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 md_wakeup_thread(mddev->thread);
406 md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
408 EXPORT_SYMBOL_GPL(mddev_resume);
410 int mddev_congested(struct mddev *mddev, int bits)
412 return mddev->suspended;
414 EXPORT_SYMBOL(mddev_congested);
417 * Generic flush handling for md
420 static void md_end_flush(struct bio *bio, int err)
422 struct md_rdev *rdev = bio->bi_private;
423 struct mddev *mddev = rdev->mddev;
425 rdev_dec_pending(rdev, mddev);
427 if (atomic_dec_and_test(&mddev->flush_pending)) {
428 /* The pre-request flush has finished */
429 queue_work(md_wq, &mddev->flush_work);
434 static void md_submit_flush_data(struct work_struct *ws);
436 static void submit_flushes(struct work_struct *ws)
438 struct mddev *mddev = container_of(ws, struct mddev, flush_work);
439 struct md_rdev *rdev;
441 INIT_WORK(&mddev->flush_work, md_submit_flush_data);
442 atomic_set(&mddev->flush_pending, 1);
444 rdev_for_each_rcu(rdev, mddev)
445 if (rdev->raid_disk >= 0 &&
446 !test_bit(Faulty, &rdev->flags)) {
447 /* Take two references, one is dropped
448 * when request finishes, one after
449 * we reclaim rcu_read_lock
452 atomic_inc(&rdev->nr_pending);
453 atomic_inc(&rdev->nr_pending);
455 bi = bio_alloc_mddev(GFP_KERNEL, 0, mddev);
456 bi->bi_end_io = md_end_flush;
457 bi->bi_private = rdev;
458 bi->bi_bdev = rdev->bdev;
459 atomic_inc(&mddev->flush_pending);
460 submit_bio(WRITE_FLUSH, bi);
462 rdev_dec_pending(rdev, mddev);
465 if (atomic_dec_and_test(&mddev->flush_pending))
466 queue_work(md_wq, &mddev->flush_work);
469 static void md_submit_flush_data(struct work_struct *ws)
471 struct mddev *mddev = container_of(ws, struct mddev, flush_work);
472 struct bio *bio = mddev->flush_bio;
474 if (bio->bi_size == 0)
475 /* an empty barrier - all done */
478 bio->bi_rw &= ~REQ_FLUSH;
479 mddev->pers->make_request(mddev, bio);
482 mddev->flush_bio = NULL;
483 wake_up(&mddev->sb_wait);
486 void md_flush_request(struct mddev *mddev, struct bio *bio)
488 spin_lock_irq(&mddev->write_lock);
489 wait_event_lock_irq(mddev->sb_wait,
491 mddev->write_lock, /*nothing*/);
492 mddev->flush_bio = bio;
493 spin_unlock_irq(&mddev->write_lock);
495 INIT_WORK(&mddev->flush_work, submit_flushes);
496 queue_work(md_wq, &mddev->flush_work);
498 EXPORT_SYMBOL(md_flush_request);
500 /* Support for plugging.
501 * This mirrors the plugging support in request_queue, but does not
502 * require having a whole queue or request structures.
503 * We allocate an md_plug_cb for each md device and each thread it gets
504 * plugged on. This links tot the private plug_handle structure in the
505 * personality data where we keep a count of the number of outstanding
506 * plugs so other code can see if a plug is active.
509 struct blk_plug_cb cb;
513 static void plugger_unplug(struct blk_plug_cb *cb)
515 struct md_plug_cb *mdcb = container_of(cb, struct md_plug_cb, cb);
516 if (atomic_dec_and_test(&mdcb->mddev->plug_cnt))
517 md_wakeup_thread(mdcb->mddev->thread);
521 /* Check that an unplug wakeup will come shortly.
522 * If not, wakeup the md thread immediately
524 int mddev_check_plugged(struct mddev *mddev)
526 struct blk_plug *plug = current->plug;
527 struct md_plug_cb *mdcb;
532 list_for_each_entry(mdcb, &plug->cb_list, cb.list) {
533 if (mdcb->cb.callback == plugger_unplug &&
534 mdcb->mddev == mddev) {
535 /* Already on the list, move to top */
536 if (mdcb != list_first_entry(&plug->cb_list,
539 list_move(&mdcb->cb.list, &plug->cb_list);
543 /* Not currently on the callback list */
544 mdcb = kmalloc(sizeof(*mdcb), GFP_ATOMIC);
549 mdcb->cb.callback = plugger_unplug;
550 atomic_inc(&mddev->plug_cnt);
551 list_add(&mdcb->cb.list, &plug->cb_list);
554 EXPORT_SYMBOL_GPL(mddev_check_plugged);
556 static inline struct mddev *mddev_get(struct mddev *mddev)
558 atomic_inc(&mddev->active);
562 static void mddev_delayed_delete(struct work_struct *ws);
564 static void mddev_put(struct mddev *mddev)
566 struct bio_set *bs = NULL;
568 if (!atomic_dec_and_lock(&mddev->active, &all_mddevs_lock))
570 if (!mddev->raid_disks && list_empty(&mddev->disks) &&
571 mddev->ctime == 0 && !mddev->hold_active) {
572 /* Array is not configured at all, and not held active,
574 list_del_init(&mddev->all_mddevs);
576 mddev->bio_set = NULL;
577 if (mddev->gendisk) {
578 /* We did a probe so need to clean up. Call
579 * queue_work inside the spinlock so that
580 * flush_workqueue() after mddev_find will
581 * succeed in waiting for the work to be done.
583 INIT_WORK(&mddev->del_work, mddev_delayed_delete);
584 queue_work(md_misc_wq, &mddev->del_work);
588 spin_unlock(&all_mddevs_lock);
593 void mddev_init(struct mddev *mddev)
595 mutex_init(&mddev->open_mutex);
596 mutex_init(&mddev->reconfig_mutex);
597 mutex_init(&mddev->bitmap_info.mutex);
598 INIT_LIST_HEAD(&mddev->disks);
599 INIT_LIST_HEAD(&mddev->all_mddevs);
600 init_timer(&mddev->safemode_timer);
601 atomic_set(&mddev->active, 1);
602 atomic_set(&mddev->openers, 0);
603 atomic_set(&mddev->active_io, 0);
604 atomic_set(&mddev->plug_cnt, 0);
605 spin_lock_init(&mddev->write_lock);
606 atomic_set(&mddev->flush_pending, 0);
607 init_waitqueue_head(&mddev->sb_wait);
608 init_waitqueue_head(&mddev->recovery_wait);
609 mddev->reshape_position = MaxSector;
610 mddev->resync_min = 0;
611 mddev->resync_max = MaxSector;
612 mddev->level = LEVEL_NONE;
614 EXPORT_SYMBOL_GPL(mddev_init);
616 static struct mddev * mddev_find(dev_t unit)
618 struct mddev *mddev, *new = NULL;
620 if (unit && MAJOR(unit) != MD_MAJOR)
621 unit &= ~((1<<MdpMinorShift)-1);
624 spin_lock(&all_mddevs_lock);
627 list_for_each_entry(mddev, &all_mddevs, all_mddevs)
628 if (mddev->unit == unit) {
630 spin_unlock(&all_mddevs_lock);
636 list_add(&new->all_mddevs, &all_mddevs);
637 spin_unlock(&all_mddevs_lock);
638 new->hold_active = UNTIL_IOCTL;
642 /* find an unused unit number */
643 static int next_minor = 512;
644 int start = next_minor;
648 dev = MKDEV(MD_MAJOR, next_minor);
650 if (next_minor > MINORMASK)
652 if (next_minor == start) {
653 /* Oh dear, all in use. */
654 spin_unlock(&all_mddevs_lock);
660 list_for_each_entry(mddev, &all_mddevs, all_mddevs)
661 if (mddev->unit == dev) {
667 new->md_minor = MINOR(dev);
668 new->hold_active = UNTIL_STOP;
669 list_add(&new->all_mddevs, &all_mddevs);
670 spin_unlock(&all_mddevs_lock);
673 spin_unlock(&all_mddevs_lock);
675 new = kzalloc(sizeof(*new), GFP_KERNEL);
680 if (MAJOR(unit) == MD_MAJOR)
681 new->md_minor = MINOR(unit);
683 new->md_minor = MINOR(unit) >> MdpMinorShift;
690 static inline int mddev_lock(struct mddev * mddev)
692 return mutex_lock_interruptible(&mddev->reconfig_mutex);
695 static inline int mddev_is_locked(struct mddev *mddev)
697 return mutex_is_locked(&mddev->reconfig_mutex);
700 static inline int mddev_trylock(struct mddev * mddev)
702 return mutex_trylock(&mddev->reconfig_mutex);
705 static struct attribute_group md_redundancy_group;
707 static void mddev_unlock(struct mddev * mddev)
709 if (mddev->to_remove) {
710 /* These cannot be removed under reconfig_mutex as
711 * an access to the files will try to take reconfig_mutex
712 * while holding the file unremovable, which leads to
714 * So hold set sysfs_active while the remove in happeing,
715 * and anything else which might set ->to_remove or my
716 * otherwise change the sysfs namespace will fail with
717 * -EBUSY if sysfs_active is still set.
718 * We set sysfs_active under reconfig_mutex and elsewhere
719 * test it under the same mutex to ensure its correct value
722 struct attribute_group *to_remove = mddev->to_remove;
723 mddev->to_remove = NULL;
724 mddev->sysfs_active = 1;
725 mutex_unlock(&mddev->reconfig_mutex);
727 if (mddev->kobj.sd) {
728 if (to_remove != &md_redundancy_group)
729 sysfs_remove_group(&mddev->kobj, to_remove);
730 if (mddev->pers == NULL ||
731 mddev->pers->sync_request == NULL) {
732 sysfs_remove_group(&mddev->kobj, &md_redundancy_group);
733 if (mddev->sysfs_action)
734 sysfs_put(mddev->sysfs_action);
735 mddev->sysfs_action = NULL;
738 mddev->sysfs_active = 0;
740 mutex_unlock(&mddev->reconfig_mutex);
742 /* As we've dropped the mutex we need a spinlock to
743 * make sure the thread doesn't disappear
745 spin_lock(&pers_lock);
746 md_wakeup_thread(mddev->thread);
747 spin_unlock(&pers_lock);
750 static struct md_rdev * find_rdev_nr(struct mddev *mddev, int nr)
752 struct md_rdev *rdev;
754 rdev_for_each(rdev, mddev)
755 if (rdev->desc_nr == nr)
761 static struct md_rdev * find_rdev(struct mddev * mddev, dev_t dev)
763 struct md_rdev *rdev;
765 rdev_for_each(rdev, mddev)
766 if (rdev->bdev->bd_dev == dev)
772 static struct md_personality *find_pers(int level, char *clevel)
774 struct md_personality *pers;
775 list_for_each_entry(pers, &pers_list, list) {
776 if (level != LEVEL_NONE && pers->level == level)
778 if (strcmp(pers->name, clevel)==0)
784 /* return the offset of the super block in 512byte sectors */
785 static inline sector_t calc_dev_sboffset(struct md_rdev *rdev)
787 sector_t num_sectors = i_size_read(rdev->bdev->bd_inode) / 512;
788 return MD_NEW_SIZE_SECTORS(num_sectors);
791 static int alloc_disk_sb(struct md_rdev * rdev)
796 rdev->sb_page = alloc_page(GFP_KERNEL);
797 if (!rdev->sb_page) {
798 printk(KERN_ALERT "md: out of memory.\n");
805 static void free_disk_sb(struct md_rdev * rdev)
808 put_page(rdev->sb_page);
810 rdev->sb_page = NULL;
815 put_page(rdev->bb_page);
816 rdev->bb_page = NULL;
821 static void super_written(struct bio *bio, int error)
823 struct md_rdev *rdev = bio->bi_private;
824 struct mddev *mddev = rdev->mddev;
826 if (error || !test_bit(BIO_UPTODATE, &bio->bi_flags)) {
827 printk("md: super_written gets error=%d, uptodate=%d\n",
828 error, test_bit(BIO_UPTODATE, &bio->bi_flags));
829 WARN_ON(test_bit(BIO_UPTODATE, &bio->bi_flags));
830 md_error(mddev, rdev);
833 if (atomic_dec_and_test(&mddev->pending_writes))
834 wake_up(&mddev->sb_wait);
838 void md_super_write(struct mddev *mddev, struct md_rdev *rdev,
839 sector_t sector, int size, struct page *page)
841 /* write first size bytes of page to sector of rdev
842 * Increment mddev->pending_writes before returning
843 * and decrement it on completion, waking up sb_wait
844 * if zero is reached.
845 * If an error occurred, call md_error
847 struct bio *bio = bio_alloc_mddev(GFP_NOIO, 1, mddev);
849 bio->bi_bdev = rdev->meta_bdev ? rdev->meta_bdev : rdev->bdev;
850 bio->bi_sector = sector;
851 bio_add_page(bio, page, size, 0);
852 bio->bi_private = rdev;
853 bio->bi_end_io = super_written;
855 atomic_inc(&mddev->pending_writes);
856 submit_bio(WRITE_FLUSH_FUA, bio);
859 void md_super_wait(struct mddev *mddev)
861 /* wait for all superblock writes that were scheduled to complete */
864 prepare_to_wait(&mddev->sb_wait, &wq, TASK_UNINTERRUPTIBLE);
865 if (atomic_read(&mddev->pending_writes)==0)
869 finish_wait(&mddev->sb_wait, &wq);
872 static void bi_complete(struct bio *bio, int error)
874 complete((struct completion*)bio->bi_private);
877 int sync_page_io(struct md_rdev *rdev, sector_t sector, int size,
878 struct page *page, int rw, bool metadata_op)
880 struct bio *bio = bio_alloc_mddev(GFP_NOIO, 1, rdev->mddev);
881 struct completion event;
886 bio->bi_bdev = (metadata_op && rdev->meta_bdev) ?
887 rdev->meta_bdev : rdev->bdev;
889 bio->bi_sector = sector + rdev->sb_start;
891 bio->bi_sector = sector + rdev->data_offset;
892 bio_add_page(bio, page, size, 0);
893 init_completion(&event);
894 bio->bi_private = &event;
895 bio->bi_end_io = bi_complete;
897 wait_for_completion(&event);
899 ret = test_bit(BIO_UPTODATE, &bio->bi_flags);
903 EXPORT_SYMBOL_GPL(sync_page_io);
905 static int read_disk_sb(struct md_rdev * rdev, int size)
907 char b[BDEVNAME_SIZE];
908 if (!rdev->sb_page) {
916 if (!sync_page_io(rdev, 0, size, rdev->sb_page, READ, true))
922 printk(KERN_WARNING "md: disabled device %s, could not read superblock.\n",
923 bdevname(rdev->bdev,b));
927 static int uuid_equal(mdp_super_t *sb1, mdp_super_t *sb2)
929 return sb1->set_uuid0 == sb2->set_uuid0 &&
930 sb1->set_uuid1 == sb2->set_uuid1 &&
931 sb1->set_uuid2 == sb2->set_uuid2 &&
932 sb1->set_uuid3 == sb2->set_uuid3;
935 static int sb_equal(mdp_super_t *sb1, mdp_super_t *sb2)
938 mdp_super_t *tmp1, *tmp2;
940 tmp1 = kmalloc(sizeof(*tmp1),GFP_KERNEL);
941 tmp2 = kmalloc(sizeof(*tmp2),GFP_KERNEL);
943 if (!tmp1 || !tmp2) {
945 printk(KERN_INFO "md.c sb_equal(): failed to allocate memory!\n");
953 * nr_disks is not constant
958 ret = (memcmp(tmp1, tmp2, MD_SB_GENERIC_CONSTANT_WORDS * 4) == 0);
966 static u32 md_csum_fold(u32 csum)
968 csum = (csum & 0xffff) + (csum >> 16);
969 return (csum & 0xffff) + (csum >> 16);
972 static unsigned int calc_sb_csum(mdp_super_t * sb)
975 u32 *sb32 = (u32*)sb;
977 unsigned int disk_csum, csum;
979 disk_csum = sb->sb_csum;
982 for (i = 0; i < MD_SB_BYTES/4 ; i++)
984 csum = (newcsum & 0xffffffff) + (newcsum>>32);
988 /* This used to use csum_partial, which was wrong for several
989 * reasons including that different results are returned on
990 * different architectures. It isn't critical that we get exactly
991 * the same return value as before (we always csum_fold before
992 * testing, and that removes any differences). However as we
993 * know that csum_partial always returned a 16bit value on
994 * alphas, do a fold to maximise conformity to previous behaviour.
996 sb->sb_csum = md_csum_fold(disk_csum);
998 sb->sb_csum = disk_csum;
1005 * Handle superblock details.
1006 * We want to be able to handle multiple superblock formats
1007 * so we have a common interface to them all, and an array of
1008 * different handlers.
1009 * We rely on user-space to write the initial superblock, and support
1010 * reading and updating of superblocks.
1011 * Interface methods are:
1012 * int load_super(struct md_rdev *dev, struct md_rdev *refdev, int minor_version)
1013 * loads and validates a superblock on dev.
1014 * if refdev != NULL, compare superblocks on both devices
1016 * 0 - dev has a superblock that is compatible with refdev
1017 * 1 - dev has a superblock that is compatible and newer than refdev
1018 * so dev should be used as the refdev in future
1019 * -EINVAL superblock incompatible or invalid
1020 * -othererror e.g. -EIO
1022 * int validate_super(struct mddev *mddev, struct md_rdev *dev)
1023 * Verify that dev is acceptable into mddev.
1024 * The first time, mddev->raid_disks will be 0, and data from
1025 * dev should be merged in. Subsequent calls check that dev
1026 * is new enough. Return 0 or -EINVAL
1028 * void sync_super(struct mddev *mddev, struct md_rdev *dev)
1029 * Update the superblock for rdev with data in mddev
1030 * This does not write to disc.
1036 struct module *owner;
1037 int (*load_super)(struct md_rdev *rdev, struct md_rdev *refdev,
1039 int (*validate_super)(struct mddev *mddev, struct md_rdev *rdev);
1040 void (*sync_super)(struct mddev *mddev, struct md_rdev *rdev);
1041 unsigned long long (*rdev_size_change)(struct md_rdev *rdev,
1042 sector_t num_sectors);
1046 * Check that the given mddev has no bitmap.
1048 * This function is called from the run method of all personalities that do not
1049 * support bitmaps. It prints an error message and returns non-zero if mddev
1050 * has a bitmap. Otherwise, it returns 0.
1053 int md_check_no_bitmap(struct mddev *mddev)
1055 if (!mddev->bitmap_info.file && !mddev->bitmap_info.offset)
1057 printk(KERN_ERR "%s: bitmaps are not supported for %s\n",
1058 mdname(mddev), mddev->pers->name);
1061 EXPORT_SYMBOL(md_check_no_bitmap);
1064 * load_super for 0.90.0
1066 static int super_90_load(struct md_rdev *rdev, struct md_rdev *refdev, int minor_version)
1068 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
1073 * Calculate the position of the superblock (512byte sectors),
1074 * it's at the end of the disk.
1076 * It also happens to be a multiple of 4Kb.
1078 rdev->sb_start = calc_dev_sboffset(rdev);
1080 ret = read_disk_sb(rdev, MD_SB_BYTES);
1081 if (ret) return ret;
1085 bdevname(rdev->bdev, b);
1086 sb = page_address(rdev->sb_page);
1088 if (sb->md_magic != MD_SB_MAGIC) {
1089 printk(KERN_ERR "md: invalid raid superblock magic on %s\n",
1094 if (sb->major_version != 0 ||
1095 sb->minor_version < 90 ||
1096 sb->minor_version > 91) {
1097 printk(KERN_WARNING "Bad version number %d.%d on %s\n",
1098 sb->major_version, sb->minor_version,
1103 if (sb->raid_disks <= 0)
1106 if (md_csum_fold(calc_sb_csum(sb)) != md_csum_fold(sb->sb_csum)) {
1107 printk(KERN_WARNING "md: invalid superblock checksum on %s\n",
1112 rdev->preferred_minor = sb->md_minor;
1113 rdev->data_offset = 0;
1114 rdev->sb_size = MD_SB_BYTES;
1115 rdev->badblocks.shift = -1;
1117 if (sb->level == LEVEL_MULTIPATH)
1120 rdev->desc_nr = sb->this_disk.number;
1126 mdp_super_t *refsb = page_address(refdev->sb_page);
1127 if (!uuid_equal(refsb, sb)) {
1128 printk(KERN_WARNING "md: %s has different UUID to %s\n",
1129 b, bdevname(refdev->bdev,b2));
1132 if (!sb_equal(refsb, sb)) {
1133 printk(KERN_WARNING "md: %s has same UUID"
1134 " but different superblock to %s\n",
1135 b, bdevname(refdev->bdev, b2));
1139 ev2 = md_event(refsb);
1145 rdev->sectors = rdev->sb_start;
1146 /* Limit to 4TB as metadata cannot record more than that */
1147 if (rdev->sectors >= (2ULL << 32))
1148 rdev->sectors = (2ULL << 32) - 2;
1150 if (rdev->sectors < ((sector_t)sb->size) * 2 && sb->level >= 1)
1151 /* "this cannot possibly happen" ... */
1159 * validate_super for 0.90.0
1161 static int super_90_validate(struct mddev *mddev, struct md_rdev *rdev)
1164 mdp_super_t *sb = page_address(rdev->sb_page);
1165 __u64 ev1 = md_event(sb);
1167 rdev->raid_disk = -1;
1168 clear_bit(Faulty, &rdev->flags);
1169 clear_bit(In_sync, &rdev->flags);
1170 clear_bit(WriteMostly, &rdev->flags);
1172 if (mddev->raid_disks == 0) {
1173 mddev->major_version = 0;
1174 mddev->minor_version = sb->minor_version;
1175 mddev->patch_version = sb->patch_version;
1176 mddev->external = 0;
1177 mddev->chunk_sectors = sb->chunk_size >> 9;
1178 mddev->ctime = sb->ctime;
1179 mddev->utime = sb->utime;
1180 mddev->level = sb->level;
1181 mddev->clevel[0] = 0;
1182 mddev->layout = sb->layout;
1183 mddev->raid_disks = sb->raid_disks;
1184 mddev->dev_sectors = ((sector_t)sb->size) * 2;
1185 mddev->events = ev1;
1186 mddev->bitmap_info.offset = 0;
1187 mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9;
1189 if (mddev->minor_version >= 91) {
1190 mddev->reshape_position = sb->reshape_position;
1191 mddev->delta_disks = sb->delta_disks;
1192 mddev->new_level = sb->new_level;
1193 mddev->new_layout = sb->new_layout;
1194 mddev->new_chunk_sectors = sb->new_chunk >> 9;
1196 mddev->reshape_position = MaxSector;
1197 mddev->delta_disks = 0;
1198 mddev->new_level = mddev->level;
1199 mddev->new_layout = mddev->layout;
1200 mddev->new_chunk_sectors = mddev->chunk_sectors;
1203 if (sb->state & (1<<MD_SB_CLEAN))
1204 mddev->recovery_cp = MaxSector;
1206 if (sb->events_hi == sb->cp_events_hi &&
1207 sb->events_lo == sb->cp_events_lo) {
1208 mddev->recovery_cp = sb->recovery_cp;
1210 mddev->recovery_cp = 0;
1213 memcpy(mddev->uuid+0, &sb->set_uuid0, 4);
1214 memcpy(mddev->uuid+4, &sb->set_uuid1, 4);
1215 memcpy(mddev->uuid+8, &sb->set_uuid2, 4);
1216 memcpy(mddev->uuid+12,&sb->set_uuid3, 4);
1218 mddev->max_disks = MD_SB_DISKS;
1220 if (sb->state & (1<<MD_SB_BITMAP_PRESENT) &&
1221 mddev->bitmap_info.file == NULL)
1222 mddev->bitmap_info.offset =
1223 mddev->bitmap_info.default_offset;
1225 } else if (mddev->pers == NULL) {
1226 /* Insist on good event counter while assembling, except
1227 * for spares (which don't need an event count) */
1229 if (sb->disks[rdev->desc_nr].state & (
1230 (1<<MD_DISK_SYNC) | (1 << MD_DISK_ACTIVE)))
1231 if (ev1 < mddev->events)
1233 } else if (mddev->bitmap) {
1234 /* if adding to array with a bitmap, then we can accept an
1235 * older device ... but not too old.
1237 if (ev1 < mddev->bitmap->events_cleared)
1240 if (ev1 < mddev->events)
1241 /* just a hot-add of a new device, leave raid_disk at -1 */
1245 if (mddev->level != LEVEL_MULTIPATH) {
1246 desc = sb->disks + rdev->desc_nr;
1248 if (desc->state & (1<<MD_DISK_FAULTY))
1249 set_bit(Faulty, &rdev->flags);
1250 else if (desc->state & (1<<MD_DISK_SYNC) /* &&
1251 desc->raid_disk < mddev->raid_disks */) {
1252 set_bit(In_sync, &rdev->flags);
1253 rdev->raid_disk = desc->raid_disk;
1254 } else if (desc->state & (1<<MD_DISK_ACTIVE)) {
1255 /* active but not in sync implies recovery up to
1256 * reshape position. We don't know exactly where
1257 * that is, so set to zero for now */
1258 if (mddev->minor_version >= 91) {
1259 rdev->recovery_offset = 0;
1260 rdev->raid_disk = desc->raid_disk;
1263 if (desc->state & (1<<MD_DISK_WRITEMOSTLY))
1264 set_bit(WriteMostly, &rdev->flags);
1265 } else /* MULTIPATH are always insync */
1266 set_bit(In_sync, &rdev->flags);
1271 * sync_super for 0.90.0
1273 static void super_90_sync(struct mddev *mddev, struct md_rdev *rdev)
1276 struct md_rdev *rdev2;
1277 int next_spare = mddev->raid_disks;
1280 /* make rdev->sb match mddev data..
1283 * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
1284 * 3/ any empty disks < next_spare become removed
1286 * disks[0] gets initialised to REMOVED because
1287 * we cannot be sure from other fields if it has
1288 * been initialised or not.
1291 int active=0, working=0,failed=0,spare=0,nr_disks=0;
1293 rdev->sb_size = MD_SB_BYTES;
1295 sb = page_address(rdev->sb_page);
1297 memset(sb, 0, sizeof(*sb));
1299 sb->md_magic = MD_SB_MAGIC;
1300 sb->major_version = mddev->major_version;
1301 sb->patch_version = mddev->patch_version;
1302 sb->gvalid_words = 0; /* ignored */
1303 memcpy(&sb->set_uuid0, mddev->uuid+0, 4);
1304 memcpy(&sb->set_uuid1, mddev->uuid+4, 4);
1305 memcpy(&sb->set_uuid2, mddev->uuid+8, 4);
1306 memcpy(&sb->set_uuid3, mddev->uuid+12,4);
1308 sb->ctime = mddev->ctime;
1309 sb->level = mddev->level;
1310 sb->size = mddev->dev_sectors / 2;
1311 sb->raid_disks = mddev->raid_disks;
1312 sb->md_minor = mddev->md_minor;
1313 sb->not_persistent = 0;
1314 sb->utime = mddev->utime;
1316 sb->events_hi = (mddev->events>>32);
1317 sb->events_lo = (u32)mddev->events;
1319 if (mddev->reshape_position == MaxSector)
1320 sb->minor_version = 90;
1322 sb->minor_version = 91;
1323 sb->reshape_position = mddev->reshape_position;
1324 sb->new_level = mddev->new_level;
1325 sb->delta_disks = mddev->delta_disks;
1326 sb->new_layout = mddev->new_layout;
1327 sb->new_chunk = mddev->new_chunk_sectors << 9;
1329 mddev->minor_version = sb->minor_version;
1332 sb->recovery_cp = mddev->recovery_cp;
1333 sb->cp_events_hi = (mddev->events>>32);
1334 sb->cp_events_lo = (u32)mddev->events;
1335 if (mddev->recovery_cp == MaxSector)
1336 sb->state = (1<< MD_SB_CLEAN);
1338 sb->recovery_cp = 0;
1340 sb->layout = mddev->layout;
1341 sb->chunk_size = mddev->chunk_sectors << 9;
1343 if (mddev->bitmap && mddev->bitmap_info.file == NULL)
1344 sb->state |= (1<<MD_SB_BITMAP_PRESENT);
1346 sb->disks[0].state = (1<<MD_DISK_REMOVED);
1347 rdev_for_each(rdev2, mddev) {
1350 int is_active = test_bit(In_sync, &rdev2->flags);
1352 if (rdev2->raid_disk >= 0 &&
1353 sb->minor_version >= 91)
1354 /* we have nowhere to store the recovery_offset,
1355 * but if it is not below the reshape_position,
1356 * we can piggy-back on that.
1359 if (rdev2->raid_disk < 0 ||
1360 test_bit(Faulty, &rdev2->flags))
1363 desc_nr = rdev2->raid_disk;
1365 desc_nr = next_spare++;
1366 rdev2->desc_nr = desc_nr;
1367 d = &sb->disks[rdev2->desc_nr];
1369 d->number = rdev2->desc_nr;
1370 d->major = MAJOR(rdev2->bdev->bd_dev);
1371 d->minor = MINOR(rdev2->bdev->bd_dev);
1373 d->raid_disk = rdev2->raid_disk;
1375 d->raid_disk = rdev2->desc_nr; /* compatibility */
1376 if (test_bit(Faulty, &rdev2->flags))
1377 d->state = (1<<MD_DISK_FAULTY);
1378 else if (is_active) {
1379 d->state = (1<<MD_DISK_ACTIVE);
1380 if (test_bit(In_sync, &rdev2->flags))
1381 d->state |= (1<<MD_DISK_SYNC);
1389 if (test_bit(WriteMostly, &rdev2->flags))
1390 d->state |= (1<<MD_DISK_WRITEMOSTLY);
1392 /* now set the "removed" and "faulty" bits on any missing devices */
1393 for (i=0 ; i < mddev->raid_disks ; i++) {
1394 mdp_disk_t *d = &sb->disks[i];
1395 if (d->state == 0 && d->number == 0) {
1398 d->state = (1<<MD_DISK_REMOVED);
1399 d->state |= (1<<MD_DISK_FAULTY);
1403 sb->nr_disks = nr_disks;
1404 sb->active_disks = active;
1405 sb->working_disks = working;
1406 sb->failed_disks = failed;
1407 sb->spare_disks = spare;
1409 sb->this_disk = sb->disks[rdev->desc_nr];
1410 sb->sb_csum = calc_sb_csum(sb);
1414 * rdev_size_change for 0.90.0
1416 static unsigned long long
1417 super_90_rdev_size_change(struct md_rdev *rdev, sector_t num_sectors)
1419 if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
1420 return 0; /* component must fit device */
1421 if (rdev->mddev->bitmap_info.offset)
1422 return 0; /* can't move bitmap */
1423 rdev->sb_start = calc_dev_sboffset(rdev);
1424 if (!num_sectors || num_sectors > rdev->sb_start)
1425 num_sectors = rdev->sb_start;
1426 /* Limit to 4TB as metadata cannot record more than that.
1427 * 4TB == 2^32 KB, or 2*2^32 sectors.
1429 if (num_sectors >= (2ULL << 32))
1430 num_sectors = (2ULL << 32) - 2;
1431 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
1433 md_super_wait(rdev->mddev);
1439 * version 1 superblock
1442 static __le32 calc_sb_1_csum(struct mdp_superblock_1 * sb)
1446 unsigned long long newcsum;
1447 int size = 256 + le32_to_cpu(sb->max_dev)*2;
1448 __le32 *isuper = (__le32*)sb;
1451 disk_csum = sb->sb_csum;
1454 for (i=0; size>=4; size -= 4 )
1455 newcsum += le32_to_cpu(*isuper++);
1458 newcsum += le16_to_cpu(*(__le16*) isuper);
1460 csum = (newcsum & 0xffffffff) + (newcsum >> 32);
1461 sb->sb_csum = disk_csum;
1462 return cpu_to_le32(csum);
1465 static int md_set_badblocks(struct badblocks *bb, sector_t s, int sectors,
1467 static int super_1_load(struct md_rdev *rdev, struct md_rdev *refdev, int minor_version)
1469 struct mdp_superblock_1 *sb;
1472 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
1476 * Calculate the position of the superblock in 512byte sectors.
1477 * It is always aligned to a 4K boundary and
1478 * depeding on minor_version, it can be:
1479 * 0: At least 8K, but less than 12K, from end of device
1480 * 1: At start of device
1481 * 2: 4K from start of device.
1483 switch(minor_version) {
1485 sb_start = i_size_read(rdev->bdev->bd_inode) >> 9;
1487 sb_start &= ~(sector_t)(4*2-1);
1498 rdev->sb_start = sb_start;
1500 /* superblock is rarely larger than 1K, but it can be larger,
1501 * and it is safe to read 4k, so we do that
1503 ret = read_disk_sb(rdev, 4096);
1504 if (ret) return ret;
1507 sb = page_address(rdev->sb_page);
1509 if (sb->magic != cpu_to_le32(MD_SB_MAGIC) ||
1510 sb->major_version != cpu_to_le32(1) ||
1511 le32_to_cpu(sb->max_dev) > (4096-256)/2 ||
1512 le64_to_cpu(sb->super_offset) != rdev->sb_start ||
1513 (le32_to_cpu(sb->feature_map) & ~MD_FEATURE_ALL) != 0)
1516 if (calc_sb_1_csum(sb) != sb->sb_csum) {
1517 printk("md: invalid superblock checksum on %s\n",
1518 bdevname(rdev->bdev,b));
1521 if (le64_to_cpu(sb->data_size) < 10) {
1522 printk("md: data_size too small on %s\n",
1523 bdevname(rdev->bdev,b));
1527 rdev->preferred_minor = 0xffff;
1528 rdev->data_offset = le64_to_cpu(sb->data_offset);
1529 atomic_set(&rdev->corrected_errors, le32_to_cpu(sb->cnt_corrected_read));
1531 rdev->sb_size = le32_to_cpu(sb->max_dev) * 2 + 256;
1532 bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
1533 if (rdev->sb_size & bmask)
1534 rdev->sb_size = (rdev->sb_size | bmask) + 1;
1537 && rdev->data_offset < sb_start + (rdev->sb_size/512))
1540 if (sb->level == cpu_to_le32(LEVEL_MULTIPATH))
1543 rdev->desc_nr = le32_to_cpu(sb->dev_number);
1545 if (!rdev->bb_page) {
1546 rdev->bb_page = alloc_page(GFP_KERNEL);
1550 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BAD_BLOCKS) &&
1551 rdev->badblocks.count == 0) {
1552 /* need to load the bad block list.
1553 * Currently we limit it to one page.
1559 int sectors = le16_to_cpu(sb->bblog_size);
1560 if (sectors > (PAGE_SIZE / 512))
1562 offset = le32_to_cpu(sb->bblog_offset);
1565 bb_sector = (long long)offset;
1566 if (!sync_page_io(rdev, bb_sector, sectors << 9,
1567 rdev->bb_page, READ, true))
1569 bbp = (u64 *)page_address(rdev->bb_page);
1570 rdev->badblocks.shift = sb->bblog_shift;
1571 for (i = 0 ; i < (sectors << (9-3)) ; i++, bbp++) {
1572 u64 bb = le64_to_cpu(*bbp);
1573 int count = bb & (0x3ff);
1574 u64 sector = bb >> 10;
1575 sector <<= sb->bblog_shift;
1576 count <<= sb->bblog_shift;
1579 if (md_set_badblocks(&rdev->badblocks,
1580 sector, count, 1) == 0)
1583 } else if (sb->bblog_offset == 0)
1584 rdev->badblocks.shift = -1;
1590 struct mdp_superblock_1 *refsb = page_address(refdev->sb_page);
1592 if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 ||
1593 sb->level != refsb->level ||
1594 sb->layout != refsb->layout ||
1595 sb->chunksize != refsb->chunksize) {
1596 printk(KERN_WARNING "md: %s has strangely different"
1597 " superblock to %s\n",
1598 bdevname(rdev->bdev,b),
1599 bdevname(refdev->bdev,b2));
1602 ev1 = le64_to_cpu(sb->events);
1603 ev2 = le64_to_cpu(refsb->events);
1611 rdev->sectors = (i_size_read(rdev->bdev->bd_inode) >> 9) -
1612 le64_to_cpu(sb->data_offset);
1614 rdev->sectors = rdev->sb_start;
1615 if (rdev->sectors < le64_to_cpu(sb->data_size))
1617 rdev->sectors = le64_to_cpu(sb->data_size);
1618 if (le64_to_cpu(sb->size) > rdev->sectors)
1623 static int super_1_validate(struct mddev *mddev, struct md_rdev *rdev)
1625 struct mdp_superblock_1 *sb = page_address(rdev->sb_page);
1626 __u64 ev1 = le64_to_cpu(sb->events);
1628 rdev->raid_disk = -1;
1629 clear_bit(Faulty, &rdev->flags);
1630 clear_bit(In_sync, &rdev->flags);
1631 clear_bit(WriteMostly, &rdev->flags);
1633 if (mddev->raid_disks == 0) {
1634 mddev->major_version = 1;
1635 mddev->patch_version = 0;
1636 mddev->external = 0;
1637 mddev->chunk_sectors = le32_to_cpu(sb->chunksize);
1638 mddev->ctime = le64_to_cpu(sb->ctime) & ((1ULL << 32)-1);
1639 mddev->utime = le64_to_cpu(sb->utime) & ((1ULL << 32)-1);
1640 mddev->level = le32_to_cpu(sb->level);
1641 mddev->clevel[0] = 0;
1642 mddev->layout = le32_to_cpu(sb->layout);
1643 mddev->raid_disks = le32_to_cpu(sb->raid_disks);
1644 mddev->dev_sectors = le64_to_cpu(sb->size);
1645 mddev->events = ev1;
1646 mddev->bitmap_info.offset = 0;
1647 mddev->bitmap_info.default_offset = 1024 >> 9;
1649 mddev->recovery_cp = le64_to_cpu(sb->resync_offset);
1650 memcpy(mddev->uuid, sb->set_uuid, 16);
1652 mddev->max_disks = (4096-256)/2;
1654 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) &&
1655 mddev->bitmap_info.file == NULL )
1656 mddev->bitmap_info.offset =
1657 (__s32)le32_to_cpu(sb->bitmap_offset);
1659 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) {
1660 mddev->reshape_position = le64_to_cpu(sb->reshape_position);
1661 mddev->delta_disks = le32_to_cpu(sb->delta_disks);
1662 mddev->new_level = le32_to_cpu(sb->new_level);
1663 mddev->new_layout = le32_to_cpu(sb->new_layout);
1664 mddev->new_chunk_sectors = le32_to_cpu(sb->new_chunk);
1666 mddev->reshape_position = MaxSector;
1667 mddev->delta_disks = 0;
1668 mddev->new_level = mddev->level;
1669 mddev->new_layout = mddev->layout;
1670 mddev->new_chunk_sectors = mddev->chunk_sectors;
1673 } else if (mddev->pers == NULL) {
1674 /* Insist of good event counter while assembling, except for
1675 * spares (which don't need an event count) */
1677 if (rdev->desc_nr >= 0 &&
1678 rdev->desc_nr < le32_to_cpu(sb->max_dev) &&
1679 le16_to_cpu(sb->dev_roles[rdev->desc_nr]) < 0xfffe)
1680 if (ev1 < mddev->events)
1682 } else if (mddev->bitmap) {
1683 /* If adding to array with a bitmap, then we can accept an
1684 * older device, but not too old.
1686 if (ev1 < mddev->bitmap->events_cleared)
1689 if (ev1 < mddev->events)
1690 /* just a hot-add of a new device, leave raid_disk at -1 */
1693 if (mddev->level != LEVEL_MULTIPATH) {
1695 if (rdev->desc_nr < 0 ||
1696 rdev->desc_nr >= le32_to_cpu(sb->max_dev)) {
1700 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
1702 case 0xffff: /* spare */
1704 case 0xfffe: /* faulty */
1705 set_bit(Faulty, &rdev->flags);
1708 if ((le32_to_cpu(sb->feature_map) &
1709 MD_FEATURE_RECOVERY_OFFSET))
1710 rdev->recovery_offset = le64_to_cpu(sb->recovery_offset);
1712 set_bit(In_sync, &rdev->flags);
1713 rdev->raid_disk = role;
1716 if (sb->devflags & WriteMostly1)
1717 set_bit(WriteMostly, &rdev->flags);
1718 if (le32_to_cpu(sb->feature_map) & MD_FEATURE_REPLACEMENT)
1719 set_bit(Replacement, &rdev->flags);
1720 } else /* MULTIPATH are always insync */
1721 set_bit(In_sync, &rdev->flags);
1726 static void super_1_sync(struct mddev *mddev, struct md_rdev *rdev)
1728 struct mdp_superblock_1 *sb;
1729 struct md_rdev *rdev2;
1731 /* make rdev->sb match mddev and rdev data. */
1733 sb = page_address(rdev->sb_page);
1735 sb->feature_map = 0;
1737 sb->recovery_offset = cpu_to_le64(0);
1738 memset(sb->pad1, 0, sizeof(sb->pad1));
1739 memset(sb->pad3, 0, sizeof(sb->pad3));
1741 sb->utime = cpu_to_le64((__u64)mddev->utime);
1742 sb->events = cpu_to_le64(mddev->events);
1744 sb->resync_offset = cpu_to_le64(mddev->recovery_cp);
1746 sb->resync_offset = cpu_to_le64(0);
1748 sb->cnt_corrected_read = cpu_to_le32(atomic_read(&rdev->corrected_errors));
1750 sb->raid_disks = cpu_to_le32(mddev->raid_disks);
1751 sb->size = cpu_to_le64(mddev->dev_sectors);
1752 sb->chunksize = cpu_to_le32(mddev->chunk_sectors);
1753 sb->level = cpu_to_le32(mddev->level);
1754 sb->layout = cpu_to_le32(mddev->layout);
1756 if (test_bit(WriteMostly, &rdev->flags))
1757 sb->devflags |= WriteMostly1;
1759 sb->devflags &= ~WriteMostly1;
1761 if (mddev->bitmap && mddev->bitmap_info.file == NULL) {
1762 sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_info.offset);
1763 sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET);
1766 if (rdev->raid_disk >= 0 &&
1767 !test_bit(In_sync, &rdev->flags)) {
1769 cpu_to_le32(MD_FEATURE_RECOVERY_OFFSET);
1770 sb->recovery_offset =
1771 cpu_to_le64(rdev->recovery_offset);
1773 if (test_bit(Replacement, &rdev->flags))
1775 cpu_to_le32(MD_FEATURE_REPLACEMENT);
1777 if (mddev->reshape_position != MaxSector) {
1778 sb->feature_map |= cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE);
1779 sb->reshape_position = cpu_to_le64(mddev->reshape_position);
1780 sb->new_layout = cpu_to_le32(mddev->new_layout);
1781 sb->delta_disks = cpu_to_le32(mddev->delta_disks);
1782 sb->new_level = cpu_to_le32(mddev->new_level);
1783 sb->new_chunk = cpu_to_le32(mddev->new_chunk_sectors);
1786 if (rdev->badblocks.count == 0)
1787 /* Nothing to do for bad blocks*/ ;
1788 else if (sb->bblog_offset == 0)
1789 /* Cannot record bad blocks on this device */
1790 md_error(mddev, rdev);
1792 struct badblocks *bb = &rdev->badblocks;
1793 u64 *bbp = (u64 *)page_address(rdev->bb_page);
1795 sb->feature_map |= cpu_to_le32(MD_FEATURE_BAD_BLOCKS);
1800 seq = read_seqbegin(&bb->lock);
1802 memset(bbp, 0xff, PAGE_SIZE);
1804 for (i = 0 ; i < bb->count ; i++) {
1805 u64 internal_bb = *p++;
1806 u64 store_bb = ((BB_OFFSET(internal_bb) << 10)
1807 | BB_LEN(internal_bb));
1808 *bbp++ = cpu_to_le64(store_bb);
1811 if (read_seqretry(&bb->lock, seq))
1814 bb->sector = (rdev->sb_start +
1815 (int)le32_to_cpu(sb->bblog_offset));
1816 bb->size = le16_to_cpu(sb->bblog_size);
1821 rdev_for_each(rdev2, mddev)
1822 if (rdev2->desc_nr+1 > max_dev)
1823 max_dev = rdev2->desc_nr+1;
1825 if (max_dev > le32_to_cpu(sb->max_dev)) {
1827 sb->max_dev = cpu_to_le32(max_dev);
1828 rdev->sb_size = max_dev * 2 + 256;
1829 bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
1830 if (rdev->sb_size & bmask)
1831 rdev->sb_size = (rdev->sb_size | bmask) + 1;
1833 max_dev = le32_to_cpu(sb->max_dev);
1835 for (i=0; i<max_dev;i++)
1836 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1838 rdev_for_each(rdev2, mddev) {
1840 if (test_bit(Faulty, &rdev2->flags))
1841 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1842 else if (test_bit(In_sync, &rdev2->flags))
1843 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1844 else if (rdev2->raid_disk >= 0)
1845 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1847 sb->dev_roles[i] = cpu_to_le16(0xffff);
1850 sb->sb_csum = calc_sb_1_csum(sb);
1853 static unsigned long long
1854 super_1_rdev_size_change(struct md_rdev *rdev, sector_t num_sectors)
1856 struct mdp_superblock_1 *sb;
1857 sector_t max_sectors;
1858 if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
1859 return 0; /* component must fit device */
1860 if (rdev->sb_start < rdev->data_offset) {
1861 /* minor versions 1 and 2; superblock before data */
1862 max_sectors = i_size_read(rdev->bdev->bd_inode) >> 9;
1863 max_sectors -= rdev->data_offset;
1864 if (!num_sectors || num_sectors > max_sectors)
1865 num_sectors = max_sectors;
1866 } else if (rdev->mddev->bitmap_info.offset) {
1867 /* minor version 0 with bitmap we can't move */
1870 /* minor version 0; superblock after data */
1872 sb_start = (i_size_read(rdev->bdev->bd_inode) >> 9) - 8*2;
1873 sb_start &= ~(sector_t)(4*2 - 1);
1874 max_sectors = rdev->sectors + sb_start - rdev->sb_start;
1875 if (!num_sectors || num_sectors > max_sectors)
1876 num_sectors = max_sectors;
1877 rdev->sb_start = sb_start;
1879 sb = page_address(rdev->sb_page);
1880 sb->data_size = cpu_to_le64(num_sectors);
1881 sb->super_offset = rdev->sb_start;
1882 sb->sb_csum = calc_sb_1_csum(sb);
1883 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
1885 md_super_wait(rdev->mddev);
1889 static struct super_type super_types[] = {
1892 .owner = THIS_MODULE,
1893 .load_super = super_90_load,
1894 .validate_super = super_90_validate,
1895 .sync_super = super_90_sync,
1896 .rdev_size_change = super_90_rdev_size_change,
1900 .owner = THIS_MODULE,
1901 .load_super = super_1_load,
1902 .validate_super = super_1_validate,
1903 .sync_super = super_1_sync,
1904 .rdev_size_change = super_1_rdev_size_change,
1908 static void sync_super(struct mddev *mddev, struct md_rdev *rdev)
1910 if (mddev->sync_super) {
1911 mddev->sync_super(mddev, rdev);
1915 BUG_ON(mddev->major_version >= ARRAY_SIZE(super_types));
1917 super_types[mddev->major_version].sync_super(mddev, rdev);
1920 static int match_mddev_units(struct mddev *mddev1, struct mddev *mddev2)
1922 struct md_rdev *rdev, *rdev2;
1925 rdev_for_each_rcu(rdev, mddev1)
1926 rdev_for_each_rcu(rdev2, mddev2)
1927 if (rdev->bdev->bd_contains ==
1928 rdev2->bdev->bd_contains) {
1936 static LIST_HEAD(pending_raid_disks);
1939 * Try to register data integrity profile for an mddev
1941 * This is called when an array is started and after a disk has been kicked
1942 * from the array. It only succeeds if all working and active component devices
1943 * are integrity capable with matching profiles.
1945 int md_integrity_register(struct mddev *mddev)
1947 struct md_rdev *rdev, *reference = NULL;
1949 if (list_empty(&mddev->disks))
1950 return 0; /* nothing to do */
1951 if (!mddev->gendisk || blk_get_integrity(mddev->gendisk))
1952 return 0; /* shouldn't register, or already is */
1953 rdev_for_each(rdev, mddev) {
1954 /* skip spares and non-functional disks */
1955 if (test_bit(Faulty, &rdev->flags))
1957 if (rdev->raid_disk < 0)
1960 /* Use the first rdev as the reference */
1964 /* does this rdev's profile match the reference profile? */
1965 if (blk_integrity_compare(reference->bdev->bd_disk,
1966 rdev->bdev->bd_disk) < 0)
1969 if (!reference || !bdev_get_integrity(reference->bdev))
1972 * All component devices are integrity capable and have matching
1973 * profiles, register the common profile for the md device.
1975 if (blk_integrity_register(mddev->gendisk,
1976 bdev_get_integrity(reference->bdev)) != 0) {
1977 printk(KERN_ERR "md: failed to register integrity for %s\n",
1981 printk(KERN_NOTICE "md: data integrity enabled on %s\n", mdname(mddev));
1982 if (bioset_integrity_create(mddev->bio_set, BIO_POOL_SIZE)) {
1983 printk(KERN_ERR "md: failed to create integrity pool for %s\n",
1989 EXPORT_SYMBOL(md_integrity_register);
1991 /* Disable data integrity if non-capable/non-matching disk is being added */
1992 void md_integrity_add_rdev(struct md_rdev *rdev, struct mddev *mddev)
1994 struct blk_integrity *bi_rdev = bdev_get_integrity(rdev->bdev);
1995 struct blk_integrity *bi_mddev = blk_get_integrity(mddev->gendisk);
1997 if (!bi_mddev) /* nothing to do */
1999 if (rdev->raid_disk < 0) /* skip spares */
2001 if (bi_rdev && blk_integrity_compare(mddev->gendisk,
2002 rdev->bdev->bd_disk) >= 0)
2004 printk(KERN_NOTICE "disabling data integrity on %s\n", mdname(mddev));
2005 blk_integrity_unregister(mddev->gendisk);
2007 EXPORT_SYMBOL(md_integrity_add_rdev);
2009 static int bind_rdev_to_array(struct md_rdev * rdev, struct mddev * mddev)
2011 char b[BDEVNAME_SIZE];
2021 /* prevent duplicates */
2022 if (find_rdev(mddev, rdev->bdev->bd_dev))
2025 /* make sure rdev->sectors exceeds mddev->dev_sectors */
2026 if (rdev->sectors && (mddev->dev_sectors == 0 ||
2027 rdev->sectors < mddev->dev_sectors)) {
2029 /* Cannot change size, so fail
2030 * If mddev->level <= 0, then we don't care
2031 * about aligning sizes (e.g. linear)
2033 if (mddev->level > 0)
2036 mddev->dev_sectors = rdev->sectors;
2039 /* Verify rdev->desc_nr is unique.
2040 * If it is -1, assign a free number, else
2041 * check number is not in use
2043 if (rdev->desc_nr < 0) {
2045 if (mddev->pers) choice = mddev->raid_disks;
2046 while (find_rdev_nr(mddev, choice))
2048 rdev->desc_nr = choice;
2050 if (find_rdev_nr(mddev, rdev->desc_nr))
2053 if (mddev->max_disks && rdev->desc_nr >= mddev->max_disks) {
2054 printk(KERN_WARNING "md: %s: array is limited to %d devices\n",
2055 mdname(mddev), mddev->max_disks);
2058 bdevname(rdev->bdev,b);
2059 while ( (s=strchr(b, '/')) != NULL)
2062 rdev->mddev = mddev;
2063 printk(KERN_INFO "md: bind<%s>\n", b);
2065 if ((err = kobject_add(&rdev->kobj, &mddev->kobj, "dev-%s", b)))
2068 ko = &part_to_dev(rdev->bdev->bd_part)->kobj;
2069 if (sysfs_create_link(&rdev->kobj, ko, "block"))
2070 /* failure here is OK */;
2071 rdev->sysfs_state = sysfs_get_dirent_safe(rdev->kobj.sd, "state");
2073 list_add_rcu(&rdev->same_set, &mddev->disks);
2074 bd_link_disk_holder(rdev->bdev, mddev->gendisk);
2076 /* May as well allow recovery to be retried once */
2077 mddev->recovery_disabled++;
2082 printk(KERN_WARNING "md: failed to register dev-%s for %s\n",
2087 static void md_delayed_delete(struct work_struct *ws)
2089 struct md_rdev *rdev = container_of(ws, struct md_rdev, del_work);
2090 kobject_del(&rdev->kobj);
2091 kobject_put(&rdev->kobj);
2094 static void unbind_rdev_from_array(struct md_rdev * rdev)
2096 char b[BDEVNAME_SIZE];
2101 bd_unlink_disk_holder(rdev->bdev, rdev->mddev->gendisk);
2102 list_del_rcu(&rdev->same_set);
2103 printk(KERN_INFO "md: unbind<%s>\n", bdevname(rdev->bdev,b));
2105 sysfs_remove_link(&rdev->kobj, "block");
2106 sysfs_put(rdev->sysfs_state);
2107 rdev->sysfs_state = NULL;
2108 kfree(rdev->badblocks.page);
2109 rdev->badblocks.count = 0;
2110 rdev->badblocks.page = NULL;
2111 /* We need to delay this, otherwise we can deadlock when
2112 * writing to 'remove' to "dev/state". We also need
2113 * to delay it due to rcu usage.
2116 INIT_WORK(&rdev->del_work, md_delayed_delete);
2117 kobject_get(&rdev->kobj);
2118 queue_work(md_misc_wq, &rdev->del_work);
2122 * prevent the device from being mounted, repartitioned or
2123 * otherwise reused by a RAID array (or any other kernel
2124 * subsystem), by bd_claiming the device.
2126 static int lock_rdev(struct md_rdev *rdev, dev_t dev, int shared)
2129 struct block_device *bdev;
2130 char b[BDEVNAME_SIZE];
2132 bdev = blkdev_get_by_dev(dev, FMODE_READ|FMODE_WRITE|FMODE_EXCL,
2133 shared ? (struct md_rdev *)lock_rdev : rdev);
2135 printk(KERN_ERR "md: could not open %s.\n",
2136 __bdevname(dev, b));
2137 return PTR_ERR(bdev);
2143 static void unlock_rdev(struct md_rdev *rdev)
2145 struct block_device *bdev = rdev->bdev;
2149 blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
2152 void md_autodetect_dev(dev_t dev);
2154 static void export_rdev(struct md_rdev * rdev)
2156 char b[BDEVNAME_SIZE];
2157 printk(KERN_INFO "md: export_rdev(%s)\n",
2158 bdevname(rdev->bdev,b));
2163 if (test_bit(AutoDetected, &rdev->flags))
2164 md_autodetect_dev(rdev->bdev->bd_dev);
2167 kobject_put(&rdev->kobj);
2170 static void kick_rdev_from_array(struct md_rdev * rdev)
2172 unbind_rdev_from_array(rdev);
2176 static void export_array(struct mddev *mddev)
2178 struct md_rdev *rdev, *tmp;
2180 rdev_for_each_safe(rdev, tmp, mddev) {
2185 kick_rdev_from_array(rdev);
2187 if (!list_empty(&mddev->disks))
2189 mddev->raid_disks = 0;
2190 mddev->major_version = 0;
2193 static void print_desc(mdp_disk_t *desc)
2195 printk(" DISK<N:%d,(%d,%d),R:%d,S:%d>\n", desc->number,
2196 desc->major,desc->minor,desc->raid_disk,desc->state);
2199 static void print_sb_90(mdp_super_t *sb)
2204 "md: SB: (V:%d.%d.%d) ID:<%08x.%08x.%08x.%08x> CT:%08x\n",
2205 sb->major_version, sb->minor_version, sb->patch_version,
2206 sb->set_uuid0, sb->set_uuid1, sb->set_uuid2, sb->set_uuid3,
2208 printk(KERN_INFO "md: L%d S%08d ND:%d RD:%d md%d LO:%d CS:%d\n",
2209 sb->level, sb->size, sb->nr_disks, sb->raid_disks,
2210 sb->md_minor, sb->layout, sb->chunk_size);
2211 printk(KERN_INFO "md: UT:%08x ST:%d AD:%d WD:%d"
2212 " FD:%d SD:%d CSUM:%08x E:%08lx\n",
2213 sb->utime, sb->state, sb->active_disks, sb->working_disks,
2214 sb->failed_disks, sb->spare_disks,
2215 sb->sb_csum, (unsigned long)sb->events_lo);
2218 for (i = 0; i < MD_SB_DISKS; i++) {
2221 desc = sb->disks + i;
2222 if (desc->number || desc->major || desc->minor ||
2223 desc->raid_disk || (desc->state && (desc->state != 4))) {
2224 printk(" D %2d: ", i);
2228 printk(KERN_INFO "md: THIS: ");
2229 print_desc(&sb->this_disk);
2232 static void print_sb_1(struct mdp_superblock_1 *sb)
2236 uuid = sb->set_uuid;
2238 "md: SB: (V:%u) (F:0x%08x) Array-ID:<%pU>\n"
2239 "md: Name: \"%s\" CT:%llu\n",
2240 le32_to_cpu(sb->major_version),
2241 le32_to_cpu(sb->feature_map),
2244 (unsigned long long)le64_to_cpu(sb->ctime)
2245 & MD_SUPERBLOCK_1_TIME_SEC_MASK);
2247 uuid = sb->device_uuid;
2249 "md: L%u SZ%llu RD:%u LO:%u CS:%u DO:%llu DS:%llu SO:%llu"
2251 "md: Dev:%08x UUID: %pU\n"
2252 "md: (F:0x%08x) UT:%llu Events:%llu ResyncOffset:%llu CSUM:0x%08x\n"
2253 "md: (MaxDev:%u) \n",
2254 le32_to_cpu(sb->level),
2255 (unsigned long long)le64_to_cpu(sb->size),
2256 le32_to_cpu(sb->raid_disks),
2257 le32_to_cpu(sb->layout),
2258 le32_to_cpu(sb->chunksize),
2259 (unsigned long long)le64_to_cpu(sb->data_offset),
2260 (unsigned long long)le64_to_cpu(sb->data_size),
2261 (unsigned long long)le64_to_cpu(sb->super_offset),
2262 (unsigned long long)le64_to_cpu(sb->recovery_offset),
2263 le32_to_cpu(sb->dev_number),
2266 (unsigned long long)le64_to_cpu(sb->utime) & MD_SUPERBLOCK_1_TIME_SEC_MASK,
2267 (unsigned long long)le64_to_cpu(sb->events),
2268 (unsigned long long)le64_to_cpu(sb->resync_offset),
2269 le32_to_cpu(sb->sb_csum),
2270 le32_to_cpu(sb->max_dev)
2274 static void print_rdev(struct md_rdev *rdev, int major_version)
2276 char b[BDEVNAME_SIZE];
2277 printk(KERN_INFO "md: rdev %s, Sect:%08llu F:%d S:%d DN:%u\n",
2278 bdevname(rdev->bdev, b), (unsigned long long)rdev->sectors,
2279 test_bit(Faulty, &rdev->flags), test_bit(In_sync, &rdev->flags),
2281 if (rdev->sb_loaded) {
2282 printk(KERN_INFO "md: rdev superblock (MJ:%d):\n", major_version);
2283 switch (major_version) {
2285 print_sb_90(page_address(rdev->sb_page));
2288 print_sb_1(page_address(rdev->sb_page));
2292 printk(KERN_INFO "md: no rdev superblock!\n");
2295 static void md_print_devices(void)
2297 struct list_head *tmp;
2298 struct md_rdev *rdev;
2299 struct mddev *mddev;
2300 char b[BDEVNAME_SIZE];
2303 printk("md: **********************************\n");
2304 printk("md: * <COMPLETE RAID STATE PRINTOUT> *\n");
2305 printk("md: **********************************\n");
2306 for_each_mddev(mddev, tmp) {
2309 bitmap_print_sb(mddev->bitmap);
2311 printk("%s: ", mdname(mddev));
2312 rdev_for_each(rdev, mddev)
2313 printk("<%s>", bdevname(rdev->bdev,b));
2316 rdev_for_each(rdev, mddev)
2317 print_rdev(rdev, mddev->major_version);
2319 printk("md: **********************************\n");
2324 static void sync_sbs(struct mddev * mddev, int nospares)
2326 /* Update each superblock (in-memory image), but
2327 * if we are allowed to, skip spares which already
2328 * have the right event counter, or have one earlier
2329 * (which would mean they aren't being marked as dirty
2330 * with the rest of the array)
2332 struct md_rdev *rdev;
2333 rdev_for_each(rdev, mddev) {
2334 if (rdev->sb_events == mddev->events ||
2336 rdev->raid_disk < 0 &&
2337 rdev->sb_events+1 == mddev->events)) {
2338 /* Don't update this superblock */
2339 rdev->sb_loaded = 2;
2341 sync_super(mddev, rdev);
2342 rdev->sb_loaded = 1;
2347 static void md_update_sb(struct mddev * mddev, int force_change)
2349 struct md_rdev *rdev;
2352 int any_badblocks_changed = 0;
2355 /* First make sure individual recovery_offsets are correct */
2356 rdev_for_each(rdev, mddev) {
2357 if (rdev->raid_disk >= 0 &&
2358 mddev->delta_disks >= 0 &&
2359 !test_bit(In_sync, &rdev->flags) &&
2360 mddev->curr_resync_completed > rdev->recovery_offset)
2361 rdev->recovery_offset = mddev->curr_resync_completed;
2364 if (!mddev->persistent) {
2365 clear_bit(MD_CHANGE_CLEAN, &mddev->flags);
2366 clear_bit(MD_CHANGE_DEVS, &mddev->flags);
2367 if (!mddev->external) {
2368 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
2369 rdev_for_each(rdev, mddev) {
2370 if (rdev->badblocks.changed) {
2371 rdev->badblocks.changed = 0;
2372 md_ack_all_badblocks(&rdev->badblocks);
2373 md_error(mddev, rdev);
2375 clear_bit(Blocked, &rdev->flags);
2376 clear_bit(BlockedBadBlocks, &rdev->flags);
2377 wake_up(&rdev->blocked_wait);
2380 wake_up(&mddev->sb_wait);
2384 spin_lock_irq(&mddev->write_lock);
2386 mddev->utime = get_seconds();
2388 if (test_and_clear_bit(MD_CHANGE_DEVS, &mddev->flags))
2390 if (test_and_clear_bit(MD_CHANGE_CLEAN, &mddev->flags))
2391 /* just a clean<-> dirty transition, possibly leave spares alone,
2392 * though if events isn't the right even/odd, we will have to do
2398 if (mddev->degraded)
2399 /* If the array is degraded, then skipping spares is both
2400 * dangerous and fairly pointless.
2401 * Dangerous because a device that was removed from the array
2402 * might have a event_count that still looks up-to-date,
2403 * so it can be re-added without a resync.
2404 * Pointless because if there are any spares to skip,
2405 * then a recovery will happen and soon that array won't
2406 * be degraded any more and the spare can go back to sleep then.
2410 sync_req = mddev->in_sync;
2412 /* If this is just a dirty<->clean transition, and the array is clean
2413 * and 'events' is odd, we can roll back to the previous clean state */
2415 && (mddev->in_sync && mddev->recovery_cp == MaxSector)
2416 && mddev->can_decrease_events
2417 && mddev->events != 1) {
2419 mddev->can_decrease_events = 0;
2421 /* otherwise we have to go forward and ... */
2423 mddev->can_decrease_events = nospares;
2426 if (!mddev->events) {
2428 * oops, this 64-bit counter should never wrap.
2429 * Either we are in around ~1 trillion A.C., assuming
2430 * 1 reboot per second, or we have a bug:
2436 rdev_for_each(rdev, mddev) {
2437 if (rdev->badblocks.changed)
2438 any_badblocks_changed++;
2439 if (test_bit(Faulty, &rdev->flags))
2440 set_bit(FaultRecorded, &rdev->flags);
2443 sync_sbs(mddev, nospares);
2444 spin_unlock_irq(&mddev->write_lock);
2446 pr_debug("md: updating %s RAID superblock on device (in sync %d)\n",
2447 mdname(mddev), mddev->in_sync);
2449 bitmap_update_sb(mddev->bitmap);
2450 rdev_for_each(rdev, mddev) {
2451 char b[BDEVNAME_SIZE];
2453 if (rdev->sb_loaded != 1)
2454 continue; /* no noise on spare devices */
2456 if (!test_bit(Faulty, &rdev->flags) &&
2457 rdev->saved_raid_disk == -1) {
2458 md_super_write(mddev,rdev,
2459 rdev->sb_start, rdev->sb_size,
2461 pr_debug("md: (write) %s's sb offset: %llu\n",
2462 bdevname(rdev->bdev, b),
2463 (unsigned long long)rdev->sb_start);
2464 rdev->sb_events = mddev->events;
2465 if (rdev->badblocks.size) {
2466 md_super_write(mddev, rdev,
2467 rdev->badblocks.sector,
2468 rdev->badblocks.size << 9,
2470 rdev->badblocks.size = 0;
2473 } else if (test_bit(Faulty, &rdev->flags))
2474 pr_debug("md: %s (skipping faulty)\n",
2475 bdevname(rdev->bdev, b));
2477 pr_debug("(skipping incremental s/r ");
2479 if (mddev->level == LEVEL_MULTIPATH)
2480 /* only need to write one superblock... */
2483 md_super_wait(mddev);
2484 /* if there was a failure, MD_CHANGE_DEVS was set, and we re-write super */
2486 spin_lock_irq(&mddev->write_lock);
2487 if (mddev->in_sync != sync_req ||
2488 test_bit(MD_CHANGE_DEVS, &mddev->flags)) {
2489 /* have to write it out again */
2490 spin_unlock_irq(&mddev->write_lock);
2493 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
2494 spin_unlock_irq(&mddev->write_lock);
2495 wake_up(&mddev->sb_wait);
2496 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
2497 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
2499 rdev_for_each(rdev, mddev) {
2500 if (test_and_clear_bit(FaultRecorded, &rdev->flags))
2501 clear_bit(Blocked, &rdev->flags);
2503 if (any_badblocks_changed)
2504 md_ack_all_badblocks(&rdev->badblocks);
2505 clear_bit(BlockedBadBlocks, &rdev->flags);
2506 wake_up(&rdev->blocked_wait);
2510 /* words written to sysfs files may, or may not, be \n terminated.
2511 * We want to accept with case. For this we use cmd_match.
2513 static int cmd_match(const char *cmd, const char *str)
2515 /* See if cmd, written into a sysfs file, matches
2516 * str. They must either be the same, or cmd can
2517 * have a trailing newline
2519 while (*cmd && *str && *cmd == *str) {
2530 struct rdev_sysfs_entry {
2531 struct attribute attr;
2532 ssize_t (*show)(struct md_rdev *, char *);
2533 ssize_t (*store)(struct md_rdev *, const char *, size_t);
2537 state_show(struct md_rdev *rdev, char *page)
2542 if (test_bit(Faulty, &rdev->flags) ||
2543 rdev->badblocks.unacked_exist) {
2544 len+= sprintf(page+len, "%sfaulty",sep);
2547 if (test_bit(In_sync, &rdev->flags)) {
2548 len += sprintf(page+len, "%sin_sync",sep);
2551 if (test_bit(WriteMostly, &rdev->flags)) {
2552 len += sprintf(page+len, "%swrite_mostly",sep);
2555 if (test_bit(Blocked, &rdev->flags) ||
2556 (rdev->badblocks.unacked_exist
2557 && !test_bit(Faulty, &rdev->flags))) {
2558 len += sprintf(page+len, "%sblocked", sep);
2561 if (!test_bit(Faulty, &rdev->flags) &&
2562 !test_bit(In_sync, &rdev->flags)) {
2563 len += sprintf(page+len, "%sspare", sep);
2566 if (test_bit(WriteErrorSeen, &rdev->flags)) {
2567 len += sprintf(page+len, "%swrite_error", sep);
2570 if (test_bit(WantReplacement, &rdev->flags)) {
2571 len += sprintf(page+len, "%swant_replacement", sep);
2574 if (test_bit(Replacement, &rdev->flags)) {
2575 len += sprintf(page+len, "%sreplacement", sep);
2579 return len+sprintf(page+len, "\n");
2583 state_store(struct md_rdev *rdev, const char *buf, size_t len)
2586 * faulty - simulates an error
2587 * remove - disconnects the device
2588 * writemostly - sets write_mostly
2589 * -writemostly - clears write_mostly
2590 * blocked - sets the Blocked flags
2591 * -blocked - clears the Blocked and possibly simulates an error
2592 * insync - sets Insync providing device isn't active
2593 * write_error - sets WriteErrorSeen
2594 * -write_error - clears WriteErrorSeen
2597 if (cmd_match(buf, "faulty") && rdev->mddev->pers) {
2598 md_error(rdev->mddev, rdev);
2599 if (test_bit(Faulty, &rdev->flags))
2603 } else if (cmd_match(buf, "remove")) {
2604 if (rdev->raid_disk >= 0)
2607 struct mddev *mddev = rdev->mddev;
2608 kick_rdev_from_array(rdev);
2610 md_update_sb(mddev, 1);
2611 md_new_event(mddev);
2614 } else if (cmd_match(buf, "writemostly")) {
2615 set_bit(WriteMostly, &rdev->flags);
2617 } else if (cmd_match(buf, "-writemostly")) {
2618 clear_bit(WriteMostly, &rdev->flags);
2620 } else if (cmd_match(buf, "blocked")) {
2621 set_bit(Blocked, &rdev->flags);
2623 } else if (cmd_match(buf, "-blocked")) {
2624 if (!test_bit(Faulty, &rdev->flags) &&
2625 rdev->badblocks.unacked_exist) {
2626 /* metadata handler doesn't understand badblocks,
2627 * so we need to fail the device
2629 md_error(rdev->mddev, rdev);
2631 clear_bit(Blocked, &rdev->flags);
2632 clear_bit(BlockedBadBlocks, &rdev->flags);
2633 wake_up(&rdev->blocked_wait);
2634 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2635 md_wakeup_thread(rdev->mddev->thread);
2638 } else if (cmd_match(buf, "insync") && rdev->raid_disk == -1) {
2639 set_bit(In_sync, &rdev->flags);
2641 } else if (cmd_match(buf, "write_error")) {
2642 set_bit(WriteErrorSeen, &rdev->flags);
2644 } else if (cmd_match(buf, "-write_error")) {
2645 clear_bit(WriteErrorSeen, &rdev->flags);
2647 } else if (cmd_match(buf, "want_replacement")) {
2648 /* Any non-spare device that is not a replacement can
2649 * become want_replacement at any time, but we then need to
2650 * check if recovery is needed.
2652 if (rdev->raid_disk >= 0 &&
2653 !test_bit(Replacement, &rdev->flags))
2654 set_bit(WantReplacement, &rdev->flags);
2655 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2656 md_wakeup_thread(rdev->mddev->thread);
2658 } else if (cmd_match(buf, "-want_replacement")) {
2659 /* Clearing 'want_replacement' is always allowed.
2660 * Once replacements starts it is too late though.
2663 clear_bit(WantReplacement, &rdev->flags);
2664 } else if (cmd_match(buf, "replacement")) {
2665 /* Can only set a device as a replacement when array has not
2666 * yet been started. Once running, replacement is automatic
2667 * from spares, or by assigning 'slot'.
2669 if (rdev->mddev->pers)
2672 set_bit(Replacement, &rdev->flags);
2675 } else if (cmd_match(buf, "-replacement")) {
2676 /* Similarly, can only clear Replacement before start */
2677 if (rdev->mddev->pers)
2680 clear_bit(Replacement, &rdev->flags);
2685 sysfs_notify_dirent_safe(rdev->sysfs_state);
2686 return err ? err : len;
2688 static struct rdev_sysfs_entry rdev_state =
2689 __ATTR(state, S_IRUGO|S_IWUSR, state_show, state_store);
2692 errors_show(struct md_rdev *rdev, char *page)
2694 return sprintf(page, "%d\n", atomic_read(&rdev->corrected_errors));
2698 errors_store(struct md_rdev *rdev, const char *buf, size_t len)
2701 unsigned long n = simple_strtoul(buf, &e, 10);
2702 if (*buf && (*e == 0 || *e == '\n')) {
2703 atomic_set(&rdev->corrected_errors, n);
2708 static struct rdev_sysfs_entry rdev_errors =
2709 __ATTR(errors, S_IRUGO|S_IWUSR, errors_show, errors_store);
2712 slot_show(struct md_rdev *rdev, char *page)
2714 if (rdev->raid_disk < 0)
2715 return sprintf(page, "none\n");
2717 return sprintf(page, "%d\n", rdev->raid_disk);
2721 slot_store(struct md_rdev *rdev, const char *buf, size_t len)
2725 int slot = simple_strtoul(buf, &e, 10);
2726 if (strncmp(buf, "none", 4)==0)
2728 else if (e==buf || (*e && *e!= '\n'))
2730 if (rdev->mddev->pers && slot == -1) {
2731 /* Setting 'slot' on an active array requires also
2732 * updating the 'rd%d' link, and communicating
2733 * with the personality with ->hot_*_disk.
2734 * For now we only support removing
2735 * failed/spare devices. This normally happens automatically,
2736 * but not when the metadata is externally managed.
2738 if (rdev->raid_disk == -1)
2740 /* personality does all needed checks */
2741 if (rdev->mddev->pers->hot_remove_disk == NULL)
2743 err = rdev->mddev->pers->
2744 hot_remove_disk(rdev->mddev, rdev);
2747 sysfs_unlink_rdev(rdev->mddev, rdev);
2748 rdev->raid_disk = -1;
2749 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2750 md_wakeup_thread(rdev->mddev->thread);
2751 } else if (rdev->mddev->pers) {
2752 /* Activating a spare .. or possibly reactivating
2753 * if we ever get bitmaps working here.
2756 if (rdev->raid_disk != -1)
2759 if (test_bit(MD_RECOVERY_RUNNING, &rdev->mddev->recovery))
2762 if (rdev->mddev->pers->hot_add_disk == NULL)
2765 if (slot >= rdev->mddev->raid_disks &&
2766 slot >= rdev->mddev->raid_disks + rdev->mddev->delta_disks)
2769 rdev->raid_disk = slot;
2770 if (test_bit(In_sync, &rdev->flags))
2771 rdev->saved_raid_disk = slot;
2773 rdev->saved_raid_disk = -1;
2774 clear_bit(In_sync, &rdev->flags);
2775 err = rdev->mddev->pers->
2776 hot_add_disk(rdev->mddev, rdev);
2778 rdev->raid_disk = -1;
2781 sysfs_notify_dirent_safe(rdev->sysfs_state);
2782 if (sysfs_link_rdev(rdev->mddev, rdev))
2783 /* failure here is OK */;
2784 /* don't wakeup anyone, leave that to userspace. */
2786 if (slot >= rdev->mddev->raid_disks &&
2787 slot >= rdev->mddev->raid_disks + rdev->mddev->delta_disks)
2789 rdev->raid_disk = slot;
2790 /* assume it is working */
2791 clear_bit(Faulty, &rdev->flags);
2792 clear_bit(WriteMostly, &rdev->flags);
2793 set_bit(In_sync, &rdev->flags);
2794 sysfs_notify_dirent_safe(rdev->sysfs_state);
2800 static struct rdev_sysfs_entry rdev_slot =
2801 __ATTR(slot, S_IRUGO|S_IWUSR, slot_show, slot_store);
2804 offset_show(struct md_rdev *rdev, char *page)
2806 return sprintf(page, "%llu\n", (unsigned long long)rdev->data_offset);
2810 offset_store(struct md_rdev *rdev, const char *buf, size_t len)
2813 unsigned long long offset = simple_strtoull(buf, &e, 10);
2814 if (e==buf || (*e && *e != '\n'))
2816 if (rdev->mddev->pers && rdev->raid_disk >= 0)
2818 if (rdev->sectors && rdev->mddev->external)
2819 /* Must set offset before size, so overlap checks
2822 rdev->data_offset = offset;
2826 static struct rdev_sysfs_entry rdev_offset =
2827 __ATTR(offset, S_IRUGO|S_IWUSR, offset_show, offset_store);
2830 rdev_size_show(struct md_rdev *rdev, char *page)
2832 return sprintf(page, "%llu\n", (unsigned long long)rdev->sectors / 2);
2835 static int overlaps(sector_t s1, sector_t l1, sector_t s2, sector_t l2)
2837 /* check if two start/length pairs overlap */
2845 static int strict_blocks_to_sectors(const char *buf, sector_t *sectors)
2847 unsigned long long blocks;
2850 if (strict_strtoull(buf, 10, &blocks) < 0)
2853 if (blocks & 1ULL << (8 * sizeof(blocks) - 1))
2854 return -EINVAL; /* sector conversion overflow */
2857 if (new != blocks * 2)
2858 return -EINVAL; /* unsigned long long to sector_t overflow */
2865 rdev_size_store(struct md_rdev *rdev, const char *buf, size_t len)
2867 struct mddev *my_mddev = rdev->mddev;
2868 sector_t oldsectors = rdev->sectors;
2871 if (strict_blocks_to_sectors(buf, §ors) < 0)
2873 if (my_mddev->pers && rdev->raid_disk >= 0) {
2874 if (my_mddev->persistent) {
2875 sectors = super_types[my_mddev->major_version].
2876 rdev_size_change(rdev, sectors);
2879 } else if (!sectors)
2880 sectors = (i_size_read(rdev->bdev->bd_inode) >> 9) -
2883 if (sectors < my_mddev->dev_sectors)
2884 return -EINVAL; /* component must fit device */
2886 rdev->sectors = sectors;
2887 if (sectors > oldsectors && my_mddev->external) {
2888 /* need to check that all other rdevs with the same ->bdev
2889 * do not overlap. We need to unlock the mddev to avoid
2890 * a deadlock. We have already changed rdev->sectors, and if
2891 * we have to change it back, we will have the lock again.
2893 struct mddev *mddev;
2895 struct list_head *tmp;
2897 mddev_unlock(my_mddev);
2898 for_each_mddev(mddev, tmp) {
2899 struct md_rdev *rdev2;
2902 rdev_for_each(rdev2, mddev)
2903 if (rdev->bdev == rdev2->bdev &&
2905 overlaps(rdev->data_offset, rdev->sectors,
2911 mddev_unlock(mddev);
2917 mddev_lock(my_mddev);
2919 /* Someone else could have slipped in a size
2920 * change here, but doing so is just silly.
2921 * We put oldsectors back because we *know* it is
2922 * safe, and trust userspace not to race with
2925 rdev->sectors = oldsectors;
2932 static struct rdev_sysfs_entry rdev_size =
2933 __ATTR(size, S_IRUGO|S_IWUSR, rdev_size_show, rdev_size_store);
2936 static ssize_t recovery_start_show(struct md_rdev *rdev, char *page)
2938 unsigned long long recovery_start = rdev->recovery_offset;
2940 if (test_bit(In_sync, &rdev->flags) ||
2941 recovery_start == MaxSector)
2942 return sprintf(page, "none\n");
2944 return sprintf(page, "%llu\n", recovery_start);
2947 static ssize_t recovery_start_store(struct md_rdev *rdev, const char *buf, size_t len)
2949 unsigned long long recovery_start;
2951 if (cmd_match(buf, "none"))
2952 recovery_start = MaxSector;
2953 else if (strict_strtoull(buf, 10, &recovery_start))
2956 if (rdev->mddev->pers &&
2957 rdev->raid_disk >= 0)
2960 rdev->recovery_offset = recovery_start;
2961 if (recovery_start == MaxSector)
2962 set_bit(In_sync, &rdev->flags);
2964 clear_bit(In_sync, &rdev->flags);
2968 static struct rdev_sysfs_entry rdev_recovery_start =
2969 __ATTR(recovery_start, S_IRUGO|S_IWUSR, recovery_start_show, recovery_start_store);
2973 badblocks_show(struct badblocks *bb, char *page, int unack);
2975 badblocks_store(struct badblocks *bb, const char *page, size_t len, int unack);
2977 static ssize_t bb_show(struct md_rdev *rdev, char *page)
2979 return badblocks_show(&rdev->badblocks, page, 0);
2981 static ssize_t bb_store(struct md_rdev *rdev, const char *page, size_t len)
2983 int rv = badblocks_store(&rdev->badblocks, page, len, 0);
2984 /* Maybe that ack was all we needed */
2985 if (test_and_clear_bit(BlockedBadBlocks, &rdev->flags))
2986 wake_up(&rdev->blocked_wait);
2989 static struct rdev_sysfs_entry rdev_bad_blocks =
2990 __ATTR(bad_blocks, S_IRUGO|S_IWUSR, bb_show, bb_store);
2993 static ssize_t ubb_show(struct md_rdev *rdev, char *page)
2995 return badblocks_show(&rdev->badblocks, page, 1);
2997 static ssize_t ubb_store(struct md_rdev *rdev, const char *page, size_t len)
2999 return badblocks_store(&rdev->badblocks, page, len, 1);
3001 static struct rdev_sysfs_entry rdev_unack_bad_blocks =
3002 __ATTR(unacknowledged_bad_blocks, S_IRUGO|S_IWUSR, ubb_show, ubb_store);
3004 static struct attribute *rdev_default_attrs[] = {
3010 &rdev_recovery_start.attr,
3011 &rdev_bad_blocks.attr,
3012 &rdev_unack_bad_blocks.attr,
3016 rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
3018 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
3019 struct md_rdev *rdev = container_of(kobj, struct md_rdev, kobj);
3020 struct mddev *mddev = rdev->mddev;
3026 rv = mddev ? mddev_lock(mddev) : -EBUSY;
3028 if (rdev->mddev == NULL)
3031 rv = entry->show(rdev, page);
3032 mddev_unlock(mddev);
3038 rdev_attr_store(struct kobject *kobj, struct attribute *attr,
3039 const char *page, size_t length)
3041 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
3042 struct md_rdev *rdev = container_of(kobj, struct md_rdev, kobj);
3044 struct mddev *mddev = rdev->mddev;
3048 if (!capable(CAP_SYS_ADMIN))
3050 rv = mddev ? mddev_lock(mddev): -EBUSY;
3052 if (rdev->mddev == NULL)
3055 rv = entry->store(rdev, page, length);
3056 mddev_unlock(mddev);
3061 static void rdev_free(struct kobject *ko)
3063 struct md_rdev *rdev = container_of(ko, struct md_rdev, kobj);
3066 static const struct sysfs_ops rdev_sysfs_ops = {
3067 .show = rdev_attr_show,
3068 .store = rdev_attr_store,
3070 static struct kobj_type rdev_ktype = {
3071 .release = rdev_free,
3072 .sysfs_ops = &rdev_sysfs_ops,
3073 .default_attrs = rdev_default_attrs,
3076 int md_rdev_init(struct md_rdev *rdev)
3079 rdev->saved_raid_disk = -1;
3080 rdev->raid_disk = -1;
3082 rdev->data_offset = 0;
3083 rdev->sb_events = 0;
3084 rdev->last_read_error.tv_sec = 0;
3085 rdev->last_read_error.tv_nsec = 0;
3086 rdev->sb_loaded = 0;
3087 rdev->bb_page = NULL;
3088 atomic_set(&rdev->nr_pending, 0);
3089 atomic_set(&rdev->read_errors, 0);
3090 atomic_set(&rdev->corrected_errors, 0);
3092 INIT_LIST_HEAD(&rdev->same_set);
3093 init_waitqueue_head(&rdev->blocked_wait);
3095 /* Add space to store bad block list.
3096 * This reserves the space even on arrays where it cannot
3097 * be used - I wonder if that matters
3099 rdev->badblocks.count = 0;
3100 rdev->badblocks.shift = 0;
3101 rdev->badblocks.page = kmalloc(PAGE_SIZE, GFP_KERNEL);
3102 seqlock_init(&rdev->badblocks.lock);
3103 if (rdev->badblocks.page == NULL)
3108 EXPORT_SYMBOL_GPL(md_rdev_init);
3110 * Import a device. If 'super_format' >= 0, then sanity check the superblock
3112 * mark the device faulty if:
3114 * - the device is nonexistent (zero size)
3115 * - the device has no valid superblock
3117 * a faulty rdev _never_ has rdev->sb set.
3119 static struct md_rdev *md_import_device(dev_t newdev, int super_format, int super_minor)
3121 char b[BDEVNAME_SIZE];
3123 struct md_rdev *rdev;
3126 rdev = kzalloc(sizeof(*rdev), GFP_KERNEL);
3128 printk(KERN_ERR "md: could not alloc mem for new device!\n");
3129 return ERR_PTR(-ENOMEM);
3132 err = md_rdev_init(rdev);
3135 err = alloc_disk_sb(rdev);
3139 err = lock_rdev(rdev, newdev, super_format == -2);
3143 kobject_init(&rdev->kobj, &rdev_ktype);
3145 size = i_size_read(rdev->bdev->bd_inode) >> BLOCK_SIZE_BITS;
3148 "md: %s has zero or unknown size, marking faulty!\n",
3149 bdevname(rdev->bdev,b));
3154 if (super_format >= 0) {
3155 err = super_types[super_format].
3156 load_super(rdev, NULL, super_minor);
3157 if (err == -EINVAL) {
3159 "md: %s does not have a valid v%d.%d "
3160 "superblock, not importing!\n",
3161 bdevname(rdev->bdev,b),
3162 super_format, super_minor);
3167 "md: could not read %s's sb, not importing!\n",
3168 bdevname(rdev->bdev,b));
3172 if (super_format == -1)
3173 /* hot-add for 0.90, or non-persistent: so no badblocks */
3174 rdev->badblocks.shift = -1;
3182 kfree(rdev->badblocks.page);
3184 return ERR_PTR(err);
3188 * Check a full RAID array for plausibility
3192 static void analyze_sbs(struct mddev * mddev)
3195 struct md_rdev *rdev, *freshest, *tmp;
3196 char b[BDEVNAME_SIZE];
3199 rdev_for_each_safe(rdev, tmp, mddev)
3200 switch (super_types[mddev->major_version].
3201 load_super(rdev, freshest, mddev->minor_version)) {
3209 "md: fatal superblock inconsistency in %s"
3210 " -- removing from array\n",
3211 bdevname(rdev->bdev,b));
3212 kick_rdev_from_array(rdev);
3216 super_types[mddev->major_version].
3217 validate_super(mddev, freshest);
3220 rdev_for_each_safe(rdev, tmp, mddev) {
3221 if (mddev->max_disks &&
3222 (rdev->desc_nr >= mddev->max_disks ||
3223 i > mddev->max_disks)) {
3225 "md: %s: %s: only %d devices permitted\n",
3226 mdname(mddev), bdevname(rdev->bdev, b),
3228 kick_rdev_from_array(rdev);
3231 if (rdev != freshest)
3232 if (super_types[mddev->major_version].
3233 validate_super(mddev, rdev)) {
3234 printk(KERN_WARNING "md: kicking non-fresh %s"
3236 bdevname(rdev->bdev,b));
3237 kick_rdev_from_array(rdev);
3240 if (mddev->level == LEVEL_MULTIPATH) {
3241 rdev->desc_nr = i++;
3242 rdev->raid_disk = rdev->desc_nr;
3243 set_bit(In_sync, &rdev->flags);
3244 } else if (rdev->raid_disk >= (mddev->raid_disks - min(0, mddev->delta_disks))) {
3245 rdev->raid_disk = -1;
3246 clear_bit(In_sync, &rdev->flags);
3251 /* Read a fixed-point number.
3252 * Numbers in sysfs attributes should be in "standard" units where
3253 * possible, so time should be in seconds.
3254 * However we internally use a a much smaller unit such as
3255 * milliseconds or jiffies.
3256 * This function takes a decimal number with a possible fractional
3257 * component, and produces an integer which is the result of
3258 * multiplying that number by 10^'scale'.
3259 * all without any floating-point arithmetic.
3261 int strict_strtoul_scaled(const char *cp, unsigned long *res, int scale)
3263 unsigned long result = 0;
3265 while (isdigit(*cp) || (*cp == '.' && decimals < 0)) {
3268 else if (decimals < scale) {
3271 result = result * 10 + value;
3283 while (decimals < scale) {
3292 static void md_safemode_timeout(unsigned long data);
3295 safe_delay_show(struct mddev *mddev, char *page)
3297 int msec = (mddev->safemode_delay*1000)/HZ;
3298 return sprintf(page, "%d.%03d\n", msec/1000, msec%1000);
3301 safe_delay_store(struct mddev *mddev, const char *cbuf, size_t len)
3305 if (strict_strtoul_scaled(cbuf, &msec, 3) < 0)
3308 mddev->safemode_delay = 0;
3310 unsigned long old_delay = mddev->safemode_delay;
3311 mddev->safemode_delay = (msec*HZ)/1000;
3312 if (mddev->safemode_delay == 0)
3313 mddev->safemode_delay = 1;
3314 if (mddev->safemode_delay < old_delay)
3315 md_safemode_timeout((unsigned long)mddev);
3319 static struct md_sysfs_entry md_safe_delay =
3320 __ATTR(safe_mode_delay, S_IRUGO|S_IWUSR,safe_delay_show, safe_delay_store);
3323 level_show(struct mddev *mddev, char *page)
3325 struct md_personality *p = mddev->pers;
3327 return sprintf(page, "%s\n", p->name);
3328 else if (mddev->clevel[0])
3329 return sprintf(page, "%s\n", mddev->clevel);
3330 else if (mddev->level != LEVEL_NONE)
3331 return sprintf(page, "%d\n", mddev->level);
3337 level_store(struct mddev *mddev, const char *buf, size_t len)
3341 struct md_personality *pers;
3344 struct md_rdev *rdev;
3346 if (mddev->pers == NULL) {
3349 if (len >= sizeof(mddev->clevel))
3351 strncpy(mddev->clevel, buf, len);
3352 if (mddev->clevel[len-1] == '\n')
3354 mddev->clevel[len] = 0;
3355 mddev->level = LEVEL_NONE;
3359 /* request to change the personality. Need to ensure:
3360 * - array is not engaged in resync/recovery/reshape
3361 * - old personality can be suspended
3362 * - new personality will access other array.
3365 if (mddev->sync_thread ||
3366 mddev->reshape_position != MaxSector ||
3367 mddev->sysfs_active)
3370 if (!mddev->pers->quiesce) {
3371 printk(KERN_WARNING "md: %s: %s does not support online personality change\n",
3372 mdname(mddev), mddev->pers->name);
3376 /* Now find the new personality */
3377 if (len == 0 || len >= sizeof(clevel))
3379 strncpy(clevel, buf, len);
3380 if (clevel[len-1] == '\n')
3383 if (strict_strtol(clevel, 10, &level))
3386 if (request_module("md-%s", clevel) != 0)
3387 request_module("md-level-%s", clevel);
3388 spin_lock(&pers_lock);
3389 pers = find_pers(level, clevel);
3390 if (!pers || !try_module_get(pers->owner)) {
3391 spin_unlock(&pers_lock);
3392 printk(KERN_WARNING "md: personality %s not loaded\n", clevel);
3395 spin_unlock(&pers_lock);
3397 if (pers == mddev->pers) {
3398 /* Nothing to do! */
3399 module_put(pers->owner);
3402 if (!pers->takeover) {
3403 module_put(pers->owner);
3404 printk(KERN_WARNING "md: %s: %s does not support personality takeover\n",
3405 mdname(mddev), clevel);
3409 rdev_for_each(rdev, mddev)
3410 rdev->new_raid_disk = rdev->raid_disk;
3412 /* ->takeover must set new_* and/or delta_disks
3413 * if it succeeds, and may set them when it fails.
3415 priv = pers->takeover(mddev);
3417 mddev->new_level = mddev->level;
3418 mddev->new_layout = mddev->layout;
3419 mddev->new_chunk_sectors = mddev->chunk_sectors;
3420 mddev->raid_disks -= mddev->delta_disks;
3421 mddev->delta_disks = 0;
3422 module_put(pers->owner);
3423 printk(KERN_WARNING "md: %s: %s would not accept array\n",
3424 mdname(mddev), clevel);
3425 return PTR_ERR(priv);
3428 /* Looks like we have a winner */
3429 mddev_suspend(mddev);
3430 mddev->pers->stop(mddev);
3432 if (mddev->pers->sync_request == NULL &&
3433 pers->sync_request != NULL) {
3434 /* need to add the md_redundancy_group */
3435 if (sysfs_create_group(&mddev->kobj, &md_redundancy_group))
3437 "md: cannot register extra attributes for %s\n",
3439 mddev->sysfs_action = sysfs_get_dirent(mddev->kobj.sd, NULL, "sync_action");
3441 if (mddev->pers->sync_request != NULL &&
3442 pers->sync_request == NULL) {
3443 /* need to remove the md_redundancy_group */
3444 if (mddev->to_remove == NULL)
3445 mddev->to_remove = &md_redundancy_group;
3448 if (mddev->pers->sync_request == NULL &&
3450 /* We are converting from a no-redundancy array
3451 * to a redundancy array and metadata is managed
3452 * externally so we need to be sure that writes
3453 * won't block due to a need to transition
3455 * until external management is started.
3458 mddev->safemode_delay = 0;
3459 mddev->safemode = 0;
3462 rdev_for_each(rdev, mddev) {
3463 if (rdev->raid_disk < 0)
3465 if (rdev->new_raid_disk >= mddev->raid_disks)
3466 rdev->new_raid_disk = -1;
3467 if (rdev->new_raid_disk == rdev->raid_disk)
3469 sysfs_unlink_rdev(mddev, rdev);
3471 rdev_for_each(rdev, mddev) {
3472 if (rdev->raid_disk < 0)
3474 if (rdev->new_raid_disk == rdev->raid_disk)
3476 rdev->raid_disk = rdev->new_raid_disk;
3477 if (rdev->raid_disk < 0)
3478 clear_bit(In_sync, &rdev->flags);
3480 if (sysfs_link_rdev(mddev, rdev))
3481 printk(KERN_WARNING "md: cannot register rd%d"
3482 " for %s after level change\n",
3483 rdev->raid_disk, mdname(mddev));
3487 module_put(mddev->pers->owner);
3489 mddev->private = priv;
3490 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
3491 mddev->level = mddev->new_level;
3492 mddev->layout = mddev->new_layout;
3493 mddev->chunk_sectors = mddev->new_chunk_sectors;
3494 mddev->delta_disks = 0;
3495 mddev->degraded = 0;
3496 if (mddev->pers->sync_request == NULL) {
3497 /* this is now an array without redundancy, so
3498 * it must always be in_sync
3501 del_timer_sync(&mddev->safemode_timer);
3504 mddev_resume(mddev);
3505 set_bit(MD_CHANGE_DEVS, &mddev->flags);
3506 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3507 md_wakeup_thread(mddev->thread);
3508 sysfs_notify(&mddev->kobj, NULL, "level");
3509 md_new_event(mddev);
3513 static struct md_sysfs_entry md_level =
3514 __ATTR(level, S_IRUGO|S_IWUSR, level_show, level_store);
3518 layout_show(struct mddev *mddev, char *page)
3520 /* just a number, not meaningful for all levels */
3521 if (mddev->reshape_position != MaxSector &&
3522 mddev->layout != mddev->new_layout)
3523 return sprintf(page, "%d (%d)\n",
3524 mddev->new_layout, mddev->layout);
3525 return sprintf(page, "%d\n", mddev->layout);
3529 layout_store(struct mddev *mddev, const char *buf, size_t len)
3532 unsigned long n = simple_strtoul(buf, &e, 10);
3534 if (!*buf || (*e && *e != '\n'))
3539 if (mddev->pers->check_reshape == NULL)
3541 mddev->new_layout = n;
3542 err = mddev->pers->check_reshape(mddev);
3544 mddev->new_layout = mddev->layout;
3548 mddev->new_layout = n;
3549 if (mddev->reshape_position == MaxSector)
3554 static struct md_sysfs_entry md_layout =
3555 __ATTR(layout, S_IRUGO|S_IWUSR, layout_show, layout_store);
3559 raid_disks_show(struct mddev *mddev, char *page)
3561 if (mddev->raid_disks == 0)
3563 if (mddev->reshape_position != MaxSector &&
3564 mddev->delta_disks != 0)
3565 return sprintf(page, "%d (%d)\n", mddev->raid_disks,
3566 mddev->raid_disks - mddev->delta_disks);
3567 return sprintf(page, "%d\n", mddev->raid_disks);
3570 static int update_raid_disks(struct mddev *mddev, int raid_disks);
3573 raid_disks_store(struct mddev *mddev, const char *buf, size_t len)
3577 unsigned long n = simple_strtoul(buf, &e, 10);
3579 if (!*buf || (*e && *e != '\n'))
3583 rv = update_raid_disks(mddev, n);
3584 else if (mddev->reshape_position != MaxSector) {
3585 int olddisks = mddev->raid_disks - mddev->delta_disks;
3586 mddev->delta_disks = n - olddisks;
3587 mddev->raid_disks = n;
3589 mddev->raid_disks = n;
3590 return rv ? rv : len;
3592 static struct md_sysfs_entry md_raid_disks =
3593 __ATTR(raid_disks, S_IRUGO|S_IWUSR, raid_disks_show, raid_disks_store);
3596 chunk_size_show(struct mddev *mddev, char *page)
3598 if (mddev->reshape_position != MaxSector &&
3599 mddev->chunk_sectors != mddev->new_chunk_sectors)
3600 return sprintf(page, "%d (%d)\n",
3601 mddev->new_chunk_sectors << 9,
3602 mddev->chunk_sectors << 9);
3603 return sprintf(page, "%d\n", mddev->chunk_sectors << 9);
3607 chunk_size_store(struct mddev *mddev, const char *buf, size_t len)
3610 unsigned long n = simple_strtoul(buf, &e, 10);
3612 if (!*buf || (*e && *e != '\n'))
3617 if (mddev->pers->check_reshape == NULL)
3619 mddev->new_chunk_sectors = n >> 9;
3620 err = mddev->pers->check_reshape(mddev);
3622 mddev->new_chunk_sectors = mddev->chunk_sectors;
3626 mddev->new_chunk_sectors = n >> 9;
3627 if (mddev->reshape_position == MaxSector)
3628 mddev->chunk_sectors = n >> 9;
3632 static struct md_sysfs_entry md_chunk_size =
3633 __ATTR(chunk_size, S_IRUGO|S_IWUSR, chunk_size_show, chunk_size_store);
3636 resync_start_show(struct mddev *mddev, char *page)
3638 if (mddev->recovery_cp == MaxSector)
3639 return sprintf(page, "none\n");
3640 return sprintf(page, "%llu\n", (unsigned long long)mddev->recovery_cp);
3644 resync_start_store(struct mddev *mddev, const char *buf, size_t len)
3647 unsigned long long n = simple_strtoull(buf, &e, 10);
3649 if (mddev->pers && !test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
3651 if (cmd_match(buf, "none"))
3653 else if (!*buf || (*e && *e != '\n'))
3656 mddev->recovery_cp = n;
3659 static struct md_sysfs_entry md_resync_start =
3660 __ATTR(resync_start, S_IRUGO|S_IWUSR, resync_start_show, resync_start_store);
3663 * The array state can be:
3666 * No devices, no size, no level
3667 * Equivalent to STOP_ARRAY ioctl
3669 * May have some settings, but array is not active
3670 * all IO results in error
3671 * When written, doesn't tear down array, but just stops it
3672 * suspended (not supported yet)
3673 * All IO requests will block. The array can be reconfigured.
3674 * Writing this, if accepted, will block until array is quiescent
3676 * no resync can happen. no superblocks get written.
3677 * write requests fail
3679 * like readonly, but behaves like 'clean' on a write request.
3681 * clean - no pending writes, but otherwise active.
3682 * When written to inactive array, starts without resync
3683 * If a write request arrives then
3684 * if metadata is known, mark 'dirty' and switch to 'active'.
3685 * if not known, block and switch to write-pending
3686 * If written to an active array that has pending writes, then fails.
3688 * fully active: IO and resync can be happening.
3689 * When written to inactive array, starts with resync
3692 * clean, but writes are blocked waiting for 'active' to be written.
3695 * like active, but no writes have been seen for a while (100msec).
3698 enum array_state { clear, inactive, suspended, readonly, read_auto, clean, active,
3699 write_pending, active_idle, bad_word};
3700 static char *array_states[] = {
3701 "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active",
3702 "write-pending", "active-idle", NULL };
3704 static int match_word(const char *word, char **list)
3707 for (n=0; list[n]; n++)
3708 if (cmd_match(word, list[n]))
3714 array_state_show(struct mddev *mddev, char *page)
3716 enum array_state st = inactive;
3729 else if (test_bit(MD_CHANGE_PENDING, &mddev->flags))
3731 else if (mddev->safemode)
3737 if (list_empty(&mddev->disks) &&
3738 mddev->raid_disks == 0 &&
3739 mddev->dev_sectors == 0)
3744 return sprintf(page, "%s\n", array_states[st]);
3747 static int do_md_stop(struct mddev * mddev, int ro, int is_open);
3748 static int md_set_readonly(struct mddev * mddev, int is_open);
3749 static int do_md_run(struct mddev * mddev);
3750 static int restart_array(struct mddev *mddev);
3753 array_state_store(struct mddev *mddev, const char *buf, size_t len)
3756 enum array_state st = match_word(buf, array_states);
3761 /* stopping an active array */
3762 if (atomic_read(&mddev->openers) > 0)
3764 err = do_md_stop(mddev, 0, 0);
3767 /* stopping an active array */
3769 if (atomic_read(&mddev->openers) > 0)
3771 err = do_md_stop(mddev, 2, 0);
3773 err = 0; /* already inactive */
3776 break; /* not supported yet */
3779 err = md_set_readonly(mddev, 0);
3782 set_disk_ro(mddev->gendisk, 1);
3783 err = do_md_run(mddev);
3789 err = md_set_readonly(mddev, 0);
3790 else if (mddev->ro == 1)
3791 err = restart_array(mddev);
3794 set_disk_ro(mddev->gendisk, 0);
3798 err = do_md_run(mddev);
3803 restart_array(mddev);
3804 spin_lock_irq(&mddev->write_lock);
3805 if (atomic_read(&mddev->writes_pending) == 0) {
3806 if (mddev->in_sync == 0) {
3808 if (mddev->safemode == 1)
3809 mddev->safemode = 0;
3810 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
3815 spin_unlock_irq(&mddev->write_lock);
3821 restart_array(mddev);
3822 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
3823 wake_up(&mddev->sb_wait);
3827 set_disk_ro(mddev->gendisk, 0);
3828 err = do_md_run(mddev);
3833 /* these cannot be set */
3839 if (mddev->hold_active == UNTIL_IOCTL)
3840 mddev->hold_active = 0;
3841 sysfs_notify_dirent_safe(mddev->sysfs_state);
3845 static struct md_sysfs_entry md_array_state =
3846 __ATTR(array_state, S_IRUGO|S_IWUSR, array_state_show, array_state_store);
3849 max_corrected_read_errors_show(struct mddev *mddev, char *page) {
3850 return sprintf(page, "%d\n",
3851 atomic_read(&mddev->max_corr_read_errors));
3855 max_corrected_read_errors_store(struct mddev *mddev, const char *buf, size_t len)
3858 unsigned long n = simple_strtoul(buf, &e, 10);
3860 if (*buf && (*e == 0 || *e == '\n')) {
3861 atomic_set(&mddev->max_corr_read_errors, n);
3867 static struct md_sysfs_entry max_corr_read_errors =
3868 __ATTR(max_read_errors, S_IRUGO|S_IWUSR, max_corrected_read_errors_show,
3869 max_corrected_read_errors_store);
3872 null_show(struct mddev *mddev, char *page)
3878 new_dev_store(struct mddev *mddev, const char *buf, size_t len)
3880 /* buf must be %d:%d\n? giving major and minor numbers */
3881 /* The new device is added to the array.
3882 * If the array has a persistent superblock, we read the
3883 * superblock to initialise info and check validity.
3884 * Otherwise, only checking done is that in bind_rdev_to_array,
3885 * which mainly checks size.
3888 int major = simple_strtoul(buf, &e, 10);
3891 struct md_rdev *rdev;
3894 if (!*buf || *e != ':' || !e[1] || e[1] == '\n')
3896 minor = simple_strtoul(e+1, &e, 10);
3897 if (*e && *e != '\n')
3899 dev = MKDEV(major, minor);
3900 if (major != MAJOR(dev) ||
3901 minor != MINOR(dev))
3905 if (mddev->persistent) {
3906 rdev = md_import_device(dev, mddev->major_version,
3907 mddev->minor_version);
3908 if (!IS_ERR(rdev) && !list_empty(&mddev->disks)) {
3909 struct md_rdev *rdev0
3910 = list_entry(mddev->disks.next,
3911 struct md_rdev, same_set);
3912 err = super_types[mddev->major_version]
3913 .load_super(rdev, rdev0, mddev->minor_version);
3917 } else if (mddev->external)
3918 rdev = md_import_device(dev, -2, -1);
3920 rdev = md_import_device(dev, -1, -1);
3923 return PTR_ERR(rdev);
3924 err = bind_rdev_to_array(rdev, mddev);
3928 return err ? err : len;
3931 static struct md_sysfs_entry md_new_device =
3932 __ATTR(new_dev, S_IWUSR, null_show, new_dev_store);
3935 bitmap_store(struct mddev *mddev, const char *buf, size_t len)
3938 unsigned long chunk, end_chunk;
3942 /* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */
3944 chunk = end_chunk = simple_strtoul(buf, &end, 0);
3945 if (buf == end) break;
3946 if (*end == '-') { /* range */
3948 end_chunk = simple_strtoul(buf, &end, 0);
3949 if (buf == end) break;
3951 if (*end && !isspace(*end)) break;
3952 bitmap_dirty_bits(mddev->bitmap, chunk, end_chunk);
3953 buf = skip_spaces(end);
3955 bitmap_unplug(mddev->bitmap); /* flush the bits to disk */
3960 static struct md_sysfs_entry md_bitmap =
3961 __ATTR(bitmap_set_bits, S_IWUSR, null_show, bitmap_store);
3964 size_show(struct mddev *mddev, char *page)
3966 return sprintf(page, "%llu\n",
3967 (unsigned long long)mddev->dev_sectors / 2);
3970 static int update_size(struct mddev *mddev, sector_t num_sectors);
3973 size_store(struct mddev *mddev, const char *buf, size_t len)
3975 /* If array is inactive, we can reduce the component size, but
3976 * not increase it (except from 0).
3977 * If array is active, we can try an on-line resize
3980 int err = strict_blocks_to_sectors(buf, §ors);
3985 err = update_size(mddev, sectors);
3986 md_update_sb(mddev, 1);
3988 if (mddev->dev_sectors == 0 ||
3989 mddev->dev_sectors > sectors)
3990 mddev->dev_sectors = sectors;
3994 return err ? err : len;
3997 static struct md_sysfs_entry md_size =
3998 __ATTR(component_size, S_IRUGO|S_IWUSR, size_show, size_store);
4003 * 'none' for arrays with no metadata (good luck...)
4004 * 'external' for arrays with externally managed metadata,
4005 * or N.M for internally known formats
4008 metadata_show(struct mddev *mddev, char *page)
4010 if (mddev->persistent)
4011 return sprintf(page, "%d.%d\n",
4012 mddev->major_version, mddev->minor_version);
4013 else if (mddev->external)
4014 return sprintf(page, "external:%s\n", mddev->metadata_type);
4016 return sprintf(page, "none\n");
4020 metadata_store(struct mddev *mddev, const char *buf, size_t len)
4024 /* Changing the details of 'external' metadata is
4025 * always permitted. Otherwise there must be
4026 * no devices attached to the array.
4028 if (mddev->external && strncmp(buf, "external:", 9) == 0)
4030 else if (!list_empty(&mddev->disks))
4033 if (cmd_match(buf, "none")) {
4034 mddev->persistent = 0;
4035 mddev->external = 0;
4036 mddev->major_version = 0;
4037 mddev->minor_version = 90;
4040 if (strncmp(buf, "external:", 9) == 0) {
4041 size_t namelen = len-9;
4042 if (namelen >= sizeof(mddev->metadata_type))
4043 namelen = sizeof(mddev->metadata_type)-1;
4044 strncpy(mddev->metadata_type, buf+9, namelen);
4045 mddev->metadata_type[namelen] = 0;
4046 if (namelen && mddev->metadata_type[namelen-1] == '\n')
4047 mddev->metadata_type[--namelen] = 0;
4048 mddev->persistent = 0;
4049 mddev->external = 1;
4050 mddev->major_version = 0;
4051 mddev->minor_version = 90;
4054 major = simple_strtoul(buf, &e, 10);
4055 if (e==buf || *e != '.')
4058 minor = simple_strtoul(buf, &e, 10);
4059 if (e==buf || (*e && *e != '\n') )
4061 if (major >= ARRAY_SIZE(super_types) || super_types[major].name == NULL)
4063 mddev->major_version = major;
4064 mddev->minor_version = minor;
4065 mddev->persistent = 1;
4066 mddev->external = 0;
4070 static struct md_sysfs_entry md_metadata =
4071 __ATTR(metadata_version, S_IRUGO|S_IWUSR, metadata_show, metadata_store);
4074 action_show(struct mddev *mddev, char *page)
4076 char *type = "idle";
4077 if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
4079 else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
4080 (!mddev->ro && test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))) {
4081 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
4083 else if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
4084 if (!test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
4086 else if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
4090 } else if (test_bit(MD_RECOVERY_RECOVER, &mddev->recovery))
4093 return sprintf(page, "%s\n", type);
4096 static void reap_sync_thread(struct mddev *mddev);
4099 action_store(struct mddev *mddev, const char *page, size_t len)
4101 if (!mddev->pers || !mddev->pers->sync_request)
4104 if (cmd_match(page, "frozen"))
4105 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4107 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4109 if (cmd_match(page, "idle") || cmd_match(page, "frozen")) {
4110 if (mddev->sync_thread) {
4111 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
4112 reap_sync_thread(mddev);
4114 } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
4115 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
4117 else if (cmd_match(page, "resync"))
4118 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4119 else if (cmd_match(page, "recover")) {
4120 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
4121 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4122 } else if (cmd_match(page, "reshape")) {
4124 if (mddev->pers->start_reshape == NULL)
4126 err = mddev->pers->start_reshape(mddev);
4129 sysfs_notify(&mddev->kobj, NULL, "degraded");
4131 if (cmd_match(page, "check"))
4132 set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
4133 else if (!cmd_match(page, "repair"))
4135 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
4136 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
4138 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4139 md_wakeup_thread(mddev->thread);
4140 sysfs_notify_dirent_safe(mddev->sysfs_action);
4145 mismatch_cnt_show(struct mddev *mddev, char *page)
4147 return sprintf(page, "%llu\n",
4148 (unsigned long long) mddev->resync_mismatches);
4151 static struct md_sysfs_entry md_scan_mode =
4152 __ATTR(sync_action, S_IRUGO|S_IWUSR, action_show, action_store);
4155 static struct md_sysfs_entry md_mismatches = __ATTR_RO(mismatch_cnt);
4158 sync_min_show(struct mddev *mddev, char *page)
4160 return sprintf(page, "%d (%s)\n", speed_min(mddev),
4161 mddev->sync_speed_min ? "local": "system");
4165 sync_min_store(struct mddev *mddev, const char *buf, size_t len)
4169 if (strncmp(buf, "system", 6)==0) {
4170 mddev->sync_speed_min = 0;
4173 min = simple_strtoul(buf, &e, 10);
4174 if (buf == e || (*e && *e != '\n') || min <= 0)
4176 mddev->sync_speed_min = min;
4180 static struct md_sysfs_entry md_sync_min =
4181 __ATTR(sync_speed_min, S_IRUGO|S_IWUSR, sync_min_show, sync_min_store);
4184 sync_max_show(struct mddev *mddev, char *page)
4186 return sprintf(page, "%d (%s)\n", speed_max(mddev),
4187 mddev->sync_speed_max ? "local": "system");
4191 sync_max_store(struct mddev *mddev, const char *buf, size_t len)
4195 if (strncmp(buf, "system", 6)==0) {
4196 mddev->sync_speed_max = 0;
4199 max = simple_strtoul(buf, &e, 10);
4200 if (buf == e || (*e && *e != '\n') || max <= 0)
4202 mddev->sync_speed_max = max;
4206 static struct md_sysfs_entry md_sync_max =
4207 __ATTR(sync_speed_max, S_IRUGO|S_IWUSR, sync_max_show, sync_max_store);
4210 degraded_show(struct mddev *mddev, char *page)
4212 return sprintf(page, "%d\n", mddev->degraded);
4214 static struct md_sysfs_entry md_degraded = __ATTR_RO(degraded);
4217 sync_force_parallel_show(struct mddev *mddev, char *page)
4219 return sprintf(page, "%d\n", mddev->parallel_resync);
4223 sync_force_parallel_store(struct mddev *mddev, const char *buf, size_t len)
4227 if (strict_strtol(buf, 10, &n))
4230 if (n != 0 && n != 1)
4233 mddev->parallel_resync = n;
4235 if (mddev->sync_thread)
4236 wake_up(&resync_wait);
4241 /* force parallel resync, even with shared block devices */
4242 static struct md_sysfs_entry md_sync_force_parallel =
4243 __ATTR(sync_force_parallel, S_IRUGO|S_IWUSR,
4244 sync_force_parallel_show, sync_force_parallel_store);
4247 sync_speed_show(struct mddev *mddev, char *page)
4249 unsigned long resync, dt, db;
4250 if (mddev->curr_resync == 0)
4251 return sprintf(page, "none\n");
4252 resync = mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active);
4253 dt = (jiffies - mddev->resync_mark) / HZ;
4255 db = resync - mddev->resync_mark_cnt;
4256 return sprintf(page, "%lu\n", db/dt/2); /* K/sec */
4259 static struct md_sysfs_entry md_sync_speed = __ATTR_RO(sync_speed);
4262 sync_completed_show(struct mddev *mddev, char *page)
4264 unsigned long long max_sectors, resync;
4266 if (!test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4267 return sprintf(page, "none\n");
4269 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
4270 max_sectors = mddev->resync_max_sectors;
4272 max_sectors = mddev->dev_sectors;
4274 resync = mddev->curr_resync_completed;
4275 return sprintf(page, "%llu / %llu\n", resync, max_sectors);
4278 static struct md_sysfs_entry md_sync_completed = __ATTR_RO(sync_completed);
4281 min_sync_show(struct mddev *mddev, char *page)
4283 return sprintf(page, "%llu\n",
4284 (unsigned long long)mddev->resync_min);
4287 min_sync_store(struct mddev *mddev, const char *buf, size_t len)
4289 unsigned long long min;
4290 if (strict_strtoull(buf, 10, &min))
4292 if (min > mddev->resync_max)
4294 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4297 /* Must be a multiple of chunk_size */
4298 if (mddev->chunk_sectors) {
4299 sector_t temp = min;
4300 if (sector_div(temp, mddev->chunk_sectors))
4303 mddev->resync_min = min;
4308 static struct md_sysfs_entry md_min_sync =
4309 __ATTR(sync_min, S_IRUGO|S_IWUSR, min_sync_show, min_sync_store);
4312 max_sync_show(struct mddev *mddev, char *page)
4314 if (mddev->resync_max == MaxSector)
4315 return sprintf(page, "max\n");
4317 return sprintf(page, "%llu\n",
4318 (unsigned long long)mddev->resync_max);
4321 max_sync_store(struct mddev *mddev, const char *buf, size_t len)
4323 if (strncmp(buf, "max", 3) == 0)
4324 mddev->resync_max = MaxSector;
4326 unsigned long long max;
4327 if (strict_strtoull(buf, 10, &max))
4329 if (max < mddev->resync_min)
4331 if (max < mddev->resync_max &&
4333 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4336 /* Must be a multiple of chunk_size */
4337 if (mddev->chunk_sectors) {
4338 sector_t temp = max;
4339 if (sector_div(temp, mddev->chunk_sectors))
4342 mddev->resync_max = max;
4344 wake_up(&mddev->recovery_wait);
4348 static struct md_sysfs_entry md_max_sync =
4349 __ATTR(sync_max, S_IRUGO|S_IWUSR, max_sync_show, max_sync_store);
4352 suspend_lo_show(struct mddev *mddev, char *page)
4354 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_lo);
4358 suspend_lo_store(struct mddev *mddev, const char *buf, size_t len)
4361 unsigned long long new = simple_strtoull(buf, &e, 10);
4362 unsigned long long old = mddev->suspend_lo;
4364 if (mddev->pers == NULL ||
4365 mddev->pers->quiesce == NULL)
4367 if (buf == e || (*e && *e != '\n'))
4370 mddev->suspend_lo = new;
4372 /* Shrinking suspended region */
4373 mddev->pers->quiesce(mddev, 2);
4375 /* Expanding suspended region - need to wait */
4376 mddev->pers->quiesce(mddev, 1);
4377 mddev->pers->quiesce(mddev, 0);
4381 static struct md_sysfs_entry md_suspend_lo =
4382 __ATTR(suspend_lo, S_IRUGO|S_IWUSR, suspend_lo_show, suspend_lo_store);
4386 suspend_hi_show(struct mddev *mddev, char *page)
4388 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_hi);
4392 suspend_hi_store(struct mddev *mddev, const char *buf, size_t len)
4395 unsigned long long new = simple_strtoull(buf, &e, 10);
4396 unsigned long long old = mddev->suspend_hi;
4398 if (mddev->pers == NULL ||
4399 mddev->pers->quiesce == NULL)
4401 if (buf == e || (*e && *e != '\n'))
4404 mddev->suspend_hi = new;
4406 /* Shrinking suspended region */
4407 mddev->pers->quiesce(mddev, 2);
4409 /* Expanding suspended region - need to wait */
4410 mddev->pers->quiesce(mddev, 1);
4411 mddev->pers->quiesce(mddev, 0);
4415 static struct md_sysfs_entry md_suspend_hi =
4416 __ATTR(suspend_hi, S_IRUGO|S_IWUSR, suspend_hi_show, suspend_hi_store);
4419 reshape_position_show(struct mddev *mddev, char *page)
4421 if (mddev->reshape_position != MaxSector)
4422 return sprintf(page, "%llu\n",
4423 (unsigned long long)mddev->reshape_position);
4424 strcpy(page, "none\n");
4429 reshape_position_store(struct mddev *mddev, const char *buf, size_t len)
4432 unsigned long long new = simple_strtoull(buf, &e, 10);
4435 if (buf == e || (*e && *e != '\n'))
4437 mddev->reshape_position = new;
4438 mddev->delta_disks = 0;
4439 mddev->new_level = mddev->level;
4440 mddev->new_layout = mddev->layout;
4441 mddev->new_chunk_sectors = mddev->chunk_sectors;
4445 static struct md_sysfs_entry md_reshape_position =
4446 __ATTR(reshape_position, S_IRUGO|S_IWUSR, reshape_position_show,
4447 reshape_position_store);
4450 array_size_show(struct mddev *mddev, char *page)
4452 if (mddev->external_size)
4453 return sprintf(page, "%llu\n",
4454 (unsigned long long)mddev->array_sectors/2);
4456 return sprintf(page, "default\n");
4460 array_size_store(struct mddev *mddev, const char *buf, size_t len)
4464 if (strncmp(buf, "default", 7) == 0) {
4466 sectors = mddev->pers->size(mddev, 0, 0);
4468 sectors = mddev->array_sectors;
4470 mddev->external_size = 0;
4472 if (strict_blocks_to_sectors(buf, §ors) < 0)
4474 if (mddev->pers && mddev->pers->size(mddev, 0, 0) < sectors)
4477 mddev->external_size = 1;
4480 mddev->array_sectors = sectors;
4482 set_capacity(mddev->gendisk, mddev->array_sectors);
4483 revalidate_disk(mddev->gendisk);
4488 static struct md_sysfs_entry md_array_size =
4489 __ATTR(array_size, S_IRUGO|S_IWUSR, array_size_show,
4492 static struct attribute *md_default_attrs[] = {
4495 &md_raid_disks.attr,
4496 &md_chunk_size.attr,
4498 &md_resync_start.attr,
4500 &md_new_device.attr,
4501 &md_safe_delay.attr,
4502 &md_array_state.attr,
4503 &md_reshape_position.attr,
4504 &md_array_size.attr,
4505 &max_corr_read_errors.attr,
4509 static struct attribute *md_redundancy_attrs[] = {
4511 &md_mismatches.attr,
4514 &md_sync_speed.attr,
4515 &md_sync_force_parallel.attr,
4516 &md_sync_completed.attr,
4519 &md_suspend_lo.attr,
4520 &md_suspend_hi.attr,
4525 static struct attribute_group md_redundancy_group = {
4527 .attrs = md_redundancy_attrs,
4532 md_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
4534 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
4535 struct mddev *mddev = container_of(kobj, struct mddev, kobj);
4540 spin_lock(&all_mddevs_lock);
4541 if (list_empty(&mddev->all_mddevs)) {
4542 spin_unlock(&all_mddevs_lock);
4546 spin_unlock(&all_mddevs_lock);
4548 rv = mddev_lock(mddev);
4550 rv = entry->show(mddev, page);
4551 mddev_unlock(mddev);
4558 md_attr_store(struct kobject *kobj, struct attribute *attr,
4559 const char *page, size_t length)
4561 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
4562 struct mddev *mddev = container_of(kobj, struct mddev, kobj);
4567 if (!capable(CAP_SYS_ADMIN))
4569 spin_lock(&all_mddevs_lock);
4570 if (list_empty(&mddev->all_mddevs)) {
4571 spin_unlock(&all_mddevs_lock);
4575 spin_unlock(&all_mddevs_lock);
4576 rv = mddev_lock(mddev);
4578 rv = entry->store(mddev, page, length);
4579 mddev_unlock(mddev);
4585 static void md_free(struct kobject *ko)
4587 struct mddev *mddev = container_of(ko, struct mddev, kobj);
4589 if (mddev->sysfs_state)
4590 sysfs_put(mddev->sysfs_state);
4592 if (mddev->gendisk) {
4593 del_gendisk(mddev->gendisk);
4594 put_disk(mddev->gendisk);
4597 blk_cleanup_queue(mddev->queue);
4602 static const struct sysfs_ops md_sysfs_ops = {
4603 .show = md_attr_show,
4604 .store = md_attr_store,
4606 static struct kobj_type md_ktype = {
4608 .sysfs_ops = &md_sysfs_ops,
4609 .default_attrs = md_default_attrs,
4614 static void mddev_delayed_delete(struct work_struct *ws)
4616 struct mddev *mddev = container_of(ws, struct mddev, del_work);
4618 sysfs_remove_group(&mddev->kobj, &md_bitmap_group);
4619 kobject_del(&mddev->kobj);
4620 kobject_put(&mddev->kobj);
4623 static int md_alloc(dev_t dev, char *name)
4625 static DEFINE_MUTEX(disks_mutex);
4626 struct mddev *mddev = mddev_find(dev);
4627 struct gendisk *disk;
4636 partitioned = (MAJOR(mddev->unit) != MD_MAJOR);
4637 shift = partitioned ? MdpMinorShift : 0;
4638 unit = MINOR(mddev->unit) >> shift;
4640 /* wait for any previous instance of this device to be
4641 * completely removed (mddev_delayed_delete).
4643 flush_workqueue(md_misc_wq);
4645 mutex_lock(&disks_mutex);
4651 /* Need to ensure that 'name' is not a duplicate.
4653 struct mddev *mddev2;
4654 spin_lock(&all_mddevs_lock);
4656 list_for_each_entry(mddev2, &all_mddevs, all_mddevs)
4657 if (mddev2->gendisk &&
4658 strcmp(mddev2->gendisk->disk_name, name) == 0) {
4659 spin_unlock(&all_mddevs_lock);
4662 spin_unlock(&all_mddevs_lock);
4666 mddev->queue = blk_alloc_queue(GFP_KERNEL);
4669 mddev->queue->queuedata = mddev;
4671 blk_queue_make_request(mddev->queue, md_make_request);
4672 blk_set_stacking_limits(&mddev->queue->limits);
4674 disk = alloc_disk(1 << shift);
4676 blk_cleanup_queue(mddev->queue);
4677 mddev->queue = NULL;
4680 disk->major = MAJOR(mddev->unit);
4681 disk->first_minor = unit << shift;
4683 strcpy(disk->disk_name, name);
4684 else if (partitioned)
4685 sprintf(disk->disk_name, "md_d%d", unit);
4687 sprintf(disk->disk_name, "md%d", unit);
4688 disk->fops = &md_fops;
4689 disk->private_data = mddev;
4690 disk->queue = mddev->queue;
4691 blk_queue_flush(mddev->queue, REQ_FLUSH | REQ_FUA);
4692 /* Allow extended partitions. This makes the
4693 * 'mdp' device redundant, but we can't really
4696 disk->flags |= GENHD_FL_EXT_DEVT;
4697 mddev->gendisk = disk;
4698 /* As soon as we call add_disk(), another thread could get
4699 * through to md_open, so make sure it doesn't get too far
4701 mutex_lock(&mddev->open_mutex);
4704 error = kobject_init_and_add(&mddev->kobj, &md_ktype,
4705 &disk_to_dev(disk)->kobj, "%s", "md");
4707 /* This isn't possible, but as kobject_init_and_add is marked
4708 * __must_check, we must do something with the result
4710 printk(KERN_WARNING "md: cannot register %s/md - name in use\n",
4714 if (mddev->kobj.sd &&
4715 sysfs_create_group(&mddev->kobj, &md_bitmap_group))
4716 printk(KERN_DEBUG "pointless warning\n");
4717 mutex_unlock(&mddev->open_mutex);
4719 mutex_unlock(&disks_mutex);
4720 if (!error && mddev->kobj.sd) {
4721 kobject_uevent(&mddev->kobj, KOBJ_ADD);
4722 mddev->sysfs_state = sysfs_get_dirent_safe(mddev->kobj.sd, "array_state");
4728 static struct kobject *md_probe(dev_t dev, int *part, void *data)
4730 md_alloc(dev, NULL);
4734 static int add_named_array(const char *val, struct kernel_param *kp)
4736 /* val must be "md_*" where * is not all digits.
4737 * We allocate an array with a large free minor number, and
4738 * set the name to val. val must not already be an active name.
4740 int len = strlen(val);
4741 char buf[DISK_NAME_LEN];
4743 while (len && val[len-1] == '\n')
4745 if (len >= DISK_NAME_LEN)
4747 strlcpy(buf, val, len+1);
4748 if (strncmp(buf, "md_", 3) != 0)
4750 return md_alloc(0, buf);
4753 static void md_safemode_timeout(unsigned long data)
4755 struct mddev *mddev = (struct mddev *) data;
4757 if (!atomic_read(&mddev->writes_pending)) {
4758 mddev->safemode = 1;
4759 if (mddev->external)
4760 sysfs_notify_dirent_safe(mddev->sysfs_state);
4762 md_wakeup_thread(mddev->thread);
4765 static int start_dirty_degraded;
4767 int md_run(struct mddev *mddev)
4770 struct md_rdev *rdev;
4771 struct md_personality *pers;
4773 if (list_empty(&mddev->disks))
4774 /* cannot run an array with no devices.. */
4779 /* Cannot run until previous stop completes properly */
4780 if (mddev->sysfs_active)
4784 * Analyze all RAID superblock(s)
4786 if (!mddev->raid_disks) {
4787 if (!mddev->persistent)
4792 if (mddev->level != LEVEL_NONE)
4793 request_module("md-level-%d", mddev->level);
4794 else if (mddev->clevel[0])
4795 request_module("md-%s", mddev->clevel);
4798 * Drop all container device buffers, from now on
4799 * the only valid external interface is through the md
4802 rdev_for_each(rdev, mddev) {
4803 if (test_bit(Faulty, &rdev->flags))
4805 sync_blockdev(rdev->bdev);
4806 invalidate_bdev(rdev->bdev);
4808 /* perform some consistency tests on the device.
4809 * We don't want the data to overlap the metadata,
4810 * Internal Bitmap issues have been handled elsewhere.
4812 if (rdev->meta_bdev) {
4813 /* Nothing to check */;
4814 } else if (rdev->data_offset < rdev->sb_start) {
4815 if (mddev->dev_sectors &&
4816 rdev->data_offset + mddev->dev_sectors
4818 printk("md: %s: data overlaps metadata\n",
4823 if (rdev->sb_start + rdev->sb_size/512
4824 > rdev->data_offset) {
4825 printk("md: %s: metadata overlaps data\n",
4830 sysfs_notify_dirent_safe(rdev->sysfs_state);
4833 if (mddev->bio_set == NULL)
4834 mddev->bio_set = bioset_create(BIO_POOL_SIZE,
4835 sizeof(struct mddev *));
4837 spin_lock(&pers_lock);
4838 pers = find_pers(mddev->level, mddev->clevel);
4839 if (!pers || !try_module_get(pers->owner)) {
4840 spin_unlock(&pers_lock);
4841 if (mddev->level != LEVEL_NONE)
4842 printk(KERN_WARNING "md: personality for level %d is not loaded!\n",
4845 printk(KERN_WARNING "md: personality for level %s is not loaded!\n",
4850 spin_unlock(&pers_lock);
4851 if (mddev->level != pers->level) {
4852 mddev->level = pers->level;
4853 mddev->new_level = pers->level;
4855 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
4857 if (mddev->reshape_position != MaxSector &&
4858 pers->start_reshape == NULL) {
4859 /* This personality cannot handle reshaping... */
4861 module_put(pers->owner);
4865 if (pers->sync_request) {
4866 /* Warn if this is a potentially silly
4869 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
4870 struct md_rdev *rdev2;
4873 rdev_for_each(rdev, mddev)
4874 rdev_for_each(rdev2, mddev) {
4876 rdev->bdev->bd_contains ==
4877 rdev2->bdev->bd_contains) {
4879 "%s: WARNING: %s appears to be"
4880 " on the same physical disk as"
4883 bdevname(rdev->bdev,b),
4884 bdevname(rdev2->bdev,b2));
4891 "True protection against single-disk"
4892 " failure might be compromised.\n");
4895 mddev->recovery = 0;
4896 /* may be over-ridden by personality */
4897 mddev->resync_max_sectors = mddev->dev_sectors;
4899 mddev->ok_start_degraded = start_dirty_degraded;
4901 if (start_readonly && mddev->ro == 0)
4902 mddev->ro = 2; /* read-only, but switch on first write */
4904 err = mddev->pers->run(mddev);
4906 printk(KERN_ERR "md: pers->run() failed ...\n");
4907 else if (mddev->pers->size(mddev, 0, 0) < mddev->array_sectors) {
4908 WARN_ONCE(!mddev->external_size, "%s: default size too small,"
4909 " but 'external_size' not in effect?\n", __func__);
4911 "md: invalid array_size %llu > default size %llu\n",
4912 (unsigned long long)mddev->array_sectors / 2,
4913 (unsigned long long)mddev->pers->size(mddev, 0, 0) / 2);
4915 mddev->pers->stop(mddev);
4917 if (err == 0 && mddev->pers->sync_request) {
4918 err = bitmap_create(mddev);
4920 printk(KERN_ERR "%s: failed to create bitmap (%d)\n",
4921 mdname(mddev), err);
4922 mddev->pers->stop(mddev);
4926 module_put(mddev->pers->owner);
4928 bitmap_destroy(mddev);
4931 if (mddev->pers->sync_request) {
4932 if (mddev->kobj.sd &&
4933 sysfs_create_group(&mddev->kobj, &md_redundancy_group))
4935 "md: cannot register extra attributes for %s\n",
4937 mddev->sysfs_action = sysfs_get_dirent_safe(mddev->kobj.sd, "sync_action");
4938 } else if (mddev->ro == 2) /* auto-readonly not meaningful */
4941 atomic_set(&mddev->writes_pending,0);
4942 atomic_set(&mddev->max_corr_read_errors,
4943 MD_DEFAULT_MAX_CORRECTED_READ_ERRORS);
4944 mddev->safemode = 0;
4945 mddev->safemode_timer.function = md_safemode_timeout;
4946 mddev->safemode_timer.data = (unsigned long) mddev;
4947 mddev->safemode_delay = (200 * HZ)/1000 +1; /* 200 msec delay */
4951 rdev_for_each(rdev, mddev)
4952 if (rdev->raid_disk >= 0)
4953 if (sysfs_link_rdev(mddev, rdev))
4954 /* failure here is OK */;
4956 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4959 md_update_sb(mddev, 0);
4961 md_new_event(mddev);
4962 sysfs_notify_dirent_safe(mddev->sysfs_state);
4963 sysfs_notify_dirent_safe(mddev->sysfs_action);
4964 sysfs_notify(&mddev->kobj, NULL, "degraded");
4967 EXPORT_SYMBOL_GPL(md_run);
4969 static int do_md_run(struct mddev *mddev)
4973 err = md_run(mddev);
4976 err = bitmap_load(mddev);
4978 bitmap_destroy(mddev);
4982 md_wakeup_thread(mddev->thread);
4983 md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
4985 set_capacity(mddev->gendisk, mddev->array_sectors);
4986 revalidate_disk(mddev->gendisk);
4988 kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
4993 static int restart_array(struct mddev *mddev)
4995 struct gendisk *disk = mddev->gendisk;
4997 /* Complain if it has no devices */
4998 if (list_empty(&mddev->disks))
5004 mddev->safemode = 0;
5006 set_disk_ro(disk, 0);
5007 printk(KERN_INFO "md: %s switched to read-write mode.\n",
5009 /* Kick recovery or resync if necessary */
5010 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5011 md_wakeup_thread(mddev->thread);
5012 md_wakeup_thread(mddev->sync_thread);
5013 sysfs_notify_dirent_safe(mddev->sysfs_state);
5017 /* similar to deny_write_access, but accounts for our holding a reference
5018 * to the file ourselves */
5019 static int deny_bitmap_write_access(struct file * file)
5021 struct inode *inode = file->f_mapping->host;
5023 spin_lock(&inode->i_lock);
5024 if (atomic_read(&inode->i_writecount) > 1) {
5025 spin_unlock(&inode->i_lock);
5028 atomic_set(&inode->i_writecount, -1);
5029 spin_unlock(&inode->i_lock);
5034 void restore_bitmap_write_access(struct file *file)
5036 struct inode *inode = file->f_mapping->host;
5038 spin_lock(&inode->i_lock);
5039 atomic_set(&inode->i_writecount, 1);
5040 spin_unlock(&inode->i_lock);
5043 static void md_clean(struct mddev *mddev)
5045 mddev->array_sectors = 0;
5046 mddev->external_size = 0;
5047 mddev->dev_sectors = 0;
5048 mddev->raid_disks = 0;
5049 mddev->recovery_cp = 0;
5050 mddev->resync_min = 0;
5051 mddev->resync_max = MaxSector;
5052 mddev->reshape_position = MaxSector;
5053 mddev->external = 0;
5054 mddev->persistent = 0;
5055 mddev->level = LEVEL_NONE;
5056 mddev->clevel[0] = 0;
5059 mddev->metadata_type[0] = 0;
5060 mddev->chunk_sectors = 0;
5061 mddev->ctime = mddev->utime = 0;
5063 mddev->max_disks = 0;
5065 mddev->can_decrease_events = 0;
5066 mddev->delta_disks = 0;
5067 mddev->new_level = LEVEL_NONE;
5068 mddev->new_layout = 0;
5069 mddev->new_chunk_sectors = 0;
5070 mddev->curr_resync = 0;
5071 mddev->resync_mismatches = 0;
5072 mddev->suspend_lo = mddev->suspend_hi = 0;
5073 mddev->sync_speed_min = mddev->sync_speed_max = 0;
5074 mddev->recovery = 0;
5077 mddev->degraded = 0;
5078 mddev->safemode = 0;
5079 mddev->merge_check_needed = 0;
5080 mddev->bitmap_info.offset = 0;
5081 mddev->bitmap_info.default_offset = 0;
5082 mddev->bitmap_info.chunksize = 0;
5083 mddev->bitmap_info.daemon_sleep = 0;
5084 mddev->bitmap_info.max_write_behind = 0;
5087 static void __md_stop_writes(struct mddev *mddev)
5089 if (mddev->sync_thread) {
5090 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5091 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5092 reap_sync_thread(mddev);
5095 del_timer_sync(&mddev->safemode_timer);
5097 bitmap_flush(mddev);
5098 md_super_wait(mddev);
5100 if (!mddev->in_sync || mddev->flags) {
5101 /* mark array as shutdown cleanly */
5103 md_update_sb(mddev, 1);
5107 void md_stop_writes(struct mddev *mddev)
5110 __md_stop_writes(mddev);
5111 mddev_unlock(mddev);
5113 EXPORT_SYMBOL_GPL(md_stop_writes);
5115 void md_stop(struct mddev *mddev)
5118 mddev->pers->stop(mddev);
5119 if (mddev->pers->sync_request && mddev->to_remove == NULL)
5120 mddev->to_remove = &md_redundancy_group;
5121 module_put(mddev->pers->owner);
5123 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5125 EXPORT_SYMBOL_GPL(md_stop);
5127 static int md_set_readonly(struct mddev *mddev, int is_open)
5130 mutex_lock(&mddev->open_mutex);
5131 if (atomic_read(&mddev->openers) > is_open) {
5132 printk("md: %s still in use.\n",mdname(mddev));
5137 __md_stop_writes(mddev);
5143 set_disk_ro(mddev->gendisk, 1);
5144 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5145 sysfs_notify_dirent_safe(mddev->sysfs_state);
5149 mutex_unlock(&mddev->open_mutex);
5154 * 0 - completely stop and dis-assemble array
5155 * 2 - stop but do not disassemble array
5157 static int do_md_stop(struct mddev * mddev, int mode, int is_open)
5159 struct gendisk *disk = mddev->gendisk;
5160 struct md_rdev *rdev;
5162 mutex_lock(&mddev->open_mutex);
5163 if (atomic_read(&mddev->openers) > is_open ||
5164 mddev->sysfs_active) {
5165 printk("md: %s still in use.\n",mdname(mddev));
5166 mutex_unlock(&mddev->open_mutex);
5172 set_disk_ro(disk, 0);
5174 __md_stop_writes(mddev);
5176 mddev->queue->merge_bvec_fn = NULL;
5177 mddev->queue->backing_dev_info.congested_fn = NULL;
5179 /* tell userspace to handle 'inactive' */
5180 sysfs_notify_dirent_safe(mddev->sysfs_state);
5182 rdev_for_each(rdev, mddev)
5183 if (rdev->raid_disk >= 0)
5184 sysfs_unlink_rdev(mddev, rdev);
5186 set_capacity(disk, 0);
5187 mutex_unlock(&mddev->open_mutex);
5189 revalidate_disk(disk);
5194 mutex_unlock(&mddev->open_mutex);
5196 * Free resources if final stop
5199 printk(KERN_INFO "md: %s stopped.\n", mdname(mddev));
5201 bitmap_destroy(mddev);
5202 if (mddev->bitmap_info.file) {
5203 restore_bitmap_write_access(mddev->bitmap_info.file);
5204 fput(mddev->bitmap_info.file);
5205 mddev->bitmap_info.file = NULL;
5207 mddev->bitmap_info.offset = 0;
5209 export_array(mddev);
5212 kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
5213 if (mddev->hold_active == UNTIL_STOP)
5214 mddev->hold_active = 0;
5216 blk_integrity_unregister(disk);
5217 md_new_event(mddev);
5218 sysfs_notify_dirent_safe(mddev->sysfs_state);
5223 static void autorun_array(struct mddev *mddev)
5225 struct md_rdev *rdev;
5228 if (list_empty(&mddev->disks))
5231 printk(KERN_INFO "md: running: ");
5233 rdev_for_each(rdev, mddev) {
5234 char b[BDEVNAME_SIZE];
5235 printk("<%s>", bdevname(rdev->bdev,b));
5239 err = do_md_run(mddev);
5241 printk(KERN_WARNING "md: do_md_run() returned %d\n", err);
5242 do_md_stop(mddev, 0, 0);
5247 * lets try to run arrays based on all disks that have arrived
5248 * until now. (those are in pending_raid_disks)
5250 * the method: pick the first pending disk, collect all disks with
5251 * the same UUID, remove all from the pending list and put them into
5252 * the 'same_array' list. Then order this list based on superblock
5253 * update time (freshest comes first), kick out 'old' disks and
5254 * compare superblocks. If everything's fine then run it.
5256 * If "unit" is allocated, then bump its reference count
5258 static void autorun_devices(int part)
5260 struct md_rdev *rdev0, *rdev, *tmp;
5261 struct mddev *mddev;
5262 char b[BDEVNAME_SIZE];
5264 printk(KERN_INFO "md: autorun ...\n");
5265 while (!list_empty(&pending_raid_disks)) {
5268 LIST_HEAD(candidates);
5269 rdev0 = list_entry(pending_raid_disks.next,
5270 struct md_rdev, same_set);
5272 printk(KERN_INFO "md: considering %s ...\n",
5273 bdevname(rdev0->bdev,b));
5274 INIT_LIST_HEAD(&candidates);
5275 rdev_for_each_list(rdev, tmp, &pending_raid_disks)
5276 if (super_90_load(rdev, rdev0, 0) >= 0) {
5277 printk(KERN_INFO "md: adding %s ...\n",
5278 bdevname(rdev->bdev,b));
5279 list_move(&rdev->same_set, &candidates);
5282 * now we have a set of devices, with all of them having
5283 * mostly sane superblocks. It's time to allocate the
5287 dev = MKDEV(mdp_major,
5288 rdev0->preferred_minor << MdpMinorShift);
5289 unit = MINOR(dev) >> MdpMinorShift;
5291 dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
5294 if (rdev0->preferred_minor != unit) {
5295 printk(KERN_INFO "md: unit number in %s is bad: %d\n",
5296 bdevname(rdev0->bdev, b), rdev0->preferred_minor);
5300 md_probe(dev, NULL, NULL);
5301 mddev = mddev_find(dev);
5302 if (!mddev || !mddev->gendisk) {
5306 "md: cannot allocate memory for md drive.\n");
5309 if (mddev_lock(mddev))
5310 printk(KERN_WARNING "md: %s locked, cannot run\n",
5312 else if (mddev->raid_disks || mddev->major_version
5313 || !list_empty(&mddev->disks)) {
5315 "md: %s already running, cannot run %s\n",
5316 mdname(mddev), bdevname(rdev0->bdev,b));
5317 mddev_unlock(mddev);
5319 printk(KERN_INFO "md: created %s\n", mdname(mddev));
5320 mddev->persistent = 1;
5321 rdev_for_each_list(rdev, tmp, &candidates) {
5322 list_del_init(&rdev->same_set);
5323 if (bind_rdev_to_array(rdev, mddev))
5326 autorun_array(mddev);
5327 mddev_unlock(mddev);
5329 /* on success, candidates will be empty, on error
5332 rdev_for_each_list(rdev, tmp, &candidates) {
5333 list_del_init(&rdev->same_set);
5338 printk(KERN_INFO "md: ... autorun DONE.\n");
5340 #endif /* !MODULE */
5342 static int get_version(void __user * arg)
5346 ver.major = MD_MAJOR_VERSION;
5347 ver.minor = MD_MINOR_VERSION;
5348 ver.patchlevel = MD_PATCHLEVEL_VERSION;
5350 if (copy_to_user(arg, &ver, sizeof(ver)))
5356 static int get_array_info(struct mddev * mddev, void __user * arg)
5358 mdu_array_info_t info;
5359 int nr,working,insync,failed,spare;
5360 struct md_rdev *rdev;
5362 nr=working=insync=failed=spare=0;
5363 rdev_for_each(rdev, mddev) {
5365 if (test_bit(Faulty, &rdev->flags))
5369 if (test_bit(In_sync, &rdev->flags))
5376 info.major_version = mddev->major_version;
5377 info.minor_version = mddev->minor_version;
5378 info.patch_version = MD_PATCHLEVEL_VERSION;
5379 info.ctime = mddev->ctime;
5380 info.level = mddev->level;
5381 info.size = mddev->dev_sectors / 2;
5382 if (info.size != mddev->dev_sectors / 2) /* overflow */
5385 info.raid_disks = mddev->raid_disks;
5386 info.md_minor = mddev->md_minor;
5387 info.not_persistent= !mddev->persistent;
5389 info.utime = mddev->utime;
5392 info.state = (1<<MD_SB_CLEAN);
5393 if (mddev->bitmap && mddev->bitmap_info.offset)
5394 info.state = (1<<MD_SB_BITMAP_PRESENT);
5395 info.active_disks = insync;
5396 info.working_disks = working;
5397 info.failed_disks = failed;
5398 info.spare_disks = spare;
5400 info.layout = mddev->layout;
5401 info.chunk_size = mddev->chunk_sectors << 9;
5403 if (copy_to_user(arg, &info, sizeof(info)))
5409 static int get_bitmap_file(struct mddev * mddev, void __user * arg)
5411 mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */
5412 char *ptr, *buf = NULL;
5415 if (md_allow_write(mddev))
5416 file = kmalloc(sizeof(*file), GFP_NOIO);
5418 file = kmalloc(sizeof(*file), GFP_KERNEL);
5423 /* bitmap disabled, zero the first byte and copy out */
5424 if (!mddev->bitmap || !mddev->bitmap->file) {
5425 file->pathname[0] = '\0';
5429 buf = kmalloc(sizeof(file->pathname), GFP_KERNEL);
5433 ptr = d_path(&mddev->bitmap->file->f_path, buf, sizeof(file->pathname));
5437 strcpy(file->pathname, ptr);
5441 if (copy_to_user(arg, file, sizeof(*file)))
5449 static int get_disk_info(struct mddev * mddev, void __user * arg)
5451 mdu_disk_info_t info;
5452 struct md_rdev *rdev;
5454 if (copy_from_user(&info, arg, sizeof(info)))
5457 rdev = find_rdev_nr(mddev, info.number);
5459 info.major = MAJOR(rdev->bdev->bd_dev);
5460 info.minor = MINOR(rdev->bdev->bd_dev);
5461 info.raid_disk = rdev->raid_disk;
5463 if (test_bit(Faulty, &rdev->flags))
5464 info.state |= (1<<MD_DISK_FAULTY);
5465 else if (test_bit(In_sync, &rdev->flags)) {
5466 info.state |= (1<<MD_DISK_ACTIVE);
5467 info.state |= (1<<MD_DISK_SYNC);
5469 if (test_bit(WriteMostly, &rdev->flags))
5470 info.state |= (1<<MD_DISK_WRITEMOSTLY);
5472 info.major = info.minor = 0;
5473 info.raid_disk = -1;
5474 info.state = (1<<MD_DISK_REMOVED);
5477 if (copy_to_user(arg, &info, sizeof(info)))
5483 static int add_new_disk(struct mddev * mddev, mdu_disk_info_t *info)
5485 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
5486 struct md_rdev *rdev;
5487 dev_t dev = MKDEV(info->major,info->minor);
5489 if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
5492 if (!mddev->raid_disks) {
5494 /* expecting a device which has a superblock */
5495 rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
5498 "md: md_import_device returned %ld\n",
5500 return PTR_ERR(rdev);
5502 if (!list_empty(&mddev->disks)) {
5503 struct md_rdev *rdev0
5504 = list_entry(mddev->disks.next,
5505 struct md_rdev, same_set);
5506 err = super_types[mddev->major_version]
5507 .load_super(rdev, rdev0, mddev->minor_version);
5510 "md: %s has different UUID to %s\n",
5511 bdevname(rdev->bdev,b),
5512 bdevname(rdev0->bdev,b2));
5517 err = bind_rdev_to_array(rdev, mddev);
5524 * add_new_disk can be used once the array is assembled
5525 * to add "hot spares". They must already have a superblock
5530 if (!mddev->pers->hot_add_disk) {
5532 "%s: personality does not support diskops!\n",
5536 if (mddev->persistent)
5537 rdev = md_import_device(dev, mddev->major_version,
5538 mddev->minor_version);
5540 rdev = md_import_device(dev, -1, -1);
5543 "md: md_import_device returned %ld\n",
5545 return PTR_ERR(rdev);
5547 /* set saved_raid_disk if appropriate */
5548 if (!mddev->persistent) {
5549 if (info->state & (1<<MD_DISK_SYNC) &&
5550 info->raid_disk < mddev->raid_disks) {
5551 rdev->raid_disk = info->raid_disk;
5552 set_bit(In_sync, &rdev->flags);
5554 rdev->raid_disk = -1;
5556 super_types[mddev->major_version].
5557 validate_super(mddev, rdev);
5558 if ((info->state & (1<<MD_DISK_SYNC)) &&
5559 (!test_bit(In_sync, &rdev->flags) ||
5560 rdev->raid_disk != info->raid_disk)) {
5561 /* This was a hot-add request, but events doesn't
5562 * match, so reject it.
5568 if (test_bit(In_sync, &rdev->flags))
5569 rdev->saved_raid_disk = rdev->raid_disk;
5571 rdev->saved_raid_disk = -1;
5573 clear_bit(In_sync, &rdev->flags); /* just to be sure */
5574 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
5575 set_bit(WriteMostly, &rdev->flags);
5577 clear_bit(WriteMostly, &rdev->flags);
5579 rdev->raid_disk = -1;
5580 err = bind_rdev_to_array(rdev, mddev);
5581 if (!err && !mddev->pers->hot_remove_disk) {
5582 /* If there is hot_add_disk but no hot_remove_disk
5583 * then added disks for geometry changes,
5584 * and should be added immediately.
5586 super_types[mddev->major_version].
5587 validate_super(mddev, rdev);
5588 err = mddev->pers->hot_add_disk(mddev, rdev);
5590 unbind_rdev_from_array(rdev);
5595 sysfs_notify_dirent_safe(rdev->sysfs_state);
5597 md_update_sb(mddev, 1);
5598 if (mddev->degraded)
5599 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
5600 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5602 md_new_event(mddev);
5603 md_wakeup_thread(mddev->thread);
5607 /* otherwise, add_new_disk is only allowed
5608 * for major_version==0 superblocks
5610 if (mddev->major_version != 0) {
5611 printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n",
5616 if (!(info->state & (1<<MD_DISK_FAULTY))) {
5618 rdev = md_import_device(dev, -1, 0);
5621 "md: error, md_import_device() returned %ld\n",
5623 return PTR_ERR(rdev);
5625 rdev->desc_nr = info->number;
5626 if (info->raid_disk < mddev->raid_disks)
5627 rdev->raid_disk = info->raid_disk;
5629 rdev->raid_disk = -1;
5631 if (rdev->raid_disk < mddev->raid_disks)
5632 if (info->state & (1<<MD_DISK_SYNC))
5633 set_bit(In_sync, &rdev->flags);
5635 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
5636 set_bit(WriteMostly, &rdev->flags);
5638 if (!mddev->persistent) {
5639 printk(KERN_INFO "md: nonpersistent superblock ...\n");
5640 rdev->sb_start = i_size_read(rdev->bdev->bd_inode) / 512;
5642 rdev->sb_start = calc_dev_sboffset(rdev);
5643 rdev->sectors = rdev->sb_start;
5645 err = bind_rdev_to_array(rdev, mddev);
5655 static int hot_remove_disk(struct mddev * mddev, dev_t dev)
5657 char b[BDEVNAME_SIZE];
5658 struct md_rdev *rdev;
5660 rdev = find_rdev(mddev, dev);
5664 if (rdev->raid_disk >= 0)
5667 kick_rdev_from_array(rdev);
5668 md_update_sb(mddev, 1);
5669 md_new_event(mddev);
5673 printk(KERN_WARNING "md: cannot remove active disk %s from %s ...\n",
5674 bdevname(rdev->bdev,b), mdname(mddev));
5678 static int hot_add_disk(struct mddev * mddev, dev_t dev)
5680 char b[BDEVNAME_SIZE];
5682 struct md_rdev *rdev;
5687 if (mddev->major_version != 0) {
5688 printk(KERN_WARNING "%s: HOT_ADD may only be used with"
5689 " version-0 superblocks.\n",
5693 if (!mddev->pers->hot_add_disk) {
5695 "%s: personality does not support diskops!\n",
5700 rdev = md_import_device(dev, -1, 0);
5703 "md: error, md_import_device() returned %ld\n",
5708 if (mddev->persistent)
5709 rdev->sb_start = calc_dev_sboffset(rdev);
5711 rdev->sb_start = i_size_read(rdev->bdev->bd_inode) / 512;
5713 rdev->sectors = rdev->sb_start;
5715 if (test_bit(Faulty, &rdev->flags)) {
5717 "md: can not hot-add faulty %s disk to %s!\n",
5718 bdevname(rdev->bdev,b), mdname(mddev));
5722 clear_bit(In_sync, &rdev->flags);
5724 rdev->saved_raid_disk = -1;
5725 err = bind_rdev_to_array(rdev, mddev);
5730 * The rest should better be atomic, we can have disk failures
5731 * noticed in interrupt contexts ...
5734 rdev->raid_disk = -1;
5736 md_update_sb(mddev, 1);
5739 * Kick recovery, maybe this spare has to be added to the
5740 * array immediately.
5742 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5743 md_wakeup_thread(mddev->thread);
5744 md_new_event(mddev);
5752 static int set_bitmap_file(struct mddev *mddev, int fd)
5757 if (!mddev->pers->quiesce)
5759 if (mddev->recovery || mddev->sync_thread)
5761 /* we should be able to change the bitmap.. */
5767 return -EEXIST; /* cannot add when bitmap is present */
5768 mddev->bitmap_info.file = fget(fd);
5770 if (mddev->bitmap_info.file == NULL) {
5771 printk(KERN_ERR "%s: error: failed to get bitmap file\n",
5776 err = deny_bitmap_write_access(mddev->bitmap_info.file);
5778 printk(KERN_ERR "%s: error: bitmap file is already in use\n",
5780 fput(mddev->bitmap_info.file);
5781 mddev->bitmap_info.file = NULL;
5784 mddev->bitmap_info.offset = 0; /* file overrides offset */
5785 } else if (mddev->bitmap == NULL)
5786 return -ENOENT; /* cannot remove what isn't there */
5789 mddev->pers->quiesce(mddev, 1);
5791 err = bitmap_create(mddev);
5793 err = bitmap_load(mddev);
5795 if (fd < 0 || err) {
5796 bitmap_destroy(mddev);
5797 fd = -1; /* make sure to put the file */
5799 mddev->pers->quiesce(mddev, 0);
5802 if (mddev->bitmap_info.file) {
5803 restore_bitmap_write_access(mddev->bitmap_info.file);
5804 fput(mddev->bitmap_info.file);
5806 mddev->bitmap_info.file = NULL;
5813 * set_array_info is used two different ways
5814 * The original usage is when creating a new array.
5815 * In this usage, raid_disks is > 0 and it together with
5816 * level, size, not_persistent,layout,chunksize determine the
5817 * shape of the array.
5818 * This will always create an array with a type-0.90.0 superblock.
5819 * The newer usage is when assembling an array.
5820 * In this case raid_disks will be 0, and the major_version field is
5821 * use to determine which style super-blocks are to be found on the devices.
5822 * The minor and patch _version numbers are also kept incase the
5823 * super_block handler wishes to interpret them.
5825 static int set_array_info(struct mddev * mddev, mdu_array_info_t *info)
5828 if (info->raid_disks == 0) {
5829 /* just setting version number for superblock loading */
5830 if (info->major_version < 0 ||
5831 info->major_version >= ARRAY_SIZE(super_types) ||
5832 super_types[info->major_version].name == NULL) {
5833 /* maybe try to auto-load a module? */
5835 "md: superblock version %d not known\n",
5836 info->major_version);
5839 mddev->major_version = info->major_version;
5840 mddev->minor_version = info->minor_version;
5841 mddev->patch_version = info->patch_version;
5842 mddev->persistent = !info->not_persistent;
5843 /* ensure mddev_put doesn't delete this now that there
5844 * is some minimal configuration.
5846 mddev->ctime = get_seconds();
5849 mddev->major_version = MD_MAJOR_VERSION;
5850 mddev->minor_version = MD_MINOR_VERSION;
5851 mddev->patch_version = MD_PATCHLEVEL_VERSION;
5852 mddev->ctime = get_seconds();
5854 mddev->level = info->level;
5855 mddev->clevel[0] = 0;
5856 mddev->dev_sectors = 2 * (sector_t)info->size;
5857 mddev->raid_disks = info->raid_disks;
5858 /* don't set md_minor, it is determined by which /dev/md* was
5861 if (info->state & (1<<MD_SB_CLEAN))
5862 mddev->recovery_cp = MaxSector;
5864 mddev->recovery_cp = 0;
5865 mddev->persistent = ! info->not_persistent;
5866 mddev->external = 0;
5868 mddev->layout = info->layout;
5869 mddev->chunk_sectors = info->chunk_size >> 9;
5871 mddev->max_disks = MD_SB_DISKS;
5873 if (mddev->persistent)
5875 set_bit(MD_CHANGE_DEVS, &mddev->flags);
5877 mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9;
5878 mddev->bitmap_info.offset = 0;
5880 mddev->reshape_position = MaxSector;
5883 * Generate a 128 bit UUID
5885 get_random_bytes(mddev->uuid, 16);
5887 mddev->new_level = mddev->level;
5888 mddev->new_chunk_sectors = mddev->chunk_sectors;
5889 mddev->new_layout = mddev->layout;
5890 mddev->delta_disks = 0;
5895 void md_set_array_sectors(struct mddev *mddev, sector_t array_sectors)
5897 WARN(!mddev_is_locked(mddev), "%s: unlocked mddev!\n", __func__);
5899 if (mddev->external_size)
5902 mddev->array_sectors = array_sectors;
5904 EXPORT_SYMBOL(md_set_array_sectors);
5906 static int update_size(struct mddev *mddev, sector_t num_sectors)
5908 struct md_rdev *rdev;
5910 int fit = (num_sectors == 0);
5912 if (mddev->pers->resize == NULL)
5914 /* The "num_sectors" is the number of sectors of each device that
5915 * is used. This can only make sense for arrays with redundancy.
5916 * linear and raid0 always use whatever space is available. We can only
5917 * consider changing this number if no resync or reconstruction is
5918 * happening, and if the new size is acceptable. It must fit before the
5919 * sb_start or, if that is <data_offset, it must fit before the size
5920 * of each device. If num_sectors is zero, we find the largest size
5923 if (mddev->sync_thread)
5926 /* Sorry, cannot grow a bitmap yet, just remove it,
5930 rdev_for_each(rdev, mddev) {
5931 sector_t avail = rdev->sectors;
5933 if (fit && (num_sectors == 0 || num_sectors > avail))
5934 num_sectors = avail;
5935 if (avail < num_sectors)
5938 rv = mddev->pers->resize(mddev, num_sectors);
5940 revalidate_disk(mddev->gendisk);
5944 static int update_raid_disks(struct mddev *mddev, int raid_disks)
5947 /* change the number of raid disks */
5948 if (mddev->pers->check_reshape == NULL)
5950 if (raid_disks <= 0 ||
5951 (mddev->max_disks && raid_disks >= mddev->max_disks))
5953 if (mddev->sync_thread || mddev->reshape_position != MaxSector)
5955 mddev->delta_disks = raid_disks - mddev->raid_disks;
5957 rv = mddev->pers->check_reshape(mddev);
5959 mddev->delta_disks = 0;
5965 * update_array_info is used to change the configuration of an
5967 * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
5968 * fields in the info are checked against the array.
5969 * Any differences that cannot be handled will cause an error.
5970 * Normally, only one change can be managed at a time.
5972 static int update_array_info(struct mddev *mddev, mdu_array_info_t *info)
5978 /* calculate expected state,ignoring low bits */
5979 if (mddev->bitmap && mddev->bitmap_info.offset)
5980 state |= (1 << MD_SB_BITMAP_PRESENT);
5982 if (mddev->major_version != info->major_version ||
5983 mddev->minor_version != info->minor_version ||
5984 /* mddev->patch_version != info->patch_version || */
5985 mddev->ctime != info->ctime ||
5986 mddev->level != info->level ||
5987 /* mddev->layout != info->layout || */
5988 !mddev->persistent != info->not_persistent||
5989 mddev->chunk_sectors != info->chunk_size >> 9 ||
5990 /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
5991 ((state^info->state) & 0xfffffe00)
5994 /* Check there is only one change */
5995 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
5997 if (mddev->raid_disks != info->raid_disks)
5999 if (mddev->layout != info->layout)
6001 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT))
6008 if (mddev->layout != info->layout) {
6010 * we don't need to do anything at the md level, the
6011 * personality will take care of it all.
6013 if (mddev->pers->check_reshape == NULL)
6016 mddev->new_layout = info->layout;
6017 rv = mddev->pers->check_reshape(mddev);
6019 mddev->new_layout = mddev->layout;
6023 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
6024 rv = update_size(mddev, (sector_t)info->size * 2);
6026 if (mddev->raid_disks != info->raid_disks)
6027 rv = update_raid_disks(mddev, info->raid_disks);
6029 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) {
6030 if (mddev->pers->quiesce == NULL)
6032 if (mddev->recovery || mddev->sync_thread)
6034 if (info->state & (1<<MD_SB_BITMAP_PRESENT)) {
6035 /* add the bitmap */
6038 if (mddev->bitmap_info.default_offset == 0)
6040 mddev->bitmap_info.offset =
6041 mddev->bitmap_info.default_offset;
6042 mddev->pers->quiesce(mddev, 1);
6043 rv = bitmap_create(mddev);
6045 rv = bitmap_load(mddev);
6047 bitmap_destroy(mddev);
6048 mddev->pers->quiesce(mddev, 0);
6050 /* remove the bitmap */
6053 if (mddev->bitmap->file)
6055 mddev->pers->quiesce(mddev, 1);
6056 bitmap_destroy(mddev);
6057 mddev->pers->quiesce(mddev, 0);
6058 mddev->bitmap_info.offset = 0;
6061 md_update_sb(mddev, 1);
6065 static int set_disk_faulty(struct mddev *mddev, dev_t dev)
6067 struct md_rdev *rdev;
6069 if (mddev->pers == NULL)
6072 rdev = find_rdev(mddev, dev);
6076 md_error(mddev, rdev);
6077 if (!test_bit(Faulty, &rdev->flags))
6083 * We have a problem here : there is no easy way to give a CHS
6084 * virtual geometry. We currently pretend that we have a 2 heads
6085 * 4 sectors (with a BIG number of cylinders...). This drives
6086 * dosfs just mad... ;-)
6088 static int md_getgeo(struct block_device *bdev, struct hd_geometry *geo)
6090 struct mddev *mddev = bdev->bd_disk->private_data;
6094 geo->cylinders = mddev->array_sectors / 8;
6098 static int md_ioctl(struct block_device *bdev, fmode_t mode,
6099 unsigned int cmd, unsigned long arg)
6102 void __user *argp = (void __user *)arg;
6103 struct mddev *mddev = NULL;
6108 case GET_ARRAY_INFO:
6112 if (!capable(CAP_SYS_ADMIN))
6117 * Commands dealing with the RAID driver but not any
6123 err = get_version(argp);
6126 case PRINT_RAID_DEBUG:
6134 autostart_arrays(arg);
6141 * Commands creating/starting a new array:
6144 mddev = bdev->bd_disk->private_data;
6151 err = mddev_lock(mddev);
6154 "md: ioctl lock interrupted, reason %d, cmd %d\n",
6161 case SET_ARRAY_INFO:
6163 mdu_array_info_t info;
6165 memset(&info, 0, sizeof(info));
6166 else if (copy_from_user(&info, argp, sizeof(info))) {
6171 err = update_array_info(mddev, &info);
6173 printk(KERN_WARNING "md: couldn't update"
6174 " array info. %d\n", err);
6179 if (!list_empty(&mddev->disks)) {
6181 "md: array %s already has disks!\n",
6186 if (mddev->raid_disks) {
6188 "md: array %s already initialised!\n",
6193 err = set_array_info(mddev, &info);
6195 printk(KERN_WARNING "md: couldn't set"
6196 " array info. %d\n", err);
6206 * Commands querying/configuring an existing array:
6208 /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
6209 * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */
6210 if ((!mddev->raid_disks && !mddev->external)
6211 && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY
6212 && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE
6213 && cmd != GET_BITMAP_FILE) {
6219 * Commands even a read-only array can execute:
6223 case GET_ARRAY_INFO:
6224 err = get_array_info(mddev, argp);
6227 case GET_BITMAP_FILE:
6228 err = get_bitmap_file(mddev, argp);
6232 err = get_disk_info(mddev, argp);
6235 case RESTART_ARRAY_RW:
6236 err = restart_array(mddev);
6240 err = do_md_stop(mddev, 0, 1);
6244 err = md_set_readonly(mddev, 1);
6248 if (get_user(ro, (int __user *)(arg))) {
6254 /* if the bdev is going readonly the value of mddev->ro
6255 * does not matter, no writes are coming
6260 /* are we are already prepared for writes? */
6264 /* transitioning to readauto need only happen for
6265 * arrays that call md_write_start
6268 err = restart_array(mddev);
6271 set_disk_ro(mddev->gendisk, 0);
6278 * The remaining ioctls are changing the state of the
6279 * superblock, so we do not allow them on read-only arrays.
6280 * However non-MD ioctls (e.g. get-size) will still come through
6281 * here and hit the 'default' below, so only disallow
6282 * 'md' ioctls, and switch to rw mode if started auto-readonly.
6284 if (_IOC_TYPE(cmd) == MD_MAJOR && mddev->ro && mddev->pers) {
6285 if (mddev->ro == 2) {
6287 sysfs_notify_dirent_safe(mddev->sysfs_state);
6288 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6289 md_wakeup_thread(mddev->thread);
6300 mdu_disk_info_t info;
6301 if (copy_from_user(&info, argp, sizeof(info)))
6304 err = add_new_disk(mddev, &info);
6308 case HOT_REMOVE_DISK:
6309 err = hot_remove_disk(mddev, new_decode_dev(arg));
6313 err = hot_add_disk(mddev, new_decode_dev(arg));
6316 case SET_DISK_FAULTY:
6317 err = set_disk_faulty(mddev, new_decode_dev(arg));
6321 err = do_md_run(mddev);
6324 case SET_BITMAP_FILE:
6325 err = set_bitmap_file(mddev, (int)arg);
6335 if (mddev->hold_active == UNTIL_IOCTL &&
6337 mddev->hold_active = 0;
6338 mddev_unlock(mddev);
6347 #ifdef CONFIG_COMPAT
6348 static int md_compat_ioctl(struct block_device *bdev, fmode_t mode,
6349 unsigned int cmd, unsigned long arg)
6352 case HOT_REMOVE_DISK:
6354 case SET_DISK_FAULTY:
6355 case SET_BITMAP_FILE:
6356 /* These take in integer arg, do not convert */
6359 arg = (unsigned long)compat_ptr(arg);
6363 return md_ioctl(bdev, mode, cmd, arg);
6365 #endif /* CONFIG_COMPAT */
6367 static int md_open(struct block_device *bdev, fmode_t mode)
6370 * Succeed if we can lock the mddev, which confirms that
6371 * it isn't being stopped right now.
6373 struct mddev *mddev = mddev_find(bdev->bd_dev);
6376 if (mddev->gendisk != bdev->bd_disk) {
6377 /* we are racing with mddev_put which is discarding this
6381 /* Wait until bdev->bd_disk is definitely gone */
6382 flush_workqueue(md_misc_wq);
6383 /* Then retry the open from the top */
6384 return -ERESTARTSYS;
6386 BUG_ON(mddev != bdev->bd_disk->private_data);
6388 if ((err = mutex_lock_interruptible(&mddev->open_mutex)))
6392 atomic_inc(&mddev->openers);
6393 mutex_unlock(&mddev->open_mutex);
6395 check_disk_change(bdev);
6400 static int md_release(struct gendisk *disk, fmode_t mode)
6402 struct mddev *mddev = disk->private_data;
6405 atomic_dec(&mddev->openers);
6411 static int md_media_changed(struct gendisk *disk)
6413 struct mddev *mddev = disk->private_data;
6415 return mddev->changed;
6418 static int md_revalidate(struct gendisk *disk)
6420 struct mddev *mddev = disk->private_data;
6425 static const struct block_device_operations md_fops =
6427 .owner = THIS_MODULE,
6429 .release = md_release,
6431 #ifdef CONFIG_COMPAT
6432 .compat_ioctl = md_compat_ioctl,
6434 .getgeo = md_getgeo,
6435 .media_changed = md_media_changed,
6436 .revalidate_disk= md_revalidate,
6439 static int md_thread(void * arg)
6441 struct md_thread *thread = arg;
6444 * md_thread is a 'system-thread', it's priority should be very
6445 * high. We avoid resource deadlocks individually in each
6446 * raid personality. (RAID5 does preallocation) We also use RR and
6447 * the very same RT priority as kswapd, thus we will never get
6448 * into a priority inversion deadlock.
6450 * we definitely have to have equal or higher priority than
6451 * bdflush, otherwise bdflush will deadlock if there are too
6452 * many dirty RAID5 blocks.
6455 allow_signal(SIGKILL);
6456 while (!kthread_should_stop()) {
6458 /* We need to wait INTERRUPTIBLE so that
6459 * we don't add to the load-average.
6460 * That means we need to be sure no signals are
6463 if (signal_pending(current))
6464 flush_signals(current);
6466 wait_event_interruptible_timeout
6468 test_bit(THREAD_WAKEUP, &thread->flags)
6469 || kthread_should_stop(),
6472 clear_bit(THREAD_WAKEUP, &thread->flags);
6473 if (!kthread_should_stop())
6474 thread->run(thread->mddev);
6480 void md_wakeup_thread(struct md_thread *thread)
6483 pr_debug("md: waking up MD thread %s.\n", thread->tsk->comm);
6484 set_bit(THREAD_WAKEUP, &thread->flags);
6485 wake_up(&thread->wqueue);
6489 struct md_thread *md_register_thread(void (*run) (struct mddev *), struct mddev *mddev,
6492 struct md_thread *thread;
6494 thread = kzalloc(sizeof(struct md_thread), GFP_KERNEL);
6498 init_waitqueue_head(&thread->wqueue);
6501 thread->mddev = mddev;
6502 thread->timeout = MAX_SCHEDULE_TIMEOUT;
6503 thread->tsk = kthread_run(md_thread, thread,
6505 mdname(thread->mddev),
6506 name ?: mddev->pers->name);
6507 if (IS_ERR(thread->tsk)) {
6514 void md_unregister_thread(struct md_thread **threadp)
6516 struct md_thread *thread = *threadp;
6519 pr_debug("interrupting MD-thread pid %d\n", task_pid_nr(thread->tsk));
6520 /* Locking ensures that mddev_unlock does not wake_up a
6521 * non-existent thread
6523 spin_lock(&pers_lock);
6525 spin_unlock(&pers_lock);
6527 kthread_stop(thread->tsk);
6531 void md_error(struct mddev *mddev, struct md_rdev *rdev)
6538 if (!rdev || test_bit(Faulty, &rdev->flags))
6541 if (!mddev->pers || !mddev->pers->error_handler)
6543 mddev->pers->error_handler(mddev,rdev);
6544 if (mddev->degraded)
6545 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
6546 sysfs_notify_dirent_safe(rdev->sysfs_state);
6547 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6548 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6549 md_wakeup_thread(mddev->thread);
6550 if (mddev->event_work.func)
6551 queue_work(md_misc_wq, &mddev->event_work);
6552 md_new_event_inintr(mddev);
6555 /* seq_file implementation /proc/mdstat */
6557 static void status_unused(struct seq_file *seq)
6560 struct md_rdev *rdev;
6562 seq_printf(seq, "unused devices: ");
6564 list_for_each_entry(rdev, &pending_raid_disks, same_set) {
6565 char b[BDEVNAME_SIZE];
6567 seq_printf(seq, "%s ",
6568 bdevname(rdev->bdev,b));
6571 seq_printf(seq, "<none>");
6573 seq_printf(seq, "\n");
6577 static void status_resync(struct seq_file *seq, struct mddev * mddev)
6579 sector_t max_sectors, resync, res;
6580 unsigned long dt, db;
6583 unsigned int per_milli;
6585 resync = mddev->curr_resync - atomic_read(&mddev->recovery_active);
6587 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
6588 max_sectors = mddev->resync_max_sectors;
6590 max_sectors = mddev->dev_sectors;
6593 * Should not happen.
6599 /* Pick 'scale' such that (resync>>scale)*1000 will fit
6600 * in a sector_t, and (max_sectors>>scale) will fit in a
6601 * u32, as those are the requirements for sector_div.
6602 * Thus 'scale' must be at least 10
6605 if (sizeof(sector_t) > sizeof(unsigned long)) {
6606 while ( max_sectors/2 > (1ULL<<(scale+32)))
6609 res = (resync>>scale)*1000;
6610 sector_div(res, (u32)((max_sectors>>scale)+1));
6614 int i, x = per_milli/50, y = 20-x;
6615 seq_printf(seq, "[");
6616 for (i = 0; i < x; i++)
6617 seq_printf(seq, "=");
6618 seq_printf(seq, ">");
6619 for (i = 0; i < y; i++)
6620 seq_printf(seq, ".");
6621 seq_printf(seq, "] ");
6623 seq_printf(seq, " %s =%3u.%u%% (%llu/%llu)",
6624 (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)?
6626 (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)?
6628 (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
6629 "resync" : "recovery"))),
6630 per_milli/10, per_milli % 10,
6631 (unsigned long long) resync/2,
6632 (unsigned long long) max_sectors/2);
6635 * dt: time from mark until now
6636 * db: blocks written from mark until now
6637 * rt: remaining time
6639 * rt is a sector_t, so could be 32bit or 64bit.
6640 * So we divide before multiply in case it is 32bit and close
6642 * We scale the divisor (db) by 32 to avoid losing precision
6643 * near the end of resync when the number of remaining sectors
6645 * We then divide rt by 32 after multiplying by db to compensate.
6646 * The '+1' avoids division by zero if db is very small.
6648 dt = ((jiffies - mddev->resync_mark) / HZ);
6650 db = (mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active))
6651 - mddev->resync_mark_cnt;
6653 rt = max_sectors - resync; /* number of remaining sectors */
6654 sector_div(rt, db/32+1);
6658 seq_printf(seq, " finish=%lu.%lumin", (unsigned long)rt / 60,
6659 ((unsigned long)rt % 60)/6);
6661 seq_printf(seq, " speed=%ldK/sec", db/2/dt);
6664 static void *md_seq_start(struct seq_file *seq, loff_t *pos)
6666 struct list_head *tmp;
6668 struct mddev *mddev;
6676 spin_lock(&all_mddevs_lock);
6677 list_for_each(tmp,&all_mddevs)
6679 mddev = list_entry(tmp, struct mddev, all_mddevs);
6681 spin_unlock(&all_mddevs_lock);
6684 spin_unlock(&all_mddevs_lock);
6686 return (void*)2;/* tail */
6690 static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
6692 struct list_head *tmp;
6693 struct mddev *next_mddev, *mddev = v;
6699 spin_lock(&all_mddevs_lock);
6701 tmp = all_mddevs.next;
6703 tmp = mddev->all_mddevs.next;
6704 if (tmp != &all_mddevs)
6705 next_mddev = mddev_get(list_entry(tmp,struct mddev,all_mddevs));
6707 next_mddev = (void*)2;
6710 spin_unlock(&all_mddevs_lock);
6718 static void md_seq_stop(struct seq_file *seq, void *v)
6720 struct mddev *mddev = v;
6722 if (mddev && v != (void*)1 && v != (void*)2)
6726 static int md_seq_show(struct seq_file *seq, void *v)
6728 struct mddev *mddev = v;
6730 struct md_rdev *rdev;
6732 if (v == (void*)1) {
6733 struct md_personality *pers;
6734 seq_printf(seq, "Personalities : ");
6735 spin_lock(&pers_lock);
6736 list_for_each_entry(pers, &pers_list, list)
6737 seq_printf(seq, "[%s] ", pers->name);
6739 spin_unlock(&pers_lock);
6740 seq_printf(seq, "\n");
6741 seq->poll_event = atomic_read(&md_event_count);
6744 if (v == (void*)2) {
6749 if (mddev_lock(mddev) < 0)
6752 if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
6753 seq_printf(seq, "%s : %sactive", mdname(mddev),
6754 mddev->pers ? "" : "in");
6757 seq_printf(seq, " (read-only)");
6759 seq_printf(seq, " (auto-read-only)");
6760 seq_printf(seq, " %s", mddev->pers->name);
6764 rdev_for_each(rdev, mddev) {
6765 char b[BDEVNAME_SIZE];
6766 seq_printf(seq, " %s[%d]",
6767 bdevname(rdev->bdev,b), rdev->desc_nr);
6768 if (test_bit(WriteMostly, &rdev->flags))
6769 seq_printf(seq, "(W)");
6770 if (test_bit(Faulty, &rdev->flags)) {
6771 seq_printf(seq, "(F)");
6774 if (rdev->raid_disk < 0)
6775 seq_printf(seq, "(S)"); /* spare */
6776 if (test_bit(Replacement, &rdev->flags))
6777 seq_printf(seq, "(R)");
6778 sectors += rdev->sectors;
6781 if (!list_empty(&mddev->disks)) {
6783 seq_printf(seq, "\n %llu blocks",
6784 (unsigned long long)
6785 mddev->array_sectors / 2);
6787 seq_printf(seq, "\n %llu blocks",
6788 (unsigned long long)sectors / 2);
6790 if (mddev->persistent) {
6791 if (mddev->major_version != 0 ||
6792 mddev->minor_version != 90) {
6793 seq_printf(seq," super %d.%d",
6794 mddev->major_version,
6795 mddev->minor_version);
6797 } else if (mddev->external)
6798 seq_printf(seq, " super external:%s",
6799 mddev->metadata_type);
6801 seq_printf(seq, " super non-persistent");
6804 mddev->pers->status(seq, mddev);
6805 seq_printf(seq, "\n ");
6806 if (mddev->pers->sync_request) {
6807 if (mddev->curr_resync > 2) {
6808 status_resync(seq, mddev);
6809 seq_printf(seq, "\n ");
6810 } else if (mddev->curr_resync == 1 || mddev->curr_resync == 2)
6811 seq_printf(seq, "\tresync=DELAYED\n ");
6812 else if (mddev->recovery_cp < MaxSector)
6813 seq_printf(seq, "\tresync=PENDING\n ");
6816 seq_printf(seq, "\n ");
6818 bitmap_status(seq, mddev->bitmap);
6820 seq_printf(seq, "\n");
6822 mddev_unlock(mddev);
6827 static const struct seq_operations md_seq_ops = {
6828 .start = md_seq_start,
6829 .next = md_seq_next,
6830 .stop = md_seq_stop,
6831 .show = md_seq_show,
6834 static int md_seq_open(struct inode *inode, struct file *file)
6836 struct seq_file *seq;
6839 error = seq_open(file, &md_seq_ops);
6843 seq = file->private_data;
6844 seq->poll_event = atomic_read(&md_event_count);
6848 static unsigned int mdstat_poll(struct file *filp, poll_table *wait)
6850 struct seq_file *seq = filp->private_data;
6853 poll_wait(filp, &md_event_waiters, wait);
6855 /* always allow read */
6856 mask = POLLIN | POLLRDNORM;
6858 if (seq->poll_event != atomic_read(&md_event_count))
6859 mask |= POLLERR | POLLPRI;
6863 static const struct file_operations md_seq_fops = {
6864 .owner = THIS_MODULE,
6865 .open = md_seq_open,
6867 .llseek = seq_lseek,
6868 .release = seq_release_private,
6869 .poll = mdstat_poll,
6872 int register_md_personality(struct md_personality *p)
6874 spin_lock(&pers_lock);
6875 list_add_tail(&p->list, &pers_list);
6876 printk(KERN_INFO "md: %s personality registered for level %d\n", p->name, p->level);
6877 spin_unlock(&pers_lock);
6881 int unregister_md_personality(struct md_personality *p)
6883 printk(KERN_INFO "md: %s personality unregistered\n", p->name);
6884 spin_lock(&pers_lock);
6885 list_del_init(&p->list);
6886 spin_unlock(&pers_lock);
6890 static int is_mddev_idle(struct mddev *mddev, int init)
6892 struct md_rdev * rdev;
6898 rdev_for_each_rcu(rdev, mddev) {
6899 struct gendisk *disk = rdev->bdev->bd_contains->bd_disk;
6900 curr_events = (int)part_stat_read(&disk->part0, sectors[0]) +
6901 (int)part_stat_read(&disk->part0, sectors[1]) -
6902 atomic_read(&disk->sync_io);
6903 /* sync IO will cause sync_io to increase before the disk_stats
6904 * as sync_io is counted when a request starts, and
6905 * disk_stats is counted when it completes.
6906 * So resync activity will cause curr_events to be smaller than
6907 * when there was no such activity.
6908 * non-sync IO will cause disk_stat to increase without
6909 * increasing sync_io so curr_events will (eventually)
6910 * be larger than it was before. Once it becomes
6911 * substantially larger, the test below will cause
6912 * the array to appear non-idle, and resync will slow
6914 * If there is a lot of outstanding resync activity when
6915 * we set last_event to curr_events, then all that activity
6916 * completing might cause the array to appear non-idle
6917 * and resync will be slowed down even though there might
6918 * not have been non-resync activity. This will only
6919 * happen once though. 'last_events' will soon reflect
6920 * the state where there is little or no outstanding
6921 * resync requests, and further resync activity will
6922 * always make curr_events less than last_events.
6925 if (init || curr_events - rdev->last_events > 64) {
6926 rdev->last_events = curr_events;
6934 void md_done_sync(struct mddev *mddev, int blocks, int ok)
6936 /* another "blocks" (512byte) blocks have been synced */
6937 atomic_sub(blocks, &mddev->recovery_active);
6938 wake_up(&mddev->recovery_wait);
6940 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6941 md_wakeup_thread(mddev->thread);
6942 // stop recovery, signal do_sync ....
6947 /* md_write_start(mddev, bi)
6948 * If we need to update some array metadata (e.g. 'active' flag
6949 * in superblock) before writing, schedule a superblock update
6950 * and wait for it to complete.
6952 void md_write_start(struct mddev *mddev, struct bio *bi)
6955 if (bio_data_dir(bi) != WRITE)
6958 BUG_ON(mddev->ro == 1);
6959 if (mddev->ro == 2) {
6960 /* need to switch to read/write */
6962 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6963 md_wakeup_thread(mddev->thread);
6964 md_wakeup_thread(mddev->sync_thread);
6967 atomic_inc(&mddev->writes_pending);
6968 if (mddev->safemode == 1)
6969 mddev->safemode = 0;
6970 if (mddev->in_sync) {
6971 spin_lock_irq(&mddev->write_lock);
6972 if (mddev->in_sync) {
6974 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
6975 set_bit(MD_CHANGE_PENDING, &mddev->flags);
6976 md_wakeup_thread(mddev->thread);
6979 spin_unlock_irq(&mddev->write_lock);
6982 sysfs_notify_dirent_safe(mddev->sysfs_state);
6983 wait_event(mddev->sb_wait,
6984 !test_bit(MD_CHANGE_PENDING, &mddev->flags));
6987 void md_write_end(struct mddev *mddev)
6989 if (atomic_dec_and_test(&mddev->writes_pending)) {
6990 if (mddev->safemode == 2)
6991 md_wakeup_thread(mddev->thread);
6992 else if (mddev->safemode_delay)
6993 mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay);
6997 /* md_allow_write(mddev)
6998 * Calling this ensures that the array is marked 'active' so that writes
6999 * may proceed without blocking. It is important to call this before
7000 * attempting a GFP_KERNEL allocation while holding the mddev lock.
7001 * Must be called with mddev_lock held.
7003 * In the ->external case MD_CHANGE_CLEAN can not be cleared until mddev->lock
7004 * is dropped, so return -EAGAIN after notifying userspace.
7006 int md_allow_write(struct mddev *mddev)
7012 if (!mddev->pers->sync_request)
7015 spin_lock_irq(&mddev->write_lock);
7016 if (mddev->in_sync) {
7018 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
7019 set_bit(MD_CHANGE_PENDING, &mddev->flags);
7020 if (mddev->safemode_delay &&
7021 mddev->safemode == 0)
7022 mddev->safemode = 1;
7023 spin_unlock_irq(&mddev->write_lock);
7024 md_update_sb(mddev, 0);
7025 sysfs_notify_dirent_safe(mddev->sysfs_state);
7027 spin_unlock_irq(&mddev->write_lock);
7029 if (test_bit(MD_CHANGE_PENDING, &mddev->flags))
7034 EXPORT_SYMBOL_GPL(md_allow_write);
7036 #define SYNC_MARKS 10
7037 #define SYNC_MARK_STEP (3*HZ)
7038 void md_do_sync(struct mddev *mddev)
7040 struct mddev *mddev2;
7041 unsigned int currspeed = 0,
7043 sector_t max_sectors,j, io_sectors;
7044 unsigned long mark[SYNC_MARKS];
7045 sector_t mark_cnt[SYNC_MARKS];
7047 struct list_head *tmp;
7048 sector_t last_check;
7050 struct md_rdev *rdev;
7053 /* just incase thread restarts... */
7054 if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
7056 if (mddev->ro) /* never try to sync a read-only array */
7059 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
7060 if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
7061 desc = "data-check";
7062 else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
7063 desc = "requested-resync";
7066 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
7071 /* we overload curr_resync somewhat here.
7072 * 0 == not engaged in resync at all
7073 * 2 == checking that there is no conflict with another sync
7074 * 1 == like 2, but have yielded to allow conflicting resync to
7076 * other == active in resync - this many blocks
7078 * Before starting a resync we must have set curr_resync to
7079 * 2, and then checked that every "conflicting" array has curr_resync
7080 * less than ours. When we find one that is the same or higher
7081 * we wait on resync_wait. To avoid deadlock, we reduce curr_resync
7082 * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
7083 * This will mean we have to start checking from the beginning again.
7088 mddev->curr_resync = 2;
7091 if (kthread_should_stop())
7092 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
7094 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
7096 for_each_mddev(mddev2, tmp) {
7097 if (mddev2 == mddev)
7099 if (!mddev->parallel_resync
7100 && mddev2->curr_resync
7101 && match_mddev_units(mddev, mddev2)) {
7103 if (mddev < mddev2 && mddev->curr_resync == 2) {
7104 /* arbitrarily yield */
7105 mddev->curr_resync = 1;
7106 wake_up(&resync_wait);
7108 if (mddev > mddev2 && mddev->curr_resync == 1)
7109 /* no need to wait here, we can wait the next
7110 * time 'round when curr_resync == 2
7113 /* We need to wait 'interruptible' so as not to
7114 * contribute to the load average, and not to
7115 * be caught by 'softlockup'
7117 prepare_to_wait(&resync_wait, &wq, TASK_INTERRUPTIBLE);
7118 if (!kthread_should_stop() &&
7119 mddev2->curr_resync >= mddev->curr_resync) {
7120 printk(KERN_INFO "md: delaying %s of %s"
7121 " until %s has finished (they"
7122 " share one or more physical units)\n",
7123 desc, mdname(mddev), mdname(mddev2));
7125 if (signal_pending(current))
7126 flush_signals(current);
7128 finish_wait(&resync_wait, &wq);
7131 finish_wait(&resync_wait, &wq);
7134 } while (mddev->curr_resync < 2);
7137 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
7138 /* resync follows the size requested by the personality,
7139 * which defaults to physical size, but can be virtual size
7141 max_sectors = mddev->resync_max_sectors;
7142 mddev->resync_mismatches = 0;
7143 /* we don't use the checkpoint if there's a bitmap */
7144 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
7145 j = mddev->resync_min;
7146 else if (!mddev->bitmap)
7147 j = mddev->recovery_cp;
7149 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
7150 max_sectors = mddev->dev_sectors;
7152 /* recovery follows the physical size of devices */
7153 max_sectors = mddev->dev_sectors;
7156 rdev_for_each_rcu(rdev, mddev)
7157 if (rdev->raid_disk >= 0 &&
7158 !test_bit(Faulty, &rdev->flags) &&
7159 !test_bit(In_sync, &rdev->flags) &&
7160 rdev->recovery_offset < j)
7161 j = rdev->recovery_offset;
7165 printk(KERN_INFO "md: %s of RAID array %s\n", desc, mdname(mddev));
7166 printk(KERN_INFO "md: minimum _guaranteed_ speed:"
7167 " %d KB/sec/disk.\n", speed_min(mddev));
7168 printk(KERN_INFO "md: using maximum available idle IO bandwidth "
7169 "(but not more than %d KB/sec) for %s.\n",
7170 speed_max(mddev), desc);
7172 is_mddev_idle(mddev, 1); /* this initializes IO event counters */
7175 for (m = 0; m < SYNC_MARKS; m++) {
7177 mark_cnt[m] = io_sectors;
7180 mddev->resync_mark = mark[last_mark];
7181 mddev->resync_mark_cnt = mark_cnt[last_mark];
7184 * Tune reconstruction:
7186 window = 32*(PAGE_SIZE/512);
7187 printk(KERN_INFO "md: using %dk window, over a total of %lluk.\n",
7188 window/2, (unsigned long long)max_sectors/2);
7190 atomic_set(&mddev->recovery_active, 0);
7195 "md: resuming %s of %s from checkpoint.\n",
7196 desc, mdname(mddev));
7197 mddev->curr_resync = j;
7199 mddev->curr_resync_completed = j;
7201 while (j < max_sectors) {
7206 if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
7207 ((mddev->curr_resync > mddev->curr_resync_completed &&
7208 (mddev->curr_resync - mddev->curr_resync_completed)
7209 > (max_sectors >> 4)) ||
7210 (j - mddev->curr_resync_completed)*2
7211 >= mddev->resync_max - mddev->curr_resync_completed
7213 /* time to update curr_resync_completed */
7214 wait_event(mddev->recovery_wait,
7215 atomic_read(&mddev->recovery_active) == 0);
7216 mddev->curr_resync_completed = j;
7217 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
7218 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
7221 while (j >= mddev->resync_max && !kthread_should_stop()) {
7222 /* As this condition is controlled by user-space,
7223 * we can block indefinitely, so use '_interruptible'
7224 * to avoid triggering warnings.
7226 flush_signals(current); /* just in case */
7227 wait_event_interruptible(mddev->recovery_wait,
7228 mddev->resync_max > j
7229 || kthread_should_stop());
7232 if (kthread_should_stop())
7235 sectors = mddev->pers->sync_request(mddev, j, &skipped,
7236 currspeed < speed_min(mddev));
7238 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
7242 if (!skipped) { /* actual IO requested */
7243 io_sectors += sectors;
7244 atomic_add(sectors, &mddev->recovery_active);
7247 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
7251 if (j>1) mddev->curr_resync = j;
7252 mddev->curr_mark_cnt = io_sectors;
7253 if (last_check == 0)
7254 /* this is the earliest that rebuild will be
7255 * visible in /proc/mdstat
7257 md_new_event(mddev);
7259 if (last_check + window > io_sectors || j == max_sectors)
7262 last_check = io_sectors;
7264 if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) {
7266 int next = (last_mark+1) % SYNC_MARKS;
7268 mddev->resync_mark = mark[next];
7269 mddev->resync_mark_cnt = mark_cnt[next];
7270 mark[next] = jiffies;
7271 mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active);
7276 if (kthread_should_stop())
7281 * this loop exits only if either when we are slower than
7282 * the 'hard' speed limit, or the system was IO-idle for
7284 * the system might be non-idle CPU-wise, but we only care
7285 * about not overloading the IO subsystem. (things like an
7286 * e2fsck being done on the RAID array should execute fast)
7290 currspeed = ((unsigned long)(io_sectors-mddev->resync_mark_cnt))/2
7291 /((jiffies-mddev->resync_mark)/HZ +1) +1;
7293 if (currspeed > speed_min(mddev)) {
7294 if ((currspeed > speed_max(mddev)) ||
7295 !is_mddev_idle(mddev, 0)) {
7301 printk(KERN_INFO "md: %s: %s done.\n",mdname(mddev), desc);
7303 * this also signals 'finished resyncing' to md_stop
7306 wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
7308 /* tell personality that we are finished */
7309 mddev->pers->sync_request(mddev, max_sectors, &skipped, 1);
7311 if (!test_bit(MD_RECOVERY_CHECK, &mddev->recovery) &&
7312 mddev->curr_resync > 2) {
7313 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
7314 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
7315 if (mddev->curr_resync >= mddev->recovery_cp) {
7317 "md: checkpointing %s of %s.\n",
7318 desc, mdname(mddev));
7319 mddev->recovery_cp =
7320 mddev->curr_resync_completed;
7323 mddev->recovery_cp = MaxSector;
7325 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery))
7326 mddev->curr_resync = MaxSector;
7328 rdev_for_each_rcu(rdev, mddev)
7329 if (rdev->raid_disk >= 0 &&
7330 mddev->delta_disks >= 0 &&
7331 !test_bit(Faulty, &rdev->flags) &&
7332 !test_bit(In_sync, &rdev->flags) &&
7333 rdev->recovery_offset < mddev->curr_resync)
7334 rdev->recovery_offset = mddev->curr_resync;
7339 set_bit(MD_CHANGE_DEVS, &mddev->flags);
7341 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
7342 /* We completed so min/max setting can be forgotten if used. */
7343 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
7344 mddev->resync_min = 0;
7345 mddev->resync_max = MaxSector;
7346 } else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
7347 mddev->resync_min = mddev->curr_resync_completed;
7348 mddev->curr_resync = 0;
7349 wake_up(&resync_wait);
7350 set_bit(MD_RECOVERY_DONE, &mddev->recovery);
7351 md_wakeup_thread(mddev->thread);
7356 * got a signal, exit.
7359 "md: md_do_sync() got signal ... exiting\n");
7360 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
7364 EXPORT_SYMBOL_GPL(md_do_sync);
7366 static int remove_and_add_spares(struct mddev *mddev)
7368 struct md_rdev *rdev;
7372 mddev->curr_resync_completed = 0;
7374 rdev_for_each(rdev, mddev)
7375 if (rdev->raid_disk >= 0 &&
7376 !test_bit(Blocked, &rdev->flags) &&
7377 (test_bit(Faulty, &rdev->flags) ||
7378 ! test_bit(In_sync, &rdev->flags)) &&
7379 atomic_read(&rdev->nr_pending)==0) {
7380 if (mddev->pers->hot_remove_disk(
7381 mddev, rdev) == 0) {
7382 sysfs_unlink_rdev(mddev, rdev);
7383 rdev->raid_disk = -1;
7388 sysfs_notify(&mddev->kobj, NULL,
7392 rdev_for_each(rdev, mddev) {
7393 if (rdev->raid_disk >= 0 &&
7394 !test_bit(In_sync, &rdev->flags) &&
7395 !test_bit(Faulty, &rdev->flags))
7397 if (rdev->raid_disk < 0
7398 && !test_bit(Faulty, &rdev->flags)) {
7399 rdev->recovery_offset = 0;
7401 hot_add_disk(mddev, rdev) == 0) {
7402 if (sysfs_link_rdev(mddev, rdev))
7403 /* failure here is OK */;
7405 md_new_event(mddev);
7406 set_bit(MD_CHANGE_DEVS, &mddev->flags);
7413 static void reap_sync_thread(struct mddev *mddev)
7415 struct md_rdev *rdev;
7417 /* resync has finished, collect result */
7418 md_unregister_thread(&mddev->sync_thread);
7419 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
7420 !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
7422 /* activate any spares */
7423 if (mddev->pers->spare_active(mddev))
7424 sysfs_notify(&mddev->kobj, NULL,
7427 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
7428 mddev->pers->finish_reshape)
7429 mddev->pers->finish_reshape(mddev);
7431 /* If array is no-longer degraded, then any saved_raid_disk
7432 * information must be scrapped. Also if any device is now
7433 * In_sync we must scrape the saved_raid_disk for that device
7434 * do the superblock for an incrementally recovered device
7437 rdev_for_each(rdev, mddev)
7438 if (!mddev->degraded ||
7439 test_bit(In_sync, &rdev->flags))
7440 rdev->saved_raid_disk = -1;
7442 md_update_sb(mddev, 1);
7443 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
7444 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
7445 clear_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
7446 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
7447 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
7448 /* flag recovery needed just to double check */
7449 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7450 sysfs_notify_dirent_safe(mddev->sysfs_action);
7451 md_new_event(mddev);
7452 if (mddev->event_work.func)
7453 queue_work(md_misc_wq, &mddev->event_work);
7457 * This routine is regularly called by all per-raid-array threads to
7458 * deal with generic issues like resync and super-block update.
7459 * Raid personalities that don't have a thread (linear/raid0) do not
7460 * need this as they never do any recovery or update the superblock.
7462 * It does not do any resync itself, but rather "forks" off other threads
7463 * to do that as needed.
7464 * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
7465 * "->recovery" and create a thread at ->sync_thread.
7466 * When the thread finishes it sets MD_RECOVERY_DONE
7467 * and wakeups up this thread which will reap the thread and finish up.
7468 * This thread also removes any faulty devices (with nr_pending == 0).
7470 * The overall approach is:
7471 * 1/ if the superblock needs updating, update it.
7472 * 2/ If a recovery thread is running, don't do anything else.
7473 * 3/ If recovery has finished, clean up, possibly marking spares active.
7474 * 4/ If there are any faulty devices, remove them.
7475 * 5/ If array is degraded, try to add spares devices
7476 * 6/ If array has spares or is not in-sync, start a resync thread.
7478 void md_check_recovery(struct mddev *mddev)
7480 if (mddev->suspended)
7484 bitmap_daemon_work(mddev);
7486 if (signal_pending(current)) {
7487 if (mddev->pers->sync_request && !mddev->external) {
7488 printk(KERN_INFO "md: %s in immediate safe mode\n",
7490 mddev->safemode = 2;
7492 flush_signals(current);
7495 if (mddev->ro && !test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
7498 (mddev->flags & ~ (1<<MD_CHANGE_PENDING)) ||
7499 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
7500 test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
7501 (mddev->external == 0 && mddev->safemode == 1) ||
7502 (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending)
7503 && !mddev->in_sync && mddev->recovery_cp == MaxSector)
7507 if (mddev_trylock(mddev)) {
7511 /* Only thing we do on a ro array is remove
7514 struct md_rdev *rdev;
7515 rdev_for_each(rdev, mddev)
7516 if (rdev->raid_disk >= 0 &&
7517 !test_bit(Blocked, &rdev->flags) &&
7518 test_bit(Faulty, &rdev->flags) &&
7519 atomic_read(&rdev->nr_pending)==0) {
7520 if (mddev->pers->hot_remove_disk(
7521 mddev, rdev) == 0) {
7522 sysfs_unlink_rdev(mddev, rdev);
7523 rdev->raid_disk = -1;
7526 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7530 if (!mddev->external) {
7532 spin_lock_irq(&mddev->write_lock);
7533 if (mddev->safemode &&
7534 !atomic_read(&mddev->writes_pending) &&
7536 mddev->recovery_cp == MaxSector) {
7539 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
7541 if (mddev->safemode == 1)
7542 mddev->safemode = 0;
7543 spin_unlock_irq(&mddev->write_lock);
7545 sysfs_notify_dirent_safe(mddev->sysfs_state);
7549 md_update_sb(mddev, 0);
7551 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
7552 !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
7553 /* resync/recovery still happening */
7554 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7557 if (mddev->sync_thread) {
7558 reap_sync_thread(mddev);
7561 /* Set RUNNING before clearing NEEDED to avoid
7562 * any transients in the value of "sync_action".
7564 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
7565 /* Clear some bits that don't mean anything, but
7568 clear_bit(MD_RECOVERY_INTR, &mddev->recovery);
7569 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
7571 if (!test_and_clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
7572 test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
7574 /* no recovery is running.
7575 * remove any failed drives, then
7576 * add spares if possible.
7577 * Spare are also removed and re-added, to allow
7578 * the personality to fail the re-add.
7581 if (mddev->reshape_position != MaxSector) {
7582 if (mddev->pers->check_reshape == NULL ||
7583 mddev->pers->check_reshape(mddev) != 0)
7584 /* Cannot proceed */
7586 set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
7587 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
7588 } else if ((spares = remove_and_add_spares(mddev))) {
7589 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
7590 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
7591 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
7592 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
7593 } else if (mddev->recovery_cp < MaxSector) {
7594 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
7595 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
7596 } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
7597 /* nothing to be done ... */
7600 if (mddev->pers->sync_request) {
7601 if (spares && mddev->bitmap && ! mddev->bitmap->file) {
7602 /* We are adding a device or devices to an array
7603 * which has the bitmap stored on all devices.
7604 * So make sure all bitmap pages get written
7606 bitmap_write_all(mddev->bitmap);
7608 mddev->sync_thread = md_register_thread(md_do_sync,
7611 if (!mddev->sync_thread) {
7612 printk(KERN_ERR "%s: could not start resync"
7615 /* leave the spares where they are, it shouldn't hurt */
7616 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
7617 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
7618 clear_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
7619 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
7620 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
7622 md_wakeup_thread(mddev->sync_thread);
7623 sysfs_notify_dirent_safe(mddev->sysfs_action);
7624 md_new_event(mddev);
7627 if (!mddev->sync_thread) {
7628 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
7629 if (test_and_clear_bit(MD_RECOVERY_RECOVER,
7631 if (mddev->sysfs_action)
7632 sysfs_notify_dirent_safe(mddev->sysfs_action);
7634 mddev_unlock(mddev);
7638 void md_wait_for_blocked_rdev(struct md_rdev *rdev, struct mddev *mddev)
7640 sysfs_notify_dirent_safe(rdev->sysfs_state);
7641 wait_event_timeout(rdev->blocked_wait,
7642 !test_bit(Blocked, &rdev->flags) &&
7643 !test_bit(BlockedBadBlocks, &rdev->flags),
7644 msecs_to_jiffies(5000));
7645 rdev_dec_pending(rdev, mddev);
7647 EXPORT_SYMBOL(md_wait_for_blocked_rdev);
7650 /* Bad block management.
7651 * We can record which blocks on each device are 'bad' and so just
7652 * fail those blocks, or that stripe, rather than the whole device.
7653 * Entries in the bad-block table are 64bits wide. This comprises:
7654 * Length of bad-range, in sectors: 0-511 for lengths 1-512
7655 * Start of bad-range, sector offset, 54 bits (allows 8 exbibytes)
7656 * A 'shift' can be set so that larger blocks are tracked and
7657 * consequently larger devices can be covered.
7658 * 'Acknowledged' flag - 1 bit. - the most significant bit.
7660 * Locking of the bad-block table uses a seqlock so md_is_badblock
7661 * might need to retry if it is very unlucky.
7662 * We will sometimes want to check for bad blocks in a bi_end_io function,
7663 * so we use the write_seqlock_irq variant.
7665 * When looking for a bad block we specify a range and want to
7666 * know if any block in the range is bad. So we binary-search
7667 * to the last range that starts at-or-before the given endpoint,
7668 * (or "before the sector after the target range")
7669 * then see if it ends after the given start.
7671 * 0 if there are no known bad blocks in the range
7672 * 1 if there are known bad block which are all acknowledged
7673 * -1 if there are bad blocks which have not yet been acknowledged in metadata.
7674 * plus the start/length of the first bad section we overlap.
7676 int md_is_badblock(struct badblocks *bb, sector_t s, int sectors,
7677 sector_t *first_bad, int *bad_sectors)
7683 sector_t target = s + sectors;
7686 if (bb->shift > 0) {
7687 /* round the start down, and the end up */
7689 target += (1<<bb->shift) - 1;
7690 target >>= bb->shift;
7691 sectors = target - s;
7693 /* 'target' is now the first block after the bad range */
7696 seq = read_seqbegin(&bb->lock);
7700 /* Binary search between lo and hi for 'target'
7701 * i.e. for the last range that starts before 'target'
7703 /* INVARIANT: ranges before 'lo' and at-or-after 'hi'
7704 * are known not to be the last range before target.
7705 * VARIANT: hi-lo is the number of possible
7706 * ranges, and decreases until it reaches 1
7708 while (hi - lo > 1) {
7709 int mid = (lo + hi) / 2;
7710 sector_t a = BB_OFFSET(p[mid]);
7712 /* This could still be the one, earlier ranges
7716 /* This and later ranges are definitely out. */
7719 /* 'lo' might be the last that started before target, but 'hi' isn't */
7721 /* need to check all range that end after 's' to see if
7722 * any are unacknowledged.
7725 BB_OFFSET(p[lo]) + BB_LEN(p[lo]) > s) {
7726 if (BB_OFFSET(p[lo]) < target) {
7727 /* starts before the end, and finishes after
7728 * the start, so they must overlap
7730 if (rv != -1 && BB_ACK(p[lo]))
7734 *first_bad = BB_OFFSET(p[lo]);
7735 *bad_sectors = BB_LEN(p[lo]);
7741 if (read_seqretry(&bb->lock, seq))
7746 EXPORT_SYMBOL_GPL(md_is_badblock);
7749 * Add a range of bad blocks to the table.
7750 * This might extend the table, or might contract it
7751 * if two adjacent ranges can be merged.
7752 * We binary-search to find the 'insertion' point, then
7753 * decide how best to handle it.
7755 static int md_set_badblocks(struct badblocks *bb, sector_t s, int sectors,
7763 /* badblocks are disabled */
7767 /* round the start down, and the end up */
7768 sector_t next = s + sectors;
7770 next += (1<<bb->shift) - 1;
7775 write_seqlock_irq(&bb->lock);
7780 /* Find the last range that starts at-or-before 's' */
7781 while (hi - lo > 1) {
7782 int mid = (lo + hi) / 2;
7783 sector_t a = BB_OFFSET(p[mid]);
7789 if (hi > lo && BB_OFFSET(p[lo]) > s)
7793 /* we found a range that might merge with the start
7796 sector_t a = BB_OFFSET(p[lo]);
7797 sector_t e = a + BB_LEN(p[lo]);
7798 int ack = BB_ACK(p[lo]);
7800 /* Yes, we can merge with a previous range */
7801 if (s == a && s + sectors >= e)
7802 /* new range covers old */
7805 ack = ack && acknowledged;
7807 if (e < s + sectors)
7809 if (e - a <= BB_MAX_LEN) {
7810 p[lo] = BB_MAKE(a, e-a, ack);
7813 /* does not all fit in one range,
7814 * make p[lo] maximal
7816 if (BB_LEN(p[lo]) != BB_MAX_LEN)
7817 p[lo] = BB_MAKE(a, BB_MAX_LEN, ack);
7823 if (sectors && hi < bb->count) {
7824 /* 'hi' points to the first range that starts after 's'.
7825 * Maybe we can merge with the start of that range */
7826 sector_t a = BB_OFFSET(p[hi]);
7827 sector_t e = a + BB_LEN(p[hi]);
7828 int ack = BB_ACK(p[hi]);
7829 if (a <= s + sectors) {
7830 /* merging is possible */
7831 if (e <= s + sectors) {
7836 ack = ack && acknowledged;
7839 if (e - a <= BB_MAX_LEN) {
7840 p[hi] = BB_MAKE(a, e-a, ack);
7843 p[hi] = BB_MAKE(a, BB_MAX_LEN, ack);
7851 if (sectors == 0 && hi < bb->count) {
7852 /* we might be able to combine lo and hi */
7853 /* Note: 's' is at the end of 'lo' */
7854 sector_t a = BB_OFFSET(p[hi]);
7855 int lolen = BB_LEN(p[lo]);
7856 int hilen = BB_LEN(p[hi]);
7857 int newlen = lolen + hilen - (s - a);
7858 if (s >= a && newlen < BB_MAX_LEN) {
7859 /* yes, we can combine them */
7860 int ack = BB_ACK(p[lo]) && BB_ACK(p[hi]);
7861 p[lo] = BB_MAKE(BB_OFFSET(p[lo]), newlen, ack);
7862 memmove(p + hi, p + hi + 1,
7863 (bb->count - hi - 1) * 8);
7868 /* didn't merge (it all).
7869 * Need to add a range just before 'hi' */
7870 if (bb->count >= MD_MAX_BADBLOCKS) {
7871 /* No room for more */
7875 int this_sectors = sectors;
7876 memmove(p + hi + 1, p + hi,
7877 (bb->count - hi) * 8);
7880 if (this_sectors > BB_MAX_LEN)
7881 this_sectors = BB_MAX_LEN;
7882 p[hi] = BB_MAKE(s, this_sectors, acknowledged);
7883 sectors -= this_sectors;
7890 bb->unacked_exist = 1;
7891 write_sequnlock_irq(&bb->lock);
7896 int rdev_set_badblocks(struct md_rdev *rdev, sector_t s, int sectors,
7899 int rv = md_set_badblocks(&rdev->badblocks,
7900 s + rdev->data_offset, sectors, acknowledged);
7902 /* Make sure they get written out promptly */
7903 sysfs_notify_dirent_safe(rdev->sysfs_state);
7904 set_bit(MD_CHANGE_CLEAN, &rdev->mddev->flags);
7905 md_wakeup_thread(rdev->mddev->thread);
7909 EXPORT_SYMBOL_GPL(rdev_set_badblocks);
7912 * Remove a range of bad blocks from the table.
7913 * This may involve extending the table if we spilt a region,
7914 * but it must not fail. So if the table becomes full, we just
7915 * drop the remove request.
7917 static int md_clear_badblocks(struct badblocks *bb, sector_t s, int sectors)
7921 sector_t target = s + sectors;
7924 if (bb->shift > 0) {
7925 /* When clearing we round the start up and the end down.
7926 * This should not matter as the shift should align with
7927 * the block size and no rounding should ever be needed.
7928 * However it is better the think a block is bad when it
7929 * isn't than to think a block is not bad when it is.
7931 s += (1<<bb->shift) - 1;
7933 target >>= bb->shift;
7934 sectors = target - s;
7937 write_seqlock_irq(&bb->lock);
7942 /* Find the last range that starts before 'target' */
7943 while (hi - lo > 1) {
7944 int mid = (lo + hi) / 2;
7945 sector_t a = BB_OFFSET(p[mid]);
7952 /* p[lo] is the last range that could overlap the
7953 * current range. Earlier ranges could also overlap,
7954 * but only this one can overlap the end of the range.
7956 if (BB_OFFSET(p[lo]) + BB_LEN(p[lo]) > target) {
7957 /* Partial overlap, leave the tail of this range */
7958 int ack = BB_ACK(p[lo]);
7959 sector_t a = BB_OFFSET(p[lo]);
7960 sector_t end = a + BB_LEN(p[lo]);
7963 /* we need to split this range */
7964 if (bb->count >= MD_MAX_BADBLOCKS) {
7968 memmove(p+lo+1, p+lo, (bb->count - lo) * 8);
7970 p[lo] = BB_MAKE(a, s-a, ack);
7973 p[lo] = BB_MAKE(target, end - target, ack);
7974 /* there is no longer an overlap */
7979 BB_OFFSET(p[lo]) + BB_LEN(p[lo]) > s) {
7980 /* This range does overlap */
7981 if (BB_OFFSET(p[lo]) < s) {
7982 /* Keep the early parts of this range. */
7983 int ack = BB_ACK(p[lo]);
7984 sector_t start = BB_OFFSET(p[lo]);
7985 p[lo] = BB_MAKE(start, s - start, ack);
7986 /* now low doesn't overlap, so.. */
7991 /* 'lo' is strictly before, 'hi' is strictly after,
7992 * anything between needs to be discarded
7995 memmove(p+lo+1, p+hi, (bb->count - hi) * 8);
7996 bb->count -= (hi - lo - 1);
8002 write_sequnlock_irq(&bb->lock);
8006 int rdev_clear_badblocks(struct md_rdev *rdev, sector_t s, int sectors)
8008 return md_clear_badblocks(&rdev->badblocks,
8009 s + rdev->data_offset,
8012 EXPORT_SYMBOL_GPL(rdev_clear_badblocks);
8015 * Acknowledge all bad blocks in a list.
8016 * This only succeeds if ->changed is clear. It is used by
8017 * in-kernel metadata updates
8019 void md_ack_all_badblocks(struct badblocks *bb)
8021 if (bb->page == NULL || bb->changed)
8022 /* no point even trying */
8024 write_seqlock_irq(&bb->lock);
8026 if (bb->changed == 0 && bb->unacked_exist) {
8029 for (i = 0; i < bb->count ; i++) {
8030 if (!BB_ACK(p[i])) {
8031 sector_t start = BB_OFFSET(p[i]);
8032 int len = BB_LEN(p[i]);
8033 p[i] = BB_MAKE(start, len, 1);
8036 bb->unacked_exist = 0;
8038 write_sequnlock_irq(&bb->lock);
8040 EXPORT_SYMBOL_GPL(md_ack_all_badblocks);
8042 /* sysfs access to bad-blocks list.
8043 * We present two files.
8044 * 'bad-blocks' lists sector numbers and lengths of ranges that
8045 * are recorded as bad. The list is truncated to fit within
8046 * the one-page limit of sysfs.
8047 * Writing "sector length" to this file adds an acknowledged
8049 * 'unacknowledged-bad-blocks' lists bad blocks that have not yet
8050 * been acknowledged. Writing to this file adds bad blocks
8051 * without acknowledging them. This is largely for testing.
8055 badblocks_show(struct badblocks *bb, char *page, int unack)
8066 seq = read_seqbegin(&bb->lock);
8071 while (len < PAGE_SIZE && i < bb->count) {
8072 sector_t s = BB_OFFSET(p[i]);
8073 unsigned int length = BB_LEN(p[i]);
8074 int ack = BB_ACK(p[i]);
8080 len += snprintf(page+len, PAGE_SIZE-len, "%llu %u\n",
8081 (unsigned long long)s << bb->shift,
8082 length << bb->shift);
8084 if (unack && len == 0)
8085 bb->unacked_exist = 0;
8087 if (read_seqretry(&bb->lock, seq))
8096 badblocks_store(struct badblocks *bb, const char *page, size_t len, int unack)
8098 unsigned long long sector;
8102 /* Allow clearing via sysfs *only* for testing/debugging.
8103 * Normally only a successful write may clear a badblock
8106 if (page[0] == '-') {
8110 #endif /* DO_DEBUG */
8112 switch (sscanf(page, "%llu %d%c", §or, &length, &newline)) {
8114 if (newline != '\n')
8126 md_clear_badblocks(bb, sector, length);
8129 #endif /* DO_DEBUG */
8130 if (md_set_badblocks(bb, sector, length, !unack))
8136 static int md_notify_reboot(struct notifier_block *this,
8137 unsigned long code, void *x)
8139 struct list_head *tmp;
8140 struct mddev *mddev;
8143 for_each_mddev(mddev, tmp) {
8144 if (mddev_trylock(mddev)) {
8146 __md_stop_writes(mddev);
8147 mddev->safemode = 2;
8148 mddev_unlock(mddev);
8153 * certain more exotic SCSI devices are known to be
8154 * volatile wrt too early system reboots. While the
8155 * right place to handle this issue is the given
8156 * driver, we do want to have a safe RAID driver ...
8164 static struct notifier_block md_notifier = {
8165 .notifier_call = md_notify_reboot,
8167 .priority = INT_MAX, /* before any real devices */
8170 static void md_geninit(void)
8172 pr_debug("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
8174 proc_create("mdstat", S_IRUGO, NULL, &md_seq_fops);
8177 static int __init md_init(void)
8181 md_wq = alloc_workqueue("md", WQ_MEM_RECLAIM, 0);
8185 md_misc_wq = alloc_workqueue("md_misc", 0, 0);
8189 if ((ret = register_blkdev(MD_MAJOR, "md")) < 0)
8192 if ((ret = register_blkdev(0, "mdp")) < 0)
8196 blk_register_region(MKDEV(MD_MAJOR, 0), 1UL<<MINORBITS, THIS_MODULE,
8197 md_probe, NULL, NULL);
8198 blk_register_region(MKDEV(mdp_major, 0), 1UL<<MINORBITS, THIS_MODULE,
8199 md_probe, NULL, NULL);
8201 register_reboot_notifier(&md_notifier);
8202 raid_table_header = register_sysctl_table(raid_root_table);
8208 unregister_blkdev(MD_MAJOR, "md");
8210 destroy_workqueue(md_misc_wq);
8212 destroy_workqueue(md_wq);
8220 * Searches all registered partitions for autorun RAID arrays
8224 static LIST_HEAD(all_detected_devices);
8225 struct detected_devices_node {
8226 struct list_head list;
8230 void md_autodetect_dev(dev_t dev)
8232 struct detected_devices_node *node_detected_dev;
8234 node_detected_dev = kzalloc(sizeof(*node_detected_dev), GFP_KERNEL);
8235 if (node_detected_dev) {
8236 node_detected_dev->dev = dev;
8237 list_add_tail(&node_detected_dev->list, &all_detected_devices);
8239 printk(KERN_CRIT "md: md_autodetect_dev: kzalloc failed"
8240 ", skipping dev(%d,%d)\n", MAJOR(dev), MINOR(dev));
8245 static void autostart_arrays(int part)
8247 struct md_rdev *rdev;
8248 struct detected_devices_node *node_detected_dev;
8250 int i_scanned, i_passed;
8255 printk(KERN_INFO "md: Autodetecting RAID arrays.\n");
8257 while (!list_empty(&all_detected_devices) && i_scanned < INT_MAX) {
8259 node_detected_dev = list_entry(all_detected_devices.next,
8260 struct detected_devices_node, list);
8261 list_del(&node_detected_dev->list);
8262 dev = node_detected_dev->dev;
8263 kfree(node_detected_dev);
8264 rdev = md_import_device(dev,0, 90);
8268 if (test_bit(Faulty, &rdev->flags)) {
8272 set_bit(AutoDetected, &rdev->flags);
8273 list_add(&rdev->same_set, &pending_raid_disks);
8277 printk(KERN_INFO "md: Scanned %d and added %d devices.\n",
8278 i_scanned, i_passed);
8280 autorun_devices(part);
8283 #endif /* !MODULE */
8285 static __exit void md_exit(void)
8287 struct mddev *mddev;
8288 struct list_head *tmp;
8290 blk_unregister_region(MKDEV(MD_MAJOR,0), 1U << MINORBITS);
8291 blk_unregister_region(MKDEV(mdp_major,0), 1U << MINORBITS);
8293 unregister_blkdev(MD_MAJOR,"md");
8294 unregister_blkdev(mdp_major, "mdp");
8295 unregister_reboot_notifier(&md_notifier);
8296 unregister_sysctl_table(raid_table_header);
8297 remove_proc_entry("mdstat", NULL);
8298 for_each_mddev(mddev, tmp) {
8299 export_array(mddev);
8300 mddev->hold_active = 0;
8302 destroy_workqueue(md_misc_wq);
8303 destroy_workqueue(md_wq);
8306 subsys_initcall(md_init);
8307 module_exit(md_exit)
8309 static int get_ro(char *buffer, struct kernel_param *kp)
8311 return sprintf(buffer, "%d", start_readonly);
8313 static int set_ro(const char *val, struct kernel_param *kp)
8316 int num = simple_strtoul(val, &e, 10);
8317 if (*val && (*e == '\0' || *e == '\n')) {
8318 start_readonly = num;
8324 module_param_call(start_ro, set_ro, get_ro, NULL, S_IRUSR|S_IWUSR);
8325 module_param(start_dirty_degraded, int, S_IRUGO|S_IWUSR);
8327 module_param_call(new_array, add_named_array, NULL, NULL, S_IWUSR);
8329 EXPORT_SYMBOL(register_md_personality);
8330 EXPORT_SYMBOL(unregister_md_personality);
8331 EXPORT_SYMBOL(md_error);
8332 EXPORT_SYMBOL(md_done_sync);
8333 EXPORT_SYMBOL(md_write_start);
8334 EXPORT_SYMBOL(md_write_end);
8335 EXPORT_SYMBOL(md_register_thread);
8336 EXPORT_SYMBOL(md_unregister_thread);
8337 EXPORT_SYMBOL(md_wakeup_thread);
8338 EXPORT_SYMBOL(md_check_recovery);
8339 MODULE_LICENSE("GPL");
8340 MODULE_DESCRIPTION("MD RAID framework");
8342 MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR);