);
2311 static void print_sb_90(mdp_super_t *sb)
2316 "md: SB: (V:%d.%d.%d) ID:<%08x.%08x.%08x.%08x> CT:%08x\n",
2317 sb->major_version, sb->minor_version, sb->patch_version,
2318 sb->set_uuid0, sb->set_uuid1, sb->set_uuid2, sb->set_uuid3,
2320 printk(KERN_INFO "md: L%d S%08d ND:%d RD:%d md%d LO:%d CS:%d\n",
2321 sb->level, sb->size, sb->nr_disks, sb->raid_disks,
2322 sb->md_minor, sb->layout, sb->chunk_size);
2323 printk(KERN_INFO "md: UT:%08x ST:%d AD:%d WD:%d"
2324 " FD:%d SD:%d CSUM:%08x E:%08lx\n",
2325 sb->utime, sb->state, sb->active_disks, sb->working_disks,
2326 sb->failed_disks, sb->spare_disks,
2327 sb->sb_csum, (unsigned long)sb->events_lo);
2330 for (i = 0; i < MD_SB_DISKS; i++) {
2333 desc = sb->disks + i;
2334 if (desc->number || desc->major || desc->minor ||
2335 desc->raid_disk || (desc->state && (desc->state != 4))) {
2336 printk(" D %2d: ", i);
2340 printk(KERN_INFO "md: THIS: ");
2341 print_desc(&sb->this_disk);
2344 static void print_sb_1(struct mdp_superblock_1 *sb)
2348 uuid = sb->set_uuid;
2350 "md: SB: (V:%u) (F:0x%08x) Array-ID:<%pU>\n"
2351 "md: Name: \"%s\" CT:%llu\n",
2352 le32_to_cpu(sb->major_version),
2353 le32_to_cpu(sb->feature_map),
2356 (unsigned long long)le64_to_cpu(sb->ctime)
2357 & MD_SUPERBLOCK_1_TIME_SEC_MASK);
2359 uuid = sb->device_uuid;
2361 "md: L%u SZ%llu RD:%u LO:%u CS:%u DO:%llu DS:%llu SO:%llu"
2363 "md: Dev:%08x UUID: %pU\n"
2364 "md: (F:0x%08x) UT:%llu Events:%llu ResyncOffset:%llu CSUM:0x%08x\n"
2365 "md: (MaxDev:%u) \n",
2366 le32_to_cpu(sb->level),
2367 (unsigned long long)le64_to_cpu(sb->size),
2368 le32_to_cpu(sb->raid_disks),
2369 le32_to_cpu(sb->layout),
2370 le32_to_cpu(sb->chunksize),
2371 (unsigned long long)le64_to_cpu(sb->data_offset),
2372 (unsigned long long)le64_to_cpu(sb->data_size),
2373 (unsigned long long)le64_to_cpu(sb->super_offset),
2374 (unsigned long long)le64_to_cpu(sb->recovery_offset),
2375 le32_to_cpu(sb->dev_number),
2378 (unsigned long long)le64_to_cpu(sb->utime) & MD_SUPERBLOCK_1_TIME_SEC_MASK,
2379 (unsigned long long)le64_to_cpu(sb->events),
2380 (unsigned long long)le64_to_cpu(sb->resync_offset),
2381 le32_to_cpu(sb->sb_csum),
2382 le32_to_cpu(sb->max_dev)
2386 static void print_rdev(struct md_rdev *rdev, int major_version)
2388 char b[BDEVNAME_SIZE];
2389 printk(KERN_INFO "md: rdev %s, Sect:%08llu F:%d S:%d DN:%u\n",
2390 bdevname(rdev->bdev, b), (unsigned long long)rdev->sectors,
2391 test_bit(Faulty, &rdev->flags), test_bit(In_sync, &rdev->flags),
2393 if (rdev->sb_loaded) {
2394 printk(KERN_INFO "md: rdev superblock (MJ:%d):\n", major_version);
2395 switch (major_version) {
2397 print_sb_90(page_address(rdev->sb_page));
2400 print_sb_1(page_address(rdev->sb_page));
2404 printk(KERN_INFO "md: no rdev superblock!\n");
2407 static void md_print_devices(void)
2409 struct list_head *tmp;
2410 struct md_rdev *rdev;
2411 struct mddev *mddev;
2412 char b[BDEVNAME_SIZE];
2415 printk("md: **********************************\n");
2416 printk("md: * <COMPLETE RAID STATE PRINTOUT> *\n");
2417 printk("md: **********************************\n");
2418 for_each_mddev(mddev, tmp) {
2421 bitmap_print_sb(mddev->bitmap);
2423 printk("%s: ", mdname(mddev));
2424 rdev_for_each(rdev, mddev)
2425 printk("<%s>", bdevname(rdev->bdev,b));
2428 rdev_for_each(rdev, mddev)
2429 print_rdev(rdev, mddev->major_version);
2431 printk("md: **********************************\n");
2436 static void sync_sbs(struct mddev * mddev, int nospares)
2438 /* Update each superblock (in-memory image), but
2439 * if we are allowed to, skip spares which already
2440 * have the right event counter, or have one earlier
2441 * (which would mean they aren't being marked as dirty
2442 * with the rest of the array)
2444 struct md_rdev *rdev;
2445 rdev_for_each(rdev, mddev) {
2446 if (rdev->sb_events == mddev->events ||
2448 rdev->raid_disk < 0 &&
2449 rdev->sb_events+1 == mddev->events)) {
2450 /* Don't update this superblock */
2451 rdev->sb_loaded = 2;
2453 sync_super(mddev, rdev);
2454 rdev->sb_loaded = 1;
2459 static void md_update_sb(struct mddev * mddev, int force_change)
2461 struct md_rdev *rdev;
2464 int any_badblocks_changed = 0;
2467 /* First make sure individual recovery_offsets are correct */
2468 rdev_for_each(rdev, mddev) {
2469 if (rdev->raid_disk >= 0 &&
2470 mddev->delta_disks >= 0 &&
2471 !test_bit(In_sync, &rdev->flags) &&
2472 mddev->curr_resync_completed > rdev->recovery_offset)
2473 rdev->recovery_offset = mddev->curr_resync_completed;
2476 if (!mddev->persistent) {
2477 clear_bit(MD_CHANGE_CLEAN, &mddev->flags);
2478 clear_bit(MD_CHANGE_DEVS, &mddev->flags);
2479 if (!mddev->external) {
2480 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
2481 rdev_for_each(rdev, mddev) {
2482 if (rdev->badblocks.changed) {
2483 rdev->badblocks.changed = 0;
2484 md_ack_all_badblocks(&rdev->badblocks);
2485 md_error(mddev, rdev);
2487 clear_bit(Blocked, &rdev->flags);
2488 clear_bit(BlockedBadBlocks, &rdev->flags);
2489 wake_up(&rdev->blocked_wait);
2492 wake_up(&mddev->sb_wait);
2496 spin_lock_irq(&mddev->write_lock);
2498 mddev->utime = get_seconds();
2500 if (test_and_clear_bit(MD_CHANGE_DEVS, &mddev->flags))
2502 if (test_and_clear_bit(MD_CHANGE_CLEAN, &mddev->flags))
2503 /* just a clean<-> dirty transition, possibly leave spares alone,
2504 * though if events isn't the right even/odd, we will have to do
2510 if (mddev->degraded)
2511 /* If the array is degraded, then skipping spares is both
2512 * dangerous and fairly pointless.
2513 * Dangerous because a device that was removed from the array
2514 * might have a event_count that still looks up-to-date,
2515 * so it can be re-added without a resync.
2516 * Pointless because if there are any spares to skip,
2517 * then a recovery will happen and soon that array won't
2518 * be degraded any more and the spare can go back to sleep then.
2522 sync_req = mddev->in_sync;
2524 /* If this is just a dirty<->clean transition, and the array is clean
2525 * and 'events' is odd, we can roll back to the previous clean state */
2527 && (mddev->in_sync && mddev->recovery_cp == MaxSector)
2528 && mddev->can_decrease_events
2529 && mddev->events != 1) {
2531 mddev->can_decrease_events = 0;
2533 /* otherwise we have to go forward and ... */
2535 mddev->can_decrease_events = nospares;
2538 if (!mddev->events) {
2540 * oops, this 64-bit counter should never wrap.
2541 * Either we are in around ~1 trillion A.C., assuming
2542 * 1 reboot per second, or we have a bug:
2548 rdev_for_each(rdev, mddev) {
2549 if (rdev->badblocks.changed)
2550 any_badblocks_changed++;
2551 if (test_bit(Faulty, &rdev->flags))
2552 set_bit(FaultRecorded, &rdev->flags);
2555 sync_sbs(mddev, nospares);
2556 spin_unlock_irq(&mddev->write_lock);
2558 pr_debug("md: updating %s RAID superblock on device (in sync %d)\n",
2559 mdname(mddev), mddev->in_sync);
2561 bitmap_update_sb(mddev->bitmap);
2562 rdev_for_each(rdev, mddev) {
2563 char b[BDEVNAME_SIZE];
2565 if (rdev->sb_loaded != 1)
2566 continue; /* no noise on spare devices */
2568 if (!test_bit(Faulty, &rdev->flags) &&
2569 rdev->saved_raid_disk == -1) {
2570 md_super_write(mddev,rdev,
2571 rdev->sb_start, rdev->sb_size,
2573 pr_debug("md: (write) %s's sb offset: %llu\n",
2574 bdevname(rdev->bdev, b),
2575 (unsigned long long)rdev->sb_start);
2576 rdev->sb_events = mddev->events;
2577 if (rdev->badblocks.size) {
2578 md_super_write(mddev, rdev,
2579 rdev->badblocks.sector,
2580 rdev->badblocks.size << 9,
2582 rdev->badblocks.size = 0;
2585 } else if (test_bit(Faulty, &rdev->flags))
2586 pr_debug("md: %s (skipping faulty)\n",
2587 bdevname(rdev->bdev, b));
2589 pr_debug("(skipping incremental s/r ");
2591 if (mddev->level == LEVEL_MULTIPATH)
2592 /* only need to write one superblock... */
2595 md_super_wait(mddev);
2596 /* if there was a failure, MD_CHANGE_DEVS was set, and we re-write super */
2598 spin_lock_irq(&mddev->write_lock);
2599 if (mddev->in_sync != sync_req ||
2600 test_bit(MD_CHANGE_DEVS, &mddev->flags)) {
2601 /* have to write it out again */
2602 spin_unlock_irq(&mddev->write_lock);
2605 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
2606 spin_unlock_irq(&mddev->write_lock);
2607 wake_up(&mddev->sb_wait);
2608 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
2609 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
2611 rdev_for_each(rdev, mddev) {
2612 if (test_and_clear_bit(FaultRecorded, &rdev->flags))
2613 clear_bit(Blocked, &rdev->flags);
2615 if (any_badblocks_changed)
2616 md_ack_all_badblocks(&rdev->badblocks);
2617 clear_bit(BlockedBadBlocks, &rdev->flags);
2618 wake_up(&rdev->blocked_wait);
2622 /* words written to sysfs files may, or may not, be \n terminated.
2623 * We want to accept with case. For this we use cmd_match.
2625 static int cmd_match(const char *cmd, const char *str)
2627 /* See if cmd, written into a sysfs file, matches
2628 * str. They must either be the same, or cmd can
2629 * have a trailing newline
2631 while (*cmd && *str && *cmd == *str) {
2642 struct rdev_sysfs_entry {
2643 struct attribute attr;
2644 ssize_t (*show)(struct md_rdev *, char *);
2645 ssize_t (*store)(struct md_rdev *, const char *, size_t);
2649 state_show(struct md_rdev *rdev, char *page)
2654 if (test_bit(Faulty, &rdev->flags) ||
2655 rdev->badblocks.unacked_exist) {
2656 len+= sprintf(page+len, "%sfaulty",sep);
2659 if (test_bit(In_sync, &rdev->flags)) {
2660 len += sprintf(page+len, "%sin_sync",sep);
2663 if (test_bit(WriteMostly, &rdev->flags)) {
2664 len += sprintf(page+len, "%swrite_mostly",sep);
2667 if (test_bit(Blocked, &rdev->flags) ||
2668 (rdev->badblocks.unacked_exist
2669 && !test_bit(Faulty, &rdev->flags))) {
2670 len += sprintf(page+len, "%sblocked", sep);
2673 if (!test_bit(Faulty, &rdev->flags) &&
2674 !test_bit(In_sync, &rdev->flags)) {
2675 len += sprintf(page+len, "%sspare", sep);
2678 if (test_bit(WriteErrorSeen, &rdev->flags)) {
2679 len += sprintf(page+len, "%swrite_error", sep);
2682 if (test_bit(WantReplacement, &rdev->flags)) {
2683 len += sprintf(page+len, "%swant_replacement", sep);
2686 if (test_bit(Replacement, &rdev->flags)) {
2687 len += sprintf(page+len, "%sreplacement", sep);
2691 return len+sprintf(page+len, "\n");
2695 state_store(struct md_rdev *rdev, const char *buf, size_t len)
2698 * faulty - simulates an error
2699 * remove - disconnects the device
2700 * writemostly - sets write_mostly
2701 * -writemostly - clears write_mostly
2702 * blocked - sets the Blocked flags
2703 * -blocked - clears the Blocked and possibly simulates an error
2704 * insync - sets Insync providing device isn't active
2705 * write_error - sets WriteErrorSeen
2706 * -write_error - clears WriteErrorSeen
2709 if (cmd_match(buf, "faulty") && rdev->mddev->pers) {
2710 md_error(rdev->mddev, rdev);
2711 if (test_bit(Faulty, &rdev->flags))
2715 } else if (cmd_match(buf, "remove")) {
2716 if (rdev->raid_disk >= 0)
2719 struct mddev *mddev = rdev->mddev;
2720 kick_rdev_from_array(rdev);
2722 md_update_sb(mddev, 1);
2723 md_new_event(mddev);
2726 } else if (cmd_match(buf, "writemostly")) {
2727 set_bit(WriteMostly, &rdev->flags);
2729 } else if (cmd_match(buf, "-writemostly")) {
2730 clear_bit(WriteMostly, &rdev->flags);
2732 } else if (cmd_match(buf, "blocked")) {
2733 set_bit(Blocked, &rdev->flags);
2735 } else if (cmd_match(buf, "-blocked")) {
2736 if (!test_bit(Faulty, &rdev->flags) &&
2737 rdev->badblocks.unacked_exist) {
2738 /* metadata handler doesn't understand badblocks,
2739 * so we need to fail the device
2741 md_error(rdev->mddev, rdev);
2743 clear_bit(Blocked, &rdev->flags);
2744 clear_bit(BlockedBadBlocks, &rdev->flags);
2745 wake_up(&rdev->blocked_wait);
2746 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2747 md_wakeup_thread(rdev->mddev->thread);
2750 } else if (cmd_match(buf, "insync") && rdev->raid_disk == -1) {
2751 set_bit(In_sync, &rdev->flags);
2753 } else if (cmd_match(buf, "write_error")) {
2754 set_bit(WriteErrorSeen, &rdev->flags);
2756 } else if (cmd_match(buf, "-write_error")) {
2757 clear_bit(WriteErrorSeen, &rdev->flags);
2759 } else if (cmd_match(buf, "want_replacement")) {
2760 /* Any non-spare device that is not a replacement can
2761 * become want_replacement at any time, but we then need to
2762 * check if recovery is needed.
2764 if (rdev->raid_disk >= 0 &&
2765 !test_bit(Replacement, &rdev->flags))
2766 set_bit(WantReplacement, &rdev->flags);
2767 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2768 md_wakeup_thread(rdev->mddev->thread);
2770 } else if (cmd_match(buf, "-want_replacement")) {
2771 /* Clearing 'want_replacement' is always allowed.
2772 * Once replacements starts it is too late though.
2775 clear_bit(WantReplacement, &rdev->flags);
2776 } else if (cmd_match(buf, "replacement")) {
2777 /* Can only set a device as a replacement when array has not
2778 * yet been started. Once running, replacement is automatic
2779 * from spares, or by assigning 'slot'.
2781 if (rdev->mddev->pers)
2784 set_bit(Replacement, &rdev->flags);
2787 } else if (cmd_match(buf, "-replacement")) {
2788 /* Similarly, can only clear Replacement before start */
2789 if (rdev->mddev->pers)
2792 clear_bit(Replacement, &rdev->flags);
2797 sysfs_notify_dirent_safe(rdev->sysfs_state);
2798 return err ? err : len;
2800 static struct rdev_sysfs_entry rdev_state =
2801 __ATTR(state, S_IRUGO|S_IWUSR, state_show, state_store);
2804 errors_show(struct md_rdev *rdev, char *page)
2806 return sprintf(page, "%d\n", atomic_read(&rdev->corrected_errors));
2810 errors_store(struct md_rdev *rdev, const char *buf, size_t len)
2813 unsigned long n = simple_strtoul(buf, &e, 10);
2814 if (*buf && (*e == 0 || *e == '\n')) {
2815 atomic_set(&rdev->corrected_errors, n);
2820 static struct rdev_sysfs_entry rdev_errors =
2821 __ATTR(errors, S_IRUGO|S_IWUSR, errors_show, errors_store);
2824 slot_show(struct md_rdev *rdev, char *page)
2826 if (rdev->raid_disk < 0)
2827 return sprintf(page, "none\n");
2829 return sprintf(page, "%d\n", rdev->raid_disk);
2833 slot_store(struct md_rdev *rdev, const char *buf, size_t len)
2837 int slot = simple_strtoul(buf, &e, 10);
2838 if (strncmp(buf, "none", 4)==0)
2840 else if (e==buf || (*e && *e!= '\n'))
2842 if (rdev->mddev->pers && slot == -1) {
2843 /* Setting 'slot' on an active array requires also
2844 * updating the 'rd%d' link, and communicating
2845 * with the personality with ->hot_*_disk.
2846 * For now we only support removing
2847 * failed/spare devices. This normally happens automatically,
2848 * but not when the metadata is externally managed.
2850 if (rdev->raid_disk == -1)
2852 /* personality does all needed checks */
2853 if (rdev->mddev->pers->hot_remove_disk == NULL)
2855 err = rdev->mddev->pers->
2856 hot_remove_disk(rdev->mddev, rdev);
2859 sysfs_unlink_rdev(rdev->mddev, rdev);
2860 rdev->raid_disk = -1;
2861 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2862 md_wakeup_thread(rdev->mddev->thread);
2863 } else if (rdev->mddev->pers) {
2864 /* Activating a spare .. or possibly reactivating
2865 * if we ever get bitmaps working here.
2868 if (rdev->raid_disk != -1)
2871 if (test_bit(MD_RECOVERY_RUNNING, &rdev->mddev->recovery))
2874 if (rdev->mddev->pers->hot_add_disk == NULL)
2877 if (slot >= rdev->mddev->raid_disks &&
2878 slot >= rdev->mddev->raid_disks + rdev->mddev->delta_disks)
2881 rdev->raid_disk = slot;
2882 if (test_bit(In_sync, &rdev->flags))
2883 rdev->saved_raid_disk = slot;
2885 rdev->saved_raid_disk = -1;
2886 clear_bit(In_sync, &rdev->flags);
2887 err = rdev->mddev->pers->
2888 hot_add_disk(rdev->mddev, rdev);
2890 rdev->raid_disk = -1;
2893 sysfs_notify_dirent_safe(rdev->sysfs_state);
2894 if (sysfs_link_rdev(rdev->mddev, rdev))
2895 /* failure here is OK */;
2896 /* don't wakeup anyone, leave that to userspace. */
2898 if (slot >= rdev->mddev->raid_disks &&
2899 slot >= rdev->mddev->raid_disks + rdev->mddev->delta_disks)
2901 rdev->raid_disk = slot;
2902 /* assume it is working */
2903 clear_bit(Faulty, &rdev->flags);
2904 clear_bit(WriteMostly, &rdev->flags);
2905 set_bit(In_sync, &rdev->flags);
2906 sysfs_notify_dirent_safe(rdev->sysfs_state);
2912 static struct rdev_sysfs_entry rdev_slot =
2913 __ATTR(slot, S_IRUGO|S_IWUSR, slot_show, slot_store);
2916 offset_show(struct md_rdev *rdev, char *page)
2918 return sprintf(page, "%llu\n", (unsigned long long)rdev->data_offset);
2922 offset_store(struct md_rdev *rdev, const char *buf, size_t len)
2924 unsigned long long offset;
2925 if (strict_strtoull(buf, 10, &offset) < 0)
2927 if (rdev->mddev->pers && rdev->raid_disk >= 0)
2929 if (rdev->sectors && rdev->mddev->external)
2930 /* Must set offset before size, so overlap checks
2933 rdev->data_offset = offset;
2934 rdev->new_data_offset = offset;
2938 static struct rdev_sysfs_entry rdev_offset =
2939 __ATTR(offset, S_IRUGO|S_IWUSR, offset_show, offset_store);
2941 static ssize_t new_offset_show(struct md_rdev *rdev, char *page)
2943 return sprintf(page, "%llu\n",
2944 (unsigned long long)rdev->new_data_offset);
2947 static ssize_t new_offset_store(struct md_rdev *rdev,
2948 const char *buf, size_t len)
2950 unsigned long long new_offset;
2951 struct mddev *mddev = rdev->mddev;
2953 if (strict_strtoull(buf, 10, &new_offset) < 0)
2956 if (mddev->sync_thread)
2958 if (new_offset == rdev->data_offset)
2959 /* reset is always permitted */
2961 else if (new_offset > rdev->data_offset) {
2962 /* must not push array size beyond rdev_sectors */
2963 if (new_offset - rdev->data_offset
2964 + mddev->dev_sectors > rdev->sectors)
2967 /* Metadata worries about other space details. */
2969 /* decreasing the offset is inconsistent with a backwards
2972 if (new_offset < rdev->data_offset &&
2973 mddev->reshape_backwards)
2975 /* Increasing offset is inconsistent with forwards
2976 * reshape. reshape_direction should be set to
2977 * 'backwards' first.
2979 if (new_offset > rdev->data_offset &&
2980 !mddev->reshape_backwards)
2983 if (mddev->pers && mddev->persistent &&
2984 !super_types[mddev->major_version]
2985 .allow_new_offset(rdev, new_offset))
2987 rdev->new_data_offset = new_offset;
2988 if (new_offset > rdev->data_offset)
2989 mddev->reshape_backwards = 1;
2990 else if (new_offset < rdev->data_offset)
2991 mddev->reshape_backwards = 0;
2995 static struct rdev_sysfs_entry rdev_new_offset =
2996 __ATTR(new_offset, S_IRUGO|S_IWUSR, new_offset_show, new_offset_store);
2999 rdev_size_show(struct md_rdev *rdev, char *page)
3001 return sprintf(page, "%llu\n", (unsigned long long)rdev->sectors / 2);
3004 static int overlaps(sector_t s1, sector_t l1, sector_t s2, sector_t l2)
3006 /* check if two start/length pairs overlap */
3014 static int strict_blocks_to_sectors(const char *buf, sector_t *sectors)
3016 unsigned long long blocks;
3019 if (strict_strtoull(buf, 10, &blocks) < 0)
3022 if (blocks & 1ULL << (8 * sizeof(blocks) - 1))
3023 return -EINVAL; /* sector conversion overflow */
3026 if (new != blocks * 2)
3027 return -EINVAL; /* unsigned long long to sector_t overflow */
3034 rdev_size_store(struct md_rdev *rdev, const char *buf, size_t len)
3036 struct mddev *my_mddev = rdev->mddev;
3037 sector_t oldsectors = rdev->sectors;
3040 if (strict_blocks_to_sectors(buf, §ors) < 0)
3042 if (rdev->data_offset != rdev->new_data_offset)
3043 return -EINVAL; /* too confusing */
3044 if (my_mddev->pers && rdev->raid_disk >= 0) {
3045 if (my_mddev->persistent) {
3046 sectors = super_types[my_mddev->major_version].
3047 rdev_size_change(rdev, sectors);
3050 } else if (!sectors)
3051 sectors = (i_size_read(rdev->bdev->bd_inode) >> 9) -
3054 if (sectors < my_mddev->dev_sectors)
3055 return -EINVAL; /* component must fit device */
3057 rdev->sectors = sectors;
3058 if (sectors > oldsectors && my_mddev->external) {
3059 /* need to check that all other rdevs with the same ->bdev
3060 * do not overlap. We need to unlock the mddev to avoid
3061 * a deadlock. We have already changed rdev->sectors, and if
3062 * we have to change it back, we will have the lock again.
3064 struct mddev *mddev;
3066 struct list_head *tmp;
3068 mddev_unlock(my_mddev);
3069 for_each_mddev(mddev, tmp) {
3070 struct md_rdev *rdev2;
3073 rdev_for_each(rdev2, mddev)
3074 if (rdev->bdev == rdev2->bdev &&
3076 overlaps(rdev->data_offset, rdev->sectors,
3082 mddev_unlock(mddev);
3088 mddev_lock(my_mddev);
3090 /* Someone else could have slipped in a size
3091 * change here, but doing so is just silly.
3092 * We put oldsectors back because we *know* it is
3093 * safe, and trust userspace not to race with
3096 rdev->sectors = oldsectors;
3103 static struct rdev_sysfs_entry rdev_size =
3104 __ATTR(size, S_IRUGO|S_IWUSR, rdev_size_show, rdev_size_store);
3107 static ssize_t recovery_start_show(struct md_rdev *rdev, char *page)
3109 unsigned long long recovery_start = rdev->recovery_offset;
3111 if (test_bit(In_sync, &rdev->flags) ||
3112 recovery_start == MaxSector)
3113 return sprintf(page, "none\n");
3115 return sprintf(page, "%llu\n", recovery_start);
3118 static ssize_t recovery_start_store(struct md_rdev *rdev, const char *buf, size_t len)
3120 unsigned long long recovery_start;
3122 if (cmd_match(buf, "none"))
3123 recovery_start = MaxSector;
3124 else if (strict_strtoull(buf, 10, &recovery_start))
3127 if (rdev->mddev->pers &&
3128 rdev->raid_disk >= 0)
3131 rdev->recovery_offset = recovery_start;
3132 if (recovery_start == MaxSector)
3133 set_bit(In_sync, &rdev->flags);
3135 clear_bit(In_sync, &rdev->flags);
3139 static struct rdev_sysfs_entry rdev_recovery_start =
3140 __ATTR(recovery_start, S_IRUGO|S_IWUSR, recovery_start_show, recovery_start_store);
3144 badblocks_show(struct badblocks *bb, char *page, int unack);
3146 badblocks_store(struct badblocks *bb, const char *page, size_t len, int unack);
3148 static ssize_t bb_show(struct md_rdev *rdev, char *page)
3150 return badblocks_show(&rdev->badblocks, page, 0);
3152 static ssize_t bb_store(struct md_rdev *rdev, const char *page, size_t len)
3154 int rv = badblocks_store(&rdev->badblocks, page, len, 0);
3155 /* Maybe that ack was all we needed */
3156 if (test_and_clear_bit(BlockedBadBlocks, &rdev->flags))
3157 wake_up(&rdev->blocked_wait);
3160 static struct rdev_sysfs_entry rdev_bad_blocks =
3161 __ATTR(bad_blocks, S_IRUGO|S_IWUSR, bb_show, bb_store);
3164 static ssize_t ubb_show(struct md_rdev *rdev, char *page)
3166 return badblocks_show(&rdev->badblocks, page, 1);
3168 static ssize_t ubb_store(struct md_rdev *rdev, const char *page, size_t len)
3170 return badblocks_store(&rdev->badblocks, page, len, 1);
3172 static struct rdev_sysfs_entry rdev_unack_bad_blocks =
3173 __ATTR(unacknowledged_bad_blocks, S_IRUGO|S_IWUSR, ubb_show, ubb_store);
3175 static struct attribute *rdev_default_attrs[] = {
3180 &rdev_new_offset.attr,
3182 &rdev_recovery_start.attr,
3183 &rdev_bad_blocks.attr,
3184 &rdev_unack_bad_blocks.attr,
3188 rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
3190 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
3191 struct md_rdev *rdev = container_of(kobj, struct md_rdev, kobj);
3192 struct mddev *mddev = rdev->mddev;
3198 rv = mddev ? mddev_lock(mddev) : -EBUSY;
3200 if (rdev->mddev == NULL)
3203 rv = entry->show(rdev, page);
3204 mddev_unlock(mddev);
3210 rdev_attr_store(struct kobject *kobj, struct attribute *attr,
3211 const char *page, size_t length)
3213 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
3214 struct md_rdev *rdev = container_of(kobj, struct md_rdev, kobj);
3216 struct mddev *mddev = rdev->mddev;
3220 if (!capable(CAP_SYS_ADMIN))
3222 rv = mddev ? mddev_lock(mddev): -EBUSY;
3224 if (rdev->mddev == NULL)
3227 rv = entry->store(rdev, page, length);
3228 mddev_unlock(mddev);
3233 static void rdev_free(struct kobject *ko)
3235 struct md_rdev *rdev = container_of(ko, struct md_rdev, kobj);
3238 static const struct sysfs_ops rdev_sysfs_ops = {
3239 .show = rdev_attr_show,
3240 .store = rdev_attr_store,
3242 static struct kobj_type rdev_ktype = {
3243 .release = rdev_free,
3244 .sysfs_ops = &rdev_sysfs_ops,
3245 .default_attrs = rdev_default_attrs,
3248 int md_rdev_init(struct md_rdev *rdev)
3251 rdev->saved_raid_disk = -1;
3252 rdev->raid_disk = -1;
3254 rdev->data_offset = 0;
3255 rdev->new_data_offset = 0;
3256 rdev->sb_events = 0;
3257 rdev->last_read_error.tv_sec = 0;
3258 rdev->last_read_error.tv_nsec = 0;
3259 rdev->sb_loaded = 0;
3260 rdev->bb_page = NULL;
3261 atomic_set(&rdev->nr_pending, 0);
3262 atomic_set(&rdev->read_errors, 0);
3263 atomic_set(&rdev->corrected_errors, 0);
3265 INIT_LIST_HEAD(&rdev->same_set);
3266 init_waitqueue_head(&rdev->blocked_wait);
3268 /* Add space to store bad block list.
3269 * This reserves the space even on arrays where it cannot
3270 * be used - I wonder if that matters
3272 rdev->badblocks.count = 0;
3273 rdev->badblocks.shift = 0;
3274 rdev->badblocks.page = kmalloc(PAGE_SIZE, GFP_KERNEL);
3275 seqlock_init(&rdev->badblocks.lock);
3276 if (rdev->badblocks.page == NULL)
3281 EXPORT_SYMBOL_GPL(md_rdev_init);
3283 * Import a device. If 'super_format' >= 0, then sanity check the superblock
3285 * mark the device faulty if:
3287 * - the device is nonexistent (zero size)
3288 * - the device has no valid superblock
3290 * a faulty rdev _never_ has rdev->sb set.
3292 static struct md_rdev *md_import_device(dev_t newdev, int super_format, int super_minor)
3294 char b[BDEVNAME_SIZE];
3296 struct md_rdev *rdev;
3299 rdev = kzalloc(sizeof(*rdev), GFP_KERNEL);
3301 printk(KERN_ERR "md: could not alloc mem for new device!\n");
3302 return ERR_PTR(-ENOMEM);
3305 err = md_rdev_init(rdev);
3308 err = alloc_disk_sb(rdev);
3312 err = lock_rdev(rdev, newdev, super_format == -2);
3316 kobject_init(&rdev->kobj, &rdev_ktype);
3318 size = i_size_read(rdev->bdev->bd_inode) >> BLOCK_SIZE_BITS;
3321 "md: %s has zero or unknown size, marking faulty!\n",
3322 bdevname(rdev->bdev,b));
3327 if (super_format >= 0) {
3328 err = super_types[super_format].
3329 load_super(rdev, NULL, super_minor);
3330 if (err == -EINVAL) {
3332 "md: %s does not have a valid v%d.%d "
3333 "superblock, not importing!\n",
3334 bdevname(rdev->bdev,b),
3335 super_format, super_minor);
3340 "md: could not read %s's sb, not importing!\n",
3341 bdevname(rdev->bdev,b));
3345 if (super_format == -1)
3346 /* hot-add for 0.90, or non-persistent: so no badblocks */
3347 rdev->badblocks.shift = -1;
3354 md_rdev_clear(rdev);
3356 return ERR_PTR(err);
3360 * Check a full RAID array for plausibility
3364 static void analyze_sbs(struct mddev * mddev)
3367 struct md_rdev *rdev, *freshest, *tmp;
3368 char b[BDEVNAME_SIZE];
3371 rdev_for_each_safe(rdev, tmp, mddev)
3372 switch (super_types[mddev->major_version].
3373 load_super(rdev, freshest, mddev->minor_version)) {
3381 "md: fatal superblock inconsistency in %s"
3382 " -- removing from array\n",
3383 bdevname(rdev->bdev,b));
3384 kick_rdev_from_array(rdev);
3388 super_types[mddev->major_version].
3389 validate_super(mddev, freshest);
3392 rdev_for_each_safe(rdev, tmp, mddev) {
3393 if (mddev->max_disks &&
3394 (rdev->desc_nr >= mddev->max_disks ||
3395 i > mddev->max_disks)) {
3397 "md: %s: %s: only %d devices permitted\n",
3398 mdname(mddev), bdevname(rdev->bdev, b),
3400 kick_rdev_from_array(rdev);
3403 if (rdev != freshest)
3404 if (super_types[mddev->major_version].
3405 validate_super(mddev, rdev)) {
3406 printk(KERN_WARNING "md: kicking non-fresh %s"
3408 bdevname(rdev->bdev,b));
3409 kick_rdev_from_array(rdev);
3412 if (mddev->level == LEVEL_MULTIPATH) {
3413 rdev->desc_nr = i++;
3414 rdev->raid_disk = rdev->desc_nr;
3415 set_bit(In_sync, &rdev->flags);
3416 } else if (rdev->raid_disk >= (mddev->raid_disks - min(0, mddev->delta_disks))) {
3417 rdev->raid_disk = -1;
3418 clear_bit(In_sync, &rdev->flags);
3423 /* Read a fixed-point number.
3424 * Numbers in sysfs attributes should be in "standard" units where
3425 * possible, so time should be in seconds.
3426 * However we internally use a a much smaller unit such as
3427 * milliseconds or jiffies.
3428 * This function takes a decimal number with a possible fractional
3429 * component, and produces an integer which is the result of
3430 * multiplying that number by 10^'scale'.
3431 * all without any floating-point arithmetic.
3433 int strict_strtoul_scaled(const char *cp, unsigned long *res, int scale)
3435 unsigned long result = 0;
3437 while (isdigit(*cp) || (*cp == '.' && decimals < 0)) {
3440 else if (decimals < scale) {
3443 result = result * 10 + value;
3455 while (decimals < scale) {
3464 static void md_safemode_timeout(unsigned long data);
3467 safe_delay_show(struct mddev *mddev, char *page)
3469 int msec = (mddev->safemode_delay*1000)/HZ;
3470 return sprintf(page, "%d.%03d\n", msec/1000, msec%1000);
3473 safe_delay_store(struct mddev *mddev, const char *cbuf, size_t len)
3477 if (strict_strtoul_scaled(cbuf, &msec, 3) < 0)
3480 mddev->safemode_delay = 0;
3482 unsigned long old_delay = mddev->safemode_delay;
3483 mddev->safemode_delay = (msec*HZ)/1000;
3484 if (mddev->safemode_delay == 0)
3485 mddev->safemode_delay = 1;
3486 if (mddev->safemode_delay < old_delay)
3487 md_safemode_timeout((unsigned long)mddev);
3491 static struct md_sysfs_entry md_safe_delay =
3492 __ATTR(safe_mode_delay, S_IRUGO|S_IWUSR,safe_delay_show, safe_delay_store);
3495 level_show(struct mddev *mddev, char *page)
3497 struct md_personality *p = mddev->pers;
3499 return sprintf(page, "%s\n", p->name);
3500 else if (mddev->clevel[0])
3501 return sprintf(page, "%s\n", mddev->clevel);
3502 else if (mddev->level != LEVEL_NONE)
3503 return sprintf(page, "%d\n", mddev->level);
3509 level_store(struct mddev *mddev, const char *buf, size_t len)
3513 struct md_personality *pers;
3516 struct md_rdev *rdev;
3518 if (mddev->pers == NULL) {
3521 if (len >= sizeof(mddev->clevel))
3523 strncpy(mddev->clevel, buf, len);
3524 if (mddev->clevel[len-1] == '\n')
3526 mddev->clevel[len] = 0;
3527 mddev->level = LEVEL_NONE;
3531 /* request to change the personality. Need to ensure:
3532 * - array is not engaged in resync/recovery/reshape
3533 * - old personality can be suspended
3534 * - new personality will access other array.
3537 if (mddev->sync_thread ||
3538 mddev->reshape_position != MaxSector ||
3539 mddev->sysfs_active)
3542 if (!mddev->pers->quiesce) {
3543 printk(KERN_WARNING "md: %s: %s does not support online personality change\n",
3544 mdname(mddev), mddev->pers->name);
3548 /* Now find the new personality */
3549 if (len == 0 || len >= sizeof(clevel))
3551 strncpy(clevel, buf, len);
3552 if (clevel[len-1] == '\n')
3555 if (strict_strtol(clevel, 10, &level))
3558 if (request_module("md-%s", clevel) != 0)
3559 request_module("md-level-%s", clevel);
3560 spin_lock(&pers_lock);
3561 pers = find_pers(level, clevel);
3562 if (!pers || !try_module_get(pers->owner)) {
3563 spin_unlock(&pers_lock);
3564 printk(KERN_WARNING "md: personality %s not loaded\n", clevel);
3567 spin_unlock(&pers_lock);
3569 if (pers == mddev->pers) {
3570 /* Nothing to do! */
3571 module_put(pers->owner);
3574 if (!pers->takeover) {
3575 module_put(pers->owner);
3576 printk(KERN_WARNING "md: %s: %s does not support personality takeover\n",
3577 mdname(mddev), clevel);
3581 rdev_for_each(rdev, mddev)
3582 rdev->new_raid_disk = rdev->raid_disk;
3584 /* ->takeover must set new_* and/or delta_disks
3585 * if it succeeds, and may set them when it fails.
3587 priv = pers->takeover(mddev);
3589 mddev->new_level = mddev->level;
3590 mddev->new_layout = mddev->layout;
3591 mddev->new_chunk_sectors = mddev->chunk_sectors;
3592 mddev->raid_disks -= mddev->delta_disks;
3593 mddev->delta_disks = 0;
3594 mddev->reshape_backwards = 0;
3595 module_put(pers->owner);
3596 printk(KERN_WARNING "md: %s: %s would not accept array\n",
3597 mdname(mddev), clevel);
3598 return PTR_ERR(priv);
3601 /* Looks like we have a winner */
3602 mddev_suspend(mddev);
3603 mddev->pers->stop(mddev);
3605 if (mddev->pers->sync_request == NULL &&
3606 pers->sync_request != NULL) {
3607 /* need to add the md_redundancy_group */
3608 if (sysfs_create_group(&mddev->kobj, &md_redundancy_group))
3610 "md: cannot register extra attributes for %s\n",
3612 mddev->sysfs_action = sysfs_get_dirent(mddev->kobj.sd, NULL, "sync_action");
3614 if (mddev->pers->sync_request != NULL &&
3615 pers->sync_request == NULL) {
3616 /* need to remove the md_redundancy_group */
3617 if (mddev->to_remove == NULL)
3618 mddev->to_remove = &md_redundancy_group;
3621 if (mddev->pers->sync_request == NULL &&
3623 /* We are converting from a no-redundancy array
3624 * to a redundancy array and metadata is managed
3625 * externally so we need to be sure that writes
3626 * won't block due to a need to transition
3628 * until external management is started.
3631 mddev->safemode_delay = 0;
3632 mddev->safemode = 0;
3635 rdev_for_each(rdev, mddev) {
3636 if (rdev->raid_disk < 0)
3638 if (rdev->new_raid_disk >= mddev->raid_disks)
3639 rdev->new_raid_disk = -1;
3640 if (rdev->new_raid_disk == rdev->raid_disk)
3642 sysfs_unlink_rdev(mddev, rdev);
3644 rdev_for_each(rdev, mddev) {
3645 if (rdev->raid_disk < 0)
3647 if (rdev->new_raid_disk == rdev->raid_disk)
3649 rdev->raid_disk = rdev->new_raid_disk;
3650 if (rdev->raid_disk < 0)
3651 clear_bit(In_sync, &rdev->flags);
3653 if (sysfs_link_rdev(mddev, rdev))
3654 printk(KERN_WARNING "md: cannot register rd%d"
3655 " for %s after level change\n",
3656 rdev->raid_disk, mdname(mddev));
3660 module_put(mddev->pers->owner);
3662 mddev->private = priv;
3663 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
3664 mddev->level = mddev->new_level;
3665 mddev->layout = mddev->new_layout;
3666 mddev->chunk_sectors = mddev->new_chunk_sectors;
3667 mddev->delta_disks = 0;
3668 mddev->reshape_backwards = 0;
3669 mddev->degraded = 0;
3670 if (mddev->pers->sync_request == NULL) {
3671 /* this is now an array without redundancy, so
3672 * it must always be in_sync
3675 del_timer_sync(&mddev->safemode_timer);
3678 set_bit(MD_CHANGE_DEVS, &mddev->flags);
3679 mddev_resume(mddev);
3680 sysfs_notify(&mddev->kobj, NULL, "level");
3681 md_new_event(mddev);
3685 static struct md_sysfs_entry md_level =
3686 __ATTR(level, S_IRUGO|S_IWUSR, level_show, level_store);
3690 layout_show(struct mddev *mddev, char *page)
3692 /* just a number, not meaningful for all levels */
3693 if (mddev->reshape_position != MaxSector &&
3694 mddev->layout != mddev->new_layout)
3695 return sprintf(page, "%d (%d)\n",
3696 mddev->new_layout, mddev->layout);
3697 return sprintf(page, "%d\n", mddev->layout);
3701 layout_store(struct mddev *mddev, const char *buf, size_t len)
3704 unsigned long n = simple_strtoul(buf, &e, 10);
3706 if (!*buf || (*e && *e != '\n'))
3711 if (mddev->pers->check_reshape == NULL)
3713 mddev->new_layout = n;
3714 err = mddev->pers->check_reshape(mddev);
3716 mddev->new_layout = mddev->layout;
3720 mddev->new_layout = n;
3721 if (mddev->reshape_position == MaxSector)
3726 static struct md_sysfs_entry md_layout =
3727 __ATTR(layout, S_IRUGO|S_IWUSR, layout_show, layout_store);
3731 raid_disks_show(struct mddev *mddev, char *page)
3733 if (mddev->raid_disks == 0)
3735 if (mddev->reshape_position != MaxSector &&
3736 mddev->delta_disks != 0)
3737 return sprintf(page, "%d (%d)\n", mddev->raid_disks,
3738 mddev->raid_disks - mddev->delta_disks);
3739 return sprintf(page, "%d\n", mddev->raid_disks);
3742 static int update_raid_disks(struct mddev *mddev, int raid_disks);
3745 raid_disks_store(struct mddev *mddev, const char *buf, size_t len)
3749 unsigned long n = simple_strtoul(buf, &e, 10);
3751 if (!*buf || (*e && *e != '\n'))
3755 rv = update_raid_disks(mddev, n);
3756 else if (mddev->reshape_position != MaxSector) {
3757 struct md_rdev *rdev;
3758 int olddisks = mddev->raid_disks - mddev->delta_disks;
3760 rdev_for_each(rdev, mddev) {
3762 rdev->data_offset < rdev->new_data_offset)
3765 rdev->data_offset > rdev->new_data_offset)
3768 mddev->delta_disks = n - olddisks;
3769 mddev->raid_disks = n;
3770 mddev->reshape_backwards = (mddev->delta_disks < 0);
3772 mddev->raid_disks = n;
3773 return rv ? rv : len;
3775 static struct md_sysfs_entry md_raid_disks =
3776 __ATTR(raid_disks, S_IRUGO|S_IWUSR, raid_disks_show, raid_disks_store);
3779 chunk_size_show(struct mddev *mddev, char *page)
3781 if (mddev->reshape_position != MaxSector &&
3782 mddev->chunk_sectors != mddev->new_chunk_sectors)
3783 return sprintf(page, "%d (%d)\n",
3784 mddev->new_chunk_sectors << 9,
3785 mddev->chunk_sectors << 9);
3786 return sprintf(page, "%d\n", mddev->chunk_sectors << 9);
3790 chunk_size_store(struct mddev *mddev, const char *buf, size_t len)
3793 unsigned long n = simple_strtoul(buf, &e, 10);
3795 if (!*buf || (*e && *e != '\n'))
3800 if (mddev->pers->check_reshape == NULL)
3802 mddev->new_chunk_sectors = n >> 9;
3803 err = mddev->pers->check_reshape(mddev);
3805 mddev->new_chunk_sectors = mddev->chunk_sectors;
3809 mddev->new_chunk_sectors = n >> 9;
3810 if (mddev->reshape_position == MaxSector)
3811 mddev->chunk_sectors = n >> 9;
3815 static struct md_sysfs_entry md_chunk_size =
3816 __ATTR(chunk_size, S_IRUGO|S_IWUSR, chunk_size_show, chunk_size_store);
3819 resync_start_show(struct mddev *mddev, char *page)
3821 if (mddev->recovery_cp == MaxSector)
3822 return sprintf(page, "none\n");
3823 return sprintf(page, "%llu\n", (unsigned long long)mddev->recovery_cp);
3827 resync_start_store(struct mddev *mddev, const char *buf, size_t len)
3830 unsigned long long n = simple_strtoull(buf, &e, 10);
3832 if (mddev->pers && !test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
3834 if (cmd_match(buf, "none"))
3836 else if (!*buf || (*e && *e != '\n'))
3839 mddev->recovery_cp = n;
3842 static struct md_sysfs_entry md_resync_start =
3843 __ATTR(resync_start, S_IRUGO|S_IWUSR, resync_start_show, resync_start_store);
3846 * The array state can be:
3849 * No devices, no size, no level
3850 * Equivalent to STOP_ARRAY ioctl
3852 * May have some settings, but array is not active
3853 * all IO results in error
3854 * When written, doesn't tear down array, but just stops it
3855 * suspended (not supported yet)
3856 * All IO requests will block. The array can be reconfigured.
3857 * Writing this, if accepted, will block until array is quiescent
3859 * no resync can happen. no superblocks get written.
3860 * write requests fail
3862 * like readonly, but behaves like 'clean' on a write request.
3864 * clean - no pending writes, but otherwise active.
3865 * When written to inactive array, starts without resync
3866 * If a write request arrives then
3867 * if metadata is known, mark 'dirty' and switch to 'active'.
3868 * if not known, block and switch to write-pending
3869 * If written to an active array that has pending writes, then fails.
3871 * fully active: IO and resync can be happening.
3872 * When written to inactive array, starts with resync
3875 * clean, but writes are blocked waiting for 'active' to be written.
3878 * like active, but no writes have been seen for a while (100msec).
3881 enum array_state { clear, inactive, suspended, readonly, read_auto, clean, active,
3882 write_pending, active_idle, bad_word};
3883 static char *array_states[] = {
3884 "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active",
3885 "write-pending", "active-idle", NULL };
3887 static int match_word(const char *word, char **list)
3890 for (n=0; list[n]; n++)
3891 if (cmd_match(word, list[n]))
3897 array_state_show(struct mddev *mddev, char *page)
3899 enum array_state st = inactive;
3912 else if (test_bit(MD_CHANGE_PENDING, &mddev->flags))
3914 else if (mddev->safemode)
3920 if (list_empty(&mddev->disks) &&
3921 mddev->raid_disks == 0 &&
3922 mddev->dev_sectors == 0)
3927 return sprintf(page, "%s\n", array_states[st]);
3930 static int do_md_stop(struct mddev * mddev, int ro, struct block_device *bdev);
3931 static int md_set_readonly(struct mddev * mddev, struct block_device *bdev);
3932 static int do_md_run(struct mddev * mddev);
3933 static int restart_array(struct mddev *mddev);
3936 array_state_store(struct mddev *mddev, const char *buf, size_t len)
3939 enum array_state st = match_word(buf, array_states);
3944 /* stopping an active array */
3945 if (atomic_read(&mddev->openers) > 0)
3947 err = do_md_stop(mddev, 0, NULL);
3950 /* stopping an active array */
3952 if (atomic_read(&mddev->openers) > 0)
3954 err = do_md_stop(mddev, 2, NULL);
3956 err = 0; /* already inactive */
3959 break; /* not supported yet */
3962 err = md_set_readonly(mddev, NULL);
3965 set_disk_ro(mddev->gendisk, 1);
3966 err = do_md_run(mddev);
3972 err = md_set_readonly(mddev, NULL);
3973 else if (mddev->ro == 1)
3974 err = restart_array(mddev);
3977 set_disk_ro(mddev->gendisk, 0);
3981 err = do_md_run(mddev);
3986 restart_array(mddev);
3987 spin_lock_irq(&mddev->write_lock);
3988 if (atomic_read(&mddev->writes_pending) == 0) {
3989 if (mddev->in_sync == 0) {
3991 if (mddev->safemode == 1)
3992 mddev->safemode = 0;
3993 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
3998 spin_unlock_irq(&mddev->write_lock);
4004 restart_array(mddev);
4005 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
4006 wake_up(&mddev->sb_wait);
4010 set_disk_ro(mddev->gendisk, 0);
4011 err = do_md_run(mddev);
4016 /* these cannot be set */
4022 if (mddev->hold_active == UNTIL_IOCTL)
4023 mddev->hold_active = 0;
4024 sysfs_notify_dirent_safe(mddev->sysfs_state);
4028 static struct md_sysfs_entry md_array_state =
4029 __ATTR(array_state, S_IRUGO|S_IWUSR, array_state_show, array_state_store);
4032 max_corrected_read_errors_show(struct mddev *mddev, char *page) {
4033 return sprintf(page, "%d\n",
4034 atomic_read(&mddev->max_corr_read_errors));
4038 max_corrected_read_errors_store(struct mddev *mddev, const char *buf, size_t len)
4041 unsigned long n = simple_strtoul(buf, &e, 10);
4043 if (*buf && (*e == 0 || *e == '\n')) {
4044 atomic_set(&mddev->max_corr_read_errors, n);
4050 static struct md_sysfs_entry max_corr_read_errors =
4051 __ATTR(max_read_errors, S_IRUGO|S_IWUSR, max_corrected_read_errors_show,
4052 max_corrected_read_errors_store);
4055 null_show(struct mddev *mddev, char *page)
4061 new_dev_store(struct mddev *mddev, const char *buf, size_t len)
4063 /* buf must be %d:%d\n? giving major and minor numbers */
4064 /* The new device is added to the array.
4065 * If the array has a persistent superblock, we read the
4066 * superblock to initialise info and check validity.
4067 * Otherwise, only checking done is that in bind_rdev_to_array,
4068 * which mainly checks size.
4071 int major = simple_strtoul(buf, &e, 10);
4074 struct md_rdev *rdev;
4077 if (!*buf || *e != ':' || !e[1] || e[1] == '\n')
4079 minor = simple_strtoul(e+1, &e, 10);
4080 if (*e && *e != '\n')
4082 dev = MKDEV(major, minor);
4083 if (major != MAJOR(dev) ||
4084 minor != MINOR(dev))
4088 if (mddev->persistent) {
4089 rdev = md_import_device(dev, mddev->major_version,
4090 mddev->minor_version);
4091 if (!IS_ERR(rdev) && !list_empty(&mddev->disks)) {
4092 struct md_rdev *rdev0
4093 = list_entry(mddev->disks.next,
4094 struct md_rdev, same_set);
4095 err = super_types[mddev->major_version]
4096 .load_super(rdev, rdev0, mddev->minor_version);
4100 } else if (mddev->external)
4101 rdev = md_import_device(dev, -2, -1);
4103 rdev = md_import_device(dev, -1, -1);
4106 return PTR_ERR(rdev);
4107 err = bind_rdev_to_array(rdev, mddev);
4111 return err ? err : len;
4114 static struct md_sysfs_entry md_new_device =
4115 __ATTR(new_dev, S_IWUSR, null_show, new_dev_store);
4118 bitmap_store(struct mddev *mddev, const char *buf, size_t len)
4121 unsigned long chunk, end_chunk;
4125 /* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */
4127 chunk = end_chunk = simple_strtoul(buf, &end, 0);
4128 if (buf == end) break;
4129 if (*end == '-') { /* range */
4131 end_chunk = simple_strtoul(buf, &end, 0);
4132 if (buf == end) break;
4134 if (*end && !isspace(*end)) break;
4135 bitmap_dirty_bits(mddev->bitmap, chunk, end_chunk);
4136 buf = skip_spaces(end);
4138 bitmap_unplug(mddev->bitmap); /* flush the bits to disk */
4143 static struct md_sysfs_entry md_bitmap =
4144 __ATTR(bitmap_set_bits, S_IWUSR, null_show, bitmap_store);
4147 size_show(struct mddev *mddev, char *page)
4149 return sprintf(page, "%llu\n",
4150 (unsigned long long)mddev->dev_sectors / 2);
4153 static int update_size(struct mddev *mddev, sector_t num_sectors);
4156 size_store(struct mddev *mddev, const char *buf, size_t len)
4158 /* If array is inactive, we can reduce the component size, but
4159 * not increase it (except from 0).
4160 * If array is active, we can try an on-line resize
4163 int err = strict_blocks_to_sectors(buf, §ors);
4168 err = update_size(mddev, sectors);
4169 md_update_sb(mddev, 1);
4171 if (mddev->dev_sectors == 0 ||
4172 mddev->dev_sectors > sectors)
4173 mddev->dev_sectors = sectors;
4177 return err ? err : len;
4180 static struct md_sysfs_entry md_size =
4181 __ATTR(component_size, S_IRUGO|S_IWUSR, size_show, size_store);
4186 * 'none' for arrays with no metadata (good luck...)
4187 * 'external' for arrays with externally managed metadata,
4188 * or N.M for internally known formats
4191 metadata_show(struct mddev *mddev, char *page)
4193 if (mddev->persistent)
4194 return sprintf(page, "%d.%d\n",
4195 mddev->major_version, mddev->minor_version);
4196 else if (mddev->external)
4197 return sprintf(page, "external:%s\n", mddev->metadata_type);
4199 return sprintf(page, "none\n");
4203 metadata_store(struct mddev *mddev, const char *buf, size_t len)
4207 /* Changing the details of 'external' metadata is
4208 * always permitted. Otherwise there must be
4209 * no devices attached to the array.
4211 if (mddev->external && strncmp(buf, "external:", 9) == 0)
4213 else if (!list_empty(&mddev->disks))
4216 if (cmd_match(buf, "none")) {
4217 mddev->persistent = 0;
4218 mddev->external = 0;
4219 mddev->major_version = 0;
4220 mddev->minor_version = 90;
4223 if (strncmp(buf, "external:", 9) == 0) {
4224 size_t namelen = len-9;
4225 if (namelen >= sizeof(mddev->metadata_type))
4226 namelen = sizeof(mddev->metadata_type)-1;
4227 strncpy(mddev->metadata_type, buf+9, namelen);
4228 mddev->metadata_type[namelen] = 0;
4229 if (namelen && mddev->metadata_type[namelen-1] == '\n')
4230 mddev->metadata_type[--namelen] = 0;
4231 mddev->persistent = 0;
4232 mddev->external = 1;
4233 mddev->major_version = 0;
4234 mddev->minor_version = 90;
4237 major = simple_strtoul(buf, &e, 10);
4238 if (e==buf || *e != '.')
4241 minor = simple_strtoul(buf, &e, 10);
4242 if (e==buf || (*e && *e != '\n') )
4244 if (major >= ARRAY_SIZE(super_types) || super_types[major].name == NULL)
4246 mddev->major_version = major;
4247 mddev->minor_version = minor;
4248 mddev->persistent = 1;
4249 mddev->external = 0;
4253 static struct md_sysfs_entry md_metadata =
4254 __ATTR(metadata_version, S_IRUGO|S_IWUSR, metadata_show, metadata_store);
4257 action_show(struct mddev *mddev, char *page)
4259 char *type = "idle";
4260 if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
4262 else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
4263 (!mddev->ro && test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))) {
4264 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
4266 else if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
4267 if (!test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
4269 else if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
4273 } else if (test_bit(MD_RECOVERY_RECOVER, &mddev->recovery))
4276 return sprintf(page, "%s\n", type);
4279 static void reap_sync_thread(struct mddev *mddev);
4282 action_store(struct mddev *mddev, const char *page, size_t len)
4284 if (!mddev->pers || !mddev->pers->sync_request)
4287 if (cmd_match(page, "frozen"))
4288 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4290 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4292 if (cmd_match(page, "idle") || cmd_match(page, "frozen")) {
4293 if (mddev->sync_thread) {
4294 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
4295 reap_sync_thread(mddev);
4297 } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
4298 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
4300 else if (cmd_match(page, "resync"))
4301 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4302 else if (cmd_match(page, "recover")) {
4303 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
4304 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4305 } else if (cmd_match(page, "reshape")) {
4307 if (mddev->pers->start_reshape == NULL)
4309 err = mddev->pers->start_reshape(mddev);
4312 sysfs_notify(&mddev->kobj, NULL, "degraded");
4314 if (cmd_match(page, "check"))
4315 set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
4316 else if (!cmd_match(page, "repair"))
4318 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
4319 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
4321 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4322 md_wakeup_thread(mddev->thread);
4323 sysfs_notify_dirent_safe(mddev->sysfs_action);
4328 mismatch_cnt_show(struct mddev *mddev, char *page)
4330 return sprintf(page, "%llu\n",
4331 (unsigned long long) mddev->resync_mismatches);
4334 static struct md_sysfs_entry md_scan_mode =
4335 __ATTR(sync_action, S_IRUGO|S_IWUSR, action_show, action_store);
4338 static struct md_sysfs_entry md_mismatches = __ATTR_RO(mismatch_cnt);
4341 sync_min_show(struct mddev *mddev, char *page)
4343 return sprintf(page, "%d (%s)\n", speed_min(mddev),
4344 mddev->sync_speed_min ? "local": "system");
4348 sync_min_store(struct mddev *mddev, const char *buf, size_t len)
4352 if (strncmp(buf, "system", 6)==0) {
4353 mddev->sync_speed_min = 0;
4356 min = simple_strtoul(buf, &e, 10);
4357 if (buf == e || (*e && *e != '\n') || min <= 0)
4359 mddev->sync_speed_min = min;
4363 static struct md_sysfs_entry md_sync_min =
4364 __ATTR(sync_speed_min, S_IRUGO|S_IWUSR, sync_min_show, sync_min_store);
4367 sync_max_show(struct mddev *mddev, char *page)
4369 return sprintf(page, "%d (%s)\n", speed_max(mddev),
4370 mddev->sync_speed_max ? "local": "system");
4374 sync_max_store(struct mddev *mddev, const char *buf, size_t len)
4378 if (strncmp(buf, "system", 6)==0) {
4379 mddev->sync_speed_max = 0;
4382 max = simple_strtoul(buf, &e, 10);
4383 if (buf == e || (*e && *e != '\n') || max <= 0)
4385 mddev->sync_speed_max = max;
4389 static struct md_sysfs_entry md_sync_max =
4390 __ATTR(sync_speed_max, S_IRUGO|S_IWUSR, sync_max_show, sync_max_store);
4393 degraded_show(struct mddev *mddev, char *page)
4395 return sprintf(page, "%d\n", mddev->degraded);
4397 static struct md_sysfs_entry md_degraded = __ATTR_RO(degraded);
4400 sync_force_parallel_show(struct mddev *mddev, char *page)
4402 return sprintf(page, "%d\n", mddev->parallel_resync);
4406 sync_force_parallel_store(struct mddev *mddev, const char *buf, size_t len)
4410 if (strict_strtol(buf, 10, &n))
4413 if (n != 0 && n != 1)
4416 mddev->parallel_resync = n;
4418 if (mddev->sync_thread)
4419 wake_up(&resync_wait);
4424 /* force parallel resync, even with shared block devices */
4425 static struct md_sysfs_entry md_sync_force_parallel =
4426 __ATTR(sync_force_parallel, S_IRUGO|S_IWUSR,
4427 sync_force_parallel_show, sync_force_parallel_store);
4430 sync_speed_show(struct mddev *mddev, char *page)
4432 unsigned long resync, dt, db;
4433 if (mddev->curr_resync == 0)
4434 return sprintf(page, "none\n");
4435 resync = mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active);
4436 dt = (jiffies - mddev->resync_mark) / HZ;
4438 db = resync - mddev->resync_mark_cnt;
4439 return sprintf(page, "%lu\n", db/dt/2); /* K/sec */
4442 static struct md_sysfs_entry md_sync_speed = __ATTR_RO(sync_speed);
4445 sync_completed_show(struct mddev *mddev, char *page)
4447 unsigned long long max_sectors, resync;
4449 if (!test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4450 return sprintf(page, "none\n");
4452 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ||
4453 test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
4454 max_sectors = mddev->resync_max_sectors;
4456 max_sectors = mddev->dev_sectors;
4458 resync = mddev->curr_resync_completed;
4459 return sprintf(page, "%llu / %llu\n", resync, max_sectors);
4462 static struct md_sysfs_entry md_sync_completed = __ATTR_RO(sync_completed);
4465 min_sync_show(struct mddev *mddev, char *page)
4467 return sprintf(page, "%llu\n",
4468 (unsigned long long)mddev->resync_min);
4471 min_sync_store(struct mddev *mddev, const char *buf, size_t len)
4473 unsigned long long min;
4474 if (strict_strtoull(buf, 10, &min))
4476 if (min > mddev->resync_max)
4478 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4481 /* Must be a multiple of chunk_size */
4482 if (mddev->chunk_sectors) {
4483 sector_t temp = min;
4484 if (sector_div(temp, mddev->chunk_sectors))
4487 mddev->resync_min = min;
4492 static struct md_sysfs_entry md_min_sync =
4493 __ATTR(sync_min, S_IRUGO|S_IWUSR, min_sync_show, min_sync_store);
4496 max_sync_show(struct mddev *mddev, char *page)
4498 if (mddev->resync_max == MaxSector)
4499 return sprintf(page, "max\n");
4501 return sprintf(page, "%llu\n",
4502 (unsigned long long)mddev->resync_max);
4505 max_sync_store(struct mddev *mddev, const char *buf, size_t len)
4507 if (strncmp(buf, "max", 3) == 0)
4508 mddev->resync_max = MaxSector;
4510 unsigned long long max;
4511 if (strict_strtoull(buf, 10, &max))
4513 if (max < mddev->resync_min)
4515 if (max < mddev->resync_max &&
4517 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4520 /* Must be a multiple of chunk_size */
4521 if (mddev->chunk_sectors) {
4522 sector_t temp = max;
4523 if (sector_div(temp, mddev->chunk_sectors))
4526 mddev->resync_max = max;
4528 wake_up(&mddev->recovery_wait);
4532 static struct md_sysfs_entry md_max_sync =
4533 __ATTR(sync_max, S_IRUGO|S_IWUSR, max_sync_show, max_sync_store);
4536 suspend_lo_show(struct mddev *mddev, char *page)
4538 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_lo);
4542 suspend_lo_store(struct mddev *mddev, const char *buf, size_t len)
4545 unsigned long long new = simple_strtoull(buf, &e, 10);
4546 unsigned long long old = mddev->suspend_lo;
4548 if (mddev->pers == NULL ||
4549 mddev->pers->quiesce == NULL)
4551 if (buf == e || (*e && *e != '\n'))
4554 mddev->suspend_lo = new;
4556 /* Shrinking suspended region */
4557 mddev->pers->quiesce(mddev, 2);
4559 /* Expanding suspended region - need to wait */
4560 mddev->pers->quiesce(mddev, 1);
4561 mddev->pers->quiesce(mddev, 0);
4565 static struct md_sysfs_entry md_suspend_lo =
4566 __ATTR(suspend_lo, S_IRUGO|S_IWUSR, suspend_lo_show, suspend_lo_store);
4570 suspend_hi_show(struct mddev *mddev, char *page)
4572 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_hi);
4576 suspend_hi_store(struct mddev *mddev, const char *buf, size_t len)
4579 unsigned long long new = simple_strtoull(buf, &e, 10);
4580 unsigned long long old = mddev->suspend_hi;
4582 if (mddev->pers == NULL ||
4583 mddev->pers->quiesce == NULL)
4585 if (buf == e || (*e && *e != '\n'))
4588 mddev->suspend_hi = new;
4590 /* Shrinking suspended region */
4591 mddev->pers->quiesce(mddev, 2);
4593 /* Expanding suspended region - need to wait */
4594 mddev->pers->quiesce(mddev, 1);
4595 mddev->pers->quiesce(mddev, 0);
4599 static struct md_sysfs_entry md_suspend_hi =
4600 __ATTR(suspend_hi, S_IRUGO|S_IWUSR, suspend_hi_show, suspend_hi_store);
4603 reshape_position_show(struct mddev *mddev, char *page)
4605 if (mddev->reshape_position != MaxSector)
4606 return sprintf(page, "%llu\n",
4607 (unsigned long long)mddev->reshape_position);
4608 strcpy(page, "none\n");
4613 reshape_position_store(struct mddev *mddev, const char *buf, size_t len)
4615 struct md_rdev *rdev;
4617 unsigned long long new = simple_strtoull(buf, &e, 10);
4620 if (buf == e || (*e && *e != '\n'))
4622 mddev->reshape_position = new;
4623 mddev->delta_disks = 0;
4624 mddev->reshape_backwards = 0;
4625 mddev->new_level = mddev->level;
4626 mddev->new_layout = mddev->layout;
4627 mddev->new_chunk_sectors = mddev->chunk_sectors;
4628 rdev_for_each(rdev, mddev)
4629 rdev->new_data_offset = rdev->data_offset;
4633 static struct md_sysfs_entry md_reshape_position =
4634 __ATTR(reshape_position, S_IRUGO|S_IWUSR, reshape_position_show,
4635 reshape_position_store);
4638 reshape_direction_show(struct mddev *mddev, char *page)
4640 return sprintf(page, "%s\n",
4641 mddev->reshape_backwards ? "backwards" : "forwards");
4645 reshape_direction_store(struct mddev *mddev, const char *buf, size_t len)
4648 if (cmd_match(buf, "forwards"))
4650 else if (cmd_match(buf, "backwards"))
4654 if (mddev->reshape_backwards == backwards)
4657 /* check if we are allowed to change */
4658 if (mddev->delta_disks)
4661 if (mddev->persistent &&
4662 mddev->major_version == 0)
4665 mddev->reshape_backwards = backwards;
4669 static struct md_sysfs_entry md_reshape_direction =
4670 __ATTR(reshape_direction, S_IRUGO|S_IWUSR, reshape_direction_show,
4671 reshape_direction_store);
4674 array_size_show(struct mddev *mddev, char *page)
4676 if (mddev->external_size)
4677 return sprintf(page, "%llu\n",
4678 (unsigned long long)mddev->array_sectors/2);
4680 return sprintf(page, "default\n");
4684 array_size_store(struct mddev *mddev, const char *buf, size_t len)
4688 if (strncmp(buf, "default", 7) == 0) {
4690 sectors = mddev->pers->size(mddev, 0, 0);
4692 sectors = mddev->array_sectors;
4694 mddev->external_size = 0;
4696 if (strict_blocks_to_sectors(buf, §ors) < 0)
4698 if (mddev->pers && mddev->pers->size(mddev, 0, 0) < sectors)
4701 mddev->external_size = 1;
4704 mddev->array_sectors = sectors;
4706 set_capacity(mddev->gendisk, mddev->array_sectors);
4707 revalidate_disk(mddev->gendisk);
4712 static struct md_sysfs_entry md_array_size =
4713 __ATTR(array_size, S_IRUGO|S_IWUSR, array_size_show,
4716 static struct attribute *md_default_attrs[] = {
4719 &md_raid_disks.attr,
4720 &md_chunk_size.attr,
4722 &md_resync_start.attr,
4724 &md_new_device.attr,
4725 &md_safe_delay.attr,
4726 &md_array_state.attr,
4727 &md_reshape_position.attr,
4728 &md_reshape_direction.attr,
4729 &md_array_size.attr,
4730 &max_corr_read_errors.attr,
4734 static struct attribute *md_redundancy_attrs[] = {
4736 &md_mismatches.attr,
4739 &md_sync_speed.attr,
4740 &md_sync_force_parallel.attr,
4741 &md_sync_completed.attr,
4744 &md_suspend_lo.attr,
4745 &md_suspend_hi.attr,
4750 static struct attribute_group md_redundancy_group = {
4752 .attrs = md_redundancy_attrs,
4757 md_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
4759 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
4760 struct mddev *mddev = container_of(kobj, struct mddev, kobj);
4765 spin_lock(&all_mddevs_lock);
4766 if (list_empty(&mddev->all_mddevs)) {
4767 spin_unlock(&all_mddevs_lock);
4771 spin_unlock(&all_mddevs_lock);
4773 rv = mddev_lock(mddev);
4775 rv = entry->show(mddev, page);
4776 mddev_unlock(mddev);
4783 md_attr_store(struct kobject *kobj, struct attribute *attr,
4784 const char *page, size_t length)
4786 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
4787 struct mddev *mddev = container_of(kobj, struct mddev, kobj);
4792 if (!capable(CAP_SYS_ADMIN))
4794 spin_lock(&all_mddevs_lock);
4795 if (list_empty(&mddev->all_mddevs)) {
4796 spin_unlock(&all_mddevs_lock);
4800 spin_unlock(&all_mddevs_lock);
4801 rv = mddev_lock(mddev);
4803 rv = entry->store(mddev, page, length);
4804 mddev_unlock(mddev);
4810 static void md_free(struct kobject *ko)
4812 struct mddev *mddev = container_of(ko, struct mddev, kobj);
4814 if (mddev->sysfs_state)
4815 sysfs_put(mddev->sysfs_state);
4817 if (mddev->gendisk) {
4818 del_gendisk(mddev->gendisk);
4819 put_disk(mddev->gendisk);
4822 blk_cleanup_queue(mddev->queue);
4827 static const struct sysfs_ops md_sysfs_ops = {
4828 .show = md_attr_show,
4829 .store = md_attr_store,
4831 static struct kobj_type md_ktype = {
4833 .sysfs_ops = &md_sysfs_ops,
4834 .default_attrs = md_default_attrs,
4839 static void mddev_delayed_delete(struct work_struct *ws)
4841 struct mddev *mddev = container_of(ws, struct mddev, del_work);
4843 sysfs_remove_group(&mddev->kobj, &md_bitmap_group);
4844 kobject_del(&mddev->kobj);
4845 kobject_put(&mddev->kobj);
4848 static int md_alloc(dev_t dev, char *name)
4850 static DEFINE_MUTEX(disks_mutex);
4851 struct mddev *mddev = mddev_find(dev);
4852 struct gendisk *disk;
4861 partitioned = (MAJOR(mddev->unit) != MD_MAJOR);
4862 shift = partitioned ? MdpMinorShift : 0;
4863 unit = MINOR(mddev->unit) >> shift;
4865 /* wait for any previous instance of this device to be
4866 * completely removed (mddev_delayed_delete).
4868 flush_workqueue(md_misc_wq);
4870 mutex_lock(&disks_mutex);
4876 /* Need to ensure that 'name' is not a duplicate.
4878 struct mddev *mddev2;
4879 spin_lock(&all_mddevs_lock);
4881 list_for_each_entry(mddev2, &all_mddevs, all_mddevs)
4882 if (mddev2->gendisk &&
4883 strcmp(mddev2->gendisk->disk_name, name) == 0) {
4884 spin_unlock(&all_mddevs_lock);
4887 spin_unlock(&all_mddevs_lock);
4891 mddev->queue = blk_alloc_queue(GFP_KERNEL);
4894 mddev->queue->queuedata = mddev;
4896 blk_queue_make_request(mddev->queue, md_make_request);
4897 blk_set_stacking_limits(&mddev->queue->limits);
4899 disk = alloc_disk(1 << shift);
4901 blk_cleanup_queue(mddev->queue);
4902 mddev->queue = NULL;
4905 disk->major = MAJOR(mddev->unit);
4906 disk->first_minor = unit << shift;
4908 strcpy(disk->disk_name, name);
4909 else if (partitioned)
4910 sprintf(disk->disk_name, "md_d%d", unit);
4912 sprintf(disk->disk_name, "md%d", unit);
4913 disk->fops = &md_fops;
4914 disk->private_data = mddev;
4915 disk->queue = mddev->queue;
4916 blk_queue_flush(mddev->queue, REQ_FLUSH | REQ_FUA);
4917 /* Allow extended partitions. This makes the
4918 * 'mdp' device redundant, but we can't really
4921 disk->flags |= GENHD_FL_EXT_DEVT;
4922 mddev->gendisk = disk;
4923 /* As soon as we call add_disk(), another thread could get
4924 * through to md_open, so make sure it doesn't get too far
4926 mutex_lock(&mddev->open_mutex);
4929 error = kobject_init_and_add(&mddev->kobj, &md_ktype,
4930 &disk_to_dev(disk)->kobj, "%s", "md");
4932 /* This isn't possible, but as kobject_init_and_add is marked
4933 * __must_check, we must do something with the result
4935 printk(KERN_WARNING "md: cannot register %s/md - name in use\n",
4939 if (mddev->kobj.sd &&
4940 sysfs_create_group(&mddev->kobj, &md_bitmap_group))
4941 printk(KERN_DEBUG "pointless warning\n");
4942 mutex_unlock(&mddev->open_mutex);
4944 mutex_unlock(&disks_mutex);
4945 if (!error && mddev->kobj.sd) {
4946 kobject_uevent(&mddev->kobj, KOBJ_ADD);
4947 mddev->sysfs_state = sysfs_get_dirent_safe(mddev->kobj.sd, "array_state");
4953 static struct kobject *md_probe(dev_t dev, int *part, void *data)
4955 md_alloc(dev, NULL);
4959 static int add_named_array(const char *val, struct kernel_param *kp)
4961 /* val must be "md_*" where * is not all digits.
4962 * We allocate an array with a large free minor number, and
4963 * set the name to val. val must not already be an active name.
4965 int len = strlen(val);
4966 char buf[DISK_NAME_LEN];
4968 while (len && val[len-1] == '\n')
4970 if (len >= DISK_NAME_LEN)
4972 strlcpy(buf, val, len+1);
4973 if (strncmp(buf, "md_", 3) != 0)
4975 return md_alloc(0, buf);
4978 static void md_safemode_timeout(unsigned long data)
4980 struct mddev *mddev = (struct mddev *) data;
4982 if (!atomic_read(&mddev->writes_pending)) {
4983 mddev->safemode = 1;
4984 if (mddev->external)
4985 sysfs_notify_dirent_safe(mddev->sysfs_state);
4987 md_wakeup_thread(mddev->thread);
4990 static int start_dirty_degraded;
4992 int md_run(struct mddev *mddev)
4995 struct md_rdev *rdev;
4996 struct md_personality *pers;
4998 if (list_empty(&mddev->disks))
4999 /* cannot run an array with no devices.. */
5004 /* Cannot run until previous stop completes properly */
5005 if (mddev->sysfs_active)
5009 * Analyze all RAID superblock(s)
5011 if (!mddev->raid_disks) {
5012 if (!mddev->persistent)
5017 if (mddev->level != LEVEL_NONE)
5018 request_module("md-level-%d", mddev->level);
5019 else if (mddev->clevel[0])
5020 request_module("md-%s", mddev->clevel);
5023 * Drop all container device buffers, from now on
5024 * the only valid external interface is through the md
5027 rdev_for_each(rdev, mddev) {
5028 if (test_bit(Faulty, &rdev->flags))
5030 sync_blockdev(rdev->bdev);
5031 invalidate_bdev(rdev->bdev);
5033 /* perform some consistency tests on the device.
5034 * We don't want the data to overlap the metadata,
5035 * Internal Bitmap issues have been handled elsewhere.
5037 if (rdev->meta_bdev) {
5038 /* Nothing to check */;
5039 } else if (rdev->data_offset < rdev->sb_start) {
5040 if (mddev->dev_sectors &&
5041 rdev->data_offset + mddev->dev_sectors
5043 printk("md: %s: data overlaps metadata\n",
5048 if (rdev->sb_start + rdev->sb_size/512
5049 > rdev->data_offset) {
5050 printk("md: %s: metadata overlaps data\n",
5055 sysfs_notify_dirent_safe(rdev->sysfs_state);
5058 if (mddev->bio_set == NULL)
5059 mddev->bio_set = bioset_create(BIO_POOL_SIZE,
5060 sizeof(struct mddev *));
5062 spin_lock(&pers_lock);
5063 pers = find_pers(mddev->level, mddev->clevel);
5064 if (!pers || !try_module_get(pers->owner)) {
5065 spin_unlock(&pers_lock);
5066 if (mddev->level != LEVEL_NONE)
5067 printk(KERN_WARNING "md: personality for level %d is not loaded!\n",
5070 printk(KERN_WARNING "md: personality for level %s is not loaded!\n",
5075 spin_unlock(&pers_lock);
5076 if (mddev->level != pers->level) {
5077 mddev->level = pers->level;
5078 mddev->new_level = pers->level;
5080 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
5082 if (mddev->reshape_position != MaxSector &&
5083 pers->start_reshape == NULL) {
5084 /* This personality cannot handle reshaping... */
5086 module_put(pers->owner);
5090 if (pers->sync_request) {
5091 /* Warn if this is a potentially silly
5094 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
5095 struct md_rdev *rdev2;
5098 rdev_for_each(rdev, mddev)
5099 rdev_for_each(rdev2, mddev) {
5101 rdev->bdev->bd_contains ==
5102 rdev2->bdev->bd_contains) {
5104 "%s: WARNING: %s appears to be"
5105 " on the same physical disk as"
5108 bdevname(rdev->bdev,b),
5109 bdevname(rdev2->bdev,b2));
5116 "True protection against single-disk"
5117 " failure might be compromised.\n");
5120 mddev->recovery = 0;
5121 /* may be over-ridden by personality */
5122 mddev->resync_max_sectors = mddev->dev_sectors;
5124 mddev->ok_start_degraded = start_dirty_degraded;
5126 if (start_readonly && mddev->ro == 0)
5127 mddev->ro = 2; /* read-only, but switch on first write */
5129 err = mddev->pers->run(mddev);
5131 printk(KERN_ERR "md: pers->run() failed ...\n");
5132 else if (mddev->pers->size(mddev, 0, 0) < mddev->array_sectors) {
5133 WARN_ONCE(!mddev->external_size, "%s: default size too small,"
5134 " but 'external_size' not in effect?\n", __func__);
5136 "md: invalid array_size %llu > default size %llu\n",
5137 (unsigned long long)mddev->array_sectors / 2,
5138 (unsigned long long)mddev->pers->size(mddev, 0, 0) / 2);
5140 mddev->pers->stop(mddev);
5142 if (err == 0 && mddev->pers->sync_request &&
5143 (mddev->bitmap_info.file || mddev->bitmap_info.offset)) {
5144 err = bitmap_create(mddev);
5146 printk(KERN_ERR "%s: failed to create bitmap (%d)\n",
5147 mdname(mddev), err);
5148 mddev->pers->stop(mddev);
5152 module_put(mddev->pers->owner);
5154 bitmap_destroy(mddev);
5157 if (mddev->pers->sync_request) {
5158 if (mddev->kobj.sd &&
5159 sysfs_create_group(&mddev->kobj, &md_redundancy_group))
5161 "md: cannot register extra attributes for %s\n",
5163 mddev->sysfs_action = sysfs_get_dirent_safe(mddev->kobj.sd, "sync_action");
5164 } else if (mddev->ro == 2) /* auto-readonly not meaningful */
5167 atomic_set(&mddev->writes_pending,0);
5168 atomic_set(&mddev->max_corr_read_errors,
5169 MD_DEFAULT_MAX_CORRECTED_READ_ERRORS);
5170 mddev->safemode = 0;
5171 mddev->safemode_timer.function = md_safemode_timeout;
5172 mddev->safemode_timer.data = (unsigned long) mddev;
5173 mddev->safemode_delay = (200 * HZ)/1000 +1; /* 200 msec delay */
5177 rdev_for_each(rdev, mddev)
5178 if (rdev->raid_disk >= 0)
5179 if (sysfs_link_rdev(mddev, rdev))
5180 /* failure here is OK */;
5182 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5185 md_update_sb(mddev, 0);
5187 md_new_event(mddev);
5188 sysfs_notify_dirent_safe(mddev->sysfs_state);
5189 sysfs_notify_dirent_safe(mddev->sysfs_action);
5190 sysfs_notify(&mddev->kobj, NULL, "degraded");
5193 EXPORT_SYMBOL_GPL(md_run);
5195 static int do_md_run(struct mddev *mddev)
5199 err = md_run(mddev);
5202 err = bitmap_load(mddev);
5204 bitmap_destroy(mddev);
5208 md_wakeup_thread(mddev->thread);
5209 md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
5211 set_capacity(mddev->gendisk, mddev->array_sectors);
5212 revalidate_disk(mddev->gendisk);
5214 kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
5219 static int restart_array(struct mddev *mddev)
5221 struct gendisk *disk = mddev->gendisk;
5223 /* Complain if it has no devices */
5224 if (list_empty(&mddev->disks))
5230 mddev->safemode = 0;
5232 set_disk_ro(disk, 0);
5233 printk(KERN_INFO "md: %s switched to read-write mode.\n",
5235 /* Kick recovery or resync if necessary */
5236 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5237 md_wakeup_thread(mddev->thread);
5238 md_wakeup_thread(mddev->sync_thread);
5239 sysfs_notify_dirent_safe(mddev->sysfs_state);
5243 /* similar to deny_write_access, but accounts for our holding a reference
5244 * to the file ourselves */
5245 static int deny_bitmap_write_access(struct file * file)
5247 struct inode *inode = file->f_mapping->host;
5249 spin_lock(&inode->i_lock);
5250 if (atomic_read(&inode->i_writecount) > 1) {
5251 spin_unlock(&inode->i_lock);
5254 atomic_set(&inode->i_writecount, -1);
5255 spin_unlock(&inode->i_lock);
5260 void restore_bitmap_write_access(struct file *file)
5262 struct inode *inode = file->f_mapping->host;
5264 spin_lock(&inode->i_lock);
5265 atomic_set(&inode->i_writecount, 1);
5266 spin_unlock(&inode->i_lock);
5269 static void md_clean(struct mddev *mddev)
5271 mddev->array_sectors = 0;
5272 mddev->external_size = 0;
5273 mddev->dev_sectors = 0;
5274 mddev->raid_disks = 0;
5275 mddev->recovery_cp = 0;
5276 mddev->resync_min = 0;
5277 mddev->resync_max = MaxSector;
5278 mddev->reshape_position = MaxSector;
5279 mddev->external = 0;
5280 mddev->persistent = 0;
5281 mddev->level = LEVEL_NONE;
5282 mddev->clevel[0] = 0;
5285 mddev->metadata_type[0] = 0;
5286 mddev->chunk_sectors = 0;
5287 mddev->ctime = mddev->utime = 0;
5289 mddev->max_disks = 0;
5291 mddev->can_decrease_events = 0;
5292 mddev->delta_disks = 0;
5293 mddev->reshape_backwards = 0;
5294 mddev->new_level = LEVEL_NONE;
5295 mddev->new_layout = 0;
5296 mddev->new_chunk_sectors = 0;
5297 mddev->curr_resync = 0;
5298 mddev->resync_mismatches = 0;
5299 mddev->suspend_lo = mddev->suspend_hi = 0;
5300 mddev->sync_speed_min = mddev->sync_speed_max = 0;
5301 mddev->recovery = 0;
5304 mddev->degraded = 0;
5305 mddev->safemode = 0;
5306 mddev->merge_check_needed = 0;
5307 mddev->bitmap_info.offset = 0;
5308 mddev->bitmap_info.default_offset = 0;
5309 mddev->bitmap_info.default_space = 0;
5310 mddev->bitmap_info.chunksize = 0;
5311 mddev->bitmap_info.daemon_sleep = 0;
5312 mddev->bitmap_info.max_write_behind = 0;
5315 static void __md_stop_writes(struct mddev *mddev)
5317 if (mddev->sync_thread) {
5318 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5319 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5320 reap_sync_thread(mddev);
5323 del_timer_sync(&mddev->safemode_timer);
5325 bitmap_flush(mddev);
5326 md_super_wait(mddev);
5328 if (!mddev->in_sync || mddev->flags) {
5329 /* mark array as shutdown cleanly */
5331 md_update_sb(mddev, 1);
5335 void md_stop_writes(struct mddev *mddev)
5338 __md_stop_writes(mddev);
5339 mddev_unlock(mddev);
5341 EXPORT_SYMBOL_GPL(md_stop_writes);
5343 void md_stop(struct mddev *mddev)
5346 mddev->pers->stop(mddev);
5347 if (mddev->pers->sync_request && mddev->to_remove == NULL)
5348 mddev->to_remove = &md_redundancy_group;
5349 module_put(mddev->pers->owner);
5351 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5353 EXPORT_SYMBOL_GPL(md_stop);
5355 static int md_set_readonly(struct mddev *mddev, struct block_device *bdev)
5358 mutex_lock(&mddev->open_mutex);
5359 if (atomic_read(&mddev->openers) > !!bdev) {
5360 printk("md: %s still in use.\n",mdname(mddev));
5365 sync_blockdev(bdev);
5367 __md_stop_writes(mddev);
5373 set_disk_ro(mddev->gendisk, 1);
5374 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5375 sysfs_notify_dirent_safe(mddev->sysfs_state);
5379 mutex_unlock(&mddev->open_mutex);
5384 * 0 - completely stop and dis-assemble array
5385 * 2 - stop but do not disassemble array
5387 static int do_md_stop(struct mddev * mddev, int mode,
5388 struct block_device *bdev)
5390 struct gendisk *disk = mddev->gendisk;
5391 struct md_rdev *rdev;
5393 mutex_lock(&mddev->open_mutex);
5394 if (atomic_read(&mddev->openers) > !!bdev ||
5395 mddev->sysfs_active) {
5396 printk("md: %s still in use.\n",mdname(mddev));
5397 mutex_unlock(&mddev->open_mutex);
5401 /* It is possible IO was issued on some other
5402 * open file which was closed before we took ->open_mutex.
5403 * As that was not the last close __blkdev_put will not
5404 * have called sync_blockdev, so we must.
5406 sync_blockdev(bdev);
5410 set_disk_ro(disk, 0);
5412 __md_stop_writes(mddev);
5414 mddev->queue->merge_bvec_fn = NULL;
5415 mddev->queue->backing_dev_info.congested_fn = NULL;
5417 /* tell userspace to handle 'inactive' */
5418 sysfs_notify_dirent_safe(mddev->sysfs_state);
5420 rdev_for_each(rdev, mddev)
5421 if (rdev->raid_disk >= 0)
5422 sysfs_unlink_rdev(mddev, rdev);
5424 set_capacity(disk, 0);
5425 mutex_unlock(&mddev->open_mutex);
5427 revalidate_disk(disk);
5432 mutex_unlock(&mddev->open_mutex);
5434 * Free resources if final stop
5437 printk(KERN_INFO "md: %s stopped.\n", mdname(mddev));
5439 bitmap_destroy(mddev);
5440 if (mddev->bitmap_info.file) {
5441 restore_bitmap_write_access(mddev->bitmap_info.file);
5442 fput(mddev->bitmap_info.file);
5443 mddev->bitmap_info.file = NULL;
5445 mddev->bitmap_info.offset = 0;
5447 export_array(mddev);
5450 kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
5451 if (mddev->hold_active == UNTIL_STOP)
5452 mddev->hold_active = 0;
5454 blk_integrity_unregister(disk);
5455 md_new_event(mddev);
5456 sysfs_notify_dirent_safe(mddev->sysfs_state);
5461 static void autorun_array(struct mddev *mddev)
5463 struct md_rdev *rdev;
5466 if (list_empty(&mddev->disks))
5469 printk(KERN_INFO "md: running: ");
5471 rdev_for_each(rdev, mddev) {
5472 char b[BDEVNAME_SIZE];
5473 printk("<%s>", bdevname(rdev->bdev,b));
5477 err = do_md_run(mddev);
5479 printk(KERN_WARNING "md: do_md_run() returned %d\n", err);
5480 do_md_stop(mddev, 0, NULL);
5485 * lets try to run arrays based on all disks that have arrived
5486 * until now. (those are in pending_raid_disks)
5488 * the method: pick the first pending disk, collect all disks with
5489 * the same UUID, remove all from the pending list and put them into
5490 * the 'same_array' list. Then order this list based on superblock
5491 * update time (freshest comes first), kick out 'old' disks and
5492 * compare superblocks. If everything's fine then run it.
5494 * If "unit" is allocated, then bump its reference count
5496 static void autorun_devices(int part)
5498 struct md_rdev *rdev0, *rdev, *tmp;
5499 struct mddev *mddev;
5500 char b[BDEVNAME_SIZE];
5502 printk(KERN_INFO "md: autorun ...\n");
5503 while (!list_empty(&pending_raid_disks)) {
5506 LIST_HEAD(candidates);
5507 rdev0 = list_entry(pending_raid_disks.next,
5508 struct md_rdev, same_set);
5510 printk(KERN_INFO "md: considering %s ...\n",
5511 bdevname(rdev0->bdev,b));
5512 INIT_LIST_HEAD(&candidates);
5513 rdev_for_each_list(rdev, tmp, &pending_raid_disks)
5514 if (super_90_load(rdev, rdev0, 0) >= 0) {
5515 printk(KERN_INFO "md: adding %s ...\n",
5516 bdevname(rdev->bdev,b));
5517 list_move(&rdev->same_set, &candidates);
5520 * now we have a set of devices, with all of them having
5521 * mostly sane superblocks. It's time to allocate the
5525 dev = MKDEV(mdp_major,
5526 rdev0->preferred_minor << MdpMinorShift);
5527 unit = MINOR(dev) >> MdpMinorShift;
5529 dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
5532 if (rdev0->preferred_minor != unit) {
5533 printk(KERN_INFO "md: unit number in %s is bad: %d\n",
5534 bdevname(rdev0->bdev, b), rdev0->preferred_minor);
5538 md_probe(dev, NULL, NULL);
5539 mddev = mddev_find(dev);
5540 if (!mddev || !mddev->gendisk) {
5544 "md: cannot allocate memory for md drive.\n");
5547 if (mddev_lock(mddev))
5548 printk(KERN_WARNING "md: %s locked, cannot run\n",
5550 else if (mddev->raid_disks || mddev->major_version
5551 || !list_empty(&mddev->disks)) {
5553 "md: %s already running, cannot run %s\n",
5554 mdname(mddev), bdevname(rdev0->bdev,b));
5555 mddev_unlock(mddev);
5557 printk(KERN_INFO "md: created %s\n", mdname(mddev));
5558 mddev->persistent = 1;
5559 rdev_for_each_list(rdev, tmp, &candidates) {
5560 list_del_init(&rdev->same_set);
5561 if (bind_rdev_to_array(rdev, mddev))
5564 autorun_array(mddev);
5565 mddev_unlock(mddev);
5567 /* on success, candidates will be empty, on error
5570 rdev_for_each_list(rdev, tmp, &candidates) {
5571 list_del_init(&rdev->same_set);
5576 printk(KERN_INFO "md: ... autorun DONE.\n");
5578 #endif /* !MODULE */
5580 static int get_version(void __user * arg)
5584 ver.major = MD_MAJOR_VERSION;
5585 ver.minor = MD_MINOR_VERSION;
5586 ver.patchlevel = MD_PATCHLEVEL_VERSION;
5588 if (copy_to_user(arg, &ver, sizeof(ver)))
5594 static int get_array_info(struct mddev * mddev, void __user * arg)
5596 mdu_array_info_t info;
5597 int nr,working,insync,failed,spare;
5598 struct md_rdev *rdev;
5600 nr=working=insync=failed=spare=0;
5601 rdev_for_each(rdev, mddev) {
5603 if (test_bit(Faulty, &rdev->flags))
5607 if (test_bit(In_sync, &rdev->flags))
5614 info.major_version = mddev->major_version;
5615 info.minor_version = mddev->minor_version;
5616 info.patch_version = MD_PATCHLEVEL_VERSION;
5617 info.ctime = mddev->ctime;
5618 info.level = mddev->level;
5619 info.size = mddev->dev_sectors / 2;
5620 if (info.size != mddev->dev_sectors / 2) /* overflow */
5623 info.raid_disks = mddev->raid_disks;
5624 info.md_minor = mddev->md_minor;
5625 info.not_persistent= !mddev->persistent;
5627 info.utime = mddev->utime;
5630 info.state = (1<<MD_SB_CLEAN);
5631 if (mddev->bitmap && mddev->bitmap_info.offset)
5632 info.state = (1<<MD_SB_BITMAP_PRESENT);
5633 info.active_disks = insync;
5634 info.working_disks = working;
5635 info.failed_disks = failed;
5636 info.spare_disks = spare;
5638 info.layout = mddev->layout;
5639 info.chunk_size = mddev->chunk_sectors << 9;
5641 if (copy_to_user(arg, &info, sizeof(info)))
5647 static int get_bitmap_file(struct mddev * mddev, void __user * arg)
5649 mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */
5650 char *ptr, *buf = NULL;
5653 if (md_allow_write(mddev))
5654 file = kmalloc(sizeof(*file), GFP_NOIO);
5656 file = kmalloc(sizeof(*file), GFP_KERNEL);
5661 /* bitmap disabled, zero the first byte and copy out */
5662 if (!mddev->bitmap || !mddev->bitmap->storage.file) {
5663 file->pathname[0] = '\0';
5667 buf = kmalloc(sizeof(file->pathname), GFP_KERNEL);
5671 ptr = d_path(&mddev->bitmap->storage.file->f_path,
5672 buf, sizeof(file->pathname));
5676 strcpy(file->pathname, ptr);
5680 if (copy_to_user(arg, file, sizeof(*file)))
5688 static int get_disk_info(struct mddev * mddev, void __user * arg)
5690 mdu_disk_info_t info;
5691 struct md_rdev *rdev;
5693 if (copy_from_user(&info, arg, sizeof(info)))
5696 rdev = find_rdev_nr(mddev, info.number);
5698 info.major = MAJOR(rdev->bdev->bd_dev);
5699 info.minor = MINOR(rdev->bdev->bd_dev);
5700 info.raid_disk = rdev->raid_disk;
5702 if (test_bit(Faulty, &rdev->flags))
5703 info.state |= (1<<MD_DISK_FAULTY);
5704 else if (test_bit(In_sync, &rdev->flags)) {
5705 info.state |= (1<<MD_DISK_ACTIVE);
5706 info.state |= (1<<MD_DISK_SYNC);
5708 if (test_bit(WriteMostly, &rdev->flags))
5709 info.state |= (1<<MD_DISK_WRITEMOSTLY);
5711 info.major = info.minor = 0;
5712 info.raid_disk = -1;
5713 info.state = (1<<MD_DISK_REMOVED);
5716 if (copy_to_user(arg, &info, sizeof(info)))
5722 static int add_new_disk(struct mddev * mddev, mdu_disk_info_t *info)
5724 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
5725 struct md_rdev *rdev;
5726 dev_t dev = MKDEV(info->major,info->minor);
5728 if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
5731 if (!mddev->raid_disks) {
5733 /* expecting a device which has a superblock */
5734 rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
5737 "md: md_import_device returned %ld\n",
5739 return PTR_ERR(rdev);
5741 if (!list_empty(&mddev->disks)) {
5742 struct md_rdev *rdev0
5743 = list_entry(mddev->disks.next,
5744 struct md_rdev, same_set);
5745 err = super_types[mddev->major_version]
5746 .load_super(rdev, rdev0, mddev->minor_version);
5749 "md: %s has different UUID to %s\n",
5750 bdevname(rdev->bdev,b),
5751 bdevname(rdev0->bdev,b2));
5756 err = bind_rdev_to_array(rdev, mddev);
5763 * add_new_disk can be used once the array is assembled
5764 * to add "hot spares". They must already have a superblock
5769 if (!mddev->pers->hot_add_disk) {
5771 "%s: personality does not support diskops!\n",
5775 if (mddev->persistent)
5776 rdev = md_import_device(dev, mddev->major_version,
5777 mddev->minor_version);
5779 rdev = md_import_device(dev, -1, -1);
5782 "md: md_import_device returned %ld\n",
5784 return PTR_ERR(rdev);
5786 /* set saved_raid_disk if appropriate */
5787 if (!mddev->persistent) {
5788 if (info->state & (1<<MD_DISK_SYNC) &&
5789 info->raid_disk < mddev->raid_disks) {
5790 rdev->raid_disk = info->raid_disk;
5791 set_bit(In_sync, &rdev->flags);
5793 rdev->raid_disk = -1;
5795 super_types[mddev->major_version].
5796 validate_super(mddev, rdev);
5797 if ((info->state & (1<<MD_DISK_SYNC)) &&
5798 rdev->raid_disk != info->raid_disk) {
5799 /* This was a hot-add request, but events doesn't
5800 * match, so reject it.
5806 if (test_bit(In_sync, &rdev->flags))
5807 rdev->saved_raid_disk = rdev->raid_disk;
5809 rdev->saved_raid_disk = -1;
5811 clear_bit(In_sync, &rdev->flags); /* just to be sure */
5812 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
5813 set_bit(WriteMostly, &rdev->flags);
5815 clear_bit(WriteMostly, &rdev->flags);
5817 rdev->raid_disk = -1;
5818 err = bind_rdev_to_array(rdev, mddev);
5819 if (!err && !mddev->pers->hot_remove_disk) {
5820 /* If there is hot_add_disk but no hot_remove_disk
5821 * then added disks for geometry changes,
5822 * and should be added immediately.
5824 super_types[mddev->major_version].
5825 validate_super(mddev, rdev);
5826 err = mddev->pers->hot_add_disk(mddev, rdev);
5828 unbind_rdev_from_array(rdev);
5833 sysfs_notify_dirent_safe(rdev->sysfs_state);
5835 md_update_sb(mddev, 1);
5836 if (mddev->degraded)
5837 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
5838 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5840 md_new_event(mddev);
5841 md_wakeup_thread(mddev->thread);
5845 /* otherwise, add_new_disk is only allowed
5846 * for major_version==0 superblocks
5848 if (mddev->major_version != 0) {
5849 printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n",
5854 if (!(info->state & (1<<MD_DISK_FAULTY))) {
5856 rdev = md_import_device(dev, -1, 0);
5859 "md: error, md_import_device() returned %ld\n",
5861 return PTR_ERR(rdev);
5863 rdev->desc_nr = info->number;
5864 if (info->raid_disk < mddev->raid_disks)
5865 rdev->raid_disk = info->raid_disk;
5867 rdev->raid_disk = -1;
5869 if (rdev->raid_disk < mddev->raid_disks)
5870 if (info->state & (1<<MD_DISK_SYNC))
5871 set_bit(In_sync, &rdev->flags);
5873 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
5874 set_bit(WriteMostly, &rdev->flags);
5876 if (!mddev->persistent) {
5877 printk(KERN_INFO "md: nonpersistent superblock ...\n");
5878 rdev->sb_start = i_size_read(rdev->bdev->bd_inode) / 512;
5880 rdev->sb_start = calc_dev_sboffset(rdev);
5881 rdev->sectors = rdev->sb_start;
5883 err = bind_rdev_to_array(rdev, mddev);
5893 static int hot_remove_disk(struct mddev * mddev, dev_t dev)
5895 char b[BDEVNAME_SIZE];
5896 struct md_rdev *rdev;
5898 rdev = find_rdev(mddev, dev);
5902 if (rdev->raid_disk >= 0)
5905 kick_rdev_from_array(rdev);
5906 md_update_sb(mddev, 1);
5907 md_new_event(mddev);
5911 printk(KERN_WARNING "md: cannot remove active disk %s from %s ...\n",
5912 bdevname(rdev->bdev,b), mdname(mddev));
5916 static int hot_add_disk(struct mddev * mddev, dev_t dev)
5918 char b[BDEVNAME_SIZE];
5920 struct md_rdev *rdev;
5925 if (mddev->major_version != 0) {
5926 printk(KERN_WARNING "%s: HOT_ADD may only be used with"
5927 " version-0 superblocks.\n",
5931 if (!mddev->pers->hot_add_disk) {
5933 "%s: personality does not support diskops!\n",
5938 rdev = md_import_device(dev, -1, 0);
5941 "md: error, md_import_device() returned %ld\n",
5946 if (mddev->persistent)
5947 rdev->sb_start = calc_dev_sboffset(rdev);
5949 rdev->sb_start = i_size_read(rdev->bdev->bd_inode) / 512;
5951 rdev->sectors = rdev->sb_start;
5953 if (test_bit(Faulty, &rdev->flags)) {
5955 "md: can not hot-add faulty %s disk to %s!\n",
5956 bdevname(rdev->bdev,b), mdname(mddev));
5960 clear_bit(In_sync, &rdev->flags);
5962 rdev->saved_raid_disk = -1;
5963 err = bind_rdev_to_array(rdev, mddev);
5968 * The rest should better be atomic, we can have disk failures
5969 * noticed in interrupt contexts ...
5972 rdev->raid_disk = -1;
5974 md_update_sb(mddev, 1);
5977 * Kick recovery, maybe this spare has to be added to the
5978 * array immediately.
5980 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5981 md_wakeup_thread(mddev->thread);
5982 md_new_event(mddev);
5990 static int set_bitmap_file(struct mddev *mddev, int fd)
5995 if (!mddev->pers->quiesce)
5997 if (mddev->recovery || mddev->sync_thread)
5999 /* we should be able to change the bitmap.. */
6005 return -EEXIST; /* cannot add when bitmap is present */
6006 mddev->bitmap_info.file = fget(fd);
6008 if (mddev->bitmap_info.file == NULL) {
6009 printk(KERN_ERR "%s: error: failed to get bitmap file\n",
6014 err = deny_bitmap_write_access(mddev->bitmap_info.file);
6016 printk(KERN_ERR "%s: error: bitmap file is already in use\n",
6018 fput(mddev->bitmap_info.file);
6019 mddev->bitmap_info.file = NULL;
6022 mddev->bitmap_info.offset = 0; /* file overrides offset */
6023 } else if (mddev->bitmap == NULL)
6024 return -ENOENT; /* cannot remove what isn't there */
6027 mddev->pers->quiesce(mddev, 1);
6029 err = bitmap_create(mddev);
6031 err = bitmap_load(mddev);
6033 if (fd < 0 || err) {
6034 bitmap_destroy(mddev);
6035 fd = -1; /* make sure to put the file */
6037 mddev->pers->quiesce(mddev, 0);
6040 if (mddev->bitmap_info.file) {
6041 restore_bitmap_write_access(mddev->bitmap_info.file);
6042 fput(mddev->bitmap_info.file);
6044 mddev->bitmap_info.file = NULL;
6051 * set_array_info is used two different ways
6052 * The original usage is when creating a new array.
6053 * In this usage, raid_disks is > 0 and it together with
6054 * level, size, not_persistent,layout,chunksize determine the
6055 * shape of the array.
6056 * This will always create an array with a type-0.90.0 superblock.
6057 * The newer usage is when assembling an array.
6058 * In this case raid_disks will be 0, and the major_version field is
6059 * use to determine which style super-blocks are to be found on the devices.
6060 * The minor and patch _version numbers are also kept incase the
6061 * super_block handler wishes to interpret them.
6063 static int set_array_info(struct mddev * mddev, mdu_array_info_t *info)
6066 if (info->raid_disks == 0) {
6067 /* just setting version number for superblock loading */
6068 if (info->major_version < 0 ||
6069 info->major_version >= ARRAY_SIZE(super_types) ||
6070 super_types[info->major_version].name == NULL) {
6071 /* maybe try to auto-load a module? */
6073 "md: superblock version %d not known\n",
6074 info->major_version);
6077 mddev->major_version = info->major_version;
6078 mddev->minor_version = info->minor_version;
6079 mddev->patch_version = info->patch_version;
6080 mddev->persistent = !info->not_persistent;
6081 /* ensure mddev_put doesn't delete this now that there
6082 * is some minimal configuration.
6084 mddev->ctime = get_seconds();
6087 mddev->major_version = MD_MAJOR_VERSION;
6088 mddev->minor_version = MD_MINOR_VERSION;
6089 mddev->patch_version = MD_PATCHLEVEL_VERSION;
6090 mddev->ctime = get_seconds();
6092 mddev->level = info->level;
6093 mddev->clevel[0] = 0;
6094 mddev->dev_sectors = 2 * (sector_t)info->size;
6095 mddev->raid_disks = info->raid_disks;
6096 /* don't set md_minor, it is determined by which /dev/md* was
6099 if (info->state & (1<<MD_SB_CLEAN))
6100 mddev->recovery_cp = MaxSector;
6102 mddev->recovery_cp = 0;
6103 mddev->persistent = ! info->not_persistent;
6104 mddev->external = 0;
6106 mddev->layout = info->layout;
6107 mddev->chunk_sectors = info->chunk_size >> 9;
6109 mddev->max_disks = MD_SB_DISKS;
6111 if (mddev->persistent)
6113 set_bit(MD_CHANGE_DEVS, &mddev->flags);
6115 mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9;
6116 mddev->bitmap_info.default_space = 64*2 - (MD_SB_BYTES >> 9);
6117 mddev->bitmap_info.offset = 0;
6119 mddev->reshape_position = MaxSector;
6122 * Generate a 128 bit UUID
6124 get_random_bytes(mddev->uuid, 16);
6126 mddev->new_level = mddev->level;
6127 mddev->new_chunk_sectors = mddev->chunk_sectors;
6128 mddev->new_layout = mddev->layout;
6129 mddev->delta_disks = 0;
6130 mddev->reshape_backwards = 0;
6135 void md_set_array_sectors(struct mddev *mddev, sector_t array_sectors)
6137 WARN(!mddev_is_locked(mddev), "%s: unlocked mddev!\n", __func__);
6139 if (mddev->external_size)
6142 mddev->array_sectors = array_sectors;
6144 EXPORT_SYMBOL(md_set_array_sectors);
6146 static int update_size(struct mddev *mddev, sector_t num_sectors)
6148 struct md_rdev *rdev;
6150 int fit = (num_sectors == 0);
6152 if (mddev->pers->resize == NULL)
6154 /* The "num_sectors" is the number of sectors of each device that
6155 * is used. This can only make sense for arrays with redundancy.
6156 * linear and raid0 always use whatever space is available. We can only
6157 * consider changing this number if no resync or reconstruction is
6158 * happening, and if the new size is acceptable. It must fit before the
6159 * sb_start or, if that is <data_offset, it must fit before the size
6160 * of each device. If num_sectors is zero, we find the largest size
6163 if (mddev->sync_thread)
6166 rdev_for_each(rdev, mddev) {
6167 sector_t avail = rdev->sectors;
6169 if (fit && (num_sectors == 0 || num_sectors > avail))
6170 num_sectors = avail;
6171 if (avail < num_sectors)
6174 rv = mddev->pers->resize(mddev, num_sectors);
6176 revalidate_disk(mddev->gendisk);
6180 static int update_raid_disks(struct mddev *mddev, int raid_disks)
6183 struct md_rdev *rdev;
6184 /* change the number of raid disks */
6185 if (mddev->pers->check_reshape == NULL)
6187 if (raid_disks <= 0 ||
6188 (mddev->max_disks && raid_disks >= mddev->max_disks))
6190 if (mddev->sync_thread || mddev->reshape_position != MaxSector)
6193 rdev_for_each(rdev, mddev) {
6194 if (mddev->raid_disks < raid_disks &&
6195 rdev->data_offset < rdev->new_data_offset)
6197 if (mddev->raid_disks > raid_disks &&
6198 rdev->data_offset > rdev->new_data_offset)
6202 mddev->delta_disks = raid_disks - mddev->raid_disks;
6203 if (mddev->delta_disks < 0)
6204 mddev->reshape_backwards = 1;
6205 else if (mddev->delta_disks > 0)
6206 mddev->reshape_backwards = 0;
6208 rv = mddev->pers->check_reshape(mddev);
6210 mddev->delta_disks = 0;
6211 mddev->reshape_backwards = 0;
6218 * update_array_info is used to change the configuration of an
6220 * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
6221 * fields in the info are checked against the array.
6222 * Any differences that cannot be handled will cause an error.
6223 * Normally, only one change can be managed at a time.
6225 static int update_array_info(struct mddev *mddev, mdu_array_info_t *info)
6231 /* calculate expected state,ignoring low bits */
6232 if (mddev->bitmap && mddev->bitmap_info.offset)
6233 state |= (1 << MD_SB_BITMAP_PRESENT);
6235 if (mddev->major_version != info->major_version ||
6236 mddev->minor_version != info->minor_version ||
6237 /* mddev->patch_version != info->patch_version || */
6238 mddev->ctime != info->ctime ||
6239 mddev->level != info->level ||
6240 /* mddev->layout != info->layout || */
6241 !mddev->persistent != info->not_persistent||
6242 mddev->chunk_sectors != info->chunk_size >> 9 ||
6243 /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
6244 ((state^info->state) & 0xfffffe00)
6247 /* Check there is only one change */
6248 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
6250 if (mddev->raid_disks != info->raid_disks)
6252 if (mddev->layout != info->layout)
6254 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT))
6261 if (mddev->layout != info->layout) {
6263 * we don't need to do anything at the md level, the
6264 * personality will take care of it all.
6266 if (mddev->pers->check_reshape == NULL)
6269 mddev->new_layout = info->layout;
6270 rv = mddev->pers->check_reshape(mddev);
6272 mddev->new_layout = mddev->layout;
6276 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
6277 rv = update_size(mddev, (sector_t)info->size * 2);
6279 if (mddev->raid_disks != info->raid_disks)
6280 rv = update_raid_disks(mddev, info->raid_disks);
6282 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) {
6283 if (mddev->pers->quiesce == NULL)
6285 if (mddev->recovery || mddev->sync_thread)
6287 if (info->state & (1<<MD_SB_BITMAP_PRESENT)) {
6288 /* add the bitmap */
6291 if (mddev->bitmap_info.default_offset == 0)
6293 mddev->bitmap_info.offset =
6294 mddev->bitmap_info.default_offset;
6295 mddev->bitmap_info.space =
6296 mddev->bitmap_info.default_space;
6297 mddev->pers->quiesce(mddev, 1);
6298 rv = bitmap_create(mddev);
6300 rv = bitmap_load(mddev);
6302 bitmap_destroy(mddev);
6303 mddev->pers->quiesce(mddev, 0);
6305 /* remove the bitmap */
6308 if (mddev->bitmap->storage.file)
6310 mddev->pers->quiesce(mddev, 1);
6311 bitmap_destroy(mddev);
6312 mddev->pers->quiesce(mddev, 0);
6313 mddev->bitmap_info.offset = 0;
6316 md_update_sb(mddev, 1);
6320 static int set_disk_faulty(struct mddev *mddev, dev_t dev)
6322 struct md_rdev *rdev;
6324 if (mddev->pers == NULL)
6327 rdev = find_rdev(mddev, dev);
6331 md_error(mddev, rdev);
6332 if (!test_bit(Faulty, &rdev->flags))
6338 * We have a problem here : there is no easy way to give a CHS
6339 * virtual geometry. We currently pretend that we have a 2 heads
6340 * 4 sectors (with a BIG number of cylinders...). This drives
6341 * dosfs just mad... ;-)
6343 static int md_getgeo(struct block_device *bdev, struct hd_geometry *geo)
6345 struct mddev *mddev = bdev->bd_disk->private_data;
6349 geo->cylinders = mddev->array_sectors / 8;
6353 static int md_ioctl(struct block_device *bdev, fmode_t mode,
6354 unsigned int cmd, unsigned long arg)
6357 void __user *argp = (void __user *)arg;
6358 struct mddev *mddev = NULL;
6363 case GET_ARRAY_INFO:
6367 if (!capable(CAP_SYS_ADMIN))
6372 * Commands dealing with the RAID driver but not any
6378 err = get_version(argp);
6381 case PRINT_RAID_DEBUG:
6389 autostart_arrays(arg);
6396 * Commands creating/starting a new array:
6399 mddev = bdev->bd_disk->private_data;
6406 err = mddev_lock(mddev);
6409 "md: ioctl lock interrupted, reason %d, cmd %d\n",
6416 case SET_ARRAY_INFO:
6418 mdu_array_info_t info;
6420 memset(&info, 0, sizeof(info));
6421 else if (copy_from_user(&info, argp, sizeof(info))) {
6426 err = update_array_info(mddev, &info);
6428 printk(KERN_WARNING "md: couldn't update"
6429 " array info. %d\n", err);
6434 if (!list_empty(&mddev->disks)) {
6436 "md: array %s already has disks!\n",
6441 if (mddev->raid_disks) {
6443 "md: array %s already initialised!\n",
6448 err = set_array_info(mddev, &info);
6450 printk(KERN_WARNING "md: couldn't set"
6451 " array info. %d\n", err);
6461 * Commands querying/configuring an existing array:
6463 /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
6464 * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */
6465 if ((!mddev->raid_disks && !mddev->external)
6466 && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY
6467 && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE
6468 && cmd != GET_BITMAP_FILE) {
6474 * Commands even a read-only array can execute:
6478 case GET_ARRAY_INFO:
6479 err = get_array_info(mddev, argp);
6482 case GET_BITMAP_FILE:
6483 err = get_bitmap_file(mddev, argp);
6487 err = get_disk_info(mddev, argp);
6490 case RESTART_ARRAY_RW:
6491 err = restart_array(mddev);
6495 err = do_md_stop(mddev, 0, bdev);
6499 err = md_set_readonly(mddev, bdev);
6503 if (get_user(ro, (int __user *)(arg))) {
6509 /* if the bdev is going readonly the value of mddev->ro
6510 * does not matter, no writes are coming
6515 /* are we are already prepared for writes? */
6519 /* transitioning to readauto need only happen for
6520 * arrays that call md_write_start
6523 err = restart_array(mddev);
6526 set_disk_ro(mddev->gendisk, 0);
6533 * The remaining ioctls are changing the state of the
6534 * superblock, so we do not allow them on read-only arrays.
6535 * However non-MD ioctls (e.g. get-size) will still come through
6536 * here and hit the 'default' below, so only disallow
6537 * 'md' ioctls, and switch to rw mode if started auto-readonly.
6539 if (_IOC_TYPE(cmd) == MD_MAJOR && mddev->ro && mddev->pers) {
6540 if (mddev->ro == 2) {
6542 sysfs_notify_dirent_safe(mddev->sysfs_state);
6543 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6544 md_wakeup_thread(mddev->thread);
6555 mdu_disk_info_t info;
6556 if (copy_from_user(&info, argp, sizeof(info)))
6559 err = add_new_disk(mddev, &info);
6563 case HOT_REMOVE_DISK:
6564 err = hot_remove_disk(mddev, new_decode_dev(arg));
6568 err = hot_add_disk(mddev, new_decode_dev(arg));
6571 case SET_DISK_FAULTY:
6572 err = set_disk_faulty(mddev, new_decode_dev(arg));
6576 err = do_md_run(mddev);
6579 case SET_BITMAP_FILE:
6580 err = set_bitmap_file(mddev, (int)arg);
6590 if (mddev->hold_active == UNTIL_IOCTL &&
6592 mddev->hold_active = 0;
6593 mddev_unlock(mddev);
6602 #ifdef CONFIG_COMPAT
6603 static int md_compat_ioctl(struct block_device *bdev, fmode_t mode,
6604 unsigned int cmd, unsigned long arg)
6607 case HOT_REMOVE_DISK:
6609 case SET_DISK_FAULTY:
6610 case SET_BITMAP_FILE:
6611 /* These take in integer arg, do not convert */
6614 arg = (unsigned long)compat_ptr(arg);
6618 return md_ioctl(bdev, mode, cmd, arg);
6620 #endif /* CONFIG_COMPAT */
6622 static int md_open(struct block_device *bdev, fmode_t mode)
6625 * Succeed if we can lock the mddev, which confirms that
6626 * it isn't being stopped right now.
6628 struct mddev *mddev = mddev_find(bdev->bd_dev);
6634 if (mddev->gendisk != bdev->bd_disk) {
6635 /* we are racing with mddev_put which is discarding this
6639 /* Wait until bdev->bd_disk is definitely gone */
6640 flush_workqueue(md_misc_wq);
6641 /* Then retry the open from the top */
6642 return -ERESTARTSYS;
6644 BUG_ON(mddev != bdev->bd_disk->private_data);
6646 if ((err = mutex_lock_interruptible(&mddev->open_mutex)))
6650 atomic_inc(&mddev->openers);
6651 mutex_unlock(&mddev->open_mutex);
6653 check_disk_change(bdev);
6658 static int md_release(struct gendisk *disk, fmode_t mode)
6660 struct mddev *mddev = disk->private_data;
6663 atomic_dec(&mddev->openers);
6669 static int md_media_changed(struct gendisk *disk)
6671 struct mddev *mddev = disk->private_data;
6673 return mddev->changed;
6676 static int md_revalidate(struct gendisk *disk)
6678 struct mddev *mddev = disk->private_data;
6683 static const struct block_device_operations md_fops =
6685 .owner = THIS_MODULE,
6687 .release = md_release,
6689 #ifdef CONFIG_COMPAT
6690 .compat_ioctl = md_compat_ioctl,
6692 .getgeo = md_getgeo,
6693 .media_changed = md_media_changed,
6694 .revalidate_disk= md_revalidate,
6697 static int md_thread(void * arg)
6699 struct md_thread *thread = arg;
6702 * md_thread is a 'system-thread', it's priority should be very
6703 * high. We avoid resource deadlocks individually in each
6704 * raid personality. (RAID5 does preallocation) We also use RR and
6705 * the very same RT priority as kswapd, thus we will never get
6706 * into a priority inversion deadlock.
6708 * we definitely have to have equal or higher priority than
6709 * bdflush, otherwise bdflush will deadlock if there are too
6710 * many dirty RAID5 blocks.
6713 allow_signal(SIGKILL);
6714 while (!kthread_should_stop()) {
6716 /* We need to wait INTERRUPTIBLE so that
6717 * we don't add to the load-average.
6718 * That means we need to be sure no signals are
6721 if (signal_pending(current))
6722 flush_signals(current);
6724 wait_event_interruptible_timeout
6726 test_bit(THREAD_WAKEUP, &thread->flags)
6727 || kthread_should_stop(),
6730 clear_bit(THREAD_WAKEUP, &thread->flags);
6731 if (!kthread_should_stop())
6732 thread->run(thread->mddev);
6738 void md_wakeup_thread(struct md_thread *thread)
6741 pr_debug("md: waking up MD thread %s.\n", thread->tsk->comm);
6742 set_bit(THREAD_WAKEUP, &thread->flags);
6743 wake_up(&thread->wqueue);
6747 struct md_thread *md_register_thread(void (*run) (struct mddev *), struct mddev *mddev,
6750 struct md_thread *thread;
6752 thread = kzalloc(sizeof(struct md_thread), GFP_KERNEL);
6756 init_waitqueue_head(&thread->wqueue);
6759 thread->mddev = mddev;
6760 thread->timeout = MAX_SCHEDULE_TIMEOUT;
6761 thread->tsk = kthread_run(md_thread, thread,
6763 mdname(thread->mddev),
6765 if (IS_ERR(thread->tsk)) {
6772 void md_unregister_thread(struct md_thread **threadp)
6774 struct md_thread *thread = *threadp;
6777 pr_debug("interrupting MD-thread pid %d\n", task_pid_nr(thread->tsk));
6778 /* Locking ensures that mddev_unlock does not wake_up a
6779 * non-existent thread
6781 spin_lock(&pers_lock);
6783 spin_unlock(&pers_lock);
6785 kthread_stop(thread->tsk);
6789 void md_error(struct mddev *mddev, struct md_rdev *rdev)
6796 if (!rdev || test_bit(Faulty, &rdev->flags))
6799 if (!mddev->pers || !mddev->pers->error_handler)
6801 mddev->pers->error_handler(mddev,rdev);
6802 if (mddev->degraded)
6803 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
6804 sysfs_notify_dirent_safe(rdev->sysfs_state);
6805 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6806 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6807 md_wakeup_thread(mddev->thread);
6808 if (mddev->event_work.func)
6809 queue_work(md_misc_wq, &mddev->event_work);
6810 md_new_event_inintr(mddev);
6813 /* seq_file implementation /proc/mdstat */
6815 static void status_unused(struct seq_file *seq)
6818 struct md_rdev *rdev;
6820 seq_printf(seq, "unused devices: ");
6822 list_for_each_entry(rdev, &pending_raid_disks, same_set) {
6823 char b[BDEVNAME_SIZE];
6825 seq_printf(seq, "%s ",
6826 bdevname(rdev->bdev,b));
6829 seq_printf(seq, "<none>");
6831 seq_printf(seq, "\n");
6835 static void status_resync(struct seq_file *seq, struct mddev * mddev)
6837 sector_t max_sectors, resync, res;
6838 unsigned long dt, db;
6841 unsigned int per_milli;
6843 resync = mddev->curr_resync - atomic_read(&mddev->recovery_active);
6845 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ||
6846 test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
6847 max_sectors = mddev->resync_max_sectors;
6849 max_sectors = mddev->dev_sectors;
6852 * Should not happen.
6858 /* Pick 'scale' such that (resync>>scale)*1000 will fit
6859 * in a sector_t, and (max_sectors>>scale) will fit in a
6860 * u32, as those are the requirements for sector_div.
6861 * Thus 'scale' must be at least 10
6864 if (sizeof(sector_t) > sizeof(unsigned long)) {
6865 while ( max_sectors/2 > (1ULL<<(scale+32)))
6868 res = (resync>>scale)*1000;
6869 sector_div(res, (u32)((max_sectors>>scale)+1));
6873 int i, x = per_milli/50, y = 20-x;
6874 seq_printf(seq, "[");
6875 for (i = 0; i < x; i++)
6876 seq_printf(seq, "=");
6877 seq_printf(seq, ">");
6878 for (i = 0; i < y; i++)
6879 seq_printf(seq, ".");
6880 seq_printf(seq, "] ");
6882 seq_printf(seq, " %s =%3u.%u%% (%llu/%llu)",
6883 (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)?
6885 (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)?
6887 (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
6888 "resync" : "recovery"))),
6889 per_milli/10, per_milli % 10,
6890 (unsigned long long) resync/2,
6891 (unsigned long long) max_sectors/2);
6894 * dt: time from mark until now
6895 * db: blocks written from mark until now
6896 * rt: remaining time
6898 * rt is a sector_t, so could be 32bit or 64bit.
6899 * So we divide before multiply in case it is 32bit and close
6901 * We scale the divisor (db) by 32 to avoid losing precision
6902 * near the end of resync when the number of remaining sectors
6904 * We then divide rt by 32 after multiplying by db to compensate.
6905 * The '+1' avoids division by zero if db is very small.
6907 dt = ((jiffies - mddev->resync_mark) / HZ);
6909 db = (mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active))
6910 - mddev->resync_mark_cnt;
6912 rt = max_sectors - resync; /* number of remaining sectors */
6913 sector_div(rt, db/32+1);
6917 seq_printf(seq, " finish=%lu.%lumin", (unsigned long)rt / 60,
6918 ((unsigned long)rt % 60)/6);
6920 seq_printf(seq, " speed=%ldK/sec", db/2/dt);
6923 static void *md_seq_start(struct seq_file *seq, loff_t *pos)
6925 struct list_head *tmp;
6927 struct mddev *mddev;
6935 spin_lock(&all_mddevs_lock);
6936 list_for_each(tmp,&all_mddevs)
6938 mddev = list_entry(tmp, struct mddev, all_mddevs);
6940 spin_unlock(&all_mddevs_lock);
6943 spin_unlock(&all_mddevs_lock);
6945 return (void*)2;/* tail */
6949 static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
6951 struct list_head *tmp;
6952 struct mddev *next_mddev, *mddev = v;
6958 spin_lock(&all_mddevs_lock);
6960 tmp = all_mddevs.next;
6962 tmp = mddev->all_mddevs.next;
6963 if (tmp != &all_mddevs)
6964 next_mddev = mddev_get(list_entry(tmp,struct mddev,all_mddevs));
6966 next_mddev = (void*)2;
6969 spin_unlock(&all_mddevs_lock);
6977 static void md_seq_stop(struct seq_file *seq, void *v)
6979 struct mddev *mddev = v;
6981 if (mddev && v != (void*)1 && v != (void*)2)
6985 static int md_seq_show(struct seq_file *seq, void *v)
6987 struct mddev *mddev = v;
6989 struct md_rdev *rdev;
6991 if (v == (void*)1) {
6992 struct md_personality *pers;
6993 seq_printf(seq, "Personalities : ");
6994 spin_lock(&pers_lock);
6995 list_for_each_entry(pers, &pers_list, list)
6996 seq_printf(seq, "[%s] ", pers->name);
6998 spin_unlock(&pers_lock);
6999 seq_printf(seq, "\n");
7000 seq->poll_event = atomic_read(&md_event_count);
7003 if (v == (void*)2) {
7008 if (mddev_lock(mddev) < 0)
7011 if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
7012 seq_printf(seq, "%s : %sactive", mdname(mddev),
7013 mddev->pers ? "" : "in");
7016 seq_printf(seq, " (read-only)");
7018 seq_printf(seq, " (auto-read-only)");
7019 seq_printf(seq, " %s", mddev->pers->name);
7023 rdev_for_each(rdev, mddev) {
7024 char b[BDEVNAME_SIZE];
7025 seq_printf(seq, " %s[%d]",
7026 bdevname(rdev->bdev,b), rdev->desc_nr);
7027 if (test_bit(WriteMostly, &rdev->flags))
7028 seq_printf(seq, "(W)");
7029 if (test_bit(Faulty, &rdev->flags)) {
7030 seq_printf(seq, "(F)");
7033 if (rdev->raid_disk < 0)
7034 seq_printf(seq, "(S)"); /* spare */
7035 if (test_bit(Replacement, &rdev->flags))
7036 seq_printf(seq, "(R)");
7037 sectors += rdev->sectors;
7040 if (!list_empty(&mddev->disks)) {
7042 seq_printf(seq, "\n %llu blocks",
7043 (unsigned long long)
7044 mddev->array_sectors / 2);
7046 seq_printf(seq, "\n %llu blocks",
7047 (unsigned long long)sectors / 2);
7049 if (mddev->persistent) {
7050 if (mddev->major_version != 0 ||
7051 mddev->minor_version != 90) {
7052 seq_printf(seq," super %d.%d",
7053 mddev->major_version,
7054 mddev->minor_version);
7056 } else if (mddev->external)
7057 seq_printf(seq, " super external:%s",
7058 mddev->metadata_type);
7060 seq_printf(seq, " super non-persistent");
7063 mddev->pers->status(seq, mddev);
7064 seq_printf(seq, "\n ");
7065 if (mddev->pers->sync_request) {
7066 if (mddev->curr_resync > 2) {
7067 status_resync(seq, mddev);
7068 seq_printf(seq, "\n ");
7069 } else if (mddev->curr_resync == 1 || mddev->curr_resync == 2)
7070 seq_printf(seq, "\tresync=DELAYED\n ");
7071 else if (mddev->recovery_cp < MaxSector)
7072 seq_printf(seq, "\tresync=PENDING\n ");
7075 seq_printf(seq, "\n ");
7077 bitmap_status(seq, mddev->bitmap);
7079 seq_printf(seq, "\n");
7081 mddev_unlock(mddev);
7086 static const struct seq_operations md_seq_ops = {
7087 .start = md_seq_start,
7088 .next = md_seq_next,
7089 .stop = md_seq_stop,
7090 .show = md_seq_show,
7093 static int md_seq_open(struct inode *inode, struct file *file)
7095 struct seq_file *seq;
7098 error = seq_open(file, &md_seq_ops);
7102 seq = file->private_data;
7103 seq->poll_event = atomic_read(&md_event_count);
7107 static unsigned int mdstat_poll(struct file *filp, poll_table *wait)
7109 struct seq_file *seq = filp->private_data;
7112 poll_wait(filp, &md_event_waiters, wait);
7114 /* always allow read */
7115 mask = POLLIN | POLLRDNORM;
7117 if (seq->poll_event != atomic_read(&md_event_count))
7118 mask |= POLLERR | POLLPRI;
7122 static const struct file_operations md_seq_fops = {
7123 .owner = THIS_MODULE,
7124 .open = md_seq_open,
7126 .llseek = seq_lseek,
7127 .release = seq_release_private,
7128 .poll = mdstat_poll,
7131 int register_md_personality(struct md_personality *p)
7133 spin_lock(&pers_lock);
7134 list_add_tail(&p->list, &pers_list);
7135 printk(KERN_INFO "md: %s personality registered for level %d\n", p->name, p->level);
7136 spin_unlock(&pers_lock);
7140 int unregister_md_personality(struct md_personality *p)
7142 printk(KERN_INFO "md: %s personality unregistered\n", p->name);
7143 spin_lock(&pers_lock);
7144 list_del_init(&p->list);
7145 spin_unlock(&pers_lock);
7149 static int is_mddev_idle(struct mddev *mddev, int init)
7151 struct md_rdev * rdev;
7157 rdev_for_each_rcu(rdev, mddev) {
7158 struct gendisk *disk = rdev->bdev->bd_contains->bd_disk;
7159 curr_events = (int)part_stat_read(&disk->part0, sectors[0]) +
7160 (int)part_stat_read(&disk->part0, sectors[1]) -
7161 atomic_read(&disk->sync_io);
7162 /* sync IO will cause sync_io to increase before the disk_stats
7163 * as sync_io is counted when a request starts, and
7164 * disk_stats is counted when it completes.
7165 * So resync activity will cause curr_events to be smaller than
7166 * when there was no such activity.
7167 * non-sync IO will cause disk_stat to increase without
7168 * increasing sync_io so curr_events will (eventually)
7169 * be larger than it was before. Once it becomes
7170 * substantially larger, the test below will cause
7171 * the array to appear non-idle, and resync will slow
7173 * If there is a lot of outstanding resync activity when
7174 * we set last_event to curr_events, then all that activity
7175 * completing might cause the array to appear non-idle
7176 * and resync will be slowed down even though there might
7177 * not have been non-resync activity. This will only
7178 * happen once though. 'last_events' will soon reflect
7179 * the state where there is little or no outstanding
7180 * resync requests, and further resync activity will
7181 * always make curr_events less than last_events.
7184 if (init || curr_events - rdev->last_events > 64) {
7185 rdev->last_events = curr_events;
7193 void md_done_sync(struct mddev *mddev, int blocks, int ok)
7195 /* another "blocks" (512byte) blocks have been synced */
7196 atomic_sub(blocks, &mddev->recovery_active);
7197 wake_up(&mddev->recovery_wait);
7199 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
7200 md_wakeup_thread(mddev->thread);
7201 // stop recovery, signal do_sync ....
7206 /* md_write_start(mddev, bi)
7207 * If we need to update some array metadata (e.g. 'active' flag
7208 * in superblock) before writing, schedule a superblock update
7209 * and wait for it to complete.
7211 void md_write_start(struct mddev *mddev, struct bio *bi)
7214 if (bio_data_dir(bi) != WRITE)
7217 BUG_ON(mddev->ro == 1);
7218 if (mddev->ro == 2) {
7219 /* need to switch to read/write */
7221 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7222 md_wakeup_thread(mddev->thread);
7223 md_wakeup_thread(mddev->sync_thread);
7226 atomic_inc(&mddev->writes_pending);
7227 if (mddev->safemode == 1)
7228 mddev->safemode = 0;
7229 if (mddev->in_sync) {
7230 spin_lock_irq(&mddev->write_lock);
7231 if (mddev->in_sync) {
7233 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
7234 set_bit(MD_CHANGE_PENDING, &mddev->flags);
7235 md_wakeup_thread(mddev->thread);
7238 spin_unlock_irq(&mddev->write_lock);
7241 sysfs_notify_dirent_safe(mddev->sysfs_state);
7242 wait_event(mddev->sb_wait,
7243 !test_bit(MD_CHANGE_PENDING, &mddev->flags));
7246 void md_write_end(struct mddev *mddev)
7248 if (atomic_dec_and_test(&mddev->writes_pending)) {
7249 if (mddev->safemode == 2)
7250 md_wakeup_thread(mddev->thread);
7251 else if (mddev->safemode_delay)
7252 mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay);
7256 /* md_allow_write(mddev)
7257 * Calling this ensures that the array is marked 'active' so that writes
7258 * may proceed without blocking. It is important to call this before
7259 * attempting a GFP_KERNEL allocation while holding the mddev lock.
7260 * Must be called with mddev_lock held.
7262 * In the ->external case MD_CHANGE_CLEAN can not be cleared until mddev->lock
7263 * is dropped, so return -EAGAIN after notifying userspace.
7265 int md_allow_write(struct mddev *mddev)
7271 if (!mddev->pers->sync_request)
7274 spin_lock_irq(&mddev->write_lock);
7275 if (mddev->in_sync) {
7277 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
7278 set_bit(MD_CHANGE_PENDING, &mddev->flags);
7279 if (mddev->safemode_delay &&
7280 mddev->safemode == 0)
7281 mddev->safemode = 1;
7282 spin_unlock_irq(&mddev->write_lock);
7283 md_update_sb(mddev, 0);
7284 sysfs_notify_dirent_safe(mddev->sysfs_state);
7286 spin_unlock_irq(&mddev->write_lock);
7288 if (test_bit(MD_CHANGE_PENDING, &mddev->flags))
7293 EXPORT_SYMBOL_GPL(md_allow_write);
7295 #define SYNC_MARKS 10
7296 #define SYNC_MARK_STEP (3*HZ)
7297 void md_do_sync(struct mddev *mddev)
7299 struct mddev *mddev2;
7300 unsigned int currspeed = 0,
7302 sector_t max_sectors,j, io_sectors;
7303 unsigned long mark[SYNC_MARKS];
7304 sector_t mark_cnt[SYNC_MARKS];
7306 struct list_head *tmp;
7307 sector_t last_check;
7309 struct md_rdev *rdev;
7311 struct blk_plug plug;
7313 /* just incase thread restarts... */
7314 if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
7316 if (mddev->ro) /* never try to sync a read-only array */
7319 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
7320 if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
7321 desc = "data-check";
7322 else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
7323 desc = "requested-resync";
7326 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
7331 /* we overload curr_resync somewhat here.
7332 * 0 == not engaged in resync at all
7333 * 2 == checking that there is no conflict with another sync
7334 * 1 == like 2, but have yielded to allow conflicting resync to
7336 * other == active in resync - this many blocks
7338 * Before starting a resync we must have set curr_resync to
7339 * 2, and then checked that every "conflicting" array has curr_resync
7340 * less than ours. When we find one that is the same or higher
7341 * we wait on resync_wait. To avoid deadlock, we reduce curr_resync
7342 * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
7343 * This will mean we have to start checking from the beginning again.
7348 mddev->curr_resync = 2;
7351 if (kthread_should_stop())
7352 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
7354 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
7356 for_each_mddev(mddev2, tmp) {
7357 if (mddev2 == mddev)
7359 if (!mddev->parallel_resync
7360 && mddev2->curr_resync
7361 && match_mddev_units(mddev, mddev2)) {
7363 if (mddev < mddev2 && mddev->curr_resync == 2) {
7364 /* arbitrarily yield */
7365 mddev->curr_resync = 1;
7366 wake_up(&resync_wait);
7368 if (mddev > mddev2 && mddev->curr_resync == 1)
7369 /* no need to wait here, we can wait the next
7370 * time 'round when curr_resync == 2
7373 /* We need to wait 'interruptible' so as not to
7374 * contribute to the load average, and not to
7375 * be caught by 'softlockup'
7377 prepare_to_wait(&resync_wait, &wq, TASK_INTERRUPTIBLE);
7378 if (!kthread_should_stop() &&
7379 mddev2->curr_resync >= mddev->curr_resync) {
7380 printk(KERN_INFO "md: delaying %s of %s"
7381 " until %s has finished (they"
7382 " share one or more physical units)\n",
7383 desc, mdname(mddev), mdname(mddev2));
7385 if (signal_pending(current))
7386 flush_signals(current);
7388 finish_wait(&resync_wait, &wq);
7391 finish_wait(&resync_wait, &wq);
7394 } while (mddev->curr_resync < 2);
7397 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
7398 /* resync follows the size requested by the personality,
7399 * which defaults to physical size, but can be virtual size
7401 max_sectors = mddev->resync_max_sectors;
7402 mddev->resync_mismatches = 0;
7403 /* we don't use the checkpoint if there's a bitmap */
7404 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
7405 j = mddev->resync_min;
7406 else if (!mddev->bitmap)
7407 j = mddev->recovery_cp;
7409 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
7410 max_sectors = mddev->resync_max_sectors;
7412 /* recovery follows the physical size of devices */
7413 max_sectors = mddev->dev_sectors;
7416 rdev_for_each_rcu(rdev, mddev)
7417 if (rdev->raid_disk >= 0 &&
7418 !test_bit(Faulty, &rdev->flags) &&
7419 !test_bit(In_sync, &rdev->flags) &&
7420 rdev->recovery_offset < j)
7421 j = rdev->recovery_offset;
7425 printk(KERN_INFO "md: %s of RAID array %s\n", desc, mdname(mddev));
7426 printk(KERN_INFO "md: minimum _guaranteed_ speed:"
7427 " %d KB/sec/disk.\n", speed_min(mddev));
7428 printk(KERN_INFO "md: using maximum available idle IO bandwidth "
7429 "(but not more than %d KB/sec) for %s.\n",
7430 speed_max(mddev), desc);
7432 is_mddev_idle(mddev, 1); /* this initializes IO event counters */
7435 for (m = 0; m < SYNC_MARKS; m++) {
7437 mark_cnt[m] = io_sectors;
7440 mddev->resync_mark = mark[last_mark];
7441 mddev->resync_mark_cnt = mark_cnt[last_mark];
7444 * Tune reconstruction:
7446 window = 32*(PAGE_SIZE/512);
7447 printk(KERN_INFO "md: using %dk window, over a total of %lluk.\n",
7448 window/2, (unsigned long long)max_sectors/2);
7450 atomic_set(&mddev->recovery_active, 0);
7455 "md: resuming %s of %s from checkpoint.\n",
7456 desc, mdname(mddev));
7457 mddev->curr_resync = j;
7459 mddev->curr_resync_completed = j;
7461 blk_start_plug(&plug);
7462 while (j < max_sectors) {
7467 if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
7468 ((mddev->curr_resync > mddev->curr_resync_completed &&
7469 (mddev->curr_resync - mddev->curr_resync_completed)
7470 > (max_sectors >> 4)) ||
7471 (j - mddev->curr_resync_completed)*2
7472 >= mddev->resync_max - mddev->curr_resync_completed
7474 /* time to update curr_resync_completed */
7475 wait_event(mddev->recovery_wait,
7476 atomic_read(&mddev->recovery_active) == 0);
7477 mddev->curr_resync_completed = j;
7478 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
7479 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
7482 while (j >= mddev->resync_max && !kthread_should_stop()) {
7483 /* As this condition is controlled by user-space,
7484 * we can block indefinitely, so use '_interruptible'
7485 * to avoid triggering warnings.
7487 flush_signals(current); /* just in case */
7488 wait_event_interruptible(mddev->recovery_wait,
7489 mddev->resync_max > j
7490 || kthread_should_stop());
7493 if (kthread_should_stop())
7496 sectors = mddev->pers->sync_request(mddev, j, &skipped,
7497 currspeed < speed_min(mddev));
7499 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
7503 if (!skipped) { /* actual IO requested */
7504 io_sectors += sectors;
7505 atomic_add(sectors, &mddev->recovery_active);
7508 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
7512 if (j>1) mddev->curr_resync = j;
7513 mddev->curr_mark_cnt = io_sectors;
7514 if (last_check == 0)
7515 /* this is the earliest that rebuild will be
7516 * visible in /proc/mdstat
7518 md_new_event(mddev);
7520 if (last_check + window > io_sectors || j == max_sectors)
7523 last_check = io_sectors;
7525 if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) {
7527 int next = (last_mark+1) % SYNC_MARKS;
7529 mddev->resync_mark = mark[next];
7530 mddev->resync_mark_cnt = mark_cnt[next];
7531 mark[next] = jiffies;
7532 mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active);
7537 if (kthread_should_stop())
7542 * this loop exits only if either when we are slower than
7543 * the 'hard' speed limit, or the system was IO-idle for
7545 * the system might be non-idle CPU-wise, but we only care
7546 * about not overloading the IO subsystem. (things like an
7547 * e2fsck being done on the RAID array should execute fast)
7551 currspeed = ((unsigned long)(io_sectors-mddev->resync_mark_cnt))/2
7552 /((jiffies-mddev->resync_mark)/HZ +1) +1;
7554 if (currspeed > speed_min(mddev)) {
7555 if ((currspeed > speed_max(mddev)) ||
7556 !is_mddev_idle(mddev, 0)) {
7562 printk(KERN_INFO "md: %s: %s done.\n",mdname(mddev), desc);
7564 * this also signals 'finished resyncing' to md_stop
7567 blk_finish_plug(&plug);
7568 wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
7570 /* tell personality that we are finished */
7571 mddev->pers->sync_request(mddev, max_sectors, &skipped, 1);
7573 if (!test_bit(MD_RECOVERY_CHECK, &mddev->recovery) &&
7574 mddev->curr_resync > 2) {
7575 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
7576 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
7577 if (mddev->curr_resync >= mddev->recovery_cp) {
7579 "md: checkpointing %s of %s.\n",
7580 desc, mdname(mddev));
7581 mddev->recovery_cp =
7582 mddev->curr_resync_completed;
7585 mddev->recovery_cp = MaxSector;
7587 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery))
7588 mddev->curr_resync = MaxSector;
7590 rdev_for_each_rcu(rdev, mddev)
7591 if (rdev->raid_disk >= 0 &&
7592 mddev->delta_disks >= 0 &&
7593 !test_bit(Faulty, &rdev->flags) &&
7594 !test_bit(In_sync, &rdev->flags) &&
7595 rdev->recovery_offset < mddev->curr_resync)
7596 rdev->recovery_offset = mddev->curr_resync;
7601 set_bit(MD_CHANGE_DEVS, &mddev->flags);
7603 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
7604 /* We completed so min/max setting can be forgotten if used. */
7605 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
7606 mddev->resync_min = 0;
7607 mddev->resync_max = MaxSector;
7608 } else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
7609 mddev->resync_min = mddev->curr_resync_completed;
7610 mddev->curr_resync = 0;
7611 wake_up(&resync_wait);
7612 set_bit(MD_RECOVERY_DONE, &mddev->recovery);
7613 md_wakeup_thread(mddev->thread);
7618 * got a signal, exit.
7621 "md: md_do_sync() got signal ... exiting\n");
7622 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
7626 EXPORT_SYMBOL_GPL(md_do_sync);
7628 static int remove_and_add_spares(struct mddev *mddev)
7630 struct md_rdev *rdev;
7634 mddev->curr_resync_completed = 0;
7636 rdev_for_each(rdev, mddev)
7637 if (rdev->raid_disk >= 0 &&
7638 !test_bit(Blocked, &rdev->flags) &&
7639 (test_bit(Faulty, &rdev->flags) ||
7640 ! test_bit(In_sync, &rdev->flags)) &&
7641 atomic_read(&rdev->nr_pending)==0) {
7642 if (mddev->pers->hot_remove_disk(
7643 mddev, rdev) == 0) {
7644 sysfs_unlink_rdev(mddev, rdev);
7645 rdev->raid_disk = -1;
7650 sysfs_notify(&mddev->kobj, NULL,
7654 rdev_for_each(rdev, mddev) {
7655 if (rdev->raid_disk >= 0 &&
7656 !test_bit(In_sync, &rdev->flags) &&
7657 !test_bit(Faulty, &rdev->flags))
7659 if (rdev->raid_disk < 0
7660 && !test_bit(Faulty, &rdev->flags)) {
7661 rdev->recovery_offset = 0;
7663 hot_add_disk(mddev, rdev) == 0) {
7664 if (sysfs_link_rdev(mddev, rdev))
7665 /* failure here is OK */;
7667 md_new_event(mddev);
7668 set_bit(MD_CHANGE_DEVS, &mddev->flags);
7675 static void reap_sync_thread(struct mddev *mddev)
7677 struct md_rdev *rdev;
7679 /* resync has finished, collect result */
7680 md_unregister_thread(&mddev->sync_thread);
7681 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
7682 !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
7684 /* activate any spares */
7685 if (mddev->pers->spare_active(mddev))
7686 sysfs_notify(&mddev->kobj, NULL,
7689 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
7690 mddev->pers->finish_reshape)
7691 mddev->pers->finish_reshape(mddev);
7693 /* If array is no-longer degraded, then any saved_raid_disk
7694 * information must be scrapped. Also if any device is now
7695 * In_sync we must scrape the saved_raid_disk for that device
7696 * do the superblock for an incrementally recovered device
7699 rdev_for_each(rdev, mddev)
7700 if (!mddev->degraded ||
7701 test_bit(In_sync, &rdev->flags))
7702 rdev->saved_raid_disk = -1;
7704 md_update_sb(mddev, 1);
7705 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
7706 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
7707 clear_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
7708 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
7709 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
7710 /* flag recovery needed just to double check */
7711 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7712 sysfs_notify_dirent_safe(mddev->sysfs_action);
7713 md_new_event(mddev);
7714 if (mddev->event_work.func)
7715 queue_work(md_misc_wq, &mddev->event_work);
7719 * This routine is regularly called by all per-raid-array threads to
7720 * deal with generic issues like resync and super-block update.
7721 * Raid personalities that don't have a thread (linear/raid0) do not
7722 * need this as they never do any recovery or update the superblock.
7724 * It does not do any resync itself, but rather "forks" off other threads
7725 * to do that as needed.
7726 * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
7727 * "->recovery" and create a thread at ->sync_thread.
7728 * When the thread finishes it sets MD_RECOVERY_DONE
7729 * and wakeups up this thread which will reap the thread and finish up.
7730 * This thread also removes any faulty devices (with nr_pending == 0).
7732 * The overall approach is:
7733 * 1/ if the superblock needs updating, update it.
7734 * 2/ If a recovery thread is running, don't do anything else.
7735 * 3/ If recovery has finished, clean up, possibly marking spares active.
7736 * 4/ If there are any faulty devices, remove them.
7737 * 5/ If array is degraded, try to add spares devices
7738 * 6/ If array has spares or is not in-sync, start a resync thread.
7740 void md_check_recovery(struct mddev *mddev)
7742 if (mddev->suspended)
7746 bitmap_daemon_work(mddev);
7748 if (signal_pending(current)) {
7749 if (mddev->pers->sync_request && !mddev->external) {
7750 printk(KERN_INFO "md: %s in immediate safe mode\n",
7752 mddev->safemode = 2;
7754 flush_signals(current);
7757 if (mddev->ro && !test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
7760 (mddev->flags & ~ (1<<MD_CHANGE_PENDING)) ||
7761 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
7762 test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
7763 (mddev->external == 0 && mddev->safemode == 1) ||
7764 (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending)
7765 && !mddev->in_sync && mddev->recovery_cp == MaxSector)
7769 if (mddev_trylock(mddev)) {
7773 /* Only thing we do on a ro array is remove
7776 struct md_rdev *rdev;
7777 rdev_for_each(rdev, mddev)
7778 if (rdev->raid_disk >= 0 &&
7779 !test_bit(Blocked, &rdev->flags) &&
7780 test_bit(Faulty, &rdev->flags) &&
7781 atomic_read(&rdev->nr_pending)==0) {
7782 if (mddev->pers->hot_remove_disk(
7783 mddev, rdev) == 0) {
7784 sysfs_unlink_rdev(mddev, rdev);
7785 rdev->raid_disk = -1;
7788 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7792 if (!mddev->external) {
7794 spin_lock_irq(&mddev->write_lock);
7795 if (mddev->safemode &&
7796 !atomic_read(&mddev->writes_pending) &&
7798 mddev->recovery_cp == MaxSector) {
7801 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
7803 if (mddev->safemode == 1)
7804 mddev->safemode = 0;
7805 spin_unlock_irq(&mddev->write_lock);
7807 sysfs_notify_dirent_safe(mddev->sysfs_state);
7811 md_update_sb(mddev, 0);
7813 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
7814 !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
7815 /* resync/recovery still happening */
7816 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7819 if (mddev->sync_thread) {
7820 reap_sync_thread(mddev);
7823 /* Set RUNNING before clearing NEEDED to avoid
7824 * any transients in the value of "sync_action".
7826 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
7827 /* Clear some bits that don't mean anything, but
7830 clear_bit(MD_RECOVERY_INTR, &mddev->recovery);
7831 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
7833 if (!test_and_clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
7834 test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
7836 /* no recovery is running.
7837 * remove any failed drives, then
7838 * add spares if possible.
7839 * Spare are also removed and re-added, to allow
7840 * the personality to fail the re-add.
7843 if (mddev->reshape_position != MaxSector) {
7844 if (mddev->pers->check_reshape == NULL ||
7845 mddev->pers->check_reshape(mddev) != 0)
7846 /* Cannot proceed */
7848 set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
7849 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
7850 } else if ((spares = remove_and_add_spares(mddev))) {
7851 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
7852 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
7853 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
7854 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
7855 } else if (mddev->recovery_cp < MaxSector) {
7856 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
7857 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
7858 } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
7859 /* nothing to be done ... */
7862 if (mddev->pers->sync_request) {
7864 /* We are adding a device or devices to an array
7865 * which has the bitmap stored on all devices.
7866 * So make sure all bitmap pages get written
7868 bitmap_write_all(mddev->bitmap);
7870 mddev->sync_thread = md_register_thread(md_do_sync,
7873 if (!mddev->sync_thread) {
7874 printk(KERN_ERR "%s: could not start resync"
7877 /* leave the spares where they are, it shouldn't hurt */
7878 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
7879 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
7880 clear_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
7881 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
7882 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
7884 md_wakeup_thread(mddev->sync_thread);
7885 sysfs_notify_dirent_safe(mddev->sysfs_action);
7886 md_new_event(mddev);
7889 if (!mddev->sync_thread) {
7890 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
7891 if (test_and_clear_bit(MD_RECOVERY_RECOVER,
7893 if (mddev->sysfs_action)
7894 sysfs_notify_dirent_safe(mddev->sysfs_action);
7896 mddev_unlock(mddev);
7900 void md_wait_for_blocked_rdev(struct md_rdev *rdev, struct mddev *mddev)
7902 sysfs_notify_dirent_safe(rdev->sysfs_state);
7903 wait_event_timeout(rdev->blocked_wait,
7904 !test_bit(Blocked, &rdev->flags) &&
7905 !test_bit(BlockedBadBlocks, &rdev->flags),
7906 msecs_to_jiffies(5000));
7907 rdev_dec_pending(rdev, mddev);
7909 EXPORT_SYMBOL(md_wait_for_blocked_rdev);
7911 void md_finish_reshape(struct mddev *mddev)
7913 /* called be personality module when reshape completes. */
7914 struct md_rdev *rdev;
7916 rdev_for_each(rdev, mddev) {
7917 if (rdev->data_offset > rdev->new_data_offset)
7918 rdev->sectors += rdev->data_offset - rdev->new_data_offset;
7920 rdev->sectors -= rdev->new_data_offset - rdev->data_offset;
7921 rdev->data_offset = rdev->new_data_offset;
7924 EXPORT_SYMBOL(md_finish_reshape);
7926 /* Bad block management.
7927 * We can record which blocks on each device are 'bad' and so just
7928 * fail those blocks, or that stripe, rather than the whole device.
7929 * Entries in the bad-block table are 64bits wide. This comprises:
7930 * Length of bad-range, in sectors: 0-511 for lengths 1-512
7931 * Start of bad-range, sector offset, 54 bits (allows 8 exbibytes)
7932 * A 'shift' can be set so that larger blocks are tracked and
7933 * consequently larger devices can be covered.
7934 * 'Acknowledged' flag - 1 bit. - the most significant bit.
7936 * Locking of the bad-block table uses a seqlock so md_is_badblock
7937 * might need to retry if it is very unlucky.
7938 * We will sometimes want to check for bad blocks in a bi_end_io function,
7939 * so we use the write_seqlock_irq variant.
7941 * When looking for a bad block we specify a range and want to
7942 * know if any block in the range is bad. So we binary-search
7943 * to the last range that starts at-or-before the given endpoint,
7944 * (or "before the sector after the target range")
7945 * then see if it ends after the given start.
7947 * 0 if there are no known bad blocks in the range
7948 * 1 if there are known bad block which are all acknowledged
7949 * -1 if there are bad blocks which have not yet been acknowledged in metadata.
7950 * plus the start/length of the first bad section we overlap.
7952 int md_is_badblock(struct badblocks *bb, sector_t s, int sectors,
7953 sector_t *first_bad, int *bad_sectors)
7959 sector_t target = s + sectors;
7962 if (bb->shift > 0) {
7963 /* round the start down, and the end up */
7965 target += (1<<bb->shift) - 1;
7966 target >>= bb->shift;
7967 sectors = target - s;
7969 /* 'target' is now the first block after the bad range */
7972 seq = read_seqbegin(&bb->lock);
7976 /* Binary search between lo and hi for 'target'
7977 * i.e. for the last range that starts before 'target'
7979 /* INVARIANT: ranges before 'lo' and at-or-after 'hi'
7980 * are known not to be the last range before target.
7981 * VARIANT: hi-lo is the number of possible
7982 * ranges, and decreases until it reaches 1
7984 while (hi - lo > 1) {
7985 int mid = (lo + hi) / 2;
7986 sector_t a = BB_OFFSET(p[mid]);
7988 /* This could still be the one, earlier ranges
7992 /* This and later ranges are definitely out. */
7995 /* 'lo' might be the last that started before target, but 'hi' isn't */
7997 /* need to check all range that end after 's' to see if
7998 * any are unacknowledged.
8001 BB_OFFSET(p[lo]) + BB_LEN(p[lo]) > s) {
8002 if (BB_OFFSET(p[lo]) < target) {
8003 /* starts before the end, and finishes after
8004 * the start, so they must overlap
8006 if (rv != -1 && BB_ACK(p[lo]))
8010 *first_bad = BB_OFFSET(p[lo]);
8011 *bad_sectors = BB_LEN(p[lo]);
8017 if (read_seqretry(&bb->lock, seq))
8022 EXPORT_SYMBOL_GPL(md_is_badblock);
8025 * Add a range of bad blocks to the table.
8026 * This might extend the table, or might contract it
8027 * if two adjacent ranges can be merged.
8028 * We binary-search to find the 'insertion' point, then
8029 * decide how best to handle it.
8031 static int md_set_badblocks(struct badblocks *bb, sector_t s, int sectors,
8039 /* badblocks are disabled */
8043 /* round the start down, and the end up */
8044 sector_t next = s + sectors;
8046 next += (1<<bb->shift) - 1;
8051 write_seqlock_irq(&bb->lock);
8056 /* Find the last range that starts at-or-before 's' */
8057 while (hi - lo > 1) {
8058 int mid = (lo + hi) / 2;
8059 sector_t a = BB_OFFSET(p[mid]);
8065 if (hi > lo && BB_OFFSET(p[lo]) > s)
8069 /* we found a range that might merge with the start
8072 sector_t a = BB_OFFSET(p[lo]);
8073 sector_t e = a + BB_LEN(p[lo]);
8074 int ack = BB_ACK(p[lo]);
8076 /* Yes, we can merge with a previous range */
8077 if (s == a && s + sectors >= e)
8078 /* new range covers old */
8081 ack = ack && acknowledged;
8083 if (e < s + sectors)
8085 if (e - a <= BB_MAX_LEN) {
8086 p[lo] = BB_MAKE(a, e-a, ack);
8089 /* does not all fit in one range,
8090 * make p[lo] maximal
8092 if (BB_LEN(p[lo]) != BB_MAX_LEN)
8093 p[lo] = BB_MAKE(a, BB_MAX_LEN, ack);
8099 if (sectors && hi < bb->count) {
8100 /* 'hi' points to the first range that starts after 's'.
8101 * Maybe we can merge with the start of that range */
8102 sector_t a = BB_OFFSET(p[hi]);
8103 sector_t e = a + BB_LEN(p[hi]);
8104 int ack = BB_ACK(p[hi]);
8105 if (a <= s + sectors) {
8106 /* merging is possible */
8107 if (e <= s + sectors) {
8112 ack = ack && acknowledged;
8115 if (e - a <= BB_MAX_LEN) {
8116 p[hi] = BB_MAKE(a, e-a, ack);
8119 p[hi] = BB_MAKE(a, BB_MAX_LEN, ack);
8127 if (sectors == 0 && hi < bb->count) {
8128 /* we might be able to combine lo and hi */
8129 /* Note: 's' is at the end of 'lo' */
8130 sector_t a = BB_OFFSET(p[hi]);
8131 int lolen = BB_LEN(p[lo]);
8132 int hilen = BB_LEN(p[hi]);
8133 int newlen = lolen + hilen - (s - a);
8134 if (s >= a && newlen < BB_MAX_LEN) {
8135 /* yes, we can combine them */
8136 int ack = BB_ACK(p[lo]) && BB_ACK(p[hi]);
8137 p[lo] = BB_MAKE(BB_OFFSET(p[lo]), newlen, ack);
8138 memmove(p + hi, p + hi + 1,
8139 (bb->count - hi - 1) * 8);
8144 /* didn't merge (it all).
8145 * Need to add a range just before 'hi' */
8146 if (bb->count >= MD_MAX_BADBLOCKS) {
8147 /* No room for more */
8151 int this_sectors = sectors;
8152 memmove(p + hi + 1, p + hi,
8153 (bb->count - hi) * 8);
8156 if (this_sectors > BB_MAX_LEN)
8157 this_sectors = BB_MAX_LEN;
8158 p[hi] = BB_MAKE(s, this_sectors, acknowledged);
8159 sectors -= this_sectors;
8166 bb->unacked_exist = 1;
8167 write_sequnlock_irq(&bb->lock);
8172 int rdev_set_badblocks(struct md_rdev *rdev, sector_t s, int sectors,
8177 s += rdev->new_data_offset;
8179 s += rdev->data_offset;
8180 rv = md_set_badblocks(&rdev->badblocks,
8183 /* Make sure they get written out promptly */
8184 sysfs_notify_dirent_safe(rdev->sysfs_state);
8185 set_bit(MD_CHANGE_CLEAN, &rdev->mddev->flags);
8186 md_wakeup_thread(rdev->mddev->thread);
8190 EXPORT_SYMBOL_GPL(rdev_set_badblocks);
8193 * Remove a range of bad blocks from the table.
8194 * This may involve extending the table if we spilt a region,
8195 * but it must not fail. So if the table becomes full, we just
8196 * drop the remove request.
8198 static int md_clear_badblocks(struct badblocks *bb, sector_t s, int sectors)
8202 sector_t target = s + sectors;
8205 if (bb->shift > 0) {
8206 /* When clearing we round the start up and the end down.
8207 * This should not matter as the shift should align with
8208 * the block size and no rounding should ever be needed.
8209 * However it is better the think a block is bad when it
8210 * isn't than to think a block is not bad when it is.
8212 s += (1<<bb->shift) - 1;
8214 target >>= bb->shift;
8215 sectors = target - s;
8218 write_seqlock_irq(&bb->lock);
8223 /* Find the last range that starts before 'target' */
8224 while (hi - lo > 1) {
8225 int mid = (lo + hi) / 2;
8226 sector_t a = BB_OFFSET(p[mid]);
8233 /* p[lo] is the last range that could overlap the
8234 * current range. Earlier ranges could also overlap,
8235 * but only this one can overlap the end of the range.
8237 if (BB_OFFSET(p[lo]) + BB_LEN(p[lo]) > target) {
8238 /* Partial overlap, leave the tail of this range */
8239 int ack = BB_ACK(p[lo]);
8240 sector_t a = BB_OFFSET(p[lo]);
8241 sector_t end = a + BB_LEN(p[lo]);
8244 /* we need to split this range */
8245 if (bb->count >= MD_MAX_BADBLOCKS) {
8249 memmove(p+lo+1, p+lo, (bb->count - lo) * 8);
8251 p[lo] = BB_MAKE(a, s-a, ack);
8254 p[lo] = BB_MAKE(target, end - target, ack);
8255 /* there is no longer an overlap */
8260 BB_OFFSET(p[lo]) + BB_LEN(p[lo]) > s) {
8261 /* This range does overlap */
8262 if (BB_OFFSET(p[lo]) < s) {
8263 /* Keep the early parts of this range. */
8264 int ack = BB_ACK(p[lo]);
8265 sector_t start = BB_OFFSET(p[lo]);
8266 p[lo] = BB_MAKE(start, s - start, ack);
8267 /* now low doesn't overlap, so.. */
8272 /* 'lo' is strictly before, 'hi' is strictly after,
8273 * anything between needs to be discarded
8276 memmove(p+lo+1, p+hi, (bb->count - hi) * 8);
8277 bb->count -= (hi - lo - 1);
8283 write_sequnlock_irq(&bb->lock);
8287 int rdev_clear_badblocks(struct md_rdev *rdev, sector_t s, int sectors,
8291 s += rdev->new_data_offset;
8293 s += rdev->data_offset;
8294 return md_clear_badblocks(&rdev->badblocks,
8297 EXPORT_SYMBOL_GPL(rdev_clear_badblocks);
8300 * Acknowledge all bad blocks in a list.
8301 * This only succeeds if ->changed is clear. It is used by
8302 * in-kernel metadata updates
8304 void md_ack_all_badblocks(struct badblocks *bb)
8306 if (bb->page == NULL || bb->changed)
8307 /* no point even trying */
8309 write_seqlock_irq(&bb->lock);
8311 if (bb->changed == 0 && bb->unacked_exist) {
8314 for (i = 0; i < bb->count ; i++) {
8315 if (!BB_ACK(p[i])) {
8316 sector_t start = BB_OFFSET(p[i]);
8317 int len = BB_LEN(p[i]);
8318 p[i] = BB_MAKE(start, len, 1);
8321 bb->unacked_exist = 0;
8323 write_sequnlock_irq(&bb->lock);
8325 EXPORT_SYMBOL_GPL(md_ack_all_badblocks);
8327 /* sysfs access to bad-blocks list.
8328 * We present two files.
8329 * 'bad-blocks' lists sector numbers and lengths of ranges that
8330 * are recorded as bad. The list is truncated to fit within
8331 * the one-page limit of sysfs.
8332 * Writing "sector length" to this file adds an acknowledged
8334 * 'unacknowledged-bad-blocks' lists bad blocks that have not yet
8335 * been acknowledged. Writing to this file adds bad blocks
8336 * without acknowledging them. This is largely for testing.
8340 badblocks_show(struct badblocks *bb, char *page, int unack)
8351 seq = read_seqbegin(&bb->lock);
8356 while (len < PAGE_SIZE && i < bb->count) {
8357 sector_t s = BB_OFFSET(p[i]);
8358 unsigned int length = BB_LEN(p[i]);
8359 int ack = BB_ACK(p[i]);
8365 len += snprintf(page+len, PAGE_SIZE-len, "%llu %u\n",
8366 (unsigned long long)s << bb->shift,
8367 length << bb->shift);
8369 if (unack && len == 0)
8370 bb->unacked_exist = 0;
8372 if (read_seqretry(&bb->lock, seq))
8381 badblocks_store(struct badblocks *bb, const char *page, size_t len, int unack)
8383 unsigned long long sector;
8387 /* Allow clearing via sysfs *only* for testing/debugging.
8388 * Normally only a successful write may clear a badblock
8391 if (page[0] == '-') {
8395 #endif /* DO_DEBUG */
8397 switch (sscanf(page, "%llu %d%c", §or, &length, &newline)) {
8399 if (newline != '\n')
8411 md_clear_badblocks(bb, sector, length);
8414 #endif /* DO_DEBUG */
8415 if (md_set_badblocks(bb, sector, length, !unack))
8421 static int md_notify_reboot(struct notifier_block *this,
8422 unsigned long code, void *x)
8424 struct list_head *tmp;
8425 struct mddev *mddev;
8428 for_each_mddev(mddev, tmp) {
8429 if (mddev_trylock(mddev)) {
8431 __md_stop_writes(mddev);
8432 mddev->safemode = 2;
8433 mddev_unlock(mddev);
8438 * certain more exotic SCSI devices are known to be
8439 * volatile wrt too early system reboots. While the
8440 * right place to handle this issue is the given
8441 * driver, we do want to have a safe RAID driver ...
8449 static struct notifier_block md_notifier = {
8450 .notifier_call = md_notify_reboot,
8452 .priority = INT_MAX, /* before any real devices */
8455 static void md_geninit(void)
8457 pr_debug("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
8459 proc_create("mdstat", S_IRUGO, NULL, &md_seq_fops);
8462 static int __init md_init(void)
8466 md_wq = alloc_workqueue("md", WQ_MEM_RECLAIM, 0);
8470 md_misc_wq = alloc_workqueue("md_misc", 0, 0);
8474 if ((ret = register_blkdev(MD_MAJOR, "md")) < 0)
8477 if ((ret = register_blkdev(0, "mdp")) < 0)
8481 blk_register_region(MKDEV(MD_MAJOR, 0), 1UL<<MINORBITS, THIS_MODULE,
8482 md_probe, NULL, NULL);
8483 blk_register_region(MKDEV(mdp_major, 0), 1UL<<MINORBITS, THIS_MODULE,
8484 md_probe, NULL, NULL);
8486 register_reboot_notifier(&md_notifier);
8487 raid_table_header = register_sysctl_table(raid_root_table);
8493 unregister_blkdev(MD_MAJOR, "md");
8495 destroy_workqueue(md_misc_wq);
8497 destroy_workqueue(md_wq);
8505 * Searches all registered partitions for autorun RAID arrays
8509 static LIST_HEAD(all_detected_devices);
8510 struct detected_devices_node {
8511 struct list_head list;
8515 void md_autodetect_dev(dev_t dev)
8517 struct detected_devices_node *node_detected_dev;
8519 node_detected_dev = kzalloc(sizeof(*node_detected_dev), GFP_KERNEL);
8520 if (node_detected_dev) {
8521 node_detected_dev->dev = dev;
8522 list_add_tail(&node_detected_dev->list, &all_detected_devices);
8524 printk(KERN_CRIT "md: md_autodetect_dev: kzalloc failed"
8525 ", skipping dev(%d,%d)\n", MAJOR(dev), MINOR(dev));
8530 static void autostart_arrays(int part)
8532 struct md_rdev *rdev;
8533 struct detected_devices_node *node_detected_dev;
8535 int i_scanned, i_passed;
8540 printk(KERN_INFO "md: Autodetecting RAID arrays.\n");
8542 while (!list_empty(&all_detected_devices) && i_scanned < INT_MAX) {
8544 node_detected_dev = list_entry(all_detected_devices.next,
8545 struct detected_devices_node, list);
8546 list_del(&node_detected_dev->list);
8547 dev = node_detected_dev->dev;
8548 kfree(node_detected_dev);
8549 rdev = md_import_device(dev,0, 90);
8553 if (test_bit(Faulty, &rdev->flags)) {
8557 set_bit(AutoDetected, &rdev->flags);
8558 list_add(&rdev->same_set, &pending_raid_disks);
8562 printk(KERN_INFO "md: Scanned %d and added %d devices.\n",
8563 i_scanned, i_passed);
8565 autorun_devices(part);
8568 #endif /* !MODULE */
8570 static __exit void md_exit(void)
8572 struct mddev *mddev;
8573 struct list_head *tmp;
8575 blk_unregister_region(MKDEV(MD_MAJOR,0), 1U << MINORBITS);
8576 blk_unregister_region(MKDEV(mdp_major,0), 1U << MINORBITS);
8578 unregister_blkdev(MD_MAJOR,"md");
8579 unregister_blkdev(mdp_major, "mdp");
8580 unregister_reboot_notifier(&md_notifier);
8581 unregister_sysctl_table(raid_table_header);
8582 remove_proc_entry("mdstat", NULL);
8583 for_each_mddev(mddev, tmp) {
8584 export_array(mddev);
8585 mddev->hold_active = 0;
8587 destroy_workqueue(md_misc_wq);
8588 destroy_workqueue(md_wq);
8591 subsys_initcall(md_init);
8592 module_exit(md_exit)
8594 static int get_ro(char *buffer, struct kernel_param *kp)
8596 return sprintf(buffer, "%d", start_readonly);
8598 static int set_ro(const char *val, struct kernel_param *kp)
8601 int num = simple_strtoul(val, &e, 10);
8602 if (*val && (*e == '\0' || *e == '\n')) {
8603 start_readonly = num;
8609 module_param_call(start_ro, set_ro, get_ro, NULL, S_IRUSR|S_IWUSR);
8610 module_param(start_dirty_degraded, int, S_IRUGO|S_IWUSR);
8612 module_param_call(new_array, add_named_array, NULL, NULL, S_IWUSR);
8614 EXPORT_SYMBOL(register_md_personality);
8615 EXPORT_SYMBOL(unregister_md_personality);
8616 EXPORT_SYMBOL(md_error);
8617 EXPORT_SYMBOL(md_done_sync);
8618 EXPORT_SYMBOL(md_write_start);
8619 EXPORT_SYMBOL(md_write_end);
8620 EXPORT_SYMBOL(md_register_thread);
8621 EXPORT_SYMBOL(md_unregister_thread);
8622 EXPORT_SYMBOL(md_wakeup_thread);
8623 EXPORT_SYMBOL(md_check_recovery);
8624 MODULE_LICENSE("GPL");
8625 MODULE_DESCRIPTION("MD RAID framework");
8627 MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR);