2 md.c : Multiple Devices driver for Linux
3 Copyright (C) 1998, 1999, 2000 Ingo Molnar
5 completely rewritten, based on the MD driver code from Marc Zyngier
9 - RAID-1/RAID-5 extensions by Miguel de Icaza, Gadi Oxman, Ingo Molnar
10 - RAID-6 extensions by H. Peter Anvin <hpa@zytor.com>
11 - boot support for linear and striped mode by Harald Hoyer <HarryH@Royal.Net>
12 - kerneld support by Boris Tobotras <boris@xtalk.msk.su>
13 - kmod support by: Cyrus Durgin
14 - RAID0 bugfixes: Mark Anthony Lisher <markal@iname.com>
15 - Devfs support by Richard Gooch <rgooch@atnf.csiro.au>
17 - lots of fixes and improvements to the RAID1/RAID5 and generic
18 RAID code (such as request based resynchronization):
20 Neil Brown <neilb@cse.unsw.edu.au>.
22 - persistent bitmap code
23 Copyright (C) 2003-2004, Paul Clements, SteelEye Technology, Inc.
25 This program is free software; you can redistribute it and/or modify
26 it under the terms of the GNU General Public License as published by
27 the Free Software Foundation; either version 2, or (at your option)
30 You should have received a copy of the GNU General Public License
31 (for example /usr/src/linux/COPYING); if not, write to the Free
32 Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
35 #include <linux/kthread.h>
36 #include <linux/blkdev.h>
37 #include <linux/sysctl.h>
38 #include <linux/seq_file.h>
39 #include <linux/mutex.h>
40 #include <linux/buffer_head.h> /* for invalidate_bdev */
41 #include <linux/poll.h>
42 #include <linux/ctype.h>
43 #include <linux/string.h>
44 #include <linux/hdreg.h>
45 #include <linux/proc_fs.h>
46 #include <linux/random.h>
47 #include <linux/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 #define dprintk(x...) ((void)(DEBUG && printk(x)))
61 static void autostart_arrays(int part);
64 /* pers_list is a list of registered personalities protected
66 * pers_lock does extra service to protect accesses to
67 * mddev->thread when the mutex cannot be held.
69 static LIST_HEAD(pers_list);
70 static DEFINE_SPINLOCK(pers_lock);
72 static void md_print_devices(void);
74 static DECLARE_WAIT_QUEUE_HEAD(resync_wait);
75 static struct workqueue_struct *md_wq;
76 static struct workqueue_struct *md_misc_wq;
78 #define MD_BUG(x...) { printk("md: bug in file %s, line %d\n", __FILE__, __LINE__); md_print_devices(); }
81 * Default number of read corrections we'll attempt on an rdev
82 * before ejecting it from the array. We divide the read error
83 * count by 2 for every hour elapsed between read errors.
85 #define MD_DEFAULT_MAX_CORRECTED_READ_ERRORS 20
87 * Current RAID-1,4,5 parallel reconstruction 'guaranteed speed limit'
88 * is 1000 KB/sec, so the extra system load does not show up that much.
89 * Increase it if you want to have more _guaranteed_ speed. Note that
90 * the RAID driver will use the maximum available bandwidth if the IO
91 * subsystem is idle. There is also an 'absolute maximum' reconstruction
92 * speed limit - in case reconstruction slows down your system despite
95 * you can change it via /proc/sys/dev/raid/speed_limit_min and _max.
96 * or /sys/block/mdX/md/sync_speed_{min,max}
99 static int sysctl_speed_limit_min = 1000;
100 static int sysctl_speed_limit_max = 200000;
101 static inline int speed_min(mddev_t *mddev)
103 return mddev->sync_speed_min ?
104 mddev->sync_speed_min : sysctl_speed_limit_min;
107 static inline int speed_max(mddev_t *mddev)
109 return mddev->sync_speed_max ?
110 mddev->sync_speed_max : sysctl_speed_limit_max;
113 static struct ctl_table_header *raid_table_header;
115 static ctl_table raid_table[] = {
117 .procname = "speed_limit_min",
118 .data = &sysctl_speed_limit_min,
119 .maxlen = sizeof(int),
120 .mode = S_IRUGO|S_IWUSR,
121 .proc_handler = proc_dointvec,
124 .procname = "speed_limit_max",
125 .data = &sysctl_speed_limit_max,
126 .maxlen = sizeof(int),
127 .mode = S_IRUGO|S_IWUSR,
128 .proc_handler = proc_dointvec,
133 static ctl_table raid_dir_table[] = {
137 .mode = S_IRUGO|S_IXUGO,
143 static ctl_table raid_root_table[] = {
148 .child = raid_dir_table,
153 static const struct block_device_operations md_fops;
155 static int start_readonly;
158 * like bio_clone, but with a local bio set
161 static void mddev_bio_destructor(struct bio *bio)
163 mddev_t *mddev, **mddevp;
168 bio_free(bio, mddev->bio_set);
171 struct bio *bio_alloc_mddev(gfp_t gfp_mask, int nr_iovecs,
177 if (!mddev || !mddev->bio_set)
178 return bio_alloc(gfp_mask, nr_iovecs);
180 b = bio_alloc_bioset(gfp_mask, nr_iovecs,
186 b->bi_destructor = mddev_bio_destructor;
189 EXPORT_SYMBOL_GPL(bio_alloc_mddev);
191 struct bio *bio_clone_mddev(struct bio *bio, gfp_t gfp_mask,
197 if (!mddev || !mddev->bio_set)
198 return bio_clone(bio, gfp_mask);
200 b = bio_alloc_bioset(gfp_mask, bio->bi_max_vecs,
206 b->bi_destructor = mddev_bio_destructor;
208 if (bio_integrity(bio)) {
211 ret = bio_integrity_clone(b, bio, gfp_mask, mddev->bio_set);
221 EXPORT_SYMBOL_GPL(bio_clone_mddev);
223 void md_trim_bio(struct bio *bio, int offset, int size)
225 /* 'bio' is a cloned bio which we need to trim to match
226 * the given offset and size.
227 * This requires adjusting bi_sector, bi_size, and bi_io_vec
230 struct bio_vec *bvec;
234 if (offset == 0 && size == bio->bi_size)
237 bio->bi_sector += offset;
240 clear_bit(BIO_SEG_VALID, &bio->bi_flags);
242 while (bio->bi_idx < bio->bi_vcnt &&
243 bio->bi_io_vec[bio->bi_idx].bv_len <= offset) {
244 /* remove this whole bio_vec */
245 offset -= bio->bi_io_vec[bio->bi_idx].bv_len;
248 if (bio->bi_idx < bio->bi_vcnt) {
249 bio->bi_io_vec[bio->bi_idx].bv_offset += offset;
250 bio->bi_io_vec[bio->bi_idx].bv_len -= offset;
252 /* avoid any complications with bi_idx being non-zero*/
254 memmove(bio->bi_io_vec, bio->bi_io_vec+bio->bi_idx,
255 (bio->bi_vcnt - bio->bi_idx) * sizeof(struct bio_vec));
256 bio->bi_vcnt -= bio->bi_idx;
259 /* Make sure vcnt and last bv are not too big */
260 bio_for_each_segment(bvec, bio, i) {
261 if (sofar + bvec->bv_len > size)
262 bvec->bv_len = size - sofar;
263 if (bvec->bv_len == 0) {
267 sofar += bvec->bv_len;
270 EXPORT_SYMBOL_GPL(md_trim_bio);
273 * We have a system wide 'event count' that is incremented
274 * on any 'interesting' event, and readers of /proc/mdstat
275 * can use 'poll' or 'select' to find out when the event
279 * start array, stop array, error, add device, remove device,
280 * start build, activate spare
282 static DECLARE_WAIT_QUEUE_HEAD(md_event_waiters);
283 static atomic_t md_event_count;
284 void md_new_event(mddev_t *mddev)
286 atomic_inc(&md_event_count);
287 wake_up(&md_event_waiters);
289 EXPORT_SYMBOL_GPL(md_new_event);
291 /* Alternate version that can be called from interrupts
292 * when calling sysfs_notify isn't needed.
294 static void md_new_event_inintr(mddev_t *mddev)
296 atomic_inc(&md_event_count);
297 wake_up(&md_event_waiters);
301 * Enables to iterate over all existing md arrays
302 * all_mddevs_lock protects this list.
304 static LIST_HEAD(all_mddevs);
305 static DEFINE_SPINLOCK(all_mddevs_lock);
309 * iterates through all used mddevs in the system.
310 * We take care to grab the all_mddevs_lock whenever navigating
311 * the list, and to always hold a refcount when unlocked.
312 * Any code which breaks out of this loop while own
313 * a reference to the current mddev and must mddev_put it.
315 #define for_each_mddev(mddev,tmp) \
317 for (({ spin_lock(&all_mddevs_lock); \
318 tmp = all_mddevs.next; \
320 ({ if (tmp != &all_mddevs) \
321 mddev_get(list_entry(tmp, mddev_t, all_mddevs));\
322 spin_unlock(&all_mddevs_lock); \
323 if (mddev) mddev_put(mddev); \
324 mddev = list_entry(tmp, mddev_t, all_mddevs); \
325 tmp != &all_mddevs;}); \
326 ({ spin_lock(&all_mddevs_lock); \
331 /* Rather than calling directly into the personality make_request function,
332 * IO requests come here first so that we can check if the device is
333 * being suspended pending a reconfiguration.
334 * We hold a refcount over the call to ->make_request. By the time that
335 * call has finished, the bio has been linked into some internal structure
336 * and so is visible to ->quiesce(), so we don't need the refcount any more.
338 static void md_make_request(struct request_queue *q, struct bio *bio)
340 const int rw = bio_data_dir(bio);
341 mddev_t *mddev = q->queuedata;
343 unsigned int sectors;
345 if (mddev == NULL || mddev->pers == NULL
350 smp_rmb(); /* Ensure implications of 'active' are visible */
352 if (mddev->suspended) {
355 prepare_to_wait(&mddev->sb_wait, &__wait,
356 TASK_UNINTERRUPTIBLE);
357 if (!mddev->suspended)
363 finish_wait(&mddev->sb_wait, &__wait);
365 atomic_inc(&mddev->active_io);
369 * save the sectors now since our bio can
370 * go away inside make_request
372 sectors = bio_sectors(bio);
373 mddev->pers->make_request(mddev, bio);
375 cpu = part_stat_lock();
376 part_stat_inc(cpu, &mddev->gendisk->part0, ios[rw]);
377 part_stat_add(cpu, &mddev->gendisk->part0, sectors[rw], sectors);
380 if (atomic_dec_and_test(&mddev->active_io) && mddev->suspended)
381 wake_up(&mddev->sb_wait);
384 /* mddev_suspend makes sure no new requests are submitted
385 * to the device, and that any requests that have been submitted
386 * are completely handled.
387 * Once ->stop is called and completes, the module will be completely
390 void mddev_suspend(mddev_t *mddev)
392 BUG_ON(mddev->suspended);
393 mddev->suspended = 1;
395 wait_event(mddev->sb_wait, atomic_read(&mddev->active_io) == 0);
396 mddev->pers->quiesce(mddev, 1);
398 EXPORT_SYMBOL_GPL(mddev_suspend);
400 void mddev_resume(mddev_t *mddev)
402 mddev->suspended = 0;
403 wake_up(&mddev->sb_wait);
404 mddev->pers->quiesce(mddev, 0);
406 md_wakeup_thread(mddev->thread);
407 md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
409 EXPORT_SYMBOL_GPL(mddev_resume);
411 int mddev_congested(mddev_t *mddev, int bits)
413 return mddev->suspended;
415 EXPORT_SYMBOL(mddev_congested);
418 * Generic flush handling for md
421 static void md_end_flush(struct bio *bio, int err)
423 mdk_rdev_t *rdev = bio->bi_private;
424 mddev_t *mddev = rdev->mddev;
426 rdev_dec_pending(rdev, mddev);
428 if (atomic_dec_and_test(&mddev->flush_pending)) {
429 /* The pre-request flush has finished */
430 queue_work(md_wq, &mddev->flush_work);
435 static void md_submit_flush_data(struct work_struct *ws);
437 static void submit_flushes(struct work_struct *ws)
439 mddev_t *mddev = container_of(ws, mddev_t, flush_work);
442 INIT_WORK(&mddev->flush_work, md_submit_flush_data);
443 atomic_set(&mddev->flush_pending, 1);
445 list_for_each_entry_rcu(rdev, &mddev->disks, same_set)
446 if (rdev->raid_disk >= 0 &&
447 !test_bit(Faulty, &rdev->flags)) {
448 /* Take two references, one is dropped
449 * when request finishes, one after
450 * we reclaim rcu_read_lock
453 atomic_inc(&rdev->nr_pending);
454 atomic_inc(&rdev->nr_pending);
456 bi = bio_alloc_mddev(GFP_KERNEL, 0, mddev);
457 bi->bi_end_io = md_end_flush;
458 bi->bi_private = rdev;
459 bi->bi_bdev = rdev->bdev;
460 atomic_inc(&mddev->flush_pending);
461 submit_bio(WRITE_FLUSH, bi);
463 rdev_dec_pending(rdev, mddev);
466 if (atomic_dec_and_test(&mddev->flush_pending))
467 queue_work(md_wq, &mddev->flush_work);
470 static void md_submit_flush_data(struct work_struct *ws)
472 mddev_t *mddev = container_of(ws, mddev_t, flush_work);
473 struct bio *bio = mddev->flush_bio;
475 if (bio->bi_size == 0)
476 /* an empty barrier - all done */
479 bio->bi_rw &= ~REQ_FLUSH;
480 mddev->pers->make_request(mddev, bio);
483 mddev->flush_bio = NULL;
484 wake_up(&mddev->sb_wait);
487 void md_flush_request(mddev_t *mddev, struct bio *bio)
489 spin_lock_irq(&mddev->write_lock);
490 wait_event_lock_irq(mddev->sb_wait,
492 mddev->write_lock, /*nothing*/);
493 mddev->flush_bio = bio;
494 spin_unlock_irq(&mddev->write_lock);
496 INIT_WORK(&mddev->flush_work, submit_flushes);
497 queue_work(md_wq, &mddev->flush_work);
499 EXPORT_SYMBOL(md_flush_request);
501 /* Support for plugging.
502 * This mirrors the plugging support in request_queue, but does not
503 * require having a whole queue or request structures.
504 * We allocate an md_plug_cb for each md device and each thread it gets
505 * plugged on. This links tot the private plug_handle structure in the
506 * personality data where we keep a count of the number of outstanding
507 * plugs so other code can see if a plug is active.
510 struct blk_plug_cb cb;
514 static void plugger_unplug(struct blk_plug_cb *cb)
516 struct md_plug_cb *mdcb = container_of(cb, struct md_plug_cb, cb);
517 if (atomic_dec_and_test(&mdcb->mddev->plug_cnt))
518 md_wakeup_thread(mdcb->mddev->thread);
522 /* Check that an unplug wakeup will come shortly.
523 * If not, wakeup the md thread immediately
525 int mddev_check_plugged(mddev_t *mddev)
527 struct blk_plug *plug = current->plug;
528 struct md_plug_cb *mdcb;
533 list_for_each_entry(mdcb, &plug->cb_list, cb.list) {
534 if (mdcb->cb.callback == plugger_unplug &&
535 mdcb->mddev == mddev) {
536 /* Already on the list, move to top */
537 if (mdcb != list_first_entry(&plug->cb_list,
540 list_move(&mdcb->cb.list, &plug->cb_list);
544 /* Not currently on the callback list */
545 mdcb = kmalloc(sizeof(*mdcb), GFP_ATOMIC);
550 mdcb->cb.callback = plugger_unplug;
551 atomic_inc(&mddev->plug_cnt);
552 list_add(&mdcb->cb.list, &plug->cb_list);
555 EXPORT_SYMBOL_GPL(mddev_check_plugged);
557 static inline mddev_t *mddev_get(mddev_t *mddev)
559 atomic_inc(&mddev->active);
563 static void mddev_delayed_delete(struct work_struct *ws);
565 static void mddev_put(mddev_t *mddev)
567 struct bio_set *bs = NULL;
569 if (!atomic_dec_and_lock(&mddev->active, &all_mddevs_lock))
571 if (!mddev->raid_disks && list_empty(&mddev->disks) &&
572 mddev->ctime == 0 && !mddev->hold_active) {
573 /* Array is not configured at all, and not held active,
575 list_del(&mddev->all_mddevs);
577 mddev->bio_set = NULL;
578 if (mddev->gendisk) {
579 /* We did a probe so need to clean up. Call
580 * queue_work inside the spinlock so that
581 * flush_workqueue() after mddev_find will
582 * succeed in waiting for the work to be done.
584 INIT_WORK(&mddev->del_work, mddev_delayed_delete);
585 queue_work(md_misc_wq, &mddev->del_work);
589 spin_unlock(&all_mddevs_lock);
594 void mddev_init(mddev_t *mddev)
596 mutex_init(&mddev->open_mutex);
597 mutex_init(&mddev->reconfig_mutex);
598 mutex_init(&mddev->bitmap_info.mutex);
599 INIT_LIST_HEAD(&mddev->disks);
600 INIT_LIST_HEAD(&mddev->all_mddevs);
601 init_timer(&mddev->safemode_timer);
602 atomic_set(&mddev->active, 1);
603 atomic_set(&mddev->openers, 0);
604 atomic_set(&mddev->active_io, 0);
605 atomic_set(&mddev->plug_cnt, 0);
606 spin_lock_init(&mddev->write_lock);
607 atomic_set(&mddev->flush_pending, 0);
608 init_waitqueue_head(&mddev->sb_wait);
609 init_waitqueue_head(&mddev->recovery_wait);
610 mddev->reshape_position = MaxSector;
611 mddev->resync_min = 0;
612 mddev->resync_max = MaxSector;
613 mddev->level = LEVEL_NONE;
615 EXPORT_SYMBOL_GPL(mddev_init);
617 static mddev_t * mddev_find(dev_t unit)
619 mddev_t *mddev, *new = NULL;
621 if (unit && MAJOR(unit) != MD_MAJOR)
622 unit &= ~((1<<MdpMinorShift)-1);
625 spin_lock(&all_mddevs_lock);
628 list_for_each_entry(mddev, &all_mddevs, all_mddevs)
629 if (mddev->unit == unit) {
631 spin_unlock(&all_mddevs_lock);
637 list_add(&new->all_mddevs, &all_mddevs);
638 spin_unlock(&all_mddevs_lock);
639 new->hold_active = UNTIL_IOCTL;
643 /* find an unused unit number */
644 static int next_minor = 512;
645 int start = next_minor;
649 dev = MKDEV(MD_MAJOR, next_minor);
651 if (next_minor > MINORMASK)
653 if (next_minor == start) {
654 /* Oh dear, all in use. */
655 spin_unlock(&all_mddevs_lock);
661 list_for_each_entry(mddev, &all_mddevs, all_mddevs)
662 if (mddev->unit == dev) {
668 new->md_minor = MINOR(dev);
669 new->hold_active = UNTIL_STOP;
670 list_add(&new->all_mddevs, &all_mddevs);
671 spin_unlock(&all_mddevs_lock);
674 spin_unlock(&all_mddevs_lock);
676 new = kzalloc(sizeof(*new), GFP_KERNEL);
681 if (MAJOR(unit) == MD_MAJOR)
682 new->md_minor = MINOR(unit);
684 new->md_minor = MINOR(unit) >> MdpMinorShift;
691 static inline int mddev_lock(mddev_t * mddev)
693 return mutex_lock_interruptible(&mddev->reconfig_mutex);
696 static inline int mddev_is_locked(mddev_t *mddev)
698 return mutex_is_locked(&mddev->reconfig_mutex);
701 static inline int mddev_trylock(mddev_t * mddev)
703 return mutex_trylock(&mddev->reconfig_mutex);
706 static struct attribute_group md_redundancy_group;
708 static void mddev_unlock(mddev_t * mddev)
710 if (mddev->to_remove) {
711 /* These cannot be removed under reconfig_mutex as
712 * an access to the files will try to take reconfig_mutex
713 * while holding the file unremovable, which leads to
715 * So hold set sysfs_active while the remove in happeing,
716 * and anything else which might set ->to_remove or my
717 * otherwise change the sysfs namespace will fail with
718 * -EBUSY if sysfs_active is still set.
719 * We set sysfs_active under reconfig_mutex and elsewhere
720 * test it under the same mutex to ensure its correct value
723 struct attribute_group *to_remove = mddev->to_remove;
724 mddev->to_remove = NULL;
725 mddev->sysfs_active = 1;
726 mutex_unlock(&mddev->reconfig_mutex);
728 if (mddev->kobj.sd) {
729 if (to_remove != &md_redundancy_group)
730 sysfs_remove_group(&mddev->kobj, to_remove);
731 if (mddev->pers == NULL ||
732 mddev->pers->sync_request == NULL) {
733 sysfs_remove_group(&mddev->kobj, &md_redundancy_group);
734 if (mddev->sysfs_action)
735 sysfs_put(mddev->sysfs_action);
736 mddev->sysfs_action = NULL;
739 mddev->sysfs_active = 0;
741 mutex_unlock(&mddev->reconfig_mutex);
743 /* was we've dropped the mutex we need a spinlock to
744 * make sur the thread doesn't disappear
746 spin_lock(&pers_lock);
747 md_wakeup_thread(mddev->thread);
748 spin_unlock(&pers_lock);
751 static mdk_rdev_t * find_rdev_nr(mddev_t *mddev, int nr)
755 list_for_each_entry(rdev, &mddev->disks, same_set)
756 if (rdev->desc_nr == nr)
762 static mdk_rdev_t * find_rdev(mddev_t * mddev, dev_t dev)
766 list_for_each_entry(rdev, &mddev->disks, same_set)
767 if (rdev->bdev->bd_dev == dev)
773 static struct mdk_personality *find_pers(int level, char *clevel)
775 struct mdk_personality *pers;
776 list_for_each_entry(pers, &pers_list, list) {
777 if (level != LEVEL_NONE && pers->level == level)
779 if (strcmp(pers->name, clevel)==0)
785 /* return the offset of the super block in 512byte sectors */
786 static inline sector_t calc_dev_sboffset(mdk_rdev_t *rdev)
788 sector_t num_sectors = i_size_read(rdev->bdev->bd_inode) / 512;
789 return MD_NEW_SIZE_SECTORS(num_sectors);
792 static int alloc_disk_sb(mdk_rdev_t * rdev)
797 rdev->sb_page = alloc_page(GFP_KERNEL);
798 if (!rdev->sb_page) {
799 printk(KERN_ALERT "md: out of memory.\n");
806 static void free_disk_sb(mdk_rdev_t * rdev)
809 put_page(rdev->sb_page);
811 rdev->sb_page = NULL;
816 put_page(rdev->bb_page);
817 rdev->bb_page = NULL;
822 static void super_written(struct bio *bio, int error)
824 mdk_rdev_t *rdev = bio->bi_private;
825 mddev_t *mddev = rdev->mddev;
827 if (error || !test_bit(BIO_UPTODATE, &bio->bi_flags)) {
828 printk("md: super_written gets error=%d, uptodate=%d\n",
829 error, test_bit(BIO_UPTODATE, &bio->bi_flags));
830 WARN_ON(test_bit(BIO_UPTODATE, &bio->bi_flags));
831 md_error(mddev, rdev);
834 if (atomic_dec_and_test(&mddev->pending_writes))
835 wake_up(&mddev->sb_wait);
839 void md_super_write(mddev_t *mddev, mdk_rdev_t *rdev,
840 sector_t sector, int size, struct page *page)
842 /* write first size bytes of page to sector of rdev
843 * Increment mddev->pending_writes before returning
844 * and decrement it on completion, waking up sb_wait
845 * if zero is reached.
846 * If an error occurred, call md_error
848 struct bio *bio = bio_alloc_mddev(GFP_NOIO, 1, mddev);
850 bio->bi_bdev = rdev->meta_bdev ? rdev->meta_bdev : rdev->bdev;
851 bio->bi_sector = sector;
852 bio_add_page(bio, page, size, 0);
853 bio->bi_private = rdev;
854 bio->bi_end_io = super_written;
856 atomic_inc(&mddev->pending_writes);
857 submit_bio(WRITE_FLUSH_FUA, bio);
860 void md_super_wait(mddev_t *mddev)
862 /* wait for all superblock writes that were scheduled to complete */
865 prepare_to_wait(&mddev->sb_wait, &wq, TASK_UNINTERRUPTIBLE);
866 if (atomic_read(&mddev->pending_writes)==0)
870 finish_wait(&mddev->sb_wait, &wq);
873 static void bi_complete(struct bio *bio, int error)
875 complete((struct completion*)bio->bi_private);
878 int sync_page_io(mdk_rdev_t *rdev, sector_t sector, int size,
879 struct page *page, int rw, bool metadata_op)
881 struct bio *bio = bio_alloc_mddev(GFP_NOIO, 1, rdev->mddev);
882 struct completion event;
887 bio->bi_bdev = (metadata_op && rdev->meta_bdev) ?
888 rdev->meta_bdev : rdev->bdev;
890 bio->bi_sector = sector + rdev->sb_start;
892 bio->bi_sector = sector + rdev->data_offset;
893 bio_add_page(bio, page, size, 0);
894 init_completion(&event);
895 bio->bi_private = &event;
896 bio->bi_end_io = bi_complete;
898 wait_for_completion(&event);
900 ret = test_bit(BIO_UPTODATE, &bio->bi_flags);
904 EXPORT_SYMBOL_GPL(sync_page_io);
906 static int read_disk_sb(mdk_rdev_t * rdev, int size)
908 char b[BDEVNAME_SIZE];
909 if (!rdev->sb_page) {
917 if (!sync_page_io(rdev, 0, size, rdev->sb_page, READ, true))
923 printk(KERN_WARNING "md: disabled device %s, could not read superblock.\n",
924 bdevname(rdev->bdev,b));
928 static int uuid_equal(mdp_super_t *sb1, mdp_super_t *sb2)
930 return sb1->set_uuid0 == sb2->set_uuid0 &&
931 sb1->set_uuid1 == sb2->set_uuid1 &&
932 sb1->set_uuid2 == sb2->set_uuid2 &&
933 sb1->set_uuid3 == sb2->set_uuid3;
936 static int sb_equal(mdp_super_t *sb1, mdp_super_t *sb2)
939 mdp_super_t *tmp1, *tmp2;
941 tmp1 = kmalloc(sizeof(*tmp1),GFP_KERNEL);
942 tmp2 = kmalloc(sizeof(*tmp2),GFP_KERNEL);
944 if (!tmp1 || !tmp2) {
946 printk(KERN_INFO "md.c sb_equal(): failed to allocate memory!\n");
954 * nr_disks is not constant
959 ret = (memcmp(tmp1, tmp2, MD_SB_GENERIC_CONSTANT_WORDS * 4) == 0);
967 static u32 md_csum_fold(u32 csum)
969 csum = (csum & 0xffff) + (csum >> 16);
970 return (csum & 0xffff) + (csum >> 16);
973 static unsigned int calc_sb_csum(mdp_super_t * sb)
976 u32 *sb32 = (u32*)sb;
978 unsigned int disk_csum, csum;
980 disk_csum = sb->sb_csum;
983 for (i = 0; i < MD_SB_BYTES/4 ; i++)
985 csum = (newcsum & 0xffffffff) + (newcsum>>32);
989 /* This used to use csum_partial, which was wrong for several
990 * reasons including that different results are returned on
991 * different architectures. It isn't critical that we get exactly
992 * the same return value as before (we always csum_fold before
993 * testing, and that removes any differences). However as we
994 * know that csum_partial always returned a 16bit value on
995 * alphas, do a fold to maximise conformity to previous behaviour.
997 sb->sb_csum = md_csum_fold(disk_csum);
999 sb->sb_csum = disk_csum;
1006 * Handle superblock details.
1007 * We want to be able to handle multiple superblock formats
1008 * so we have a common interface to them all, and an array of
1009 * different handlers.
1010 * We rely on user-space to write the initial superblock, and support
1011 * reading and updating of superblocks.
1012 * Interface methods are:
1013 * int load_super(mdk_rdev_t *dev, mdk_rdev_t *refdev, int minor_version)
1014 * loads and validates a superblock on dev.
1015 * if refdev != NULL, compare superblocks on both devices
1017 * 0 - dev has a superblock that is compatible with refdev
1018 * 1 - dev has a superblock that is compatible and newer than refdev
1019 * so dev should be used as the refdev in future
1020 * -EINVAL superblock incompatible or invalid
1021 * -othererror e.g. -EIO
1023 * int validate_super(mddev_t *mddev, mdk_rdev_t *dev)
1024 * Verify that dev is acceptable into mddev.
1025 * The first time, mddev->raid_disks will be 0, and data from
1026 * dev should be merged in. Subsequent calls check that dev
1027 * is new enough. Return 0 or -EINVAL
1029 * void sync_super(mddev_t *mddev, mdk_rdev_t *dev)
1030 * Update the superblock for rdev with data in mddev
1031 * This does not write to disc.
1037 struct module *owner;
1038 int (*load_super)(mdk_rdev_t *rdev, mdk_rdev_t *refdev,
1040 int (*validate_super)(mddev_t *mddev, mdk_rdev_t *rdev);
1041 void (*sync_super)(mddev_t *mddev, mdk_rdev_t *rdev);
1042 unsigned long long (*rdev_size_change)(mdk_rdev_t *rdev,
1043 sector_t num_sectors);
1047 * Check that the given mddev has no bitmap.
1049 * This function is called from the run method of all personalities that do not
1050 * support bitmaps. It prints an error message and returns non-zero if mddev
1051 * has a bitmap. Otherwise, it returns 0.
1054 int md_check_no_bitmap(mddev_t *mddev)
1056 if (!mddev->bitmap_info.file && !mddev->bitmap_info.offset)
1058 printk(KERN_ERR "%s: bitmaps are not supported for %s\n",
1059 mdname(mddev), mddev->pers->name);
1062 EXPORT_SYMBOL(md_check_no_bitmap);
1065 * load_super for 0.90.0
1067 static int super_90_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
1069 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
1074 * Calculate the position of the superblock (512byte sectors),
1075 * it's at the end of the disk.
1077 * It also happens to be a multiple of 4Kb.
1079 rdev->sb_start = calc_dev_sboffset(rdev);
1081 ret = read_disk_sb(rdev, MD_SB_BYTES);
1082 if (ret) return ret;
1086 bdevname(rdev->bdev, b);
1087 sb = page_address(rdev->sb_page);
1089 if (sb->md_magic != MD_SB_MAGIC) {
1090 printk(KERN_ERR "md: invalid raid superblock magic on %s\n",
1095 if (sb->major_version != 0 ||
1096 sb->minor_version < 90 ||
1097 sb->minor_version > 91) {
1098 printk(KERN_WARNING "Bad version number %d.%d on %s\n",
1099 sb->major_version, sb->minor_version,
1104 if (sb->raid_disks <= 0)
1107 if (md_csum_fold(calc_sb_csum(sb)) != md_csum_fold(sb->sb_csum)) {
1108 printk(KERN_WARNING "md: invalid superblock checksum on %s\n",
1113 rdev->preferred_minor = sb->md_minor;
1114 rdev->data_offset = 0;
1115 rdev->sb_size = MD_SB_BYTES;
1116 rdev->badblocks.shift = -1;
1118 if (sb->level == LEVEL_MULTIPATH)
1121 rdev->desc_nr = sb->this_disk.number;
1127 mdp_super_t *refsb = page_address(refdev->sb_page);
1128 if (!uuid_equal(refsb, sb)) {
1129 printk(KERN_WARNING "md: %s has different UUID to %s\n",
1130 b, bdevname(refdev->bdev,b2));
1133 if (!sb_equal(refsb, sb)) {
1134 printk(KERN_WARNING "md: %s has same UUID"
1135 " but different superblock to %s\n",
1136 b, bdevname(refdev->bdev, b2));
1140 ev2 = md_event(refsb);
1146 rdev->sectors = rdev->sb_start;
1147 /* Limit to 4TB as metadata cannot record more than that */
1148 if (rdev->sectors >= (2ULL << 32))
1149 rdev->sectors = (2ULL << 32) - 2;
1151 if (rdev->sectors < ((sector_t)sb->size) * 2 && sb->level >= 1)
1152 /* "this cannot possibly happen" ... */
1160 * validate_super for 0.90.0
1162 static int super_90_validate(mddev_t *mddev, mdk_rdev_t *rdev)
1165 mdp_super_t *sb = page_address(rdev->sb_page);
1166 __u64 ev1 = md_event(sb);
1168 rdev->raid_disk = -1;
1169 clear_bit(Faulty, &rdev->flags);
1170 clear_bit(In_sync, &rdev->flags);
1171 clear_bit(WriteMostly, &rdev->flags);
1173 if (mddev->raid_disks == 0) {
1174 mddev->major_version = 0;
1175 mddev->minor_version = sb->minor_version;
1176 mddev->patch_version = sb->patch_version;
1177 mddev->external = 0;
1178 mddev->chunk_sectors = sb->chunk_size >> 9;
1179 mddev->ctime = sb->ctime;
1180 mddev->utime = sb->utime;
1181 mddev->level = sb->level;
1182 mddev->clevel[0] = 0;
1183 mddev->layout = sb->layout;
1184 mddev->raid_disks = sb->raid_disks;
1185 mddev->dev_sectors = ((sector_t)sb->size) * 2;
1186 mddev->events = ev1;
1187 mddev->bitmap_info.offset = 0;
1188 mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9;
1190 if (mddev->minor_version >= 91) {
1191 mddev->reshape_position = sb->reshape_position;
1192 mddev->delta_disks = sb->delta_disks;
1193 mddev->new_level = sb->new_level;
1194 mddev->new_layout = sb->new_layout;
1195 mddev->new_chunk_sectors = sb->new_chunk >> 9;
1197 mddev->reshape_position = MaxSector;
1198 mddev->delta_disks = 0;
1199 mddev->new_level = mddev->level;
1200 mddev->new_layout = mddev->layout;
1201 mddev->new_chunk_sectors = mddev->chunk_sectors;
1204 if (sb->state & (1<<MD_SB_CLEAN))
1205 mddev->recovery_cp = MaxSector;
1207 if (sb->events_hi == sb->cp_events_hi &&
1208 sb->events_lo == sb->cp_events_lo) {
1209 mddev->recovery_cp = sb->recovery_cp;
1211 mddev->recovery_cp = 0;
1214 memcpy(mddev->uuid+0, &sb->set_uuid0, 4);
1215 memcpy(mddev->uuid+4, &sb->set_uuid1, 4);
1216 memcpy(mddev->uuid+8, &sb->set_uuid2, 4);
1217 memcpy(mddev->uuid+12,&sb->set_uuid3, 4);
1219 mddev->max_disks = MD_SB_DISKS;
1221 if (sb->state & (1<<MD_SB_BITMAP_PRESENT) &&
1222 mddev->bitmap_info.file == NULL)
1223 mddev->bitmap_info.offset =
1224 mddev->bitmap_info.default_offset;
1226 } else if (mddev->pers == NULL) {
1227 /* Insist on good event counter while assembling, except
1228 * for spares (which don't need an event count) */
1230 if (sb->disks[rdev->desc_nr].state & (
1231 (1<<MD_DISK_SYNC) | (1 << MD_DISK_ACTIVE)))
1232 if (ev1 < mddev->events)
1234 } else if (mddev->bitmap) {
1235 /* if adding to array with a bitmap, then we can accept an
1236 * older device ... but not too old.
1238 if (ev1 < mddev->bitmap->events_cleared)
1241 if (ev1 < mddev->events)
1242 /* just a hot-add of a new device, leave raid_disk at -1 */
1246 if (mddev->level != LEVEL_MULTIPATH) {
1247 desc = sb->disks + rdev->desc_nr;
1249 if (desc->state & (1<<MD_DISK_FAULTY))
1250 set_bit(Faulty, &rdev->flags);
1251 else if (desc->state & (1<<MD_DISK_SYNC) /* &&
1252 desc->raid_disk < mddev->raid_disks */) {
1253 set_bit(In_sync, &rdev->flags);
1254 rdev->raid_disk = desc->raid_disk;
1255 } else if (desc->state & (1<<MD_DISK_ACTIVE)) {
1256 /* active but not in sync implies recovery up to
1257 * reshape position. We don't know exactly where
1258 * that is, so set to zero for now */
1259 if (mddev->minor_version >= 91) {
1260 rdev->recovery_offset = 0;
1261 rdev->raid_disk = desc->raid_disk;
1264 if (desc->state & (1<<MD_DISK_WRITEMOSTLY))
1265 set_bit(WriteMostly, &rdev->flags);
1266 } else /* MULTIPATH are always insync */
1267 set_bit(In_sync, &rdev->flags);
1272 * sync_super for 0.90.0
1274 static void super_90_sync(mddev_t *mddev, mdk_rdev_t *rdev)
1278 int next_spare = mddev->raid_disks;
1281 /* make rdev->sb match mddev data..
1284 * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
1285 * 3/ any empty disks < next_spare become removed
1287 * disks[0] gets initialised to REMOVED because
1288 * we cannot be sure from other fields if it has
1289 * been initialised or not.
1292 int active=0, working=0,failed=0,spare=0,nr_disks=0;
1294 rdev->sb_size = MD_SB_BYTES;
1296 sb = page_address(rdev->sb_page);
1298 memset(sb, 0, sizeof(*sb));
1300 sb->md_magic = MD_SB_MAGIC;
1301 sb->major_version = mddev->major_version;
1302 sb->patch_version = mddev->patch_version;
1303 sb->gvalid_words = 0; /* ignored */
1304 memcpy(&sb->set_uuid0, mddev->uuid+0, 4);
1305 memcpy(&sb->set_uuid1, mddev->uuid+4, 4);
1306 memcpy(&sb->set_uuid2, mddev->uuid+8, 4);
1307 memcpy(&sb->set_uuid3, mddev->uuid+12,4);
1309 sb->ctime = mddev->ctime;
1310 sb->level = mddev->level;
1311 sb->size = mddev->dev_sectors / 2;
1312 sb->raid_disks = mddev->raid_disks;
1313 sb->md_minor = mddev->md_minor;
1314 sb->not_persistent = 0;
1315 sb->utime = mddev->utime;
1317 sb->events_hi = (mddev->events>>32);
1318 sb->events_lo = (u32)mddev->events;
1320 if (mddev->reshape_position == MaxSector)
1321 sb->minor_version = 90;
1323 sb->minor_version = 91;
1324 sb->reshape_position = mddev->reshape_position;
1325 sb->new_level = mddev->new_level;
1326 sb->delta_disks = mddev->delta_disks;
1327 sb->new_layout = mddev->new_layout;
1328 sb->new_chunk = mddev->new_chunk_sectors << 9;
1330 mddev->minor_version = sb->minor_version;
1333 sb->recovery_cp = mddev->recovery_cp;
1334 sb->cp_events_hi = (mddev->events>>32);
1335 sb->cp_events_lo = (u32)mddev->events;
1336 if (mddev->recovery_cp == MaxSector)
1337 sb->state = (1<< MD_SB_CLEAN);
1339 sb->recovery_cp = 0;
1341 sb->layout = mddev->layout;
1342 sb->chunk_size = mddev->chunk_sectors << 9;
1344 if (mddev->bitmap && mddev->bitmap_info.file == NULL)
1345 sb->state |= (1<<MD_SB_BITMAP_PRESENT);
1347 sb->disks[0].state = (1<<MD_DISK_REMOVED);
1348 list_for_each_entry(rdev2, &mddev->disks, same_set) {
1351 int is_active = test_bit(In_sync, &rdev2->flags);
1353 if (rdev2->raid_disk >= 0 &&
1354 sb->minor_version >= 91)
1355 /* we have nowhere to store the recovery_offset,
1356 * but if it is not below the reshape_position,
1357 * we can piggy-back on that.
1360 if (rdev2->raid_disk < 0 ||
1361 test_bit(Faulty, &rdev2->flags))
1364 desc_nr = rdev2->raid_disk;
1366 desc_nr = next_spare++;
1367 rdev2->desc_nr = desc_nr;
1368 d = &sb->disks[rdev2->desc_nr];
1370 d->number = rdev2->desc_nr;
1371 d->major = MAJOR(rdev2->bdev->bd_dev);
1372 d->minor = MINOR(rdev2->bdev->bd_dev);
1374 d->raid_disk = rdev2->raid_disk;
1376 d->raid_disk = rdev2->desc_nr; /* compatibility */
1377 if (test_bit(Faulty, &rdev2->flags))
1378 d->state = (1<<MD_DISK_FAULTY);
1379 else if (is_active) {
1380 d->state = (1<<MD_DISK_ACTIVE);
1381 if (test_bit(In_sync, &rdev2->flags))
1382 d->state |= (1<<MD_DISK_SYNC);
1390 if (test_bit(WriteMostly, &rdev2->flags))
1391 d->state |= (1<<MD_DISK_WRITEMOSTLY);
1393 /* now set the "removed" and "faulty" bits on any missing devices */
1394 for (i=0 ; i < mddev->raid_disks ; i++) {
1395 mdp_disk_t *d = &sb->disks[i];
1396 if (d->state == 0 && d->number == 0) {
1399 d->state = (1<<MD_DISK_REMOVED);
1400 d->state |= (1<<MD_DISK_FAULTY);
1404 sb->nr_disks = nr_disks;
1405 sb->active_disks = active;
1406 sb->working_disks = working;
1407 sb->failed_disks = failed;
1408 sb->spare_disks = spare;
1410 sb->this_disk = sb->disks[rdev->desc_nr];
1411 sb->sb_csum = calc_sb_csum(sb);
1415 * rdev_size_change for 0.90.0
1417 static unsigned long long
1418 super_90_rdev_size_change(mdk_rdev_t *rdev, sector_t num_sectors)
1420 if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
1421 return 0; /* component must fit device */
1422 if (rdev->mddev->bitmap_info.offset)
1423 return 0; /* can't move bitmap */
1424 rdev->sb_start = calc_dev_sboffset(rdev);
1425 if (!num_sectors || num_sectors > rdev->sb_start)
1426 num_sectors = rdev->sb_start;
1427 /* Limit to 4TB as metadata cannot record more than that.
1428 * 4TB == 2^32 KB, or 2*2^32 sectors.
1430 if (num_sectors >= (2ULL << 32))
1431 num_sectors = (2ULL << 32) - 2;
1432 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
1434 md_super_wait(rdev->mddev);
1440 * version 1 superblock
1443 static __le32 calc_sb_1_csum(struct mdp_superblock_1 * sb)
1447 unsigned long long newcsum;
1448 int size = 256 + le32_to_cpu(sb->max_dev)*2;
1449 __le32 *isuper = (__le32*)sb;
1452 disk_csum = sb->sb_csum;
1455 for (i=0; size>=4; size -= 4 )
1456 newcsum += le32_to_cpu(*isuper++);
1459 newcsum += le16_to_cpu(*(__le16*) isuper);
1461 csum = (newcsum & 0xffffffff) + (newcsum >> 32);
1462 sb->sb_csum = disk_csum;
1463 return cpu_to_le32(csum);
1466 static int md_set_badblocks(struct badblocks *bb, sector_t s, int sectors,
1468 static int super_1_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
1470 struct mdp_superblock_1 *sb;
1473 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
1477 * Calculate the position of the superblock in 512byte sectors.
1478 * It is always aligned to a 4K boundary and
1479 * depeding on minor_version, it can be:
1480 * 0: At least 8K, but less than 12K, from end of device
1481 * 1: At start of device
1482 * 2: 4K from start of device.
1484 switch(minor_version) {
1486 sb_start = i_size_read(rdev->bdev->bd_inode) >> 9;
1488 sb_start &= ~(sector_t)(4*2-1);
1499 rdev->sb_start = sb_start;
1501 /* superblock is rarely larger than 1K, but it can be larger,
1502 * and it is safe to read 4k, so we do that
1504 ret = read_disk_sb(rdev, 4096);
1505 if (ret) return ret;
1508 sb = page_address(rdev->sb_page);
1510 if (sb->magic != cpu_to_le32(MD_SB_MAGIC) ||
1511 sb->major_version != cpu_to_le32(1) ||
1512 le32_to_cpu(sb->max_dev) > (4096-256)/2 ||
1513 le64_to_cpu(sb->super_offset) != rdev->sb_start ||
1514 (le32_to_cpu(sb->feature_map) & ~MD_FEATURE_ALL) != 0)
1517 if (calc_sb_1_csum(sb) != sb->sb_csum) {
1518 printk("md: invalid superblock checksum on %s\n",
1519 bdevname(rdev->bdev,b));
1522 if (le64_to_cpu(sb->data_size) < 10) {
1523 printk("md: data_size too small on %s\n",
1524 bdevname(rdev->bdev,b));
1528 rdev->preferred_minor = 0xffff;
1529 rdev->data_offset = le64_to_cpu(sb->data_offset);
1530 atomic_set(&rdev->corrected_errors, le32_to_cpu(sb->cnt_corrected_read));
1532 rdev->sb_size = le32_to_cpu(sb->max_dev) * 2 + 256;
1533 bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
1534 if (rdev->sb_size & bmask)
1535 rdev->sb_size = (rdev->sb_size | bmask) + 1;
1538 && rdev->data_offset < sb_start + (rdev->sb_size/512))
1541 if (sb->level == cpu_to_le32(LEVEL_MULTIPATH))
1544 rdev->desc_nr = le32_to_cpu(sb->dev_number);
1546 if (!rdev->bb_page) {
1547 rdev->bb_page = alloc_page(GFP_KERNEL);
1551 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BAD_BLOCKS) &&
1552 rdev->badblocks.count == 0) {
1553 /* need to load the bad block list.
1554 * Currently we limit it to one page.
1560 int sectors = le16_to_cpu(sb->bblog_size);
1561 if (sectors > (PAGE_SIZE / 512))
1563 offset = le32_to_cpu(sb->bblog_offset);
1566 bb_sector = (long long)offset;
1567 if (!sync_page_io(rdev, bb_sector, sectors << 9,
1568 rdev->bb_page, READ, true))
1570 bbp = (u64 *)page_address(rdev->bb_page);
1571 rdev->badblocks.shift = sb->bblog_shift;
1572 for (i = 0 ; i < (sectors << (9-3)) ; i++, bbp++) {
1573 u64 bb = le64_to_cpu(*bbp);
1574 int count = bb & (0x3ff);
1575 u64 sector = bb >> 10;
1576 sector <<= sb->bblog_shift;
1577 count <<= sb->bblog_shift;
1580 if (md_set_badblocks(&rdev->badblocks,
1581 sector, count, 1) == 0)
1584 } else if (sb->bblog_offset == 0)
1585 rdev->badblocks.shift = -1;
1591 struct mdp_superblock_1 *refsb = page_address(refdev->sb_page);
1593 if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 ||
1594 sb->level != refsb->level ||
1595 sb->layout != refsb->layout ||
1596 sb->chunksize != refsb->chunksize) {
1597 printk(KERN_WARNING "md: %s has strangely different"
1598 " superblock to %s\n",
1599 bdevname(rdev->bdev,b),
1600 bdevname(refdev->bdev,b2));
1603 ev1 = le64_to_cpu(sb->events);
1604 ev2 = le64_to_cpu(refsb->events);
1612 rdev->sectors = (i_size_read(rdev->bdev->bd_inode) >> 9) -
1613 le64_to_cpu(sb->data_offset);
1615 rdev->sectors = rdev->sb_start;
1616 if (rdev->sectors < le64_to_cpu(sb->data_size))
1618 rdev->sectors = le64_to_cpu(sb->data_size);
1619 if (le64_to_cpu(sb->size) > rdev->sectors)
1624 static int super_1_validate(mddev_t *mddev, mdk_rdev_t *rdev)
1626 struct mdp_superblock_1 *sb = page_address(rdev->sb_page);
1627 __u64 ev1 = le64_to_cpu(sb->events);
1629 rdev->raid_disk = -1;
1630 clear_bit(Faulty, &rdev->flags);
1631 clear_bit(In_sync, &rdev->flags);
1632 clear_bit(WriteMostly, &rdev->flags);
1634 if (mddev->raid_disks == 0) {
1635 mddev->major_version = 1;
1636 mddev->patch_version = 0;
1637 mddev->external = 0;
1638 mddev->chunk_sectors = le32_to_cpu(sb->chunksize);
1639 mddev->ctime = le64_to_cpu(sb->ctime) & ((1ULL << 32)-1);
1640 mddev->utime = le64_to_cpu(sb->utime) & ((1ULL << 32)-1);
1641 mddev->level = le32_to_cpu(sb->level);
1642 mddev->clevel[0] = 0;
1643 mddev->layout = le32_to_cpu(sb->layout);
1644 mddev->raid_disks = le32_to_cpu(sb->raid_disks);
1645 mddev->dev_sectors = le64_to_cpu(sb->size);
1646 mddev->events = ev1;
1647 mddev->bitmap_info.offset = 0;
1648 mddev->bitmap_info.default_offset = 1024 >> 9;
1650 mddev->recovery_cp = le64_to_cpu(sb->resync_offset);
1651 memcpy(mddev->uuid, sb->set_uuid, 16);
1653 mddev->max_disks = (4096-256)/2;
1655 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) &&
1656 mddev->bitmap_info.file == NULL )
1657 mddev->bitmap_info.offset =
1658 (__s32)le32_to_cpu(sb->bitmap_offset);
1660 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) {
1661 mddev->reshape_position = le64_to_cpu(sb->reshape_position);
1662 mddev->delta_disks = le32_to_cpu(sb->delta_disks);
1663 mddev->new_level = le32_to_cpu(sb->new_level);
1664 mddev->new_layout = le32_to_cpu(sb->new_layout);
1665 mddev->new_chunk_sectors = le32_to_cpu(sb->new_chunk);
1667 mddev->reshape_position = MaxSector;
1668 mddev->delta_disks = 0;
1669 mddev->new_level = mddev->level;
1670 mddev->new_layout = mddev->layout;
1671 mddev->new_chunk_sectors = mddev->chunk_sectors;
1674 } else if (mddev->pers == NULL) {
1675 /* Insist of good event counter while assembling, except for
1676 * spares (which don't need an event count) */
1678 if (rdev->desc_nr >= 0 &&
1679 rdev->desc_nr < le32_to_cpu(sb->max_dev) &&
1680 le16_to_cpu(sb->dev_roles[rdev->desc_nr]) < 0xfffe)
1681 if (ev1 < mddev->events)
1683 } else if (mddev->bitmap) {
1684 /* If adding to array with a bitmap, then we can accept an
1685 * older device, but not too old.
1687 if (ev1 < mddev->bitmap->events_cleared)
1690 if (ev1 < mddev->events)
1691 /* just a hot-add of a new device, leave raid_disk at -1 */
1694 if (mddev->level != LEVEL_MULTIPATH) {
1696 if (rdev->desc_nr < 0 ||
1697 rdev->desc_nr >= le32_to_cpu(sb->max_dev)) {
1701 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
1703 case 0xffff: /* spare */
1705 case 0xfffe: /* faulty */
1706 set_bit(Faulty, &rdev->flags);
1709 if ((le32_to_cpu(sb->feature_map) &
1710 MD_FEATURE_RECOVERY_OFFSET))
1711 rdev->recovery_offset = le64_to_cpu(sb->recovery_offset);
1713 set_bit(In_sync, &rdev->flags);
1714 rdev->raid_disk = role;
1717 if (sb->devflags & WriteMostly1)
1718 set_bit(WriteMostly, &rdev->flags);
1719 } else /* MULTIPATH are always insync */
1720 set_bit(In_sync, &rdev->flags);
1725 static void super_1_sync(mddev_t *mddev, mdk_rdev_t *rdev)
1727 struct mdp_superblock_1 *sb;
1730 /* make rdev->sb match mddev and rdev data. */
1732 sb = page_address(rdev->sb_page);
1734 sb->feature_map = 0;
1736 sb->recovery_offset = cpu_to_le64(0);
1737 memset(sb->pad1, 0, sizeof(sb->pad1));
1738 memset(sb->pad3, 0, sizeof(sb->pad3));
1740 sb->utime = cpu_to_le64((__u64)mddev->utime);
1741 sb->events = cpu_to_le64(mddev->events);
1743 sb->resync_offset = cpu_to_le64(mddev->recovery_cp);
1745 sb->resync_offset = cpu_to_le64(0);
1747 sb->cnt_corrected_read = cpu_to_le32(atomic_read(&rdev->corrected_errors));
1749 sb->raid_disks = cpu_to_le32(mddev->raid_disks);
1750 sb->size = cpu_to_le64(mddev->dev_sectors);
1751 sb->chunksize = cpu_to_le32(mddev->chunk_sectors);
1752 sb->level = cpu_to_le32(mddev->level);
1753 sb->layout = cpu_to_le32(mddev->layout);
1755 if (test_bit(WriteMostly, &rdev->flags))
1756 sb->devflags |= WriteMostly1;
1758 sb->devflags &= ~WriteMostly1;
1760 if (mddev->bitmap && mddev->bitmap_info.file == NULL) {
1761 sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_info.offset);
1762 sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET);
1765 if (rdev->raid_disk >= 0 &&
1766 !test_bit(In_sync, &rdev->flags)) {
1768 cpu_to_le32(MD_FEATURE_RECOVERY_OFFSET);
1769 sb->recovery_offset =
1770 cpu_to_le64(rdev->recovery_offset);
1773 if (mddev->reshape_position != MaxSector) {
1774 sb->feature_map |= cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE);
1775 sb->reshape_position = cpu_to_le64(mddev->reshape_position);
1776 sb->new_layout = cpu_to_le32(mddev->new_layout);
1777 sb->delta_disks = cpu_to_le32(mddev->delta_disks);
1778 sb->new_level = cpu_to_le32(mddev->new_level);
1779 sb->new_chunk = cpu_to_le32(mddev->new_chunk_sectors);
1782 if (rdev->badblocks.count == 0)
1783 /* Nothing to do for bad blocks*/ ;
1784 else if (sb->bblog_offset == 0)
1785 /* Cannot record bad blocks on this device */
1786 md_error(mddev, rdev);
1788 struct badblocks *bb = &rdev->badblocks;
1789 u64 *bbp = (u64 *)page_address(rdev->bb_page);
1791 sb->feature_map |= cpu_to_le32(MD_FEATURE_BAD_BLOCKS);
1796 seq = read_seqbegin(&bb->lock);
1798 memset(bbp, 0xff, PAGE_SIZE);
1800 for (i = 0 ; i < bb->count ; i++) {
1801 u64 internal_bb = *p++;
1802 u64 store_bb = ((BB_OFFSET(internal_bb) << 10)
1803 | BB_LEN(internal_bb));
1804 *bbp++ = cpu_to_le64(store_bb);
1806 if (read_seqretry(&bb->lock, seq))
1809 bb->sector = (rdev->sb_start +
1810 (int)le32_to_cpu(sb->bblog_offset));
1811 bb->size = le16_to_cpu(sb->bblog_size);
1817 list_for_each_entry(rdev2, &mddev->disks, same_set)
1818 if (rdev2->desc_nr+1 > max_dev)
1819 max_dev = rdev2->desc_nr+1;
1821 if (max_dev > le32_to_cpu(sb->max_dev)) {
1823 sb->max_dev = cpu_to_le32(max_dev);
1824 rdev->sb_size = max_dev * 2 + 256;
1825 bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
1826 if (rdev->sb_size & bmask)
1827 rdev->sb_size = (rdev->sb_size | bmask) + 1;
1829 max_dev = le32_to_cpu(sb->max_dev);
1831 for (i=0; i<max_dev;i++)
1832 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1834 list_for_each_entry(rdev2, &mddev->disks, same_set) {
1836 if (test_bit(Faulty, &rdev2->flags))
1837 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1838 else if (test_bit(In_sync, &rdev2->flags))
1839 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1840 else if (rdev2->raid_disk >= 0)
1841 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1843 sb->dev_roles[i] = cpu_to_le16(0xffff);
1846 sb->sb_csum = calc_sb_1_csum(sb);
1849 static unsigned long long
1850 super_1_rdev_size_change(mdk_rdev_t *rdev, sector_t num_sectors)
1852 struct mdp_superblock_1 *sb;
1853 sector_t max_sectors;
1854 if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
1855 return 0; /* component must fit device */
1856 if (rdev->sb_start < rdev->data_offset) {
1857 /* minor versions 1 and 2; superblock before data */
1858 max_sectors = i_size_read(rdev->bdev->bd_inode) >> 9;
1859 max_sectors -= rdev->data_offset;
1860 if (!num_sectors || num_sectors > max_sectors)
1861 num_sectors = max_sectors;
1862 } else if (rdev->mddev->bitmap_info.offset) {
1863 /* minor version 0 with bitmap we can't move */
1866 /* minor version 0; superblock after data */
1868 sb_start = (i_size_read(rdev->bdev->bd_inode) >> 9) - 8*2;
1869 sb_start &= ~(sector_t)(4*2 - 1);
1870 max_sectors = rdev->sectors + sb_start - rdev->sb_start;
1871 if (!num_sectors || num_sectors > max_sectors)
1872 num_sectors = max_sectors;
1873 rdev->sb_start = sb_start;
1875 sb = page_address(rdev->sb_page);
1876 sb->data_size = cpu_to_le64(num_sectors);
1877 sb->super_offset = rdev->sb_start;
1878 sb->sb_csum = calc_sb_1_csum(sb);
1879 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
1881 md_super_wait(rdev->mddev);
1885 static struct super_type super_types[] = {
1888 .owner = THIS_MODULE,
1889 .load_super = super_90_load,
1890 .validate_super = super_90_validate,
1891 .sync_super = super_90_sync,
1892 .rdev_size_change = super_90_rdev_size_change,
1896 .owner = THIS_MODULE,
1897 .load_super = super_1_load,
1898 .validate_super = super_1_validate,
1899 .sync_super = super_1_sync,
1900 .rdev_size_change = super_1_rdev_size_change,
1904 static void sync_super(mddev_t *mddev, mdk_rdev_t *rdev)
1906 if (mddev->sync_super) {
1907 mddev->sync_super(mddev, rdev);
1911 BUG_ON(mddev->major_version >= ARRAY_SIZE(super_types));
1913 super_types[mddev->major_version].sync_super(mddev, rdev);
1916 static int match_mddev_units(mddev_t *mddev1, mddev_t *mddev2)
1918 mdk_rdev_t *rdev, *rdev2;
1921 rdev_for_each_rcu(rdev, mddev1)
1922 rdev_for_each_rcu(rdev2, mddev2)
1923 if (rdev->bdev->bd_contains ==
1924 rdev2->bdev->bd_contains) {
1932 static LIST_HEAD(pending_raid_disks);
1935 * Try to register data integrity profile for an mddev
1937 * This is called when an array is started and after a disk has been kicked
1938 * from the array. It only succeeds if all working and active component devices
1939 * are integrity capable with matching profiles.
1941 int md_integrity_register(mddev_t *mddev)
1943 mdk_rdev_t *rdev, *reference = NULL;
1945 if (list_empty(&mddev->disks))
1946 return 0; /* nothing to do */
1947 if (!mddev->gendisk || blk_get_integrity(mddev->gendisk))
1948 return 0; /* shouldn't register, or already is */
1949 list_for_each_entry(rdev, &mddev->disks, same_set) {
1950 /* skip spares and non-functional disks */
1951 if (test_bit(Faulty, &rdev->flags))
1953 if (rdev->raid_disk < 0)
1956 /* Use the first rdev as the reference */
1960 /* does this rdev's profile match the reference profile? */
1961 if (blk_integrity_compare(reference->bdev->bd_disk,
1962 rdev->bdev->bd_disk) < 0)
1965 if (!reference || !bdev_get_integrity(reference->bdev))
1968 * All component devices are integrity capable and have matching
1969 * profiles, register the common profile for the md device.
1971 if (blk_integrity_register(mddev->gendisk,
1972 bdev_get_integrity(reference->bdev)) != 0) {
1973 printk(KERN_ERR "md: failed to register integrity for %s\n",
1977 printk(KERN_NOTICE "md: data integrity enabled on %s\n", mdname(mddev));
1978 if (bioset_integrity_create(mddev->bio_set, BIO_POOL_SIZE)) {
1979 printk(KERN_ERR "md: failed to create integrity pool for %s\n",
1985 EXPORT_SYMBOL(md_integrity_register);
1987 /* Disable data integrity if non-capable/non-matching disk is being added */
1988 void md_integrity_add_rdev(mdk_rdev_t *rdev, mddev_t *mddev)
1990 struct blk_integrity *bi_rdev = bdev_get_integrity(rdev->bdev);
1991 struct blk_integrity *bi_mddev = blk_get_integrity(mddev->gendisk);
1993 if (!bi_mddev) /* nothing to do */
1995 if (rdev->raid_disk < 0) /* skip spares */
1997 if (bi_rdev && blk_integrity_compare(mddev->gendisk,
1998 rdev->bdev->bd_disk) >= 0)
2000 printk(KERN_NOTICE "disabling data integrity on %s\n", mdname(mddev));
2001 blk_integrity_unregister(mddev->gendisk);
2003 EXPORT_SYMBOL(md_integrity_add_rdev);
2005 static int bind_rdev_to_array(mdk_rdev_t * rdev, mddev_t * mddev)
2007 char b[BDEVNAME_SIZE];
2017 /* prevent duplicates */
2018 if (find_rdev(mddev, rdev->bdev->bd_dev))
2021 /* make sure rdev->sectors exceeds mddev->dev_sectors */
2022 if (rdev->sectors && (mddev->dev_sectors == 0 ||
2023 rdev->sectors < mddev->dev_sectors)) {
2025 /* Cannot change size, so fail
2026 * If mddev->level <= 0, then we don't care
2027 * about aligning sizes (e.g. linear)
2029 if (mddev->level > 0)
2032 mddev->dev_sectors = rdev->sectors;
2035 /* Verify rdev->desc_nr is unique.
2036 * If it is -1, assign a free number, else
2037 * check number is not in use
2039 if (rdev->desc_nr < 0) {
2041 if (mddev->pers) choice = mddev->raid_disks;
2042 while (find_rdev_nr(mddev, choice))
2044 rdev->desc_nr = choice;
2046 if (find_rdev_nr(mddev, rdev->desc_nr))
2049 if (mddev->max_disks && rdev->desc_nr >= mddev->max_disks) {
2050 printk(KERN_WARNING "md: %s: array is limited to %d devices\n",
2051 mdname(mddev), mddev->max_disks);
2054 bdevname(rdev->bdev,b);
2055 while ( (s=strchr(b, '/')) != NULL)
2058 rdev->mddev = mddev;
2059 printk(KERN_INFO "md: bind<%s>\n", b);
2061 if ((err = kobject_add(&rdev->kobj, &mddev->kobj, "dev-%s", b)))
2064 ko = &part_to_dev(rdev->bdev->bd_part)->kobj;
2065 if (sysfs_create_link(&rdev->kobj, ko, "block"))
2066 /* failure here is OK */;
2067 rdev->sysfs_state = sysfs_get_dirent_safe(rdev->kobj.sd, "state");
2069 list_add_rcu(&rdev->same_set, &mddev->disks);
2070 bd_link_disk_holder(rdev->bdev, mddev->gendisk);
2072 /* May as well allow recovery to be retried once */
2073 mddev->recovery_disabled++;
2078 printk(KERN_WARNING "md: failed to register dev-%s for %s\n",
2083 static void md_delayed_delete(struct work_struct *ws)
2085 mdk_rdev_t *rdev = container_of(ws, mdk_rdev_t, del_work);
2086 kobject_del(&rdev->kobj);
2087 kobject_put(&rdev->kobj);
2090 static void unbind_rdev_from_array(mdk_rdev_t * rdev)
2092 char b[BDEVNAME_SIZE];
2097 bd_unlink_disk_holder(rdev->bdev, rdev->mddev->gendisk);
2098 list_del_rcu(&rdev->same_set);
2099 printk(KERN_INFO "md: unbind<%s>\n", bdevname(rdev->bdev,b));
2101 sysfs_remove_link(&rdev->kobj, "block");
2102 sysfs_put(rdev->sysfs_state);
2103 rdev->sysfs_state = NULL;
2104 kfree(rdev->badblocks.page);
2105 rdev->badblocks.count = 0;
2106 rdev->badblocks.page = NULL;
2107 /* We need to delay this, otherwise we can deadlock when
2108 * writing to 'remove' to "dev/state". We also need
2109 * to delay it due to rcu usage.
2112 INIT_WORK(&rdev->del_work, md_delayed_delete);
2113 kobject_get(&rdev->kobj);
2114 queue_work(md_misc_wq, &rdev->del_work);
2118 * prevent the device from being mounted, repartitioned or
2119 * otherwise reused by a RAID array (or any other kernel
2120 * subsystem), by bd_claiming the device.
2122 static int lock_rdev(mdk_rdev_t *rdev, dev_t dev, int shared)
2125 struct block_device *bdev;
2126 char b[BDEVNAME_SIZE];
2128 bdev = blkdev_get_by_dev(dev, FMODE_READ|FMODE_WRITE|FMODE_EXCL,
2129 shared ? (mdk_rdev_t *)lock_rdev : rdev);
2131 printk(KERN_ERR "md: could not open %s.\n",
2132 __bdevname(dev, b));
2133 return PTR_ERR(bdev);
2139 static void unlock_rdev(mdk_rdev_t *rdev)
2141 struct block_device *bdev = rdev->bdev;
2145 blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
2148 void md_autodetect_dev(dev_t dev);
2150 static void export_rdev(mdk_rdev_t * rdev)
2152 char b[BDEVNAME_SIZE];
2153 printk(KERN_INFO "md: export_rdev(%s)\n",
2154 bdevname(rdev->bdev,b));
2159 if (test_bit(AutoDetected, &rdev->flags))
2160 md_autodetect_dev(rdev->bdev->bd_dev);
2163 kobject_put(&rdev->kobj);
2166 static void kick_rdev_from_array(mdk_rdev_t * rdev)
2168 unbind_rdev_from_array(rdev);
2172 static void export_array(mddev_t *mddev)
2174 mdk_rdev_t *rdev, *tmp;
2176 rdev_for_each(rdev, tmp, mddev) {
2181 kick_rdev_from_array(rdev);
2183 if (!list_empty(&mddev->disks))
2185 mddev->raid_disks = 0;
2186 mddev->major_version = 0;
2189 static void print_desc(mdp_disk_t *desc)
2191 printk(" DISK<N:%d,(%d,%d),R:%d,S:%d>\n", desc->number,
2192 desc->major,desc->minor,desc->raid_disk,desc->state);
2195 static void print_sb_90(mdp_super_t *sb)
2200 "md: SB: (V:%d.%d.%d) ID:<%08x.%08x.%08x.%08x> CT:%08x\n",
2201 sb->major_version, sb->minor_version, sb->patch_version,
2202 sb->set_uuid0, sb->set_uuid1, sb->set_uuid2, sb->set_uuid3,
2204 printk(KERN_INFO "md: L%d S%08d ND:%d RD:%d md%d LO:%d CS:%d\n",
2205 sb->level, sb->size, sb->nr_disks, sb->raid_disks,
2206 sb->md_minor, sb->layout, sb->chunk_size);
2207 printk(KERN_INFO "md: UT:%08x ST:%d AD:%d WD:%d"
2208 " FD:%d SD:%d CSUM:%08x E:%08lx\n",
2209 sb->utime, sb->state, sb->active_disks, sb->working_disks,
2210 sb->failed_disks, sb->spare_disks,
2211 sb->sb_csum, (unsigned long)sb->events_lo);
2214 for (i = 0; i < MD_SB_DISKS; i++) {
2217 desc = sb->disks + i;
2218 if (desc->number || desc->major || desc->minor ||
2219 desc->raid_disk || (desc->state && (desc->state != 4))) {
2220 printk(" D %2d: ", i);
2224 printk(KERN_INFO "md: THIS: ");
2225 print_desc(&sb->this_disk);
2228 static void print_sb_1(struct mdp_superblock_1 *sb)
2232 uuid = sb->set_uuid;
2234 "md: SB: (V:%u) (F:0x%08x) Array-ID:<%pU>\n"
2235 "md: Name: \"%s\" CT:%llu\n",
2236 le32_to_cpu(sb->major_version),
2237 le32_to_cpu(sb->feature_map),
2240 (unsigned long long)le64_to_cpu(sb->ctime)
2241 & MD_SUPERBLOCK_1_TIME_SEC_MASK);
2243 uuid = sb->device_uuid;
2245 "md: L%u SZ%llu RD:%u LO:%u CS:%u DO:%llu DS:%llu SO:%llu"
2247 "md: Dev:%08x UUID: %pU\n"
2248 "md: (F:0x%08x) UT:%llu Events:%llu ResyncOffset:%llu CSUM:0x%08x\n"
2249 "md: (MaxDev:%u) \n",
2250 le32_to_cpu(sb->level),
2251 (unsigned long long)le64_to_cpu(sb->size),
2252 le32_to_cpu(sb->raid_disks),
2253 le32_to_cpu(sb->layout),
2254 le32_to_cpu(sb->chunksize),
2255 (unsigned long long)le64_to_cpu(sb->data_offset),
2256 (unsigned long long)le64_to_cpu(sb->data_size),
2257 (unsigned long long)le64_to_cpu(sb->super_offset),
2258 (unsigned long long)le64_to_cpu(sb->recovery_offset),
2259 le32_to_cpu(sb->dev_number),
2262 (unsigned long long)le64_to_cpu(sb->utime) & MD_SUPERBLOCK_1_TIME_SEC_MASK,
2263 (unsigned long long)le64_to_cpu(sb->events),
2264 (unsigned long long)le64_to_cpu(sb->resync_offset),
2265 le32_to_cpu(sb->sb_csum),
2266 le32_to_cpu(sb->max_dev)
2270 static void print_rdev(mdk_rdev_t *rdev, int major_version)
2272 char b[BDEVNAME_SIZE];
2273 printk(KERN_INFO "md: rdev %s, Sect:%08llu F:%d S:%d DN:%u\n",
2274 bdevname(rdev->bdev, b), (unsigned long long)rdev->sectors,
2275 test_bit(Faulty, &rdev->flags), test_bit(In_sync, &rdev->flags),
2277 if (rdev->sb_loaded) {
2278 printk(KERN_INFO "md: rdev superblock (MJ:%d):\n", major_version);
2279 switch (major_version) {
2281 print_sb_90(page_address(rdev->sb_page));
2284 print_sb_1(page_address(rdev->sb_page));
2288 printk(KERN_INFO "md: no rdev superblock!\n");
2291 static void md_print_devices(void)
2293 struct list_head *tmp;
2296 char b[BDEVNAME_SIZE];
2299 printk("md: **********************************\n");
2300 printk("md: * <COMPLETE RAID STATE PRINTOUT> *\n");
2301 printk("md: **********************************\n");
2302 for_each_mddev(mddev, tmp) {
2305 bitmap_print_sb(mddev->bitmap);
2307 printk("%s: ", mdname(mddev));
2308 list_for_each_entry(rdev, &mddev->disks, same_set)
2309 printk("<%s>", bdevname(rdev->bdev,b));
2312 list_for_each_entry(rdev, &mddev->disks, same_set)
2313 print_rdev(rdev, mddev->major_version);
2315 printk("md: **********************************\n");
2320 static void sync_sbs(mddev_t * mddev, int nospares)
2322 /* Update each superblock (in-memory image), but
2323 * if we are allowed to, skip spares which already
2324 * have the right event counter, or have one earlier
2325 * (which would mean they aren't being marked as dirty
2326 * with the rest of the array)
2329 list_for_each_entry(rdev, &mddev->disks, same_set) {
2330 if (rdev->sb_events == mddev->events ||
2332 rdev->raid_disk < 0 &&
2333 rdev->sb_events+1 == mddev->events)) {
2334 /* Don't update this superblock */
2335 rdev->sb_loaded = 2;
2337 sync_super(mddev, rdev);
2338 rdev->sb_loaded = 1;
2343 static void md_update_sb(mddev_t * mddev, int force_change)
2348 int any_badblocks_changed = 0;
2351 /* First make sure individual recovery_offsets are correct */
2352 list_for_each_entry(rdev, &mddev->disks, same_set) {
2353 if (rdev->raid_disk >= 0 &&
2354 mddev->delta_disks >= 0 &&
2355 !test_bit(In_sync, &rdev->flags) &&
2356 mddev->curr_resync_completed > rdev->recovery_offset)
2357 rdev->recovery_offset = mddev->curr_resync_completed;
2360 if (!mddev->persistent) {
2361 clear_bit(MD_CHANGE_CLEAN, &mddev->flags);
2362 clear_bit(MD_CHANGE_DEVS, &mddev->flags);
2363 if (!mddev->external) {
2364 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
2365 list_for_each_entry(rdev, &mddev->disks, same_set) {
2366 if (rdev->badblocks.changed) {
2367 md_ack_all_badblocks(&rdev->badblocks);
2368 md_error(mddev, rdev);
2370 clear_bit(Blocked, &rdev->flags);
2371 clear_bit(BlockedBadBlocks, &rdev->flags);
2372 wake_up(&rdev->blocked_wait);
2375 wake_up(&mddev->sb_wait);
2379 spin_lock_irq(&mddev->write_lock);
2381 mddev->utime = get_seconds();
2383 if (test_and_clear_bit(MD_CHANGE_DEVS, &mddev->flags))
2385 if (test_and_clear_bit(MD_CHANGE_CLEAN, &mddev->flags))
2386 /* just a clean<-> dirty transition, possibly leave spares alone,
2387 * though if events isn't the right even/odd, we will have to do
2393 if (mddev->degraded)
2394 /* If the array is degraded, then skipping spares is both
2395 * dangerous and fairly pointless.
2396 * Dangerous because a device that was removed from the array
2397 * might have a event_count that still looks up-to-date,
2398 * so it can be re-added without a resync.
2399 * Pointless because if there are any spares to skip,
2400 * then a recovery will happen and soon that array won't
2401 * be degraded any more and the spare can go back to sleep then.
2405 sync_req = mddev->in_sync;
2407 /* If this is just a dirty<->clean transition, and the array is clean
2408 * and 'events' is odd, we can roll back to the previous clean state */
2410 && (mddev->in_sync && mddev->recovery_cp == MaxSector)
2411 && mddev->can_decrease_events
2412 && mddev->events != 1) {
2414 mddev->can_decrease_events = 0;
2416 /* otherwise we have to go forward and ... */
2418 mddev->can_decrease_events = nospares;
2421 if (!mddev->events) {
2423 * oops, this 64-bit counter should never wrap.
2424 * Either we are in around ~1 trillion A.C., assuming
2425 * 1 reboot per second, or we have a bug:
2431 list_for_each_entry(rdev, &mddev->disks, same_set) {
2432 if (rdev->badblocks.changed)
2433 any_badblocks_changed++;
2434 if (test_bit(Faulty, &rdev->flags))
2435 set_bit(FaultRecorded, &rdev->flags);
2438 sync_sbs(mddev, nospares);
2439 spin_unlock_irq(&mddev->write_lock);
2442 "md: updating %s RAID superblock on device (in sync %d)\n",
2443 mdname(mddev),mddev->in_sync);
2445 bitmap_update_sb(mddev->bitmap);
2446 list_for_each_entry(rdev, &mddev->disks, same_set) {
2447 char b[BDEVNAME_SIZE];
2448 dprintk(KERN_INFO "md: ");
2449 if (rdev->sb_loaded != 1)
2450 continue; /* no noise on spare devices */
2451 if (test_bit(Faulty, &rdev->flags))
2452 dprintk("(skipping faulty ");
2454 dprintk("%s ", bdevname(rdev->bdev,b));
2455 if (!test_bit(Faulty, &rdev->flags)) {
2456 md_super_write(mddev,rdev,
2457 rdev->sb_start, rdev->sb_size,
2459 dprintk(KERN_INFO "(write) %s's sb offset: %llu\n",
2460 bdevname(rdev->bdev,b),
2461 (unsigned long long)rdev->sb_start);
2462 rdev->sb_events = mddev->events;
2463 if (rdev->badblocks.size) {
2464 md_super_write(mddev, rdev,
2465 rdev->badblocks.sector,
2466 rdev->badblocks.size << 9,
2468 rdev->badblocks.size = 0;
2473 if (mddev->level == LEVEL_MULTIPATH)
2474 /* only need to write one superblock... */
2477 md_super_wait(mddev);
2478 /* if there was a failure, MD_CHANGE_DEVS was set, and we re-write super */
2480 spin_lock_irq(&mddev->write_lock);
2481 if (mddev->in_sync != sync_req ||
2482 test_bit(MD_CHANGE_DEVS, &mddev->flags)) {
2483 /* have to write it out again */
2484 spin_unlock_irq(&mddev->write_lock);
2487 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
2488 spin_unlock_irq(&mddev->write_lock);
2489 wake_up(&mddev->sb_wait);
2490 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
2491 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
2493 list_for_each_entry(rdev, &mddev->disks, same_set) {
2494 if (test_and_clear_bit(FaultRecorded, &rdev->flags))
2495 clear_bit(Blocked, &rdev->flags);
2497 if (any_badblocks_changed)
2498 md_ack_all_badblocks(&rdev->badblocks);
2499 clear_bit(BlockedBadBlocks, &rdev->flags);
2500 wake_up(&rdev->blocked_wait);
2504 /* words written to sysfs files may, or may not, be \n terminated.
2505 * We want to accept with case. For this we use cmd_match.
2507 static int cmd_match(const char *cmd, const char *str)
2509 /* See if cmd, written into a sysfs file, matches
2510 * str. They must either be the same, or cmd can
2511 * have a trailing newline
2513 while (*cmd && *str && *cmd == *str) {
2524 struct rdev_sysfs_entry {
2525 struct attribute attr;
2526 ssize_t (*show)(mdk_rdev_t *, char *);
2527 ssize_t (*store)(mdk_rdev_t *, const char *, size_t);
2531 state_show(mdk_rdev_t *rdev, char *page)
2536 if (test_bit(Faulty, &rdev->flags) ||
2537 rdev->badblocks.unacked_exist) {
2538 len+= sprintf(page+len, "%sfaulty",sep);
2541 if (test_bit(In_sync, &rdev->flags)) {
2542 len += sprintf(page+len, "%sin_sync",sep);
2545 if (test_bit(WriteMostly, &rdev->flags)) {
2546 len += sprintf(page+len, "%swrite_mostly",sep);
2549 if (test_bit(Blocked, &rdev->flags) ||
2550 rdev->badblocks.unacked_exist) {
2551 len += sprintf(page+len, "%sblocked", sep);
2554 if (!test_bit(Faulty, &rdev->flags) &&
2555 !test_bit(In_sync, &rdev->flags)) {
2556 len += sprintf(page+len, "%sspare", sep);
2559 if (test_bit(WriteErrorSeen, &rdev->flags)) {
2560 len += sprintf(page+len, "%swrite_error", sep);
2563 return len+sprintf(page+len, "\n");
2567 state_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2570 * faulty - simulates an error
2571 * remove - disconnects the device
2572 * writemostly - sets write_mostly
2573 * -writemostly - clears write_mostly
2574 * blocked - sets the Blocked flags
2575 * -blocked - clears the Blocked and possibly simulates an error
2576 * insync - sets Insync providing device isn't active
2577 * write_error - sets WriteErrorSeen
2578 * -write_error - clears WriteErrorSeen
2581 if (cmd_match(buf, "faulty") && rdev->mddev->pers) {
2582 md_error(rdev->mddev, rdev);
2583 if (test_bit(Faulty, &rdev->flags))
2587 } else if (cmd_match(buf, "remove")) {
2588 if (rdev->raid_disk >= 0)
2591 mddev_t *mddev = rdev->mddev;
2592 kick_rdev_from_array(rdev);
2594 md_update_sb(mddev, 1);
2595 md_new_event(mddev);
2598 } else if (cmd_match(buf, "writemostly")) {
2599 set_bit(WriteMostly, &rdev->flags);
2601 } else if (cmd_match(buf, "-writemostly")) {
2602 clear_bit(WriteMostly, &rdev->flags);
2604 } else if (cmd_match(buf, "blocked")) {
2605 set_bit(Blocked, &rdev->flags);
2607 } else if (cmd_match(buf, "-blocked")) {
2608 if (!test_bit(Faulty, &rdev->flags) &&
2609 rdev->badblocks.unacked_exist) {
2610 /* metadata handler doesn't understand badblocks,
2611 * so we need to fail the device
2613 md_error(rdev->mddev, rdev);
2615 clear_bit(Blocked, &rdev->flags);
2616 clear_bit(BlockedBadBlocks, &rdev->flags);
2617 wake_up(&rdev->blocked_wait);
2618 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2619 md_wakeup_thread(rdev->mddev->thread);
2622 } else if (cmd_match(buf, "insync") && rdev->raid_disk == -1) {
2623 set_bit(In_sync, &rdev->flags);
2625 } else if (cmd_match(buf, "write_error")) {
2626 set_bit(WriteErrorSeen, &rdev->flags);
2628 } else if (cmd_match(buf, "-write_error")) {
2629 clear_bit(WriteErrorSeen, &rdev->flags);
2633 sysfs_notify_dirent_safe(rdev->sysfs_state);
2634 return err ? err : len;
2636 static struct rdev_sysfs_entry rdev_state =
2637 __ATTR(state, S_IRUGO|S_IWUSR, state_show, state_store);
2640 errors_show(mdk_rdev_t *rdev, char *page)
2642 return sprintf(page, "%d\n", atomic_read(&rdev->corrected_errors));
2646 errors_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2649 unsigned long n = simple_strtoul(buf, &e, 10);
2650 if (*buf && (*e == 0 || *e == '\n')) {
2651 atomic_set(&rdev->corrected_errors, n);
2656 static struct rdev_sysfs_entry rdev_errors =
2657 __ATTR(errors, S_IRUGO|S_IWUSR, errors_show, errors_store);
2660 slot_show(mdk_rdev_t *rdev, char *page)
2662 if (rdev->raid_disk < 0)
2663 return sprintf(page, "none\n");
2665 return sprintf(page, "%d\n", rdev->raid_disk);
2669 slot_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2673 int slot = simple_strtoul(buf, &e, 10);
2674 if (strncmp(buf, "none", 4)==0)
2676 else if (e==buf || (*e && *e!= '\n'))
2678 if (rdev->mddev->pers && slot == -1) {
2679 /* Setting 'slot' on an active array requires also
2680 * updating the 'rd%d' link, and communicating
2681 * with the personality with ->hot_*_disk.
2682 * For now we only support removing
2683 * failed/spare devices. This normally happens automatically,
2684 * but not when the metadata is externally managed.
2686 if (rdev->raid_disk == -1)
2688 /* personality does all needed checks */
2689 if (rdev->mddev->pers->hot_remove_disk == NULL)
2691 err = rdev->mddev->pers->
2692 hot_remove_disk(rdev->mddev, rdev->raid_disk);
2695 sysfs_unlink_rdev(rdev->mddev, rdev);
2696 rdev->raid_disk = -1;
2697 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2698 md_wakeup_thread(rdev->mddev->thread);
2699 } else if (rdev->mddev->pers) {
2701 /* Activating a spare .. or possibly reactivating
2702 * if we ever get bitmaps working here.
2705 if (rdev->raid_disk != -1)
2708 if (test_bit(MD_RECOVERY_RUNNING, &rdev->mddev->recovery))
2711 if (rdev->mddev->pers->hot_add_disk == NULL)
2714 list_for_each_entry(rdev2, &rdev->mddev->disks, same_set)
2715 if (rdev2->raid_disk == slot)
2718 if (slot >= rdev->mddev->raid_disks &&
2719 slot >= rdev->mddev->raid_disks + rdev->mddev->delta_disks)
2722 rdev->raid_disk = slot;
2723 if (test_bit(In_sync, &rdev->flags))
2724 rdev->saved_raid_disk = slot;
2726 rdev->saved_raid_disk = -1;
2727 err = rdev->mddev->pers->
2728 hot_add_disk(rdev->mddev, rdev);
2730 rdev->raid_disk = -1;
2733 sysfs_notify_dirent_safe(rdev->sysfs_state);
2734 if (sysfs_link_rdev(rdev->mddev, rdev))
2735 /* failure here is OK */;
2736 /* don't wakeup anyone, leave that to userspace. */
2738 if (slot >= rdev->mddev->raid_disks &&
2739 slot >= rdev->mddev->raid_disks + rdev->mddev->delta_disks)
2741 rdev->raid_disk = slot;
2742 /* assume it is working */
2743 clear_bit(Faulty, &rdev->flags);
2744 clear_bit(WriteMostly, &rdev->flags);
2745 set_bit(In_sync, &rdev->flags);
2746 sysfs_notify_dirent_safe(rdev->sysfs_state);
2752 static struct rdev_sysfs_entry rdev_slot =
2753 __ATTR(slot, S_IRUGO|S_IWUSR, slot_show, slot_store);
2756 offset_show(mdk_rdev_t *rdev, char *page)
2758 return sprintf(page, "%llu\n", (unsigned long long)rdev->data_offset);
2762 offset_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2765 unsigned long long offset = simple_strtoull(buf, &e, 10);
2766 if (e==buf || (*e && *e != '\n'))
2768 if (rdev->mddev->pers && rdev->raid_disk >= 0)
2770 if (rdev->sectors && rdev->mddev->external)
2771 /* Must set offset before size, so overlap checks
2774 rdev->data_offset = offset;
2778 static struct rdev_sysfs_entry rdev_offset =
2779 __ATTR(offset, S_IRUGO|S_IWUSR, offset_show, offset_store);
2782 rdev_size_show(mdk_rdev_t *rdev, char *page)
2784 return sprintf(page, "%llu\n", (unsigned long long)rdev->sectors / 2);
2787 static int overlaps(sector_t s1, sector_t l1, sector_t s2, sector_t l2)
2789 /* check if two start/length pairs overlap */
2797 static int strict_blocks_to_sectors(const char *buf, sector_t *sectors)
2799 unsigned long long blocks;
2802 if (strict_strtoull(buf, 10, &blocks) < 0)
2805 if (blocks & 1ULL << (8 * sizeof(blocks) - 1))
2806 return -EINVAL; /* sector conversion overflow */
2809 if (new != blocks * 2)
2810 return -EINVAL; /* unsigned long long to sector_t overflow */
2817 rdev_size_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2819 mddev_t *my_mddev = rdev->mddev;
2820 sector_t oldsectors = rdev->sectors;
2823 if (strict_blocks_to_sectors(buf, §ors) < 0)
2825 if (my_mddev->pers && rdev->raid_disk >= 0) {
2826 if (my_mddev->persistent) {
2827 sectors = super_types[my_mddev->major_version].
2828 rdev_size_change(rdev, sectors);
2831 } else if (!sectors)
2832 sectors = (i_size_read(rdev->bdev->bd_inode) >> 9) -
2835 if (sectors < my_mddev->dev_sectors)
2836 return -EINVAL; /* component must fit device */
2838 rdev->sectors = sectors;
2839 if (sectors > oldsectors && my_mddev->external) {
2840 /* need to check that all other rdevs with the same ->bdev
2841 * do not overlap. We need to unlock the mddev to avoid
2842 * a deadlock. We have already changed rdev->sectors, and if
2843 * we have to change it back, we will have the lock again.
2847 struct list_head *tmp;
2849 mddev_unlock(my_mddev);
2850 for_each_mddev(mddev, tmp) {
2854 list_for_each_entry(rdev2, &mddev->disks, same_set)
2855 if (rdev->bdev == rdev2->bdev &&
2857 overlaps(rdev->data_offset, rdev->sectors,
2863 mddev_unlock(mddev);
2869 mddev_lock(my_mddev);
2871 /* Someone else could have slipped in a size
2872 * change here, but doing so is just silly.
2873 * We put oldsectors back because we *know* it is
2874 * safe, and trust userspace not to race with
2877 rdev->sectors = oldsectors;
2884 static struct rdev_sysfs_entry rdev_size =
2885 __ATTR(size, S_IRUGO|S_IWUSR, rdev_size_show, rdev_size_store);
2888 static ssize_t recovery_start_show(mdk_rdev_t *rdev, char *page)
2890 unsigned long long recovery_start = rdev->recovery_offset;
2892 if (test_bit(In_sync, &rdev->flags) ||
2893 recovery_start == MaxSector)
2894 return sprintf(page, "none\n");
2896 return sprintf(page, "%llu\n", recovery_start);
2899 static ssize_t recovery_start_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2901 unsigned long long recovery_start;
2903 if (cmd_match(buf, "none"))
2904 recovery_start = MaxSector;
2905 else if (strict_strtoull(buf, 10, &recovery_start))
2908 if (rdev->mddev->pers &&
2909 rdev->raid_disk >= 0)
2912 rdev->recovery_offset = recovery_start;
2913 if (recovery_start == MaxSector)
2914 set_bit(In_sync, &rdev->flags);
2916 clear_bit(In_sync, &rdev->flags);
2920 static struct rdev_sysfs_entry rdev_recovery_start =
2921 __ATTR(recovery_start, S_IRUGO|S_IWUSR, recovery_start_show, recovery_start_store);
2925 badblocks_show(struct badblocks *bb, char *page, int unack);
2927 badblocks_store(struct badblocks *bb, const char *page, size_t len, int unack);
2929 static ssize_t bb_show(mdk_rdev_t *rdev, char *page)
2931 return badblocks_show(&rdev->badblocks, page, 0);
2933 static ssize_t bb_store(mdk_rdev_t *rdev, const char *page, size_t len)
2935 int rv = badblocks_store(&rdev->badblocks, page, len, 0);
2936 /* Maybe that ack was all we needed */
2937 if (test_and_clear_bit(BlockedBadBlocks, &rdev->flags))
2938 wake_up(&rdev->blocked_wait);
2941 static struct rdev_sysfs_entry rdev_bad_blocks =
2942 __ATTR(bad_blocks, S_IRUGO|S_IWUSR, bb_show, bb_store);
2945 static ssize_t ubb_show(mdk_rdev_t *rdev, char *page)
2947 return badblocks_show(&rdev->badblocks, page, 1);
2949 static ssize_t ubb_store(mdk_rdev_t *rdev, const char *page, size_t len)
2951 return badblocks_store(&rdev->badblocks, page, len, 1);
2953 static struct rdev_sysfs_entry rdev_unack_bad_blocks =
2954 __ATTR(unacknowledged_bad_blocks, S_IRUGO|S_IWUSR, ubb_show, ubb_store);
2956 static struct attribute *rdev_default_attrs[] = {
2962 &rdev_recovery_start.attr,
2963 &rdev_bad_blocks.attr,
2964 &rdev_unack_bad_blocks.attr,
2968 rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
2970 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
2971 mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
2972 mddev_t *mddev = rdev->mddev;
2978 rv = mddev ? mddev_lock(mddev) : -EBUSY;
2980 if (rdev->mddev == NULL)
2983 rv = entry->show(rdev, page);
2984 mddev_unlock(mddev);
2990 rdev_attr_store(struct kobject *kobj, struct attribute *attr,
2991 const char *page, size_t length)
2993 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
2994 mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
2996 mddev_t *mddev = rdev->mddev;
3000 if (!capable(CAP_SYS_ADMIN))
3002 rv = mddev ? mddev_lock(mddev): -EBUSY;
3004 if (rdev->mddev == NULL)
3007 rv = entry->store(rdev, page, length);
3008 mddev_unlock(mddev);
3013 static void rdev_free(struct kobject *ko)
3015 mdk_rdev_t *rdev = container_of(ko, mdk_rdev_t, kobj);
3018 static const struct sysfs_ops rdev_sysfs_ops = {
3019 .show = rdev_attr_show,
3020 .store = rdev_attr_store,
3022 static struct kobj_type rdev_ktype = {
3023 .release = rdev_free,
3024 .sysfs_ops = &rdev_sysfs_ops,
3025 .default_attrs = rdev_default_attrs,
3028 int md_rdev_init(mdk_rdev_t *rdev)
3031 rdev->saved_raid_disk = -1;
3032 rdev->raid_disk = -1;
3034 rdev->data_offset = 0;
3035 rdev->sb_events = 0;
3036 rdev->last_read_error.tv_sec = 0;
3037 rdev->last_read_error.tv_nsec = 0;
3038 rdev->sb_loaded = 0;
3039 rdev->bb_page = NULL;
3040 atomic_set(&rdev->nr_pending, 0);
3041 atomic_set(&rdev->read_errors, 0);
3042 atomic_set(&rdev->corrected_errors, 0);
3044 INIT_LIST_HEAD(&rdev->same_set);
3045 init_waitqueue_head(&rdev->blocked_wait);
3047 /* Add space to store bad block list.
3048 * This reserves the space even on arrays where it cannot
3049 * be used - I wonder if that matters
3051 rdev->badblocks.count = 0;
3052 rdev->badblocks.shift = 0;
3053 rdev->badblocks.page = kmalloc(PAGE_SIZE, GFP_KERNEL);
3054 seqlock_init(&rdev->badblocks.lock);
3055 if (rdev->badblocks.page == NULL)
3060 EXPORT_SYMBOL_GPL(md_rdev_init);
3062 * Import a device. If 'super_format' >= 0, then sanity check the superblock
3064 * mark the device faulty if:
3066 * - the device is nonexistent (zero size)
3067 * - the device has no valid superblock
3069 * a faulty rdev _never_ has rdev->sb set.
3071 static mdk_rdev_t *md_import_device(dev_t newdev, int super_format, int super_minor)
3073 char b[BDEVNAME_SIZE];
3078 rdev = kzalloc(sizeof(*rdev), GFP_KERNEL);
3080 printk(KERN_ERR "md: could not alloc mem for new device!\n");
3081 return ERR_PTR(-ENOMEM);
3084 err = md_rdev_init(rdev);
3087 err = alloc_disk_sb(rdev);
3091 err = lock_rdev(rdev, newdev, super_format == -2);
3095 kobject_init(&rdev->kobj, &rdev_ktype);
3097 size = i_size_read(rdev->bdev->bd_inode) >> BLOCK_SIZE_BITS;
3100 "md: %s has zero or unknown size, marking faulty!\n",
3101 bdevname(rdev->bdev,b));
3106 if (super_format >= 0) {
3107 err = super_types[super_format].
3108 load_super(rdev, NULL, super_minor);
3109 if (err == -EINVAL) {
3111 "md: %s does not have a valid v%d.%d "
3112 "superblock, not importing!\n",
3113 bdevname(rdev->bdev,b),
3114 super_format, super_minor);
3119 "md: could not read %s's sb, not importing!\n",
3120 bdevname(rdev->bdev,b));
3124 if (super_format == -1)
3125 /* hot-add for 0.90, or non-persistent: so no badblocks */
3126 rdev->badblocks.shift = -1;
3134 kfree(rdev->badblocks.page);
3136 return ERR_PTR(err);
3140 * Check a full RAID array for plausibility
3144 static void analyze_sbs(mddev_t * mddev)
3147 mdk_rdev_t *rdev, *freshest, *tmp;
3148 char b[BDEVNAME_SIZE];
3151 rdev_for_each(rdev, tmp, mddev)
3152 switch (super_types[mddev->major_version].
3153 load_super(rdev, freshest, mddev->minor_version)) {
3161 "md: fatal superblock inconsistency in %s"
3162 " -- removing from array\n",
3163 bdevname(rdev->bdev,b));
3164 kick_rdev_from_array(rdev);
3168 super_types[mddev->major_version].
3169 validate_super(mddev, freshest);
3172 rdev_for_each(rdev, tmp, mddev) {
3173 if (mddev->max_disks &&
3174 (rdev->desc_nr >= mddev->max_disks ||
3175 i > mddev->max_disks)) {
3177 "md: %s: %s: only %d devices permitted\n",
3178 mdname(mddev), bdevname(rdev->bdev, b),
3180 kick_rdev_from_array(rdev);
3183 if (rdev != freshest)
3184 if (super_types[mddev->major_version].
3185 validate_super(mddev, rdev)) {
3186 printk(KERN_WARNING "md: kicking non-fresh %s"
3188 bdevname(rdev->bdev,b));
3189 kick_rdev_from_array(rdev);
3192 if (mddev->level == LEVEL_MULTIPATH) {
3193 rdev->desc_nr = i++;
3194 rdev->raid_disk = rdev->desc_nr;
3195 set_bit(In_sync, &rdev->flags);
3196 } else if (rdev->raid_disk >= (mddev->raid_disks - min(0, mddev->delta_disks))) {
3197 rdev->raid_disk = -1;
3198 clear_bit(In_sync, &rdev->flags);
3203 /* Read a fixed-point number.
3204 * Numbers in sysfs attributes should be in "standard" units where
3205 * possible, so time should be in seconds.
3206 * However we internally use a a much smaller unit such as
3207 * milliseconds or jiffies.
3208 * This function takes a decimal number with a possible fractional
3209 * component, and produces an integer which is the result of
3210 * multiplying that number by 10^'scale'.
3211 * all without any floating-point arithmetic.
3213 int strict_strtoul_scaled(const char *cp, unsigned long *res, int scale)
3215 unsigned long result = 0;
3217 while (isdigit(*cp) || (*cp == '.' && decimals < 0)) {
3220 else if (decimals < scale) {
3223 result = result * 10 + value;
3235 while (decimals < scale) {
3244 static void md_safemode_timeout(unsigned long data);
3247 safe_delay_show(mddev_t *mddev, char *page)
3249 int msec = (mddev->safemode_delay*1000)/HZ;
3250 return sprintf(page, "%d.%03d\n", msec/1000, msec%1000);
3253 safe_delay_store(mddev_t *mddev, const char *cbuf, size_t len)
3257 if (strict_strtoul_scaled(cbuf, &msec, 3) < 0)
3260 mddev->safemode_delay = 0;
3262 unsigned long old_delay = mddev->safemode_delay;
3263 mddev->safemode_delay = (msec*HZ)/1000;
3264 if (mddev->safemode_delay == 0)
3265 mddev->safemode_delay = 1;
3266 if (mddev->safemode_delay < old_delay)
3267 md_safemode_timeout((unsigned long)mddev);
3271 static struct md_sysfs_entry md_safe_delay =
3272 __ATTR(safe_mode_delay, S_IRUGO|S_IWUSR,safe_delay_show, safe_delay_store);
3275 level_show(mddev_t *mddev, char *page)
3277 struct mdk_personality *p = mddev->pers;
3279 return sprintf(page, "%s\n", p->name);
3280 else if (mddev->clevel[0])
3281 return sprintf(page, "%s\n", mddev->clevel);
3282 else if (mddev->level != LEVEL_NONE)
3283 return sprintf(page, "%d\n", mddev->level);
3289 level_store(mddev_t *mddev, const char *buf, size_t len)
3293 struct mdk_personality *pers;
3298 if (mddev->pers == NULL) {
3301 if (len >= sizeof(mddev->clevel))
3303 strncpy(mddev->clevel, buf, len);
3304 if (mddev->clevel[len-1] == '\n')
3306 mddev->clevel[len] = 0;
3307 mddev->level = LEVEL_NONE;
3311 /* request to change the personality. Need to ensure:
3312 * - array is not engaged in resync/recovery/reshape
3313 * - old personality can be suspended
3314 * - new personality will access other array.
3317 if (mddev->sync_thread ||
3318 mddev->reshape_position != MaxSector ||
3319 mddev->sysfs_active)
3322 if (!mddev->pers->quiesce) {
3323 printk(KERN_WARNING "md: %s: %s does not support online personality change\n",
3324 mdname(mddev), mddev->pers->name);
3328 /* Now find the new personality */
3329 if (len == 0 || len >= sizeof(clevel))
3331 strncpy(clevel, buf, len);
3332 if (clevel[len-1] == '\n')
3335 if (strict_strtol(clevel, 10, &level))
3338 if (request_module("md-%s", clevel) != 0)
3339 request_module("md-level-%s", clevel);
3340 spin_lock(&pers_lock);
3341 pers = find_pers(level, clevel);
3342 if (!pers || !try_module_get(pers->owner)) {
3343 spin_unlock(&pers_lock);
3344 printk(KERN_WARNING "md: personality %s not loaded\n", clevel);
3347 spin_unlock(&pers_lock);
3349 if (pers == mddev->pers) {
3350 /* Nothing to do! */
3351 module_put(pers->owner);
3354 if (!pers->takeover) {
3355 module_put(pers->owner);
3356 printk(KERN_WARNING "md: %s: %s does not support personality takeover\n",
3357 mdname(mddev), clevel);
3361 list_for_each_entry(rdev, &mddev->disks, same_set)
3362 rdev->new_raid_disk = rdev->raid_disk;
3364 /* ->takeover must set new_* and/or delta_disks
3365 * if it succeeds, and may set them when it fails.
3367 priv = pers->takeover(mddev);
3369 mddev->new_level = mddev->level;
3370 mddev->new_layout = mddev->layout;
3371 mddev->new_chunk_sectors = mddev->chunk_sectors;
3372 mddev->raid_disks -= mddev->delta_disks;
3373 mddev->delta_disks = 0;
3374 module_put(pers->owner);
3375 printk(KERN_WARNING "md: %s: %s would not accept array\n",
3376 mdname(mddev), clevel);
3377 return PTR_ERR(priv);
3380 /* Looks like we have a winner */
3381 mddev_suspend(mddev);
3382 mddev->pers->stop(mddev);
3384 if (mddev->pers->sync_request == NULL &&
3385 pers->sync_request != NULL) {
3386 /* need to add the md_redundancy_group */
3387 if (sysfs_create_group(&mddev->kobj, &md_redundancy_group))
3389 "md: cannot register extra attributes for %s\n",
3391 mddev->sysfs_action = sysfs_get_dirent(mddev->kobj.sd, NULL, "sync_action");
3393 if (mddev->pers->sync_request != NULL &&
3394 pers->sync_request == NULL) {
3395 /* need to remove the md_redundancy_group */
3396 if (mddev->to_remove == NULL)
3397 mddev->to_remove = &md_redundancy_group;
3400 if (mddev->pers->sync_request == NULL &&
3402 /* We are converting from a no-redundancy array
3403 * to a redundancy array and metadata is managed
3404 * externally so we need to be sure that writes
3405 * won't block due to a need to transition
3407 * until external management is started.
3410 mddev->safemode_delay = 0;
3411 mddev->safemode = 0;
3414 list_for_each_entry(rdev, &mddev->disks, same_set) {
3415 if (rdev->raid_disk < 0)
3417 if (rdev->new_raid_disk >= mddev->raid_disks)
3418 rdev->new_raid_disk = -1;
3419 if (rdev->new_raid_disk == rdev->raid_disk)
3421 sysfs_unlink_rdev(mddev, rdev);
3423 list_for_each_entry(rdev, &mddev->disks, same_set) {
3424 if (rdev->raid_disk < 0)
3426 if (rdev->new_raid_disk == rdev->raid_disk)
3428 rdev->raid_disk = rdev->new_raid_disk;
3429 if (rdev->raid_disk < 0)
3430 clear_bit(In_sync, &rdev->flags);
3432 if (sysfs_link_rdev(mddev, rdev))
3433 printk(KERN_WARNING "md: cannot register rd%d"
3434 " for %s after level change\n",
3435 rdev->raid_disk, mdname(mddev));
3439 module_put(mddev->pers->owner);
3441 mddev->private = priv;
3442 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
3443 mddev->level = mddev->new_level;
3444 mddev->layout = mddev->new_layout;
3445 mddev->chunk_sectors = mddev->new_chunk_sectors;
3446 mddev->delta_disks = 0;
3447 mddev->degraded = 0;
3448 if (mddev->pers->sync_request == NULL) {
3449 /* this is now an array without redundancy, so
3450 * it must always be in_sync
3453 del_timer_sync(&mddev->safemode_timer);
3456 mddev_resume(mddev);
3457 set_bit(MD_CHANGE_DEVS, &mddev->flags);
3458 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3459 md_wakeup_thread(mddev->thread);
3460 sysfs_notify(&mddev->kobj, NULL, "level");
3461 md_new_event(mddev);
3465 static struct md_sysfs_entry md_level =
3466 __ATTR(level, S_IRUGO|S_IWUSR, level_show, level_store);
3470 layout_show(mddev_t *mddev, char *page)
3472 /* just a number, not meaningful for all levels */
3473 if (mddev->reshape_position != MaxSector &&
3474 mddev->layout != mddev->new_layout)
3475 return sprintf(page, "%d (%d)\n",
3476 mddev->new_layout, mddev->layout);
3477 return sprintf(page, "%d\n", mddev->layout);
3481 layout_store(mddev_t *mddev, const char *buf, size_t len)
3484 unsigned long n = simple_strtoul(buf, &e, 10);
3486 if (!*buf || (*e && *e != '\n'))
3491 if (mddev->pers->check_reshape == NULL)
3493 mddev->new_layout = n;
3494 err = mddev->pers->check_reshape(mddev);
3496 mddev->new_layout = mddev->layout;
3500 mddev->new_layout = n;
3501 if (mddev->reshape_position == MaxSector)
3506 static struct md_sysfs_entry md_layout =
3507 __ATTR(layout, S_IRUGO|S_IWUSR, layout_show, layout_store);
3511 raid_disks_show(mddev_t *mddev, char *page)
3513 if (mddev->raid_disks == 0)
3515 if (mddev->reshape_position != MaxSector &&
3516 mddev->delta_disks != 0)
3517 return sprintf(page, "%d (%d)\n", mddev->raid_disks,
3518 mddev->raid_disks - mddev->delta_disks);
3519 return sprintf(page, "%d\n", mddev->raid_disks);
3522 static int update_raid_disks(mddev_t *mddev, int raid_disks);
3525 raid_disks_store(mddev_t *mddev, const char *buf, size_t len)
3529 unsigned long n = simple_strtoul(buf, &e, 10);
3531 if (!*buf || (*e && *e != '\n'))
3535 rv = update_raid_disks(mddev, n);
3536 else if (mddev->reshape_position != MaxSector) {
3537 int olddisks = mddev->raid_disks - mddev->delta_disks;
3538 mddev->delta_disks = n - olddisks;
3539 mddev->raid_disks = n;
3541 mddev->raid_disks = n;
3542 return rv ? rv : len;
3544 static struct md_sysfs_entry md_raid_disks =
3545 __ATTR(raid_disks, S_IRUGO|S_IWUSR, raid_disks_show, raid_disks_store);
3548 chunk_size_show(mddev_t *mddev, char *page)
3550 if (mddev->reshape_position != MaxSector &&
3551 mddev->chunk_sectors != mddev->new_chunk_sectors)
3552 return sprintf(page, "%d (%d)\n",
3553 mddev->new_chunk_sectors << 9,
3554 mddev->chunk_sectors << 9);
3555 return sprintf(page, "%d\n", mddev->chunk_sectors << 9);
3559 chunk_size_store(mddev_t *mddev, const char *buf, size_t len)
3562 unsigned long n = simple_strtoul(buf, &e, 10);
3564 if (!*buf || (*e && *e != '\n'))
3569 if (mddev->pers->check_reshape == NULL)
3571 mddev->new_chunk_sectors = n >> 9;
3572 err = mddev->pers->check_reshape(mddev);
3574 mddev->new_chunk_sectors = mddev->chunk_sectors;
3578 mddev->new_chunk_sectors = n >> 9;
3579 if (mddev->reshape_position == MaxSector)
3580 mddev->chunk_sectors = n >> 9;
3584 static struct md_sysfs_entry md_chunk_size =
3585 __ATTR(chunk_size, S_IRUGO|S_IWUSR, chunk_size_show, chunk_size_store);
3588 resync_start_show(mddev_t *mddev, char *page)
3590 if (mddev->recovery_cp == MaxSector)
3591 return sprintf(page, "none\n");
3592 return sprintf(page, "%llu\n", (unsigned long long)mddev->recovery_cp);
3596 resync_start_store(mddev_t *mddev, const char *buf, size_t len)
3599 unsigned long long n = simple_strtoull(buf, &e, 10);
3601 if (mddev->pers && !test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
3603 if (cmd_match(buf, "none"))
3605 else if (!*buf || (*e && *e != '\n'))
3608 mddev->recovery_cp = n;
3611 static struct md_sysfs_entry md_resync_start =
3612 __ATTR(resync_start, S_IRUGO|S_IWUSR, resync_start_show, resync_start_store);
3615 * The array state can be:
3618 * No devices, no size, no level
3619 * Equivalent to STOP_ARRAY ioctl
3621 * May have some settings, but array is not active
3622 * all IO results in error
3623 * When written, doesn't tear down array, but just stops it
3624 * suspended (not supported yet)
3625 * All IO requests will block. The array can be reconfigured.
3626 * Writing this, if accepted, will block until array is quiescent
3628 * no resync can happen. no superblocks get written.
3629 * write requests fail
3631 * like readonly, but behaves like 'clean' on a write request.
3633 * clean - no pending writes, but otherwise active.
3634 * When written to inactive array, starts without resync
3635 * If a write request arrives then
3636 * if metadata is known, mark 'dirty' and switch to 'active'.
3637 * if not known, block and switch to write-pending
3638 * If written to an active array that has pending writes, then fails.
3640 * fully active: IO and resync can be happening.
3641 * When written to inactive array, starts with resync
3644 * clean, but writes are blocked waiting for 'active' to be written.
3647 * like active, but no writes have been seen for a while (100msec).
3650 enum array_state { clear, inactive, suspended, readonly, read_auto, clean, active,
3651 write_pending, active_idle, bad_word};
3652 static char *array_states[] = {
3653 "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active",
3654 "write-pending", "active-idle", NULL };
3656 static int match_word(const char *word, char **list)
3659 for (n=0; list[n]; n++)
3660 if (cmd_match(word, list[n]))
3666 array_state_show(mddev_t *mddev, char *page)
3668 enum array_state st = inactive;
3681 else if (test_bit(MD_CHANGE_PENDING, &mddev->flags))
3683 else if (mddev->safemode)
3689 if (list_empty(&mddev->disks) &&
3690 mddev->raid_disks == 0 &&
3691 mddev->dev_sectors == 0)
3696 return sprintf(page, "%s\n", array_states[st]);
3699 static int do_md_stop(mddev_t * mddev, int ro, int is_open);
3700 static int md_set_readonly(mddev_t * mddev, int is_open);
3701 static int do_md_run(mddev_t * mddev);
3702 static int restart_array(mddev_t *mddev);
3705 array_state_store(mddev_t *mddev, const char *buf, size_t len)
3708 enum array_state st = match_word(buf, array_states);
3713 /* stopping an active array */
3714 if (atomic_read(&mddev->openers) > 0)
3716 err = do_md_stop(mddev, 0, 0);
3719 /* stopping an active array */
3721 if (atomic_read(&mddev->openers) > 0)
3723 err = do_md_stop(mddev, 2, 0);
3725 err = 0; /* already inactive */
3728 break; /* not supported yet */
3731 err = md_set_readonly(mddev, 0);
3734 set_disk_ro(mddev->gendisk, 1);
3735 err = do_md_run(mddev);
3741 err = md_set_readonly(mddev, 0);
3742 else if (mddev->ro == 1)
3743 err = restart_array(mddev);
3746 set_disk_ro(mddev->gendisk, 0);
3750 err = do_md_run(mddev);
3755 restart_array(mddev);
3756 spin_lock_irq(&mddev->write_lock);
3757 if (atomic_read(&mddev->writes_pending) == 0) {
3758 if (mddev->in_sync == 0) {
3760 if (mddev->safemode == 1)
3761 mddev->safemode = 0;
3762 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
3767 spin_unlock_irq(&mddev->write_lock);
3773 restart_array(mddev);
3774 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
3775 wake_up(&mddev->sb_wait);
3779 set_disk_ro(mddev->gendisk, 0);
3780 err = do_md_run(mddev);
3785 /* these cannot be set */
3791 sysfs_notify_dirent_safe(mddev->sysfs_state);
3795 static struct md_sysfs_entry md_array_state =
3796 __ATTR(array_state, S_IRUGO|S_IWUSR, array_state_show, array_state_store);
3799 max_corrected_read_errors_show(mddev_t *mddev, char *page) {
3800 return sprintf(page, "%d\n",
3801 atomic_read(&mddev->max_corr_read_errors));
3805 max_corrected_read_errors_store(mddev_t *mddev, const char *buf, size_t len)
3808 unsigned long n = simple_strtoul(buf, &e, 10);
3810 if (*buf && (*e == 0 || *e == '\n')) {
3811 atomic_set(&mddev->max_corr_read_errors, n);
3817 static struct md_sysfs_entry max_corr_read_errors =
3818 __ATTR(max_read_errors, S_IRUGO|S_IWUSR, max_corrected_read_errors_show,
3819 max_corrected_read_errors_store);
3822 null_show(mddev_t *mddev, char *page)
3828 new_dev_store(mddev_t *mddev, const char *buf, size_t len)
3830 /* buf must be %d:%d\n? giving major and minor numbers */
3831 /* The new device is added to the array.
3832 * If the array has a persistent superblock, we read the
3833 * superblock to initialise info and check validity.
3834 * Otherwise, only checking done is that in bind_rdev_to_array,
3835 * which mainly checks size.
3838 int major = simple_strtoul(buf, &e, 10);
3844 if (!*buf || *e != ':' || !e[1] || e[1] == '\n')
3846 minor = simple_strtoul(e+1, &e, 10);
3847 if (*e && *e != '\n')
3849 dev = MKDEV(major, minor);
3850 if (major != MAJOR(dev) ||
3851 minor != MINOR(dev))
3855 if (mddev->persistent) {
3856 rdev = md_import_device(dev, mddev->major_version,
3857 mddev->minor_version);
3858 if (!IS_ERR(rdev) && !list_empty(&mddev->disks)) {
3859 mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
3860 mdk_rdev_t, same_set);
3861 err = super_types[mddev->major_version]
3862 .load_super(rdev, rdev0, mddev->minor_version);
3866 } else if (mddev->external)
3867 rdev = md_import_device(dev, -2, -1);
3869 rdev = md_import_device(dev, -1, -1);
3872 return PTR_ERR(rdev);
3873 err = bind_rdev_to_array(rdev, mddev);
3877 return err ? err : len;
3880 static struct md_sysfs_entry md_new_device =
3881 __ATTR(new_dev, S_IWUSR, null_show, new_dev_store);
3884 bitmap_store(mddev_t *mddev, const char *buf, size_t len)
3887 unsigned long chunk, end_chunk;
3891 /* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */
3893 chunk = end_chunk = simple_strtoul(buf, &end, 0);
3894 if (buf == end) break;
3895 if (*end == '-') { /* range */
3897 end_chunk = simple_strtoul(buf, &end, 0);
3898 if (buf == end) break;
3900 if (*end && !isspace(*end)) break;
3901 bitmap_dirty_bits(mddev->bitmap, chunk, end_chunk);
3902 buf = skip_spaces(end);
3904 bitmap_unplug(mddev->bitmap); /* flush the bits to disk */
3909 static struct md_sysfs_entry md_bitmap =
3910 __ATTR(bitmap_set_bits, S_IWUSR, null_show, bitmap_store);
3913 size_show(mddev_t *mddev, char *page)
3915 return sprintf(page, "%llu\n",
3916 (unsigned long long)mddev->dev_sectors / 2);
3919 static int update_size(mddev_t *mddev, sector_t num_sectors);
3922 size_store(mddev_t *mddev, const char *buf, size_t len)
3924 /* If array is inactive, we can reduce the component size, but
3925 * not increase it (except from 0).
3926 * If array is active, we can try an on-line resize
3929 int err = strict_blocks_to_sectors(buf, §ors);
3934 err = update_size(mddev, sectors);
3935 md_update_sb(mddev, 1);
3937 if (mddev->dev_sectors == 0 ||
3938 mddev->dev_sectors > sectors)
3939 mddev->dev_sectors = sectors;
3943 return err ? err : len;
3946 static struct md_sysfs_entry md_size =
3947 __ATTR(component_size, S_IRUGO|S_IWUSR, size_show, size_store);
3952 * 'none' for arrays with no metadata (good luck...)
3953 * 'external' for arrays with externally managed metadata,
3954 * or N.M for internally known formats
3957 metadata_show(mddev_t *mddev, char *page)
3959 if (mddev->persistent)
3960 return sprintf(page, "%d.%d\n",
3961 mddev->major_version, mddev->minor_version);
3962 else if (mddev->external)
3963 return sprintf(page, "external:%s\n", mddev->metadata_type);
3965 return sprintf(page, "none\n");
3969 metadata_store(mddev_t *mddev, const char *buf, size_t len)
3973 /* Changing the details of 'external' metadata is
3974 * always permitted. Otherwise there must be
3975 * no devices attached to the array.
3977 if (mddev->external && strncmp(buf, "external:", 9) == 0)
3979 else if (!list_empty(&mddev->disks))
3982 if (cmd_match(buf, "none")) {
3983 mddev->persistent = 0;
3984 mddev->external = 0;
3985 mddev->major_version = 0;
3986 mddev->minor_version = 90;
3989 if (strncmp(buf, "external:", 9) == 0) {
3990 size_t namelen = len-9;
3991 if (namelen >= sizeof(mddev->metadata_type))
3992 namelen = sizeof(mddev->metadata_type)-1;
3993 strncpy(mddev->metadata_type, buf+9, namelen);
3994 mddev->metadata_type[namelen] = 0;
3995 if (namelen && mddev->metadata_type[namelen-1] == '\n')
3996 mddev->metadata_type[--namelen] = 0;
3997 mddev->persistent = 0;
3998 mddev->external = 1;
3999 mddev->major_version = 0;
4000 mddev->minor_version = 90;
4003 major = simple_strtoul(buf, &e, 10);
4004 if (e==buf || *e != '.')
4007 minor = simple_strtoul(buf, &e, 10);
4008 if (e==buf || (*e && *e != '\n') )
4010 if (major >= ARRAY_SIZE(super_types) || super_types[major].name == NULL)
4012 mddev->major_version = major;
4013 mddev->minor_version = minor;
4014 mddev->persistent = 1;
4015 mddev->external = 0;
4019 static struct md_sysfs_entry md_metadata =
4020 __ATTR(metadata_version, S_IRUGO|S_IWUSR, metadata_show, metadata_store);
4023 action_show(mddev_t *mddev, char *page)
4025 char *type = "idle";
4026 if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
4028 else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
4029 (!mddev->ro && test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))) {
4030 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
4032 else if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
4033 if (!test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
4035 else if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
4039 } else if (test_bit(MD_RECOVERY_RECOVER, &mddev->recovery))
4042 return sprintf(page, "%s\n", type);
4045 static void reap_sync_thread(mddev_t *mddev);
4048 action_store(mddev_t *mddev, const char *page, size_t len)
4050 if (!mddev->pers || !mddev->pers->sync_request)
4053 if (cmd_match(page, "frozen"))
4054 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4056 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4058 if (cmd_match(page, "idle") || cmd_match(page, "frozen")) {
4059 if (mddev->sync_thread) {
4060 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
4061 reap_sync_thread(mddev);
4063 } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
4064 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
4066 else if (cmd_match(page, "resync"))
4067 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4068 else if (cmd_match(page, "recover")) {
4069 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
4070 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4071 } else if (cmd_match(page, "reshape")) {
4073 if (mddev->pers->start_reshape == NULL)
4075 err = mddev->pers->start_reshape(mddev);
4078 sysfs_notify(&mddev->kobj, NULL, "degraded");
4080 if (cmd_match(page, "check"))
4081 set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
4082 else if (!cmd_match(page, "repair"))
4084 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
4085 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
4087 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4088 md_wakeup_thread(mddev->thread);
4089 sysfs_notify_dirent_safe(mddev->sysfs_action);
4094 mismatch_cnt_show(mddev_t *mddev, char *page)
4096 return sprintf(page, "%llu\n",
4097 (unsigned long long) mddev->resync_mismatches);
4100 static struct md_sysfs_entry md_scan_mode =
4101 __ATTR(sync_action, S_IRUGO|S_IWUSR, action_show, action_store);
4104 static struct md_sysfs_entry md_mismatches = __ATTR_RO(mismatch_cnt);
4107 sync_min_show(mddev_t *mddev, char *page)
4109 return sprintf(page, "%d (%s)\n", speed_min(mddev),
4110 mddev->sync_speed_min ? "local": "system");
4114 sync_min_store(mddev_t *mddev, const char *buf, size_t len)
4118 if (strncmp(buf, "system", 6)==0) {
4119 mddev->sync_speed_min = 0;
4122 min = simple_strtoul(buf, &e, 10);
4123 if (buf == e || (*e && *e != '\n') || min <= 0)
4125 mddev->sync_speed_min = min;
4129 static struct md_sysfs_entry md_sync_min =
4130 __ATTR(sync_speed_min, S_IRUGO|S_IWUSR, sync_min_show, sync_min_store);
4133 sync_max_show(mddev_t *mddev, char *page)
4135 return sprintf(page, "%d (%s)\n", speed_max(mddev),
4136 mddev->sync_speed_max ? "local": "system");
4140 sync_max_store(mddev_t *mddev, const char *buf, size_t len)
4144 if (strncmp(buf, "system", 6)==0) {
4145 mddev->sync_speed_max = 0;
4148 max = simple_strtoul(buf, &e, 10);
4149 if (buf == e || (*e && *e != '\n') || max <= 0)
4151 mddev->sync_speed_max = max;
4155 static struct md_sysfs_entry md_sync_max =
4156 __ATTR(sync_speed_max, S_IRUGO|S_IWUSR, sync_max_show, sync_max_store);
4159 degraded_show(mddev_t *mddev, char *page)
4161 return sprintf(page, "%d\n", mddev->degraded);
4163 static struct md_sysfs_entry md_degraded = __ATTR_RO(degraded);
4166 sync_force_parallel_show(mddev_t *mddev, char *page)
4168 return sprintf(page, "%d\n", mddev->parallel_resync);
4172 sync_force_parallel_store(mddev_t *mddev, const char *buf, size_t len)
4176 if (strict_strtol(buf, 10, &n))
4179 if (n != 0 && n != 1)
4182 mddev->parallel_resync = n;
4184 if (mddev->sync_thread)
4185 wake_up(&resync_wait);
4190 /* force parallel resync, even with shared block devices */
4191 static struct md_sysfs_entry md_sync_force_parallel =
4192 __ATTR(sync_force_parallel, S_IRUGO|S_IWUSR,
4193 sync_force_parallel_show, sync_force_parallel_store);
4196 sync_speed_show(mddev_t *mddev, char *page)
4198 unsigned long resync, dt, db;
4199 if (mddev->curr_resync == 0)
4200 return sprintf(page, "none\n");
4201 resync = mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active);
4202 dt = (jiffies - mddev->resync_mark) / HZ;
4204 db = resync - mddev->resync_mark_cnt;
4205 return sprintf(page, "%lu\n", db/dt/2); /* K/sec */
4208 static struct md_sysfs_entry md_sync_speed = __ATTR_RO(sync_speed);
4211 sync_completed_show(mddev_t *mddev, char *page)
4213 unsigned long long max_sectors, resync;
4215 if (!test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4216 return sprintf(page, "none\n");
4218 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
4219 max_sectors = mddev->resync_max_sectors;
4221 max_sectors = mddev->dev_sectors;
4223 resync = mddev->curr_resync_completed;
4224 return sprintf(page, "%llu / %llu\n", resync, max_sectors);
4227 static struct md_sysfs_entry md_sync_completed = __ATTR_RO(sync_completed);
4230 min_sync_show(mddev_t *mddev, char *page)
4232 return sprintf(page, "%llu\n",
4233 (unsigned long long)mddev->resync_min);
4236 min_sync_store(mddev_t *mddev, const char *buf, size_t len)
4238 unsigned long long min;
4239 if (strict_strtoull(buf, 10, &min))
4241 if (min > mddev->resync_max)
4243 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4246 /* Must be a multiple of chunk_size */
4247 if (mddev->chunk_sectors) {
4248 sector_t temp = min;
4249 if (sector_div(temp, mddev->chunk_sectors))
4252 mddev->resync_min = min;
4257 static struct md_sysfs_entry md_min_sync =
4258 __ATTR(sync_min, S_IRUGO|S_IWUSR, min_sync_show, min_sync_store);
4261 max_sync_show(mddev_t *mddev, char *page)
4263 if (mddev->resync_max == MaxSector)
4264 return sprintf(page, "max\n");
4266 return sprintf(page, "%llu\n",
4267 (unsigned long long)mddev->resync_max);
4270 max_sync_store(mddev_t *mddev, const char *buf, size_t len)
4272 if (strncmp(buf, "max", 3) == 0)
4273 mddev->resync_max = MaxSector;
4275 unsigned long long max;
4276 if (strict_strtoull(buf, 10, &max))
4278 if (max < mddev->resync_min)
4280 if (max < mddev->resync_max &&
4282 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4285 /* Must be a multiple of chunk_size */
4286 if (mddev->chunk_sectors) {
4287 sector_t temp = max;
4288 if (sector_div(temp, mddev->chunk_sectors))
4291 mddev->resync_max = max;
4293 wake_up(&mddev->recovery_wait);
4297 static struct md_sysfs_entry md_max_sync =
4298 __ATTR(sync_max, S_IRUGO|S_IWUSR, max_sync_show, max_sync_store);
4301 suspend_lo_show(mddev_t *mddev, char *page)
4303 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_lo);
4307 suspend_lo_store(mddev_t *mddev, const char *buf, size_t len)
4310 unsigned long long new = simple_strtoull(buf, &e, 10);
4311 unsigned long long old = mddev->suspend_lo;
4313 if (mddev->pers == NULL ||
4314 mddev->pers->quiesce == NULL)
4316 if (buf == e || (*e && *e != '\n'))
4319 mddev->suspend_lo = new;
4321 /* Shrinking suspended region */
4322 mddev->pers->quiesce(mddev, 2);
4324 /* Expanding suspended region - need to wait */
4325 mddev->pers->quiesce(mddev, 1);
4326 mddev->pers->quiesce(mddev, 0);
4330 static struct md_sysfs_entry md_suspend_lo =
4331 __ATTR(suspend_lo, S_IRUGO|S_IWUSR, suspend_lo_show, suspend_lo_store);
4335 suspend_hi_show(mddev_t *mddev, char *page)
4337 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_hi);
4341 suspend_hi_store(mddev_t *mddev, const char *buf, size_t len)
4344 unsigned long long new = simple_strtoull(buf, &e, 10);
4345 unsigned long long old = mddev->suspend_hi;
4347 if (mddev->pers == NULL ||
4348 mddev->pers->quiesce == NULL)
4350 if (buf == e || (*e && *e != '\n'))
4353 mddev->suspend_hi = new;
4355 /* Shrinking suspended region */
4356 mddev->pers->quiesce(mddev, 2);
4358 /* Expanding suspended region - need to wait */
4359 mddev->pers->quiesce(mddev, 1);
4360 mddev->pers->quiesce(mddev, 0);
4364 static struct md_sysfs_entry md_suspend_hi =
4365 __ATTR(suspend_hi, S_IRUGO|S_IWUSR, suspend_hi_show, suspend_hi_store);
4368 reshape_position_show(mddev_t *mddev, char *page)
4370 if (mddev->reshape_position != MaxSector)
4371 return sprintf(page, "%llu\n",
4372 (unsigned long long)mddev->reshape_position);
4373 strcpy(page, "none\n");
4378 reshape_position_store(mddev_t *mddev, const char *buf, size_t len)
4381 unsigned long long new = simple_strtoull(buf, &e, 10);
4384 if (buf == e || (*e && *e != '\n'))
4386 mddev->reshape_position = new;
4387 mddev->delta_disks = 0;
4388 mddev->new_level = mddev->level;
4389 mddev->new_layout = mddev->layout;
4390 mddev->new_chunk_sectors = mddev->chunk_sectors;
4394 static struct md_sysfs_entry md_reshape_position =
4395 __ATTR(reshape_position, S_IRUGO|S_IWUSR, reshape_position_show,
4396 reshape_position_store);
4399 array_size_show(mddev_t *mddev, char *page)
4401 if (mddev->external_size)
4402 return sprintf(page, "%llu\n",
4403 (unsigned long long)mddev->array_sectors/2);
4405 return sprintf(page, "default\n");
4409 array_size_store(mddev_t *mddev, const char *buf, size_t len)
4413 if (strncmp(buf, "default", 7) == 0) {
4415 sectors = mddev->pers->size(mddev, 0, 0);
4417 sectors = mddev->array_sectors;
4419 mddev->external_size = 0;
4421 if (strict_blocks_to_sectors(buf, §ors) < 0)
4423 if (mddev->pers && mddev->pers->size(mddev, 0, 0) < sectors)
4426 mddev->external_size = 1;
4429 mddev->array_sectors = sectors;
4431 set_capacity(mddev->gendisk, mddev->array_sectors);
4432 revalidate_disk(mddev->gendisk);
4437 static struct md_sysfs_entry md_array_size =
4438 __ATTR(array_size, S_IRUGO|S_IWUSR, array_size_show,
4441 static struct attribute *md_default_attrs[] = {
4444 &md_raid_disks.attr,
4445 &md_chunk_size.attr,
4447 &md_resync_start.attr,
4449 &md_new_device.attr,
4450 &md_safe_delay.attr,
4451 &md_array_state.attr,
4452 &md_reshape_position.attr,
4453 &md_array_size.attr,
4454 &max_corr_read_errors.attr,
4458 static struct attribute *md_redundancy_attrs[] = {
4460 &md_mismatches.attr,
4463 &md_sync_speed.attr,
4464 &md_sync_force_parallel.attr,
4465 &md_sync_completed.attr,
4468 &md_suspend_lo.attr,
4469 &md_suspend_hi.attr,
4474 static struct attribute_group md_redundancy_group = {
4476 .attrs = md_redundancy_attrs,
4481 md_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
4483 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
4484 mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
4489 rv = mddev_lock(mddev);
4491 rv = entry->show(mddev, page);
4492 mddev_unlock(mddev);
4498 md_attr_store(struct kobject *kobj, struct attribute *attr,
4499 const char *page, size_t length)
4501 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
4502 mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
4507 if (!capable(CAP_SYS_ADMIN))
4509 rv = mddev_lock(mddev);
4510 if (mddev->hold_active == UNTIL_IOCTL)
4511 mddev->hold_active = 0;
4513 rv = entry->store(mddev, page, length);
4514 mddev_unlock(mddev);
4519 static void md_free(struct kobject *ko)
4521 mddev_t *mddev = container_of(ko, mddev_t, kobj);
4523 if (mddev->sysfs_state)
4524 sysfs_put(mddev->sysfs_state);
4526 if (mddev->gendisk) {
4527 del_gendisk(mddev->gendisk);
4528 put_disk(mddev->gendisk);
4531 blk_cleanup_queue(mddev->queue);
4536 static const struct sysfs_ops md_sysfs_ops = {
4537 .show = md_attr_show,
4538 .store = md_attr_store,
4540 static struct kobj_type md_ktype = {
4542 .sysfs_ops = &md_sysfs_ops,
4543 .default_attrs = md_default_attrs,
4548 static void mddev_delayed_delete(struct work_struct *ws)
4550 mddev_t *mddev = container_of(ws, mddev_t, del_work);
4552 sysfs_remove_group(&mddev->kobj, &md_bitmap_group);
4553 kobject_del(&mddev->kobj);
4554 kobject_put(&mddev->kobj);
4557 static int md_alloc(dev_t dev, char *name)
4559 static DEFINE_MUTEX(disks_mutex);
4560 mddev_t *mddev = mddev_find(dev);
4561 struct gendisk *disk;
4570 partitioned = (MAJOR(mddev->unit) != MD_MAJOR);
4571 shift = partitioned ? MdpMinorShift : 0;
4572 unit = MINOR(mddev->unit) >> shift;
4574 /* wait for any previous instance of this device to be
4575 * completely removed (mddev_delayed_delete).
4577 flush_workqueue(md_misc_wq);
4579 mutex_lock(&disks_mutex);
4585 /* Need to ensure that 'name' is not a duplicate.
4588 spin_lock(&all_mddevs_lock);
4590 list_for_each_entry(mddev2, &all_mddevs, all_mddevs)
4591 if (mddev2->gendisk &&
4592 strcmp(mddev2->gendisk->disk_name, name) == 0) {
4593 spin_unlock(&all_mddevs_lock);
4596 spin_unlock(&all_mddevs_lock);
4600 mddev->queue = blk_alloc_queue(GFP_KERNEL);
4603 mddev->queue->queuedata = mddev;
4605 blk_queue_make_request(mddev->queue, md_make_request);
4607 disk = alloc_disk(1 << shift);
4609 blk_cleanup_queue(mddev->queue);
4610 mddev->queue = NULL;
4613 disk->major = MAJOR(mddev->unit);
4614 disk->first_minor = unit << shift;
4616 strcpy(disk->disk_name, name);
4617 else if (partitioned)
4618 sprintf(disk->disk_name, "md_d%d", unit);
4620 sprintf(disk->disk_name, "md%d", unit);
4621 disk->fops = &md_fops;
4622 disk->private_data = mddev;
4623 disk->queue = mddev->queue;
4624 blk_queue_flush(mddev->queue, REQ_FLUSH | REQ_FUA);
4625 /* Allow extended partitions. This makes the
4626 * 'mdp' device redundant, but we can't really
4629 disk->flags |= GENHD_FL_EXT_DEVT;
4630 mddev->gendisk = disk;
4631 /* As soon as we call add_disk(), another thread could get
4632 * through to md_open, so make sure it doesn't get too far
4634 mutex_lock(&mddev->open_mutex);
4637 error = kobject_init_and_add(&mddev->kobj, &md_ktype,
4638 &disk_to_dev(disk)->kobj, "%s", "md");
4640 /* This isn't possible, but as kobject_init_and_add is marked
4641 * __must_check, we must do something with the result
4643 printk(KERN_WARNING "md: cannot register %s/md - name in use\n",
4647 if (mddev->kobj.sd &&
4648 sysfs_create_group(&mddev->kobj, &md_bitmap_group))
4649 printk(KERN_DEBUG "pointless warning\n");
4650 mutex_unlock(&mddev->open_mutex);
4652 mutex_unlock(&disks_mutex);
4653 if (!error && mddev->kobj.sd) {
4654 kobject_uevent(&mddev->kobj, KOBJ_ADD);
4655 mddev->sysfs_state = sysfs_get_dirent_safe(mddev->kobj.sd, "array_state");
4661 static struct kobject *md_probe(dev_t dev, int *part, void *data)
4663 md_alloc(dev, NULL);
4667 static int add_named_array(const char *val, struct kernel_param *kp)
4669 /* val must be "md_*" where * is not all digits.
4670 * We allocate an array with a large free minor number, and
4671 * set the name to val. val must not already be an active name.
4673 int len = strlen(val);
4674 char buf[DISK_NAME_LEN];
4676 while (len && val[len-1] == '\n')
4678 if (len >= DISK_NAME_LEN)
4680 strlcpy(buf, val, len+1);
4681 if (strncmp(buf, "md_", 3) != 0)
4683 return md_alloc(0, buf);
4686 static void md_safemode_timeout(unsigned long data)
4688 mddev_t *mddev = (mddev_t *) data;
4690 if (!atomic_read(&mddev->writes_pending)) {
4691 mddev->safemode = 1;
4692 if (mddev->external)
4693 sysfs_notify_dirent_safe(mddev->sysfs_state);
4695 md_wakeup_thread(mddev->thread);
4698 static int start_dirty_degraded;
4700 int md_run(mddev_t *mddev)
4704 struct mdk_personality *pers;
4706 if (list_empty(&mddev->disks))
4707 /* cannot run an array with no devices.. */
4712 /* Cannot run until previous stop completes properly */
4713 if (mddev->sysfs_active)
4717 * Analyze all RAID superblock(s)
4719 if (!mddev->raid_disks) {
4720 if (!mddev->persistent)
4725 if (mddev->level != LEVEL_NONE)
4726 request_module("md-level-%d", mddev->level);
4727 else if (mddev->clevel[0])
4728 request_module("md-%s", mddev->clevel);
4731 * Drop all container device buffers, from now on
4732 * the only valid external interface is through the md
4735 list_for_each_entry(rdev, &mddev->disks, same_set) {
4736 if (test_bit(Faulty, &rdev->flags))
4738 sync_blockdev(rdev->bdev);
4739 invalidate_bdev(rdev->bdev);
4741 /* perform some consistency tests on the device.
4742 * We don't want the data to overlap the metadata,
4743 * Internal Bitmap issues have been handled elsewhere.
4745 if (rdev->meta_bdev) {
4746 /* Nothing to check */;
4747 } else if (rdev->data_offset < rdev->sb_start) {
4748 if (mddev->dev_sectors &&
4749 rdev->data_offset + mddev->dev_sectors
4751 printk("md: %s: data overlaps metadata\n",
4756 if (rdev->sb_start + rdev->sb_size/512
4757 > rdev->data_offset) {
4758 printk("md: %s: metadata overlaps data\n",
4763 sysfs_notify_dirent_safe(rdev->sysfs_state);
4766 if (mddev->bio_set == NULL)
4767 mddev->bio_set = bioset_create(BIO_POOL_SIZE,
4770 spin_lock(&pers_lock);
4771 pers = find_pers(mddev->level, mddev->clevel);
4772 if (!pers || !try_module_get(pers->owner)) {
4773 spin_unlock(&pers_lock);
4774 if (mddev->level != LEVEL_NONE)
4775 printk(KERN_WARNING "md: personality for level %d is not loaded!\n",
4778 printk(KERN_WARNING "md: personality for level %s is not loaded!\n",
4783 spin_unlock(&pers_lock);
4784 if (mddev->level != pers->level) {
4785 mddev->level = pers->level;
4786 mddev->new_level = pers->level;
4788 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
4790 if (mddev->reshape_position != MaxSector &&
4791 pers->start_reshape == NULL) {
4792 /* This personality cannot handle reshaping... */
4794 module_put(pers->owner);
4798 if (pers->sync_request) {
4799 /* Warn if this is a potentially silly
4802 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
4806 list_for_each_entry(rdev, &mddev->disks, same_set)
4807 list_for_each_entry(rdev2, &mddev->disks, same_set) {
4809 rdev->bdev->bd_contains ==
4810 rdev2->bdev->bd_contains) {
4812 "%s: WARNING: %s appears to be"
4813 " on the same physical disk as"
4816 bdevname(rdev->bdev,b),
4817 bdevname(rdev2->bdev,b2));
4824 "True protection against single-disk"
4825 " failure might be compromised.\n");
4828 mddev->recovery = 0;
4829 /* may be over-ridden by personality */
4830 mddev->resync_max_sectors = mddev->dev_sectors;
4832 mddev->ok_start_degraded = start_dirty_degraded;
4834 if (start_readonly && mddev->ro == 0)
4835 mddev->ro = 2; /* read-only, but switch on first write */
4837 err = mddev->pers->run(mddev);
4839 printk(KERN_ERR "md: pers->run() failed ...\n");
4840 else if (mddev->pers->size(mddev, 0, 0) < mddev->array_sectors) {
4841 WARN_ONCE(!mddev->external_size, "%s: default size too small,"
4842 " but 'external_size' not in effect?\n", __func__);
4844 "md: invalid array_size %llu > default size %llu\n",
4845 (unsigned long long)mddev->array_sectors / 2,
4846 (unsigned long long)mddev->pers->size(mddev, 0, 0) / 2);
4848 mddev->pers->stop(mddev);
4850 if (err == 0 && mddev->pers->sync_request) {
4851 err = bitmap_create(mddev);
4853 printk(KERN_ERR "%s: failed to create bitmap (%d)\n",
4854 mdname(mddev), err);
4855 mddev->pers->stop(mddev);
4859 module_put(mddev->pers->owner);
4861 bitmap_destroy(mddev);
4864 if (mddev->pers->sync_request) {
4865 if (mddev->kobj.sd &&
4866 sysfs_create_group(&mddev->kobj, &md_redundancy_group))
4868 "md: cannot register extra attributes for %s\n",
4870 mddev->sysfs_action = sysfs_get_dirent_safe(mddev->kobj.sd, "sync_action");
4871 } else if (mddev->ro == 2) /* auto-readonly not meaningful */
4874 atomic_set(&mddev->writes_pending,0);
4875 atomic_set(&mddev->max_corr_read_errors,
4876 MD_DEFAULT_MAX_CORRECTED_READ_ERRORS);
4877 mddev->safemode = 0;
4878 mddev->safemode_timer.function = md_safemode_timeout;
4879 mddev->safemode_timer.data = (unsigned long) mddev;
4880 mddev->safemode_delay = (200 * HZ)/1000 +1; /* 200 msec delay */
4884 list_for_each_entry(rdev, &mddev->disks, same_set)
4885 if (rdev->raid_disk >= 0)
4886 if (sysfs_link_rdev(mddev, rdev))
4887 /* failure here is OK */;
4889 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4892 md_update_sb(mddev, 0);
4894 md_new_event(mddev);
4895 sysfs_notify_dirent_safe(mddev->sysfs_state);
4896 sysfs_notify_dirent_safe(mddev->sysfs_action);
4897 sysfs_notify(&mddev->kobj, NULL, "degraded");
4900 EXPORT_SYMBOL_GPL(md_run);
4902 static int do_md_run(mddev_t *mddev)
4906 err = md_run(mddev);
4909 err = bitmap_load(mddev);
4911 bitmap_destroy(mddev);
4915 md_wakeup_thread(mddev->thread);
4916 md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
4918 set_capacity(mddev->gendisk, mddev->array_sectors);
4919 revalidate_disk(mddev->gendisk);
4921 kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
4926 static int restart_array(mddev_t *mddev)
4928 struct gendisk *disk = mddev->gendisk;
4930 /* Complain if it has no devices */
4931 if (list_empty(&mddev->disks))
4937 mddev->safemode = 0;
4939 set_disk_ro(disk, 0);
4940 printk(KERN_INFO "md: %s switched to read-write mode.\n",
4942 /* Kick recovery or resync if necessary */
4943 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4944 md_wakeup_thread(mddev->thread);
4945 md_wakeup_thread(mddev->sync_thread);
4946 sysfs_notify_dirent_safe(mddev->sysfs_state);
4950 /* similar to deny_write_access, but accounts for our holding a reference
4951 * to the file ourselves */
4952 static int deny_bitmap_write_access(struct file * file)
4954 struct inode *inode = file->f_mapping->host;
4956 spin_lock(&inode->i_lock);
4957 if (atomic_read(&inode->i_writecount) > 1) {
4958 spin_unlock(&inode->i_lock);
4961 atomic_set(&inode->i_writecount, -1);
4962 spin_unlock(&inode->i_lock);
4967 void restore_bitmap_write_access(struct file *file)
4969 struct inode *inode = file->f_mapping->host;
4971 spin_lock(&inode->i_lock);
4972 atomic_set(&inode->i_writecount, 1);
4973 spin_unlock(&inode->i_lock);
4976 static void md_clean(mddev_t *mddev)
4978 mddev->array_sectors = 0;
4979 mddev->external_size = 0;
4980 mddev->dev_sectors = 0;
4981 mddev->raid_disks = 0;
4982 mddev->recovery_cp = 0;
4983 mddev->resync_min = 0;
4984 mddev->resync_max = MaxSector;
4985 mddev->reshape_position = MaxSector;
4986 mddev->external = 0;
4987 mddev->persistent = 0;
4988 mddev->level = LEVEL_NONE;
4989 mddev->clevel[0] = 0;
4992 mddev->metadata_type[0] = 0;
4993 mddev->chunk_sectors = 0;
4994 mddev->ctime = mddev->utime = 0;
4996 mddev->max_disks = 0;
4998 mddev->can_decrease_events = 0;
4999 mddev->delta_disks = 0;
5000 mddev->new_level = LEVEL_NONE;
5001 mddev->new_layout = 0;
5002 mddev->new_chunk_sectors = 0;
5003 mddev->curr_resync = 0;
5004 mddev->resync_mismatches = 0;
5005 mddev->suspend_lo = mddev->suspend_hi = 0;
5006 mddev->sync_speed_min = mddev->sync_speed_max = 0;
5007 mddev->recovery = 0;
5010 mddev->degraded = 0;
5011 mddev->safemode = 0;
5012 mddev->bitmap_info.offset = 0;
5013 mddev->bitmap_info.default_offset = 0;
5014 mddev->bitmap_info.chunksize = 0;
5015 mddev->bitmap_info.daemon_sleep = 0;
5016 mddev->bitmap_info.max_write_behind = 0;
5019 static void __md_stop_writes(mddev_t *mddev)
5021 if (mddev->sync_thread) {
5022 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5023 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5024 reap_sync_thread(mddev);
5027 del_timer_sync(&mddev->safemode_timer);
5029 bitmap_flush(mddev);
5030 md_super_wait(mddev);
5032 if (!mddev->in_sync || mddev->flags) {
5033 /* mark array as shutdown cleanly */
5035 md_update_sb(mddev, 1);
5039 void md_stop_writes(mddev_t *mddev)
5042 __md_stop_writes(mddev);
5043 mddev_unlock(mddev);
5045 EXPORT_SYMBOL_GPL(md_stop_writes);
5047 void md_stop(mddev_t *mddev)
5050 mddev->pers->stop(mddev);
5051 if (mddev->pers->sync_request && mddev->to_remove == NULL)
5052 mddev->to_remove = &md_redundancy_group;
5053 module_put(mddev->pers->owner);
5055 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5057 EXPORT_SYMBOL_GPL(md_stop);
5059 static int md_set_readonly(mddev_t *mddev, int is_open)
5062 mutex_lock(&mddev->open_mutex);
5063 if (atomic_read(&mddev->openers) > is_open) {
5064 printk("md: %s still in use.\n",mdname(mddev));
5069 __md_stop_writes(mddev);
5075 set_disk_ro(mddev->gendisk, 1);
5076 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5077 sysfs_notify_dirent_safe(mddev->sysfs_state);
5081 mutex_unlock(&mddev->open_mutex);
5086 * 0 - completely stop and dis-assemble array
5087 * 2 - stop but do not disassemble array
5089 static int do_md_stop(mddev_t * mddev, int mode, int is_open)
5091 struct gendisk *disk = mddev->gendisk;
5094 mutex_lock(&mddev->open_mutex);
5095 if (atomic_read(&mddev->openers) > is_open ||
5096 mddev->sysfs_active) {
5097 printk("md: %s still in use.\n",mdname(mddev));
5098 mutex_unlock(&mddev->open_mutex);
5104 set_disk_ro(disk, 0);
5106 __md_stop_writes(mddev);
5108 mddev->queue->merge_bvec_fn = NULL;
5109 mddev->queue->backing_dev_info.congested_fn = NULL;
5111 /* tell userspace to handle 'inactive' */
5112 sysfs_notify_dirent_safe(mddev->sysfs_state);
5114 list_for_each_entry(rdev, &mddev->disks, same_set)
5115 if (rdev->raid_disk >= 0)
5116 sysfs_unlink_rdev(mddev, rdev);
5118 set_capacity(disk, 0);
5119 mutex_unlock(&mddev->open_mutex);
5121 revalidate_disk(disk);
5126 mutex_unlock(&mddev->open_mutex);
5128 * Free resources if final stop
5131 printk(KERN_INFO "md: %s stopped.\n", mdname(mddev));
5133 bitmap_destroy(mddev);
5134 if (mddev->bitmap_info.file) {
5135 restore_bitmap_write_access(mddev->bitmap_info.file);
5136 fput(mddev->bitmap_info.file);
5137 mddev->bitmap_info.file = NULL;
5139 mddev->bitmap_info.offset = 0;
5141 export_array(mddev);
5144 kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
5145 if (mddev->hold_active == UNTIL_STOP)
5146 mddev->hold_active = 0;
5148 blk_integrity_unregister(disk);
5149 md_new_event(mddev);
5150 sysfs_notify_dirent_safe(mddev->sysfs_state);
5155 static void autorun_array(mddev_t *mddev)
5160 if (list_empty(&mddev->disks))
5163 printk(KERN_INFO "md: running: ");
5165 list_for_each_entry(rdev, &mddev->disks, same_set) {
5166 char b[BDEVNAME_SIZE];
5167 printk("<%s>", bdevname(rdev->bdev,b));
5171 err = do_md_run(mddev);
5173 printk(KERN_WARNING "md: do_md_run() returned %d\n", err);
5174 do_md_stop(mddev, 0, 0);
5179 * lets try to run arrays based on all disks that have arrived
5180 * until now. (those are in pending_raid_disks)
5182 * the method: pick the first pending disk, collect all disks with
5183 * the same UUID, remove all from the pending list and put them into
5184 * the 'same_array' list. Then order this list based on superblock
5185 * update time (freshest comes first), kick out 'old' disks and
5186 * compare superblocks. If everything's fine then run it.
5188 * If "unit" is allocated, then bump its reference count
5190 static void autorun_devices(int part)
5192 mdk_rdev_t *rdev0, *rdev, *tmp;
5194 char b[BDEVNAME_SIZE];
5196 printk(KERN_INFO "md: autorun ...\n");
5197 while (!list_empty(&pending_raid_disks)) {
5200 LIST_HEAD(candidates);
5201 rdev0 = list_entry(pending_raid_disks.next,
5202 mdk_rdev_t, same_set);
5204 printk(KERN_INFO "md: considering %s ...\n",
5205 bdevname(rdev0->bdev,b));
5206 INIT_LIST_HEAD(&candidates);
5207 rdev_for_each_list(rdev, tmp, &pending_raid_disks)
5208 if (super_90_load(rdev, rdev0, 0) >= 0) {
5209 printk(KERN_INFO "md: adding %s ...\n",
5210 bdevname(rdev->bdev,b));
5211 list_move(&rdev->same_set, &candidates);
5214 * now we have a set of devices, with all of them having
5215 * mostly sane superblocks. It's time to allocate the
5219 dev = MKDEV(mdp_major,
5220 rdev0->preferred_minor << MdpMinorShift);
5221 unit = MINOR(dev) >> MdpMinorShift;
5223 dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
5226 if (rdev0->preferred_minor != unit) {
5227 printk(KERN_INFO "md: unit number in %s is bad: %d\n",
5228 bdevname(rdev0->bdev, b), rdev0->preferred_minor);
5232 md_probe(dev, NULL, NULL);
5233 mddev = mddev_find(dev);
5234 if (!mddev || !mddev->gendisk) {
5238 "md: cannot allocate memory for md drive.\n");
5241 if (mddev_lock(mddev))
5242 printk(KERN_WARNING "md: %s locked, cannot run\n",
5244 else if (mddev->raid_disks || mddev->major_version
5245 || !list_empty(&mddev->disks)) {
5247 "md: %s already running, cannot run %s\n",
5248 mdname(mddev), bdevname(rdev0->bdev,b));
5249 mddev_unlock(mddev);
5251 printk(KERN_INFO "md: created %s\n", mdname(mddev));
5252 mddev->persistent = 1;
5253 rdev_for_each_list(rdev, tmp, &candidates) {
5254 list_del_init(&rdev->same_set);
5255 if (bind_rdev_to_array(rdev, mddev))
5258 autorun_array(mddev);
5259 mddev_unlock(mddev);
5261 /* on success, candidates will be empty, on error
5264 rdev_for_each_list(rdev, tmp, &candidates) {
5265 list_del_init(&rdev->same_set);
5270 printk(KERN_INFO "md: ... autorun DONE.\n");
5272 #endif /* !MODULE */
5274 static int get_version(void __user * arg)
5278 ver.major = MD_MAJOR_VERSION;
5279 ver.minor = MD_MINOR_VERSION;
5280 ver.patchlevel = MD_PATCHLEVEL_VERSION;
5282 if (copy_to_user(arg, &ver, sizeof(ver)))
5288 static int get_array_info(mddev_t * mddev, void __user * arg)
5290 mdu_array_info_t info;
5291 int nr,working,insync,failed,spare;
5294 nr=working=insync=failed=spare=0;
5295 list_for_each_entry(rdev, &mddev->disks, same_set) {
5297 if (test_bit(Faulty, &rdev->flags))
5301 if (test_bit(In_sync, &rdev->flags))
5308 info.major_version = mddev->major_version;
5309 info.minor_version = mddev->minor_version;
5310 info.patch_version = MD_PATCHLEVEL_VERSION;
5311 info.ctime = mddev->ctime;
5312 info.level = mddev->level;
5313 info.size = mddev->dev_sectors / 2;
5314 if (info.size != mddev->dev_sectors / 2) /* overflow */
5317 info.raid_disks = mddev->raid_disks;
5318 info.md_minor = mddev->md_minor;
5319 info.not_persistent= !mddev->persistent;
5321 info.utime = mddev->utime;
5324 info.state = (1<<MD_SB_CLEAN);
5325 if (mddev->bitmap && mddev->bitmap_info.offset)
5326 info.state = (1<<MD_SB_BITMAP_PRESENT);
5327 info.active_disks = insync;
5328 info.working_disks = working;
5329 info.failed_disks = failed;
5330 info.spare_disks = spare;
5332 info.layout = mddev->layout;
5333 info.chunk_size = mddev->chunk_sectors << 9;
5335 if (copy_to_user(arg, &info, sizeof(info)))
5341 static int get_bitmap_file(mddev_t * mddev, void __user * arg)
5343 mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */
5344 char *ptr, *buf = NULL;
5347 if (md_allow_write(mddev))
5348 file = kmalloc(sizeof(*file), GFP_NOIO);
5350 file = kmalloc(sizeof(*file), GFP_KERNEL);
5355 /* bitmap disabled, zero the first byte and copy out */
5356 if (!mddev->bitmap || !mddev->bitmap->file) {
5357 file->pathname[0] = '\0';
5361 buf = kmalloc(sizeof(file->pathname), GFP_KERNEL);
5365 ptr = d_path(&mddev->bitmap->file->f_path, buf, sizeof(file->pathname));
5369 strcpy(file->pathname, ptr);
5373 if (copy_to_user(arg, file, sizeof(*file)))
5381 static int get_disk_info(mddev_t * mddev, void __user * arg)
5383 mdu_disk_info_t info;
5386 if (copy_from_user(&info, arg, sizeof(info)))
5389 rdev = find_rdev_nr(mddev, info.number);
5391 info.major = MAJOR(rdev->bdev->bd_dev);
5392 info.minor = MINOR(rdev->bdev->bd_dev);
5393 info.raid_disk = rdev->raid_disk;
5395 if (test_bit(Faulty, &rdev->flags))
5396 info.state |= (1<<MD_DISK_FAULTY);
5397 else if (test_bit(In_sync, &rdev->flags)) {
5398 info.state |= (1<<MD_DISK_ACTIVE);
5399 info.state |= (1<<MD_DISK_SYNC);
5401 if (test_bit(WriteMostly, &rdev->flags))
5402 info.state |= (1<<MD_DISK_WRITEMOSTLY);
5404 info.major = info.minor = 0;
5405 info.raid_disk = -1;
5406 info.state = (1<<MD_DISK_REMOVED);
5409 if (copy_to_user(arg, &info, sizeof(info)))
5415 static int add_new_disk(mddev_t * mddev, mdu_disk_info_t *info)
5417 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
5419 dev_t dev = MKDEV(info->major,info->minor);
5421 if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
5424 if (!mddev->raid_disks) {
5426 /* expecting a device which has a superblock */
5427 rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
5430 "md: md_import_device returned %ld\n",
5432 return PTR_ERR(rdev);
5434 if (!list_empty(&mddev->disks)) {
5435 mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
5436 mdk_rdev_t, same_set);
5437 err = super_types[mddev->major_version]
5438 .load_super(rdev, rdev0, mddev->minor_version);
5441 "md: %s has different UUID to %s\n",
5442 bdevname(rdev->bdev,b),
5443 bdevname(rdev0->bdev,b2));
5448 err = bind_rdev_to_array(rdev, mddev);
5455 * add_new_disk can be used once the array is assembled
5456 * to add "hot spares". They must already have a superblock
5461 if (!mddev->pers->hot_add_disk) {
5463 "%s: personality does not support diskops!\n",
5467 if (mddev->persistent)
5468 rdev = md_import_device(dev, mddev->major_version,
5469 mddev->minor_version);
5471 rdev = md_import_device(dev, -1, -1);
5474 "md: md_import_device returned %ld\n",
5476 return PTR_ERR(rdev);
5478 /* set saved_raid_disk if appropriate */
5479 if (!mddev->persistent) {
5480 if (info->state & (1<<MD_DISK_SYNC) &&
5481 info->raid_disk < mddev->raid_disks) {
5482 rdev->raid_disk = info->raid_disk;
5483 set_bit(In_sync, &rdev->flags);
5485 rdev->raid_disk = -1;
5487 super_types[mddev->major_version].
5488 validate_super(mddev, rdev);
5489 if ((info->state & (1<<MD_DISK_SYNC)) &&
5490 (!test_bit(In_sync, &rdev->flags) ||
5491 rdev->raid_disk != info->raid_disk)) {
5492 /* This was a hot-add request, but events doesn't
5493 * match, so reject it.
5499 if (test_bit(In_sync, &rdev->flags))
5500 rdev->saved_raid_disk = rdev->raid_disk;
5502 rdev->saved_raid_disk = -1;
5504 clear_bit(In_sync, &rdev->flags); /* just to be sure */
5505 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
5506 set_bit(WriteMostly, &rdev->flags);
5508 clear_bit(WriteMostly, &rdev->flags);
5510 rdev->raid_disk = -1;
5511 err = bind_rdev_to_array(rdev, mddev);
5512 if (!err && !mddev->pers->hot_remove_disk) {
5513 /* If there is hot_add_disk but no hot_remove_disk
5514 * then added disks for geometry changes,
5515 * and should be added immediately.
5517 super_types[mddev->major_version].
5518 validate_super(mddev, rdev);
5519 err = mddev->pers->hot_add_disk(mddev, rdev);
5521 unbind_rdev_from_array(rdev);
5526 sysfs_notify_dirent_safe(rdev->sysfs_state);
5528 md_update_sb(mddev, 1);
5529 if (mddev->degraded)
5530 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
5531 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5533 md_new_event(mddev);
5534 md_wakeup_thread(mddev->thread);
5538 /* otherwise, add_new_disk is only allowed
5539 * for major_version==0 superblocks
5541 if (mddev->major_version != 0) {
5542 printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n",
5547 if (!(info->state & (1<<MD_DISK_FAULTY))) {
5549 rdev = md_import_device(dev, -1, 0);
5552 "md: error, md_import_device() returned %ld\n",
5554 return PTR_ERR(rdev);
5556 rdev->desc_nr = info->number;
5557 if (info->raid_disk < mddev->raid_disks)
5558 rdev->raid_disk = info->raid_disk;
5560 rdev->raid_disk = -1;
5562 if (rdev->raid_disk < mddev->raid_disks)
5563 if (info->state & (1<<MD_DISK_SYNC))
5564 set_bit(In_sync, &rdev->flags);
5566 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
5567 set_bit(WriteMostly, &rdev->flags);
5569 if (!mddev->persistent) {
5570 printk(KERN_INFO "md: nonpersistent superblock ...\n");
5571 rdev->sb_start = i_size_read(rdev->bdev->bd_inode) / 512;
5573 rdev->sb_start = calc_dev_sboffset(rdev);
5574 rdev->sectors = rdev->sb_start;
5576 err = bind_rdev_to_array(rdev, mddev);
5586 static int hot_remove_disk(mddev_t * mddev, dev_t dev)
5588 char b[BDEVNAME_SIZE];
5591 rdev = find_rdev(mddev, dev);
5595 if (rdev->raid_disk >= 0)
5598 kick_rdev_from_array(rdev);
5599 md_update_sb(mddev, 1);
5600 md_new_event(mddev);
5604 printk(KERN_WARNING "md: cannot remove active disk %s from %s ...\n",
5605 bdevname(rdev->bdev,b), mdname(mddev));
5609 static int hot_add_disk(mddev_t * mddev, dev_t dev)
5611 char b[BDEVNAME_SIZE];
5618 if (mddev->major_version != 0) {
5619 printk(KERN_WARNING "%s: HOT_ADD may only be used with"
5620 " version-0 superblocks.\n",
5624 if (!mddev->pers->hot_add_disk) {
5626 "%s: personality does not support diskops!\n",
5631 rdev = md_import_device(dev, -1, 0);
5634 "md: error, md_import_device() returned %ld\n",
5639 if (mddev->persistent)
5640 rdev->sb_start = calc_dev_sboffset(rdev);
5642 rdev->sb_start = i_size_read(rdev->bdev->bd_inode) / 512;
5644 rdev->sectors = rdev->sb_start;
5646 if (test_bit(Faulty, &rdev->flags)) {
5648 "md: can not hot-add faulty %s disk to %s!\n",
5649 bdevname(rdev->bdev,b), mdname(mddev));
5653 clear_bit(In_sync, &rdev->flags);
5655 rdev->saved_raid_disk = -1;
5656 err = bind_rdev_to_array(rdev, mddev);
5661 * The rest should better be atomic, we can have disk failures
5662 * noticed in interrupt contexts ...
5665 rdev->raid_disk = -1;
5667 md_update_sb(mddev, 1);
5670 * Kick recovery, maybe this spare has to be added to the
5671 * array immediately.
5673 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5674 md_wakeup_thread(mddev->thread);
5675 md_new_event(mddev);
5683 static int set_bitmap_file(mddev_t *mddev, int fd)
5688 if (!mddev->pers->quiesce)
5690 if (mddev->recovery || mddev->sync_thread)
5692 /* we should be able to change the bitmap.. */
5698 return -EEXIST; /* cannot add when bitmap is present */
5699 mddev->bitmap_info.file = fget(fd);
5701 if (mddev->bitmap_info.file == NULL) {
5702 printk(KERN_ERR "%s: error: failed to get bitmap file\n",
5707 err = deny_bitmap_write_access(mddev->bitmap_info.file);
5709 printk(KERN_ERR "%s: error: bitmap file is already in use\n",
5711 fput(mddev->bitmap_info.file);
5712 mddev->bitmap_info.file = NULL;
5715 mddev->bitmap_info.offset = 0; /* file overrides offset */
5716 } else if (mddev->bitmap == NULL)
5717 return -ENOENT; /* cannot remove what isn't there */
5720 mddev->pers->quiesce(mddev, 1);
5722 err = bitmap_create(mddev);
5724 err = bitmap_load(mddev);
5726 if (fd < 0 || err) {
5727 bitmap_destroy(mddev);
5728 fd = -1; /* make sure to put the file */
5730 mddev->pers->quiesce(mddev, 0);
5733 if (mddev->bitmap_info.file) {
5734 restore_bitmap_write_access(mddev->bitmap_info.file);
5735 fput(mddev->bitmap_info.file);
5737 mddev->bitmap_info.file = NULL;
5744 * set_array_info is used two different ways
5745 * The original usage is when creating a new array.
5746 * In this usage, raid_disks is > 0 and it together with
5747 * level, size, not_persistent,layout,chunksize determine the
5748 * shape of the array.
5749 * This will always create an array with a type-0.90.0 superblock.
5750 * The newer usage is when assembling an array.
5751 * In this case raid_disks will be 0, and the major_version field is
5752 * use to determine which style super-blocks are to be found on the devices.
5753 * The minor and patch _version numbers are also kept incase the
5754 * super_block handler wishes to interpret them.
5756 static int set_array_info(mddev_t * mddev, mdu_array_info_t *info)
5759 if (info->raid_disks == 0) {
5760 /* just setting version number for superblock loading */
5761 if (info->major_version < 0 ||
5762 info->major_version >= ARRAY_SIZE(super_types) ||
5763 super_types[info->major_version].name == NULL) {
5764 /* maybe try to auto-load a module? */
5766 "md: superblock version %d not known\n",
5767 info->major_version);
5770 mddev->major_version = info->major_version;
5771 mddev->minor_version = info->minor_version;
5772 mddev->patch_version = info->patch_version;
5773 mddev->persistent = !info->not_persistent;
5774 /* ensure mddev_put doesn't delete this now that there
5775 * is some minimal configuration.
5777 mddev->ctime = get_seconds();
5780 mddev->major_version = MD_MAJOR_VERSION;
5781 mddev->minor_version = MD_MINOR_VERSION;
5782 mddev->patch_version = MD_PATCHLEVEL_VERSION;
5783 mddev->ctime = get_seconds();
5785 mddev->level = info->level;
5786 mddev->clevel[0] = 0;
5787 mddev->dev_sectors = 2 * (sector_t)info->size;
5788 mddev->raid_disks = info->raid_disks;
5789 /* don't set md_minor, it is determined by which /dev/md* was
5792 if (info->state & (1<<MD_SB_CLEAN))
5793 mddev->recovery_cp = MaxSector;
5795 mddev->recovery_cp = 0;
5796 mddev->persistent = ! info->not_persistent;
5797 mddev->external = 0;
5799 mddev->layout = info->layout;
5800 mddev->chunk_sectors = info->chunk_size >> 9;
5802 mddev->max_disks = MD_SB_DISKS;
5804 if (mddev->persistent)
5806 set_bit(MD_CHANGE_DEVS, &mddev->flags);
5808 mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9;
5809 mddev->bitmap_info.offset = 0;
5811 mddev->reshape_position = MaxSector;
5814 * Generate a 128 bit UUID
5816 get_random_bytes(mddev->uuid, 16);
5818 mddev->new_level = mddev->level;
5819 mddev->new_chunk_sectors = mddev->chunk_sectors;
5820 mddev->new_layout = mddev->layout;
5821 mddev->delta_disks = 0;
5826 void md_set_array_sectors(mddev_t *mddev, sector_t array_sectors)
5828 WARN(!mddev_is_locked(mddev), "%s: unlocked mddev!\n", __func__);
5830 if (mddev->external_size)
5833 mddev->array_sectors = array_sectors;
5835 EXPORT_SYMBOL(md_set_array_sectors);
5837 static int update_size(mddev_t *mddev, sector_t num_sectors)
5841 int fit = (num_sectors == 0);
5843 if (mddev->pers->resize == NULL)
5845 /* The "num_sectors" is the number of sectors of each device that
5846 * is used. This can only make sense for arrays with redundancy.
5847 * linear and raid0 always use whatever space is available. We can only
5848 * consider changing this number if no resync or reconstruction is
5849 * happening, and if the new size is acceptable. It must fit before the
5850 * sb_start or, if that is <data_offset, it must fit before the size
5851 * of each device. If num_sectors is zero, we find the largest size
5854 if (mddev->sync_thread)
5857 /* Sorry, cannot grow a bitmap yet, just remove it,
5861 list_for_each_entry(rdev, &mddev->disks, same_set) {
5862 sector_t avail = rdev->sectors;
5864 if (fit && (num_sectors == 0 || num_sectors > avail))
5865 num_sectors = avail;
5866 if (avail < num_sectors)
5869 rv = mddev->pers->resize(mddev, num_sectors);
5871 revalidate_disk(mddev->gendisk);
5875 static int update_raid_disks(mddev_t *mddev, int raid_disks)
5878 /* change the number of raid disks */
5879 if (mddev->pers->check_reshape == NULL)
5881 if (raid_disks <= 0 ||
5882 (mddev->max_disks && raid_disks >= mddev->max_disks))
5884 if (mddev->sync_thread || mddev->reshape_position != MaxSector)
5886 mddev->delta_disks = raid_disks - mddev->raid_disks;
5888 rv = mddev->pers->check_reshape(mddev);
5890 mddev->delta_disks = 0;
5896 * update_array_info is used to change the configuration of an
5898 * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
5899 * fields in the info are checked against the array.
5900 * Any differences that cannot be handled will cause an error.
5901 * Normally, only one change can be managed at a time.
5903 static int update_array_info(mddev_t *mddev, mdu_array_info_t *info)
5909 /* calculate expected state,ignoring low bits */
5910 if (mddev->bitmap && mddev->bitmap_info.offset)
5911 state |= (1 << MD_SB_BITMAP_PRESENT);
5913 if (mddev->major_version != info->major_version ||
5914 mddev->minor_version != info->minor_version ||
5915 /* mddev->patch_version != info->patch_version || */
5916 mddev->ctime != info->ctime ||
5917 mddev->level != info->level ||
5918 /* mddev->layout != info->layout || */
5919 !mddev->persistent != info->not_persistent||
5920 mddev->chunk_sectors != info->chunk_size >> 9 ||
5921 /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
5922 ((state^info->state) & 0xfffffe00)
5925 /* Check there is only one change */
5926 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
5928 if (mddev->raid_disks != info->raid_disks)
5930 if (mddev->layout != info->layout)
5932 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT))
5939 if (mddev->layout != info->layout) {
5941 * we don't need to do anything at the md level, the
5942 * personality will take care of it all.
5944 if (mddev->pers->check_reshape == NULL)
5947 mddev->new_layout = info->layout;
5948 rv = mddev->pers->check_reshape(mddev);
5950 mddev->new_layout = mddev->layout;
5954 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
5955 rv = update_size(mddev, (sector_t)info->size * 2);
5957 if (mddev->raid_disks != info->raid_disks)
5958 rv = update_raid_disks(mddev, info->raid_disks);
5960 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) {
5961 if (mddev->pers->quiesce == NULL)
5963 if (mddev->recovery || mddev->sync_thread)
5965 if (info->state & (1<<MD_SB_BITMAP_PRESENT)) {
5966 /* add the bitmap */
5969 if (mddev->bitmap_info.default_offset == 0)
5971 mddev->bitmap_info.offset =
5972 mddev->bitmap_info.default_offset;
5973 mddev->pers->quiesce(mddev, 1);
5974 rv = bitmap_create(mddev);
5976 rv = bitmap_load(mddev);
5978 bitmap_destroy(mddev);
5979 mddev->pers->quiesce(mddev, 0);
5981 /* remove the bitmap */
5984 if (mddev->bitmap->file)
5986 mddev->pers->quiesce(mddev, 1);
5987 bitmap_destroy(mddev);
5988 mddev->pers->quiesce(mddev, 0);
5989 mddev->bitmap_info.offset = 0;
5992 md_update_sb(mddev, 1);
5996 static int set_disk_faulty(mddev_t *mddev, dev_t dev)
6000 if (mddev->pers == NULL)
6003 rdev = find_rdev(mddev, dev);
6007 md_error(mddev, rdev);
6008 if (!test_bit(Faulty, &rdev->flags))
6014 * We have a problem here : there is no easy way to give a CHS
6015 * virtual geometry. We currently pretend that we have a 2 heads
6016 * 4 sectors (with a BIG number of cylinders...). This drives
6017 * dosfs just mad... ;-)
6019 static int md_getgeo(struct block_device *bdev, struct hd_geometry *geo)
6021 mddev_t *mddev = bdev->bd_disk->private_data;
6025 geo->cylinders = mddev->array_sectors / 8;
6029 static int md_ioctl(struct block_device *bdev, fmode_t mode,
6030 unsigned int cmd, unsigned long arg)
6033 void __user *argp = (void __user *)arg;
6034 mddev_t *mddev = NULL;
6037 if (!capable(CAP_SYS_ADMIN))
6041 * Commands dealing with the RAID driver but not any
6047 err = get_version(argp);
6050 case PRINT_RAID_DEBUG:
6058 autostart_arrays(arg);
6065 * Commands creating/starting a new array:
6068 mddev = bdev->bd_disk->private_data;
6075 err = mddev_lock(mddev);
6078 "md: ioctl lock interrupted, reason %d, cmd %d\n",
6085 case SET_ARRAY_INFO:
6087 mdu_array_info_t info;
6089 memset(&info, 0, sizeof(info));
6090 else if (copy_from_user(&info, argp, sizeof(info))) {
6095 err = update_array_info(mddev, &info);
6097 printk(KERN_WARNING "md: couldn't update"
6098 " array info. %d\n", err);
6103 if (!list_empty(&mddev->disks)) {
6105 "md: array %s already has disks!\n",
6110 if (mddev->raid_disks) {
6112 "md: array %s already initialised!\n",
6117 err = set_array_info(mddev, &info);
6119 printk(KERN_WARNING "md: couldn't set"
6120 " array info. %d\n", err);
6130 * Commands querying/configuring an existing array:
6132 /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
6133 * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */
6134 if ((!mddev->raid_disks && !mddev->external)
6135 && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY
6136 && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE
6137 && cmd != GET_BITMAP_FILE) {
6143 * Commands even a read-only array can execute:
6147 case GET_ARRAY_INFO:
6148 err = get_array_info(mddev, argp);
6151 case GET_BITMAP_FILE:
6152 err = get_bitmap_file(mddev, argp);
6156 err = get_disk_info(mddev, argp);
6159 case RESTART_ARRAY_RW:
6160 err = restart_array(mddev);
6164 err = do_md_stop(mddev, 0, 1);
6168 err = md_set_readonly(mddev, 1);
6172 if (get_user(ro, (int __user *)(arg))) {
6178 /* if the bdev is going readonly the value of mddev->ro
6179 * does not matter, no writes are coming
6184 /* are we are already prepared for writes? */
6188 /* transitioning to readauto need only happen for
6189 * arrays that call md_write_start
6192 err = restart_array(mddev);
6195 set_disk_ro(mddev->gendisk, 0);
6202 * The remaining ioctls are changing the state of the
6203 * superblock, so we do not allow them on read-only arrays.
6204 * However non-MD ioctls (e.g. get-size) will still come through
6205 * here and hit the 'default' below, so only disallow
6206 * 'md' ioctls, and switch to rw mode if started auto-readonly.
6208 if (_IOC_TYPE(cmd) == MD_MAJOR && mddev->ro && mddev->pers) {
6209 if (mddev->ro == 2) {
6211 sysfs_notify_dirent_safe(mddev->sysfs_state);
6212 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6213 md_wakeup_thread(mddev->thread);
6224 mdu_disk_info_t info;
6225 if (copy_from_user(&info, argp, sizeof(info)))
6228 err = add_new_disk(mddev, &info);
6232 case HOT_REMOVE_DISK:
6233 err = hot_remove_disk(mddev, new_decode_dev(arg));
6237 err = hot_add_disk(mddev, new_decode_dev(arg));
6240 case SET_DISK_FAULTY:
6241 err = set_disk_faulty(mddev, new_decode_dev(arg));
6245 err = do_md_run(mddev);
6248 case SET_BITMAP_FILE:
6249 err = set_bitmap_file(mddev, (int)arg);
6259 if (mddev->hold_active == UNTIL_IOCTL &&
6261 mddev->hold_active = 0;
6262 mddev_unlock(mddev);
6271 #ifdef CONFIG_COMPAT
6272 static int md_compat_ioctl(struct block_device *bdev, fmode_t mode,
6273 unsigned int cmd, unsigned long arg)
6276 case HOT_REMOVE_DISK:
6278 case SET_DISK_FAULTY:
6279 case SET_BITMAP_FILE:
6280 /* These take in integer arg, do not convert */
6283 arg = (unsigned long)compat_ptr(arg);
6287 return md_ioctl(bdev, mode, cmd, arg);
6289 #endif /* CONFIG_COMPAT */
6291 static int md_open(struct block_device *bdev, fmode_t mode)
6294 * Succeed if we can lock the mddev, which confirms that
6295 * it isn't being stopped right now.
6297 mddev_t *mddev = mddev_find(bdev->bd_dev);
6300 if (mddev->gendisk != bdev->bd_disk) {
6301 /* we are racing with mddev_put which is discarding this
6305 /* Wait until bdev->bd_disk is definitely gone */
6306 flush_workqueue(md_misc_wq);
6307 /* Then retry the open from the top */
6308 return -ERESTARTSYS;
6310 BUG_ON(mddev != bdev->bd_disk->private_data);
6312 if ((err = mutex_lock_interruptible(&mddev->open_mutex)))
6316 atomic_inc(&mddev->openers);
6317 mutex_unlock(&mddev->open_mutex);
6319 check_disk_change(bdev);
6324 static int md_release(struct gendisk *disk, fmode_t mode)
6326 mddev_t *mddev = disk->private_data;
6329 atomic_dec(&mddev->openers);
6335 static int md_media_changed(struct gendisk *disk)
6337 mddev_t *mddev = disk->private_data;
6339 return mddev->changed;
6342 static int md_revalidate(struct gendisk *disk)
6344 mddev_t *mddev = disk->private_data;
6349 static const struct block_device_operations md_fops =
6351 .owner = THIS_MODULE,
6353 .release = md_release,
6355 #ifdef CONFIG_COMPAT
6356 .compat_ioctl = md_compat_ioctl,
6358 .getgeo = md_getgeo,
6359 .media_changed = md_media_changed,
6360 .revalidate_disk= md_revalidate,
6363 static int md_thread(void * arg)
6365 mdk_thread_t *thread = arg;
6368 * md_thread is a 'system-thread', it's priority should be very
6369 * high. We avoid resource deadlocks individually in each
6370 * raid personality. (RAID5 does preallocation) We also use RR and
6371 * the very same RT priority as kswapd, thus we will never get
6372 * into a priority inversion deadlock.
6374 * we definitely have to have equal or higher priority than
6375 * bdflush, otherwise bdflush will deadlock if there are too
6376 * many dirty RAID5 blocks.
6379 allow_signal(SIGKILL);
6380 while (!kthread_should_stop()) {
6382 /* We need to wait INTERRUPTIBLE so that
6383 * we don't add to the load-average.
6384 * That means we need to be sure no signals are
6387 if (signal_pending(current))
6388 flush_signals(current);
6390 wait_event_interruptible_timeout
6392 test_bit(THREAD_WAKEUP, &thread->flags)
6393 || kthread_should_stop(),
6396 clear_bit(THREAD_WAKEUP, &thread->flags);
6397 if (!kthread_should_stop())
6398 thread->run(thread->mddev);
6404 void md_wakeup_thread(mdk_thread_t *thread)
6407 dprintk("md: waking up MD thread %s.\n", thread->tsk->comm);
6408 set_bit(THREAD_WAKEUP, &thread->flags);
6409 wake_up(&thread->wqueue);
6413 mdk_thread_t *md_register_thread(void (*run) (mddev_t *), mddev_t *mddev,
6416 mdk_thread_t *thread;
6418 thread = kzalloc(sizeof(mdk_thread_t), GFP_KERNEL);
6422 init_waitqueue_head(&thread->wqueue);
6425 thread->mddev = mddev;
6426 thread->timeout = MAX_SCHEDULE_TIMEOUT;
6427 thread->tsk = kthread_run(md_thread, thread,
6429 mdname(thread->mddev),
6430 name ?: mddev->pers->name);
6431 if (IS_ERR(thread->tsk)) {
6438 void md_unregister_thread(mdk_thread_t **threadp)
6440 mdk_thread_t *thread = *threadp;
6443 dprintk("interrupting MD-thread pid %d\n", task_pid_nr(thread->tsk));
6444 /* Locking ensures that mddev_unlock does not wake_up a
6445 * non-existent thread
6447 spin_lock(&pers_lock);
6449 spin_unlock(&pers_lock);
6451 kthread_stop(thread->tsk);
6455 void md_error(mddev_t *mddev, mdk_rdev_t *rdev)
6462 if (!rdev || test_bit(Faulty, &rdev->flags))
6465 if (!mddev->pers || !mddev->pers->error_handler)
6467 mddev->pers->error_handler(mddev,rdev);
6468 if (mddev->degraded)
6469 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
6470 sysfs_notify_dirent_safe(rdev->sysfs_state);
6471 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6472 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6473 md_wakeup_thread(mddev->thread);
6474 if (mddev->event_work.func)
6475 queue_work(md_misc_wq, &mddev->event_work);
6476 md_new_event_inintr(mddev);
6479 /* seq_file implementation /proc/mdstat */
6481 static void status_unused(struct seq_file *seq)
6486 seq_printf(seq, "unused devices: ");
6488 list_for_each_entry(rdev, &pending_raid_disks, same_set) {
6489 char b[BDEVNAME_SIZE];
6491 seq_printf(seq, "%s ",
6492 bdevname(rdev->bdev,b));
6495 seq_printf(seq, "<none>");
6497 seq_printf(seq, "\n");
6501 static void status_resync(struct seq_file *seq, mddev_t * mddev)
6503 sector_t max_sectors, resync, res;
6504 unsigned long dt, db;
6507 unsigned int per_milli;
6509 resync = mddev->curr_resync - atomic_read(&mddev->recovery_active);
6511 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
6512 max_sectors = mddev->resync_max_sectors;
6514 max_sectors = mddev->dev_sectors;
6517 * Should not happen.
6523 /* Pick 'scale' such that (resync>>scale)*1000 will fit
6524 * in a sector_t, and (max_sectors>>scale) will fit in a
6525 * u32, as those are the requirements for sector_div.
6526 * Thus 'scale' must be at least 10
6529 if (sizeof(sector_t) > sizeof(unsigned long)) {
6530 while ( max_sectors/2 > (1ULL<<(scale+32)))
6533 res = (resync>>scale)*1000;
6534 sector_div(res, (u32)((max_sectors>>scale)+1));
6538 int i, x = per_milli/50, y = 20-x;
6539 seq_printf(seq, "[");
6540 for (i = 0; i < x; i++)
6541 seq_printf(seq, "=");
6542 seq_printf(seq, ">");
6543 for (i = 0; i < y; i++)
6544 seq_printf(seq, ".");
6545 seq_printf(seq, "] ");
6547 seq_printf(seq, " %s =%3u.%u%% (%llu/%llu)",
6548 (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)?
6550 (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)?
6552 (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
6553 "resync" : "recovery"))),
6554 per_milli/10, per_milli % 10,
6555 (unsigned long long) resync/2,
6556 (unsigned long long) max_sectors/2);
6559 * dt: time from mark until now
6560 * db: blocks written from mark until now
6561 * rt: remaining time
6563 * rt is a sector_t, so could be 32bit or 64bit.
6564 * So we divide before multiply in case it is 32bit and close
6566 * We scale the divisor (db) by 32 to avoid losing precision
6567 * near the end of resync when the number of remaining sectors
6569 * We then divide rt by 32 after multiplying by db to compensate.
6570 * The '+1' avoids division by zero if db is very small.
6572 dt = ((jiffies - mddev->resync_mark) / HZ);
6574 db = (mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active))
6575 - mddev->resync_mark_cnt;
6577 rt = max_sectors - resync; /* number of remaining sectors */
6578 sector_div(rt, db/32+1);
6582 seq_printf(seq, " finish=%lu.%lumin", (unsigned long)rt / 60,
6583 ((unsigned long)rt % 60)/6);
6585 seq_printf(seq, " speed=%ldK/sec", db/2/dt);
6588 static void *md_seq_start(struct seq_file *seq, loff_t *pos)
6590 struct list_head *tmp;
6600 spin_lock(&all_mddevs_lock);
6601 list_for_each(tmp,&all_mddevs)
6603 mddev = list_entry(tmp, mddev_t, all_mddevs);
6605 spin_unlock(&all_mddevs_lock);
6608 spin_unlock(&all_mddevs_lock);
6610 return (void*)2;/* tail */
6614 static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
6616 struct list_head *tmp;
6617 mddev_t *next_mddev, *mddev = v;
6623 spin_lock(&all_mddevs_lock);
6625 tmp = all_mddevs.next;
6627 tmp = mddev->all_mddevs.next;
6628 if (tmp != &all_mddevs)
6629 next_mddev = mddev_get(list_entry(tmp,mddev_t,all_mddevs));
6631 next_mddev = (void*)2;
6634 spin_unlock(&all_mddevs_lock);
6642 static void md_seq_stop(struct seq_file *seq, void *v)
6646 if (mddev && v != (void*)1 && v != (void*)2)
6650 static int md_seq_show(struct seq_file *seq, void *v)
6655 struct bitmap *bitmap;
6657 if (v == (void*)1) {
6658 struct mdk_personality *pers;
6659 seq_printf(seq, "Personalities : ");
6660 spin_lock(&pers_lock);
6661 list_for_each_entry(pers, &pers_list, list)
6662 seq_printf(seq, "[%s] ", pers->name);
6664 spin_unlock(&pers_lock);
6665 seq_printf(seq, "\n");
6666 seq->poll_event = atomic_read(&md_event_count);
6669 if (v == (void*)2) {
6674 if (mddev_lock(mddev) < 0)
6677 if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
6678 seq_printf(seq, "%s : %sactive", mdname(mddev),
6679 mddev->pers ? "" : "in");
6682 seq_printf(seq, " (read-only)");
6684 seq_printf(seq, " (auto-read-only)");
6685 seq_printf(seq, " %s", mddev->pers->name);
6689 list_for_each_entry(rdev, &mddev->disks, same_set) {
6690 char b[BDEVNAME_SIZE];
6691 seq_printf(seq, " %s[%d]",
6692 bdevname(rdev->bdev,b), rdev->desc_nr);
6693 if (test_bit(WriteMostly, &rdev->flags))
6694 seq_printf(seq, "(W)");
6695 if (test_bit(Faulty, &rdev->flags)) {
6696 seq_printf(seq, "(F)");
6698 } else if (rdev->raid_disk < 0)
6699 seq_printf(seq, "(S)"); /* spare */
6700 sectors += rdev->sectors;
6703 if (!list_empty(&mddev->disks)) {
6705 seq_printf(seq, "\n %llu blocks",
6706 (unsigned long long)
6707 mddev->array_sectors / 2);
6709 seq_printf(seq, "\n %llu blocks",
6710 (unsigned long long)sectors / 2);
6712 if (mddev->persistent) {
6713 if (mddev->major_version != 0 ||
6714 mddev->minor_version != 90) {
6715 seq_printf(seq," super %d.%d",
6716 mddev->major_version,
6717 mddev->minor_version);
6719 } else if (mddev->external)
6720 seq_printf(seq, " super external:%s",
6721 mddev->metadata_type);
6723 seq_printf(seq, " super non-persistent");
6726 mddev->pers->status(seq, mddev);
6727 seq_printf(seq, "\n ");
6728 if (mddev->pers->sync_request) {
6729 if (mddev->curr_resync > 2) {
6730 status_resync(seq, mddev);
6731 seq_printf(seq, "\n ");
6732 } else if (mddev->curr_resync == 1 || mddev->curr_resync == 2)
6733 seq_printf(seq, "\tresync=DELAYED\n ");
6734 else if (mddev->recovery_cp < MaxSector)
6735 seq_printf(seq, "\tresync=PENDING\n ");
6738 seq_printf(seq, "\n ");
6740 if ((bitmap = mddev->bitmap)) {
6741 unsigned long chunk_kb;
6742 unsigned long flags;
6743 spin_lock_irqsave(&bitmap->lock, flags);
6744 chunk_kb = mddev->bitmap_info.chunksize >> 10;
6745 seq_printf(seq, "bitmap: %lu/%lu pages [%luKB], "
6747 bitmap->pages - bitmap->missing_pages,
6749 (bitmap->pages - bitmap->missing_pages)
6750 << (PAGE_SHIFT - 10),
6751 chunk_kb ? chunk_kb : mddev->bitmap_info.chunksize,
6752 chunk_kb ? "KB" : "B");
6754 seq_printf(seq, ", file: ");
6755 seq_path(seq, &bitmap->file->f_path, " \t\n");
6758 seq_printf(seq, "\n");
6759 spin_unlock_irqrestore(&bitmap->lock, flags);
6762 seq_printf(seq, "\n");
6764 mddev_unlock(mddev);
6769 static const struct seq_operations md_seq_ops = {
6770 .start = md_seq_start,
6771 .next = md_seq_next,
6772 .stop = md_seq_stop,
6773 .show = md_seq_show,
6776 static int md_seq_open(struct inode *inode, struct file *file)
6778 struct seq_file *seq;
6781 error = seq_open(file, &md_seq_ops);
6785 seq = file->private_data;
6786 seq->poll_event = atomic_read(&md_event_count);
6790 static unsigned int mdstat_poll(struct file *filp, poll_table *wait)
6792 struct seq_file *seq = filp->private_data;
6795 poll_wait(filp, &md_event_waiters, wait);
6797 /* always allow read */
6798 mask = POLLIN | POLLRDNORM;
6800 if (seq->poll_event != atomic_read(&md_event_count))
6801 mask |= POLLERR | POLLPRI;
6805 static const struct file_operations md_seq_fops = {
6806 .owner = THIS_MODULE,
6807 .open = md_seq_open,
6809 .llseek = seq_lseek,
6810 .release = seq_release_private,
6811 .poll = mdstat_poll,
6814 int register_md_personality(struct mdk_personality *p)
6816 spin_lock(&pers_lock);
6817 list_add_tail(&p->list, &pers_list);
6818 printk(KERN_INFO "md: %s personality registered for level %d\n", p->name, p->level);
6819 spin_unlock(&pers_lock);
6823 int unregister_md_personality(struct mdk_personality *p)
6825 printk(KERN_INFO "md: %s personality unregistered\n", p->name);
6826 spin_lock(&pers_lock);
6827 list_del_init(&p->list);
6828 spin_unlock(&pers_lock);
6832 static int is_mddev_idle(mddev_t *mddev, int init)
6840 rdev_for_each_rcu(rdev, mddev) {
6841 struct gendisk *disk = rdev->bdev->bd_contains->bd_disk;
6842 curr_events = (int)part_stat_read(&disk->part0, sectors[0]) +
6843 (int)part_stat_read(&disk->part0, sectors[1]) -
6844 atomic_read(&disk->sync_io);
6845 /* sync IO will cause sync_io to increase before the disk_stats
6846 * as sync_io is counted when a request starts, and
6847 * disk_stats is counted when it completes.
6848 * So resync activity will cause curr_events to be smaller than
6849 * when there was no such activity.
6850 * non-sync IO will cause disk_stat to increase without
6851 * increasing sync_io so curr_events will (eventually)
6852 * be larger than it was before. Once it becomes
6853 * substantially larger, the test below will cause
6854 * the array to appear non-idle, and resync will slow
6856 * If there is a lot of outstanding resync activity when
6857 * we set last_event to curr_events, then all that activity
6858 * completing might cause the array to appear non-idle
6859 * and resync will be slowed down even though there might
6860 * not have been non-resync activity. This will only
6861 * happen once though. 'last_events' will soon reflect
6862 * the state where there is little or no outstanding
6863 * resync requests, and further resync activity will
6864 * always make curr_events less than last_events.
6867 if (init || curr_events - rdev->last_events > 64) {
6868 rdev->last_events = curr_events;
6876 void md_done_sync(mddev_t *mddev, int blocks, int ok)
6878 /* another "blocks" (512byte) blocks have been synced */
6879 atomic_sub(blocks, &mddev->recovery_active);
6880 wake_up(&mddev->recovery_wait);
6882 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6883 md_wakeup_thread(mddev->thread);
6884 // stop recovery, signal do_sync ....
6889 /* md_write_start(mddev, bi)
6890 * If we need to update some array metadata (e.g. 'active' flag
6891 * in superblock) before writing, schedule a superblock update
6892 * and wait for it to complete.
6894 void md_write_start(mddev_t *mddev, struct bio *bi)
6897 if (bio_data_dir(bi) != WRITE)
6900 BUG_ON(mddev->ro == 1);
6901 if (mddev->ro == 2) {
6902 /* need to switch to read/write */
6904 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6905 md_wakeup_thread(mddev->thread);
6906 md_wakeup_thread(mddev->sync_thread);
6909 atomic_inc(&mddev->writes_pending);
6910 if (mddev->safemode == 1)
6911 mddev->safemode = 0;
6912 if (mddev->in_sync) {
6913 spin_lock_irq(&mddev->write_lock);
6914 if (mddev->in_sync) {
6916 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
6917 set_bit(MD_CHANGE_PENDING, &mddev->flags);
6918 md_wakeup_thread(mddev->thread);
6921 spin_unlock_irq(&mddev->write_lock);
6924 sysfs_notify_dirent_safe(mddev->sysfs_state);
6925 wait_event(mddev->sb_wait,
6926 !test_bit(MD_CHANGE_PENDING, &mddev->flags));
6929 void md_write_end(mddev_t *mddev)
6931 if (atomic_dec_and_test(&mddev->writes_pending)) {
6932 if (mddev->safemode == 2)
6933 md_wakeup_thread(mddev->thread);
6934 else if (mddev->safemode_delay)
6935 mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay);
6939 /* md_allow_write(mddev)
6940 * Calling this ensures that the array is marked 'active' so that writes
6941 * may proceed without blocking. It is important to call this before
6942 * attempting a GFP_KERNEL allocation while holding the mddev lock.
6943 * Must be called with mddev_lock held.
6945 * In the ->external case MD_CHANGE_CLEAN can not be cleared until mddev->lock
6946 * is dropped, so return -EAGAIN after notifying userspace.
6948 int md_allow_write(mddev_t *mddev)
6954 if (!mddev->pers->sync_request)
6957 spin_lock_irq(&mddev->write_lock);
6958 if (mddev->in_sync) {
6960 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
6961 set_bit(MD_CHANGE_PENDING, &mddev->flags);
6962 if (mddev->safemode_delay &&
6963 mddev->safemode == 0)
6964 mddev->safemode = 1;
6965 spin_unlock_irq(&mddev->write_lock);
6966 md_update_sb(mddev, 0);
6967 sysfs_notify_dirent_safe(mddev->sysfs_state);
6969 spin_unlock_irq(&mddev->write_lock);
6971 if (test_bit(MD_CHANGE_PENDING, &mddev->flags))
6976 EXPORT_SYMBOL_GPL(md_allow_write);
6978 #define SYNC_MARKS 10
6979 #define SYNC_MARK_STEP (3*HZ)
6980 void md_do_sync(mddev_t *mddev)
6983 unsigned int currspeed = 0,
6985 sector_t max_sectors,j, io_sectors;
6986 unsigned long mark[SYNC_MARKS];
6987 sector_t mark_cnt[SYNC_MARKS];
6989 struct list_head *tmp;
6990 sector_t last_check;
6995 /* just incase thread restarts... */
6996 if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
6998 if (mddev->ro) /* never try to sync a read-only array */
7001 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
7002 if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
7003 desc = "data-check";
7004 else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
7005 desc = "requested-resync";
7008 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
7013 /* we overload curr_resync somewhat here.
7014 * 0 == not engaged in resync at all
7015 * 2 == checking that there is no conflict with another sync
7016 * 1 == like 2, but have yielded to allow conflicting resync to
7018 * other == active in resync - this many blocks
7020 * Before starting a resync we must have set curr_resync to
7021 * 2, and then checked that every "conflicting" array has curr_resync
7022 * less than ours. When we find one that is the same or higher
7023 * we wait on resync_wait. To avoid deadlock, we reduce curr_resync
7024 * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
7025 * This will mean we have to start checking from the beginning again.
7030 mddev->curr_resync = 2;
7033 if (kthread_should_stop())
7034 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
7036 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
7038 for_each_mddev(mddev2, tmp) {
7039 if (mddev2 == mddev)
7041 if (!mddev->parallel_resync
7042 && mddev2->curr_resync
7043 && match_mddev_units(mddev, mddev2)) {
7045 if (mddev < mddev2 && mddev->curr_resync == 2) {
7046 /* arbitrarily yield */
7047 mddev->curr_resync = 1;
7048 wake_up(&resync_wait);
7050 if (mddev > mddev2 && mddev->curr_resync == 1)
7051 /* no need to wait here, we can wait the next
7052 * time 'round when curr_resync == 2
7055 /* We need to wait 'interruptible' so as not to
7056 * contribute to the load average, and not to
7057 * be caught by 'softlockup'
7059 prepare_to_wait(&resync_wait, &wq, TASK_INTERRUPTIBLE);
7060 if (!kthread_should_stop() &&
7061 mddev2->curr_resync >= mddev->curr_resync) {
7062 printk(KERN_INFO "md: delaying %s of %s"
7063 " until %s has finished (they"
7064 " share one or more physical units)\n",
7065 desc, mdname(mddev), mdname(mddev2));
7067 if (signal_pending(current))
7068 flush_signals(current);
7070 finish_wait(&resync_wait, &wq);
7073 finish_wait(&resync_wait, &wq);
7076 } while (mddev->curr_resync < 2);
7079 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
7080 /* resync follows the size requested by the personality,
7081 * which defaults to physical size, but can be virtual size
7083 max_sectors = mddev->resync_max_sectors;
7084 mddev->resync_mismatches = 0;
7085 /* we don't use the checkpoint if there's a bitmap */
7086 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
7087 j = mddev->resync_min;
7088 else if (!mddev->bitmap)
7089 j = mddev->recovery_cp;
7091 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
7092 max_sectors = mddev->dev_sectors;
7094 /* recovery follows the physical size of devices */
7095 max_sectors = mddev->dev_sectors;
7098 list_for_each_entry_rcu(rdev, &mddev->disks, same_set)
7099 if (rdev->raid_disk >= 0 &&
7100 !test_bit(Faulty, &rdev->flags) &&
7101 !test_bit(In_sync, &rdev->flags) &&
7102 rdev->recovery_offset < j)
7103 j = rdev->recovery_offset;
7107 printk(KERN_INFO "md: %s of RAID array %s\n", desc, mdname(mddev));
7108 printk(KERN_INFO "md: minimum _guaranteed_ speed:"
7109 " %d KB/sec/disk.\n", speed_min(mddev));
7110 printk(KERN_INFO "md: using maximum available idle IO bandwidth "
7111 "(but not more than %d KB/sec) for %s.\n",
7112 speed_max(mddev), desc);
7114 is_mddev_idle(mddev, 1); /* this initializes IO event counters */
7117 for (m = 0; m < SYNC_MARKS; m++) {
7119 mark_cnt[m] = io_sectors;
7122 mddev->resync_mark = mark[last_mark];
7123 mddev->resync_mark_cnt = mark_cnt[last_mark];
7126 * Tune reconstruction:
7128 window = 32*(PAGE_SIZE/512);
7129 printk(KERN_INFO "md: using %dk window, over a total of %lluk.\n",
7130 window/2, (unsigned long long)max_sectors/2);
7132 atomic_set(&mddev->recovery_active, 0);
7137 "md: resuming %s of %s from checkpoint.\n",
7138 desc, mdname(mddev));
7139 mddev->curr_resync = j;
7141 mddev->curr_resync_completed = j;
7143 while (j < max_sectors) {
7148 if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
7149 ((mddev->curr_resync > mddev->curr_resync_completed &&
7150 (mddev->curr_resync - mddev->curr_resync_completed)
7151 > (max_sectors >> 4)) ||
7152 (j - mddev->curr_resync_completed)*2
7153 >= mddev->resync_max - mddev->curr_resync_completed
7155 /* time to update curr_resync_completed */
7156 wait_event(mddev->recovery_wait,
7157 atomic_read(&mddev->recovery_active) == 0);
7158 mddev->curr_resync_completed = j;
7159 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
7160 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
7163 while (j >= mddev->resync_max && !kthread_should_stop()) {
7164 /* As this condition is controlled by user-space,
7165 * we can block indefinitely, so use '_interruptible'
7166 * to avoid triggering warnings.
7168 flush_signals(current); /* just in case */
7169 wait_event_interruptible(mddev->recovery_wait,
7170 mddev->resync_max > j
7171 || kthread_should_stop());
7174 if (kthread_should_stop())
7177 sectors = mddev->pers->sync_request(mddev, j, &skipped,
7178 currspeed < speed_min(mddev));
7180 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
7184 if (!skipped) { /* actual IO requested */
7185 io_sectors += sectors;
7186 atomic_add(sectors, &mddev->recovery_active);
7189 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
7193 if (j>1) mddev->curr_resync = j;
7194 mddev->curr_mark_cnt = io_sectors;
7195 if (last_check == 0)
7196 /* this is the earliest that rebuild will be
7197 * visible in /proc/mdstat
7199 md_new_event(mddev);
7201 if (last_check + window > io_sectors || j == max_sectors)
7204 last_check = io_sectors;
7206 if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) {
7208 int next = (last_mark+1) % SYNC_MARKS;
7210 mddev->resync_mark = mark[next];
7211 mddev->resync_mark_cnt = mark_cnt[next];
7212 mark[next] = jiffies;
7213 mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active);
7218 if (kthread_should_stop())
7223 * this loop exits only if either when we are slower than
7224 * the 'hard' speed limit, or the system was IO-idle for
7226 * the system might be non-idle CPU-wise, but we only care
7227 * about not overloading the IO subsystem. (things like an
7228 * e2fsck being done on the RAID array should execute fast)
7232 currspeed = ((unsigned long)(io_sectors-mddev->resync_mark_cnt))/2
7233 /((jiffies-mddev->resync_mark)/HZ +1) +1;
7235 if (currspeed > speed_min(mddev)) {
7236 if ((currspeed > speed_max(mddev)) ||
7237 !is_mddev_idle(mddev, 0)) {
7243 printk(KERN_INFO "md: %s: %s done.\n",mdname(mddev), desc);
7245 * this also signals 'finished resyncing' to md_stop
7248 wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
7250 /* tell personality that we are finished */
7251 mddev->pers->sync_request(mddev, max_sectors, &skipped, 1);
7253 if (!test_bit(MD_RECOVERY_CHECK, &mddev->recovery) &&
7254 mddev->curr_resync > 2) {
7255 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
7256 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
7257 if (mddev->curr_resync >= mddev->recovery_cp) {
7259 "md: checkpointing %s of %s.\n",
7260 desc, mdname(mddev));
7261 mddev->recovery_cp = mddev->curr_resync;
7264 mddev->recovery_cp = MaxSector;
7266 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery))
7267 mddev->curr_resync = MaxSector;
7269 list_for_each_entry_rcu(rdev, &mddev->disks, same_set)
7270 if (rdev->raid_disk >= 0 &&
7271 mddev->delta_disks >= 0 &&
7272 !test_bit(Faulty, &rdev->flags) &&
7273 !test_bit(In_sync, &rdev->flags) &&
7274 rdev->recovery_offset < mddev->curr_resync)
7275 rdev->recovery_offset = mddev->curr_resync;
7279 set_bit(MD_CHANGE_DEVS, &mddev->flags);
7282 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
7283 /* We completed so min/max setting can be forgotten if used. */
7284 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
7285 mddev->resync_min = 0;
7286 mddev->resync_max = MaxSector;
7287 } else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
7288 mddev->resync_min = mddev->curr_resync_completed;
7289 mddev->curr_resync = 0;
7290 wake_up(&resync_wait);
7291 set_bit(MD_RECOVERY_DONE, &mddev->recovery);
7292 md_wakeup_thread(mddev->thread);
7297 * got a signal, exit.
7300 "md: md_do_sync() got signal ... exiting\n");
7301 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
7305 EXPORT_SYMBOL_GPL(md_do_sync);
7307 static int remove_and_add_spares(mddev_t *mddev)
7312 mddev->curr_resync_completed = 0;
7314 list_for_each_entry(rdev, &mddev->disks, same_set)
7315 if (rdev->raid_disk >= 0 &&
7316 !test_bit(Blocked, &rdev->flags) &&
7317 (test_bit(Faulty, &rdev->flags) ||
7318 ! test_bit(In_sync, &rdev->flags)) &&
7319 atomic_read(&rdev->nr_pending)==0) {
7320 if (mddev->pers->hot_remove_disk(
7321 mddev, rdev->raid_disk)==0) {
7322 sysfs_unlink_rdev(mddev, rdev);
7323 rdev->raid_disk = -1;
7327 if (mddev->degraded) {
7328 list_for_each_entry(rdev, &mddev->disks, same_set) {
7329 if (rdev->raid_disk >= 0 &&
7330 !test_bit(In_sync, &rdev->flags) &&
7331 !test_bit(Faulty, &rdev->flags))
7333 if (rdev->raid_disk < 0
7334 && !test_bit(Faulty, &rdev->flags)) {
7335 rdev->recovery_offset = 0;
7337 hot_add_disk(mddev, rdev) == 0) {
7338 if (sysfs_link_rdev(mddev, rdev))
7339 /* failure here is OK */;
7341 md_new_event(mddev);
7342 set_bit(MD_CHANGE_DEVS, &mddev->flags);
7351 static void reap_sync_thread(mddev_t *mddev)
7355 /* resync has finished, collect result */
7356 md_unregister_thread(&mddev->sync_thread);
7357 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
7358 !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
7360 /* activate any spares */
7361 if (mddev->pers->spare_active(mddev))
7362 sysfs_notify(&mddev->kobj, NULL,
7365 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
7366 mddev->pers->finish_reshape)
7367 mddev->pers->finish_reshape(mddev);
7368 md_update_sb(mddev, 1);
7370 /* if array is no-longer degraded, then any saved_raid_disk
7371 * information must be scrapped
7373 if (!mddev->degraded)
7374 list_for_each_entry(rdev, &mddev->disks, same_set)
7375 rdev->saved_raid_disk = -1;
7377 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
7378 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
7379 clear_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
7380 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
7381 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
7382 /* flag recovery needed just to double check */
7383 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7384 sysfs_notify_dirent_safe(mddev->sysfs_action);
7385 md_new_event(mddev);
7386 if (mddev->event_work.func)
7387 queue_work(md_misc_wq, &mddev->event_work);
7391 * This routine is regularly called by all per-raid-array threads to
7392 * deal with generic issues like resync and super-block update.
7393 * Raid personalities that don't have a thread (linear/raid0) do not
7394 * need this as they never do any recovery or update the superblock.
7396 * It does not do any resync itself, but rather "forks" off other threads
7397 * to do that as needed.
7398 * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
7399 * "->recovery" and create a thread at ->sync_thread.
7400 * When the thread finishes it sets MD_RECOVERY_DONE
7401 * and wakeups up this thread which will reap the thread and finish up.
7402 * This thread also removes any faulty devices (with nr_pending == 0).
7404 * The overall approach is:
7405 * 1/ if the superblock needs updating, update it.
7406 * 2/ If a recovery thread is running, don't do anything else.
7407 * 3/ If recovery has finished, clean up, possibly marking spares active.
7408 * 4/ If there are any faulty devices, remove them.
7409 * 5/ If array is degraded, try to add spares devices
7410 * 6/ If array has spares or is not in-sync, start a resync thread.
7412 void md_check_recovery(mddev_t *mddev)
7414 if (mddev->suspended)
7418 bitmap_daemon_work(mddev);
7420 if (signal_pending(current)) {
7421 if (mddev->pers->sync_request && !mddev->external) {
7422 printk(KERN_INFO "md: %s in immediate safe mode\n",
7424 mddev->safemode = 2;
7426 flush_signals(current);
7429 if (mddev->ro && !test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
7432 (mddev->flags & ~ (1<<MD_CHANGE_PENDING)) ||
7433 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
7434 test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
7435 (mddev->external == 0 && mddev->safemode == 1) ||
7436 (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending)
7437 && !mddev->in_sync && mddev->recovery_cp == MaxSector)
7441 if (mddev_trylock(mddev)) {
7445 /* Only thing we do on a ro array is remove
7449 list_for_each_entry(rdev, &mddev->disks, same_set)
7450 if (rdev->raid_disk >= 0 &&
7451 !test_bit(Blocked, &rdev->flags) &&
7452 test_bit(Faulty, &rdev->flags) &&
7453 atomic_read(&rdev->nr_pending)==0) {
7454 if (mddev->pers->hot_remove_disk(
7455 mddev, rdev->raid_disk)==0) {
7456 sysfs_unlink_rdev(mddev, rdev);
7457 rdev->raid_disk = -1;
7460 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7464 if (!mddev->external) {
7466 spin_lock_irq(&mddev->write_lock);
7467 if (mddev->safemode &&
7468 !atomic_read(&mddev->writes_pending) &&
7470 mddev->recovery_cp == MaxSector) {
7473 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
7475 if (mddev->safemode == 1)
7476 mddev->safemode = 0;
7477 spin_unlock_irq(&mddev->write_lock);
7479 sysfs_notify_dirent_safe(mddev->sysfs_state);
7483 md_update_sb(mddev, 0);
7485 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
7486 !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
7487 /* resync/recovery still happening */
7488 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7491 if (mddev->sync_thread) {
7492 reap_sync_thread(mddev);
7495 /* Set RUNNING before clearing NEEDED to avoid
7496 * any transients in the value of "sync_action".
7498 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
7499 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7500 /* Clear some bits that don't mean anything, but
7503 clear_bit(MD_RECOVERY_INTR, &mddev->recovery);
7504 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
7506 if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
7508 /* no recovery is running.
7509 * remove any failed drives, then
7510 * add spares if possible.
7511 * Spare are also removed and re-added, to allow
7512 * the personality to fail the re-add.
7515 if (mddev->reshape_position != MaxSector) {
7516 if (mddev->pers->check_reshape == NULL ||
7517 mddev->pers->check_reshape(mddev) != 0)
7518 /* Cannot proceed */
7520 set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
7521 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
7522 } else if ((spares = remove_and_add_spares(mddev))) {
7523 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
7524 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
7525 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
7526 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
7527 } else if (mddev->recovery_cp < MaxSector) {
7528 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
7529 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
7530 } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
7531 /* nothing to be done ... */
7534 if (mddev->pers->sync_request) {
7535 if (spares && mddev->bitmap && ! mddev->bitmap->file) {
7536 /* We are adding a device or devices to an array
7537 * which has the bitmap stored on all devices.
7538 * So make sure all bitmap pages get written
7540 bitmap_write_all(mddev->bitmap);
7542 mddev->sync_thread = md_register_thread(md_do_sync,
7545 if (!mddev->sync_thread) {
7546 printk(KERN_ERR "%s: could not start resync"
7549 /* leave the spares where they are, it shouldn't hurt */
7550 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
7551 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
7552 clear_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
7553 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
7554 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
7556 md_wakeup_thread(mddev->sync_thread);
7557 sysfs_notify_dirent_safe(mddev->sysfs_action);
7558 md_new_event(mddev);
7561 if (!mddev->sync_thread) {
7562 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
7563 if (test_and_clear_bit(MD_RECOVERY_RECOVER,
7565 if (mddev->sysfs_action)
7566 sysfs_notify_dirent_safe(mddev->sysfs_action);
7568 mddev_unlock(mddev);
7572 void md_wait_for_blocked_rdev(mdk_rdev_t *rdev, mddev_t *mddev)
7574 sysfs_notify_dirent_safe(rdev->sysfs_state);
7575 wait_event_timeout(rdev->blocked_wait,
7576 !test_bit(Blocked, &rdev->flags) &&
7577 !test_bit(BlockedBadBlocks, &rdev->flags),
7578 msecs_to_jiffies(5000));
7579 rdev_dec_pending(rdev, mddev);
7581 EXPORT_SYMBOL(md_wait_for_blocked_rdev);
7584 /* Bad block management.
7585 * We can record which blocks on each device are 'bad' and so just
7586 * fail those blocks, or that stripe, rather than the whole device.
7587 * Entries in the bad-block table are 64bits wide. This comprises:
7588 * Length of bad-range, in sectors: 0-511 for lengths 1-512
7589 * Start of bad-range, sector offset, 54 bits (allows 8 exbibytes)
7590 * A 'shift' can be set so that larger blocks are tracked and
7591 * consequently larger devices can be covered.
7592 * 'Acknowledged' flag - 1 bit. - the most significant bit.
7594 * Locking of the bad-block table uses a seqlock so md_is_badblock
7595 * might need to retry if it is very unlucky.
7596 * We will sometimes want to check for bad blocks in a bi_end_io function,
7597 * so we use the write_seqlock_irq variant.
7599 * When looking for a bad block we specify a range and want to
7600 * know if any block in the range is bad. So we binary-search
7601 * to the last range that starts at-or-before the given endpoint,
7602 * (or "before the sector after the target range")
7603 * then see if it ends after the given start.
7605 * 0 if there are no known bad blocks in the range
7606 * 1 if there are known bad block which are all acknowledged
7607 * -1 if there are bad blocks which have not yet been acknowledged in metadata.
7608 * plus the start/length of the first bad section we overlap.
7610 int md_is_badblock(struct badblocks *bb, sector_t s, int sectors,
7611 sector_t *first_bad, int *bad_sectors)
7617 sector_t target = s + sectors;
7620 if (bb->shift > 0) {
7621 /* round the start down, and the end up */
7623 target += (1<<bb->shift) - 1;
7624 target >>= bb->shift;
7625 sectors = target - s;
7627 /* 'target' is now the first block after the bad range */
7630 seq = read_seqbegin(&bb->lock);
7634 /* Binary search between lo and hi for 'target'
7635 * i.e. for the last range that starts before 'target'
7637 /* INVARIANT: ranges before 'lo' and at-or-after 'hi'
7638 * are known not to be the last range before target.
7639 * VARIANT: hi-lo is the number of possible
7640 * ranges, and decreases until it reaches 1
7642 while (hi - lo > 1) {
7643 int mid = (lo + hi) / 2;
7644 sector_t a = BB_OFFSET(p[mid]);
7646 /* This could still be the one, earlier ranges
7650 /* This and later ranges are definitely out. */
7653 /* 'lo' might be the last that started before target, but 'hi' isn't */
7655 /* need to check all range that end after 's' to see if
7656 * any are unacknowledged.
7659 BB_OFFSET(p[lo]) + BB_LEN(p[lo]) > s) {
7660 if (BB_OFFSET(p[lo]) < target) {
7661 /* starts before the end, and finishes after
7662 * the start, so they must overlap
7664 if (rv != -1 && BB_ACK(p[lo]))
7668 *first_bad = BB_OFFSET(p[lo]);
7669 *bad_sectors = BB_LEN(p[lo]);
7675 if (read_seqretry(&bb->lock, seq))
7680 EXPORT_SYMBOL_GPL(md_is_badblock);
7683 * Add a range of bad blocks to the table.
7684 * This might extend the table, or might contract it
7685 * if two adjacent ranges can be merged.
7686 * We binary-search to find the 'insertion' point, then
7687 * decide how best to handle it.
7689 static int md_set_badblocks(struct badblocks *bb, sector_t s, int sectors,
7697 /* badblocks are disabled */
7701 /* round the start down, and the end up */
7702 sector_t next = s + sectors;
7704 next += (1<<bb->shift) - 1;
7709 write_seqlock_irq(&bb->lock);
7714 /* Find the last range that starts at-or-before 's' */
7715 while (hi - lo > 1) {
7716 int mid = (lo + hi) / 2;
7717 sector_t a = BB_OFFSET(p[mid]);
7723 if (hi > lo && BB_OFFSET(p[lo]) > s)
7727 /* we found a range that might merge with the start
7730 sector_t a = BB_OFFSET(p[lo]);
7731 sector_t e = a + BB_LEN(p[lo]);
7732 int ack = BB_ACK(p[lo]);
7734 /* Yes, we can merge with a previous range */
7735 if (s == a && s + sectors >= e)
7736 /* new range covers old */
7739 ack = ack && acknowledged;
7741 if (e < s + sectors)
7743 if (e - a <= BB_MAX_LEN) {
7744 p[lo] = BB_MAKE(a, e-a, ack);
7747 /* does not all fit in one range,
7748 * make p[lo] maximal
7750 if (BB_LEN(p[lo]) != BB_MAX_LEN)
7751 p[lo] = BB_MAKE(a, BB_MAX_LEN, ack);
7757 if (sectors && hi < bb->count) {
7758 /* 'hi' points to the first range that starts after 's'.
7759 * Maybe we can merge with the start of that range */
7760 sector_t a = BB_OFFSET(p[hi]);
7761 sector_t e = a + BB_LEN(p[hi]);
7762 int ack = BB_ACK(p[hi]);
7763 if (a <= s + sectors) {
7764 /* merging is possible */
7765 if (e <= s + sectors) {
7770 ack = ack && acknowledged;
7773 if (e - a <= BB_MAX_LEN) {
7774 p[hi] = BB_MAKE(a, e-a, ack);
7777 p[hi] = BB_MAKE(a, BB_MAX_LEN, ack);
7785 if (sectors == 0 && hi < bb->count) {
7786 /* we might be able to combine lo and hi */
7787 /* Note: 's' is at the end of 'lo' */
7788 sector_t a = BB_OFFSET(p[hi]);
7789 int lolen = BB_LEN(p[lo]);
7790 int hilen = BB_LEN(p[hi]);
7791 int newlen = lolen + hilen - (s - a);
7792 if (s >= a && newlen < BB_MAX_LEN) {
7793 /* yes, we can combine them */
7794 int ack = BB_ACK(p[lo]) && BB_ACK(p[hi]);
7795 p[lo] = BB_MAKE(BB_OFFSET(p[lo]), newlen, ack);
7796 memmove(p + hi, p + hi + 1,
7797 (bb->count - hi - 1) * 8);
7802 /* didn't merge (it all).
7803 * Need to add a range just before 'hi' */
7804 if (bb->count >= MD_MAX_BADBLOCKS) {
7805 /* No room for more */
7809 int this_sectors = sectors;
7810 memmove(p + hi + 1, p + hi,
7811 (bb->count - hi) * 8);
7814 if (this_sectors > BB_MAX_LEN)
7815 this_sectors = BB_MAX_LEN;
7816 p[hi] = BB_MAKE(s, this_sectors, acknowledged);
7817 sectors -= this_sectors;
7824 bb->unacked_exist = 1;
7825 write_sequnlock_irq(&bb->lock);
7830 int rdev_set_badblocks(mdk_rdev_t *rdev, sector_t s, int sectors,
7833 int rv = md_set_badblocks(&rdev->badblocks,
7834 s + rdev->data_offset, sectors, acknowledged);
7836 /* Make sure they get written out promptly */
7837 set_bit(MD_CHANGE_CLEAN, &rdev->mddev->flags);
7838 md_wakeup_thread(rdev->mddev->thread);
7842 EXPORT_SYMBOL_GPL(rdev_set_badblocks);
7845 * Remove a range of bad blocks from the table.
7846 * This may involve extending the table if we spilt a region,
7847 * but it must not fail. So if the table becomes full, we just
7848 * drop the remove request.
7850 static int md_clear_badblocks(struct badblocks *bb, sector_t s, int sectors)
7854 sector_t target = s + sectors;
7857 if (bb->shift > 0) {
7858 /* When clearing we round the start up and the end down.
7859 * This should not matter as the shift should align with
7860 * the block size and no rounding should ever be needed.
7861 * However it is better the think a block is bad when it
7862 * isn't than to think a block is not bad when it is.
7864 s += (1<<bb->shift) - 1;
7866 target >>= bb->shift;
7867 sectors = target - s;
7870 write_seqlock_irq(&bb->lock);
7875 /* Find the last range that starts before 'target' */
7876 while (hi - lo > 1) {
7877 int mid = (lo + hi) / 2;
7878 sector_t a = BB_OFFSET(p[mid]);
7885 /* p[lo] is the last range that could overlap the
7886 * current range. Earlier ranges could also overlap,
7887 * but only this one can overlap the end of the range.
7889 if (BB_OFFSET(p[lo]) + BB_LEN(p[lo]) > target) {
7890 /* Partial overlap, leave the tail of this range */
7891 int ack = BB_ACK(p[lo]);
7892 sector_t a = BB_OFFSET(p[lo]);
7893 sector_t end = a + BB_LEN(p[lo]);
7896 /* we need to split this range */
7897 if (bb->count >= MD_MAX_BADBLOCKS) {
7901 memmove(p+lo+1, p+lo, (bb->count - lo) * 8);
7903 p[lo] = BB_MAKE(a, s-a, ack);
7906 p[lo] = BB_MAKE(target, end - target, ack);
7907 /* there is no longer an overlap */
7912 BB_OFFSET(p[lo]) + BB_LEN(p[lo]) > s) {
7913 /* This range does overlap */
7914 if (BB_OFFSET(p[lo]) < s) {
7915 /* Keep the early parts of this range. */
7916 int ack = BB_ACK(p[lo]);
7917 sector_t start = BB_OFFSET(p[lo]);
7918 p[lo] = BB_MAKE(start, s - start, ack);
7919 /* now low doesn't overlap, so.. */
7924 /* 'lo' is strictly before, 'hi' is strictly after,
7925 * anything between needs to be discarded
7928 memmove(p+lo+1, p+hi, (bb->count - hi) * 8);
7929 bb->count -= (hi - lo - 1);
7935 write_sequnlock_irq(&bb->lock);
7939 int rdev_clear_badblocks(mdk_rdev_t *rdev, sector_t s, int sectors)
7941 return md_clear_badblocks(&rdev->badblocks,
7942 s + rdev->data_offset,
7945 EXPORT_SYMBOL_GPL(rdev_clear_badblocks);
7948 * Acknowledge all bad blocks in a list.
7949 * This only succeeds if ->changed is clear. It is used by
7950 * in-kernel metadata updates
7952 void md_ack_all_badblocks(struct badblocks *bb)
7954 if (bb->page == NULL || bb->changed)
7955 /* no point even trying */
7957 write_seqlock_irq(&bb->lock);
7959 if (bb->changed == 0) {
7962 for (i = 0; i < bb->count ; i++) {
7963 if (!BB_ACK(p[i])) {
7964 sector_t start = BB_OFFSET(p[i]);
7965 int len = BB_LEN(p[i]);
7966 p[i] = BB_MAKE(start, len, 1);
7969 bb->unacked_exist = 0;
7971 write_sequnlock_irq(&bb->lock);
7973 EXPORT_SYMBOL_GPL(md_ack_all_badblocks);
7975 /* sysfs access to bad-blocks list.
7976 * We present two files.
7977 * 'bad-blocks' lists sector numbers and lengths of ranges that
7978 * are recorded as bad. The list is truncated to fit within
7979 * the one-page limit of sysfs.
7980 * Writing "sector length" to this file adds an acknowledged
7982 * 'unacknowledged-bad-blocks' lists bad blocks that have not yet
7983 * been acknowledged. Writing to this file adds bad blocks
7984 * without acknowledging them. This is largely for testing.
7988 badblocks_show(struct badblocks *bb, char *page, int unack)
7999 seq = read_seqbegin(&bb->lock);
8004 while (len < PAGE_SIZE && i < bb->count) {
8005 sector_t s = BB_OFFSET(p[i]);
8006 unsigned int length = BB_LEN(p[i]);
8007 int ack = BB_ACK(p[i]);
8013 len += snprintf(page+len, PAGE_SIZE-len, "%llu %u\n",
8014 (unsigned long long)s << bb->shift,
8015 length << bb->shift);
8017 if (unack && len == 0)
8018 bb->unacked_exist = 0;
8020 if (read_seqretry(&bb->lock, seq))
8029 badblocks_store(struct badblocks *bb, const char *page, size_t len, int unack)
8031 unsigned long long sector;
8035 /* Allow clearing via sysfs *only* for testing/debugging.
8036 * Normally only a successful write may clear a badblock
8039 if (page[0] == '-') {
8043 #endif /* DO_DEBUG */
8045 switch (sscanf(page, "%llu %d%c", §or, &length, &newline)) {
8047 if (newline != '\n')
8059 md_clear_badblocks(bb, sector, length);
8062 #endif /* DO_DEBUG */
8063 if (md_set_badblocks(bb, sector, length, !unack))
8069 static int md_notify_reboot(struct notifier_block *this,
8070 unsigned long code, void *x)
8072 struct list_head *tmp;
8075 if ((code == SYS_DOWN) || (code == SYS_HALT) || (code == SYS_POWER_OFF)) {
8077 printk(KERN_INFO "md: stopping all md devices.\n");
8079 for_each_mddev(mddev, tmp)
8080 if (mddev_trylock(mddev)) {
8081 /* Force a switch to readonly even array
8082 * appears to still be in use. Hence
8085 md_set_readonly(mddev, 100);
8086 mddev_unlock(mddev);
8089 * certain more exotic SCSI devices are known to be
8090 * volatile wrt too early system reboots. While the
8091 * right place to handle this issue is the given
8092 * driver, we do want to have a safe RAID driver ...
8099 static struct notifier_block md_notifier = {
8100 .notifier_call = md_notify_reboot,
8102 .priority = INT_MAX, /* before any real devices */
8105 static void md_geninit(void)
8107 dprintk("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
8109 proc_create("mdstat", S_IRUGO, NULL, &md_seq_fops);
8112 static int __init md_init(void)
8116 md_wq = alloc_workqueue("md", WQ_MEM_RECLAIM, 0);
8120 md_misc_wq = alloc_workqueue("md_misc", 0, 0);
8124 if ((ret = register_blkdev(MD_MAJOR, "md")) < 0)
8127 if ((ret = register_blkdev(0, "mdp")) < 0)
8131 blk_register_region(MKDEV(MD_MAJOR, 0), 1UL<<MINORBITS, THIS_MODULE,
8132 md_probe, NULL, NULL);
8133 blk_register_region(MKDEV(mdp_major, 0), 1UL<<MINORBITS, THIS_MODULE,
8134 md_probe, NULL, NULL);
8136 register_reboot_notifier(&md_notifier);
8137 raid_table_header = register_sysctl_table(raid_root_table);
8143 unregister_blkdev(MD_MAJOR, "md");
8145 destroy_workqueue(md_misc_wq);
8147 destroy_workqueue(md_wq);
8155 * Searches all registered partitions for autorun RAID arrays
8159 static LIST_HEAD(all_detected_devices);
8160 struct detected_devices_node {
8161 struct list_head list;
8165 void md_autodetect_dev(dev_t dev)
8167 struct detected_devices_node *node_detected_dev;
8169 node_detected_dev = kzalloc(sizeof(*node_detected_dev), GFP_KERNEL);
8170 if (node_detected_dev) {
8171 node_detected_dev->dev = dev;
8172 list_add_tail(&node_detected_dev->list, &all_detected_devices);
8174 printk(KERN_CRIT "md: md_autodetect_dev: kzalloc failed"
8175 ", skipping dev(%d,%d)\n", MAJOR(dev), MINOR(dev));
8180 static void autostart_arrays(int part)
8183 struct detected_devices_node *node_detected_dev;
8185 int i_scanned, i_passed;
8190 printk(KERN_INFO "md: Autodetecting RAID arrays.\n");
8192 while (!list_empty(&all_detected_devices) && i_scanned < INT_MAX) {
8194 node_detected_dev = list_entry(all_detected_devices.next,
8195 struct detected_devices_node, list);
8196 list_del(&node_detected_dev->list);
8197 dev = node_detected_dev->dev;
8198 kfree(node_detected_dev);
8199 rdev = md_import_device(dev,0, 90);
8203 if (test_bit(Faulty, &rdev->flags)) {
8207 set_bit(AutoDetected, &rdev->flags);
8208 list_add(&rdev->same_set, &pending_raid_disks);
8212 printk(KERN_INFO "md: Scanned %d and added %d devices.\n",
8213 i_scanned, i_passed);
8215 autorun_devices(part);
8218 #endif /* !MODULE */
8220 static __exit void md_exit(void)
8223 struct list_head *tmp;
8225 blk_unregister_region(MKDEV(MD_MAJOR,0), 1U << MINORBITS);
8226 blk_unregister_region(MKDEV(mdp_major,0), 1U << MINORBITS);
8228 unregister_blkdev(MD_MAJOR,"md");
8229 unregister_blkdev(mdp_major, "mdp");
8230 unregister_reboot_notifier(&md_notifier);
8231 unregister_sysctl_table(raid_table_header);
8232 remove_proc_entry("mdstat", NULL);
8233 for_each_mddev(mddev, tmp) {
8234 export_array(mddev);
8235 mddev->hold_active = 0;
8237 destroy_workqueue(md_misc_wq);
8238 destroy_workqueue(md_wq);
8241 subsys_initcall(md_init);
8242 module_exit(md_exit)
8244 static int get_ro(char *buffer, struct kernel_param *kp)
8246 return sprintf(buffer, "%d", start_readonly);
8248 static int set_ro(const char *val, struct kernel_param *kp)
8251 int num = simple_strtoul(val, &e, 10);
8252 if (*val && (*e == '\0' || *e == '\n')) {
8253 start_readonly = num;
8259 module_param_call(start_ro, set_ro, get_ro, NULL, S_IRUSR|S_IWUSR);
8260 module_param(start_dirty_degraded, int, S_IRUGO|S_IWUSR);
8262 module_param_call(new_array, add_named_array, NULL, NULL, S_IWUSR);
8264 EXPORT_SYMBOL(register_md_personality);
8265 EXPORT_SYMBOL(unregister_md_personality);
8266 EXPORT_SYMBOL(md_error);
8267 EXPORT_SYMBOL(md_done_sync);
8268 EXPORT_SYMBOL(md_write_start);
8269 EXPORT_SYMBOL(md_write_end);
8270 EXPORT_SYMBOL(md_register_thread);
8271 EXPORT_SYMBOL(md_unregister_thread);
8272 EXPORT_SYMBOL(md_wakeup_thread);
8273 EXPORT_SYMBOL(md_check_recovery);
8274 MODULE_LICENSE("GPL");
8275 MODULE_DESCRIPTION("MD RAID framework");
8277 MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR);