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/badblocks.h>
38 #include <linux/sysctl.h>
39 #include <linux/seq_file.h>
41 #include <linux/poll.h>
42 #include <linux/ctype.h>
43 #include <linux/string.h>
44 #include <linux/hdreg.h>
45 #include <linux/proc_fs.h>
46 #include <linux/random.h>
47 #include <linux/module.h>
48 #include <linux/reboot.h>
49 #include <linux/file.h>
50 #include <linux/compat.h>
51 #include <linux/delay.h>
52 #include <linux/raid/md_p.h>
53 #include <linux/raid/md_u.h>
54 #include <linux/slab.h>
57 #include "md-cluster.h"
60 static void autostart_arrays(int part);
63 /* pers_list is a list of registered personalities protected
65 * pers_lock does extra service to protect accesses to
66 * mddev->thread when the mutex cannot be held.
68 static LIST_HEAD(pers_list);
69 static DEFINE_SPINLOCK(pers_lock);
71 struct md_cluster_operations *md_cluster_ops;
72 EXPORT_SYMBOL(md_cluster_ops);
73 struct module *md_cluster_mod;
74 EXPORT_SYMBOL(md_cluster_mod);
76 static DECLARE_WAIT_QUEUE_HEAD(resync_wait);
77 static struct workqueue_struct *md_wq;
78 static struct workqueue_struct *md_misc_wq;
80 static int remove_and_add_spares(struct mddev *mddev,
81 struct md_rdev *this);
82 static void mddev_detach(struct mddev *mddev);
85 * Default number of read corrections we'll attempt on an rdev
86 * before ejecting it from the array. We divide the read error
87 * count by 2 for every hour elapsed between read errors.
89 #define MD_DEFAULT_MAX_CORRECTED_READ_ERRORS 20
91 * Current RAID-1,4,5 parallel reconstruction 'guaranteed speed limit'
92 * is 1000 KB/sec, so the extra system load does not show up that much.
93 * Increase it if you want to have more _guaranteed_ speed. Note that
94 * the RAID driver will use the maximum available bandwidth if the IO
95 * subsystem is idle. There is also an 'absolute maximum' reconstruction
96 * speed limit - in case reconstruction slows down your system despite
99 * you can change it via /proc/sys/dev/raid/speed_limit_min and _max.
100 * or /sys/block/mdX/md/sync_speed_{min,max}
103 static int sysctl_speed_limit_min = 1000;
104 static int sysctl_speed_limit_max = 200000;
105 static inline int speed_min(struct mddev *mddev)
107 return mddev->sync_speed_min ?
108 mddev->sync_speed_min : sysctl_speed_limit_min;
111 static inline int speed_max(struct mddev *mddev)
113 return mddev->sync_speed_max ?
114 mddev->sync_speed_max : sysctl_speed_limit_max;
117 static struct ctl_table_header *raid_table_header;
119 static struct ctl_table raid_table[] = {
121 .procname = "speed_limit_min",
122 .data = &sysctl_speed_limit_min,
123 .maxlen = sizeof(int),
124 .mode = S_IRUGO|S_IWUSR,
125 .proc_handler = proc_dointvec,
128 .procname = "speed_limit_max",
129 .data = &sysctl_speed_limit_max,
130 .maxlen = sizeof(int),
131 .mode = S_IRUGO|S_IWUSR,
132 .proc_handler = proc_dointvec,
137 static struct ctl_table raid_dir_table[] = {
141 .mode = S_IRUGO|S_IXUGO,
147 static struct ctl_table raid_root_table[] = {
152 .child = raid_dir_table,
157 static const struct block_device_operations md_fops;
159 static int start_readonly;
162 * like bio_clone, but with a local bio set
165 struct bio *bio_alloc_mddev(gfp_t gfp_mask, int nr_iovecs,
170 if (!mddev || !mddev->bio_set)
171 return bio_alloc(gfp_mask, nr_iovecs);
173 b = bio_alloc_bioset(gfp_mask, nr_iovecs, mddev->bio_set);
178 EXPORT_SYMBOL_GPL(bio_alloc_mddev);
180 struct bio *bio_clone_mddev(struct bio *bio, gfp_t gfp_mask,
183 if (!mddev || !mddev->bio_set)
184 return bio_clone(bio, gfp_mask);
186 return bio_clone_bioset(bio, gfp_mask, mddev->bio_set);
188 EXPORT_SYMBOL_GPL(bio_clone_mddev);
191 * We have a system wide 'event count' that is incremented
192 * on any 'interesting' event, and readers of /proc/mdstat
193 * can use 'poll' or 'select' to find out when the event
197 * start array, stop array, error, add device, remove device,
198 * start build, activate spare
200 static DECLARE_WAIT_QUEUE_HEAD(md_event_waiters);
201 static atomic_t md_event_count;
202 void md_new_event(struct mddev *mddev)
204 atomic_inc(&md_event_count);
205 wake_up(&md_event_waiters);
207 EXPORT_SYMBOL_GPL(md_new_event);
210 * Enables to iterate over all existing md arrays
211 * all_mddevs_lock protects this list.
213 static LIST_HEAD(all_mddevs);
214 static DEFINE_SPINLOCK(all_mddevs_lock);
217 * iterates through all used mddevs in the system.
218 * We take care to grab the all_mddevs_lock whenever navigating
219 * the list, and to always hold a refcount when unlocked.
220 * Any code which breaks out of this loop while own
221 * a reference to the current mddev and must mddev_put it.
223 #define for_each_mddev(_mddev,_tmp) \
225 for (({ spin_lock(&all_mddevs_lock); \
226 _tmp = all_mddevs.next; \
228 ({ if (_tmp != &all_mddevs) \
229 mddev_get(list_entry(_tmp, struct mddev, all_mddevs));\
230 spin_unlock(&all_mddevs_lock); \
231 if (_mddev) mddev_put(_mddev); \
232 _mddev = list_entry(_tmp, struct mddev, all_mddevs); \
233 _tmp != &all_mddevs;}); \
234 ({ spin_lock(&all_mddevs_lock); \
235 _tmp = _tmp->next;}) \
238 /* Rather than calling directly into the personality make_request function,
239 * IO requests come here first so that we can check if the device is
240 * being suspended pending a reconfiguration.
241 * We hold a refcount over the call to ->make_request. By the time that
242 * call has finished, the bio has been linked into some internal structure
243 * and so is visible to ->quiesce(), so we don't need the refcount any more.
245 static blk_qc_t md_make_request(struct request_queue *q, struct bio *bio)
247 const int rw = bio_data_dir(bio);
248 struct mddev *mddev = q->queuedata;
249 unsigned int sectors;
252 blk_queue_split(q, &bio, q->bio_split);
254 if (mddev == NULL || mddev->pers == NULL) {
256 return BLK_QC_T_NONE;
258 if (mddev->ro == 1 && unlikely(rw == WRITE)) {
259 if (bio_sectors(bio) != 0)
260 bio->bi_error = -EROFS;
262 return BLK_QC_T_NONE;
264 smp_rmb(); /* Ensure implications of 'active' are visible */
266 if (mddev->suspended) {
269 prepare_to_wait(&mddev->sb_wait, &__wait,
270 TASK_UNINTERRUPTIBLE);
271 if (!mddev->suspended)
277 finish_wait(&mddev->sb_wait, &__wait);
279 atomic_inc(&mddev->active_io);
283 * save the sectors now since our bio can
284 * go away inside make_request
286 sectors = bio_sectors(bio);
287 mddev->pers->make_request(mddev, bio);
289 cpu = part_stat_lock();
290 part_stat_inc(cpu, &mddev->gendisk->part0, ios[rw]);
291 part_stat_add(cpu, &mddev->gendisk->part0, sectors[rw], sectors);
294 if (atomic_dec_and_test(&mddev->active_io) && mddev->suspended)
295 wake_up(&mddev->sb_wait);
297 return BLK_QC_T_NONE;
300 /* mddev_suspend makes sure no new requests are submitted
301 * to the device, and that any requests that have been submitted
302 * are completely handled.
303 * Once mddev_detach() is called and completes, the module will be
306 void mddev_suspend(struct mddev *mddev)
308 if (mddev->suspended++)
311 wait_event(mddev->sb_wait, atomic_read(&mddev->active_io) == 0);
312 mddev->pers->quiesce(mddev, 1);
314 del_timer_sync(&mddev->safemode_timer);
316 EXPORT_SYMBOL_GPL(mddev_suspend);
318 void mddev_resume(struct mddev *mddev)
320 if (--mddev->suspended)
322 wake_up(&mddev->sb_wait);
323 mddev->pers->quiesce(mddev, 0);
325 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
326 md_wakeup_thread(mddev->thread);
327 md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
329 EXPORT_SYMBOL_GPL(mddev_resume);
331 int mddev_congested(struct mddev *mddev, int bits)
333 struct md_personality *pers = mddev->pers;
337 if (mddev->suspended)
339 else if (pers && pers->congested)
340 ret = pers->congested(mddev, bits);
344 EXPORT_SYMBOL_GPL(mddev_congested);
345 static int md_congested(void *data, int bits)
347 struct mddev *mddev = data;
348 return mddev_congested(mddev, bits);
352 * Generic flush handling for md
355 static void md_end_flush(struct bio *bio)
357 struct md_rdev *rdev = bio->bi_private;
358 struct mddev *mddev = rdev->mddev;
360 rdev_dec_pending(rdev, mddev);
362 if (atomic_dec_and_test(&mddev->flush_pending)) {
363 /* The pre-request flush has finished */
364 queue_work(md_wq, &mddev->flush_work);
369 static void md_submit_flush_data(struct work_struct *ws);
371 static void submit_flushes(struct work_struct *ws)
373 struct mddev *mddev = container_of(ws, struct mddev, flush_work);
374 struct md_rdev *rdev;
376 INIT_WORK(&mddev->flush_work, md_submit_flush_data);
377 atomic_set(&mddev->flush_pending, 1);
379 rdev_for_each_rcu(rdev, mddev)
380 if (rdev->raid_disk >= 0 &&
381 !test_bit(Faulty, &rdev->flags)) {
382 /* Take two references, one is dropped
383 * when request finishes, one after
384 * we reclaim rcu_read_lock
387 atomic_inc(&rdev->nr_pending);
388 atomic_inc(&rdev->nr_pending);
390 bi = bio_alloc_mddev(GFP_NOIO, 0, mddev);
391 bi->bi_end_io = md_end_flush;
392 bi->bi_private = rdev;
393 bi->bi_bdev = rdev->bdev;
394 atomic_inc(&mddev->flush_pending);
395 submit_bio(WRITE_FLUSH, bi);
397 rdev_dec_pending(rdev, mddev);
400 if (atomic_dec_and_test(&mddev->flush_pending))
401 queue_work(md_wq, &mddev->flush_work);
404 static void md_submit_flush_data(struct work_struct *ws)
406 struct mddev *mddev = container_of(ws, struct mddev, flush_work);
407 struct bio *bio = mddev->flush_bio;
409 if (bio->bi_iter.bi_size == 0)
410 /* an empty barrier - all done */
413 bio->bi_rw &= ~REQ_FLUSH;
414 mddev->pers->make_request(mddev, bio);
417 mddev->flush_bio = NULL;
418 wake_up(&mddev->sb_wait);
421 void md_flush_request(struct mddev *mddev, struct bio *bio)
423 spin_lock_irq(&mddev->lock);
424 wait_event_lock_irq(mddev->sb_wait,
427 mddev->flush_bio = bio;
428 spin_unlock_irq(&mddev->lock);
430 INIT_WORK(&mddev->flush_work, submit_flushes);
431 queue_work(md_wq, &mddev->flush_work);
433 EXPORT_SYMBOL(md_flush_request);
435 void md_unplug(struct blk_plug_cb *cb, bool from_schedule)
437 struct mddev *mddev = cb->data;
438 md_wakeup_thread(mddev->thread);
441 EXPORT_SYMBOL(md_unplug);
443 static inline struct mddev *mddev_get(struct mddev *mddev)
445 atomic_inc(&mddev->active);
449 static void mddev_delayed_delete(struct work_struct *ws);
451 static void mddev_put(struct mddev *mddev)
453 struct bio_set *bs = NULL;
455 if (!atomic_dec_and_lock(&mddev->active, &all_mddevs_lock))
457 if (!mddev->raid_disks && list_empty(&mddev->disks) &&
458 mddev->ctime == 0 && !mddev->hold_active) {
459 /* Array is not configured at all, and not held active,
461 list_del_init(&mddev->all_mddevs);
463 mddev->bio_set = NULL;
464 if (mddev->gendisk) {
465 /* We did a probe so need to clean up. Call
466 * queue_work inside the spinlock so that
467 * flush_workqueue() after mddev_find will
468 * succeed in waiting for the work to be done.
470 INIT_WORK(&mddev->del_work, mddev_delayed_delete);
471 queue_work(md_misc_wq, &mddev->del_work);
475 spin_unlock(&all_mddevs_lock);
480 static void md_safemode_timeout(unsigned long data);
482 void mddev_init(struct mddev *mddev)
484 mutex_init(&mddev->open_mutex);
485 mutex_init(&mddev->reconfig_mutex);
486 mutex_init(&mddev->bitmap_info.mutex);
487 INIT_LIST_HEAD(&mddev->disks);
488 INIT_LIST_HEAD(&mddev->all_mddevs);
489 setup_timer(&mddev->safemode_timer, md_safemode_timeout,
490 (unsigned long) mddev);
491 atomic_set(&mddev->active, 1);
492 atomic_set(&mddev->openers, 0);
493 atomic_set(&mddev->active_io, 0);
494 spin_lock_init(&mddev->lock);
495 atomic_set(&mddev->flush_pending, 0);
496 init_waitqueue_head(&mddev->sb_wait);
497 init_waitqueue_head(&mddev->recovery_wait);
498 mddev->reshape_position = MaxSector;
499 mddev->reshape_backwards = 0;
500 mddev->last_sync_action = "none";
501 mddev->resync_min = 0;
502 mddev->resync_max = MaxSector;
503 mddev->level = LEVEL_NONE;
505 EXPORT_SYMBOL_GPL(mddev_init);
507 static struct mddev *mddev_find(dev_t unit)
509 struct mddev *mddev, *new = NULL;
511 if (unit && MAJOR(unit) != MD_MAJOR)
512 unit &= ~((1<<MdpMinorShift)-1);
515 spin_lock(&all_mddevs_lock);
518 list_for_each_entry(mddev, &all_mddevs, all_mddevs)
519 if (mddev->unit == unit) {
521 spin_unlock(&all_mddevs_lock);
527 list_add(&new->all_mddevs, &all_mddevs);
528 spin_unlock(&all_mddevs_lock);
529 new->hold_active = UNTIL_IOCTL;
533 /* find an unused unit number */
534 static int next_minor = 512;
535 int start = next_minor;
539 dev = MKDEV(MD_MAJOR, next_minor);
541 if (next_minor > MINORMASK)
543 if (next_minor == start) {
544 /* Oh dear, all in use. */
545 spin_unlock(&all_mddevs_lock);
551 list_for_each_entry(mddev, &all_mddevs, all_mddevs)
552 if (mddev->unit == dev) {
558 new->md_minor = MINOR(dev);
559 new->hold_active = UNTIL_STOP;
560 list_add(&new->all_mddevs, &all_mddevs);
561 spin_unlock(&all_mddevs_lock);
564 spin_unlock(&all_mddevs_lock);
566 new = kzalloc(sizeof(*new), GFP_KERNEL);
571 if (MAJOR(unit) == MD_MAJOR)
572 new->md_minor = MINOR(unit);
574 new->md_minor = MINOR(unit) >> MdpMinorShift;
581 static struct attribute_group md_redundancy_group;
583 void mddev_unlock(struct mddev *mddev)
585 if (mddev->to_remove) {
586 /* These cannot be removed under reconfig_mutex as
587 * an access to the files will try to take reconfig_mutex
588 * while holding the file unremovable, which leads to
590 * So hold set sysfs_active while the remove in happeing,
591 * and anything else which might set ->to_remove or my
592 * otherwise change the sysfs namespace will fail with
593 * -EBUSY if sysfs_active is still set.
594 * We set sysfs_active under reconfig_mutex and elsewhere
595 * test it under the same mutex to ensure its correct value
598 struct attribute_group *to_remove = mddev->to_remove;
599 mddev->to_remove = NULL;
600 mddev->sysfs_active = 1;
601 mutex_unlock(&mddev->reconfig_mutex);
603 if (mddev->kobj.sd) {
604 if (to_remove != &md_redundancy_group)
605 sysfs_remove_group(&mddev->kobj, to_remove);
606 if (mddev->pers == NULL ||
607 mddev->pers->sync_request == NULL) {
608 sysfs_remove_group(&mddev->kobj, &md_redundancy_group);
609 if (mddev->sysfs_action)
610 sysfs_put(mddev->sysfs_action);
611 mddev->sysfs_action = NULL;
614 mddev->sysfs_active = 0;
616 mutex_unlock(&mddev->reconfig_mutex);
618 /* As we've dropped the mutex we need a spinlock to
619 * make sure the thread doesn't disappear
621 spin_lock(&pers_lock);
622 md_wakeup_thread(mddev->thread);
623 spin_unlock(&pers_lock);
625 EXPORT_SYMBOL_GPL(mddev_unlock);
627 struct md_rdev *md_find_rdev_nr_rcu(struct mddev *mddev, int nr)
629 struct md_rdev *rdev;
631 rdev_for_each_rcu(rdev, mddev)
632 if (rdev->desc_nr == nr)
637 EXPORT_SYMBOL_GPL(md_find_rdev_nr_rcu);
639 static struct md_rdev *find_rdev(struct mddev *mddev, dev_t dev)
641 struct md_rdev *rdev;
643 rdev_for_each(rdev, mddev)
644 if (rdev->bdev->bd_dev == dev)
650 static struct md_rdev *find_rdev_rcu(struct mddev *mddev, dev_t dev)
652 struct md_rdev *rdev;
654 rdev_for_each_rcu(rdev, mddev)
655 if (rdev->bdev->bd_dev == dev)
661 static struct md_personality *find_pers(int level, char *clevel)
663 struct md_personality *pers;
664 list_for_each_entry(pers, &pers_list, list) {
665 if (level != LEVEL_NONE && pers->level == level)
667 if (strcmp(pers->name, clevel)==0)
673 /* return the offset of the super block in 512byte sectors */
674 static inline sector_t calc_dev_sboffset(struct md_rdev *rdev)
676 sector_t num_sectors = i_size_read(rdev->bdev->bd_inode) / 512;
677 return MD_NEW_SIZE_SECTORS(num_sectors);
680 static int alloc_disk_sb(struct md_rdev *rdev)
682 rdev->sb_page = alloc_page(GFP_KERNEL);
683 if (!rdev->sb_page) {
684 printk(KERN_ALERT "md: out of memory.\n");
691 void md_rdev_clear(struct md_rdev *rdev)
694 put_page(rdev->sb_page);
696 rdev->sb_page = NULL;
701 put_page(rdev->bb_page);
702 rdev->bb_page = NULL;
704 badblocks_exit(&rdev->badblocks);
706 EXPORT_SYMBOL_GPL(md_rdev_clear);
708 static void super_written(struct bio *bio)
710 struct md_rdev *rdev = bio->bi_private;
711 struct mddev *mddev = rdev->mddev;
714 printk("md: super_written gets error=%d\n", bio->bi_error);
715 md_error(mddev, rdev);
718 if (atomic_dec_and_test(&mddev->pending_writes))
719 wake_up(&mddev->sb_wait);
723 void md_super_write(struct mddev *mddev, struct md_rdev *rdev,
724 sector_t sector, int size, struct page *page)
726 /* write first size bytes of page to sector of rdev
727 * Increment mddev->pending_writes before returning
728 * and decrement it on completion, waking up sb_wait
729 * if zero is reached.
730 * If an error occurred, call md_error
732 struct bio *bio = bio_alloc_mddev(GFP_NOIO, 1, mddev);
734 bio->bi_bdev = rdev->meta_bdev ? rdev->meta_bdev : rdev->bdev;
735 bio->bi_iter.bi_sector = sector;
736 bio_add_page(bio, page, size, 0);
737 bio->bi_private = rdev;
738 bio->bi_end_io = super_written;
740 atomic_inc(&mddev->pending_writes);
741 submit_bio(WRITE_FLUSH_FUA, bio);
744 void md_super_wait(struct mddev *mddev)
746 /* wait for all superblock writes that were scheduled to complete */
747 wait_event(mddev->sb_wait, atomic_read(&mddev->pending_writes)==0);
750 int sync_page_io(struct md_rdev *rdev, sector_t sector, int size,
751 struct page *page, int rw, bool metadata_op)
753 struct bio *bio = bio_alloc_mddev(GFP_NOIO, 1, rdev->mddev);
756 bio->bi_bdev = (metadata_op && rdev->meta_bdev) ?
757 rdev->meta_bdev : rdev->bdev;
759 bio->bi_iter.bi_sector = sector + rdev->sb_start;
760 else if (rdev->mddev->reshape_position != MaxSector &&
761 (rdev->mddev->reshape_backwards ==
762 (sector >= rdev->mddev->reshape_position)))
763 bio->bi_iter.bi_sector = sector + rdev->new_data_offset;
765 bio->bi_iter.bi_sector = sector + rdev->data_offset;
766 bio_add_page(bio, page, size, 0);
767 submit_bio_wait(rw, bio);
769 ret = !bio->bi_error;
773 EXPORT_SYMBOL_GPL(sync_page_io);
775 static int read_disk_sb(struct md_rdev *rdev, int size)
777 char b[BDEVNAME_SIZE];
782 if (!sync_page_io(rdev, 0, size, rdev->sb_page, READ, true))
788 printk(KERN_WARNING "md: disabled device %s, could not read superblock.\n",
789 bdevname(rdev->bdev,b));
793 static int uuid_equal(mdp_super_t *sb1, mdp_super_t *sb2)
795 return sb1->set_uuid0 == sb2->set_uuid0 &&
796 sb1->set_uuid1 == sb2->set_uuid1 &&
797 sb1->set_uuid2 == sb2->set_uuid2 &&
798 sb1->set_uuid3 == sb2->set_uuid3;
801 static int sb_equal(mdp_super_t *sb1, mdp_super_t *sb2)
804 mdp_super_t *tmp1, *tmp2;
806 tmp1 = kmalloc(sizeof(*tmp1),GFP_KERNEL);
807 tmp2 = kmalloc(sizeof(*tmp2),GFP_KERNEL);
809 if (!tmp1 || !tmp2) {
811 printk(KERN_INFO "md.c sb_equal(): failed to allocate memory!\n");
819 * nr_disks is not constant
824 ret = (memcmp(tmp1, tmp2, MD_SB_GENERIC_CONSTANT_WORDS * 4) == 0);
831 static u32 md_csum_fold(u32 csum)
833 csum = (csum & 0xffff) + (csum >> 16);
834 return (csum & 0xffff) + (csum >> 16);
837 static unsigned int calc_sb_csum(mdp_super_t *sb)
840 u32 *sb32 = (u32*)sb;
842 unsigned int disk_csum, csum;
844 disk_csum = sb->sb_csum;
847 for (i = 0; i < MD_SB_BYTES/4 ; i++)
849 csum = (newcsum & 0xffffffff) + (newcsum>>32);
852 /* This used to use csum_partial, which was wrong for several
853 * reasons including that different results are returned on
854 * different architectures. It isn't critical that we get exactly
855 * the same return value as before (we always csum_fold before
856 * testing, and that removes any differences). However as we
857 * know that csum_partial always returned a 16bit value on
858 * alphas, do a fold to maximise conformity to previous behaviour.
860 sb->sb_csum = md_csum_fold(disk_csum);
862 sb->sb_csum = disk_csum;
868 * Handle superblock details.
869 * We want to be able to handle multiple superblock formats
870 * so we have a common interface to them all, and an array of
871 * different handlers.
872 * We rely on user-space to write the initial superblock, and support
873 * reading and updating of superblocks.
874 * Interface methods are:
875 * int load_super(struct md_rdev *dev, struct md_rdev *refdev, int minor_version)
876 * loads and validates a superblock on dev.
877 * if refdev != NULL, compare superblocks on both devices
879 * 0 - dev has a superblock that is compatible with refdev
880 * 1 - dev has a superblock that is compatible and newer than refdev
881 * so dev should be used as the refdev in future
882 * -EINVAL superblock incompatible or invalid
883 * -othererror e.g. -EIO
885 * int validate_super(struct mddev *mddev, struct md_rdev *dev)
886 * Verify that dev is acceptable into mddev.
887 * The first time, mddev->raid_disks will be 0, and data from
888 * dev should be merged in. Subsequent calls check that dev
889 * is new enough. Return 0 or -EINVAL
891 * void sync_super(struct mddev *mddev, struct md_rdev *dev)
892 * Update the superblock for rdev with data in mddev
893 * This does not write to disc.
899 struct module *owner;
900 int (*load_super)(struct md_rdev *rdev,
901 struct md_rdev *refdev,
903 int (*validate_super)(struct mddev *mddev,
904 struct md_rdev *rdev);
905 void (*sync_super)(struct mddev *mddev,
906 struct md_rdev *rdev);
907 unsigned long long (*rdev_size_change)(struct md_rdev *rdev,
908 sector_t num_sectors);
909 int (*allow_new_offset)(struct md_rdev *rdev,
910 unsigned long long new_offset);
914 * Check that the given mddev has no bitmap.
916 * This function is called from the run method of all personalities that do not
917 * support bitmaps. It prints an error message and returns non-zero if mddev
918 * has a bitmap. Otherwise, it returns 0.
921 int md_check_no_bitmap(struct mddev *mddev)
923 if (!mddev->bitmap_info.file && !mddev->bitmap_info.offset)
925 printk(KERN_ERR "%s: bitmaps are not supported for %s\n",
926 mdname(mddev), mddev->pers->name);
929 EXPORT_SYMBOL(md_check_no_bitmap);
932 * load_super for 0.90.0
934 static int super_90_load(struct md_rdev *rdev, struct md_rdev *refdev, int minor_version)
936 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
941 * Calculate the position of the superblock (512byte sectors),
942 * it's at the end of the disk.
944 * It also happens to be a multiple of 4Kb.
946 rdev->sb_start = calc_dev_sboffset(rdev);
948 ret = read_disk_sb(rdev, MD_SB_BYTES);
953 bdevname(rdev->bdev, b);
954 sb = page_address(rdev->sb_page);
956 if (sb->md_magic != MD_SB_MAGIC) {
957 printk(KERN_ERR "md: invalid raid superblock magic on %s\n",
962 if (sb->major_version != 0 ||
963 sb->minor_version < 90 ||
964 sb->minor_version > 91) {
965 printk(KERN_WARNING "Bad version number %d.%d on %s\n",
966 sb->major_version, sb->minor_version,
971 if (sb->raid_disks <= 0)
974 if (md_csum_fold(calc_sb_csum(sb)) != md_csum_fold(sb->sb_csum)) {
975 printk(KERN_WARNING "md: invalid superblock checksum on %s\n",
980 rdev->preferred_minor = sb->md_minor;
981 rdev->data_offset = 0;
982 rdev->new_data_offset = 0;
983 rdev->sb_size = MD_SB_BYTES;
984 rdev->badblocks.shift = -1;
986 if (sb->level == LEVEL_MULTIPATH)
989 rdev->desc_nr = sb->this_disk.number;
995 mdp_super_t *refsb = page_address(refdev->sb_page);
996 if (!uuid_equal(refsb, sb)) {
997 printk(KERN_WARNING "md: %s has different UUID to %s\n",
998 b, bdevname(refdev->bdev,b2));
1001 if (!sb_equal(refsb, sb)) {
1002 printk(KERN_WARNING "md: %s has same UUID"
1003 " but different superblock to %s\n",
1004 b, bdevname(refdev->bdev, b2));
1008 ev2 = md_event(refsb);
1014 rdev->sectors = rdev->sb_start;
1015 /* Limit to 4TB as metadata cannot record more than that.
1016 * (not needed for Linear and RAID0 as metadata doesn't
1019 if (IS_ENABLED(CONFIG_LBDAF) && (u64)rdev->sectors >= (2ULL << 32) &&
1021 rdev->sectors = (sector_t)(2ULL << 32) - 2;
1023 if (rdev->sectors < ((sector_t)sb->size) * 2 && sb->level >= 1)
1024 /* "this cannot possibly happen" ... */
1032 * validate_super for 0.90.0
1034 static int super_90_validate(struct mddev *mddev, struct md_rdev *rdev)
1037 mdp_super_t *sb = page_address(rdev->sb_page);
1038 __u64 ev1 = md_event(sb);
1040 rdev->raid_disk = -1;
1041 clear_bit(Faulty, &rdev->flags);
1042 clear_bit(In_sync, &rdev->flags);
1043 clear_bit(Bitmap_sync, &rdev->flags);
1044 clear_bit(WriteMostly, &rdev->flags);
1046 if (mddev->raid_disks == 0) {
1047 mddev->major_version = 0;
1048 mddev->minor_version = sb->minor_version;
1049 mddev->patch_version = sb->patch_version;
1050 mddev->external = 0;
1051 mddev->chunk_sectors = sb->chunk_size >> 9;
1052 mddev->ctime = sb->ctime;
1053 mddev->utime = sb->utime;
1054 mddev->level = sb->level;
1055 mddev->clevel[0] = 0;
1056 mddev->layout = sb->layout;
1057 mddev->raid_disks = sb->raid_disks;
1058 mddev->dev_sectors = ((sector_t)sb->size) * 2;
1059 mddev->events = ev1;
1060 mddev->bitmap_info.offset = 0;
1061 mddev->bitmap_info.space = 0;
1062 /* bitmap can use 60 K after the 4K superblocks */
1063 mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9;
1064 mddev->bitmap_info.default_space = 64*2 - (MD_SB_BYTES >> 9);
1065 mddev->reshape_backwards = 0;
1067 if (mddev->minor_version >= 91) {
1068 mddev->reshape_position = sb->reshape_position;
1069 mddev->delta_disks = sb->delta_disks;
1070 mddev->new_level = sb->new_level;
1071 mddev->new_layout = sb->new_layout;
1072 mddev->new_chunk_sectors = sb->new_chunk >> 9;
1073 if (mddev->delta_disks < 0)
1074 mddev->reshape_backwards = 1;
1076 mddev->reshape_position = MaxSector;
1077 mddev->delta_disks = 0;
1078 mddev->new_level = mddev->level;
1079 mddev->new_layout = mddev->layout;
1080 mddev->new_chunk_sectors = mddev->chunk_sectors;
1083 if (sb->state & (1<<MD_SB_CLEAN))
1084 mddev->recovery_cp = MaxSector;
1086 if (sb->events_hi == sb->cp_events_hi &&
1087 sb->events_lo == sb->cp_events_lo) {
1088 mddev->recovery_cp = sb->recovery_cp;
1090 mddev->recovery_cp = 0;
1093 memcpy(mddev->uuid+0, &sb->set_uuid0, 4);
1094 memcpy(mddev->uuid+4, &sb->set_uuid1, 4);
1095 memcpy(mddev->uuid+8, &sb->set_uuid2, 4);
1096 memcpy(mddev->uuid+12,&sb->set_uuid3, 4);
1098 mddev->max_disks = MD_SB_DISKS;
1100 if (sb->state & (1<<MD_SB_BITMAP_PRESENT) &&
1101 mddev->bitmap_info.file == NULL) {
1102 mddev->bitmap_info.offset =
1103 mddev->bitmap_info.default_offset;
1104 mddev->bitmap_info.space =
1105 mddev->bitmap_info.default_space;
1108 } else if (mddev->pers == NULL) {
1109 /* Insist on good event counter while assembling, except
1110 * for spares (which don't need an event count) */
1112 if (sb->disks[rdev->desc_nr].state & (
1113 (1<<MD_DISK_SYNC) | (1 << MD_DISK_ACTIVE)))
1114 if (ev1 < mddev->events)
1116 } else if (mddev->bitmap) {
1117 /* if adding to array with a bitmap, then we can accept an
1118 * older device ... but not too old.
1120 if (ev1 < mddev->bitmap->events_cleared)
1122 if (ev1 < mddev->events)
1123 set_bit(Bitmap_sync, &rdev->flags);
1125 if (ev1 < mddev->events)
1126 /* just a hot-add of a new device, leave raid_disk at -1 */
1130 if (mddev->level != LEVEL_MULTIPATH) {
1131 desc = sb->disks + rdev->desc_nr;
1133 if (desc->state & (1<<MD_DISK_FAULTY))
1134 set_bit(Faulty, &rdev->flags);
1135 else if (desc->state & (1<<MD_DISK_SYNC) /* &&
1136 desc->raid_disk < mddev->raid_disks */) {
1137 set_bit(In_sync, &rdev->flags);
1138 rdev->raid_disk = desc->raid_disk;
1139 rdev->saved_raid_disk = desc->raid_disk;
1140 } else if (desc->state & (1<<MD_DISK_ACTIVE)) {
1141 /* active but not in sync implies recovery up to
1142 * reshape position. We don't know exactly where
1143 * that is, so set to zero for now */
1144 if (mddev->minor_version >= 91) {
1145 rdev->recovery_offset = 0;
1146 rdev->raid_disk = desc->raid_disk;
1149 if (desc->state & (1<<MD_DISK_WRITEMOSTLY))
1150 set_bit(WriteMostly, &rdev->flags);
1151 } else /* MULTIPATH are always insync */
1152 set_bit(In_sync, &rdev->flags);
1157 * sync_super for 0.90.0
1159 static void super_90_sync(struct mddev *mddev, struct md_rdev *rdev)
1162 struct md_rdev *rdev2;
1163 int next_spare = mddev->raid_disks;
1165 /* make rdev->sb match mddev data..
1168 * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
1169 * 3/ any empty disks < next_spare become removed
1171 * disks[0] gets initialised to REMOVED because
1172 * we cannot be sure from other fields if it has
1173 * been initialised or not.
1176 int active=0, working=0,failed=0,spare=0,nr_disks=0;
1178 rdev->sb_size = MD_SB_BYTES;
1180 sb = page_address(rdev->sb_page);
1182 memset(sb, 0, sizeof(*sb));
1184 sb->md_magic = MD_SB_MAGIC;
1185 sb->major_version = mddev->major_version;
1186 sb->patch_version = mddev->patch_version;
1187 sb->gvalid_words = 0; /* ignored */
1188 memcpy(&sb->set_uuid0, mddev->uuid+0, 4);
1189 memcpy(&sb->set_uuid1, mddev->uuid+4, 4);
1190 memcpy(&sb->set_uuid2, mddev->uuid+8, 4);
1191 memcpy(&sb->set_uuid3, mddev->uuid+12,4);
1193 sb->ctime = clamp_t(time64_t, mddev->ctime, 0, U32_MAX);
1194 sb->level = mddev->level;
1195 sb->size = mddev->dev_sectors / 2;
1196 sb->raid_disks = mddev->raid_disks;
1197 sb->md_minor = mddev->md_minor;
1198 sb->not_persistent = 0;
1199 sb->utime = clamp_t(time64_t, mddev->utime, 0, U32_MAX);
1201 sb->events_hi = (mddev->events>>32);
1202 sb->events_lo = (u32)mddev->events;
1204 if (mddev->reshape_position == MaxSector)
1205 sb->minor_version = 90;
1207 sb->minor_version = 91;
1208 sb->reshape_position = mddev->reshape_position;
1209 sb->new_level = mddev->new_level;
1210 sb->delta_disks = mddev->delta_disks;
1211 sb->new_layout = mddev->new_layout;
1212 sb->new_chunk = mddev->new_chunk_sectors << 9;
1214 mddev->minor_version = sb->minor_version;
1217 sb->recovery_cp = mddev->recovery_cp;
1218 sb->cp_events_hi = (mddev->events>>32);
1219 sb->cp_events_lo = (u32)mddev->events;
1220 if (mddev->recovery_cp == MaxSector)
1221 sb->state = (1<< MD_SB_CLEAN);
1223 sb->recovery_cp = 0;
1225 sb->layout = mddev->layout;
1226 sb->chunk_size = mddev->chunk_sectors << 9;
1228 if (mddev->bitmap && mddev->bitmap_info.file == NULL)
1229 sb->state |= (1<<MD_SB_BITMAP_PRESENT);
1231 sb->disks[0].state = (1<<MD_DISK_REMOVED);
1232 rdev_for_each(rdev2, mddev) {
1235 int is_active = test_bit(In_sync, &rdev2->flags);
1237 if (rdev2->raid_disk >= 0 &&
1238 sb->minor_version >= 91)
1239 /* we have nowhere to store the recovery_offset,
1240 * but if it is not below the reshape_position,
1241 * we can piggy-back on that.
1244 if (rdev2->raid_disk < 0 ||
1245 test_bit(Faulty, &rdev2->flags))
1248 desc_nr = rdev2->raid_disk;
1250 desc_nr = next_spare++;
1251 rdev2->desc_nr = desc_nr;
1252 d = &sb->disks[rdev2->desc_nr];
1254 d->number = rdev2->desc_nr;
1255 d->major = MAJOR(rdev2->bdev->bd_dev);
1256 d->minor = MINOR(rdev2->bdev->bd_dev);
1258 d->raid_disk = rdev2->raid_disk;
1260 d->raid_disk = rdev2->desc_nr; /* compatibility */
1261 if (test_bit(Faulty, &rdev2->flags))
1262 d->state = (1<<MD_DISK_FAULTY);
1263 else if (is_active) {
1264 d->state = (1<<MD_DISK_ACTIVE);
1265 if (test_bit(In_sync, &rdev2->flags))
1266 d->state |= (1<<MD_DISK_SYNC);
1274 if (test_bit(WriteMostly, &rdev2->flags))
1275 d->state |= (1<<MD_DISK_WRITEMOSTLY);
1277 /* now set the "removed" and "faulty" bits on any missing devices */
1278 for (i=0 ; i < mddev->raid_disks ; i++) {
1279 mdp_disk_t *d = &sb->disks[i];
1280 if (d->state == 0 && d->number == 0) {
1283 d->state = (1<<MD_DISK_REMOVED);
1284 d->state |= (1<<MD_DISK_FAULTY);
1288 sb->nr_disks = nr_disks;
1289 sb->active_disks = active;
1290 sb->working_disks = working;
1291 sb->failed_disks = failed;
1292 sb->spare_disks = spare;
1294 sb->this_disk = sb->disks[rdev->desc_nr];
1295 sb->sb_csum = calc_sb_csum(sb);
1299 * rdev_size_change for 0.90.0
1301 static unsigned long long
1302 super_90_rdev_size_change(struct md_rdev *rdev, sector_t num_sectors)
1304 if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
1305 return 0; /* component must fit device */
1306 if (rdev->mddev->bitmap_info.offset)
1307 return 0; /* can't move bitmap */
1308 rdev->sb_start = calc_dev_sboffset(rdev);
1309 if (!num_sectors || num_sectors > rdev->sb_start)
1310 num_sectors = rdev->sb_start;
1311 /* Limit to 4TB as metadata cannot record more than that.
1312 * 4TB == 2^32 KB, or 2*2^32 sectors.
1314 if (IS_ENABLED(CONFIG_LBDAF) && (u64)num_sectors >= (2ULL << 32) &&
1315 rdev->mddev->level >= 1)
1316 num_sectors = (sector_t)(2ULL << 32) - 2;
1317 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
1319 md_super_wait(rdev->mddev);
1324 super_90_allow_new_offset(struct md_rdev *rdev, unsigned long long new_offset)
1326 /* non-zero offset changes not possible with v0.90 */
1327 return new_offset == 0;
1331 * version 1 superblock
1334 static __le32 calc_sb_1_csum(struct mdp_superblock_1 *sb)
1338 unsigned long long newcsum;
1339 int size = 256 + le32_to_cpu(sb->max_dev)*2;
1340 __le32 *isuper = (__le32*)sb;
1342 disk_csum = sb->sb_csum;
1345 for (; size >= 4; size -= 4)
1346 newcsum += le32_to_cpu(*isuper++);
1349 newcsum += le16_to_cpu(*(__le16*) isuper);
1351 csum = (newcsum & 0xffffffff) + (newcsum >> 32);
1352 sb->sb_csum = disk_csum;
1353 return cpu_to_le32(csum);
1356 static int super_1_load(struct md_rdev *rdev, struct md_rdev *refdev, int minor_version)
1358 struct mdp_superblock_1 *sb;
1362 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
1366 * Calculate the position of the superblock in 512byte sectors.
1367 * It is always aligned to a 4K boundary and
1368 * depeding on minor_version, it can be:
1369 * 0: At least 8K, but less than 12K, from end of device
1370 * 1: At start of device
1371 * 2: 4K from start of device.
1373 switch(minor_version) {
1375 sb_start = i_size_read(rdev->bdev->bd_inode) >> 9;
1377 sb_start &= ~(sector_t)(4*2-1);
1388 rdev->sb_start = sb_start;
1390 /* superblock is rarely larger than 1K, but it can be larger,
1391 * and it is safe to read 4k, so we do that
1393 ret = read_disk_sb(rdev, 4096);
1394 if (ret) return ret;
1396 sb = page_address(rdev->sb_page);
1398 if (sb->magic != cpu_to_le32(MD_SB_MAGIC) ||
1399 sb->major_version != cpu_to_le32(1) ||
1400 le32_to_cpu(sb->max_dev) > (4096-256)/2 ||
1401 le64_to_cpu(sb->super_offset) != rdev->sb_start ||
1402 (le32_to_cpu(sb->feature_map) & ~MD_FEATURE_ALL) != 0)
1405 if (calc_sb_1_csum(sb) != sb->sb_csum) {
1406 printk("md: invalid superblock checksum on %s\n",
1407 bdevname(rdev->bdev,b));
1410 if (le64_to_cpu(sb->data_size) < 10) {
1411 printk("md: data_size too small on %s\n",
1412 bdevname(rdev->bdev,b));
1417 memcmp(sb->pad3, sb->pad3+1, sizeof(sb->pad3) - sizeof(sb->pad3[1])))
1418 /* Some padding is non-zero, might be a new feature */
1421 rdev->preferred_minor = 0xffff;
1422 rdev->data_offset = le64_to_cpu(sb->data_offset);
1423 rdev->new_data_offset = rdev->data_offset;
1424 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE) &&
1425 (le32_to_cpu(sb->feature_map) & MD_FEATURE_NEW_OFFSET))
1426 rdev->new_data_offset += (s32)le32_to_cpu(sb->new_offset);
1427 atomic_set(&rdev->corrected_errors, le32_to_cpu(sb->cnt_corrected_read));
1429 rdev->sb_size = le32_to_cpu(sb->max_dev) * 2 + 256;
1430 bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
1431 if (rdev->sb_size & bmask)
1432 rdev->sb_size = (rdev->sb_size | bmask) + 1;
1435 && rdev->data_offset < sb_start + (rdev->sb_size/512))
1438 && rdev->new_data_offset < sb_start + (rdev->sb_size/512))
1441 if (sb->level == cpu_to_le32(LEVEL_MULTIPATH))
1444 rdev->desc_nr = le32_to_cpu(sb->dev_number);
1446 if (!rdev->bb_page) {
1447 rdev->bb_page = alloc_page(GFP_KERNEL);
1451 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BAD_BLOCKS) &&
1452 rdev->badblocks.count == 0) {
1453 /* need to load the bad block list.
1454 * Currently we limit it to one page.
1460 int sectors = le16_to_cpu(sb->bblog_size);
1461 if (sectors > (PAGE_SIZE / 512))
1463 offset = le32_to_cpu(sb->bblog_offset);
1466 bb_sector = (long long)offset;
1467 if (!sync_page_io(rdev, bb_sector, sectors << 9,
1468 rdev->bb_page, READ, true))
1470 bbp = (u64 *)page_address(rdev->bb_page);
1471 rdev->badblocks.shift = sb->bblog_shift;
1472 for (i = 0 ; i < (sectors << (9-3)) ; i++, bbp++) {
1473 u64 bb = le64_to_cpu(*bbp);
1474 int count = bb & (0x3ff);
1475 u64 sector = bb >> 10;
1476 sector <<= sb->bblog_shift;
1477 count <<= sb->bblog_shift;
1480 if (badblocks_set(&rdev->badblocks, sector, count, 1))
1483 } else if (sb->bblog_offset != 0)
1484 rdev->badblocks.shift = 0;
1490 struct mdp_superblock_1 *refsb = page_address(refdev->sb_page);
1492 if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 ||
1493 sb->level != refsb->level ||
1494 sb->layout != refsb->layout ||
1495 sb->chunksize != refsb->chunksize) {
1496 printk(KERN_WARNING "md: %s has strangely different"
1497 " superblock to %s\n",
1498 bdevname(rdev->bdev,b),
1499 bdevname(refdev->bdev,b2));
1502 ev1 = le64_to_cpu(sb->events);
1503 ev2 = le64_to_cpu(refsb->events);
1510 if (minor_version) {
1511 sectors = (i_size_read(rdev->bdev->bd_inode) >> 9);
1512 sectors -= rdev->data_offset;
1514 sectors = rdev->sb_start;
1515 if (sectors < le64_to_cpu(sb->data_size))
1517 rdev->sectors = le64_to_cpu(sb->data_size);
1521 static int super_1_validate(struct mddev *mddev, struct md_rdev *rdev)
1523 struct mdp_superblock_1 *sb = page_address(rdev->sb_page);
1524 __u64 ev1 = le64_to_cpu(sb->events);
1526 rdev->raid_disk = -1;
1527 clear_bit(Faulty, &rdev->flags);
1528 clear_bit(In_sync, &rdev->flags);
1529 clear_bit(Bitmap_sync, &rdev->flags);
1530 clear_bit(WriteMostly, &rdev->flags);
1532 if (mddev->raid_disks == 0) {
1533 mddev->major_version = 1;
1534 mddev->patch_version = 0;
1535 mddev->external = 0;
1536 mddev->chunk_sectors = le32_to_cpu(sb->chunksize);
1537 mddev->ctime = le64_to_cpu(sb->ctime);
1538 mddev->utime = le64_to_cpu(sb->utime);
1539 mddev->level = le32_to_cpu(sb->level);
1540 mddev->clevel[0] = 0;
1541 mddev->layout = le32_to_cpu(sb->layout);
1542 mddev->raid_disks = le32_to_cpu(sb->raid_disks);
1543 mddev->dev_sectors = le64_to_cpu(sb->size);
1544 mddev->events = ev1;
1545 mddev->bitmap_info.offset = 0;
1546 mddev->bitmap_info.space = 0;
1547 /* Default location for bitmap is 1K after superblock
1548 * using 3K - total of 4K
1550 mddev->bitmap_info.default_offset = 1024 >> 9;
1551 mddev->bitmap_info.default_space = (4096-1024) >> 9;
1552 mddev->reshape_backwards = 0;
1554 mddev->recovery_cp = le64_to_cpu(sb->resync_offset);
1555 memcpy(mddev->uuid, sb->set_uuid, 16);
1557 mddev->max_disks = (4096-256)/2;
1559 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) &&
1560 mddev->bitmap_info.file == NULL) {
1561 mddev->bitmap_info.offset =
1562 (__s32)le32_to_cpu(sb->bitmap_offset);
1563 /* Metadata doesn't record how much space is available.
1564 * For 1.0, we assume we can use up to the superblock
1565 * if before, else to 4K beyond superblock.
1566 * For others, assume no change is possible.
1568 if (mddev->minor_version > 0)
1569 mddev->bitmap_info.space = 0;
1570 else if (mddev->bitmap_info.offset > 0)
1571 mddev->bitmap_info.space =
1572 8 - mddev->bitmap_info.offset;
1574 mddev->bitmap_info.space =
1575 -mddev->bitmap_info.offset;
1578 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) {
1579 mddev->reshape_position = le64_to_cpu(sb->reshape_position);
1580 mddev->delta_disks = le32_to_cpu(sb->delta_disks);
1581 mddev->new_level = le32_to_cpu(sb->new_level);
1582 mddev->new_layout = le32_to_cpu(sb->new_layout);
1583 mddev->new_chunk_sectors = le32_to_cpu(sb->new_chunk);
1584 if (mddev->delta_disks < 0 ||
1585 (mddev->delta_disks == 0 &&
1586 (le32_to_cpu(sb->feature_map)
1587 & MD_FEATURE_RESHAPE_BACKWARDS)))
1588 mddev->reshape_backwards = 1;
1590 mddev->reshape_position = MaxSector;
1591 mddev->delta_disks = 0;
1592 mddev->new_level = mddev->level;
1593 mddev->new_layout = mddev->layout;
1594 mddev->new_chunk_sectors = mddev->chunk_sectors;
1597 if (le32_to_cpu(sb->feature_map) & MD_FEATURE_JOURNAL) {
1598 set_bit(MD_HAS_JOURNAL, &mddev->flags);
1599 if (mddev->recovery_cp == MaxSector)
1600 set_bit(MD_JOURNAL_CLEAN, &mddev->flags);
1602 } else if (mddev->pers == NULL) {
1603 /* Insist of good event counter while assembling, except for
1604 * spares (which don't need an event count) */
1606 if (rdev->desc_nr >= 0 &&
1607 rdev->desc_nr < le32_to_cpu(sb->max_dev) &&
1608 (le16_to_cpu(sb->dev_roles[rdev->desc_nr]) < MD_DISK_ROLE_MAX ||
1609 le16_to_cpu(sb->dev_roles[rdev->desc_nr]) == MD_DISK_ROLE_JOURNAL))
1610 if (ev1 < mddev->events)
1612 } else if (mddev->bitmap) {
1613 /* If adding to array with a bitmap, then we can accept an
1614 * older device, but not too old.
1616 if (ev1 < mddev->bitmap->events_cleared)
1618 if (ev1 < mddev->events)
1619 set_bit(Bitmap_sync, &rdev->flags);
1621 if (ev1 < mddev->events)
1622 /* just a hot-add of a new device, leave raid_disk at -1 */
1625 if (mddev->level != LEVEL_MULTIPATH) {
1627 if (rdev->desc_nr < 0 ||
1628 rdev->desc_nr >= le32_to_cpu(sb->max_dev)) {
1629 role = MD_DISK_ROLE_SPARE;
1632 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
1634 case MD_DISK_ROLE_SPARE: /* spare */
1636 case MD_DISK_ROLE_FAULTY: /* faulty */
1637 set_bit(Faulty, &rdev->flags);
1639 case MD_DISK_ROLE_JOURNAL: /* journal device */
1640 if (!(le32_to_cpu(sb->feature_map) & MD_FEATURE_JOURNAL)) {
1641 /* journal device without journal feature */
1643 "md: journal device provided without journal feature, ignoring the device\n");
1646 set_bit(Journal, &rdev->flags);
1647 rdev->journal_tail = le64_to_cpu(sb->journal_tail);
1648 rdev->raid_disk = 0;
1651 rdev->saved_raid_disk = role;
1652 if ((le32_to_cpu(sb->feature_map) &
1653 MD_FEATURE_RECOVERY_OFFSET)) {
1654 rdev->recovery_offset = le64_to_cpu(sb->recovery_offset);
1655 if (!(le32_to_cpu(sb->feature_map) &
1656 MD_FEATURE_RECOVERY_BITMAP))
1657 rdev->saved_raid_disk = -1;
1659 set_bit(In_sync, &rdev->flags);
1660 rdev->raid_disk = role;
1663 if (sb->devflags & WriteMostly1)
1664 set_bit(WriteMostly, &rdev->flags);
1665 if (le32_to_cpu(sb->feature_map) & MD_FEATURE_REPLACEMENT)
1666 set_bit(Replacement, &rdev->flags);
1667 } else /* MULTIPATH are always insync */
1668 set_bit(In_sync, &rdev->flags);
1673 static void super_1_sync(struct mddev *mddev, struct md_rdev *rdev)
1675 struct mdp_superblock_1 *sb;
1676 struct md_rdev *rdev2;
1678 /* make rdev->sb match mddev and rdev data. */
1680 sb = page_address(rdev->sb_page);
1682 sb->feature_map = 0;
1684 sb->recovery_offset = cpu_to_le64(0);
1685 memset(sb->pad3, 0, sizeof(sb->pad3));
1687 sb->utime = cpu_to_le64((__u64)mddev->utime);
1688 sb->events = cpu_to_le64(mddev->events);
1690 sb->resync_offset = cpu_to_le64(mddev->recovery_cp);
1691 else if (test_bit(MD_JOURNAL_CLEAN, &mddev->flags))
1692 sb->resync_offset = cpu_to_le64(MaxSector);
1694 sb->resync_offset = cpu_to_le64(0);
1696 sb->cnt_corrected_read = cpu_to_le32(atomic_read(&rdev->corrected_errors));
1698 sb->raid_disks = cpu_to_le32(mddev->raid_disks);
1699 sb->size = cpu_to_le64(mddev->dev_sectors);
1700 sb->chunksize = cpu_to_le32(mddev->chunk_sectors);
1701 sb->level = cpu_to_le32(mddev->level);
1702 sb->layout = cpu_to_le32(mddev->layout);
1704 if (test_bit(WriteMostly, &rdev->flags))
1705 sb->devflags |= WriteMostly1;
1707 sb->devflags &= ~WriteMostly1;
1708 sb->data_offset = cpu_to_le64(rdev->data_offset);
1709 sb->data_size = cpu_to_le64(rdev->sectors);
1711 if (mddev->bitmap && mddev->bitmap_info.file == NULL) {
1712 sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_info.offset);
1713 sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET);
1716 if (rdev->raid_disk >= 0 && !test_bit(Journal, &rdev->flags) &&
1717 !test_bit(In_sync, &rdev->flags)) {
1719 cpu_to_le32(MD_FEATURE_RECOVERY_OFFSET);
1720 sb->recovery_offset =
1721 cpu_to_le64(rdev->recovery_offset);
1722 if (rdev->saved_raid_disk >= 0 && mddev->bitmap)
1724 cpu_to_le32(MD_FEATURE_RECOVERY_BITMAP);
1726 /* Note: recovery_offset and journal_tail share space */
1727 if (test_bit(Journal, &rdev->flags))
1728 sb->journal_tail = cpu_to_le64(rdev->journal_tail);
1729 if (test_bit(Replacement, &rdev->flags))
1731 cpu_to_le32(MD_FEATURE_REPLACEMENT);
1733 if (mddev->reshape_position != MaxSector) {
1734 sb->feature_map |= cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE);
1735 sb->reshape_position = cpu_to_le64(mddev->reshape_position);
1736 sb->new_layout = cpu_to_le32(mddev->new_layout);
1737 sb->delta_disks = cpu_to_le32(mddev->delta_disks);
1738 sb->new_level = cpu_to_le32(mddev->new_level);
1739 sb->new_chunk = cpu_to_le32(mddev->new_chunk_sectors);
1740 if (mddev->delta_disks == 0 &&
1741 mddev->reshape_backwards)
1743 |= cpu_to_le32(MD_FEATURE_RESHAPE_BACKWARDS);
1744 if (rdev->new_data_offset != rdev->data_offset) {
1746 |= cpu_to_le32(MD_FEATURE_NEW_OFFSET);
1747 sb->new_offset = cpu_to_le32((__u32)(rdev->new_data_offset
1748 - rdev->data_offset));
1752 if (mddev_is_clustered(mddev))
1753 sb->feature_map |= cpu_to_le32(MD_FEATURE_CLUSTERED);
1755 if (rdev->badblocks.count == 0)
1756 /* Nothing to do for bad blocks*/ ;
1757 else if (sb->bblog_offset == 0)
1758 /* Cannot record bad blocks on this device */
1759 md_error(mddev, rdev);
1761 struct badblocks *bb = &rdev->badblocks;
1762 u64 *bbp = (u64 *)page_address(rdev->bb_page);
1764 sb->feature_map |= cpu_to_le32(MD_FEATURE_BAD_BLOCKS);
1769 seq = read_seqbegin(&bb->lock);
1771 memset(bbp, 0xff, PAGE_SIZE);
1773 for (i = 0 ; i < bb->count ; i++) {
1774 u64 internal_bb = p[i];
1775 u64 store_bb = ((BB_OFFSET(internal_bb) << 10)
1776 | BB_LEN(internal_bb));
1777 bbp[i] = cpu_to_le64(store_bb);
1780 if (read_seqretry(&bb->lock, seq))
1783 bb->sector = (rdev->sb_start +
1784 (int)le32_to_cpu(sb->bblog_offset));
1785 bb->size = le16_to_cpu(sb->bblog_size);
1790 rdev_for_each(rdev2, mddev)
1791 if (rdev2->desc_nr+1 > max_dev)
1792 max_dev = rdev2->desc_nr+1;
1794 if (max_dev > le32_to_cpu(sb->max_dev)) {
1796 sb->max_dev = cpu_to_le32(max_dev);
1797 rdev->sb_size = max_dev * 2 + 256;
1798 bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
1799 if (rdev->sb_size & bmask)
1800 rdev->sb_size = (rdev->sb_size | bmask) + 1;
1802 max_dev = le32_to_cpu(sb->max_dev);
1804 for (i=0; i<max_dev;i++)
1805 sb->dev_roles[i] = cpu_to_le16(MD_DISK_ROLE_FAULTY);
1807 if (test_bit(MD_HAS_JOURNAL, &mddev->flags))
1808 sb->feature_map |= cpu_to_le32(MD_FEATURE_JOURNAL);
1810 rdev_for_each(rdev2, mddev) {
1812 if (test_bit(Faulty, &rdev2->flags))
1813 sb->dev_roles[i] = cpu_to_le16(MD_DISK_ROLE_FAULTY);
1814 else if (test_bit(In_sync, &rdev2->flags))
1815 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1816 else if (test_bit(Journal, &rdev2->flags))
1817 sb->dev_roles[i] = cpu_to_le16(MD_DISK_ROLE_JOURNAL);
1818 else if (rdev2->raid_disk >= 0)
1819 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1821 sb->dev_roles[i] = cpu_to_le16(MD_DISK_ROLE_SPARE);
1824 sb->sb_csum = calc_sb_1_csum(sb);
1827 static unsigned long long
1828 super_1_rdev_size_change(struct md_rdev *rdev, sector_t num_sectors)
1830 struct mdp_superblock_1 *sb;
1831 sector_t max_sectors;
1832 if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
1833 return 0; /* component must fit device */
1834 if (rdev->data_offset != rdev->new_data_offset)
1835 return 0; /* too confusing */
1836 if (rdev->sb_start < rdev->data_offset) {
1837 /* minor versions 1 and 2; superblock before data */
1838 max_sectors = i_size_read(rdev->bdev->bd_inode) >> 9;
1839 max_sectors -= rdev->data_offset;
1840 if (!num_sectors || num_sectors > max_sectors)
1841 num_sectors = max_sectors;
1842 } else if (rdev->mddev->bitmap_info.offset) {
1843 /* minor version 0 with bitmap we can't move */
1846 /* minor version 0; superblock after data */
1848 sb_start = (i_size_read(rdev->bdev->bd_inode) >> 9) - 8*2;
1849 sb_start &= ~(sector_t)(4*2 - 1);
1850 max_sectors = rdev->sectors + sb_start - rdev->sb_start;
1851 if (!num_sectors || num_sectors > max_sectors)
1852 num_sectors = max_sectors;
1853 rdev->sb_start = sb_start;
1855 sb = page_address(rdev->sb_page);
1856 sb->data_size = cpu_to_le64(num_sectors);
1857 sb->super_offset = rdev->sb_start;
1858 sb->sb_csum = calc_sb_1_csum(sb);
1859 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
1861 md_super_wait(rdev->mddev);
1867 super_1_allow_new_offset(struct md_rdev *rdev,
1868 unsigned long long new_offset)
1870 /* All necessary checks on new >= old have been done */
1871 struct bitmap *bitmap;
1872 if (new_offset >= rdev->data_offset)
1875 /* with 1.0 metadata, there is no metadata to tread on
1876 * so we can always move back */
1877 if (rdev->mddev->minor_version == 0)
1880 /* otherwise we must be sure not to step on
1881 * any metadata, so stay:
1882 * 36K beyond start of superblock
1883 * beyond end of badblocks
1884 * beyond write-intent bitmap
1886 if (rdev->sb_start + (32+4)*2 > new_offset)
1888 bitmap = rdev->mddev->bitmap;
1889 if (bitmap && !rdev->mddev->bitmap_info.file &&
1890 rdev->sb_start + rdev->mddev->bitmap_info.offset +
1891 bitmap->storage.file_pages * (PAGE_SIZE>>9) > new_offset)
1893 if (rdev->badblocks.sector + rdev->badblocks.size > new_offset)
1899 static struct super_type super_types[] = {
1902 .owner = THIS_MODULE,
1903 .load_super = super_90_load,
1904 .validate_super = super_90_validate,
1905 .sync_super = super_90_sync,
1906 .rdev_size_change = super_90_rdev_size_change,
1907 .allow_new_offset = super_90_allow_new_offset,
1911 .owner = THIS_MODULE,
1912 .load_super = super_1_load,
1913 .validate_super = super_1_validate,
1914 .sync_super = super_1_sync,
1915 .rdev_size_change = super_1_rdev_size_change,
1916 .allow_new_offset = super_1_allow_new_offset,
1920 static void sync_super(struct mddev *mddev, struct md_rdev *rdev)
1922 if (mddev->sync_super) {
1923 mddev->sync_super(mddev, rdev);
1927 BUG_ON(mddev->major_version >= ARRAY_SIZE(super_types));
1929 super_types[mddev->major_version].sync_super(mddev, rdev);
1932 static int match_mddev_units(struct mddev *mddev1, struct mddev *mddev2)
1934 struct md_rdev *rdev, *rdev2;
1937 rdev_for_each_rcu(rdev, mddev1) {
1938 if (test_bit(Faulty, &rdev->flags) ||
1939 test_bit(Journal, &rdev->flags) ||
1940 rdev->raid_disk == -1)
1942 rdev_for_each_rcu(rdev2, mddev2) {
1943 if (test_bit(Faulty, &rdev2->flags) ||
1944 test_bit(Journal, &rdev2->flags) ||
1945 rdev2->raid_disk == -1)
1947 if (rdev->bdev->bd_contains ==
1948 rdev2->bdev->bd_contains) {
1958 static LIST_HEAD(pending_raid_disks);
1961 * Try to register data integrity profile for an mddev
1963 * This is called when an array is started and after a disk has been kicked
1964 * from the array. It only succeeds if all working and active component devices
1965 * are integrity capable with matching profiles.
1967 int md_integrity_register(struct mddev *mddev)
1969 struct md_rdev *rdev, *reference = NULL;
1971 if (list_empty(&mddev->disks))
1972 return 0; /* nothing to do */
1973 if (!mddev->gendisk || blk_get_integrity(mddev->gendisk))
1974 return 0; /* shouldn't register, or already is */
1975 rdev_for_each(rdev, mddev) {
1976 /* skip spares and non-functional disks */
1977 if (test_bit(Faulty, &rdev->flags))
1979 if (rdev->raid_disk < 0)
1982 /* Use the first rdev as the reference */
1986 /* does this rdev's profile match the reference profile? */
1987 if (blk_integrity_compare(reference->bdev->bd_disk,
1988 rdev->bdev->bd_disk) < 0)
1991 if (!reference || !bdev_get_integrity(reference->bdev))
1994 * All component devices are integrity capable and have matching
1995 * profiles, register the common profile for the md device.
1997 blk_integrity_register(mddev->gendisk,
1998 bdev_get_integrity(reference->bdev));
2000 printk(KERN_NOTICE "md: data integrity enabled on %s\n", mdname(mddev));
2001 if (bioset_integrity_create(mddev->bio_set, BIO_POOL_SIZE)) {
2002 printk(KERN_ERR "md: failed to create integrity pool for %s\n",
2008 EXPORT_SYMBOL(md_integrity_register);
2011 * Attempt to add an rdev, but only if it is consistent with the current
2014 int md_integrity_add_rdev(struct md_rdev *rdev, struct mddev *mddev)
2016 struct blk_integrity *bi_rdev;
2017 struct blk_integrity *bi_mddev;
2018 char name[BDEVNAME_SIZE];
2020 if (!mddev->gendisk)
2023 bi_rdev = bdev_get_integrity(rdev->bdev);
2024 bi_mddev = blk_get_integrity(mddev->gendisk);
2026 if (!bi_mddev) /* nothing to do */
2029 if (blk_integrity_compare(mddev->gendisk, rdev->bdev->bd_disk) != 0) {
2030 printk(KERN_NOTICE "%s: incompatible integrity profile for %s\n",
2031 mdname(mddev), bdevname(rdev->bdev, name));
2037 EXPORT_SYMBOL(md_integrity_add_rdev);
2039 static int bind_rdev_to_array(struct md_rdev *rdev, struct mddev *mddev)
2041 char b[BDEVNAME_SIZE];
2045 /* prevent duplicates */
2046 if (find_rdev(mddev, rdev->bdev->bd_dev))
2049 /* make sure rdev->sectors exceeds mddev->dev_sectors */
2050 if (!test_bit(Journal, &rdev->flags) &&
2052 (mddev->dev_sectors == 0 || rdev->sectors < mddev->dev_sectors)) {
2054 /* Cannot change size, so fail
2055 * If mddev->level <= 0, then we don't care
2056 * about aligning sizes (e.g. linear)
2058 if (mddev->level > 0)
2061 mddev->dev_sectors = rdev->sectors;
2064 /* Verify rdev->desc_nr is unique.
2065 * If it is -1, assign a free number, else
2066 * check number is not in use
2069 if (rdev->desc_nr < 0) {
2072 choice = mddev->raid_disks;
2073 while (md_find_rdev_nr_rcu(mddev, choice))
2075 rdev->desc_nr = choice;
2077 if (md_find_rdev_nr_rcu(mddev, rdev->desc_nr)) {
2083 if (!test_bit(Journal, &rdev->flags) &&
2084 mddev->max_disks && rdev->desc_nr >= mddev->max_disks) {
2085 printk(KERN_WARNING "md: %s: array is limited to %d devices\n",
2086 mdname(mddev), mddev->max_disks);
2089 bdevname(rdev->bdev,b);
2090 strreplace(b, '/', '!');
2092 rdev->mddev = mddev;
2093 printk(KERN_INFO "md: bind<%s>\n", b);
2095 if ((err = kobject_add(&rdev->kobj, &mddev->kobj, "dev-%s", b)))
2098 ko = &part_to_dev(rdev->bdev->bd_part)->kobj;
2099 if (sysfs_create_link(&rdev->kobj, ko, "block"))
2100 /* failure here is OK */;
2101 rdev->sysfs_state = sysfs_get_dirent_safe(rdev->kobj.sd, "state");
2103 list_add_rcu(&rdev->same_set, &mddev->disks);
2104 bd_link_disk_holder(rdev->bdev, mddev->gendisk);
2106 /* May as well allow recovery to be retried once */
2107 mddev->recovery_disabled++;
2112 printk(KERN_WARNING "md: failed to register dev-%s for %s\n",
2117 static void md_delayed_delete(struct work_struct *ws)
2119 struct md_rdev *rdev = container_of(ws, struct md_rdev, del_work);
2120 kobject_del(&rdev->kobj);
2121 kobject_put(&rdev->kobj);
2124 static void unbind_rdev_from_array(struct md_rdev *rdev)
2126 char b[BDEVNAME_SIZE];
2128 bd_unlink_disk_holder(rdev->bdev, rdev->mddev->gendisk);
2129 list_del_rcu(&rdev->same_set);
2130 printk(KERN_INFO "md: unbind<%s>\n", bdevname(rdev->bdev,b));
2132 sysfs_remove_link(&rdev->kobj, "block");
2133 sysfs_put(rdev->sysfs_state);
2134 rdev->sysfs_state = NULL;
2135 rdev->badblocks.count = 0;
2136 /* We need to delay this, otherwise we can deadlock when
2137 * writing to 'remove' to "dev/state". We also need
2138 * to delay it due to rcu usage.
2141 INIT_WORK(&rdev->del_work, md_delayed_delete);
2142 kobject_get(&rdev->kobj);
2143 queue_work(md_misc_wq, &rdev->del_work);
2147 * prevent the device from being mounted, repartitioned or
2148 * otherwise reused by a RAID array (or any other kernel
2149 * subsystem), by bd_claiming the device.
2151 static int lock_rdev(struct md_rdev *rdev, dev_t dev, int shared)
2154 struct block_device *bdev;
2155 char b[BDEVNAME_SIZE];
2157 bdev = blkdev_get_by_dev(dev, FMODE_READ|FMODE_WRITE|FMODE_EXCL,
2158 shared ? (struct md_rdev *)lock_rdev : rdev);
2160 printk(KERN_ERR "md: could not open %s.\n",
2161 __bdevname(dev, b));
2162 return PTR_ERR(bdev);
2168 static void unlock_rdev(struct md_rdev *rdev)
2170 struct block_device *bdev = rdev->bdev;
2172 blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
2175 void md_autodetect_dev(dev_t dev);
2177 static void export_rdev(struct md_rdev *rdev)
2179 char b[BDEVNAME_SIZE];
2181 printk(KERN_INFO "md: export_rdev(%s)\n",
2182 bdevname(rdev->bdev,b));
2183 md_rdev_clear(rdev);
2185 if (test_bit(AutoDetected, &rdev->flags))
2186 md_autodetect_dev(rdev->bdev->bd_dev);
2189 kobject_put(&rdev->kobj);
2192 void md_kick_rdev_from_array(struct md_rdev *rdev)
2194 unbind_rdev_from_array(rdev);
2197 EXPORT_SYMBOL_GPL(md_kick_rdev_from_array);
2199 static void export_array(struct mddev *mddev)
2201 struct md_rdev *rdev;
2203 while (!list_empty(&mddev->disks)) {
2204 rdev = list_first_entry(&mddev->disks, struct md_rdev,
2206 md_kick_rdev_from_array(rdev);
2208 mddev->raid_disks = 0;
2209 mddev->major_version = 0;
2212 static void sync_sbs(struct mddev *mddev, int nospares)
2214 /* Update each superblock (in-memory image), but
2215 * if we are allowed to, skip spares which already
2216 * have the right event counter, or have one earlier
2217 * (which would mean they aren't being marked as dirty
2218 * with the rest of the array)
2220 struct md_rdev *rdev;
2221 rdev_for_each(rdev, mddev) {
2222 if (rdev->sb_events == mddev->events ||
2224 rdev->raid_disk < 0 &&
2225 rdev->sb_events+1 == mddev->events)) {
2226 /* Don't update this superblock */
2227 rdev->sb_loaded = 2;
2229 sync_super(mddev, rdev);
2230 rdev->sb_loaded = 1;
2235 static bool does_sb_need_changing(struct mddev *mddev)
2237 struct md_rdev *rdev;
2238 struct mdp_superblock_1 *sb;
2241 /* Find a good rdev */
2242 rdev_for_each(rdev, mddev)
2243 if ((rdev->raid_disk >= 0) && !test_bit(Faulty, &rdev->flags))
2246 /* No good device found. */
2250 sb = page_address(rdev->sb_page);
2251 /* Check if a device has become faulty or a spare become active */
2252 rdev_for_each(rdev, mddev) {
2253 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
2254 /* Device activated? */
2255 if (role == 0xffff && rdev->raid_disk >=0 &&
2256 !test_bit(Faulty, &rdev->flags))
2258 /* Device turned faulty? */
2259 if (test_bit(Faulty, &rdev->flags) && (role < 0xfffd))
2263 /* Check if any mddev parameters have changed */
2264 if ((mddev->dev_sectors != le64_to_cpu(sb->size)) ||
2265 (mddev->reshape_position != le64_to_cpu(sb->reshape_position)) ||
2266 (mddev->layout != le64_to_cpu(sb->layout)) ||
2267 (mddev->raid_disks != le32_to_cpu(sb->raid_disks)) ||
2268 (mddev->chunk_sectors != le32_to_cpu(sb->chunksize)))
2274 void md_update_sb(struct mddev *mddev, int force_change)
2276 struct md_rdev *rdev;
2279 int any_badblocks_changed = 0;
2284 set_bit(MD_CHANGE_DEVS, &mddev->flags);
2288 if (mddev_is_clustered(mddev)) {
2289 if (test_and_clear_bit(MD_CHANGE_DEVS, &mddev->flags))
2291 ret = md_cluster_ops->metadata_update_start(mddev);
2292 /* Has someone else has updated the sb */
2293 if (!does_sb_need_changing(mddev)) {
2295 md_cluster_ops->metadata_update_cancel(mddev);
2296 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
2301 /* First make sure individual recovery_offsets are correct */
2302 rdev_for_each(rdev, mddev) {
2303 if (rdev->raid_disk >= 0 &&
2304 mddev->delta_disks >= 0 &&
2305 !test_bit(Journal, &rdev->flags) &&
2306 !test_bit(In_sync, &rdev->flags) &&
2307 mddev->curr_resync_completed > rdev->recovery_offset)
2308 rdev->recovery_offset = mddev->curr_resync_completed;
2311 if (!mddev->persistent) {
2312 clear_bit(MD_CHANGE_CLEAN, &mddev->flags);
2313 clear_bit(MD_CHANGE_DEVS, &mddev->flags);
2314 if (!mddev->external) {
2315 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
2316 rdev_for_each(rdev, mddev) {
2317 if (rdev->badblocks.changed) {
2318 rdev->badblocks.changed = 0;
2319 ack_all_badblocks(&rdev->badblocks);
2320 md_error(mddev, rdev);
2322 clear_bit(Blocked, &rdev->flags);
2323 clear_bit(BlockedBadBlocks, &rdev->flags);
2324 wake_up(&rdev->blocked_wait);
2327 wake_up(&mddev->sb_wait);
2331 spin_lock(&mddev->lock);
2333 mddev->utime = ktime_get_real_seconds();
2335 if (test_and_clear_bit(MD_CHANGE_DEVS, &mddev->flags))
2337 if (test_and_clear_bit(MD_CHANGE_CLEAN, &mddev->flags))
2338 /* just a clean<-> dirty transition, possibly leave spares alone,
2339 * though if events isn't the right even/odd, we will have to do
2345 if (mddev->degraded)
2346 /* If the array is degraded, then skipping spares is both
2347 * dangerous and fairly pointless.
2348 * Dangerous because a device that was removed from the array
2349 * might have a event_count that still looks up-to-date,
2350 * so it can be re-added without a resync.
2351 * Pointless because if there are any spares to skip,
2352 * then a recovery will happen and soon that array won't
2353 * be degraded any more and the spare can go back to sleep then.
2357 sync_req = mddev->in_sync;
2359 /* If this is just a dirty<->clean transition, and the array is clean
2360 * and 'events' is odd, we can roll back to the previous clean state */
2362 && (mddev->in_sync && mddev->recovery_cp == MaxSector)
2363 && mddev->can_decrease_events
2364 && mddev->events != 1) {
2366 mddev->can_decrease_events = 0;
2368 /* otherwise we have to go forward and ... */
2370 mddev->can_decrease_events = nospares;
2374 * This 64-bit counter should never wrap.
2375 * Either we are in around ~1 trillion A.C., assuming
2376 * 1 reboot per second, or we have a bug...
2378 WARN_ON(mddev->events == 0);
2380 rdev_for_each(rdev, mddev) {
2381 if (rdev->badblocks.changed)
2382 any_badblocks_changed++;
2383 if (test_bit(Faulty, &rdev->flags))
2384 set_bit(FaultRecorded, &rdev->flags);
2387 sync_sbs(mddev, nospares);
2388 spin_unlock(&mddev->lock);
2390 pr_debug("md: updating %s RAID superblock on device (in sync %d)\n",
2391 mdname(mddev), mddev->in_sync);
2393 bitmap_update_sb(mddev->bitmap);
2394 rdev_for_each(rdev, mddev) {
2395 char b[BDEVNAME_SIZE];
2397 if (rdev->sb_loaded != 1)
2398 continue; /* no noise on spare devices */
2400 if (!test_bit(Faulty, &rdev->flags)) {
2401 md_super_write(mddev,rdev,
2402 rdev->sb_start, rdev->sb_size,
2404 pr_debug("md: (write) %s's sb offset: %llu\n",
2405 bdevname(rdev->bdev, b),
2406 (unsigned long long)rdev->sb_start);
2407 rdev->sb_events = mddev->events;
2408 if (rdev->badblocks.size) {
2409 md_super_write(mddev, rdev,
2410 rdev->badblocks.sector,
2411 rdev->badblocks.size << 9,
2413 rdev->badblocks.size = 0;
2417 pr_debug("md: %s (skipping faulty)\n",
2418 bdevname(rdev->bdev, b));
2420 if (mddev->level == LEVEL_MULTIPATH)
2421 /* only need to write one superblock... */
2424 md_super_wait(mddev);
2425 /* if there was a failure, MD_CHANGE_DEVS was set, and we re-write super */
2427 spin_lock(&mddev->lock);
2428 if (mddev->in_sync != sync_req ||
2429 test_bit(MD_CHANGE_DEVS, &mddev->flags)) {
2430 /* have to write it out again */
2431 spin_unlock(&mddev->lock);
2434 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
2435 spin_unlock(&mddev->lock);
2436 wake_up(&mddev->sb_wait);
2437 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
2438 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
2440 rdev_for_each(rdev, mddev) {
2441 if (test_and_clear_bit(FaultRecorded, &rdev->flags))
2442 clear_bit(Blocked, &rdev->flags);
2444 if (any_badblocks_changed)
2445 ack_all_badblocks(&rdev->badblocks);
2446 clear_bit(BlockedBadBlocks, &rdev->flags);
2447 wake_up(&rdev->blocked_wait);
2450 if (mddev_is_clustered(mddev) && ret == 0)
2451 md_cluster_ops->metadata_update_finish(mddev);
2453 EXPORT_SYMBOL(md_update_sb);
2455 static int add_bound_rdev(struct md_rdev *rdev)
2457 struct mddev *mddev = rdev->mddev;
2459 bool add_journal = test_bit(Journal, &rdev->flags);
2461 if (!mddev->pers->hot_remove_disk || add_journal) {
2462 /* If there is hot_add_disk but no hot_remove_disk
2463 * then added disks for geometry changes,
2464 * and should be added immediately.
2466 super_types[mddev->major_version].
2467 validate_super(mddev, rdev);
2469 mddev_suspend(mddev);
2470 err = mddev->pers->hot_add_disk(mddev, rdev);
2472 mddev_resume(mddev);
2474 unbind_rdev_from_array(rdev);
2479 sysfs_notify_dirent_safe(rdev->sysfs_state);
2481 set_bit(MD_CHANGE_DEVS, &mddev->flags);
2482 if (mddev->degraded)
2483 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
2484 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2485 md_new_event(mddev);
2486 md_wakeup_thread(mddev->thread);
2490 /* words written to sysfs files may, or may not, be \n terminated.
2491 * We want to accept with case. For this we use cmd_match.
2493 static int cmd_match(const char *cmd, const char *str)
2495 /* See if cmd, written into a sysfs file, matches
2496 * str. They must either be the same, or cmd can
2497 * have a trailing newline
2499 while (*cmd && *str && *cmd == *str) {
2510 struct rdev_sysfs_entry {
2511 struct attribute attr;
2512 ssize_t (*show)(struct md_rdev *, char *);
2513 ssize_t (*store)(struct md_rdev *, const char *, size_t);
2517 state_show(struct md_rdev *rdev, char *page)
2521 unsigned long flags = ACCESS_ONCE(rdev->flags);
2523 if (test_bit(Faulty, &flags) ||
2524 rdev->badblocks.unacked_exist) {
2525 len+= sprintf(page+len, "%sfaulty",sep);
2528 if (test_bit(In_sync, &flags)) {
2529 len += sprintf(page+len, "%sin_sync",sep);
2532 if (test_bit(Journal, &flags)) {
2533 len += sprintf(page+len, "%sjournal",sep);
2536 if (test_bit(WriteMostly, &flags)) {
2537 len += sprintf(page+len, "%swrite_mostly",sep);
2540 if (test_bit(Blocked, &flags) ||
2541 (rdev->badblocks.unacked_exist
2542 && !test_bit(Faulty, &flags))) {
2543 len += sprintf(page+len, "%sblocked", sep);
2546 if (!test_bit(Faulty, &flags) &&
2547 !test_bit(Journal, &flags) &&
2548 !test_bit(In_sync, &flags)) {
2549 len += sprintf(page+len, "%sspare", sep);
2552 if (test_bit(WriteErrorSeen, &flags)) {
2553 len += sprintf(page+len, "%swrite_error", sep);
2556 if (test_bit(WantReplacement, &flags)) {
2557 len += sprintf(page+len, "%swant_replacement", sep);
2560 if (test_bit(Replacement, &flags)) {
2561 len += sprintf(page+len, "%sreplacement", sep);
2565 return len+sprintf(page+len, "\n");
2569 state_store(struct md_rdev *rdev, const char *buf, size_t len)
2572 * faulty - simulates an error
2573 * remove - disconnects the device
2574 * writemostly - sets write_mostly
2575 * -writemostly - clears write_mostly
2576 * blocked - sets the Blocked flags
2577 * -blocked - clears the Blocked and possibly simulates an error
2578 * insync - sets Insync providing device isn't active
2579 * -insync - clear Insync for a device with a slot assigned,
2580 * so that it gets rebuilt based on bitmap
2581 * write_error - sets WriteErrorSeen
2582 * -write_error - clears WriteErrorSeen
2585 if (cmd_match(buf, "faulty") && rdev->mddev->pers) {
2586 md_error(rdev->mddev, rdev);
2587 if (test_bit(Faulty, &rdev->flags))
2591 } else if (cmd_match(buf, "remove")) {
2592 if (rdev->raid_disk >= 0)
2595 struct mddev *mddev = rdev->mddev;
2597 if (mddev_is_clustered(mddev))
2598 err = md_cluster_ops->remove_disk(mddev, rdev);
2601 md_kick_rdev_from_array(rdev);
2603 md_update_sb(mddev, 1);
2604 md_new_event(mddev);
2607 } else if (cmd_match(buf, "writemostly")) {
2608 set_bit(WriteMostly, &rdev->flags);
2610 } else if (cmd_match(buf, "-writemostly")) {
2611 clear_bit(WriteMostly, &rdev->flags);
2613 } else if (cmd_match(buf, "blocked")) {
2614 set_bit(Blocked, &rdev->flags);
2616 } else if (cmd_match(buf, "-blocked")) {
2617 if (!test_bit(Faulty, &rdev->flags) &&
2618 rdev->badblocks.unacked_exist) {
2619 /* metadata handler doesn't understand badblocks,
2620 * so we need to fail the device
2622 md_error(rdev->mddev, rdev);
2624 clear_bit(Blocked, &rdev->flags);
2625 clear_bit(BlockedBadBlocks, &rdev->flags);
2626 wake_up(&rdev->blocked_wait);
2627 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2628 md_wakeup_thread(rdev->mddev->thread);
2631 } else if (cmd_match(buf, "insync") && rdev->raid_disk == -1) {
2632 set_bit(In_sync, &rdev->flags);
2634 } else if (cmd_match(buf, "-insync") && rdev->raid_disk >= 0 &&
2635 !test_bit(Journal, &rdev->flags)) {
2636 if (rdev->mddev->pers == NULL) {
2637 clear_bit(In_sync, &rdev->flags);
2638 rdev->saved_raid_disk = rdev->raid_disk;
2639 rdev->raid_disk = -1;
2642 } else if (cmd_match(buf, "write_error")) {
2643 set_bit(WriteErrorSeen, &rdev->flags);
2645 } else if (cmd_match(buf, "-write_error")) {
2646 clear_bit(WriteErrorSeen, &rdev->flags);
2648 } else if (cmd_match(buf, "want_replacement")) {
2649 /* Any non-spare device that is not a replacement can
2650 * become want_replacement at any time, but we then need to
2651 * check if recovery is needed.
2653 if (rdev->raid_disk >= 0 &&
2654 !test_bit(Journal, &rdev->flags) &&
2655 !test_bit(Replacement, &rdev->flags))
2656 set_bit(WantReplacement, &rdev->flags);
2657 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2658 md_wakeup_thread(rdev->mddev->thread);
2660 } else if (cmd_match(buf, "-want_replacement")) {
2661 /* Clearing 'want_replacement' is always allowed.
2662 * Once replacements starts it is too late though.
2665 clear_bit(WantReplacement, &rdev->flags);
2666 } else if (cmd_match(buf, "replacement")) {
2667 /* Can only set a device as a replacement when array has not
2668 * yet been started. Once running, replacement is automatic
2669 * from spares, or by assigning 'slot'.
2671 if (rdev->mddev->pers)
2674 set_bit(Replacement, &rdev->flags);
2677 } else if (cmd_match(buf, "-replacement")) {
2678 /* Similarly, can only clear Replacement before start */
2679 if (rdev->mddev->pers)
2682 clear_bit(Replacement, &rdev->flags);
2685 } else if (cmd_match(buf, "re-add")) {
2686 if (test_bit(Faulty, &rdev->flags) && (rdev->raid_disk == -1)) {
2687 /* clear_bit is performed _after_ all the devices
2688 * have their local Faulty bit cleared. If any writes
2689 * happen in the meantime in the local node, they
2690 * will land in the local bitmap, which will be synced
2691 * by this node eventually
2693 if (!mddev_is_clustered(rdev->mddev) ||
2694 (err = md_cluster_ops->gather_bitmaps(rdev)) == 0) {
2695 clear_bit(Faulty, &rdev->flags);
2696 err = add_bound_rdev(rdev);
2702 sysfs_notify_dirent_safe(rdev->sysfs_state);
2703 return err ? err : len;
2705 static struct rdev_sysfs_entry rdev_state =
2706 __ATTR_PREALLOC(state, S_IRUGO|S_IWUSR, state_show, state_store);
2709 errors_show(struct md_rdev *rdev, char *page)
2711 return sprintf(page, "%d\n", atomic_read(&rdev->corrected_errors));
2715 errors_store(struct md_rdev *rdev, const char *buf, size_t len)
2720 rv = kstrtouint(buf, 10, &n);
2723 atomic_set(&rdev->corrected_errors, n);
2726 static struct rdev_sysfs_entry rdev_errors =
2727 __ATTR(errors, S_IRUGO|S_IWUSR, errors_show, errors_store);
2730 slot_show(struct md_rdev *rdev, char *page)
2732 if (test_bit(Journal, &rdev->flags))
2733 return sprintf(page, "journal\n");
2734 else if (rdev->raid_disk < 0)
2735 return sprintf(page, "none\n");
2737 return sprintf(page, "%d\n", rdev->raid_disk);
2741 slot_store(struct md_rdev *rdev, const char *buf, size_t len)
2746 if (test_bit(Journal, &rdev->flags))
2748 if (strncmp(buf, "none", 4)==0)
2751 err = kstrtouint(buf, 10, (unsigned int *)&slot);
2755 if (rdev->mddev->pers && slot == -1) {
2756 /* Setting 'slot' on an active array requires also
2757 * updating the 'rd%d' link, and communicating
2758 * with the personality with ->hot_*_disk.
2759 * For now we only support removing
2760 * failed/spare devices. This normally happens automatically,
2761 * but not when the metadata is externally managed.
2763 if (rdev->raid_disk == -1)
2765 /* personality does all needed checks */
2766 if (rdev->mddev->pers->hot_remove_disk == NULL)
2768 clear_bit(Blocked, &rdev->flags);
2769 remove_and_add_spares(rdev->mddev, rdev);
2770 if (rdev->raid_disk >= 0)
2772 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2773 md_wakeup_thread(rdev->mddev->thread);
2774 } else if (rdev->mddev->pers) {
2775 /* Activating a spare .. or possibly reactivating
2776 * if we ever get bitmaps working here.
2780 if (rdev->raid_disk != -1)
2783 if (test_bit(MD_RECOVERY_RUNNING, &rdev->mddev->recovery))
2786 if (rdev->mddev->pers->hot_add_disk == NULL)
2789 if (slot >= rdev->mddev->raid_disks &&
2790 slot >= rdev->mddev->raid_disks + rdev->mddev->delta_disks)
2793 rdev->raid_disk = slot;
2794 if (test_bit(In_sync, &rdev->flags))
2795 rdev->saved_raid_disk = slot;
2797 rdev->saved_raid_disk = -1;
2798 clear_bit(In_sync, &rdev->flags);
2799 clear_bit(Bitmap_sync, &rdev->flags);
2800 err = rdev->mddev->pers->
2801 hot_add_disk(rdev->mddev, rdev);
2803 rdev->raid_disk = -1;
2806 sysfs_notify_dirent_safe(rdev->sysfs_state);
2807 if (sysfs_link_rdev(rdev->mddev, rdev))
2808 /* failure here is OK */;
2809 /* don't wakeup anyone, leave that to userspace. */
2811 if (slot >= rdev->mddev->raid_disks &&
2812 slot >= rdev->mddev->raid_disks + rdev->mddev->delta_disks)
2814 rdev->raid_disk = slot;
2815 /* assume it is working */
2816 clear_bit(Faulty, &rdev->flags);
2817 clear_bit(WriteMostly, &rdev->flags);
2818 set_bit(In_sync, &rdev->flags);
2819 sysfs_notify_dirent_safe(rdev->sysfs_state);
2824 static struct rdev_sysfs_entry rdev_slot =
2825 __ATTR(slot, S_IRUGO|S_IWUSR, slot_show, slot_store);
2828 offset_show(struct md_rdev *rdev, char *page)
2830 return sprintf(page, "%llu\n", (unsigned long long)rdev->data_offset);
2834 offset_store(struct md_rdev *rdev, const char *buf, size_t len)
2836 unsigned long long offset;
2837 if (kstrtoull(buf, 10, &offset) < 0)
2839 if (rdev->mddev->pers && rdev->raid_disk >= 0)
2841 if (rdev->sectors && rdev->mddev->external)
2842 /* Must set offset before size, so overlap checks
2845 rdev->data_offset = offset;
2846 rdev->new_data_offset = offset;
2850 static struct rdev_sysfs_entry rdev_offset =
2851 __ATTR(offset, S_IRUGO|S_IWUSR, offset_show, offset_store);
2853 static ssize_t new_offset_show(struct md_rdev *rdev, char *page)
2855 return sprintf(page, "%llu\n",
2856 (unsigned long long)rdev->new_data_offset);
2859 static ssize_t new_offset_store(struct md_rdev *rdev,
2860 const char *buf, size_t len)
2862 unsigned long long new_offset;
2863 struct mddev *mddev = rdev->mddev;
2865 if (kstrtoull(buf, 10, &new_offset) < 0)
2868 if (mddev->sync_thread ||
2869 test_bit(MD_RECOVERY_RUNNING,&mddev->recovery))
2871 if (new_offset == rdev->data_offset)
2872 /* reset is always permitted */
2874 else if (new_offset > rdev->data_offset) {
2875 /* must not push array size beyond rdev_sectors */
2876 if (new_offset - rdev->data_offset
2877 + mddev->dev_sectors > rdev->sectors)
2880 /* Metadata worries about other space details. */
2882 /* decreasing the offset is inconsistent with a backwards
2885 if (new_offset < rdev->data_offset &&
2886 mddev->reshape_backwards)
2888 /* Increasing offset is inconsistent with forwards
2889 * reshape. reshape_direction should be set to
2890 * 'backwards' first.
2892 if (new_offset > rdev->data_offset &&
2893 !mddev->reshape_backwards)
2896 if (mddev->pers && mddev->persistent &&
2897 !super_types[mddev->major_version]
2898 .allow_new_offset(rdev, new_offset))
2900 rdev->new_data_offset = new_offset;
2901 if (new_offset > rdev->data_offset)
2902 mddev->reshape_backwards = 1;
2903 else if (new_offset < rdev->data_offset)
2904 mddev->reshape_backwards = 0;
2908 static struct rdev_sysfs_entry rdev_new_offset =
2909 __ATTR(new_offset, S_IRUGO|S_IWUSR, new_offset_show, new_offset_store);
2912 rdev_size_show(struct md_rdev *rdev, char *page)
2914 return sprintf(page, "%llu\n", (unsigned long long)rdev->sectors / 2);
2917 static int overlaps(sector_t s1, sector_t l1, sector_t s2, sector_t l2)
2919 /* check if two start/length pairs overlap */
2927 static int strict_blocks_to_sectors(const char *buf, sector_t *sectors)
2929 unsigned long long blocks;
2932 if (kstrtoull(buf, 10, &blocks) < 0)
2935 if (blocks & 1ULL << (8 * sizeof(blocks) - 1))
2936 return -EINVAL; /* sector conversion overflow */
2939 if (new != blocks * 2)
2940 return -EINVAL; /* unsigned long long to sector_t overflow */
2947 rdev_size_store(struct md_rdev *rdev, const char *buf, size_t len)
2949 struct mddev *my_mddev = rdev->mddev;
2950 sector_t oldsectors = rdev->sectors;
2953 if (test_bit(Journal, &rdev->flags))
2955 if (strict_blocks_to_sectors(buf, §ors) < 0)
2957 if (rdev->data_offset != rdev->new_data_offset)
2958 return -EINVAL; /* too confusing */
2959 if (my_mddev->pers && rdev->raid_disk >= 0) {
2960 if (my_mddev->persistent) {
2961 sectors = super_types[my_mddev->major_version].
2962 rdev_size_change(rdev, sectors);
2965 } else if (!sectors)
2966 sectors = (i_size_read(rdev->bdev->bd_inode) >> 9) -
2968 if (!my_mddev->pers->resize)
2969 /* Cannot change size for RAID0 or Linear etc */
2972 if (sectors < my_mddev->dev_sectors)
2973 return -EINVAL; /* component must fit device */
2975 rdev->sectors = sectors;
2976 if (sectors > oldsectors && my_mddev->external) {
2977 /* Need to check that all other rdevs with the same
2978 * ->bdev do not overlap. 'rcu' is sufficient to walk
2979 * the rdev lists safely.
2980 * This check does not provide a hard guarantee, it
2981 * just helps avoid dangerous mistakes.
2983 struct mddev *mddev;
2985 struct list_head *tmp;
2988 for_each_mddev(mddev, tmp) {
2989 struct md_rdev *rdev2;
2991 rdev_for_each(rdev2, mddev)
2992 if (rdev->bdev == rdev2->bdev &&
2994 overlaps(rdev->data_offset, rdev->sectors,
3007 /* Someone else could have slipped in a size
3008 * change here, but doing so is just silly.
3009 * We put oldsectors back because we *know* it is
3010 * safe, and trust userspace not to race with
3013 rdev->sectors = oldsectors;
3020 static struct rdev_sysfs_entry rdev_size =
3021 __ATTR(size, S_IRUGO|S_IWUSR, rdev_size_show, rdev_size_store);
3023 static ssize_t recovery_start_show(struct md_rdev *rdev, char *page)
3025 unsigned long long recovery_start = rdev->recovery_offset;
3027 if (test_bit(In_sync, &rdev->flags) ||
3028 recovery_start == MaxSector)
3029 return sprintf(page, "none\n");
3031 return sprintf(page, "%llu\n", recovery_start);
3034 static ssize_t recovery_start_store(struct md_rdev *rdev, const char *buf, size_t len)
3036 unsigned long long recovery_start;
3038 if (cmd_match(buf, "none"))
3039 recovery_start = MaxSector;
3040 else if (kstrtoull(buf, 10, &recovery_start))
3043 if (rdev->mddev->pers &&
3044 rdev->raid_disk >= 0)
3047 rdev->recovery_offset = recovery_start;
3048 if (recovery_start == MaxSector)
3049 set_bit(In_sync, &rdev->flags);
3051 clear_bit(In_sync, &rdev->flags);
3055 static struct rdev_sysfs_entry rdev_recovery_start =
3056 __ATTR(recovery_start, S_IRUGO|S_IWUSR, recovery_start_show, recovery_start_store);
3058 /* sysfs access to bad-blocks list.
3059 * We present two files.
3060 * 'bad-blocks' lists sector numbers and lengths of ranges that
3061 * are recorded as bad. The list is truncated to fit within
3062 * the one-page limit of sysfs.
3063 * Writing "sector length" to this file adds an acknowledged
3065 * 'unacknowledged-bad-blocks' lists bad blocks that have not yet
3066 * been acknowledged. Writing to this file adds bad blocks
3067 * without acknowledging them. This is largely for testing.
3069 static ssize_t bb_show(struct md_rdev *rdev, char *page)
3071 return badblocks_show(&rdev->badblocks, page, 0);
3073 static ssize_t bb_store(struct md_rdev *rdev, const char *page, size_t len)
3075 int rv = badblocks_store(&rdev->badblocks, page, len, 0);
3076 /* Maybe that ack was all we needed */
3077 if (test_and_clear_bit(BlockedBadBlocks, &rdev->flags))
3078 wake_up(&rdev->blocked_wait);
3081 static struct rdev_sysfs_entry rdev_bad_blocks =
3082 __ATTR(bad_blocks, S_IRUGO|S_IWUSR, bb_show, bb_store);
3084 static ssize_t ubb_show(struct md_rdev *rdev, char *page)
3086 return badblocks_show(&rdev->badblocks, page, 1);
3088 static ssize_t ubb_store(struct md_rdev *rdev, const char *page, size_t len)
3090 return badblocks_store(&rdev->badblocks, page, len, 1);
3092 static struct rdev_sysfs_entry rdev_unack_bad_blocks =
3093 __ATTR(unacknowledged_bad_blocks, S_IRUGO|S_IWUSR, ubb_show, ubb_store);
3095 static struct attribute *rdev_default_attrs[] = {
3100 &rdev_new_offset.attr,
3102 &rdev_recovery_start.attr,
3103 &rdev_bad_blocks.attr,
3104 &rdev_unack_bad_blocks.attr,
3108 rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
3110 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
3111 struct md_rdev *rdev = container_of(kobj, struct md_rdev, kobj);
3117 return entry->show(rdev, page);
3121 rdev_attr_store(struct kobject *kobj, struct attribute *attr,
3122 const char *page, size_t length)
3124 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
3125 struct md_rdev *rdev = container_of(kobj, struct md_rdev, kobj);
3127 struct mddev *mddev = rdev->mddev;
3131 if (!capable(CAP_SYS_ADMIN))
3133 rv = mddev ? mddev_lock(mddev): -EBUSY;
3135 if (rdev->mddev == NULL)
3138 rv = entry->store(rdev, page, length);
3139 mddev_unlock(mddev);
3144 static void rdev_free(struct kobject *ko)
3146 struct md_rdev *rdev = container_of(ko, struct md_rdev, kobj);
3149 static const struct sysfs_ops rdev_sysfs_ops = {
3150 .show = rdev_attr_show,
3151 .store = rdev_attr_store,
3153 static struct kobj_type rdev_ktype = {
3154 .release = rdev_free,
3155 .sysfs_ops = &rdev_sysfs_ops,
3156 .default_attrs = rdev_default_attrs,
3159 int md_rdev_init(struct md_rdev *rdev)
3162 rdev->saved_raid_disk = -1;
3163 rdev->raid_disk = -1;
3165 rdev->data_offset = 0;
3166 rdev->new_data_offset = 0;
3167 rdev->sb_events = 0;
3168 rdev->last_read_error.tv_sec = 0;
3169 rdev->last_read_error.tv_nsec = 0;
3170 rdev->sb_loaded = 0;
3171 rdev->bb_page = NULL;
3172 atomic_set(&rdev->nr_pending, 0);
3173 atomic_set(&rdev->read_errors, 0);
3174 atomic_set(&rdev->corrected_errors, 0);
3176 INIT_LIST_HEAD(&rdev->same_set);
3177 init_waitqueue_head(&rdev->blocked_wait);
3179 /* Add space to store bad block list.
3180 * This reserves the space even on arrays where it cannot
3181 * be used - I wonder if that matters
3183 return badblocks_init(&rdev->badblocks, 0);
3185 EXPORT_SYMBOL_GPL(md_rdev_init);
3187 * Import a device. If 'super_format' >= 0, then sanity check the superblock
3189 * mark the device faulty if:
3191 * - the device is nonexistent (zero size)
3192 * - the device has no valid superblock
3194 * a faulty rdev _never_ has rdev->sb set.
3196 static struct md_rdev *md_import_device(dev_t newdev, int super_format, int super_minor)
3198 char b[BDEVNAME_SIZE];
3200 struct md_rdev *rdev;
3203 rdev = kzalloc(sizeof(*rdev), GFP_KERNEL);
3205 printk(KERN_ERR "md: could not alloc mem for new device!\n");
3206 return ERR_PTR(-ENOMEM);
3209 err = md_rdev_init(rdev);
3212 err = alloc_disk_sb(rdev);
3216 err = lock_rdev(rdev, newdev, super_format == -2);
3220 kobject_init(&rdev->kobj, &rdev_ktype);
3222 size = i_size_read(rdev->bdev->bd_inode) >> BLOCK_SIZE_BITS;
3225 "md: %s has zero or unknown size, marking faulty!\n",
3226 bdevname(rdev->bdev,b));
3231 if (super_format >= 0) {
3232 err = super_types[super_format].
3233 load_super(rdev, NULL, super_minor);
3234 if (err == -EINVAL) {
3236 "md: %s does not have a valid v%d.%d "
3237 "superblock, not importing!\n",
3238 bdevname(rdev->bdev,b),
3239 super_format, super_minor);
3244 "md: could not read %s's sb, not importing!\n",
3245 bdevname(rdev->bdev,b));
3255 md_rdev_clear(rdev);
3257 return ERR_PTR(err);
3261 * Check a full RAID array for plausibility
3264 static void analyze_sbs(struct mddev *mddev)
3267 struct md_rdev *rdev, *freshest, *tmp;
3268 char b[BDEVNAME_SIZE];
3271 rdev_for_each_safe(rdev, tmp, mddev)
3272 switch (super_types[mddev->major_version].
3273 load_super(rdev, freshest, mddev->minor_version)) {
3281 "md: fatal superblock inconsistency in %s"
3282 " -- removing from array\n",
3283 bdevname(rdev->bdev,b));
3284 md_kick_rdev_from_array(rdev);
3287 super_types[mddev->major_version].
3288 validate_super(mddev, freshest);
3291 rdev_for_each_safe(rdev, tmp, mddev) {
3292 if (mddev->max_disks &&
3293 (rdev->desc_nr >= mddev->max_disks ||
3294 i > mddev->max_disks)) {
3296 "md: %s: %s: only %d devices permitted\n",
3297 mdname(mddev), bdevname(rdev->bdev, b),
3299 md_kick_rdev_from_array(rdev);
3302 if (rdev != freshest) {
3303 if (super_types[mddev->major_version].
3304 validate_super(mddev, rdev)) {
3305 printk(KERN_WARNING "md: kicking non-fresh %s"
3307 bdevname(rdev->bdev,b));
3308 md_kick_rdev_from_array(rdev);
3312 if (mddev->level == LEVEL_MULTIPATH) {
3313 rdev->desc_nr = i++;
3314 rdev->raid_disk = rdev->desc_nr;
3315 set_bit(In_sync, &rdev->flags);
3316 } else if (rdev->raid_disk >=
3317 (mddev->raid_disks - min(0, mddev->delta_disks)) &&
3318 !test_bit(Journal, &rdev->flags)) {
3319 rdev->raid_disk = -1;
3320 clear_bit(In_sync, &rdev->flags);
3325 /* Read a fixed-point number.
3326 * Numbers in sysfs attributes should be in "standard" units where
3327 * possible, so time should be in seconds.
3328 * However we internally use a a much smaller unit such as
3329 * milliseconds or jiffies.
3330 * This function takes a decimal number with a possible fractional
3331 * component, and produces an integer which is the result of
3332 * multiplying that number by 10^'scale'.
3333 * all without any floating-point arithmetic.
3335 int strict_strtoul_scaled(const char *cp, unsigned long *res, int scale)
3337 unsigned long result = 0;
3339 while (isdigit(*cp) || (*cp == '.' && decimals < 0)) {
3342 else if (decimals < scale) {
3345 result = result * 10 + value;
3357 while (decimals < scale) {
3366 safe_delay_show(struct mddev *mddev, char *page)
3368 int msec = (mddev->safemode_delay*1000)/HZ;
3369 return sprintf(page, "%d.%03d\n", msec/1000, msec%1000);
3372 safe_delay_store(struct mddev *mddev, const char *cbuf, size_t len)
3376 if (mddev_is_clustered(mddev)) {
3377 pr_info("md: Safemode is disabled for clustered mode\n");
3381 if (strict_strtoul_scaled(cbuf, &msec, 3) < 0)
3384 mddev->safemode_delay = 0;
3386 unsigned long old_delay = mddev->safemode_delay;
3387 unsigned long new_delay = (msec*HZ)/1000;
3391 mddev->safemode_delay = new_delay;
3392 if (new_delay < old_delay || old_delay == 0)
3393 mod_timer(&mddev->safemode_timer, jiffies+1);
3397 static struct md_sysfs_entry md_safe_delay =
3398 __ATTR(safe_mode_delay, S_IRUGO|S_IWUSR,safe_delay_show, safe_delay_store);
3401 level_show(struct mddev *mddev, char *page)
3403 struct md_personality *p;
3405 spin_lock(&mddev->lock);
3408 ret = sprintf(page, "%s\n", p->name);
3409 else if (mddev->clevel[0])
3410 ret = sprintf(page, "%s\n", mddev->clevel);
3411 else if (mddev->level != LEVEL_NONE)
3412 ret = sprintf(page, "%d\n", mddev->level);
3415 spin_unlock(&mddev->lock);
3420 level_store(struct mddev *mddev, const char *buf, size_t len)
3425 struct md_personality *pers, *oldpers;
3427 void *priv, *oldpriv;
3428 struct md_rdev *rdev;
3430 if (slen == 0 || slen >= sizeof(clevel))
3433 rv = mddev_lock(mddev);
3437 if (mddev->pers == NULL) {
3438 strncpy(mddev->clevel, buf, slen);
3439 if (mddev->clevel[slen-1] == '\n')
3441 mddev->clevel[slen] = 0;
3442 mddev->level = LEVEL_NONE;
3450 /* request to change the personality. Need to ensure:
3451 * - array is not engaged in resync/recovery/reshape
3452 * - old personality can be suspended
3453 * - new personality will access other array.
3457 if (mddev->sync_thread ||
3458 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
3459 mddev->reshape_position != MaxSector ||
3460 mddev->sysfs_active)
3464 if (!mddev->pers->quiesce) {
3465 printk(KERN_WARNING "md: %s: %s does not support online personality change\n",
3466 mdname(mddev), mddev->pers->name);
3470 /* Now find the new personality */
3471 strncpy(clevel, buf, slen);
3472 if (clevel[slen-1] == '\n')
3475 if (kstrtol(clevel, 10, &level))
3478 if (request_module("md-%s", clevel) != 0)
3479 request_module("md-level-%s", clevel);
3480 spin_lock(&pers_lock);
3481 pers = find_pers(level, clevel);
3482 if (!pers || !try_module_get(pers->owner)) {
3483 spin_unlock(&pers_lock);
3484 printk(KERN_WARNING "md: personality %s not loaded\n", clevel);
3488 spin_unlock(&pers_lock);
3490 if (pers == mddev->pers) {
3491 /* Nothing to do! */
3492 module_put(pers->owner);
3496 if (!pers->takeover) {
3497 module_put(pers->owner);
3498 printk(KERN_WARNING "md: %s: %s does not support personality takeover\n",
3499 mdname(mddev), clevel);
3504 rdev_for_each(rdev, mddev)
3505 rdev->new_raid_disk = rdev->raid_disk;
3507 /* ->takeover must set new_* and/or delta_disks
3508 * if it succeeds, and may set them when it fails.
3510 priv = pers->takeover(mddev);
3512 mddev->new_level = mddev->level;
3513 mddev->new_layout = mddev->layout;
3514 mddev->new_chunk_sectors = mddev->chunk_sectors;
3515 mddev->raid_disks -= mddev->delta_disks;
3516 mddev->delta_disks = 0;
3517 mddev->reshape_backwards = 0;
3518 module_put(pers->owner);
3519 printk(KERN_WARNING "md: %s: %s would not accept array\n",
3520 mdname(mddev), clevel);
3525 /* Looks like we have a winner */
3526 mddev_suspend(mddev);
3527 mddev_detach(mddev);
3529 spin_lock(&mddev->lock);
3530 oldpers = mddev->pers;
3531 oldpriv = mddev->private;
3533 mddev->private = priv;
3534 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
3535 mddev->level = mddev->new_level;
3536 mddev->layout = mddev->new_layout;
3537 mddev->chunk_sectors = mddev->new_chunk_sectors;
3538 mddev->delta_disks = 0;
3539 mddev->reshape_backwards = 0;
3540 mddev->degraded = 0;
3541 spin_unlock(&mddev->lock);
3543 if (oldpers->sync_request == NULL &&
3545 /* We are converting from a no-redundancy array
3546 * to a redundancy array and metadata is managed
3547 * externally so we need to be sure that writes
3548 * won't block due to a need to transition
3550 * until external management is started.
3553 mddev->safemode_delay = 0;
3554 mddev->safemode = 0;
3557 oldpers->free(mddev, oldpriv);
3559 if (oldpers->sync_request == NULL &&
3560 pers->sync_request != NULL) {
3561 /* need to add the md_redundancy_group */
3562 if (sysfs_create_group(&mddev->kobj, &md_redundancy_group))
3564 "md: cannot register extra attributes for %s\n",
3566 mddev->sysfs_action = sysfs_get_dirent(mddev->kobj.sd, "sync_action");
3568 if (oldpers->sync_request != NULL &&
3569 pers->sync_request == NULL) {
3570 /* need to remove the md_redundancy_group */
3571 if (mddev->to_remove == NULL)
3572 mddev->to_remove = &md_redundancy_group;
3575 rdev_for_each(rdev, mddev) {
3576 if (rdev->raid_disk < 0)
3578 if (rdev->new_raid_disk >= mddev->raid_disks)
3579 rdev->new_raid_disk = -1;
3580 if (rdev->new_raid_disk == rdev->raid_disk)
3582 sysfs_unlink_rdev(mddev, rdev);
3584 rdev_for_each(rdev, mddev) {
3585 if (rdev->raid_disk < 0)
3587 if (rdev->new_raid_disk == rdev->raid_disk)
3589 rdev->raid_disk = rdev->new_raid_disk;
3590 if (rdev->raid_disk < 0)
3591 clear_bit(In_sync, &rdev->flags);
3593 if (sysfs_link_rdev(mddev, rdev))
3594 printk(KERN_WARNING "md: cannot register rd%d"
3595 " for %s after level change\n",
3596 rdev->raid_disk, mdname(mddev));
3600 if (pers->sync_request == NULL) {
3601 /* this is now an array without redundancy, so
3602 * it must always be in_sync
3605 del_timer_sync(&mddev->safemode_timer);
3607 blk_set_stacking_limits(&mddev->queue->limits);
3609 set_bit(MD_CHANGE_DEVS, &mddev->flags);
3610 mddev_resume(mddev);
3612 md_update_sb(mddev, 1);
3613 sysfs_notify(&mddev->kobj, NULL, "level");
3614 md_new_event(mddev);
3617 mddev_unlock(mddev);
3621 static struct md_sysfs_entry md_level =
3622 __ATTR(level, S_IRUGO|S_IWUSR, level_show, level_store);
3625 layout_show(struct mddev *mddev, char *page)
3627 /* just a number, not meaningful for all levels */
3628 if (mddev->reshape_position != MaxSector &&
3629 mddev->layout != mddev->new_layout)
3630 return sprintf(page, "%d (%d)\n",
3631 mddev->new_layout, mddev->layout);
3632 return sprintf(page, "%d\n", mddev->layout);
3636 layout_store(struct mddev *mddev, const char *buf, size_t len)
3641 err = kstrtouint(buf, 10, &n);
3644 err = mddev_lock(mddev);
3649 if (mddev->pers->check_reshape == NULL)
3654 mddev->new_layout = n;
3655 err = mddev->pers->check_reshape(mddev);
3657 mddev->new_layout = mddev->layout;
3660 mddev->new_layout = n;
3661 if (mddev->reshape_position == MaxSector)
3664 mddev_unlock(mddev);
3667 static struct md_sysfs_entry md_layout =
3668 __ATTR(layout, S_IRUGO|S_IWUSR, layout_show, layout_store);
3671 raid_disks_show(struct mddev *mddev, char *page)
3673 if (mddev->raid_disks == 0)
3675 if (mddev->reshape_position != MaxSector &&
3676 mddev->delta_disks != 0)
3677 return sprintf(page, "%d (%d)\n", mddev->raid_disks,
3678 mddev->raid_disks - mddev->delta_disks);
3679 return sprintf(page, "%d\n", mddev->raid_disks);
3682 static int update_raid_disks(struct mddev *mddev, int raid_disks);
3685 raid_disks_store(struct mddev *mddev, const char *buf, size_t len)
3690 err = kstrtouint(buf, 10, &n);
3694 err = mddev_lock(mddev);
3698 err = update_raid_disks(mddev, n);
3699 else if (mddev->reshape_position != MaxSector) {
3700 struct md_rdev *rdev;
3701 int olddisks = mddev->raid_disks - mddev->delta_disks;
3704 rdev_for_each(rdev, mddev) {
3706 rdev->data_offset < rdev->new_data_offset)
3709 rdev->data_offset > rdev->new_data_offset)
3713 mddev->delta_disks = n - olddisks;
3714 mddev->raid_disks = n;
3715 mddev->reshape_backwards = (mddev->delta_disks < 0);
3717 mddev->raid_disks = n;
3719 mddev_unlock(mddev);
3720 return err ? err : len;
3722 static struct md_sysfs_entry md_raid_disks =
3723 __ATTR(raid_disks, S_IRUGO|S_IWUSR, raid_disks_show, raid_disks_store);
3726 chunk_size_show(struct mddev *mddev, char *page)
3728 if (mddev->reshape_position != MaxSector &&
3729 mddev->chunk_sectors != mddev->new_chunk_sectors)
3730 return sprintf(page, "%d (%d)\n",
3731 mddev->new_chunk_sectors << 9,
3732 mddev->chunk_sectors << 9);
3733 return sprintf(page, "%d\n", mddev->chunk_sectors << 9);
3737 chunk_size_store(struct mddev *mddev, const char *buf, size_t len)
3742 err = kstrtoul(buf, 10, &n);
3746 err = mddev_lock(mddev);
3750 if (mddev->pers->check_reshape == NULL)
3755 mddev->new_chunk_sectors = n >> 9;
3756 err = mddev->pers->check_reshape(mddev);
3758 mddev->new_chunk_sectors = mddev->chunk_sectors;
3761 mddev->new_chunk_sectors = n >> 9;
3762 if (mddev->reshape_position == MaxSector)
3763 mddev->chunk_sectors = n >> 9;
3765 mddev_unlock(mddev);
3768 static struct md_sysfs_entry md_chunk_size =
3769 __ATTR(chunk_size, S_IRUGO|S_IWUSR, chunk_size_show, chunk_size_store);
3772 resync_start_show(struct mddev *mddev, char *page)
3774 if (mddev->recovery_cp == MaxSector)
3775 return sprintf(page, "none\n");
3776 return sprintf(page, "%llu\n", (unsigned long long)mddev->recovery_cp);
3780 resync_start_store(struct mddev *mddev, const char *buf, size_t len)
3782 unsigned long long n;
3785 if (cmd_match(buf, "none"))
3788 err = kstrtoull(buf, 10, &n);
3791 if (n != (sector_t)n)
3795 err = mddev_lock(mddev);
3798 if (mddev->pers && !test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
3802 mddev->recovery_cp = n;
3804 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
3806 mddev_unlock(mddev);
3809 static struct md_sysfs_entry md_resync_start =
3810 __ATTR_PREALLOC(resync_start, S_IRUGO|S_IWUSR,
3811 resync_start_show, resync_start_store);
3814 * The array state can be:
3817 * No devices, no size, no level
3818 * Equivalent to STOP_ARRAY ioctl
3820 * May have some settings, but array is not active
3821 * all IO results in error
3822 * When written, doesn't tear down array, but just stops it
3823 * suspended (not supported yet)
3824 * All IO requests will block. The array can be reconfigured.
3825 * Writing this, if accepted, will block until array is quiescent
3827 * no resync can happen. no superblocks get written.
3828 * write requests fail
3830 * like readonly, but behaves like 'clean' on a write request.
3832 * clean - no pending writes, but otherwise active.
3833 * When written to inactive array, starts without resync
3834 * If a write request arrives then
3835 * if metadata is known, mark 'dirty' and switch to 'active'.
3836 * if not known, block and switch to write-pending
3837 * If written to an active array that has pending writes, then fails.
3839 * fully active: IO and resync can be happening.
3840 * When written to inactive array, starts with resync
3843 * clean, but writes are blocked waiting for 'active' to be written.
3846 * like active, but no writes have been seen for a while (100msec).
3849 enum array_state { clear, inactive, suspended, readonly, read_auto, clean, active,
3850 write_pending, active_idle, bad_word};
3851 static char *array_states[] = {
3852 "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active",
3853 "write-pending", "active-idle", NULL };
3855 static int match_word(const char *word, char **list)
3858 for (n=0; list[n]; n++)
3859 if (cmd_match(word, list[n]))
3865 array_state_show(struct mddev *mddev, char *page)
3867 enum array_state st = inactive;
3880 else if (test_bit(MD_CHANGE_PENDING, &mddev->flags))
3882 else if (mddev->safemode)
3888 if (list_empty(&mddev->disks) &&
3889 mddev->raid_disks == 0 &&
3890 mddev->dev_sectors == 0)
3895 return sprintf(page, "%s\n", array_states[st]);
3898 static int do_md_stop(struct mddev *mddev, int ro, struct block_device *bdev);
3899 static int md_set_readonly(struct mddev *mddev, struct block_device *bdev);
3900 static int do_md_run(struct mddev *mddev);
3901 static int restart_array(struct mddev *mddev);
3904 array_state_store(struct mddev *mddev, const char *buf, size_t len)
3907 enum array_state st = match_word(buf, array_states);
3909 if (mddev->pers && (st == active || st == clean) && mddev->ro != 1) {
3910 /* don't take reconfig_mutex when toggling between
3913 spin_lock(&mddev->lock);
3915 restart_array(mddev);
3916 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
3917 wake_up(&mddev->sb_wait);
3919 } else /* st == clean */ {
3920 restart_array(mddev);
3921 if (atomic_read(&mddev->writes_pending) == 0) {
3922 if (mddev->in_sync == 0) {
3924 if (mddev->safemode == 1)
3925 mddev->safemode = 0;
3926 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
3932 spin_unlock(&mddev->lock);
3935 err = mddev_lock(mddev);
3943 /* stopping an active array */
3944 err = do_md_stop(mddev, 0, NULL);
3947 /* stopping an active array */
3949 err = do_md_stop(mddev, 2, NULL);
3951 err = 0; /* already inactive */
3954 break; /* not supported yet */
3957 err = md_set_readonly(mddev, NULL);
3960 set_disk_ro(mddev->gendisk, 1);
3961 err = do_md_run(mddev);
3967 err = md_set_readonly(mddev, NULL);
3968 else if (mddev->ro == 1)
3969 err = restart_array(mddev);
3972 set_disk_ro(mddev->gendisk, 0);
3976 err = do_md_run(mddev);
3981 err = restart_array(mddev);
3984 spin_lock(&mddev->lock);
3985 if (atomic_read(&mddev->writes_pending) == 0) {
3986 if (mddev->in_sync == 0) {
3988 if (mddev->safemode == 1)
3989 mddev->safemode = 0;
3990 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
3995 spin_unlock(&mddev->lock);
4001 err = restart_array(mddev);
4004 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
4005 wake_up(&mddev->sb_wait);
4009 set_disk_ro(mddev->gendisk, 0);
4010 err = do_md_run(mddev);
4015 /* these cannot be set */
4020 if (mddev->hold_active == UNTIL_IOCTL)
4021 mddev->hold_active = 0;
4022 sysfs_notify_dirent_safe(mddev->sysfs_state);
4024 mddev_unlock(mddev);
4027 static struct md_sysfs_entry md_array_state =
4028 __ATTR_PREALLOC(array_state, S_IRUGO|S_IWUSR, array_state_show, array_state_store);
4031 max_corrected_read_errors_show(struct mddev *mddev, char *page) {
4032 return sprintf(page, "%d\n",
4033 atomic_read(&mddev->max_corr_read_errors));
4037 max_corrected_read_errors_store(struct mddev *mddev, const char *buf, size_t len)
4042 rv = kstrtouint(buf, 10, &n);
4045 atomic_set(&mddev->max_corr_read_errors, n);
4049 static struct md_sysfs_entry max_corr_read_errors =
4050 __ATTR(max_read_errors, S_IRUGO|S_IWUSR, max_corrected_read_errors_show,
4051 max_corrected_read_errors_store);
4054 null_show(struct mddev *mddev, char *page)
4060 new_dev_store(struct mddev *mddev, const char *buf, size_t len)
4062 /* buf must be %d:%d\n? giving major and minor numbers */
4063 /* The new device is added to the array.
4064 * If the array has a persistent superblock, we read the
4065 * superblock to initialise info and check validity.
4066 * Otherwise, only checking done is that in bind_rdev_to_array,
4067 * which mainly checks size.
4070 int major = simple_strtoul(buf, &e, 10);
4073 struct md_rdev *rdev;
4076 if (!*buf || *e != ':' || !e[1] || e[1] == '\n')
4078 minor = simple_strtoul(e+1, &e, 10);
4079 if (*e && *e != '\n')
4081 dev = MKDEV(major, minor);
4082 if (major != MAJOR(dev) ||
4083 minor != MINOR(dev))
4086 flush_workqueue(md_misc_wq);
4088 err = mddev_lock(mddev);
4091 if (mddev->persistent) {
4092 rdev = md_import_device(dev, mddev->major_version,
4093 mddev->minor_version);
4094 if (!IS_ERR(rdev) && !list_empty(&mddev->disks)) {
4095 struct md_rdev *rdev0
4096 = list_entry(mddev->disks.next,
4097 struct md_rdev, same_set);
4098 err = super_types[mddev->major_version]
4099 .load_super(rdev, rdev0, mddev->minor_version);
4103 } else if (mddev->external)
4104 rdev = md_import_device(dev, -2, -1);
4106 rdev = md_import_device(dev, -1, -1);
4109 mddev_unlock(mddev);
4110 return PTR_ERR(rdev);
4112 err = bind_rdev_to_array(rdev, mddev);
4116 mddev_unlock(mddev);
4117 return err ? err : len;
4120 static struct md_sysfs_entry md_new_device =
4121 __ATTR(new_dev, S_IWUSR, null_show, new_dev_store);
4124 bitmap_store(struct mddev *mddev, const char *buf, size_t len)
4127 unsigned long chunk, end_chunk;
4130 err = mddev_lock(mddev);
4135 /* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */
4137 chunk = end_chunk = simple_strtoul(buf, &end, 0);
4138 if (buf == end) break;
4139 if (*end == '-') { /* range */
4141 end_chunk = simple_strtoul(buf, &end, 0);
4142 if (buf == end) break;
4144 if (*end && !isspace(*end)) break;
4145 bitmap_dirty_bits(mddev->bitmap, chunk, end_chunk);
4146 buf = skip_spaces(end);
4148 bitmap_unplug(mddev->bitmap); /* flush the bits to disk */
4150 mddev_unlock(mddev);
4154 static struct md_sysfs_entry md_bitmap =
4155 __ATTR(bitmap_set_bits, S_IWUSR, null_show, bitmap_store);
4158 size_show(struct mddev *mddev, char *page)
4160 return sprintf(page, "%llu\n",
4161 (unsigned long long)mddev->dev_sectors / 2);
4164 static int update_size(struct mddev *mddev, sector_t num_sectors);
4167 size_store(struct mddev *mddev, const char *buf, size_t len)
4169 /* If array is inactive, we can reduce the component size, but
4170 * not increase it (except from 0).
4171 * If array is active, we can try an on-line resize
4174 int err = strict_blocks_to_sectors(buf, §ors);
4178 err = mddev_lock(mddev);
4182 err = update_size(mddev, sectors);
4183 md_update_sb(mddev, 1);
4185 if (mddev->dev_sectors == 0 ||
4186 mddev->dev_sectors > sectors)
4187 mddev->dev_sectors = sectors;
4191 mddev_unlock(mddev);
4192 return err ? err : len;
4195 static struct md_sysfs_entry md_size =
4196 __ATTR(component_size, S_IRUGO|S_IWUSR, size_show, size_store);
4198 /* Metadata version.
4200 * 'none' for arrays with no metadata (good luck...)
4201 * 'external' for arrays with externally managed metadata,
4202 * or N.M for internally known formats
4205 metadata_show(struct mddev *mddev, char *page)
4207 if (mddev->persistent)
4208 return sprintf(page, "%d.%d\n",
4209 mddev->major_version, mddev->minor_version);
4210 else if (mddev->external)
4211 return sprintf(page, "external:%s\n", mddev->metadata_type);
4213 return sprintf(page, "none\n");
4217 metadata_store(struct mddev *mddev, const char *buf, size_t len)
4222 /* Changing the details of 'external' metadata is
4223 * always permitted. Otherwise there must be
4224 * no devices attached to the array.
4227 err = mddev_lock(mddev);
4231 if (mddev->external && strncmp(buf, "external:", 9) == 0)
4233 else if (!list_empty(&mddev->disks))
4237 if (cmd_match(buf, "none")) {
4238 mddev->persistent = 0;
4239 mddev->external = 0;
4240 mddev->major_version = 0;
4241 mddev->minor_version = 90;
4244 if (strncmp(buf, "external:", 9) == 0) {
4245 size_t namelen = len-9;
4246 if (namelen >= sizeof(mddev->metadata_type))
4247 namelen = sizeof(mddev->metadata_type)-1;
4248 strncpy(mddev->metadata_type, buf+9, namelen);
4249 mddev->metadata_type[namelen] = 0;
4250 if (namelen && mddev->metadata_type[namelen-1] == '\n')
4251 mddev->metadata_type[--namelen] = 0;
4252 mddev->persistent = 0;
4253 mddev->external = 1;
4254 mddev->major_version = 0;
4255 mddev->minor_version = 90;
4258 major = simple_strtoul(buf, &e, 10);
4260 if (e==buf || *e != '.')
4263 minor = simple_strtoul(buf, &e, 10);
4264 if (e==buf || (*e && *e != '\n') )
4267 if (major >= ARRAY_SIZE(super_types) || super_types[major].name == NULL)
4269 mddev->major_version = major;
4270 mddev->minor_version = minor;
4271 mddev->persistent = 1;
4272 mddev->external = 0;
4275 mddev_unlock(mddev);
4279 static struct md_sysfs_entry md_metadata =
4280 __ATTR_PREALLOC(metadata_version, S_IRUGO|S_IWUSR, metadata_show, metadata_store);
4283 action_show(struct mddev *mddev, char *page)
4285 char *type = "idle";
4286 unsigned long recovery = mddev->recovery;
4287 if (test_bit(MD_RECOVERY_FROZEN, &recovery))
4289 else if (test_bit(MD_RECOVERY_RUNNING, &recovery) ||
4290 (!mddev->ro && test_bit(MD_RECOVERY_NEEDED, &recovery))) {
4291 if (test_bit(MD_RECOVERY_RESHAPE, &recovery))
4293 else if (test_bit(MD_RECOVERY_SYNC, &recovery)) {
4294 if (!test_bit(MD_RECOVERY_REQUESTED, &recovery))
4296 else if (test_bit(MD_RECOVERY_CHECK, &recovery))
4300 } else if (test_bit(MD_RECOVERY_RECOVER, &recovery))
4302 else if (mddev->reshape_position != MaxSector)
4305 return sprintf(page, "%s\n", type);
4309 action_store(struct mddev *mddev, const char *page, size_t len)
4311 if (!mddev->pers || !mddev->pers->sync_request)
4315 if (cmd_match(page, "idle") || cmd_match(page, "frozen")) {
4316 if (cmd_match(page, "frozen"))
4317 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4319 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4320 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
4321 mddev_lock(mddev) == 0) {
4322 flush_workqueue(md_misc_wq);
4323 if (mddev->sync_thread) {
4324 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
4325 md_reap_sync_thread(mddev);
4327 mddev_unlock(mddev);
4329 } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4331 else if (cmd_match(page, "resync"))
4332 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4333 else if (cmd_match(page, "recover")) {
4334 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4335 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
4336 } else if (cmd_match(page, "reshape")) {
4338 if (mddev->pers->start_reshape == NULL)
4340 err = mddev_lock(mddev);
4342 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4345 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4346 err = mddev->pers->start_reshape(mddev);
4348 mddev_unlock(mddev);
4352 sysfs_notify(&mddev->kobj, NULL, "degraded");
4354 if (cmd_match(page, "check"))
4355 set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
4356 else if (!cmd_match(page, "repair"))
4358 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4359 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
4360 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
4362 if (mddev->ro == 2) {
4363 /* A write to sync_action is enough to justify
4364 * canceling read-auto mode
4367 md_wakeup_thread(mddev->sync_thread);
4369 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4370 md_wakeup_thread(mddev->thread);
4371 sysfs_notify_dirent_safe(mddev->sysfs_action);
4375 static struct md_sysfs_entry md_scan_mode =
4376 __ATTR_PREALLOC(sync_action, S_IRUGO|S_IWUSR, action_show, action_store);
4379 last_sync_action_show(struct mddev *mddev, char *page)
4381 return sprintf(page, "%s\n", mddev->last_sync_action);
4384 static struct md_sysfs_entry md_last_scan_mode = __ATTR_RO(last_sync_action);
4387 mismatch_cnt_show(struct mddev *mddev, char *page)
4389 return sprintf(page, "%llu\n",
4390 (unsigned long long)
4391 atomic64_read(&mddev->resync_mismatches));
4394 static struct md_sysfs_entry md_mismatches = __ATTR_RO(mismatch_cnt);
4397 sync_min_show(struct mddev *mddev, char *page)
4399 return sprintf(page, "%d (%s)\n", speed_min(mddev),
4400 mddev->sync_speed_min ? "local": "system");
4404 sync_min_store(struct mddev *mddev, const char *buf, size_t len)
4409 if (strncmp(buf, "system", 6)==0) {
4412 rv = kstrtouint(buf, 10, &min);
4418 mddev->sync_speed_min = min;
4422 static struct md_sysfs_entry md_sync_min =
4423 __ATTR(sync_speed_min, S_IRUGO|S_IWUSR, sync_min_show, sync_min_store);
4426 sync_max_show(struct mddev *mddev, char *page)
4428 return sprintf(page, "%d (%s)\n", speed_max(mddev),
4429 mddev->sync_speed_max ? "local": "system");
4433 sync_max_store(struct mddev *mddev, const char *buf, size_t len)
4438 if (strncmp(buf, "system", 6)==0) {
4441 rv = kstrtouint(buf, 10, &max);
4447 mddev->sync_speed_max = max;
4451 static struct md_sysfs_entry md_sync_max =
4452 __ATTR(sync_speed_max, S_IRUGO|S_IWUSR, sync_max_show, sync_max_store);
4455 degraded_show(struct mddev *mddev, char *page)
4457 return sprintf(page, "%d\n", mddev->degraded);
4459 static struct md_sysfs_entry md_degraded = __ATTR_RO(degraded);
4462 sync_force_parallel_show(struct mddev *mddev, char *page)
4464 return sprintf(page, "%d\n", mddev->parallel_resync);
4468 sync_force_parallel_store(struct mddev *mddev, const char *buf, size_t len)
4472 if (kstrtol(buf, 10, &n))
4475 if (n != 0 && n != 1)
4478 mddev->parallel_resync = n;
4480 if (mddev->sync_thread)
4481 wake_up(&resync_wait);
4486 /* force parallel resync, even with shared block devices */
4487 static struct md_sysfs_entry md_sync_force_parallel =
4488 __ATTR(sync_force_parallel, S_IRUGO|S_IWUSR,
4489 sync_force_parallel_show, sync_force_parallel_store);
4492 sync_speed_show(struct mddev *mddev, char *page)
4494 unsigned long resync, dt, db;
4495 if (mddev->curr_resync == 0)
4496 return sprintf(page, "none\n");
4497 resync = mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active);
4498 dt = (jiffies - mddev->resync_mark) / HZ;
4500 db = resync - mddev->resync_mark_cnt;
4501 return sprintf(page, "%lu\n", db/dt/2); /* K/sec */
4504 static struct md_sysfs_entry md_sync_speed = __ATTR_RO(sync_speed);
4507 sync_completed_show(struct mddev *mddev, char *page)
4509 unsigned long long max_sectors, resync;
4511 if (!test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4512 return sprintf(page, "none\n");
4514 if (mddev->curr_resync == 1 ||
4515 mddev->curr_resync == 2)
4516 return sprintf(page, "delayed\n");
4518 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ||
4519 test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
4520 max_sectors = mddev->resync_max_sectors;
4522 max_sectors = mddev->dev_sectors;
4524 resync = mddev->curr_resync_completed;
4525 return sprintf(page, "%llu / %llu\n", resync, max_sectors);
4528 static struct md_sysfs_entry md_sync_completed =
4529 __ATTR_PREALLOC(sync_completed, S_IRUGO, sync_completed_show, NULL);
4532 min_sync_show(struct mddev *mddev, char *page)
4534 return sprintf(page, "%llu\n",
4535 (unsigned long long)mddev->resync_min);
4538 min_sync_store(struct mddev *mddev, const char *buf, size_t len)
4540 unsigned long long min;
4543 if (kstrtoull(buf, 10, &min))
4546 spin_lock(&mddev->lock);
4548 if (min > mddev->resync_max)
4552 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4555 /* Round down to multiple of 4K for safety */
4556 mddev->resync_min = round_down(min, 8);
4560 spin_unlock(&mddev->lock);
4564 static struct md_sysfs_entry md_min_sync =
4565 __ATTR(sync_min, S_IRUGO|S_IWUSR, min_sync_show, min_sync_store);
4568 max_sync_show(struct mddev *mddev, char *page)
4570 if (mddev->resync_max == MaxSector)
4571 return sprintf(page, "max\n");
4573 return sprintf(page, "%llu\n",
4574 (unsigned long long)mddev->resync_max);
4577 max_sync_store(struct mddev *mddev, const char *buf, size_t len)
4580 spin_lock(&mddev->lock);
4581 if (strncmp(buf, "max", 3) == 0)
4582 mddev->resync_max = MaxSector;
4584 unsigned long long max;
4588 if (kstrtoull(buf, 10, &max))
4590 if (max < mddev->resync_min)
4594 if (max < mddev->resync_max &&
4596 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4599 /* Must be a multiple of chunk_size */
4600 chunk = mddev->chunk_sectors;
4602 sector_t temp = max;
4605 if (sector_div(temp, chunk))
4608 mddev->resync_max = max;
4610 wake_up(&mddev->recovery_wait);
4613 spin_unlock(&mddev->lock);
4617 static struct md_sysfs_entry md_max_sync =
4618 __ATTR(sync_max, S_IRUGO|S_IWUSR, max_sync_show, max_sync_store);
4621 suspend_lo_show(struct mddev *mddev, char *page)
4623 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_lo);
4627 suspend_lo_store(struct mddev *mddev, const char *buf, size_t len)
4629 unsigned long long old, new;
4632 err = kstrtoull(buf, 10, &new);
4635 if (new != (sector_t)new)
4638 err = mddev_lock(mddev);
4642 if (mddev->pers == NULL ||
4643 mddev->pers->quiesce == NULL)
4645 old = mddev->suspend_lo;
4646 mddev->suspend_lo = new;
4648 /* Shrinking suspended region */
4649 mddev->pers->quiesce(mddev, 2);
4651 /* Expanding suspended region - need to wait */
4652 mddev->pers->quiesce(mddev, 1);
4653 mddev->pers->quiesce(mddev, 0);
4657 mddev_unlock(mddev);
4660 static struct md_sysfs_entry md_suspend_lo =
4661 __ATTR(suspend_lo, S_IRUGO|S_IWUSR, suspend_lo_show, suspend_lo_store);
4664 suspend_hi_show(struct mddev *mddev, char *page)
4666 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_hi);
4670 suspend_hi_store(struct mddev *mddev, const char *buf, size_t len)
4672 unsigned long long old, new;
4675 err = kstrtoull(buf, 10, &new);
4678 if (new != (sector_t)new)
4681 err = mddev_lock(mddev);
4685 if (mddev->pers == NULL ||
4686 mddev->pers->quiesce == NULL)
4688 old = mddev->suspend_hi;
4689 mddev->suspend_hi = new;
4691 /* Shrinking suspended region */
4692 mddev->pers->quiesce(mddev, 2);
4694 /* Expanding suspended region - need to wait */
4695 mddev->pers->quiesce(mddev, 1);
4696 mddev->pers->quiesce(mddev, 0);
4700 mddev_unlock(mddev);
4703 static struct md_sysfs_entry md_suspend_hi =
4704 __ATTR(suspend_hi, S_IRUGO|S_IWUSR, suspend_hi_show, suspend_hi_store);
4707 reshape_position_show(struct mddev *mddev, char *page)
4709 if (mddev->reshape_position != MaxSector)
4710 return sprintf(page, "%llu\n",
4711 (unsigned long long)mddev->reshape_position);
4712 strcpy(page, "none\n");
4717 reshape_position_store(struct mddev *mddev, const char *buf, size_t len)
4719 struct md_rdev *rdev;
4720 unsigned long long new;
4723 err = kstrtoull(buf, 10, &new);
4726 if (new != (sector_t)new)
4728 err = mddev_lock(mddev);
4734 mddev->reshape_position = new;
4735 mddev->delta_disks = 0;
4736 mddev->reshape_backwards = 0;
4737 mddev->new_level = mddev->level;
4738 mddev->new_layout = mddev->layout;
4739 mddev->new_chunk_sectors = mddev->chunk_sectors;
4740 rdev_for_each(rdev, mddev)
4741 rdev->new_data_offset = rdev->data_offset;
4744 mddev_unlock(mddev);
4748 static struct md_sysfs_entry md_reshape_position =
4749 __ATTR(reshape_position, S_IRUGO|S_IWUSR, reshape_position_show,
4750 reshape_position_store);
4753 reshape_direction_show(struct mddev *mddev, char *page)
4755 return sprintf(page, "%s\n",
4756 mddev->reshape_backwards ? "backwards" : "forwards");
4760 reshape_direction_store(struct mddev *mddev, const char *buf, size_t len)
4765 if (cmd_match(buf, "forwards"))
4767 else if (cmd_match(buf, "backwards"))
4771 if (mddev->reshape_backwards == backwards)
4774 err = mddev_lock(mddev);
4777 /* check if we are allowed to change */
4778 if (mddev->delta_disks)
4780 else if (mddev->persistent &&
4781 mddev->major_version == 0)
4784 mddev->reshape_backwards = backwards;
4785 mddev_unlock(mddev);
4789 static struct md_sysfs_entry md_reshape_direction =
4790 __ATTR(reshape_direction, S_IRUGO|S_IWUSR, reshape_direction_show,
4791 reshape_direction_store);
4794 array_size_show(struct mddev *mddev, char *page)
4796 if (mddev->external_size)
4797 return sprintf(page, "%llu\n",
4798 (unsigned long long)mddev->array_sectors/2);
4800 return sprintf(page, "default\n");
4804 array_size_store(struct mddev *mddev, const char *buf, size_t len)
4809 err = mddev_lock(mddev);
4813 if (strncmp(buf, "default", 7) == 0) {
4815 sectors = mddev->pers->size(mddev, 0, 0);
4817 sectors = mddev->array_sectors;
4819 mddev->external_size = 0;
4821 if (strict_blocks_to_sectors(buf, §ors) < 0)
4823 else if (mddev->pers && mddev->pers->size(mddev, 0, 0) < sectors)
4826 mddev->external_size = 1;
4830 mddev->array_sectors = sectors;
4832 set_capacity(mddev->gendisk, mddev->array_sectors);
4833 revalidate_disk(mddev->gendisk);
4836 mddev_unlock(mddev);
4840 static struct md_sysfs_entry md_array_size =
4841 __ATTR(array_size, S_IRUGO|S_IWUSR, array_size_show,
4844 static struct attribute *md_default_attrs[] = {
4847 &md_raid_disks.attr,
4848 &md_chunk_size.attr,
4850 &md_resync_start.attr,
4852 &md_new_device.attr,
4853 &md_safe_delay.attr,
4854 &md_array_state.attr,
4855 &md_reshape_position.attr,
4856 &md_reshape_direction.attr,
4857 &md_array_size.attr,
4858 &max_corr_read_errors.attr,
4862 static struct attribute *md_redundancy_attrs[] = {
4864 &md_last_scan_mode.attr,
4865 &md_mismatches.attr,
4868 &md_sync_speed.attr,
4869 &md_sync_force_parallel.attr,
4870 &md_sync_completed.attr,
4873 &md_suspend_lo.attr,
4874 &md_suspend_hi.attr,
4879 static struct attribute_group md_redundancy_group = {
4881 .attrs = md_redundancy_attrs,
4885 md_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
4887 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
4888 struct mddev *mddev = container_of(kobj, struct mddev, kobj);
4893 spin_lock(&all_mddevs_lock);
4894 if (list_empty(&mddev->all_mddevs)) {
4895 spin_unlock(&all_mddevs_lock);
4899 spin_unlock(&all_mddevs_lock);
4901 rv = entry->show(mddev, page);
4907 md_attr_store(struct kobject *kobj, struct attribute *attr,
4908 const char *page, size_t length)
4910 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
4911 struct mddev *mddev = container_of(kobj, struct mddev, kobj);
4916 if (!capable(CAP_SYS_ADMIN))
4918 spin_lock(&all_mddevs_lock);
4919 if (list_empty(&mddev->all_mddevs)) {
4920 spin_unlock(&all_mddevs_lock);
4924 spin_unlock(&all_mddevs_lock);
4925 rv = entry->store(mddev, page, length);
4930 static void md_free(struct kobject *ko)
4932 struct mddev *mddev = container_of(ko, struct mddev, kobj);
4934 if (mddev->sysfs_state)
4935 sysfs_put(mddev->sysfs_state);
4938 blk_cleanup_queue(mddev->queue);
4939 if (mddev->gendisk) {
4940 del_gendisk(mddev->gendisk);
4941 put_disk(mddev->gendisk);
4947 static const struct sysfs_ops md_sysfs_ops = {
4948 .show = md_attr_show,
4949 .store = md_attr_store,
4951 static struct kobj_type md_ktype = {
4953 .sysfs_ops = &md_sysfs_ops,
4954 .default_attrs = md_default_attrs,
4959 static void mddev_delayed_delete(struct work_struct *ws)
4961 struct mddev *mddev = container_of(ws, struct mddev, del_work);
4963 sysfs_remove_group(&mddev->kobj, &md_bitmap_group);
4964 kobject_del(&mddev->kobj);
4965 kobject_put(&mddev->kobj);
4968 static int md_alloc(dev_t dev, char *name)
4970 static DEFINE_MUTEX(disks_mutex);
4971 struct mddev *mddev = mddev_find(dev);
4972 struct gendisk *disk;
4981 partitioned = (MAJOR(mddev->unit) != MD_MAJOR);
4982 shift = partitioned ? MdpMinorShift : 0;
4983 unit = MINOR(mddev->unit) >> shift;
4985 /* wait for any previous instance of this device to be
4986 * completely removed (mddev_delayed_delete).
4988 flush_workqueue(md_misc_wq);
4990 mutex_lock(&disks_mutex);
4996 /* Need to ensure that 'name' is not a duplicate.
4998 struct mddev *mddev2;
4999 spin_lock(&all_mddevs_lock);
5001 list_for_each_entry(mddev2, &all_mddevs, all_mddevs)
5002 if (mddev2->gendisk &&
5003 strcmp(mddev2->gendisk->disk_name, name) == 0) {
5004 spin_unlock(&all_mddevs_lock);
5007 spin_unlock(&all_mddevs_lock);
5011 mddev->queue = blk_alloc_queue(GFP_KERNEL);
5014 mddev->queue->queuedata = mddev;
5016 blk_queue_make_request(mddev->queue, md_make_request);
5017 blk_set_stacking_limits(&mddev->queue->limits);
5019 disk = alloc_disk(1 << shift);
5021 blk_cleanup_queue(mddev->queue);
5022 mddev->queue = NULL;
5025 disk->major = MAJOR(mddev->unit);
5026 disk->first_minor = unit << shift;
5028 strcpy(disk->disk_name, name);
5029 else if (partitioned)
5030 sprintf(disk->disk_name, "md_d%d", unit);
5032 sprintf(disk->disk_name, "md%d", unit);
5033 disk->fops = &md_fops;
5034 disk->private_data = mddev;
5035 disk->queue = mddev->queue;
5036 blk_queue_flush(mddev->queue, REQ_FLUSH | REQ_FUA);
5037 /* Allow extended partitions. This makes the
5038 * 'mdp' device redundant, but we can't really
5041 disk->flags |= GENHD_FL_EXT_DEVT;
5042 mddev->gendisk = disk;
5043 /* As soon as we call add_disk(), another thread could get
5044 * through to md_open, so make sure it doesn't get too far
5046 mutex_lock(&mddev->open_mutex);
5049 error = kobject_init_and_add(&mddev->kobj, &md_ktype,
5050 &disk_to_dev(disk)->kobj, "%s", "md");
5052 /* This isn't possible, but as kobject_init_and_add is marked
5053 * __must_check, we must do something with the result
5055 printk(KERN_WARNING "md: cannot register %s/md - name in use\n",
5059 if (mddev->kobj.sd &&
5060 sysfs_create_group(&mddev->kobj, &md_bitmap_group))
5061 printk(KERN_DEBUG "pointless warning\n");
5062 mutex_unlock(&mddev->open_mutex);
5064 mutex_unlock(&disks_mutex);
5065 if (!error && mddev->kobj.sd) {
5066 kobject_uevent(&mddev->kobj, KOBJ_ADD);
5067 mddev->sysfs_state = sysfs_get_dirent_safe(mddev->kobj.sd, "array_state");
5073 static struct kobject *md_probe(dev_t dev, int *part, void *data)
5075 md_alloc(dev, NULL);
5079 static int add_named_array(const char *val, struct kernel_param *kp)
5081 /* val must be "md_*" where * is not all digits.
5082 * We allocate an array with a large free minor number, and
5083 * set the name to val. val must not already be an active name.
5085 int len = strlen(val);
5086 char buf[DISK_NAME_LEN];
5088 while (len && val[len-1] == '\n')
5090 if (len >= DISK_NAME_LEN)
5092 strlcpy(buf, val, len+1);
5093 if (strncmp(buf, "md_", 3) != 0)
5095 return md_alloc(0, buf);
5098 static void md_safemode_timeout(unsigned long data)
5100 struct mddev *mddev = (struct mddev *) data;
5102 if (!atomic_read(&mddev->writes_pending)) {
5103 mddev->safemode = 1;
5104 if (mddev->external)
5105 sysfs_notify_dirent_safe(mddev->sysfs_state);
5107 md_wakeup_thread(mddev->thread);
5110 static int start_dirty_degraded;
5112 int md_run(struct mddev *mddev)
5115 struct md_rdev *rdev;
5116 struct md_personality *pers;
5118 if (list_empty(&mddev->disks))
5119 /* cannot run an array with no devices.. */
5124 /* Cannot run until previous stop completes properly */
5125 if (mddev->sysfs_active)
5129 * Analyze all RAID superblock(s)
5131 if (!mddev->raid_disks) {
5132 if (!mddev->persistent)
5137 if (mddev->level != LEVEL_NONE)
5138 request_module("md-level-%d", mddev->level);
5139 else if (mddev->clevel[0])
5140 request_module("md-%s", mddev->clevel);
5143 * Drop all container device buffers, from now on
5144 * the only valid external interface is through the md
5147 rdev_for_each(rdev, mddev) {
5148 if (test_bit(Faulty, &rdev->flags))
5150 sync_blockdev(rdev->bdev);
5151 invalidate_bdev(rdev->bdev);
5153 /* perform some consistency tests on the device.
5154 * We don't want the data to overlap the metadata,
5155 * Internal Bitmap issues have been handled elsewhere.
5157 if (rdev->meta_bdev) {
5158 /* Nothing to check */;
5159 } else if (rdev->data_offset < rdev->sb_start) {
5160 if (mddev->dev_sectors &&
5161 rdev->data_offset + mddev->dev_sectors
5163 printk("md: %s: data overlaps metadata\n",
5168 if (rdev->sb_start + rdev->sb_size/512
5169 > rdev->data_offset) {
5170 printk("md: %s: metadata overlaps data\n",
5175 sysfs_notify_dirent_safe(rdev->sysfs_state);
5178 if (mddev->bio_set == NULL)
5179 mddev->bio_set = bioset_create(BIO_POOL_SIZE, 0);
5181 spin_lock(&pers_lock);
5182 pers = find_pers(mddev->level, mddev->clevel);
5183 if (!pers || !try_module_get(pers->owner)) {
5184 spin_unlock(&pers_lock);
5185 if (mddev->level != LEVEL_NONE)
5186 printk(KERN_WARNING "md: personality for level %d is not loaded!\n",
5189 printk(KERN_WARNING "md: personality for level %s is not loaded!\n",
5193 spin_unlock(&pers_lock);
5194 if (mddev->level != pers->level) {
5195 mddev->level = pers->level;
5196 mddev->new_level = pers->level;
5198 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
5200 if (mddev->reshape_position != MaxSector &&
5201 pers->start_reshape == NULL) {
5202 /* This personality cannot handle reshaping... */
5203 module_put(pers->owner);
5207 if (pers->sync_request) {
5208 /* Warn if this is a potentially silly
5211 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
5212 struct md_rdev *rdev2;
5215 rdev_for_each(rdev, mddev)
5216 rdev_for_each(rdev2, mddev) {
5218 rdev->bdev->bd_contains ==
5219 rdev2->bdev->bd_contains) {
5221 "%s: WARNING: %s appears to be"
5222 " on the same physical disk as"
5225 bdevname(rdev->bdev,b),
5226 bdevname(rdev2->bdev,b2));
5233 "True protection against single-disk"
5234 " failure might be compromised.\n");
5237 mddev->recovery = 0;
5238 /* may be over-ridden by personality */
5239 mddev->resync_max_sectors = mddev->dev_sectors;
5241 mddev->ok_start_degraded = start_dirty_degraded;
5243 if (start_readonly && mddev->ro == 0)
5244 mddev->ro = 2; /* read-only, but switch on first write */
5246 err = pers->run(mddev);
5248 printk(KERN_ERR "md: pers->run() failed ...\n");
5249 else if (pers->size(mddev, 0, 0) < mddev->array_sectors) {
5250 WARN_ONCE(!mddev->external_size, "%s: default size too small,"
5251 " but 'external_size' not in effect?\n", __func__);
5253 "md: invalid array_size %llu > default size %llu\n",
5254 (unsigned long long)mddev->array_sectors / 2,
5255 (unsigned long long)pers->size(mddev, 0, 0) / 2);
5258 if (err == 0 && pers->sync_request &&
5259 (mddev->bitmap_info.file || mddev->bitmap_info.offset)) {
5260 struct bitmap *bitmap;
5262 bitmap = bitmap_create(mddev, -1);
5263 if (IS_ERR(bitmap)) {
5264 err = PTR_ERR(bitmap);
5265 printk(KERN_ERR "%s: failed to create bitmap (%d)\n",
5266 mdname(mddev), err);
5268 mddev->bitmap = bitmap;
5272 mddev_detach(mddev);
5274 pers->free(mddev, mddev->private);
5275 mddev->private = NULL;
5276 module_put(pers->owner);
5277 bitmap_destroy(mddev);
5281 mddev->queue->backing_dev_info.congested_data = mddev;
5282 mddev->queue->backing_dev_info.congested_fn = md_congested;
5284 if (pers->sync_request) {
5285 if (mddev->kobj.sd &&
5286 sysfs_create_group(&mddev->kobj, &md_redundancy_group))
5288 "md: cannot register extra attributes for %s\n",
5290 mddev->sysfs_action = sysfs_get_dirent_safe(mddev->kobj.sd, "sync_action");
5291 } else if (mddev->ro == 2) /* auto-readonly not meaningful */
5294 atomic_set(&mddev->writes_pending,0);
5295 atomic_set(&mddev->max_corr_read_errors,
5296 MD_DEFAULT_MAX_CORRECTED_READ_ERRORS);
5297 mddev->safemode = 0;
5298 if (mddev_is_clustered(mddev))
5299 mddev->safemode_delay = 0;
5301 mddev->safemode_delay = (200 * HZ)/1000 +1; /* 200 msec delay */
5304 spin_lock(&mddev->lock);
5306 spin_unlock(&mddev->lock);
5307 rdev_for_each(rdev, mddev)
5308 if (rdev->raid_disk >= 0)
5309 if (sysfs_link_rdev(mddev, rdev))
5310 /* failure here is OK */;
5312 if (mddev->degraded && !mddev->ro)
5313 /* This ensures that recovering status is reported immediately
5314 * via sysfs - until a lack of spares is confirmed.
5316 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
5317 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5319 if (mddev->flags & MD_UPDATE_SB_FLAGS)
5320 md_update_sb(mddev, 0);
5322 md_new_event(mddev);
5323 sysfs_notify_dirent_safe(mddev->sysfs_state);
5324 sysfs_notify_dirent_safe(mddev->sysfs_action);
5325 sysfs_notify(&mddev->kobj, NULL, "degraded");
5328 EXPORT_SYMBOL_GPL(md_run);
5330 static int do_md_run(struct mddev *mddev)
5334 err = md_run(mddev);
5337 err = bitmap_load(mddev);
5339 bitmap_destroy(mddev);
5343 if (mddev_is_clustered(mddev))
5344 md_allow_write(mddev);
5346 md_wakeup_thread(mddev->thread);
5347 md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
5349 set_capacity(mddev->gendisk, mddev->array_sectors);
5350 revalidate_disk(mddev->gendisk);
5352 kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
5357 static int restart_array(struct mddev *mddev)
5359 struct gendisk *disk = mddev->gendisk;
5361 /* Complain if it has no devices */
5362 if (list_empty(&mddev->disks))
5368 if (test_bit(MD_HAS_JOURNAL, &mddev->flags)) {
5369 struct md_rdev *rdev;
5370 bool has_journal = false;
5373 rdev_for_each_rcu(rdev, mddev) {
5374 if (test_bit(Journal, &rdev->flags) &&
5375 !test_bit(Faulty, &rdev->flags)) {
5382 /* Don't restart rw with journal missing/faulty */
5387 mddev->safemode = 0;
5389 set_disk_ro(disk, 0);
5390 printk(KERN_INFO "md: %s switched to read-write mode.\n",
5392 /* Kick recovery or resync if necessary */
5393 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5394 md_wakeup_thread(mddev->thread);
5395 md_wakeup_thread(mddev->sync_thread);
5396 sysfs_notify_dirent_safe(mddev->sysfs_state);
5400 static void md_clean(struct mddev *mddev)
5402 mddev->array_sectors = 0;
5403 mddev->external_size = 0;
5404 mddev->dev_sectors = 0;
5405 mddev->raid_disks = 0;
5406 mddev->recovery_cp = 0;
5407 mddev->resync_min = 0;
5408 mddev->resync_max = MaxSector;
5409 mddev->reshape_position = MaxSector;
5410 mddev->external = 0;
5411 mddev->persistent = 0;
5412 mddev->level = LEVEL_NONE;
5413 mddev->clevel[0] = 0;
5416 mddev->metadata_type[0] = 0;
5417 mddev->chunk_sectors = 0;
5418 mddev->ctime = mddev->utime = 0;
5420 mddev->max_disks = 0;
5422 mddev->can_decrease_events = 0;
5423 mddev->delta_disks = 0;
5424 mddev->reshape_backwards = 0;
5425 mddev->new_level = LEVEL_NONE;
5426 mddev->new_layout = 0;
5427 mddev->new_chunk_sectors = 0;
5428 mddev->curr_resync = 0;
5429 atomic64_set(&mddev->resync_mismatches, 0);
5430 mddev->suspend_lo = mddev->suspend_hi = 0;
5431 mddev->sync_speed_min = mddev->sync_speed_max = 0;
5432 mddev->recovery = 0;
5435 mddev->degraded = 0;
5436 mddev->safemode = 0;
5437 mddev->private = NULL;
5438 mddev->bitmap_info.offset = 0;
5439 mddev->bitmap_info.default_offset = 0;
5440 mddev->bitmap_info.default_space = 0;
5441 mddev->bitmap_info.chunksize = 0;
5442 mddev->bitmap_info.daemon_sleep = 0;
5443 mddev->bitmap_info.max_write_behind = 0;
5446 static void __md_stop_writes(struct mddev *mddev)
5448 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5449 flush_workqueue(md_misc_wq);
5450 if (mddev->sync_thread) {
5451 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5452 md_reap_sync_thread(mddev);
5455 del_timer_sync(&mddev->safemode_timer);
5457 bitmap_flush(mddev);
5458 md_super_wait(mddev);
5460 if (mddev->ro == 0 &&
5461 ((!mddev->in_sync && !mddev_is_clustered(mddev)) ||
5462 (mddev->flags & MD_UPDATE_SB_FLAGS))) {
5463 /* mark array as shutdown cleanly */
5464 if (!mddev_is_clustered(mddev))
5466 md_update_sb(mddev, 1);
5470 void md_stop_writes(struct mddev *mddev)
5472 mddev_lock_nointr(mddev);
5473 __md_stop_writes(mddev);
5474 mddev_unlock(mddev);
5476 EXPORT_SYMBOL_GPL(md_stop_writes);
5478 static void mddev_detach(struct mddev *mddev)
5480 struct bitmap *bitmap = mddev->bitmap;
5481 /* wait for behind writes to complete */
5482 if (bitmap && atomic_read(&bitmap->behind_writes) > 0) {
5483 printk(KERN_INFO "md:%s: behind writes in progress - waiting to stop.\n",
5485 /* need to kick something here to make sure I/O goes? */
5486 wait_event(bitmap->behind_wait,
5487 atomic_read(&bitmap->behind_writes) == 0);
5489 if (mddev->pers && mddev->pers->quiesce) {
5490 mddev->pers->quiesce(mddev, 1);
5491 mddev->pers->quiesce(mddev, 0);
5493 md_unregister_thread(&mddev->thread);
5495 blk_sync_queue(mddev->queue); /* the unplug fn references 'conf'*/
5498 static void __md_stop(struct mddev *mddev)
5500 struct md_personality *pers = mddev->pers;
5501 mddev_detach(mddev);
5502 /* Ensure ->event_work is done */
5503 flush_workqueue(md_misc_wq);
5504 spin_lock(&mddev->lock);
5506 spin_unlock(&mddev->lock);
5507 pers->free(mddev, mddev->private);
5508 mddev->private = NULL;
5509 if (pers->sync_request && mddev->to_remove == NULL)
5510 mddev->to_remove = &md_redundancy_group;
5511 module_put(pers->owner);
5512 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5515 void md_stop(struct mddev *mddev)
5517 /* stop the array and free an attached data structures.
5518 * This is called from dm-raid
5521 bitmap_destroy(mddev);
5523 bioset_free(mddev->bio_set);
5526 EXPORT_SYMBOL_GPL(md_stop);
5528 static int md_set_readonly(struct mddev *mddev, struct block_device *bdev)
5533 if (!test_bit(MD_RECOVERY_FROZEN, &mddev->recovery)) {
5535 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5536 md_wakeup_thread(mddev->thread);
5538 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
5539 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5540 if (mddev->sync_thread)
5541 /* Thread might be blocked waiting for metadata update
5542 * which will now never happen */
5543 wake_up_process(mddev->sync_thread->tsk);
5545 if (mddev->external && test_bit(MD_CHANGE_PENDING, &mddev->flags))
5547 mddev_unlock(mddev);
5548 wait_event(resync_wait, !test_bit(MD_RECOVERY_RUNNING,
5550 wait_event(mddev->sb_wait,
5551 !test_bit(MD_CHANGE_PENDING, &mddev->flags));
5552 mddev_lock_nointr(mddev);
5554 mutex_lock(&mddev->open_mutex);
5555 if ((mddev->pers && atomic_read(&mddev->openers) > !!bdev) ||
5556 mddev->sync_thread ||
5557 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
5558 (bdev && !test_bit(MD_STILL_CLOSED, &mddev->flags))) {
5559 printk("md: %s still in use.\n",mdname(mddev));
5561 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5562 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5563 md_wakeup_thread(mddev->thread);
5569 __md_stop_writes(mddev);
5575 set_disk_ro(mddev->gendisk, 1);
5576 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5577 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5578 md_wakeup_thread(mddev->thread);
5579 sysfs_notify_dirent_safe(mddev->sysfs_state);
5583 mutex_unlock(&mddev->open_mutex);
5588 * 0 - completely stop and dis-assemble array
5589 * 2 - stop but do not disassemble array
5591 static int do_md_stop(struct mddev *mddev, int mode,
5592 struct block_device *bdev)
5594 struct gendisk *disk = mddev->gendisk;
5595 struct md_rdev *rdev;
5598 if (!test_bit(MD_RECOVERY_FROZEN, &mddev->recovery)) {
5600 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5601 md_wakeup_thread(mddev->thread);
5603 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
5604 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5605 if (mddev->sync_thread)
5606 /* Thread might be blocked waiting for metadata update
5607 * which will now never happen */
5608 wake_up_process(mddev->sync_thread->tsk);
5610 mddev_unlock(mddev);
5611 wait_event(resync_wait, (mddev->sync_thread == NULL &&
5612 !test_bit(MD_RECOVERY_RUNNING,
5613 &mddev->recovery)));
5614 mddev_lock_nointr(mddev);
5616 mutex_lock(&mddev->open_mutex);
5617 if ((mddev->pers && atomic_read(&mddev->openers) > !!bdev) ||
5618 mddev->sysfs_active ||
5619 mddev->sync_thread ||
5620 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
5621 (bdev && !test_bit(MD_STILL_CLOSED, &mddev->flags))) {
5622 printk("md: %s still in use.\n",mdname(mddev));
5623 mutex_unlock(&mddev->open_mutex);
5625 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5626 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5627 md_wakeup_thread(mddev->thread);
5633 set_disk_ro(disk, 0);
5635 __md_stop_writes(mddev);
5637 mddev->queue->backing_dev_info.congested_fn = NULL;
5639 /* tell userspace to handle 'inactive' */
5640 sysfs_notify_dirent_safe(mddev->sysfs_state);
5642 rdev_for_each(rdev, mddev)
5643 if (rdev->raid_disk >= 0)
5644 sysfs_unlink_rdev(mddev, rdev);
5646 set_capacity(disk, 0);
5647 mutex_unlock(&mddev->open_mutex);
5649 revalidate_disk(disk);
5654 mutex_unlock(&mddev->open_mutex);
5656 * Free resources if final stop
5659 printk(KERN_INFO "md: %s stopped.\n", mdname(mddev));
5661 bitmap_destroy(mddev);
5662 if (mddev->bitmap_info.file) {
5663 struct file *f = mddev->bitmap_info.file;
5664 spin_lock(&mddev->lock);
5665 mddev->bitmap_info.file = NULL;
5666 spin_unlock(&mddev->lock);
5669 mddev->bitmap_info.offset = 0;
5671 export_array(mddev);
5674 kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
5675 if (mddev->hold_active == UNTIL_STOP)
5676 mddev->hold_active = 0;
5678 md_new_event(mddev);
5679 sysfs_notify_dirent_safe(mddev->sysfs_state);
5684 static void autorun_array(struct mddev *mddev)
5686 struct md_rdev *rdev;
5689 if (list_empty(&mddev->disks))
5692 printk(KERN_INFO "md: running: ");
5694 rdev_for_each(rdev, mddev) {
5695 char b[BDEVNAME_SIZE];
5696 printk("<%s>", bdevname(rdev->bdev,b));
5700 err = do_md_run(mddev);
5702 printk(KERN_WARNING "md: do_md_run() returned %d\n", err);
5703 do_md_stop(mddev, 0, NULL);
5708 * lets try to run arrays based on all disks that have arrived
5709 * until now. (those are in pending_raid_disks)
5711 * the method: pick the first pending disk, collect all disks with
5712 * the same UUID, remove all from the pending list and put them into
5713 * the 'same_array' list. Then order this list based on superblock
5714 * update time (freshest comes first), kick out 'old' disks and
5715 * compare superblocks. If everything's fine then run it.
5717 * If "unit" is allocated, then bump its reference count
5719 static void autorun_devices(int part)
5721 struct md_rdev *rdev0, *rdev, *tmp;
5722 struct mddev *mddev;
5723 char b[BDEVNAME_SIZE];
5725 printk(KERN_INFO "md: autorun ...\n");
5726 while (!list_empty(&pending_raid_disks)) {
5729 LIST_HEAD(candidates);
5730 rdev0 = list_entry(pending_raid_disks.next,
5731 struct md_rdev, same_set);
5733 printk(KERN_INFO "md: considering %s ...\n",
5734 bdevname(rdev0->bdev,b));
5735 INIT_LIST_HEAD(&candidates);
5736 rdev_for_each_list(rdev, tmp, &pending_raid_disks)
5737 if (super_90_load(rdev, rdev0, 0) >= 0) {
5738 printk(KERN_INFO "md: adding %s ...\n",
5739 bdevname(rdev->bdev,b));
5740 list_move(&rdev->same_set, &candidates);
5743 * now we have a set of devices, with all of them having
5744 * mostly sane superblocks. It's time to allocate the
5748 dev = MKDEV(mdp_major,
5749 rdev0->preferred_minor << MdpMinorShift);
5750 unit = MINOR(dev) >> MdpMinorShift;
5752 dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
5755 if (rdev0->preferred_minor != unit) {
5756 printk(KERN_INFO "md: unit number in %s is bad: %d\n",
5757 bdevname(rdev0->bdev, b), rdev0->preferred_minor);
5761 md_probe(dev, NULL, NULL);
5762 mddev = mddev_find(dev);
5763 if (!mddev || !mddev->gendisk) {
5767 "md: cannot allocate memory for md drive.\n");
5770 if (mddev_lock(mddev))
5771 printk(KERN_WARNING "md: %s locked, cannot run\n",
5773 else if (mddev->raid_disks || mddev->major_version
5774 || !list_empty(&mddev->disks)) {
5776 "md: %s already running, cannot run %s\n",
5777 mdname(mddev), bdevname(rdev0->bdev,b));
5778 mddev_unlock(mddev);
5780 printk(KERN_INFO "md: created %s\n", mdname(mddev));
5781 mddev->persistent = 1;
5782 rdev_for_each_list(rdev, tmp, &candidates) {
5783 list_del_init(&rdev->same_set);
5784 if (bind_rdev_to_array(rdev, mddev))
5787 autorun_array(mddev);
5788 mddev_unlock(mddev);
5790 /* on success, candidates will be empty, on error
5793 rdev_for_each_list(rdev, tmp, &candidates) {
5794 list_del_init(&rdev->same_set);
5799 printk(KERN_INFO "md: ... autorun DONE.\n");
5801 #endif /* !MODULE */
5803 static int get_version(void __user *arg)
5807 ver.major = MD_MAJOR_VERSION;
5808 ver.minor = MD_MINOR_VERSION;
5809 ver.patchlevel = MD_PATCHLEVEL_VERSION;
5811 if (copy_to_user(arg, &ver, sizeof(ver)))
5817 static int get_array_info(struct mddev *mddev, void __user *arg)
5819 mdu_array_info_t info;
5820 int nr,working,insync,failed,spare;
5821 struct md_rdev *rdev;
5823 nr = working = insync = failed = spare = 0;
5825 rdev_for_each_rcu(rdev, mddev) {
5827 if (test_bit(Faulty, &rdev->flags))
5831 if (test_bit(In_sync, &rdev->flags))
5839 info.major_version = mddev->major_version;
5840 info.minor_version = mddev->minor_version;
5841 info.patch_version = MD_PATCHLEVEL_VERSION;
5842 info.ctime = clamp_t(time64_t, mddev->ctime, 0, U32_MAX);
5843 info.level = mddev->level;
5844 info.size = mddev->dev_sectors / 2;
5845 if (info.size != mddev->dev_sectors / 2) /* overflow */
5848 info.raid_disks = mddev->raid_disks;
5849 info.md_minor = mddev->md_minor;
5850 info.not_persistent= !mddev->persistent;
5852 info.utime = clamp_t(time64_t, mddev->utime, 0, U32_MAX);
5855 info.state = (1<<MD_SB_CLEAN);
5856 if (mddev->bitmap && mddev->bitmap_info.offset)
5857 info.state |= (1<<MD_SB_BITMAP_PRESENT);
5858 if (mddev_is_clustered(mddev))
5859 info.state |= (1<<MD_SB_CLUSTERED);
5860 info.active_disks = insync;
5861 info.working_disks = working;
5862 info.failed_disks = failed;
5863 info.spare_disks = spare;
5865 info.layout = mddev->layout;
5866 info.chunk_size = mddev->chunk_sectors << 9;
5868 if (copy_to_user(arg, &info, sizeof(info)))
5874 static int get_bitmap_file(struct mddev *mddev, void __user * arg)
5876 mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */
5880 file = kzalloc(sizeof(*file), GFP_NOIO);
5885 spin_lock(&mddev->lock);
5886 /* bitmap enabled */
5887 if (mddev->bitmap_info.file) {
5888 ptr = file_path(mddev->bitmap_info.file, file->pathname,
5889 sizeof(file->pathname));
5893 memmove(file->pathname, ptr,
5894 sizeof(file->pathname)-(ptr-file->pathname));
5896 spin_unlock(&mddev->lock);
5899 copy_to_user(arg, file, sizeof(*file)))
5906 static int get_disk_info(struct mddev *mddev, void __user * arg)
5908 mdu_disk_info_t info;
5909 struct md_rdev *rdev;
5911 if (copy_from_user(&info, arg, sizeof(info)))
5915 rdev = md_find_rdev_nr_rcu(mddev, info.number);
5917 info.major = MAJOR(rdev->bdev->bd_dev);
5918 info.minor = MINOR(rdev->bdev->bd_dev);
5919 info.raid_disk = rdev->raid_disk;
5921 if (test_bit(Faulty, &rdev->flags))
5922 info.state |= (1<<MD_DISK_FAULTY);
5923 else if (test_bit(In_sync, &rdev->flags)) {
5924 info.state |= (1<<MD_DISK_ACTIVE);
5925 info.state |= (1<<MD_DISK_SYNC);
5927 if (test_bit(Journal, &rdev->flags))
5928 info.state |= (1<<MD_DISK_JOURNAL);
5929 if (test_bit(WriteMostly, &rdev->flags))
5930 info.state |= (1<<MD_DISK_WRITEMOSTLY);
5932 info.major = info.minor = 0;
5933 info.raid_disk = -1;
5934 info.state = (1<<MD_DISK_REMOVED);
5938 if (copy_to_user(arg, &info, sizeof(info)))
5944 static int add_new_disk(struct mddev *mddev, mdu_disk_info_t *info)
5946 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
5947 struct md_rdev *rdev;
5948 dev_t dev = MKDEV(info->major,info->minor);
5950 if (mddev_is_clustered(mddev) &&
5951 !(info->state & ((1 << MD_DISK_CLUSTER_ADD) | (1 << MD_DISK_CANDIDATE)))) {
5952 pr_err("%s: Cannot add to clustered mddev.\n",
5957 if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
5960 if (!mddev->raid_disks) {
5962 /* expecting a device which has a superblock */
5963 rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
5966 "md: md_import_device returned %ld\n",
5968 return PTR_ERR(rdev);
5970 if (!list_empty(&mddev->disks)) {
5971 struct md_rdev *rdev0
5972 = list_entry(mddev->disks.next,
5973 struct md_rdev, same_set);
5974 err = super_types[mddev->major_version]
5975 .load_super(rdev, rdev0, mddev->minor_version);
5978 "md: %s has different UUID to %s\n",
5979 bdevname(rdev->bdev,b),
5980 bdevname(rdev0->bdev,b2));
5985 err = bind_rdev_to_array(rdev, mddev);
5992 * add_new_disk can be used once the array is assembled
5993 * to add "hot spares". They must already have a superblock
5998 if (!mddev->pers->hot_add_disk) {
6000 "%s: personality does not support diskops!\n",
6004 if (mddev->persistent)
6005 rdev = md_import_device(dev, mddev->major_version,
6006 mddev->minor_version);
6008 rdev = md_import_device(dev, -1, -1);
6011 "md: md_import_device returned %ld\n",
6013 return PTR_ERR(rdev);
6015 /* set saved_raid_disk if appropriate */
6016 if (!mddev->persistent) {
6017 if (info->state & (1<<MD_DISK_SYNC) &&
6018 info->raid_disk < mddev->raid_disks) {
6019 rdev->raid_disk = info->raid_disk;
6020 set_bit(In_sync, &rdev->flags);
6021 clear_bit(Bitmap_sync, &rdev->flags);
6023 rdev->raid_disk = -1;
6024 rdev->saved_raid_disk = rdev->raid_disk;
6026 super_types[mddev->major_version].
6027 validate_super(mddev, rdev);
6028 if ((info->state & (1<<MD_DISK_SYNC)) &&
6029 rdev->raid_disk != info->raid_disk) {
6030 /* This was a hot-add request, but events doesn't
6031 * match, so reject it.
6037 clear_bit(In_sync, &rdev->flags); /* just to be sure */
6038 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
6039 set_bit(WriteMostly, &rdev->flags);
6041 clear_bit(WriteMostly, &rdev->flags);
6043 if (info->state & (1<<MD_DISK_JOURNAL)) {
6044 struct md_rdev *rdev2;
6045 bool has_journal = false;
6047 /* make sure no existing journal disk */
6048 rdev_for_each(rdev2, mddev) {
6049 if (test_bit(Journal, &rdev2->flags)) {
6058 set_bit(Journal, &rdev->flags);
6061 * check whether the device shows up in other nodes
6063 if (mddev_is_clustered(mddev)) {
6064 if (info->state & (1 << MD_DISK_CANDIDATE))
6065 set_bit(Candidate, &rdev->flags);
6066 else if (info->state & (1 << MD_DISK_CLUSTER_ADD)) {
6067 /* --add initiated by this node */
6068 err = md_cluster_ops->add_new_disk(mddev, rdev);
6076 rdev->raid_disk = -1;
6077 err = bind_rdev_to_array(rdev, mddev);
6082 if (mddev_is_clustered(mddev)) {
6083 if (info->state & (1 << MD_DISK_CANDIDATE))
6084 md_cluster_ops->new_disk_ack(mddev, (err == 0));
6087 md_cluster_ops->add_new_disk_cancel(mddev);
6089 err = add_bound_rdev(rdev);
6093 err = add_bound_rdev(rdev);
6098 /* otherwise, add_new_disk is only allowed
6099 * for major_version==0 superblocks
6101 if (mddev->major_version != 0) {
6102 printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n",
6107 if (!(info->state & (1<<MD_DISK_FAULTY))) {
6109 rdev = md_import_device(dev, -1, 0);
6112 "md: error, md_import_device() returned %ld\n",
6114 return PTR_ERR(rdev);
6116 rdev->desc_nr = info->number;
6117 if (info->raid_disk < mddev->raid_disks)
6118 rdev->raid_disk = info->raid_disk;
6120 rdev->raid_disk = -1;
6122 if (rdev->raid_disk < mddev->raid_disks)
6123 if (info->state & (1<<MD_DISK_SYNC))
6124 set_bit(In_sync, &rdev->flags);
6126 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
6127 set_bit(WriteMostly, &rdev->flags);
6129 if (!mddev->persistent) {
6130 printk(KERN_INFO "md: nonpersistent superblock ...\n");
6131 rdev->sb_start = i_size_read(rdev->bdev->bd_inode) / 512;
6133 rdev->sb_start = calc_dev_sboffset(rdev);
6134 rdev->sectors = rdev->sb_start;
6136 err = bind_rdev_to_array(rdev, mddev);
6146 static int hot_remove_disk(struct mddev *mddev, dev_t dev)
6148 char b[BDEVNAME_SIZE];
6149 struct md_rdev *rdev;
6151 rdev = find_rdev(mddev, dev);
6155 if (rdev->raid_disk < 0)
6158 clear_bit(Blocked, &rdev->flags);
6159 remove_and_add_spares(mddev, rdev);
6161 if (rdev->raid_disk >= 0)
6165 if (mddev_is_clustered(mddev))
6166 md_cluster_ops->remove_disk(mddev, rdev);
6168 md_kick_rdev_from_array(rdev);
6169 md_update_sb(mddev, 1);
6170 md_new_event(mddev);
6174 printk(KERN_WARNING "md: cannot remove active disk %s from %s ...\n",
6175 bdevname(rdev->bdev,b), mdname(mddev));
6179 static int hot_add_disk(struct mddev *mddev, dev_t dev)
6181 char b[BDEVNAME_SIZE];
6183 struct md_rdev *rdev;
6188 if (mddev->major_version != 0) {
6189 printk(KERN_WARNING "%s: HOT_ADD may only be used with"
6190 " version-0 superblocks.\n",
6194 if (!mddev->pers->hot_add_disk) {
6196 "%s: personality does not support diskops!\n",
6201 rdev = md_import_device(dev, -1, 0);
6204 "md: error, md_import_device() returned %ld\n",
6209 if (mddev->persistent)
6210 rdev->sb_start = calc_dev_sboffset(rdev);
6212 rdev->sb_start = i_size_read(rdev->bdev->bd_inode) / 512;
6214 rdev->sectors = rdev->sb_start;
6216 if (test_bit(Faulty, &rdev->flags)) {
6218 "md: can not hot-add faulty %s disk to %s!\n",
6219 bdevname(rdev->bdev,b), mdname(mddev));
6224 clear_bit(In_sync, &rdev->flags);
6226 rdev->saved_raid_disk = -1;
6227 err = bind_rdev_to_array(rdev, mddev);
6232 * The rest should better be atomic, we can have disk failures
6233 * noticed in interrupt contexts ...
6236 rdev->raid_disk = -1;
6238 md_update_sb(mddev, 1);
6240 * Kick recovery, maybe this spare has to be added to the
6241 * array immediately.
6243 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6244 md_wakeup_thread(mddev->thread);
6245 md_new_event(mddev);
6253 static int set_bitmap_file(struct mddev *mddev, int fd)
6258 if (!mddev->pers->quiesce || !mddev->thread)
6260 if (mddev->recovery || mddev->sync_thread)
6262 /* we should be able to change the bitmap.. */
6266 struct inode *inode;
6269 if (mddev->bitmap || mddev->bitmap_info.file)
6270 return -EEXIST; /* cannot add when bitmap is present */
6274 printk(KERN_ERR "%s: error: failed to get bitmap file\n",
6279 inode = f->f_mapping->host;
6280 if (!S_ISREG(inode->i_mode)) {
6281 printk(KERN_ERR "%s: error: bitmap file must be a regular file\n",
6284 } else if (!(f->f_mode & FMODE_WRITE)) {
6285 printk(KERN_ERR "%s: error: bitmap file must open for write\n",
6288 } else if (atomic_read(&inode->i_writecount) != 1) {
6289 printk(KERN_ERR "%s: error: bitmap file is already in use\n",
6297 mddev->bitmap_info.file = f;
6298 mddev->bitmap_info.offset = 0; /* file overrides offset */
6299 } else if (mddev->bitmap == NULL)
6300 return -ENOENT; /* cannot remove what isn't there */
6303 mddev->pers->quiesce(mddev, 1);
6305 struct bitmap *bitmap;
6307 bitmap = bitmap_create(mddev, -1);
6308 if (!IS_ERR(bitmap)) {
6309 mddev->bitmap = bitmap;
6310 err = bitmap_load(mddev);
6312 err = PTR_ERR(bitmap);
6314 if (fd < 0 || err) {
6315 bitmap_destroy(mddev);
6316 fd = -1; /* make sure to put the file */
6318 mddev->pers->quiesce(mddev, 0);
6321 struct file *f = mddev->bitmap_info.file;
6323 spin_lock(&mddev->lock);
6324 mddev->bitmap_info.file = NULL;
6325 spin_unlock(&mddev->lock);
6334 * set_array_info is used two different ways
6335 * The original usage is when creating a new array.
6336 * In this usage, raid_disks is > 0 and it together with
6337 * level, size, not_persistent,layout,chunksize determine the
6338 * shape of the array.
6339 * This will always create an array with a type-0.90.0 superblock.
6340 * The newer usage is when assembling an array.
6341 * In this case raid_disks will be 0, and the major_version field is
6342 * use to determine which style super-blocks are to be found on the devices.
6343 * The minor and patch _version numbers are also kept incase the
6344 * super_block handler wishes to interpret them.
6346 static int set_array_info(struct mddev *mddev, mdu_array_info_t *info)
6349 if (info->raid_disks == 0) {
6350 /* just setting version number for superblock loading */
6351 if (info->major_version < 0 ||
6352 info->major_version >= ARRAY_SIZE(super_types) ||
6353 super_types[info->major_version].name == NULL) {
6354 /* maybe try to auto-load a module? */
6356 "md: superblock version %d not known\n",
6357 info->major_version);
6360 mddev->major_version = info->major_version;
6361 mddev->minor_version = info->minor_version;
6362 mddev->patch_version = info->patch_version;
6363 mddev->persistent = !info->not_persistent;
6364 /* ensure mddev_put doesn't delete this now that there
6365 * is some minimal configuration.
6367 mddev->ctime = ktime_get_real_seconds();
6370 mddev->major_version = MD_MAJOR_VERSION;
6371 mddev->minor_version = MD_MINOR_VERSION;
6372 mddev->patch_version = MD_PATCHLEVEL_VERSION;
6373 mddev->ctime = ktime_get_real_seconds();
6375 mddev->level = info->level;
6376 mddev->clevel[0] = 0;
6377 mddev->dev_sectors = 2 * (sector_t)info->size;
6378 mddev->raid_disks = info->raid_disks;
6379 /* don't set md_minor, it is determined by which /dev/md* was
6382 if (info->state & (1<<MD_SB_CLEAN))
6383 mddev->recovery_cp = MaxSector;
6385 mddev->recovery_cp = 0;
6386 mddev->persistent = ! info->not_persistent;
6387 mddev->external = 0;
6389 mddev->layout = info->layout;
6390 mddev->chunk_sectors = info->chunk_size >> 9;
6392 mddev->max_disks = MD_SB_DISKS;
6394 if (mddev->persistent)
6396 set_bit(MD_CHANGE_DEVS, &mddev->flags);
6398 mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9;
6399 mddev->bitmap_info.default_space = 64*2 - (MD_SB_BYTES >> 9);
6400 mddev->bitmap_info.offset = 0;
6402 mddev->reshape_position = MaxSector;
6405 * Generate a 128 bit UUID
6407 get_random_bytes(mddev->uuid, 16);
6409 mddev->new_level = mddev->level;
6410 mddev->new_chunk_sectors = mddev->chunk_sectors;
6411 mddev->new_layout = mddev->layout;
6412 mddev->delta_disks = 0;
6413 mddev->reshape_backwards = 0;
6418 void md_set_array_sectors(struct mddev *mddev, sector_t array_sectors)
6420 WARN(!mddev_is_locked(mddev), "%s: unlocked mddev!\n", __func__);
6422 if (mddev->external_size)
6425 mddev->array_sectors = array_sectors;
6427 EXPORT_SYMBOL(md_set_array_sectors);
6429 static int update_size(struct mddev *mddev, sector_t num_sectors)
6431 struct md_rdev *rdev;
6433 int fit = (num_sectors == 0);
6435 if (mddev->pers->resize == NULL)
6437 /* The "num_sectors" is the number of sectors of each device that
6438 * is used. This can only make sense for arrays with redundancy.
6439 * linear and raid0 always use whatever space is available. We can only
6440 * consider changing this number if no resync or reconstruction is
6441 * happening, and if the new size is acceptable. It must fit before the
6442 * sb_start or, if that is <data_offset, it must fit before the size
6443 * of each device. If num_sectors is zero, we find the largest size
6446 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
6452 rdev_for_each(rdev, mddev) {
6453 sector_t avail = rdev->sectors;
6455 if (fit && (num_sectors == 0 || num_sectors > avail))
6456 num_sectors = avail;
6457 if (avail < num_sectors)
6460 rv = mddev->pers->resize(mddev, num_sectors);
6462 revalidate_disk(mddev->gendisk);
6466 static int update_raid_disks(struct mddev *mddev, int raid_disks)
6469 struct md_rdev *rdev;
6470 /* change the number of raid disks */
6471 if (mddev->pers->check_reshape == NULL)
6475 if (raid_disks <= 0 ||
6476 (mddev->max_disks && raid_disks >= mddev->max_disks))
6478 if (mddev->sync_thread ||
6479 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
6480 mddev->reshape_position != MaxSector)
6483 rdev_for_each(rdev, mddev) {
6484 if (mddev->raid_disks < raid_disks &&
6485 rdev->data_offset < rdev->new_data_offset)
6487 if (mddev->raid_disks > raid_disks &&
6488 rdev->data_offset > rdev->new_data_offset)
6492 mddev->delta_disks = raid_disks - mddev->raid_disks;
6493 if (mddev->delta_disks < 0)
6494 mddev->reshape_backwards = 1;
6495 else if (mddev->delta_disks > 0)
6496 mddev->reshape_backwards = 0;
6498 rv = mddev->pers->check_reshape(mddev);
6500 mddev->delta_disks = 0;
6501 mddev->reshape_backwards = 0;
6507 * update_array_info is used to change the configuration of an
6509 * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
6510 * fields in the info are checked against the array.
6511 * Any differences that cannot be handled will cause an error.
6512 * Normally, only one change can be managed at a time.
6514 static int update_array_info(struct mddev *mddev, mdu_array_info_t *info)
6520 /* calculate expected state,ignoring low bits */
6521 if (mddev->bitmap && mddev->bitmap_info.offset)
6522 state |= (1 << MD_SB_BITMAP_PRESENT);
6524 if (mddev->major_version != info->major_version ||
6525 mddev->minor_version != info->minor_version ||
6526 /* mddev->patch_version != info->patch_version || */
6527 mddev->ctime != info->ctime ||
6528 mddev->level != info->level ||
6529 /* mddev->layout != info->layout || */
6530 mddev->persistent != !info->not_persistent ||
6531 mddev->chunk_sectors != info->chunk_size >> 9 ||
6532 /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
6533 ((state^info->state) & 0xfffffe00)
6536 /* Check there is only one change */
6537 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
6539 if (mddev->raid_disks != info->raid_disks)
6541 if (mddev->layout != info->layout)
6543 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT))
6550 if (mddev->layout != info->layout) {
6552 * we don't need to do anything at the md level, the
6553 * personality will take care of it all.
6555 if (mddev->pers->check_reshape == NULL)
6558 mddev->new_layout = info->layout;
6559 rv = mddev->pers->check_reshape(mddev);
6561 mddev->new_layout = mddev->layout;
6565 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
6566 rv = update_size(mddev, (sector_t)info->size * 2);
6568 if (mddev->raid_disks != info->raid_disks)
6569 rv = update_raid_disks(mddev, info->raid_disks);
6571 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) {
6572 if (mddev->pers->quiesce == NULL || mddev->thread == NULL) {
6576 if (mddev->recovery || mddev->sync_thread) {
6580 if (info->state & (1<<MD_SB_BITMAP_PRESENT)) {
6581 struct bitmap *bitmap;
6582 /* add the bitmap */
6583 if (mddev->bitmap) {
6587 if (mddev->bitmap_info.default_offset == 0) {
6591 mddev->bitmap_info.offset =
6592 mddev->bitmap_info.default_offset;
6593 mddev->bitmap_info.space =
6594 mddev->bitmap_info.default_space;
6595 mddev->pers->quiesce(mddev, 1);
6596 bitmap = bitmap_create(mddev, -1);
6597 if (!IS_ERR(bitmap)) {
6598 mddev->bitmap = bitmap;
6599 rv = bitmap_load(mddev);
6601 rv = PTR_ERR(bitmap);
6603 bitmap_destroy(mddev);
6604 mddev->pers->quiesce(mddev, 0);
6606 /* remove the bitmap */
6607 if (!mddev->bitmap) {
6611 if (mddev->bitmap->storage.file) {
6615 if (mddev->bitmap_info.nodes) {
6616 /* hold PW on all the bitmap lock */
6617 if (md_cluster_ops->lock_all_bitmaps(mddev) <= 0) {
6618 printk("md: can't change bitmap to none since the"
6619 " array is in use by more than one node\n");
6621 md_cluster_ops->unlock_all_bitmaps(mddev);
6625 mddev->bitmap_info.nodes = 0;
6626 md_cluster_ops->leave(mddev);
6628 mddev->pers->quiesce(mddev, 1);
6629 bitmap_destroy(mddev);
6630 mddev->pers->quiesce(mddev, 0);
6631 mddev->bitmap_info.offset = 0;
6634 md_update_sb(mddev, 1);
6640 static int set_disk_faulty(struct mddev *mddev, dev_t dev)
6642 struct md_rdev *rdev;
6645 if (mddev->pers == NULL)
6649 rdev = find_rdev_rcu(mddev, dev);
6653 md_error(mddev, rdev);
6654 if (!test_bit(Faulty, &rdev->flags))
6662 * We have a problem here : there is no easy way to give a CHS
6663 * virtual geometry. We currently pretend that we have a 2 heads
6664 * 4 sectors (with a BIG number of cylinders...). This drives
6665 * dosfs just mad... ;-)
6667 static int md_getgeo(struct block_device *bdev, struct hd_geometry *geo)
6669 struct mddev *mddev = bdev->bd_disk->private_data;
6673 geo->cylinders = mddev->array_sectors / 8;
6677 static inline bool md_ioctl_valid(unsigned int cmd)
6682 case GET_ARRAY_INFO:
6683 case GET_BITMAP_FILE:
6686 case HOT_REMOVE_DISK:
6689 case RESTART_ARRAY_RW:
6691 case SET_ARRAY_INFO:
6692 case SET_BITMAP_FILE:
6693 case SET_DISK_FAULTY:
6696 case CLUSTERED_DISK_NACK:
6703 static int md_ioctl(struct block_device *bdev, fmode_t mode,
6704 unsigned int cmd, unsigned long arg)
6707 void __user *argp = (void __user *)arg;
6708 struct mddev *mddev = NULL;
6711 if (!md_ioctl_valid(cmd))
6716 case GET_ARRAY_INFO:
6720 if (!capable(CAP_SYS_ADMIN))
6725 * Commands dealing with the RAID driver but not any
6730 err = get_version(argp);
6736 autostart_arrays(arg);
6743 * Commands creating/starting a new array:
6746 mddev = bdev->bd_disk->private_data;
6753 /* Some actions do not requires the mutex */
6755 case GET_ARRAY_INFO:
6756 if (!mddev->raid_disks && !mddev->external)
6759 err = get_array_info(mddev, argp);
6763 if (!mddev->raid_disks && !mddev->external)
6766 err = get_disk_info(mddev, argp);
6769 case SET_DISK_FAULTY:
6770 err = set_disk_faulty(mddev, new_decode_dev(arg));
6773 case GET_BITMAP_FILE:
6774 err = get_bitmap_file(mddev, argp);
6779 if (cmd == ADD_NEW_DISK)
6780 /* need to ensure md_delayed_delete() has completed */
6781 flush_workqueue(md_misc_wq);
6783 if (cmd == HOT_REMOVE_DISK)
6784 /* need to ensure recovery thread has run */
6785 wait_event_interruptible_timeout(mddev->sb_wait,
6786 !test_bit(MD_RECOVERY_NEEDED,
6788 msecs_to_jiffies(5000));
6789 if (cmd == STOP_ARRAY || cmd == STOP_ARRAY_RO) {
6790 /* Need to flush page cache, and ensure no-one else opens
6793 mutex_lock(&mddev->open_mutex);
6794 if (mddev->pers && atomic_read(&mddev->openers) > 1) {
6795 mutex_unlock(&mddev->open_mutex);
6799 set_bit(MD_STILL_CLOSED, &mddev->flags);
6800 mutex_unlock(&mddev->open_mutex);
6801 sync_blockdev(bdev);
6803 err = mddev_lock(mddev);
6806 "md: ioctl lock interrupted, reason %d, cmd %d\n",
6811 if (cmd == SET_ARRAY_INFO) {
6812 mdu_array_info_t info;
6814 memset(&info, 0, sizeof(info));
6815 else if (copy_from_user(&info, argp, sizeof(info))) {
6820 err = update_array_info(mddev, &info);
6822 printk(KERN_WARNING "md: couldn't update"
6823 " array info. %d\n", err);
6828 if (!list_empty(&mddev->disks)) {
6830 "md: array %s already has disks!\n",
6835 if (mddev->raid_disks) {
6837 "md: array %s already initialised!\n",
6842 err = set_array_info(mddev, &info);
6844 printk(KERN_WARNING "md: couldn't set"
6845 " array info. %d\n", err);
6852 * Commands querying/configuring an existing array:
6854 /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
6855 * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */
6856 if ((!mddev->raid_disks && !mddev->external)
6857 && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY
6858 && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE
6859 && cmd != GET_BITMAP_FILE) {
6865 * Commands even a read-only array can execute:
6868 case RESTART_ARRAY_RW:
6869 err = restart_array(mddev);
6873 err = do_md_stop(mddev, 0, bdev);
6877 err = md_set_readonly(mddev, bdev);
6880 case HOT_REMOVE_DISK:
6881 err = hot_remove_disk(mddev, new_decode_dev(arg));
6885 /* We can support ADD_NEW_DISK on read-only arrays
6886 * on if we are re-adding a preexisting device.
6887 * So require mddev->pers and MD_DISK_SYNC.
6890 mdu_disk_info_t info;
6891 if (copy_from_user(&info, argp, sizeof(info)))
6893 else if (!(info.state & (1<<MD_DISK_SYNC)))
6894 /* Need to clear read-only for this */
6897 err = add_new_disk(mddev, &info);
6903 if (get_user(ro, (int __user *)(arg))) {
6909 /* if the bdev is going readonly the value of mddev->ro
6910 * does not matter, no writes are coming
6915 /* are we are already prepared for writes? */
6919 /* transitioning to readauto need only happen for
6920 * arrays that call md_write_start
6923 err = restart_array(mddev);
6926 set_disk_ro(mddev->gendisk, 0);
6933 * The remaining ioctls are changing the state of the
6934 * superblock, so we do not allow them on read-only arrays.
6936 if (mddev->ro && mddev->pers) {
6937 if (mddev->ro == 2) {
6939 sysfs_notify_dirent_safe(mddev->sysfs_state);
6940 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6941 /* mddev_unlock will wake thread */
6942 /* If a device failed while we were read-only, we
6943 * need to make sure the metadata is updated now.
6945 if (test_bit(MD_CHANGE_DEVS, &mddev->flags)) {
6946 mddev_unlock(mddev);
6947 wait_event(mddev->sb_wait,
6948 !test_bit(MD_CHANGE_DEVS, &mddev->flags) &&
6949 !test_bit(MD_CHANGE_PENDING, &mddev->flags));
6950 mddev_lock_nointr(mddev);
6961 mdu_disk_info_t info;
6962 if (copy_from_user(&info, argp, sizeof(info)))
6965 err = add_new_disk(mddev, &info);
6969 case CLUSTERED_DISK_NACK:
6970 if (mddev_is_clustered(mddev))
6971 md_cluster_ops->new_disk_ack(mddev, false);
6977 err = hot_add_disk(mddev, new_decode_dev(arg));
6981 err = do_md_run(mddev);
6984 case SET_BITMAP_FILE:
6985 err = set_bitmap_file(mddev, (int)arg);
6994 if (mddev->hold_active == UNTIL_IOCTL &&
6996 mddev->hold_active = 0;
6997 mddev_unlock(mddev);
7001 #ifdef CONFIG_COMPAT
7002 static int md_compat_ioctl(struct block_device *bdev, fmode_t mode,
7003 unsigned int cmd, unsigned long arg)
7006 case HOT_REMOVE_DISK:
7008 case SET_DISK_FAULTY:
7009 case SET_BITMAP_FILE:
7010 /* These take in integer arg, do not convert */
7013 arg = (unsigned long)compat_ptr(arg);
7017 return md_ioctl(bdev, mode, cmd, arg);
7019 #endif /* CONFIG_COMPAT */
7021 static int md_open(struct block_device *bdev, fmode_t mode)
7024 * Succeed if we can lock the mddev, which confirms that
7025 * it isn't being stopped right now.
7027 struct mddev *mddev = mddev_find(bdev->bd_dev);
7033 if (mddev->gendisk != bdev->bd_disk) {
7034 /* we are racing with mddev_put which is discarding this
7038 /* Wait until bdev->bd_disk is definitely gone */
7039 flush_workqueue(md_misc_wq);
7040 /* Then retry the open from the top */
7041 return -ERESTARTSYS;
7043 BUG_ON(mddev != bdev->bd_disk->private_data);
7045 if ((err = mutex_lock_interruptible(&mddev->open_mutex)))
7049 atomic_inc(&mddev->openers);
7050 clear_bit(MD_STILL_CLOSED, &mddev->flags);
7051 mutex_unlock(&mddev->open_mutex);
7053 check_disk_change(bdev);
7058 static void md_release(struct gendisk *disk, fmode_t mode)
7060 struct mddev *mddev = disk->private_data;
7063 atomic_dec(&mddev->openers);
7067 static int md_media_changed(struct gendisk *disk)
7069 struct mddev *mddev = disk->private_data;
7071 return mddev->changed;
7074 static int md_revalidate(struct gendisk *disk)
7076 struct mddev *mddev = disk->private_data;
7081 static const struct block_device_operations md_fops =
7083 .owner = THIS_MODULE,
7085 .release = md_release,
7087 #ifdef CONFIG_COMPAT
7088 .compat_ioctl = md_compat_ioctl,
7090 .getgeo = md_getgeo,
7091 .media_changed = md_media_changed,
7092 .revalidate_disk= md_revalidate,
7095 static int md_thread(void *arg)
7097 struct md_thread *thread = arg;
7100 * md_thread is a 'system-thread', it's priority should be very
7101 * high. We avoid resource deadlocks individually in each
7102 * raid personality. (RAID5 does preallocation) We also use RR and
7103 * the very same RT priority as kswapd, thus we will never get
7104 * into a priority inversion deadlock.
7106 * we definitely have to have equal or higher priority than
7107 * bdflush, otherwise bdflush will deadlock if there are too
7108 * many dirty RAID5 blocks.
7111 allow_signal(SIGKILL);
7112 while (!kthread_should_stop()) {
7114 /* We need to wait INTERRUPTIBLE so that
7115 * we don't add to the load-average.
7116 * That means we need to be sure no signals are
7119 if (signal_pending(current))
7120 flush_signals(current);
7122 wait_event_interruptible_timeout
7124 test_bit(THREAD_WAKEUP, &thread->flags)
7125 || kthread_should_stop(),
7128 clear_bit(THREAD_WAKEUP, &thread->flags);
7129 if (!kthread_should_stop())
7130 thread->run(thread);
7136 void md_wakeup_thread(struct md_thread *thread)
7139 pr_debug("md: waking up MD thread %s.\n", thread->tsk->comm);
7140 set_bit(THREAD_WAKEUP, &thread->flags);
7141 wake_up(&thread->wqueue);
7144 EXPORT_SYMBOL(md_wakeup_thread);
7146 struct md_thread *md_register_thread(void (*run) (struct md_thread *),
7147 struct mddev *mddev, const char *name)
7149 struct md_thread *thread;
7151 thread = kzalloc(sizeof(struct md_thread), GFP_KERNEL);
7155 init_waitqueue_head(&thread->wqueue);
7158 thread->mddev = mddev;
7159 thread->timeout = MAX_SCHEDULE_TIMEOUT;
7160 thread->tsk = kthread_run(md_thread, thread,
7162 mdname(thread->mddev),
7164 if (IS_ERR(thread->tsk)) {
7170 EXPORT_SYMBOL(md_register_thread);
7172 void md_unregister_thread(struct md_thread **threadp)
7174 struct md_thread *thread = *threadp;
7177 pr_debug("interrupting MD-thread pid %d\n", task_pid_nr(thread->tsk));
7178 /* Locking ensures that mddev_unlock does not wake_up a
7179 * non-existent thread
7181 spin_lock(&pers_lock);
7183 spin_unlock(&pers_lock);
7185 kthread_stop(thread->tsk);
7188 EXPORT_SYMBOL(md_unregister_thread);
7190 void md_error(struct mddev *mddev, struct md_rdev *rdev)
7192 if (!rdev || test_bit(Faulty, &rdev->flags))
7195 if (!mddev->pers || !mddev->pers->error_handler)
7197 mddev->pers->error_handler(mddev,rdev);
7198 if (mddev->degraded)
7199 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
7200 sysfs_notify_dirent_safe(rdev->sysfs_state);
7201 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
7202 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7203 md_wakeup_thread(mddev->thread);
7204 if (mddev->event_work.func)
7205 queue_work(md_misc_wq, &mddev->event_work);
7206 md_new_event(mddev);
7208 EXPORT_SYMBOL(md_error);
7210 /* seq_file implementation /proc/mdstat */
7212 static void status_unused(struct seq_file *seq)
7215 struct md_rdev *rdev;
7217 seq_printf(seq, "unused devices: ");
7219 list_for_each_entry(rdev, &pending_raid_disks, same_set) {
7220 char b[BDEVNAME_SIZE];
7222 seq_printf(seq, "%s ",
7223 bdevname(rdev->bdev,b));
7226 seq_printf(seq, "<none>");
7228 seq_printf(seq, "\n");
7231 static int status_resync(struct seq_file *seq, struct mddev *mddev)
7233 sector_t max_sectors, resync, res;
7234 unsigned long dt, db;
7237 unsigned int per_milli;
7239 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ||
7240 test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
7241 max_sectors = mddev->resync_max_sectors;
7243 max_sectors = mddev->dev_sectors;
7245 resync = mddev->curr_resync;
7247 if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
7248 /* Still cleaning up */
7249 resync = max_sectors;
7251 resync -= atomic_read(&mddev->recovery_active);
7254 if (mddev->recovery_cp < MaxSector) {
7255 seq_printf(seq, "\tresync=PENDING");
7261 seq_printf(seq, "\tresync=DELAYED");
7265 WARN_ON(max_sectors == 0);
7266 /* Pick 'scale' such that (resync>>scale)*1000 will fit
7267 * in a sector_t, and (max_sectors>>scale) will fit in a
7268 * u32, as those are the requirements for sector_div.
7269 * Thus 'scale' must be at least 10
7272 if (sizeof(sector_t) > sizeof(unsigned long)) {
7273 while ( max_sectors/2 > (1ULL<<(scale+32)))
7276 res = (resync>>scale)*1000;
7277 sector_div(res, (u32)((max_sectors>>scale)+1));
7281 int i, x = per_milli/50, y = 20-x;
7282 seq_printf(seq, "[");
7283 for (i = 0; i < x; i++)
7284 seq_printf(seq, "=");
7285 seq_printf(seq, ">");
7286 for (i = 0; i < y; i++)
7287 seq_printf(seq, ".");
7288 seq_printf(seq, "] ");
7290 seq_printf(seq, " %s =%3u.%u%% (%llu/%llu)",
7291 (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)?
7293 (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)?
7295 (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
7296 "resync" : "recovery"))),
7297 per_milli/10, per_milli % 10,
7298 (unsigned long long) resync/2,
7299 (unsigned long long) max_sectors/2);
7302 * dt: time from mark until now
7303 * db: blocks written from mark until now
7304 * rt: remaining time
7306 * rt is a sector_t, so could be 32bit or 64bit.
7307 * So we divide before multiply in case it is 32bit and close
7309 * We scale the divisor (db) by 32 to avoid losing precision
7310 * near the end of resync when the number of remaining sectors
7312 * We then divide rt by 32 after multiplying by db to compensate.
7313 * The '+1' avoids division by zero if db is very small.
7315 dt = ((jiffies - mddev->resync_mark) / HZ);
7317 db = (mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active))
7318 - mddev->resync_mark_cnt;
7320 rt = max_sectors - resync; /* number of remaining sectors */
7321 sector_div(rt, db/32+1);
7325 seq_printf(seq, " finish=%lu.%lumin", (unsigned long)rt / 60,
7326 ((unsigned long)rt % 60)/6);
7328 seq_printf(seq, " speed=%ldK/sec", db/2/dt);
7332 static void *md_seq_start(struct seq_file *seq, loff_t *pos)
7334 struct list_head *tmp;
7336 struct mddev *mddev;
7344 spin_lock(&all_mddevs_lock);
7345 list_for_each(tmp,&all_mddevs)
7347 mddev = list_entry(tmp, struct mddev, all_mddevs);
7349 spin_unlock(&all_mddevs_lock);
7352 spin_unlock(&all_mddevs_lock);
7354 return (void*)2;/* tail */
7358 static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
7360 struct list_head *tmp;
7361 struct mddev *next_mddev, *mddev = v;
7367 spin_lock(&all_mddevs_lock);
7369 tmp = all_mddevs.next;
7371 tmp = mddev->all_mddevs.next;
7372 if (tmp != &all_mddevs)
7373 next_mddev = mddev_get(list_entry(tmp,struct mddev,all_mddevs));
7375 next_mddev = (void*)2;
7378 spin_unlock(&all_mddevs_lock);
7386 static void md_seq_stop(struct seq_file *seq, void *v)
7388 struct mddev *mddev = v;
7390 if (mddev && v != (void*)1 && v != (void*)2)
7394 static int md_seq_show(struct seq_file *seq, void *v)
7396 struct mddev *mddev = v;
7398 struct md_rdev *rdev;
7400 if (v == (void*)1) {
7401 struct md_personality *pers;
7402 seq_printf(seq, "Personalities : ");
7403 spin_lock(&pers_lock);
7404 list_for_each_entry(pers, &pers_list, list)
7405 seq_printf(seq, "[%s] ", pers->name);
7407 spin_unlock(&pers_lock);
7408 seq_printf(seq, "\n");
7409 seq->poll_event = atomic_read(&md_event_count);
7412 if (v == (void*)2) {
7417 spin_lock(&mddev->lock);
7418 if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
7419 seq_printf(seq, "%s : %sactive", mdname(mddev),
7420 mddev->pers ? "" : "in");
7423 seq_printf(seq, " (read-only)");
7425 seq_printf(seq, " (auto-read-only)");
7426 seq_printf(seq, " %s", mddev->pers->name);
7431 rdev_for_each_rcu(rdev, mddev) {
7432 char b[BDEVNAME_SIZE];
7433 seq_printf(seq, " %s[%d]",
7434 bdevname(rdev->bdev,b), rdev->desc_nr);
7435 if (test_bit(WriteMostly, &rdev->flags))
7436 seq_printf(seq, "(W)");
7437 if (test_bit(Journal, &rdev->flags))
7438 seq_printf(seq, "(J)");
7439 if (test_bit(Faulty, &rdev->flags)) {
7440 seq_printf(seq, "(F)");
7443 if (rdev->raid_disk < 0)
7444 seq_printf(seq, "(S)"); /* spare */
7445 if (test_bit(Replacement, &rdev->flags))
7446 seq_printf(seq, "(R)");
7447 sectors += rdev->sectors;
7451 if (!list_empty(&mddev->disks)) {
7453 seq_printf(seq, "\n %llu blocks",
7454 (unsigned long long)
7455 mddev->array_sectors / 2);
7457 seq_printf(seq, "\n %llu blocks",
7458 (unsigned long long)sectors / 2);
7460 if (mddev->persistent) {
7461 if (mddev->major_version != 0 ||
7462 mddev->minor_version != 90) {
7463 seq_printf(seq," super %d.%d",
7464 mddev->major_version,
7465 mddev->minor_version);
7467 } else if (mddev->external)
7468 seq_printf(seq, " super external:%s",
7469 mddev->metadata_type);
7471 seq_printf(seq, " super non-persistent");
7474 mddev->pers->status(seq, mddev);
7475 seq_printf(seq, "\n ");
7476 if (mddev->pers->sync_request) {
7477 if (status_resync(seq, mddev))
7478 seq_printf(seq, "\n ");
7481 seq_printf(seq, "\n ");
7483 bitmap_status(seq, mddev->bitmap);
7485 seq_printf(seq, "\n");
7487 spin_unlock(&mddev->lock);
7492 static const struct seq_operations md_seq_ops = {
7493 .start = md_seq_start,
7494 .next = md_seq_next,
7495 .stop = md_seq_stop,
7496 .show = md_seq_show,
7499 static int md_seq_open(struct inode *inode, struct file *file)
7501 struct seq_file *seq;
7504 error = seq_open(file, &md_seq_ops);
7508 seq = file->private_data;
7509 seq->poll_event = atomic_read(&md_event_count);
7513 static int md_unloading;
7514 static unsigned int mdstat_poll(struct file *filp, poll_table *wait)
7516 struct seq_file *seq = filp->private_data;
7520 return POLLIN|POLLRDNORM|POLLERR|POLLPRI;
7521 poll_wait(filp, &md_event_waiters, wait);
7523 /* always allow read */
7524 mask = POLLIN | POLLRDNORM;
7526 if (seq->poll_event != atomic_read(&md_event_count))
7527 mask |= POLLERR | POLLPRI;
7531 static const struct file_operations md_seq_fops = {
7532 .owner = THIS_MODULE,
7533 .open = md_seq_open,
7535 .llseek = seq_lseek,
7536 .release = seq_release_private,
7537 .poll = mdstat_poll,
7540 int register_md_personality(struct md_personality *p)
7542 printk(KERN_INFO "md: %s personality registered for level %d\n",
7544 spin_lock(&pers_lock);
7545 list_add_tail(&p->list, &pers_list);
7546 spin_unlock(&pers_lock);
7549 EXPORT_SYMBOL(register_md_personality);
7551 int unregister_md_personality(struct md_personality *p)
7553 printk(KERN_INFO "md: %s personality unregistered\n", p->name);
7554 spin_lock(&pers_lock);
7555 list_del_init(&p->list);
7556 spin_unlock(&pers_lock);
7559 EXPORT_SYMBOL(unregister_md_personality);
7561 int register_md_cluster_operations(struct md_cluster_operations *ops,
7562 struct module *module)
7565 spin_lock(&pers_lock);
7566 if (md_cluster_ops != NULL)
7569 md_cluster_ops = ops;
7570 md_cluster_mod = module;
7572 spin_unlock(&pers_lock);
7575 EXPORT_SYMBOL(register_md_cluster_operations);
7577 int unregister_md_cluster_operations(void)
7579 spin_lock(&pers_lock);
7580 md_cluster_ops = NULL;
7581 spin_unlock(&pers_lock);
7584 EXPORT_SYMBOL(unregister_md_cluster_operations);
7586 int md_setup_cluster(struct mddev *mddev, int nodes)
7590 err = request_module("md-cluster");
7592 pr_err("md-cluster module not found.\n");
7596 spin_lock(&pers_lock);
7597 if (!md_cluster_ops || !try_module_get(md_cluster_mod)) {
7598 spin_unlock(&pers_lock);
7601 spin_unlock(&pers_lock);
7603 return md_cluster_ops->join(mddev, nodes);
7606 void md_cluster_stop(struct mddev *mddev)
7608 if (!md_cluster_ops)
7610 md_cluster_ops->leave(mddev);
7611 module_put(md_cluster_mod);
7614 static int is_mddev_idle(struct mddev *mddev, int init)
7616 struct md_rdev *rdev;
7622 rdev_for_each_rcu(rdev, mddev) {
7623 struct gendisk *disk = rdev->bdev->bd_contains->bd_disk;
7624 curr_events = (int)part_stat_read(&disk->part0, sectors[0]) +
7625 (int)part_stat_read(&disk->part0, sectors[1]) -
7626 atomic_read(&disk->sync_io);
7627 /* sync IO will cause sync_io to increase before the disk_stats
7628 * as sync_io is counted when a request starts, and
7629 * disk_stats is counted when it completes.
7630 * So resync activity will cause curr_events to be smaller than
7631 * when there was no such activity.
7632 * non-sync IO will cause disk_stat to increase without
7633 * increasing sync_io so curr_events will (eventually)
7634 * be larger than it was before. Once it becomes
7635 * substantially larger, the test below will cause
7636 * the array to appear non-idle, and resync will slow
7638 * If there is a lot of outstanding resync activity when
7639 * we set last_event to curr_events, then all that activity
7640 * completing might cause the array to appear non-idle
7641 * and resync will be slowed down even though there might
7642 * not have been non-resync activity. This will only
7643 * happen once though. 'last_events' will soon reflect
7644 * the state where there is little or no outstanding
7645 * resync requests, and further resync activity will
7646 * always make curr_events less than last_events.
7649 if (init || curr_events - rdev->last_events > 64) {
7650 rdev->last_events = curr_events;
7658 void md_done_sync(struct mddev *mddev, int blocks, int ok)
7660 /* another "blocks" (512byte) blocks have been synced */
7661 atomic_sub(blocks, &mddev->recovery_active);
7662 wake_up(&mddev->recovery_wait);
7664 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
7665 set_bit(MD_RECOVERY_ERROR, &mddev->recovery);
7666 md_wakeup_thread(mddev->thread);
7667 // stop recovery, signal do_sync ....
7670 EXPORT_SYMBOL(md_done_sync);
7672 /* md_write_start(mddev, bi)
7673 * If we need to update some array metadata (e.g. 'active' flag
7674 * in superblock) before writing, schedule a superblock update
7675 * and wait for it to complete.
7677 void md_write_start(struct mddev *mddev, struct bio *bi)
7680 if (bio_data_dir(bi) != WRITE)
7683 BUG_ON(mddev->ro == 1);
7684 if (mddev->ro == 2) {
7685 /* need to switch to read/write */
7687 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7688 md_wakeup_thread(mddev->thread);
7689 md_wakeup_thread(mddev->sync_thread);
7692 atomic_inc(&mddev->writes_pending);
7693 if (mddev->safemode == 1)
7694 mddev->safemode = 0;
7695 if (mddev->in_sync) {
7696 spin_lock(&mddev->lock);
7697 if (mddev->in_sync) {
7699 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
7700 set_bit(MD_CHANGE_PENDING, &mddev->flags);
7701 md_wakeup_thread(mddev->thread);
7704 spin_unlock(&mddev->lock);
7707 sysfs_notify_dirent_safe(mddev->sysfs_state);
7708 wait_event(mddev->sb_wait,
7709 !test_bit(MD_CHANGE_PENDING, &mddev->flags));
7711 EXPORT_SYMBOL(md_write_start);
7713 void md_write_end(struct mddev *mddev)
7715 if (atomic_dec_and_test(&mddev->writes_pending)) {
7716 if (mddev->safemode == 2)
7717 md_wakeup_thread(mddev->thread);
7718 else if (mddev->safemode_delay)
7719 mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay);
7722 EXPORT_SYMBOL(md_write_end);
7724 /* md_allow_write(mddev)
7725 * Calling this ensures that the array is marked 'active' so that writes
7726 * may proceed without blocking. It is important to call this before
7727 * attempting a GFP_KERNEL allocation while holding the mddev lock.
7728 * Must be called with mddev_lock held.
7730 * In the ->external case MD_CHANGE_PENDING can not be cleared until mddev->lock
7731 * is dropped, so return -EAGAIN after notifying userspace.
7733 int md_allow_write(struct mddev *mddev)
7739 if (!mddev->pers->sync_request)
7742 spin_lock(&mddev->lock);
7743 if (mddev->in_sync) {
7745 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
7746 set_bit(MD_CHANGE_PENDING, &mddev->flags);
7747 if (mddev->safemode_delay &&
7748 mddev->safemode == 0)
7749 mddev->safemode = 1;
7750 spin_unlock(&mddev->lock);
7751 md_update_sb(mddev, 0);
7752 sysfs_notify_dirent_safe(mddev->sysfs_state);
7754 spin_unlock(&mddev->lock);
7756 if (test_bit(MD_CHANGE_PENDING, &mddev->flags))
7761 EXPORT_SYMBOL_GPL(md_allow_write);
7763 #define SYNC_MARKS 10
7764 #define SYNC_MARK_STEP (3*HZ)
7765 #define UPDATE_FREQUENCY (5*60*HZ)
7766 void md_do_sync(struct md_thread *thread)
7768 struct mddev *mddev = thread->mddev;
7769 struct mddev *mddev2;
7770 unsigned int currspeed = 0,
7772 sector_t max_sectors,j, io_sectors, recovery_done;
7773 unsigned long mark[SYNC_MARKS];
7774 unsigned long update_time;
7775 sector_t mark_cnt[SYNC_MARKS];
7777 struct list_head *tmp;
7778 sector_t last_check;
7780 struct md_rdev *rdev;
7781 char *desc, *action = NULL;
7782 struct blk_plug plug;
7783 bool cluster_resync_finished = false;
7785 /* just incase thread restarts... */
7786 if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
7788 if (mddev->ro) {/* never try to sync a read-only array */
7789 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
7793 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
7794 if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)) {
7795 desc = "data-check";
7797 } else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
7798 desc = "requested-resync";
7802 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
7807 mddev->last_sync_action = action ?: desc;
7809 /* we overload curr_resync somewhat here.
7810 * 0 == not engaged in resync at all
7811 * 2 == checking that there is no conflict with another sync
7812 * 1 == like 2, but have yielded to allow conflicting resync to
7814 * other == active in resync - this many blocks
7816 * Before starting a resync we must have set curr_resync to
7817 * 2, and then checked that every "conflicting" array has curr_resync
7818 * less than ours. When we find one that is the same or higher
7819 * we wait on resync_wait. To avoid deadlock, we reduce curr_resync
7820 * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
7821 * This will mean we have to start checking from the beginning again.
7826 mddev->curr_resync = 2;
7829 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
7831 for_each_mddev(mddev2, tmp) {
7832 if (mddev2 == mddev)
7834 if (!mddev->parallel_resync
7835 && mddev2->curr_resync
7836 && match_mddev_units(mddev, mddev2)) {
7838 if (mddev < mddev2 && mddev->curr_resync == 2) {
7839 /* arbitrarily yield */
7840 mddev->curr_resync = 1;
7841 wake_up(&resync_wait);
7843 if (mddev > mddev2 && mddev->curr_resync == 1)
7844 /* no need to wait here, we can wait the next
7845 * time 'round when curr_resync == 2
7848 /* We need to wait 'interruptible' so as not to
7849 * contribute to the load average, and not to
7850 * be caught by 'softlockup'
7852 prepare_to_wait(&resync_wait, &wq, TASK_INTERRUPTIBLE);
7853 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
7854 mddev2->curr_resync >= mddev->curr_resync) {
7855 printk(KERN_INFO "md: delaying %s of %s"
7856 " until %s has finished (they"
7857 " share one or more physical units)\n",
7858 desc, mdname(mddev), mdname(mddev2));
7860 if (signal_pending(current))
7861 flush_signals(current);
7863 finish_wait(&resync_wait, &wq);
7866 finish_wait(&resync_wait, &wq);
7869 } while (mddev->curr_resync < 2);
7872 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
7873 /* resync follows the size requested by the personality,
7874 * which defaults to physical size, but can be virtual size
7876 max_sectors = mddev->resync_max_sectors;
7877 atomic64_set(&mddev->resync_mismatches, 0);
7878 /* we don't use the checkpoint if there's a bitmap */
7879 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
7880 j = mddev->resync_min;
7881 else if (!mddev->bitmap)
7882 j = mddev->recovery_cp;
7884 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
7885 max_sectors = mddev->resync_max_sectors;
7887 /* recovery follows the physical size of devices */
7888 max_sectors = mddev->dev_sectors;
7891 rdev_for_each_rcu(rdev, mddev)
7892 if (rdev->raid_disk >= 0 &&
7893 !test_bit(Journal, &rdev->flags) &&
7894 !test_bit(Faulty, &rdev->flags) &&
7895 !test_bit(In_sync, &rdev->flags) &&
7896 rdev->recovery_offset < j)
7897 j = rdev->recovery_offset;
7900 /* If there is a bitmap, we need to make sure all
7901 * writes that started before we added a spare
7902 * complete before we start doing a recovery.
7903 * Otherwise the write might complete and (via
7904 * bitmap_endwrite) set a bit in the bitmap after the
7905 * recovery has checked that bit and skipped that
7908 if (mddev->bitmap) {
7909 mddev->pers->quiesce(mddev, 1);
7910 mddev->pers->quiesce(mddev, 0);
7914 printk(KERN_INFO "md: %s of RAID array %s\n", desc, mdname(mddev));
7915 printk(KERN_INFO "md: minimum _guaranteed_ speed:"
7916 " %d KB/sec/disk.\n", speed_min(mddev));
7917 printk(KERN_INFO "md: using maximum available idle IO bandwidth "
7918 "(but not more than %d KB/sec) for %s.\n",
7919 speed_max(mddev), desc);
7921 is_mddev_idle(mddev, 1); /* this initializes IO event counters */
7924 for (m = 0; m < SYNC_MARKS; m++) {
7926 mark_cnt[m] = io_sectors;
7929 mddev->resync_mark = mark[last_mark];
7930 mddev->resync_mark_cnt = mark_cnt[last_mark];
7933 * Tune reconstruction:
7935 window = 32*(PAGE_SIZE/512);
7936 printk(KERN_INFO "md: using %dk window, over a total of %lluk.\n",
7937 window/2, (unsigned long long)max_sectors/2);
7939 atomic_set(&mddev->recovery_active, 0);
7944 "md: resuming %s of %s from checkpoint.\n",
7945 desc, mdname(mddev));
7946 mddev->curr_resync = j;
7948 mddev->curr_resync = 3; /* no longer delayed */
7949 mddev->curr_resync_completed = j;
7950 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
7951 md_new_event(mddev);
7952 update_time = jiffies;
7954 blk_start_plug(&plug);
7955 while (j < max_sectors) {
7960 if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
7961 ((mddev->curr_resync > mddev->curr_resync_completed &&
7962 (mddev->curr_resync - mddev->curr_resync_completed)
7963 > (max_sectors >> 4)) ||
7964 time_after_eq(jiffies, update_time + UPDATE_FREQUENCY) ||
7965 (j - mddev->curr_resync_completed)*2
7966 >= mddev->resync_max - mddev->curr_resync_completed ||
7967 mddev->curr_resync_completed > mddev->resync_max
7969 /* time to update curr_resync_completed */
7970 wait_event(mddev->recovery_wait,
7971 atomic_read(&mddev->recovery_active) == 0);
7972 mddev->curr_resync_completed = j;
7973 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) &&
7974 j > mddev->recovery_cp)
7975 mddev->recovery_cp = j;
7976 update_time = jiffies;
7977 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
7978 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
7981 while (j >= mddev->resync_max &&
7982 !test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
7983 /* As this condition is controlled by user-space,
7984 * we can block indefinitely, so use '_interruptible'
7985 * to avoid triggering warnings.
7987 flush_signals(current); /* just in case */
7988 wait_event_interruptible(mddev->recovery_wait,
7989 mddev->resync_max > j
7990 || test_bit(MD_RECOVERY_INTR,
7994 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
7997 sectors = mddev->pers->sync_request(mddev, j, &skipped);
7999 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
8003 if (!skipped) { /* actual IO requested */
8004 io_sectors += sectors;
8005 atomic_add(sectors, &mddev->recovery_active);
8008 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
8012 if (j > max_sectors)
8013 /* when skipping, extra large numbers can be returned. */
8016 mddev->curr_resync = j;
8017 mddev->curr_mark_cnt = io_sectors;
8018 if (last_check == 0)
8019 /* this is the earliest that rebuild will be
8020 * visible in /proc/mdstat
8022 md_new_event(mddev);
8024 if (last_check + window > io_sectors || j == max_sectors)
8027 last_check = io_sectors;
8029 if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) {
8031 int next = (last_mark+1) % SYNC_MARKS;
8033 mddev->resync_mark = mark[next];
8034 mddev->resync_mark_cnt = mark_cnt[next];
8035 mark[next] = jiffies;
8036 mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active);
8040 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
8044 * this loop exits only if either when we are slower than
8045 * the 'hard' speed limit, or the system was IO-idle for
8047 * the system might be non-idle CPU-wise, but we only care
8048 * about not overloading the IO subsystem. (things like an
8049 * e2fsck being done on the RAID array should execute fast)
8053 recovery_done = io_sectors - atomic_read(&mddev->recovery_active);
8054 currspeed = ((unsigned long)(recovery_done - mddev->resync_mark_cnt))/2
8055 /((jiffies-mddev->resync_mark)/HZ +1) +1;
8057 if (currspeed > speed_min(mddev)) {
8058 if (currspeed > speed_max(mddev)) {
8062 if (!is_mddev_idle(mddev, 0)) {
8064 * Give other IO more of a chance.
8065 * The faster the devices, the less we wait.
8067 wait_event(mddev->recovery_wait,
8068 !atomic_read(&mddev->recovery_active));
8072 printk(KERN_INFO "md: %s: %s %s.\n",mdname(mddev), desc,
8073 test_bit(MD_RECOVERY_INTR, &mddev->recovery)
8074 ? "interrupted" : "done");
8076 * this also signals 'finished resyncing' to md_stop
8078 blk_finish_plug(&plug);
8079 wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
8081 if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
8082 !test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
8083 mddev->curr_resync > 2) {
8084 mddev->curr_resync_completed = mddev->curr_resync;
8085 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
8087 /* tell personality and other nodes that we are finished */
8088 if (mddev_is_clustered(mddev)) {
8089 md_cluster_ops->resync_finish(mddev);
8090 cluster_resync_finished = true;
8092 mddev->pers->sync_request(mddev, max_sectors, &skipped);
8094 if (!test_bit(MD_RECOVERY_CHECK, &mddev->recovery) &&
8095 mddev->curr_resync > 2) {
8096 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
8097 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
8098 if (mddev->curr_resync >= mddev->recovery_cp) {
8100 "md: checkpointing %s of %s.\n",
8101 desc, mdname(mddev));
8102 if (test_bit(MD_RECOVERY_ERROR,
8104 mddev->recovery_cp =
8105 mddev->curr_resync_completed;
8107 mddev->recovery_cp =
8111 mddev->recovery_cp = MaxSector;
8113 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery))
8114 mddev->curr_resync = MaxSector;
8116 rdev_for_each_rcu(rdev, mddev)
8117 if (rdev->raid_disk >= 0 &&
8118 mddev->delta_disks >= 0 &&
8119 !test_bit(Journal, &rdev->flags) &&
8120 !test_bit(Faulty, &rdev->flags) &&
8121 !test_bit(In_sync, &rdev->flags) &&
8122 rdev->recovery_offset < mddev->curr_resync)
8123 rdev->recovery_offset = mddev->curr_resync;
8128 set_bit(MD_CHANGE_DEVS, &mddev->flags);
8130 if (mddev_is_clustered(mddev) &&
8131 test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
8132 !cluster_resync_finished)
8133 md_cluster_ops->resync_finish(mddev);
8135 spin_lock(&mddev->lock);
8136 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
8137 /* We completed so min/max setting can be forgotten if used. */
8138 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
8139 mddev->resync_min = 0;
8140 mddev->resync_max = MaxSector;
8141 } else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
8142 mddev->resync_min = mddev->curr_resync_completed;
8143 set_bit(MD_RECOVERY_DONE, &mddev->recovery);
8144 mddev->curr_resync = 0;
8145 spin_unlock(&mddev->lock);
8147 wake_up(&resync_wait);
8148 md_wakeup_thread(mddev->thread);
8151 EXPORT_SYMBOL_GPL(md_do_sync);
8153 static int remove_and_add_spares(struct mddev *mddev,
8154 struct md_rdev *this)
8156 struct md_rdev *rdev;
8160 rdev_for_each(rdev, mddev)
8161 if ((this == NULL || rdev == this) &&
8162 rdev->raid_disk >= 0 &&
8163 !test_bit(Blocked, &rdev->flags) &&
8164 (test_bit(Faulty, &rdev->flags) ||
8165 (!test_bit(In_sync, &rdev->flags) &&
8166 !test_bit(Journal, &rdev->flags))) &&
8167 atomic_read(&rdev->nr_pending)==0) {
8168 if (mddev->pers->hot_remove_disk(
8169 mddev, rdev) == 0) {
8170 sysfs_unlink_rdev(mddev, rdev);
8171 rdev->raid_disk = -1;
8175 if (removed && mddev->kobj.sd)
8176 sysfs_notify(&mddev->kobj, NULL, "degraded");
8178 if (this && removed)
8181 rdev_for_each(rdev, mddev) {
8182 if (this && this != rdev)
8184 if (test_bit(Candidate, &rdev->flags))
8186 if (rdev->raid_disk >= 0 &&
8187 !test_bit(In_sync, &rdev->flags) &&
8188 !test_bit(Journal, &rdev->flags) &&
8189 !test_bit(Faulty, &rdev->flags))
8191 if (rdev->raid_disk >= 0)
8193 if (test_bit(Faulty, &rdev->flags))
8195 if (!test_bit(Journal, &rdev->flags)) {
8197 ! (rdev->saved_raid_disk >= 0 &&
8198 !test_bit(Bitmap_sync, &rdev->flags)))
8201 rdev->recovery_offset = 0;
8204 hot_add_disk(mddev, rdev) == 0) {
8205 if (sysfs_link_rdev(mddev, rdev))
8206 /* failure here is OK */;
8207 if (!test_bit(Journal, &rdev->flags))
8209 md_new_event(mddev);
8210 set_bit(MD_CHANGE_DEVS, &mddev->flags);
8215 set_bit(MD_CHANGE_DEVS, &mddev->flags);
8219 static void md_start_sync(struct work_struct *ws)
8221 struct mddev *mddev = container_of(ws, struct mddev, del_work);
8224 if (mddev_is_clustered(mddev)) {
8225 ret = md_cluster_ops->resync_start(mddev);
8227 mddev->sync_thread = NULL;
8232 mddev->sync_thread = md_register_thread(md_do_sync,
8236 if (!mddev->sync_thread) {
8237 if (!(mddev_is_clustered(mddev) && ret == -EAGAIN))
8238 printk(KERN_ERR "%s: could not start resync"
8241 /* leave the spares where they are, it shouldn't hurt */
8242 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
8243 clear_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
8244 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
8245 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
8246 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
8247 wake_up(&resync_wait);
8248 if (test_and_clear_bit(MD_RECOVERY_RECOVER,
8250 if (mddev->sysfs_action)
8251 sysfs_notify_dirent_safe(mddev->sysfs_action);
8253 md_wakeup_thread(mddev->sync_thread);
8254 sysfs_notify_dirent_safe(mddev->sysfs_action);
8255 md_new_event(mddev);
8259 * This routine is regularly called by all per-raid-array threads to
8260 * deal with generic issues like resync and super-block update.
8261 * Raid personalities that don't have a thread (linear/raid0) do not
8262 * need this as they never do any recovery or update the superblock.
8264 * It does not do any resync itself, but rather "forks" off other threads
8265 * to do that as needed.
8266 * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
8267 * "->recovery" and create a thread at ->sync_thread.
8268 * When the thread finishes it sets MD_RECOVERY_DONE
8269 * and wakeups up this thread which will reap the thread and finish up.
8270 * This thread also removes any faulty devices (with nr_pending == 0).
8272 * The overall approach is:
8273 * 1/ if the superblock needs updating, update it.
8274 * 2/ If a recovery thread is running, don't do anything else.
8275 * 3/ If recovery has finished, clean up, possibly marking spares active.
8276 * 4/ If there are any faulty devices, remove them.
8277 * 5/ If array is degraded, try to add spares devices
8278 * 6/ If array has spares or is not in-sync, start a resync thread.
8280 void md_check_recovery(struct mddev *mddev)
8282 if (mddev->suspended)
8286 bitmap_daemon_work(mddev);
8288 if (signal_pending(current)) {
8289 if (mddev->pers->sync_request && !mddev->external) {
8290 printk(KERN_INFO "md: %s in immediate safe mode\n",
8292 mddev->safemode = 2;
8294 flush_signals(current);
8297 if (mddev->ro && !test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
8300 (mddev->flags & MD_UPDATE_SB_FLAGS & ~ (1<<MD_CHANGE_PENDING)) ||
8301 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
8302 test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
8303 test_bit(MD_RELOAD_SB, &mddev->flags) ||
8304 (mddev->external == 0 && mddev->safemode == 1) ||
8305 (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending)
8306 && !mddev->in_sync && mddev->recovery_cp == MaxSector)
8310 if (mddev_trylock(mddev)) {
8314 struct md_rdev *rdev;
8315 if (!mddev->external && mddev->in_sync)
8316 /* 'Blocked' flag not needed as failed devices
8317 * will be recorded if array switched to read/write.
8318 * Leaving it set will prevent the device
8319 * from being removed.
8321 rdev_for_each(rdev, mddev)
8322 clear_bit(Blocked, &rdev->flags);
8323 /* On a read-only array we can:
8324 * - remove failed devices
8325 * - add already-in_sync devices if the array itself
8327 * As we only add devices that are already in-sync,
8328 * we can activate the spares immediately.
8330 remove_and_add_spares(mddev, NULL);
8331 /* There is no thread, but we need to call
8332 * ->spare_active and clear saved_raid_disk
8334 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
8335 md_reap_sync_thread(mddev);
8336 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
8337 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
8338 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
8342 if (mddev_is_clustered(mddev)) {
8343 struct md_rdev *rdev;
8344 /* kick the device if another node issued a
8347 rdev_for_each(rdev, mddev) {
8348 if (test_and_clear_bit(ClusterRemove, &rdev->flags) &&
8349 rdev->raid_disk < 0)
8350 md_kick_rdev_from_array(rdev);
8353 if (test_and_clear_bit(MD_RELOAD_SB, &mddev->flags))
8354 md_reload_sb(mddev, mddev->good_device_nr);
8357 if (!mddev->external) {
8359 spin_lock(&mddev->lock);
8360 if (mddev->safemode &&
8361 !atomic_read(&mddev->writes_pending) &&
8363 mddev->recovery_cp == MaxSector) {
8366 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
8368 if (mddev->safemode == 1)
8369 mddev->safemode = 0;
8370 spin_unlock(&mddev->lock);
8372 sysfs_notify_dirent_safe(mddev->sysfs_state);
8375 if (mddev->flags & MD_UPDATE_SB_FLAGS)
8376 md_update_sb(mddev, 0);
8378 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
8379 !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
8380 /* resync/recovery still happening */
8381 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
8384 if (mddev->sync_thread) {
8385 md_reap_sync_thread(mddev);
8388 /* Set RUNNING before clearing NEEDED to avoid
8389 * any transients in the value of "sync_action".
8391 mddev->curr_resync_completed = 0;
8392 spin_lock(&mddev->lock);
8393 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
8394 spin_unlock(&mddev->lock);
8395 /* Clear some bits that don't mean anything, but
8398 clear_bit(MD_RECOVERY_INTR, &mddev->recovery);
8399 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
8401 if (!test_and_clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
8402 test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
8404 /* no recovery is running.
8405 * remove any failed drives, then
8406 * add spares if possible.
8407 * Spares are also removed and re-added, to allow
8408 * the personality to fail the re-add.
8411 if (mddev->reshape_position != MaxSector) {
8412 if (mddev->pers->check_reshape == NULL ||
8413 mddev->pers->check_reshape(mddev) != 0)
8414 /* Cannot proceed */
8416 set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
8417 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
8418 } else if ((spares = remove_and_add_spares(mddev, NULL))) {
8419 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
8420 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
8421 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
8422 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
8423 } else if (mddev->recovery_cp < MaxSector) {
8424 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
8425 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
8426 } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
8427 /* nothing to be done ... */
8430 if (mddev->pers->sync_request) {
8432 /* We are adding a device or devices to an array
8433 * which has the bitmap stored on all devices.
8434 * So make sure all bitmap pages get written
8436 bitmap_write_all(mddev->bitmap);
8438 INIT_WORK(&mddev->del_work, md_start_sync);
8439 queue_work(md_misc_wq, &mddev->del_work);
8443 if (!mddev->sync_thread) {
8444 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
8445 wake_up(&resync_wait);
8446 if (test_and_clear_bit(MD_RECOVERY_RECOVER,
8448 if (mddev->sysfs_action)
8449 sysfs_notify_dirent_safe(mddev->sysfs_action);
8452 wake_up(&mddev->sb_wait);
8453 mddev_unlock(mddev);
8456 EXPORT_SYMBOL(md_check_recovery);
8458 void md_reap_sync_thread(struct mddev *mddev)
8460 struct md_rdev *rdev;
8462 /* resync has finished, collect result */
8463 md_unregister_thread(&mddev->sync_thread);
8464 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
8465 !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
8467 /* activate any spares */
8468 if (mddev->pers->spare_active(mddev)) {
8469 sysfs_notify(&mddev->kobj, NULL,
8471 set_bit(MD_CHANGE_DEVS, &mddev->flags);
8474 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
8475 mddev->pers->finish_reshape)
8476 mddev->pers->finish_reshape(mddev);
8478 /* If array is no-longer degraded, then any saved_raid_disk
8479 * information must be scrapped.
8481 if (!mddev->degraded)
8482 rdev_for_each(rdev, mddev)
8483 rdev->saved_raid_disk = -1;
8485 md_update_sb(mddev, 1);
8486 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
8487 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
8488 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
8489 clear_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
8490 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
8491 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
8492 wake_up(&resync_wait);
8493 /* flag recovery needed just to double check */
8494 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
8495 sysfs_notify_dirent_safe(mddev->sysfs_action);
8496 md_new_event(mddev);
8497 if (mddev->event_work.func)
8498 queue_work(md_misc_wq, &mddev->event_work);
8500 EXPORT_SYMBOL(md_reap_sync_thread);
8502 void md_wait_for_blocked_rdev(struct md_rdev *rdev, struct mddev *mddev)
8504 sysfs_notify_dirent_safe(rdev->sysfs_state);
8505 wait_event_timeout(rdev->blocked_wait,
8506 !test_bit(Blocked, &rdev->flags) &&
8507 !test_bit(BlockedBadBlocks, &rdev->flags),
8508 msecs_to_jiffies(5000));
8509 rdev_dec_pending(rdev, mddev);
8511 EXPORT_SYMBOL(md_wait_for_blocked_rdev);
8513 void md_finish_reshape(struct mddev *mddev)
8515 /* called be personality module when reshape completes. */
8516 struct md_rdev *rdev;
8518 rdev_for_each(rdev, mddev) {
8519 if (rdev->data_offset > rdev->new_data_offset)
8520 rdev->sectors += rdev->data_offset - rdev->new_data_offset;
8522 rdev->sectors -= rdev->new_data_offset - rdev->data_offset;
8523 rdev->data_offset = rdev->new_data_offset;
8526 EXPORT_SYMBOL(md_finish_reshape);
8528 /* Bad block management */
8530 /* Returns 1 on success, 0 on failure */
8531 int rdev_set_badblocks(struct md_rdev *rdev, sector_t s, int sectors,
8536 s += rdev->new_data_offset;
8538 s += rdev->data_offset;
8539 rv = badblocks_set(&rdev->badblocks, s, sectors, 0);
8541 /* Make sure they get written out promptly */
8542 sysfs_notify_dirent_safe(rdev->sysfs_state);
8543 set_bit(MD_CHANGE_CLEAN, &rdev->mddev->flags);
8544 set_bit(MD_CHANGE_PENDING, &rdev->mddev->flags);
8545 md_wakeup_thread(rdev->mddev->thread);
8550 EXPORT_SYMBOL_GPL(rdev_set_badblocks);
8552 int rdev_clear_badblocks(struct md_rdev *rdev, sector_t s, int sectors,
8556 s += rdev->new_data_offset;
8558 s += rdev->data_offset;
8559 return badblocks_clear(&rdev->badblocks,
8562 EXPORT_SYMBOL_GPL(rdev_clear_badblocks);
8564 static int md_notify_reboot(struct notifier_block *this,
8565 unsigned long code, void *x)
8567 struct list_head *tmp;
8568 struct mddev *mddev;
8571 for_each_mddev(mddev, tmp) {
8572 if (mddev_trylock(mddev)) {
8574 __md_stop_writes(mddev);
8575 if (mddev->persistent)
8576 mddev->safemode = 2;
8577 mddev_unlock(mddev);
8582 * certain more exotic SCSI devices are known to be
8583 * volatile wrt too early system reboots. While the
8584 * right place to handle this issue is the given
8585 * driver, we do want to have a safe RAID driver ...
8593 static struct notifier_block md_notifier = {
8594 .notifier_call = md_notify_reboot,
8596 .priority = INT_MAX, /* before any real devices */
8599 static void md_geninit(void)
8601 pr_debug("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
8603 proc_create("mdstat", S_IRUGO, NULL, &md_seq_fops);
8606 static int __init md_init(void)
8610 md_wq = alloc_workqueue("md", WQ_MEM_RECLAIM, 0);
8614 md_misc_wq = alloc_workqueue("md_misc", 0, 0);
8618 if ((ret = register_blkdev(MD_MAJOR, "md")) < 0)
8621 if ((ret = register_blkdev(0, "mdp")) < 0)
8625 blk_register_region(MKDEV(MD_MAJOR, 0), 512, THIS_MODULE,
8626 md_probe, NULL, NULL);
8627 blk_register_region(MKDEV(mdp_major, 0), 1UL<<MINORBITS, THIS_MODULE,
8628 md_probe, NULL, NULL);
8630 register_reboot_notifier(&md_notifier);
8631 raid_table_header = register_sysctl_table(raid_root_table);
8637 unregister_blkdev(MD_MAJOR, "md");
8639 destroy_workqueue(md_misc_wq);
8641 destroy_workqueue(md_wq);
8646 static void check_sb_changes(struct mddev *mddev, struct md_rdev *rdev)
8648 struct mdp_superblock_1 *sb = page_address(rdev->sb_page);
8649 struct md_rdev *rdev2;
8651 char b[BDEVNAME_SIZE];
8653 /* Check for change of roles in the active devices */
8654 rdev_for_each(rdev2, mddev) {
8655 if (test_bit(Faulty, &rdev2->flags))
8658 /* Check if the roles changed */
8659 role = le16_to_cpu(sb->dev_roles[rdev2->desc_nr]);
8661 if (test_bit(Candidate, &rdev2->flags)) {
8662 if (role == 0xfffe) {
8663 pr_info("md: Removing Candidate device %s because add failed\n", bdevname(rdev2->bdev,b));
8664 md_kick_rdev_from_array(rdev2);
8668 clear_bit(Candidate, &rdev2->flags);
8671 if (role != rdev2->raid_disk) {
8673 if (rdev2->raid_disk == -1 && role != 0xffff) {
8674 rdev2->saved_raid_disk = role;
8675 ret = remove_and_add_spares(mddev, rdev2);
8676 pr_info("Activated spare: %s\n",
8677 bdevname(rdev2->bdev,b));
8680 * We just want to do the minimum to mark the disk
8681 * as faulty. The recovery is performed by the
8682 * one who initiated the error.
8684 if ((role == 0xfffe) || (role == 0xfffd)) {
8685 md_error(mddev, rdev2);
8686 clear_bit(Blocked, &rdev2->flags);
8691 if (mddev->raid_disks != le32_to_cpu(sb->raid_disks))
8692 update_raid_disks(mddev, le32_to_cpu(sb->raid_disks));
8694 /* Finally set the event to be up to date */
8695 mddev->events = le64_to_cpu(sb->events);
8698 static int read_rdev(struct mddev *mddev, struct md_rdev *rdev)
8701 struct page *swapout = rdev->sb_page;
8702 struct mdp_superblock_1 *sb;
8704 /* Store the sb page of the rdev in the swapout temporary
8705 * variable in case we err in the future
8707 rdev->sb_page = NULL;
8708 alloc_disk_sb(rdev);
8709 ClearPageUptodate(rdev->sb_page);
8710 rdev->sb_loaded = 0;
8711 err = super_types[mddev->major_version].load_super(rdev, NULL, mddev->minor_version);
8714 pr_warn("%s: %d Could not reload rdev(%d) err: %d. Restoring old values\n",
8715 __func__, __LINE__, rdev->desc_nr, err);
8716 put_page(rdev->sb_page);
8717 rdev->sb_page = swapout;
8718 rdev->sb_loaded = 1;
8722 sb = page_address(rdev->sb_page);
8723 /* Read the offset unconditionally, even if MD_FEATURE_RECOVERY_OFFSET
8727 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RECOVERY_OFFSET))
8728 rdev->recovery_offset = le64_to_cpu(sb->recovery_offset);
8730 /* The other node finished recovery, call spare_active to set
8731 * device In_sync and mddev->degraded
8733 if (rdev->recovery_offset == MaxSector &&
8734 !test_bit(In_sync, &rdev->flags) &&
8735 mddev->pers->spare_active(mddev))
8736 sysfs_notify(&mddev->kobj, NULL, "degraded");
8742 void md_reload_sb(struct mddev *mddev, int nr)
8744 struct md_rdev *rdev;
8748 rdev_for_each_rcu(rdev, mddev) {
8749 if (rdev->desc_nr == nr)
8753 if (!rdev || rdev->desc_nr != nr) {
8754 pr_warn("%s: %d Could not find rdev with nr %d\n", __func__, __LINE__, nr);
8758 err = read_rdev(mddev, rdev);
8762 check_sb_changes(mddev, rdev);
8764 /* Read all rdev's to update recovery_offset */
8765 rdev_for_each_rcu(rdev, mddev)
8766 read_rdev(mddev, rdev);
8768 EXPORT_SYMBOL(md_reload_sb);
8773 * Searches all registered partitions for autorun RAID arrays
8777 static LIST_HEAD(all_detected_devices);
8778 struct detected_devices_node {
8779 struct list_head list;
8783 void md_autodetect_dev(dev_t dev)
8785 struct detected_devices_node *node_detected_dev;
8787 node_detected_dev = kzalloc(sizeof(*node_detected_dev), GFP_KERNEL);
8788 if (node_detected_dev) {
8789 node_detected_dev->dev = dev;
8790 list_add_tail(&node_detected_dev->list, &all_detected_devices);
8792 printk(KERN_CRIT "md: md_autodetect_dev: kzalloc failed"
8793 ", skipping dev(%d,%d)\n", MAJOR(dev), MINOR(dev));
8797 static void autostart_arrays(int part)
8799 struct md_rdev *rdev;
8800 struct detected_devices_node *node_detected_dev;
8802 int i_scanned, i_passed;
8807 printk(KERN_INFO "md: Autodetecting RAID arrays.\n");
8809 while (!list_empty(&all_detected_devices) && i_scanned < INT_MAX) {
8811 node_detected_dev = list_entry(all_detected_devices.next,
8812 struct detected_devices_node, list);
8813 list_del(&node_detected_dev->list);
8814 dev = node_detected_dev->dev;
8815 kfree(node_detected_dev);
8816 rdev = md_import_device(dev,0, 90);
8820 if (test_bit(Faulty, &rdev->flags))
8823 set_bit(AutoDetected, &rdev->flags);
8824 list_add(&rdev->same_set, &pending_raid_disks);
8828 printk(KERN_INFO "md: Scanned %d and added %d devices.\n",
8829 i_scanned, i_passed);
8831 autorun_devices(part);
8834 #endif /* !MODULE */
8836 static __exit void md_exit(void)
8838 struct mddev *mddev;
8839 struct list_head *tmp;
8842 blk_unregister_region(MKDEV(MD_MAJOR,0), 512);
8843 blk_unregister_region(MKDEV(mdp_major,0), 1U << MINORBITS);
8845 unregister_blkdev(MD_MAJOR,"md");
8846 unregister_blkdev(mdp_major, "mdp");
8847 unregister_reboot_notifier(&md_notifier);
8848 unregister_sysctl_table(raid_table_header);
8850 /* We cannot unload the modules while some process is
8851 * waiting for us in select() or poll() - wake them up
8854 while (waitqueue_active(&md_event_waiters)) {
8855 /* not safe to leave yet */
8856 wake_up(&md_event_waiters);
8860 remove_proc_entry("mdstat", NULL);
8862 for_each_mddev(mddev, tmp) {
8863 export_array(mddev);
8864 mddev->hold_active = 0;
8866 destroy_workqueue(md_misc_wq);
8867 destroy_workqueue(md_wq);
8870 subsys_initcall(md_init);
8871 module_exit(md_exit)
8873 static int get_ro(char *buffer, struct kernel_param *kp)
8875 return sprintf(buffer, "%d", start_readonly);
8877 static int set_ro(const char *val, struct kernel_param *kp)
8879 return kstrtouint(val, 10, (unsigned int *)&start_readonly);
8882 module_param_call(start_ro, set_ro, get_ro, NULL, S_IRUSR|S_IWUSR);
8883 module_param(start_dirty_degraded, int, S_IRUGO|S_IWUSR);
8884 module_param_call(new_array, add_named_array, NULL, NULL, S_IWUSR);
8886 MODULE_LICENSE("GPL");
8887 MODULE_DESCRIPTION("MD RAID framework");
8889 MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR);