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 WARN_ON_ONCE(current == mddev->thread->tsk);
309 if (mddev->suspended++)
312 wait_event(mddev->sb_wait, atomic_read(&mddev->active_io) == 0);
313 mddev->pers->quiesce(mddev, 1);
315 del_timer_sync(&mddev->safemode_timer);
317 EXPORT_SYMBOL_GPL(mddev_suspend);
319 void mddev_resume(struct mddev *mddev)
321 if (--mddev->suspended)
323 wake_up(&mddev->sb_wait);
324 mddev->pers->quiesce(mddev, 0);
326 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
327 md_wakeup_thread(mddev->thread);
328 md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
330 EXPORT_SYMBOL_GPL(mddev_resume);
332 int mddev_congested(struct mddev *mddev, int bits)
334 struct md_personality *pers = mddev->pers;
338 if (mddev->suspended)
340 else if (pers && pers->congested)
341 ret = pers->congested(mddev, bits);
345 EXPORT_SYMBOL_GPL(mddev_congested);
346 static int md_congested(void *data, int bits)
348 struct mddev *mddev = data;
349 return mddev_congested(mddev, bits);
353 * Generic flush handling for md
356 static void md_end_flush(struct bio *bio)
358 struct md_rdev *rdev = bio->bi_private;
359 struct mddev *mddev = rdev->mddev;
361 rdev_dec_pending(rdev, mddev);
363 if (atomic_dec_and_test(&mddev->flush_pending)) {
364 /* The pre-request flush has finished */
365 queue_work(md_wq, &mddev->flush_work);
370 static void md_submit_flush_data(struct work_struct *ws);
372 static void submit_flushes(struct work_struct *ws)
374 struct mddev *mddev = container_of(ws, struct mddev, flush_work);
375 struct md_rdev *rdev;
377 INIT_WORK(&mddev->flush_work, md_submit_flush_data);
378 atomic_set(&mddev->flush_pending, 1);
380 rdev_for_each_rcu(rdev, mddev)
381 if (rdev->raid_disk >= 0 &&
382 !test_bit(Faulty, &rdev->flags)) {
383 /* Take two references, one is dropped
384 * when request finishes, one after
385 * we reclaim rcu_read_lock
388 atomic_inc(&rdev->nr_pending);
389 atomic_inc(&rdev->nr_pending);
391 bi = bio_alloc_mddev(GFP_NOIO, 0, mddev);
392 bi->bi_end_io = md_end_flush;
393 bi->bi_private = rdev;
394 bi->bi_bdev = rdev->bdev;
395 atomic_inc(&mddev->flush_pending);
396 submit_bio(WRITE_FLUSH, bi);
398 rdev_dec_pending(rdev, mddev);
401 if (atomic_dec_and_test(&mddev->flush_pending))
402 queue_work(md_wq, &mddev->flush_work);
405 static void md_submit_flush_data(struct work_struct *ws)
407 struct mddev *mddev = container_of(ws, struct mddev, flush_work);
408 struct bio *bio = mddev->flush_bio;
410 if (bio->bi_iter.bi_size == 0)
411 /* an empty barrier - all done */
414 bio->bi_rw &= ~REQ_FLUSH;
415 mddev->pers->make_request(mddev, bio);
418 mddev->flush_bio = NULL;
419 wake_up(&mddev->sb_wait);
422 void md_flush_request(struct mddev *mddev, struct bio *bio)
424 spin_lock_irq(&mddev->lock);
425 wait_event_lock_irq(mddev->sb_wait,
428 mddev->flush_bio = bio;
429 spin_unlock_irq(&mddev->lock);
431 INIT_WORK(&mddev->flush_work, submit_flushes);
432 queue_work(md_wq, &mddev->flush_work);
434 EXPORT_SYMBOL(md_flush_request);
436 void md_unplug(struct blk_plug_cb *cb, bool from_schedule)
438 struct mddev *mddev = cb->data;
439 md_wakeup_thread(mddev->thread);
442 EXPORT_SYMBOL(md_unplug);
444 static inline struct mddev *mddev_get(struct mddev *mddev)
446 atomic_inc(&mddev->active);
450 static void mddev_delayed_delete(struct work_struct *ws);
452 static void mddev_put(struct mddev *mddev)
454 struct bio_set *bs = NULL;
456 if (!atomic_dec_and_lock(&mddev->active, &all_mddevs_lock))
458 if (!mddev->raid_disks && list_empty(&mddev->disks) &&
459 mddev->ctime == 0 && !mddev->hold_active) {
460 /* Array is not configured at all, and not held active,
462 list_del_init(&mddev->all_mddevs);
464 mddev->bio_set = NULL;
465 if (mddev->gendisk) {
466 /* We did a probe so need to clean up. Call
467 * queue_work inside the spinlock so that
468 * flush_workqueue() after mddev_find will
469 * succeed in waiting for the work to be done.
471 INIT_WORK(&mddev->del_work, mddev_delayed_delete);
472 queue_work(md_misc_wq, &mddev->del_work);
476 spin_unlock(&all_mddevs_lock);
481 static void md_safemode_timeout(unsigned long data);
483 void mddev_init(struct mddev *mddev)
485 mutex_init(&mddev->open_mutex);
486 mutex_init(&mddev->reconfig_mutex);
487 mutex_init(&mddev->bitmap_info.mutex);
488 INIT_LIST_HEAD(&mddev->disks);
489 INIT_LIST_HEAD(&mddev->all_mddevs);
490 setup_timer(&mddev->safemode_timer, md_safemode_timeout,
491 (unsigned long) mddev);
492 atomic_set(&mddev->active, 1);
493 atomic_set(&mddev->openers, 0);
494 atomic_set(&mddev->active_io, 0);
495 spin_lock_init(&mddev->lock);
496 atomic_set(&mddev->flush_pending, 0);
497 init_waitqueue_head(&mddev->sb_wait);
498 init_waitqueue_head(&mddev->recovery_wait);
499 mddev->reshape_position = MaxSector;
500 mddev->reshape_backwards = 0;
501 mddev->last_sync_action = "none";
502 mddev->resync_min = 0;
503 mddev->resync_max = MaxSector;
504 mddev->level = LEVEL_NONE;
506 EXPORT_SYMBOL_GPL(mddev_init);
508 static struct mddev *mddev_find(dev_t unit)
510 struct mddev *mddev, *new = NULL;
512 if (unit && MAJOR(unit) != MD_MAJOR)
513 unit &= ~((1<<MdpMinorShift)-1);
516 spin_lock(&all_mddevs_lock);
519 list_for_each_entry(mddev, &all_mddevs, all_mddevs)
520 if (mddev->unit == unit) {
522 spin_unlock(&all_mddevs_lock);
528 list_add(&new->all_mddevs, &all_mddevs);
529 spin_unlock(&all_mddevs_lock);
530 new->hold_active = UNTIL_IOCTL;
534 /* find an unused unit number */
535 static int next_minor = 512;
536 int start = next_minor;
540 dev = MKDEV(MD_MAJOR, next_minor);
542 if (next_minor > MINORMASK)
544 if (next_minor == start) {
545 /* Oh dear, all in use. */
546 spin_unlock(&all_mddevs_lock);
552 list_for_each_entry(mddev, &all_mddevs, all_mddevs)
553 if (mddev->unit == dev) {
559 new->md_minor = MINOR(dev);
560 new->hold_active = UNTIL_STOP;
561 list_add(&new->all_mddevs, &all_mddevs);
562 spin_unlock(&all_mddevs_lock);
565 spin_unlock(&all_mddevs_lock);
567 new = kzalloc(sizeof(*new), GFP_KERNEL);
572 if (MAJOR(unit) == MD_MAJOR)
573 new->md_minor = MINOR(unit);
575 new->md_minor = MINOR(unit) >> MdpMinorShift;
582 static struct attribute_group md_redundancy_group;
584 void mddev_unlock(struct mddev *mddev)
586 if (mddev->to_remove) {
587 /* These cannot be removed under reconfig_mutex as
588 * an access to the files will try to take reconfig_mutex
589 * while holding the file unremovable, which leads to
591 * So hold set sysfs_active while the remove in happeing,
592 * and anything else which might set ->to_remove or my
593 * otherwise change the sysfs namespace will fail with
594 * -EBUSY if sysfs_active is still set.
595 * We set sysfs_active under reconfig_mutex and elsewhere
596 * test it under the same mutex to ensure its correct value
599 struct attribute_group *to_remove = mddev->to_remove;
600 mddev->to_remove = NULL;
601 mddev->sysfs_active = 1;
602 mutex_unlock(&mddev->reconfig_mutex);
604 if (mddev->kobj.sd) {
605 if (to_remove != &md_redundancy_group)
606 sysfs_remove_group(&mddev->kobj, to_remove);
607 if (mddev->pers == NULL ||
608 mddev->pers->sync_request == NULL) {
609 sysfs_remove_group(&mddev->kobj, &md_redundancy_group);
610 if (mddev->sysfs_action)
611 sysfs_put(mddev->sysfs_action);
612 mddev->sysfs_action = NULL;
615 mddev->sysfs_active = 0;
617 mutex_unlock(&mddev->reconfig_mutex);
619 /* As we've dropped the mutex we need a spinlock to
620 * make sure the thread doesn't disappear
622 spin_lock(&pers_lock);
623 md_wakeup_thread(mddev->thread);
624 spin_unlock(&pers_lock);
626 EXPORT_SYMBOL_GPL(mddev_unlock);
628 struct md_rdev *md_find_rdev_nr_rcu(struct mddev *mddev, int nr)
630 struct md_rdev *rdev;
632 rdev_for_each_rcu(rdev, mddev)
633 if (rdev->desc_nr == nr)
638 EXPORT_SYMBOL_GPL(md_find_rdev_nr_rcu);
640 static struct md_rdev *find_rdev(struct mddev *mddev, dev_t dev)
642 struct md_rdev *rdev;
644 rdev_for_each(rdev, mddev)
645 if (rdev->bdev->bd_dev == dev)
651 static struct md_rdev *find_rdev_rcu(struct mddev *mddev, dev_t dev)
653 struct md_rdev *rdev;
655 rdev_for_each_rcu(rdev, mddev)
656 if (rdev->bdev->bd_dev == dev)
662 static struct md_personality *find_pers(int level, char *clevel)
664 struct md_personality *pers;
665 list_for_each_entry(pers, &pers_list, list) {
666 if (level != LEVEL_NONE && pers->level == level)
668 if (strcmp(pers->name, clevel)==0)
674 /* return the offset of the super block in 512byte sectors */
675 static inline sector_t calc_dev_sboffset(struct md_rdev *rdev)
677 sector_t num_sectors = i_size_read(rdev->bdev->bd_inode) / 512;
678 return MD_NEW_SIZE_SECTORS(num_sectors);
681 static int alloc_disk_sb(struct md_rdev *rdev)
683 rdev->sb_page = alloc_page(GFP_KERNEL);
684 if (!rdev->sb_page) {
685 printk(KERN_ALERT "md: out of memory.\n");
692 void md_rdev_clear(struct md_rdev *rdev)
695 put_page(rdev->sb_page);
697 rdev->sb_page = NULL;
702 put_page(rdev->bb_page);
703 rdev->bb_page = NULL;
705 badblocks_exit(&rdev->badblocks);
707 EXPORT_SYMBOL_GPL(md_rdev_clear);
709 static void super_written(struct bio *bio)
711 struct md_rdev *rdev = bio->bi_private;
712 struct mddev *mddev = rdev->mddev;
715 printk("md: super_written gets error=%d\n", bio->bi_error);
716 md_error(mddev, rdev);
719 if (atomic_dec_and_test(&mddev->pending_writes))
720 wake_up(&mddev->sb_wait);
724 void md_super_write(struct mddev *mddev, struct md_rdev *rdev,
725 sector_t sector, int size, struct page *page)
727 /* write first size bytes of page to sector of rdev
728 * Increment mddev->pending_writes before returning
729 * and decrement it on completion, waking up sb_wait
730 * if zero is reached.
731 * If an error occurred, call md_error
733 struct bio *bio = bio_alloc_mddev(GFP_NOIO, 1, mddev);
735 bio->bi_bdev = rdev->meta_bdev ? rdev->meta_bdev : rdev->bdev;
736 bio->bi_iter.bi_sector = sector;
737 bio_add_page(bio, page, size, 0);
738 bio->bi_private = rdev;
739 bio->bi_end_io = super_written;
741 atomic_inc(&mddev->pending_writes);
742 submit_bio(WRITE_FLUSH_FUA, bio);
745 void md_super_wait(struct mddev *mddev)
747 /* wait for all superblock writes that were scheduled to complete */
748 wait_event(mddev->sb_wait, atomic_read(&mddev->pending_writes)==0);
751 int sync_page_io(struct md_rdev *rdev, sector_t sector, int size,
752 struct page *page, int rw, bool metadata_op)
754 struct bio *bio = bio_alloc_mddev(GFP_NOIO, 1, rdev->mddev);
757 bio->bi_bdev = (metadata_op && rdev->meta_bdev) ?
758 rdev->meta_bdev : rdev->bdev;
760 bio->bi_iter.bi_sector = sector + rdev->sb_start;
761 else if (rdev->mddev->reshape_position != MaxSector &&
762 (rdev->mddev->reshape_backwards ==
763 (sector >= rdev->mddev->reshape_position)))
764 bio->bi_iter.bi_sector = sector + rdev->new_data_offset;
766 bio->bi_iter.bi_sector = sector + rdev->data_offset;
767 bio_add_page(bio, page, size, 0);
768 submit_bio_wait(rw, bio);
770 ret = !bio->bi_error;
774 EXPORT_SYMBOL_GPL(sync_page_io);
776 static int read_disk_sb(struct md_rdev *rdev, int size)
778 char b[BDEVNAME_SIZE];
783 if (!sync_page_io(rdev, 0, size, rdev->sb_page, READ, true))
789 printk(KERN_WARNING "md: disabled device %s, could not read superblock.\n",
790 bdevname(rdev->bdev,b));
794 static int uuid_equal(mdp_super_t *sb1, mdp_super_t *sb2)
796 return sb1->set_uuid0 == sb2->set_uuid0 &&
797 sb1->set_uuid1 == sb2->set_uuid1 &&
798 sb1->set_uuid2 == sb2->set_uuid2 &&
799 sb1->set_uuid3 == sb2->set_uuid3;
802 static int sb_equal(mdp_super_t *sb1, mdp_super_t *sb2)
805 mdp_super_t *tmp1, *tmp2;
807 tmp1 = kmalloc(sizeof(*tmp1),GFP_KERNEL);
808 tmp2 = kmalloc(sizeof(*tmp2),GFP_KERNEL);
810 if (!tmp1 || !tmp2) {
812 printk(KERN_INFO "md.c sb_equal(): failed to allocate memory!\n");
820 * nr_disks is not constant
825 ret = (memcmp(tmp1, tmp2, MD_SB_GENERIC_CONSTANT_WORDS * 4) == 0);
832 static u32 md_csum_fold(u32 csum)
834 csum = (csum & 0xffff) + (csum >> 16);
835 return (csum & 0xffff) + (csum >> 16);
838 static unsigned int calc_sb_csum(mdp_super_t *sb)
841 u32 *sb32 = (u32*)sb;
843 unsigned int disk_csum, csum;
845 disk_csum = sb->sb_csum;
848 for (i = 0; i < MD_SB_BYTES/4 ; i++)
850 csum = (newcsum & 0xffffffff) + (newcsum>>32);
853 /* This used to use csum_partial, which was wrong for several
854 * reasons including that different results are returned on
855 * different architectures. It isn't critical that we get exactly
856 * the same return value as before (we always csum_fold before
857 * testing, and that removes any differences). However as we
858 * know that csum_partial always returned a 16bit value on
859 * alphas, do a fold to maximise conformity to previous behaviour.
861 sb->sb_csum = md_csum_fold(disk_csum);
863 sb->sb_csum = disk_csum;
869 * Handle superblock details.
870 * We want to be able to handle multiple superblock formats
871 * so we have a common interface to them all, and an array of
872 * different handlers.
873 * We rely on user-space to write the initial superblock, and support
874 * reading and updating of superblocks.
875 * Interface methods are:
876 * int load_super(struct md_rdev *dev, struct md_rdev *refdev, int minor_version)
877 * loads and validates a superblock on dev.
878 * if refdev != NULL, compare superblocks on both devices
880 * 0 - dev has a superblock that is compatible with refdev
881 * 1 - dev has a superblock that is compatible and newer than refdev
882 * so dev should be used as the refdev in future
883 * -EINVAL superblock incompatible or invalid
884 * -othererror e.g. -EIO
886 * int validate_super(struct mddev *mddev, struct md_rdev *dev)
887 * Verify that dev is acceptable into mddev.
888 * The first time, mddev->raid_disks will be 0, and data from
889 * dev should be merged in. Subsequent calls check that dev
890 * is new enough. Return 0 or -EINVAL
892 * void sync_super(struct mddev *mddev, struct md_rdev *dev)
893 * Update the superblock for rdev with data in mddev
894 * This does not write to disc.
900 struct module *owner;
901 int (*load_super)(struct md_rdev *rdev,
902 struct md_rdev *refdev,
904 int (*validate_super)(struct mddev *mddev,
905 struct md_rdev *rdev);
906 void (*sync_super)(struct mddev *mddev,
907 struct md_rdev *rdev);
908 unsigned long long (*rdev_size_change)(struct md_rdev *rdev,
909 sector_t num_sectors);
910 int (*allow_new_offset)(struct md_rdev *rdev,
911 unsigned long long new_offset);
915 * Check that the given mddev has no bitmap.
917 * This function is called from the run method of all personalities that do not
918 * support bitmaps. It prints an error message and returns non-zero if mddev
919 * has a bitmap. Otherwise, it returns 0.
922 int md_check_no_bitmap(struct mddev *mddev)
924 if (!mddev->bitmap_info.file && !mddev->bitmap_info.offset)
926 printk(KERN_ERR "%s: bitmaps are not supported for %s\n",
927 mdname(mddev), mddev->pers->name);
930 EXPORT_SYMBOL(md_check_no_bitmap);
933 * load_super for 0.90.0
935 static int super_90_load(struct md_rdev *rdev, struct md_rdev *refdev, int minor_version)
937 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
942 * Calculate the position of the superblock (512byte sectors),
943 * it's at the end of the disk.
945 * It also happens to be a multiple of 4Kb.
947 rdev->sb_start = calc_dev_sboffset(rdev);
949 ret = read_disk_sb(rdev, MD_SB_BYTES);
954 bdevname(rdev->bdev, b);
955 sb = page_address(rdev->sb_page);
957 if (sb->md_magic != MD_SB_MAGIC) {
958 printk(KERN_ERR "md: invalid raid superblock magic on %s\n",
963 if (sb->major_version != 0 ||
964 sb->minor_version < 90 ||
965 sb->minor_version > 91) {
966 printk(KERN_WARNING "Bad version number %d.%d on %s\n",
967 sb->major_version, sb->minor_version,
972 if (sb->raid_disks <= 0)
975 if (md_csum_fold(calc_sb_csum(sb)) != md_csum_fold(sb->sb_csum)) {
976 printk(KERN_WARNING "md: invalid superblock checksum on %s\n",
981 rdev->preferred_minor = sb->md_minor;
982 rdev->data_offset = 0;
983 rdev->new_data_offset = 0;
984 rdev->sb_size = MD_SB_BYTES;
985 rdev->badblocks.shift = -1;
987 if (sb->level == LEVEL_MULTIPATH)
990 rdev->desc_nr = sb->this_disk.number;
996 mdp_super_t *refsb = page_address(refdev->sb_page);
997 if (!uuid_equal(refsb, sb)) {
998 printk(KERN_WARNING "md: %s has different UUID to %s\n",
999 b, bdevname(refdev->bdev,b2));
1002 if (!sb_equal(refsb, sb)) {
1003 printk(KERN_WARNING "md: %s has same UUID"
1004 " but different superblock to %s\n",
1005 b, bdevname(refdev->bdev, b2));
1009 ev2 = md_event(refsb);
1015 rdev->sectors = rdev->sb_start;
1016 /* Limit to 4TB as metadata cannot record more than that.
1017 * (not needed for Linear and RAID0 as metadata doesn't
1020 if (IS_ENABLED(CONFIG_LBDAF) && (u64)rdev->sectors >= (2ULL << 32) &&
1022 rdev->sectors = (sector_t)(2ULL << 32) - 2;
1024 if (rdev->sectors < ((sector_t)sb->size) * 2 && sb->level >= 1)
1025 /* "this cannot possibly happen" ... */
1033 * validate_super for 0.90.0
1035 static int super_90_validate(struct mddev *mddev, struct md_rdev *rdev)
1038 mdp_super_t *sb = page_address(rdev->sb_page);
1039 __u64 ev1 = md_event(sb);
1041 rdev->raid_disk = -1;
1042 clear_bit(Faulty, &rdev->flags);
1043 clear_bit(In_sync, &rdev->flags);
1044 clear_bit(Bitmap_sync, &rdev->flags);
1045 clear_bit(WriteMostly, &rdev->flags);
1047 if (mddev->raid_disks == 0) {
1048 mddev->major_version = 0;
1049 mddev->minor_version = sb->minor_version;
1050 mddev->patch_version = sb->patch_version;
1051 mddev->external = 0;
1052 mddev->chunk_sectors = sb->chunk_size >> 9;
1053 mddev->ctime = sb->ctime;
1054 mddev->utime = sb->utime;
1055 mddev->level = sb->level;
1056 mddev->clevel[0] = 0;
1057 mddev->layout = sb->layout;
1058 mddev->raid_disks = sb->raid_disks;
1059 mddev->dev_sectors = ((sector_t)sb->size) * 2;
1060 mddev->events = ev1;
1061 mddev->bitmap_info.offset = 0;
1062 mddev->bitmap_info.space = 0;
1063 /* bitmap can use 60 K after the 4K superblocks */
1064 mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9;
1065 mddev->bitmap_info.default_space = 64*2 - (MD_SB_BYTES >> 9);
1066 mddev->reshape_backwards = 0;
1068 if (mddev->minor_version >= 91) {
1069 mddev->reshape_position = sb->reshape_position;
1070 mddev->delta_disks = sb->delta_disks;
1071 mddev->new_level = sb->new_level;
1072 mddev->new_layout = sb->new_layout;
1073 mddev->new_chunk_sectors = sb->new_chunk >> 9;
1074 if (mddev->delta_disks < 0)
1075 mddev->reshape_backwards = 1;
1077 mddev->reshape_position = MaxSector;
1078 mddev->delta_disks = 0;
1079 mddev->new_level = mddev->level;
1080 mddev->new_layout = mddev->layout;
1081 mddev->new_chunk_sectors = mddev->chunk_sectors;
1084 if (sb->state & (1<<MD_SB_CLEAN))
1085 mddev->recovery_cp = MaxSector;
1087 if (sb->events_hi == sb->cp_events_hi &&
1088 sb->events_lo == sb->cp_events_lo) {
1089 mddev->recovery_cp = sb->recovery_cp;
1091 mddev->recovery_cp = 0;
1094 memcpy(mddev->uuid+0, &sb->set_uuid0, 4);
1095 memcpy(mddev->uuid+4, &sb->set_uuid1, 4);
1096 memcpy(mddev->uuid+8, &sb->set_uuid2, 4);
1097 memcpy(mddev->uuid+12,&sb->set_uuid3, 4);
1099 mddev->max_disks = MD_SB_DISKS;
1101 if (sb->state & (1<<MD_SB_BITMAP_PRESENT) &&
1102 mddev->bitmap_info.file == NULL) {
1103 mddev->bitmap_info.offset =
1104 mddev->bitmap_info.default_offset;
1105 mddev->bitmap_info.space =
1106 mddev->bitmap_info.default_space;
1109 } else if (mddev->pers == NULL) {
1110 /* Insist on good event counter while assembling, except
1111 * for spares (which don't need an event count) */
1113 if (sb->disks[rdev->desc_nr].state & (
1114 (1<<MD_DISK_SYNC) | (1 << MD_DISK_ACTIVE)))
1115 if (ev1 < mddev->events)
1117 } else if (mddev->bitmap) {
1118 /* if adding to array with a bitmap, then we can accept an
1119 * older device ... but not too old.
1121 if (ev1 < mddev->bitmap->events_cleared)
1123 if (ev1 < mddev->events)
1124 set_bit(Bitmap_sync, &rdev->flags);
1126 if (ev1 < mddev->events)
1127 /* just a hot-add of a new device, leave raid_disk at -1 */
1131 if (mddev->level != LEVEL_MULTIPATH) {
1132 desc = sb->disks + rdev->desc_nr;
1134 if (desc->state & (1<<MD_DISK_FAULTY))
1135 set_bit(Faulty, &rdev->flags);
1136 else if (desc->state & (1<<MD_DISK_SYNC) /* &&
1137 desc->raid_disk < mddev->raid_disks */) {
1138 set_bit(In_sync, &rdev->flags);
1139 rdev->raid_disk = desc->raid_disk;
1140 rdev->saved_raid_disk = desc->raid_disk;
1141 } else if (desc->state & (1<<MD_DISK_ACTIVE)) {
1142 /* active but not in sync implies recovery up to
1143 * reshape position. We don't know exactly where
1144 * that is, so set to zero for now */
1145 if (mddev->minor_version >= 91) {
1146 rdev->recovery_offset = 0;
1147 rdev->raid_disk = desc->raid_disk;
1150 if (desc->state & (1<<MD_DISK_WRITEMOSTLY))
1151 set_bit(WriteMostly, &rdev->flags);
1152 } else /* MULTIPATH are always insync */
1153 set_bit(In_sync, &rdev->flags);
1158 * sync_super for 0.90.0
1160 static void super_90_sync(struct mddev *mddev, struct md_rdev *rdev)
1163 struct md_rdev *rdev2;
1164 int next_spare = mddev->raid_disks;
1166 /* make rdev->sb match mddev data..
1169 * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
1170 * 3/ any empty disks < next_spare become removed
1172 * disks[0] gets initialised to REMOVED because
1173 * we cannot be sure from other fields if it has
1174 * been initialised or not.
1177 int active=0, working=0,failed=0,spare=0,nr_disks=0;
1179 rdev->sb_size = MD_SB_BYTES;
1181 sb = page_address(rdev->sb_page);
1183 memset(sb, 0, sizeof(*sb));
1185 sb->md_magic = MD_SB_MAGIC;
1186 sb->major_version = mddev->major_version;
1187 sb->patch_version = mddev->patch_version;
1188 sb->gvalid_words = 0; /* ignored */
1189 memcpy(&sb->set_uuid0, mddev->uuid+0, 4);
1190 memcpy(&sb->set_uuid1, mddev->uuid+4, 4);
1191 memcpy(&sb->set_uuid2, mddev->uuid+8, 4);
1192 memcpy(&sb->set_uuid3, mddev->uuid+12,4);
1194 sb->ctime = clamp_t(time64_t, mddev->ctime, 0, U32_MAX);
1195 sb->level = mddev->level;
1196 sb->size = mddev->dev_sectors / 2;
1197 sb->raid_disks = mddev->raid_disks;
1198 sb->md_minor = mddev->md_minor;
1199 sb->not_persistent = 0;
1200 sb->utime = clamp_t(time64_t, mddev->utime, 0, U32_MAX);
1202 sb->events_hi = (mddev->events>>32);
1203 sb->events_lo = (u32)mddev->events;
1205 if (mddev->reshape_position == MaxSector)
1206 sb->minor_version = 90;
1208 sb->minor_version = 91;
1209 sb->reshape_position = mddev->reshape_position;
1210 sb->new_level = mddev->new_level;
1211 sb->delta_disks = mddev->delta_disks;
1212 sb->new_layout = mddev->new_layout;
1213 sb->new_chunk = mddev->new_chunk_sectors << 9;
1215 mddev->minor_version = sb->minor_version;
1218 sb->recovery_cp = mddev->recovery_cp;
1219 sb->cp_events_hi = (mddev->events>>32);
1220 sb->cp_events_lo = (u32)mddev->events;
1221 if (mddev->recovery_cp == MaxSector)
1222 sb->state = (1<< MD_SB_CLEAN);
1224 sb->recovery_cp = 0;
1226 sb->layout = mddev->layout;
1227 sb->chunk_size = mddev->chunk_sectors << 9;
1229 if (mddev->bitmap && mddev->bitmap_info.file == NULL)
1230 sb->state |= (1<<MD_SB_BITMAP_PRESENT);
1232 sb->disks[0].state = (1<<MD_DISK_REMOVED);
1233 rdev_for_each(rdev2, mddev) {
1236 int is_active = test_bit(In_sync, &rdev2->flags);
1238 if (rdev2->raid_disk >= 0 &&
1239 sb->minor_version >= 91)
1240 /* we have nowhere to store the recovery_offset,
1241 * but if it is not below the reshape_position,
1242 * we can piggy-back on that.
1245 if (rdev2->raid_disk < 0 ||
1246 test_bit(Faulty, &rdev2->flags))
1249 desc_nr = rdev2->raid_disk;
1251 desc_nr = next_spare++;
1252 rdev2->desc_nr = desc_nr;
1253 d = &sb->disks[rdev2->desc_nr];
1255 d->number = rdev2->desc_nr;
1256 d->major = MAJOR(rdev2->bdev->bd_dev);
1257 d->minor = MINOR(rdev2->bdev->bd_dev);
1259 d->raid_disk = rdev2->raid_disk;
1261 d->raid_disk = rdev2->desc_nr; /* compatibility */
1262 if (test_bit(Faulty, &rdev2->flags))
1263 d->state = (1<<MD_DISK_FAULTY);
1264 else if (is_active) {
1265 d->state = (1<<MD_DISK_ACTIVE);
1266 if (test_bit(In_sync, &rdev2->flags))
1267 d->state |= (1<<MD_DISK_SYNC);
1275 if (test_bit(WriteMostly, &rdev2->flags))
1276 d->state |= (1<<MD_DISK_WRITEMOSTLY);
1278 /* now set the "removed" and "faulty" bits on any missing devices */
1279 for (i=0 ; i < mddev->raid_disks ; i++) {
1280 mdp_disk_t *d = &sb->disks[i];
1281 if (d->state == 0 && d->number == 0) {
1284 d->state = (1<<MD_DISK_REMOVED);
1285 d->state |= (1<<MD_DISK_FAULTY);
1289 sb->nr_disks = nr_disks;
1290 sb->active_disks = active;
1291 sb->working_disks = working;
1292 sb->failed_disks = failed;
1293 sb->spare_disks = spare;
1295 sb->this_disk = sb->disks[rdev->desc_nr];
1296 sb->sb_csum = calc_sb_csum(sb);
1300 * rdev_size_change for 0.90.0
1302 static unsigned long long
1303 super_90_rdev_size_change(struct md_rdev *rdev, sector_t num_sectors)
1305 if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
1306 return 0; /* component must fit device */
1307 if (rdev->mddev->bitmap_info.offset)
1308 return 0; /* can't move bitmap */
1309 rdev->sb_start = calc_dev_sboffset(rdev);
1310 if (!num_sectors || num_sectors > rdev->sb_start)
1311 num_sectors = rdev->sb_start;
1312 /* Limit to 4TB as metadata cannot record more than that.
1313 * 4TB == 2^32 KB, or 2*2^32 sectors.
1315 if (IS_ENABLED(CONFIG_LBDAF) && (u64)num_sectors >= (2ULL << 32) &&
1316 rdev->mddev->level >= 1)
1317 num_sectors = (sector_t)(2ULL << 32) - 2;
1318 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
1320 md_super_wait(rdev->mddev);
1325 super_90_allow_new_offset(struct md_rdev *rdev, unsigned long long new_offset)
1327 /* non-zero offset changes not possible with v0.90 */
1328 return new_offset == 0;
1332 * version 1 superblock
1335 static __le32 calc_sb_1_csum(struct mdp_superblock_1 *sb)
1339 unsigned long long newcsum;
1340 int size = 256 + le32_to_cpu(sb->max_dev)*2;
1341 __le32 *isuper = (__le32*)sb;
1343 disk_csum = sb->sb_csum;
1346 for (; size >= 4; size -= 4)
1347 newcsum += le32_to_cpu(*isuper++);
1350 newcsum += le16_to_cpu(*(__le16*) isuper);
1352 csum = (newcsum & 0xffffffff) + (newcsum >> 32);
1353 sb->sb_csum = disk_csum;
1354 return cpu_to_le32(csum);
1357 static int super_1_load(struct md_rdev *rdev, struct md_rdev *refdev, int minor_version)
1359 struct mdp_superblock_1 *sb;
1363 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
1367 * Calculate the position of the superblock in 512byte sectors.
1368 * It is always aligned to a 4K boundary and
1369 * depeding on minor_version, it can be:
1370 * 0: At least 8K, but less than 12K, from end of device
1371 * 1: At start of device
1372 * 2: 4K from start of device.
1374 switch(minor_version) {
1376 sb_start = i_size_read(rdev->bdev->bd_inode) >> 9;
1378 sb_start &= ~(sector_t)(4*2-1);
1389 rdev->sb_start = sb_start;
1391 /* superblock is rarely larger than 1K, but it can be larger,
1392 * and it is safe to read 4k, so we do that
1394 ret = read_disk_sb(rdev, 4096);
1395 if (ret) return ret;
1397 sb = page_address(rdev->sb_page);
1399 if (sb->magic != cpu_to_le32(MD_SB_MAGIC) ||
1400 sb->major_version != cpu_to_le32(1) ||
1401 le32_to_cpu(sb->max_dev) > (4096-256)/2 ||
1402 le64_to_cpu(sb->super_offset) != rdev->sb_start ||
1403 (le32_to_cpu(sb->feature_map) & ~MD_FEATURE_ALL) != 0)
1406 if (calc_sb_1_csum(sb) != sb->sb_csum) {
1407 printk("md: invalid superblock checksum on %s\n",
1408 bdevname(rdev->bdev,b));
1411 if (le64_to_cpu(sb->data_size) < 10) {
1412 printk("md: data_size too small on %s\n",
1413 bdevname(rdev->bdev,b));
1418 memcmp(sb->pad3, sb->pad3+1, sizeof(sb->pad3) - sizeof(sb->pad3[1])))
1419 /* Some padding is non-zero, might be a new feature */
1422 rdev->preferred_minor = 0xffff;
1423 rdev->data_offset = le64_to_cpu(sb->data_offset);
1424 rdev->new_data_offset = rdev->data_offset;
1425 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE) &&
1426 (le32_to_cpu(sb->feature_map) & MD_FEATURE_NEW_OFFSET))
1427 rdev->new_data_offset += (s32)le32_to_cpu(sb->new_offset);
1428 atomic_set(&rdev->corrected_errors, le32_to_cpu(sb->cnt_corrected_read));
1430 rdev->sb_size = le32_to_cpu(sb->max_dev) * 2 + 256;
1431 bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
1432 if (rdev->sb_size & bmask)
1433 rdev->sb_size = (rdev->sb_size | bmask) + 1;
1436 && rdev->data_offset < sb_start + (rdev->sb_size/512))
1439 && rdev->new_data_offset < sb_start + (rdev->sb_size/512))
1442 if (sb->level == cpu_to_le32(LEVEL_MULTIPATH))
1445 rdev->desc_nr = le32_to_cpu(sb->dev_number);
1447 if (!rdev->bb_page) {
1448 rdev->bb_page = alloc_page(GFP_KERNEL);
1452 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BAD_BLOCKS) &&
1453 rdev->badblocks.count == 0) {
1454 /* need to load the bad block list.
1455 * Currently we limit it to one page.
1461 int sectors = le16_to_cpu(sb->bblog_size);
1462 if (sectors > (PAGE_SIZE / 512))
1464 offset = le32_to_cpu(sb->bblog_offset);
1467 bb_sector = (long long)offset;
1468 if (!sync_page_io(rdev, bb_sector, sectors << 9,
1469 rdev->bb_page, READ, true))
1471 bbp = (u64 *)page_address(rdev->bb_page);
1472 rdev->badblocks.shift = sb->bblog_shift;
1473 for (i = 0 ; i < (sectors << (9-3)) ; i++, bbp++) {
1474 u64 bb = le64_to_cpu(*bbp);
1475 int count = bb & (0x3ff);
1476 u64 sector = bb >> 10;
1477 sector <<= sb->bblog_shift;
1478 count <<= sb->bblog_shift;
1481 if (badblocks_set(&rdev->badblocks, sector, count, 1))
1484 } else if (sb->bblog_offset != 0)
1485 rdev->badblocks.shift = 0;
1491 struct mdp_superblock_1 *refsb = page_address(refdev->sb_page);
1493 if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 ||
1494 sb->level != refsb->level ||
1495 sb->layout != refsb->layout ||
1496 sb->chunksize != refsb->chunksize) {
1497 printk(KERN_WARNING "md: %s has strangely different"
1498 " superblock to %s\n",
1499 bdevname(rdev->bdev,b),
1500 bdevname(refdev->bdev,b2));
1503 ev1 = le64_to_cpu(sb->events);
1504 ev2 = le64_to_cpu(refsb->events);
1511 if (minor_version) {
1512 sectors = (i_size_read(rdev->bdev->bd_inode) >> 9);
1513 sectors -= rdev->data_offset;
1515 sectors = rdev->sb_start;
1516 if (sectors < le64_to_cpu(sb->data_size))
1518 rdev->sectors = le64_to_cpu(sb->data_size);
1522 static int super_1_validate(struct mddev *mddev, struct md_rdev *rdev)
1524 struct mdp_superblock_1 *sb = page_address(rdev->sb_page);
1525 __u64 ev1 = le64_to_cpu(sb->events);
1527 rdev->raid_disk = -1;
1528 clear_bit(Faulty, &rdev->flags);
1529 clear_bit(In_sync, &rdev->flags);
1530 clear_bit(Bitmap_sync, &rdev->flags);
1531 clear_bit(WriteMostly, &rdev->flags);
1533 if (mddev->raid_disks == 0) {
1534 mddev->major_version = 1;
1535 mddev->patch_version = 0;
1536 mddev->external = 0;
1537 mddev->chunk_sectors = le32_to_cpu(sb->chunksize);
1538 mddev->ctime = le64_to_cpu(sb->ctime);
1539 mddev->utime = le64_to_cpu(sb->utime);
1540 mddev->level = le32_to_cpu(sb->level);
1541 mddev->clevel[0] = 0;
1542 mddev->layout = le32_to_cpu(sb->layout);
1543 mddev->raid_disks = le32_to_cpu(sb->raid_disks);
1544 mddev->dev_sectors = le64_to_cpu(sb->size);
1545 mddev->events = ev1;
1546 mddev->bitmap_info.offset = 0;
1547 mddev->bitmap_info.space = 0;
1548 /* Default location for bitmap is 1K after superblock
1549 * using 3K - total of 4K
1551 mddev->bitmap_info.default_offset = 1024 >> 9;
1552 mddev->bitmap_info.default_space = (4096-1024) >> 9;
1553 mddev->reshape_backwards = 0;
1555 mddev->recovery_cp = le64_to_cpu(sb->resync_offset);
1556 memcpy(mddev->uuid, sb->set_uuid, 16);
1558 mddev->max_disks = (4096-256)/2;
1560 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) &&
1561 mddev->bitmap_info.file == NULL) {
1562 mddev->bitmap_info.offset =
1563 (__s32)le32_to_cpu(sb->bitmap_offset);
1564 /* Metadata doesn't record how much space is available.
1565 * For 1.0, we assume we can use up to the superblock
1566 * if before, else to 4K beyond superblock.
1567 * For others, assume no change is possible.
1569 if (mddev->minor_version > 0)
1570 mddev->bitmap_info.space = 0;
1571 else if (mddev->bitmap_info.offset > 0)
1572 mddev->bitmap_info.space =
1573 8 - mddev->bitmap_info.offset;
1575 mddev->bitmap_info.space =
1576 -mddev->bitmap_info.offset;
1579 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) {
1580 mddev->reshape_position = le64_to_cpu(sb->reshape_position);
1581 mddev->delta_disks = le32_to_cpu(sb->delta_disks);
1582 mddev->new_level = le32_to_cpu(sb->new_level);
1583 mddev->new_layout = le32_to_cpu(sb->new_layout);
1584 mddev->new_chunk_sectors = le32_to_cpu(sb->new_chunk);
1585 if (mddev->delta_disks < 0 ||
1586 (mddev->delta_disks == 0 &&
1587 (le32_to_cpu(sb->feature_map)
1588 & MD_FEATURE_RESHAPE_BACKWARDS)))
1589 mddev->reshape_backwards = 1;
1591 mddev->reshape_position = MaxSector;
1592 mddev->delta_disks = 0;
1593 mddev->new_level = mddev->level;
1594 mddev->new_layout = mddev->layout;
1595 mddev->new_chunk_sectors = mddev->chunk_sectors;
1598 if (le32_to_cpu(sb->feature_map) & MD_FEATURE_JOURNAL) {
1599 set_bit(MD_HAS_JOURNAL, &mddev->flags);
1600 if (mddev->recovery_cp == MaxSector)
1601 set_bit(MD_JOURNAL_CLEAN, &mddev->flags);
1603 } else if (mddev->pers == NULL) {
1604 /* Insist of good event counter while assembling, except for
1605 * spares (which don't need an event count) */
1607 if (rdev->desc_nr >= 0 &&
1608 rdev->desc_nr < le32_to_cpu(sb->max_dev) &&
1609 (le16_to_cpu(sb->dev_roles[rdev->desc_nr]) < MD_DISK_ROLE_MAX ||
1610 le16_to_cpu(sb->dev_roles[rdev->desc_nr]) == MD_DISK_ROLE_JOURNAL))
1611 if (ev1 < mddev->events)
1613 } else if (mddev->bitmap) {
1614 /* If adding to array with a bitmap, then we can accept an
1615 * older device, but not too old.
1617 if (ev1 < mddev->bitmap->events_cleared)
1619 if (ev1 < mddev->events)
1620 set_bit(Bitmap_sync, &rdev->flags);
1622 if (ev1 < mddev->events)
1623 /* just a hot-add of a new device, leave raid_disk at -1 */
1626 if (mddev->level != LEVEL_MULTIPATH) {
1628 if (rdev->desc_nr < 0 ||
1629 rdev->desc_nr >= le32_to_cpu(sb->max_dev)) {
1630 role = MD_DISK_ROLE_SPARE;
1633 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
1635 case MD_DISK_ROLE_SPARE: /* spare */
1637 case MD_DISK_ROLE_FAULTY: /* faulty */
1638 set_bit(Faulty, &rdev->flags);
1640 case MD_DISK_ROLE_JOURNAL: /* journal device */
1641 if (!(le32_to_cpu(sb->feature_map) & MD_FEATURE_JOURNAL)) {
1642 /* journal device without journal feature */
1644 "md: journal device provided without journal feature, ignoring the device\n");
1647 set_bit(Journal, &rdev->flags);
1648 rdev->journal_tail = le64_to_cpu(sb->journal_tail);
1649 rdev->raid_disk = 0;
1652 rdev->saved_raid_disk = role;
1653 if ((le32_to_cpu(sb->feature_map) &
1654 MD_FEATURE_RECOVERY_OFFSET)) {
1655 rdev->recovery_offset = le64_to_cpu(sb->recovery_offset);
1656 if (!(le32_to_cpu(sb->feature_map) &
1657 MD_FEATURE_RECOVERY_BITMAP))
1658 rdev->saved_raid_disk = -1;
1660 set_bit(In_sync, &rdev->flags);
1661 rdev->raid_disk = role;
1664 if (sb->devflags & WriteMostly1)
1665 set_bit(WriteMostly, &rdev->flags);
1666 if (le32_to_cpu(sb->feature_map) & MD_FEATURE_REPLACEMENT)
1667 set_bit(Replacement, &rdev->flags);
1668 } else /* MULTIPATH are always insync */
1669 set_bit(In_sync, &rdev->flags);
1674 static void super_1_sync(struct mddev *mddev, struct md_rdev *rdev)
1676 struct mdp_superblock_1 *sb;
1677 struct md_rdev *rdev2;
1679 /* make rdev->sb match mddev and rdev data. */
1681 sb = page_address(rdev->sb_page);
1683 sb->feature_map = 0;
1685 sb->recovery_offset = cpu_to_le64(0);
1686 memset(sb->pad3, 0, sizeof(sb->pad3));
1688 sb->utime = cpu_to_le64((__u64)mddev->utime);
1689 sb->events = cpu_to_le64(mddev->events);
1691 sb->resync_offset = cpu_to_le64(mddev->recovery_cp);
1692 else if (test_bit(MD_JOURNAL_CLEAN, &mddev->flags))
1693 sb->resync_offset = cpu_to_le64(MaxSector);
1695 sb->resync_offset = cpu_to_le64(0);
1697 sb->cnt_corrected_read = cpu_to_le32(atomic_read(&rdev->corrected_errors));
1699 sb->raid_disks = cpu_to_le32(mddev->raid_disks);
1700 sb->size = cpu_to_le64(mddev->dev_sectors);
1701 sb->chunksize = cpu_to_le32(mddev->chunk_sectors);
1702 sb->level = cpu_to_le32(mddev->level);
1703 sb->layout = cpu_to_le32(mddev->layout);
1705 if (test_bit(WriteMostly, &rdev->flags))
1706 sb->devflags |= WriteMostly1;
1708 sb->devflags &= ~WriteMostly1;
1709 sb->data_offset = cpu_to_le64(rdev->data_offset);
1710 sb->data_size = cpu_to_le64(rdev->sectors);
1712 if (mddev->bitmap && mddev->bitmap_info.file == NULL) {
1713 sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_info.offset);
1714 sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET);
1717 if (rdev->raid_disk >= 0 && !test_bit(Journal, &rdev->flags) &&
1718 !test_bit(In_sync, &rdev->flags)) {
1720 cpu_to_le32(MD_FEATURE_RECOVERY_OFFSET);
1721 sb->recovery_offset =
1722 cpu_to_le64(rdev->recovery_offset);
1723 if (rdev->saved_raid_disk >= 0 && mddev->bitmap)
1725 cpu_to_le32(MD_FEATURE_RECOVERY_BITMAP);
1727 /* Note: recovery_offset and journal_tail share space */
1728 if (test_bit(Journal, &rdev->flags))
1729 sb->journal_tail = cpu_to_le64(rdev->journal_tail);
1730 if (test_bit(Replacement, &rdev->flags))
1732 cpu_to_le32(MD_FEATURE_REPLACEMENT);
1734 if (mddev->reshape_position != MaxSector) {
1735 sb->feature_map |= cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE);
1736 sb->reshape_position = cpu_to_le64(mddev->reshape_position);
1737 sb->new_layout = cpu_to_le32(mddev->new_layout);
1738 sb->delta_disks = cpu_to_le32(mddev->delta_disks);
1739 sb->new_level = cpu_to_le32(mddev->new_level);
1740 sb->new_chunk = cpu_to_le32(mddev->new_chunk_sectors);
1741 if (mddev->delta_disks == 0 &&
1742 mddev->reshape_backwards)
1744 |= cpu_to_le32(MD_FEATURE_RESHAPE_BACKWARDS);
1745 if (rdev->new_data_offset != rdev->data_offset) {
1747 |= cpu_to_le32(MD_FEATURE_NEW_OFFSET);
1748 sb->new_offset = cpu_to_le32((__u32)(rdev->new_data_offset
1749 - rdev->data_offset));
1753 if (mddev_is_clustered(mddev))
1754 sb->feature_map |= cpu_to_le32(MD_FEATURE_CLUSTERED);
1756 if (rdev->badblocks.count == 0)
1757 /* Nothing to do for bad blocks*/ ;
1758 else if (sb->bblog_offset == 0)
1759 /* Cannot record bad blocks on this device */
1760 md_error(mddev, rdev);
1762 struct badblocks *bb = &rdev->badblocks;
1763 u64 *bbp = (u64 *)page_address(rdev->bb_page);
1765 sb->feature_map |= cpu_to_le32(MD_FEATURE_BAD_BLOCKS);
1770 seq = read_seqbegin(&bb->lock);
1772 memset(bbp, 0xff, PAGE_SIZE);
1774 for (i = 0 ; i < bb->count ; i++) {
1775 u64 internal_bb = p[i];
1776 u64 store_bb = ((BB_OFFSET(internal_bb) << 10)
1777 | BB_LEN(internal_bb));
1778 bbp[i] = cpu_to_le64(store_bb);
1781 if (read_seqretry(&bb->lock, seq))
1784 bb->sector = (rdev->sb_start +
1785 (int)le32_to_cpu(sb->bblog_offset));
1786 bb->size = le16_to_cpu(sb->bblog_size);
1791 rdev_for_each(rdev2, mddev)
1792 if (rdev2->desc_nr+1 > max_dev)
1793 max_dev = rdev2->desc_nr+1;
1795 if (max_dev > le32_to_cpu(sb->max_dev)) {
1797 sb->max_dev = cpu_to_le32(max_dev);
1798 rdev->sb_size = max_dev * 2 + 256;
1799 bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
1800 if (rdev->sb_size & bmask)
1801 rdev->sb_size = (rdev->sb_size | bmask) + 1;
1803 max_dev = le32_to_cpu(sb->max_dev);
1805 for (i=0; i<max_dev;i++)
1806 sb->dev_roles[i] = cpu_to_le16(MD_DISK_ROLE_FAULTY);
1808 if (test_bit(MD_HAS_JOURNAL, &mddev->flags))
1809 sb->feature_map |= cpu_to_le32(MD_FEATURE_JOURNAL);
1811 rdev_for_each(rdev2, mddev) {
1813 if (test_bit(Faulty, &rdev2->flags))
1814 sb->dev_roles[i] = cpu_to_le16(MD_DISK_ROLE_FAULTY);
1815 else if (test_bit(In_sync, &rdev2->flags))
1816 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1817 else if (test_bit(Journal, &rdev2->flags))
1818 sb->dev_roles[i] = cpu_to_le16(MD_DISK_ROLE_JOURNAL);
1819 else if (rdev2->raid_disk >= 0)
1820 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1822 sb->dev_roles[i] = cpu_to_le16(MD_DISK_ROLE_SPARE);
1825 sb->sb_csum = calc_sb_1_csum(sb);
1828 static unsigned long long
1829 super_1_rdev_size_change(struct md_rdev *rdev, sector_t num_sectors)
1831 struct mdp_superblock_1 *sb;
1832 sector_t max_sectors;
1833 if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
1834 return 0; /* component must fit device */
1835 if (rdev->data_offset != rdev->new_data_offset)
1836 return 0; /* too confusing */
1837 if (rdev->sb_start < rdev->data_offset) {
1838 /* minor versions 1 and 2; superblock before data */
1839 max_sectors = i_size_read(rdev->bdev->bd_inode) >> 9;
1840 max_sectors -= rdev->data_offset;
1841 if (!num_sectors || num_sectors > max_sectors)
1842 num_sectors = max_sectors;
1843 } else if (rdev->mddev->bitmap_info.offset) {
1844 /* minor version 0 with bitmap we can't move */
1847 /* minor version 0; superblock after data */
1849 sb_start = (i_size_read(rdev->bdev->bd_inode) >> 9) - 8*2;
1850 sb_start &= ~(sector_t)(4*2 - 1);
1851 max_sectors = rdev->sectors + sb_start - rdev->sb_start;
1852 if (!num_sectors || num_sectors > max_sectors)
1853 num_sectors = max_sectors;
1854 rdev->sb_start = sb_start;
1856 sb = page_address(rdev->sb_page);
1857 sb->data_size = cpu_to_le64(num_sectors);
1858 sb->super_offset = rdev->sb_start;
1859 sb->sb_csum = calc_sb_1_csum(sb);
1860 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
1862 md_super_wait(rdev->mddev);
1868 super_1_allow_new_offset(struct md_rdev *rdev,
1869 unsigned long long new_offset)
1871 /* All necessary checks on new >= old have been done */
1872 struct bitmap *bitmap;
1873 if (new_offset >= rdev->data_offset)
1876 /* with 1.0 metadata, there is no metadata to tread on
1877 * so we can always move back */
1878 if (rdev->mddev->minor_version == 0)
1881 /* otherwise we must be sure not to step on
1882 * any metadata, so stay:
1883 * 36K beyond start of superblock
1884 * beyond end of badblocks
1885 * beyond write-intent bitmap
1887 if (rdev->sb_start + (32+4)*2 > new_offset)
1889 bitmap = rdev->mddev->bitmap;
1890 if (bitmap && !rdev->mddev->bitmap_info.file &&
1891 rdev->sb_start + rdev->mddev->bitmap_info.offset +
1892 bitmap->storage.file_pages * (PAGE_SIZE>>9) > new_offset)
1894 if (rdev->badblocks.sector + rdev->badblocks.size > new_offset)
1900 static struct super_type super_types[] = {
1903 .owner = THIS_MODULE,
1904 .load_super = super_90_load,
1905 .validate_super = super_90_validate,
1906 .sync_super = super_90_sync,
1907 .rdev_size_change = super_90_rdev_size_change,
1908 .allow_new_offset = super_90_allow_new_offset,
1912 .owner = THIS_MODULE,
1913 .load_super = super_1_load,
1914 .validate_super = super_1_validate,
1915 .sync_super = super_1_sync,
1916 .rdev_size_change = super_1_rdev_size_change,
1917 .allow_new_offset = super_1_allow_new_offset,
1921 static void sync_super(struct mddev *mddev, struct md_rdev *rdev)
1923 if (mddev->sync_super) {
1924 mddev->sync_super(mddev, rdev);
1928 BUG_ON(mddev->major_version >= ARRAY_SIZE(super_types));
1930 super_types[mddev->major_version].sync_super(mddev, rdev);
1933 static int match_mddev_units(struct mddev *mddev1, struct mddev *mddev2)
1935 struct md_rdev *rdev, *rdev2;
1938 rdev_for_each_rcu(rdev, mddev1) {
1939 if (test_bit(Faulty, &rdev->flags) ||
1940 test_bit(Journal, &rdev->flags) ||
1941 rdev->raid_disk == -1)
1943 rdev_for_each_rcu(rdev2, mddev2) {
1944 if (test_bit(Faulty, &rdev2->flags) ||
1945 test_bit(Journal, &rdev2->flags) ||
1946 rdev2->raid_disk == -1)
1948 if (rdev->bdev->bd_contains ==
1949 rdev2->bdev->bd_contains) {
1959 static LIST_HEAD(pending_raid_disks);
1962 * Try to register data integrity profile for an mddev
1964 * This is called when an array is started and after a disk has been kicked
1965 * from the array. It only succeeds if all working and active component devices
1966 * are integrity capable with matching profiles.
1968 int md_integrity_register(struct mddev *mddev)
1970 struct md_rdev *rdev, *reference = NULL;
1972 if (list_empty(&mddev->disks))
1973 return 0; /* nothing to do */
1974 if (!mddev->gendisk || blk_get_integrity(mddev->gendisk))
1975 return 0; /* shouldn't register, or already is */
1976 rdev_for_each(rdev, mddev) {
1977 /* skip spares and non-functional disks */
1978 if (test_bit(Faulty, &rdev->flags))
1980 if (rdev->raid_disk < 0)
1983 /* Use the first rdev as the reference */
1987 /* does this rdev's profile match the reference profile? */
1988 if (blk_integrity_compare(reference->bdev->bd_disk,
1989 rdev->bdev->bd_disk) < 0)
1992 if (!reference || !bdev_get_integrity(reference->bdev))
1995 * All component devices are integrity capable and have matching
1996 * profiles, register the common profile for the md device.
1998 blk_integrity_register(mddev->gendisk,
1999 bdev_get_integrity(reference->bdev));
2001 printk(KERN_NOTICE "md: data integrity enabled on %s\n", mdname(mddev));
2002 if (bioset_integrity_create(mddev->bio_set, BIO_POOL_SIZE)) {
2003 printk(KERN_ERR "md: failed to create integrity pool for %s\n",
2009 EXPORT_SYMBOL(md_integrity_register);
2012 * Attempt to add an rdev, but only if it is consistent with the current
2015 int md_integrity_add_rdev(struct md_rdev *rdev, struct mddev *mddev)
2017 struct blk_integrity *bi_rdev;
2018 struct blk_integrity *bi_mddev;
2019 char name[BDEVNAME_SIZE];
2021 if (!mddev->gendisk)
2024 bi_rdev = bdev_get_integrity(rdev->bdev);
2025 bi_mddev = blk_get_integrity(mddev->gendisk);
2027 if (!bi_mddev) /* nothing to do */
2030 if (blk_integrity_compare(mddev->gendisk, rdev->bdev->bd_disk) != 0) {
2031 printk(KERN_NOTICE "%s: incompatible integrity profile for %s\n",
2032 mdname(mddev), bdevname(rdev->bdev, name));
2038 EXPORT_SYMBOL(md_integrity_add_rdev);
2040 static int bind_rdev_to_array(struct md_rdev *rdev, struct mddev *mddev)
2042 char b[BDEVNAME_SIZE];
2046 /* prevent duplicates */
2047 if (find_rdev(mddev, rdev->bdev->bd_dev))
2050 /* make sure rdev->sectors exceeds mddev->dev_sectors */
2051 if (!test_bit(Journal, &rdev->flags) &&
2053 (mddev->dev_sectors == 0 || rdev->sectors < mddev->dev_sectors)) {
2055 /* Cannot change size, so fail
2056 * If mddev->level <= 0, then we don't care
2057 * about aligning sizes (e.g. linear)
2059 if (mddev->level > 0)
2062 mddev->dev_sectors = rdev->sectors;
2065 /* Verify rdev->desc_nr is unique.
2066 * If it is -1, assign a free number, else
2067 * check number is not in use
2070 if (rdev->desc_nr < 0) {
2073 choice = mddev->raid_disks;
2074 while (md_find_rdev_nr_rcu(mddev, choice))
2076 rdev->desc_nr = choice;
2078 if (md_find_rdev_nr_rcu(mddev, rdev->desc_nr)) {
2084 if (!test_bit(Journal, &rdev->flags) &&
2085 mddev->max_disks && rdev->desc_nr >= mddev->max_disks) {
2086 printk(KERN_WARNING "md: %s: array is limited to %d devices\n",
2087 mdname(mddev), mddev->max_disks);
2090 bdevname(rdev->bdev,b);
2091 strreplace(b, '/', '!');
2093 rdev->mddev = mddev;
2094 printk(KERN_INFO "md: bind<%s>\n", b);
2096 if ((err = kobject_add(&rdev->kobj, &mddev->kobj, "dev-%s", b)))
2099 ko = &part_to_dev(rdev->bdev->bd_part)->kobj;
2100 if (sysfs_create_link(&rdev->kobj, ko, "block"))
2101 /* failure here is OK */;
2102 rdev->sysfs_state = sysfs_get_dirent_safe(rdev->kobj.sd, "state");
2104 list_add_rcu(&rdev->same_set, &mddev->disks);
2105 bd_link_disk_holder(rdev->bdev, mddev->gendisk);
2107 /* May as well allow recovery to be retried once */
2108 mddev->recovery_disabled++;
2113 printk(KERN_WARNING "md: failed to register dev-%s for %s\n",
2118 static void md_delayed_delete(struct work_struct *ws)
2120 struct md_rdev *rdev = container_of(ws, struct md_rdev, del_work);
2121 kobject_del(&rdev->kobj);
2122 kobject_put(&rdev->kobj);
2125 static void unbind_rdev_from_array(struct md_rdev *rdev)
2127 char b[BDEVNAME_SIZE];
2129 bd_unlink_disk_holder(rdev->bdev, rdev->mddev->gendisk);
2130 list_del_rcu(&rdev->same_set);
2131 printk(KERN_INFO "md: unbind<%s>\n", bdevname(rdev->bdev,b));
2133 sysfs_remove_link(&rdev->kobj, "block");
2134 sysfs_put(rdev->sysfs_state);
2135 rdev->sysfs_state = NULL;
2136 rdev->badblocks.count = 0;
2137 /* We need to delay this, otherwise we can deadlock when
2138 * writing to 'remove' to "dev/state". We also need
2139 * to delay it due to rcu usage.
2142 INIT_WORK(&rdev->del_work, md_delayed_delete);
2143 kobject_get(&rdev->kobj);
2144 queue_work(md_misc_wq, &rdev->del_work);
2148 * prevent the device from being mounted, repartitioned or
2149 * otherwise reused by a RAID array (or any other kernel
2150 * subsystem), by bd_claiming the device.
2152 static int lock_rdev(struct md_rdev *rdev, dev_t dev, int shared)
2155 struct block_device *bdev;
2156 char b[BDEVNAME_SIZE];
2158 bdev = blkdev_get_by_dev(dev, FMODE_READ|FMODE_WRITE|FMODE_EXCL,
2159 shared ? (struct md_rdev *)lock_rdev : rdev);
2161 printk(KERN_ERR "md: could not open %s.\n",
2162 __bdevname(dev, b));
2163 return PTR_ERR(bdev);
2169 static void unlock_rdev(struct md_rdev *rdev)
2171 struct block_device *bdev = rdev->bdev;
2173 blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
2176 void md_autodetect_dev(dev_t dev);
2178 static void export_rdev(struct md_rdev *rdev)
2180 char b[BDEVNAME_SIZE];
2182 printk(KERN_INFO "md: export_rdev(%s)\n",
2183 bdevname(rdev->bdev,b));
2184 md_rdev_clear(rdev);
2186 if (test_bit(AutoDetected, &rdev->flags))
2187 md_autodetect_dev(rdev->bdev->bd_dev);
2190 kobject_put(&rdev->kobj);
2193 void md_kick_rdev_from_array(struct md_rdev *rdev)
2195 unbind_rdev_from_array(rdev);
2198 EXPORT_SYMBOL_GPL(md_kick_rdev_from_array);
2200 static void export_array(struct mddev *mddev)
2202 struct md_rdev *rdev;
2204 while (!list_empty(&mddev->disks)) {
2205 rdev = list_first_entry(&mddev->disks, struct md_rdev,
2207 md_kick_rdev_from_array(rdev);
2209 mddev->raid_disks = 0;
2210 mddev->major_version = 0;
2213 static void sync_sbs(struct mddev *mddev, int nospares)
2215 /* Update each superblock (in-memory image), but
2216 * if we are allowed to, skip spares which already
2217 * have the right event counter, or have one earlier
2218 * (which would mean they aren't being marked as dirty
2219 * with the rest of the array)
2221 struct md_rdev *rdev;
2222 rdev_for_each(rdev, mddev) {
2223 if (rdev->sb_events == mddev->events ||
2225 rdev->raid_disk < 0 &&
2226 rdev->sb_events+1 == mddev->events)) {
2227 /* Don't update this superblock */
2228 rdev->sb_loaded = 2;
2230 sync_super(mddev, rdev);
2231 rdev->sb_loaded = 1;
2236 static bool does_sb_need_changing(struct mddev *mddev)
2238 struct md_rdev *rdev;
2239 struct mdp_superblock_1 *sb;
2242 /* Find a good rdev */
2243 rdev_for_each(rdev, mddev)
2244 if ((rdev->raid_disk >= 0) && !test_bit(Faulty, &rdev->flags))
2247 /* No good device found. */
2251 sb = page_address(rdev->sb_page);
2252 /* Check if a device has become faulty or a spare become active */
2253 rdev_for_each(rdev, mddev) {
2254 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
2255 /* Device activated? */
2256 if (role == 0xffff && rdev->raid_disk >=0 &&
2257 !test_bit(Faulty, &rdev->flags))
2259 /* Device turned faulty? */
2260 if (test_bit(Faulty, &rdev->flags) && (role < 0xfffd))
2264 /* Check if any mddev parameters have changed */
2265 if ((mddev->dev_sectors != le64_to_cpu(sb->size)) ||
2266 (mddev->reshape_position != le64_to_cpu(sb->reshape_position)) ||
2267 (mddev->layout != le64_to_cpu(sb->layout)) ||
2268 (mddev->raid_disks != le32_to_cpu(sb->raid_disks)) ||
2269 (mddev->chunk_sectors != le32_to_cpu(sb->chunksize)))
2275 void md_update_sb(struct mddev *mddev, int force_change)
2277 struct md_rdev *rdev;
2280 int any_badblocks_changed = 0;
2285 set_bit(MD_CHANGE_DEVS, &mddev->flags);
2289 if (mddev_is_clustered(mddev)) {
2290 if (test_and_clear_bit(MD_CHANGE_DEVS, &mddev->flags))
2292 ret = md_cluster_ops->metadata_update_start(mddev);
2293 /* Has someone else has updated the sb */
2294 if (!does_sb_need_changing(mddev)) {
2296 md_cluster_ops->metadata_update_cancel(mddev);
2297 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
2302 /* First make sure individual recovery_offsets are correct */
2303 rdev_for_each(rdev, mddev) {
2304 if (rdev->raid_disk >= 0 &&
2305 mddev->delta_disks >= 0 &&
2306 !test_bit(Journal, &rdev->flags) &&
2307 !test_bit(In_sync, &rdev->flags) &&
2308 mddev->curr_resync_completed > rdev->recovery_offset)
2309 rdev->recovery_offset = mddev->curr_resync_completed;
2312 if (!mddev->persistent) {
2313 clear_bit(MD_CHANGE_CLEAN, &mddev->flags);
2314 clear_bit(MD_CHANGE_DEVS, &mddev->flags);
2315 if (!mddev->external) {
2316 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
2317 rdev_for_each(rdev, mddev) {
2318 if (rdev->badblocks.changed) {
2319 rdev->badblocks.changed = 0;
2320 ack_all_badblocks(&rdev->badblocks);
2321 md_error(mddev, rdev);
2323 clear_bit(Blocked, &rdev->flags);
2324 clear_bit(BlockedBadBlocks, &rdev->flags);
2325 wake_up(&rdev->blocked_wait);
2328 wake_up(&mddev->sb_wait);
2332 spin_lock(&mddev->lock);
2334 mddev->utime = ktime_get_real_seconds();
2336 if (test_and_clear_bit(MD_CHANGE_DEVS, &mddev->flags))
2338 if (test_and_clear_bit(MD_CHANGE_CLEAN, &mddev->flags))
2339 /* just a clean<-> dirty transition, possibly leave spares alone,
2340 * though if events isn't the right even/odd, we will have to do
2346 if (mddev->degraded)
2347 /* If the array is degraded, then skipping spares is both
2348 * dangerous and fairly pointless.
2349 * Dangerous because a device that was removed from the array
2350 * might have a event_count that still looks up-to-date,
2351 * so it can be re-added without a resync.
2352 * Pointless because if there are any spares to skip,
2353 * then a recovery will happen and soon that array won't
2354 * be degraded any more and the spare can go back to sleep then.
2358 sync_req = mddev->in_sync;
2360 /* If this is just a dirty<->clean transition, and the array is clean
2361 * and 'events' is odd, we can roll back to the previous clean state */
2363 && (mddev->in_sync && mddev->recovery_cp == MaxSector)
2364 && mddev->can_decrease_events
2365 && mddev->events != 1) {
2367 mddev->can_decrease_events = 0;
2369 /* otherwise we have to go forward and ... */
2371 mddev->can_decrease_events = nospares;
2375 * This 64-bit counter should never wrap.
2376 * Either we are in around ~1 trillion A.C., assuming
2377 * 1 reboot per second, or we have a bug...
2379 WARN_ON(mddev->events == 0);
2381 rdev_for_each(rdev, mddev) {
2382 if (rdev->badblocks.changed)
2383 any_badblocks_changed++;
2384 if (test_bit(Faulty, &rdev->flags))
2385 set_bit(FaultRecorded, &rdev->flags);
2388 sync_sbs(mddev, nospares);
2389 spin_unlock(&mddev->lock);
2391 pr_debug("md: updating %s RAID superblock on device (in sync %d)\n",
2392 mdname(mddev), mddev->in_sync);
2394 bitmap_update_sb(mddev->bitmap);
2395 rdev_for_each(rdev, mddev) {
2396 char b[BDEVNAME_SIZE];
2398 if (rdev->sb_loaded != 1)
2399 continue; /* no noise on spare devices */
2401 if (!test_bit(Faulty, &rdev->flags)) {
2402 md_super_write(mddev,rdev,
2403 rdev->sb_start, rdev->sb_size,
2405 pr_debug("md: (write) %s's sb offset: %llu\n",
2406 bdevname(rdev->bdev, b),
2407 (unsigned long long)rdev->sb_start);
2408 rdev->sb_events = mddev->events;
2409 if (rdev->badblocks.size) {
2410 md_super_write(mddev, rdev,
2411 rdev->badblocks.sector,
2412 rdev->badblocks.size << 9,
2414 rdev->badblocks.size = 0;
2418 pr_debug("md: %s (skipping faulty)\n",
2419 bdevname(rdev->bdev, b));
2421 if (mddev->level == LEVEL_MULTIPATH)
2422 /* only need to write one superblock... */
2425 md_super_wait(mddev);
2426 /* if there was a failure, MD_CHANGE_DEVS was set, and we re-write super */
2428 spin_lock(&mddev->lock);
2429 if (mddev->in_sync != sync_req ||
2430 test_bit(MD_CHANGE_DEVS, &mddev->flags)) {
2431 /* have to write it out again */
2432 spin_unlock(&mddev->lock);
2435 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
2436 spin_unlock(&mddev->lock);
2437 wake_up(&mddev->sb_wait);
2438 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
2439 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
2441 rdev_for_each(rdev, mddev) {
2442 if (test_and_clear_bit(FaultRecorded, &rdev->flags))
2443 clear_bit(Blocked, &rdev->flags);
2445 if (any_badblocks_changed)
2446 ack_all_badblocks(&rdev->badblocks);
2447 clear_bit(BlockedBadBlocks, &rdev->flags);
2448 wake_up(&rdev->blocked_wait);
2451 if (mddev_is_clustered(mddev) && ret == 0)
2452 md_cluster_ops->metadata_update_finish(mddev);
2454 EXPORT_SYMBOL(md_update_sb);
2456 static int add_bound_rdev(struct md_rdev *rdev)
2458 struct mddev *mddev = rdev->mddev;
2460 bool add_journal = test_bit(Journal, &rdev->flags);
2462 if (!mddev->pers->hot_remove_disk || add_journal) {
2463 /* If there is hot_add_disk but no hot_remove_disk
2464 * then added disks for geometry changes,
2465 * and should be added immediately.
2467 super_types[mddev->major_version].
2468 validate_super(mddev, rdev);
2470 mddev_suspend(mddev);
2471 err = mddev->pers->hot_add_disk(mddev, rdev);
2473 mddev_resume(mddev);
2475 unbind_rdev_from_array(rdev);
2480 sysfs_notify_dirent_safe(rdev->sysfs_state);
2482 set_bit(MD_CHANGE_DEVS, &mddev->flags);
2483 if (mddev->degraded)
2484 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
2485 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2486 md_new_event(mddev);
2487 md_wakeup_thread(mddev->thread);
2491 /* words written to sysfs files may, or may not, be \n terminated.
2492 * We want to accept with case. For this we use cmd_match.
2494 static int cmd_match(const char *cmd, const char *str)
2496 /* See if cmd, written into a sysfs file, matches
2497 * str. They must either be the same, or cmd can
2498 * have a trailing newline
2500 while (*cmd && *str && *cmd == *str) {
2511 struct rdev_sysfs_entry {
2512 struct attribute attr;
2513 ssize_t (*show)(struct md_rdev *, char *);
2514 ssize_t (*store)(struct md_rdev *, const char *, size_t);
2518 state_show(struct md_rdev *rdev, char *page)
2522 unsigned long flags = ACCESS_ONCE(rdev->flags);
2524 if (test_bit(Faulty, &flags) ||
2525 rdev->badblocks.unacked_exist) {
2526 len+= sprintf(page+len, "%sfaulty",sep);
2529 if (test_bit(In_sync, &flags)) {
2530 len += sprintf(page+len, "%sin_sync",sep);
2533 if (test_bit(Journal, &flags)) {
2534 len += sprintf(page+len, "%sjournal",sep);
2537 if (test_bit(WriteMostly, &flags)) {
2538 len += sprintf(page+len, "%swrite_mostly",sep);
2541 if (test_bit(Blocked, &flags) ||
2542 (rdev->badblocks.unacked_exist
2543 && !test_bit(Faulty, &flags))) {
2544 len += sprintf(page+len, "%sblocked", sep);
2547 if (!test_bit(Faulty, &flags) &&
2548 !test_bit(Journal, &flags) &&
2549 !test_bit(In_sync, &flags)) {
2550 len += sprintf(page+len, "%sspare", sep);
2553 if (test_bit(WriteErrorSeen, &flags)) {
2554 len += sprintf(page+len, "%swrite_error", sep);
2557 if (test_bit(WantReplacement, &flags)) {
2558 len += sprintf(page+len, "%swant_replacement", sep);
2561 if (test_bit(Replacement, &flags)) {
2562 len += sprintf(page+len, "%sreplacement", sep);
2566 return len+sprintf(page+len, "\n");
2570 state_store(struct md_rdev *rdev, const char *buf, size_t len)
2573 * faulty - simulates an error
2574 * remove - disconnects the device
2575 * writemostly - sets write_mostly
2576 * -writemostly - clears write_mostly
2577 * blocked - sets the Blocked flags
2578 * -blocked - clears the Blocked and possibly simulates an error
2579 * insync - sets Insync providing device isn't active
2580 * -insync - clear Insync for a device with a slot assigned,
2581 * so that it gets rebuilt based on bitmap
2582 * write_error - sets WriteErrorSeen
2583 * -write_error - clears WriteErrorSeen
2586 if (cmd_match(buf, "faulty") && rdev->mddev->pers) {
2587 md_error(rdev->mddev, rdev);
2588 if (test_bit(Faulty, &rdev->flags))
2592 } else if (cmd_match(buf, "remove")) {
2593 if (rdev->raid_disk >= 0)
2596 struct mddev *mddev = rdev->mddev;
2598 if (mddev_is_clustered(mddev))
2599 err = md_cluster_ops->remove_disk(mddev, rdev);
2602 md_kick_rdev_from_array(rdev);
2604 md_update_sb(mddev, 1);
2605 md_new_event(mddev);
2608 } else if (cmd_match(buf, "writemostly")) {
2609 set_bit(WriteMostly, &rdev->flags);
2611 } else if (cmd_match(buf, "-writemostly")) {
2612 clear_bit(WriteMostly, &rdev->flags);
2614 } else if (cmd_match(buf, "blocked")) {
2615 set_bit(Blocked, &rdev->flags);
2617 } else if (cmd_match(buf, "-blocked")) {
2618 if (!test_bit(Faulty, &rdev->flags) &&
2619 rdev->badblocks.unacked_exist) {
2620 /* metadata handler doesn't understand badblocks,
2621 * so we need to fail the device
2623 md_error(rdev->mddev, rdev);
2625 clear_bit(Blocked, &rdev->flags);
2626 clear_bit(BlockedBadBlocks, &rdev->flags);
2627 wake_up(&rdev->blocked_wait);
2628 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2629 md_wakeup_thread(rdev->mddev->thread);
2632 } else if (cmd_match(buf, "insync") && rdev->raid_disk == -1) {
2633 set_bit(In_sync, &rdev->flags);
2635 } else if (cmd_match(buf, "-insync") && rdev->raid_disk >= 0 &&
2636 !test_bit(Journal, &rdev->flags)) {
2637 if (rdev->mddev->pers == NULL) {
2638 clear_bit(In_sync, &rdev->flags);
2639 rdev->saved_raid_disk = rdev->raid_disk;
2640 rdev->raid_disk = -1;
2643 } else if (cmd_match(buf, "write_error")) {
2644 set_bit(WriteErrorSeen, &rdev->flags);
2646 } else if (cmd_match(buf, "-write_error")) {
2647 clear_bit(WriteErrorSeen, &rdev->flags);
2649 } else if (cmd_match(buf, "want_replacement")) {
2650 /* Any non-spare device that is not a replacement can
2651 * become want_replacement at any time, but we then need to
2652 * check if recovery is needed.
2654 if (rdev->raid_disk >= 0 &&
2655 !test_bit(Journal, &rdev->flags) &&
2656 !test_bit(Replacement, &rdev->flags))
2657 set_bit(WantReplacement, &rdev->flags);
2658 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2659 md_wakeup_thread(rdev->mddev->thread);
2661 } else if (cmd_match(buf, "-want_replacement")) {
2662 /* Clearing 'want_replacement' is always allowed.
2663 * Once replacements starts it is too late though.
2666 clear_bit(WantReplacement, &rdev->flags);
2667 } else if (cmd_match(buf, "replacement")) {
2668 /* Can only set a device as a replacement when array has not
2669 * yet been started. Once running, replacement is automatic
2670 * from spares, or by assigning 'slot'.
2672 if (rdev->mddev->pers)
2675 set_bit(Replacement, &rdev->flags);
2678 } else if (cmd_match(buf, "-replacement")) {
2679 /* Similarly, can only clear Replacement before start */
2680 if (rdev->mddev->pers)
2683 clear_bit(Replacement, &rdev->flags);
2686 } else if (cmd_match(buf, "re-add")) {
2687 if (test_bit(Faulty, &rdev->flags) && (rdev->raid_disk == -1)) {
2688 /* clear_bit is performed _after_ all the devices
2689 * have their local Faulty bit cleared. If any writes
2690 * happen in the meantime in the local node, they
2691 * will land in the local bitmap, which will be synced
2692 * by this node eventually
2694 if (!mddev_is_clustered(rdev->mddev) ||
2695 (err = md_cluster_ops->gather_bitmaps(rdev)) == 0) {
2696 clear_bit(Faulty, &rdev->flags);
2697 err = add_bound_rdev(rdev);
2703 sysfs_notify_dirent_safe(rdev->sysfs_state);
2704 return err ? err : len;
2706 static struct rdev_sysfs_entry rdev_state =
2707 __ATTR_PREALLOC(state, S_IRUGO|S_IWUSR, state_show, state_store);
2710 errors_show(struct md_rdev *rdev, char *page)
2712 return sprintf(page, "%d\n", atomic_read(&rdev->corrected_errors));
2716 errors_store(struct md_rdev *rdev, const char *buf, size_t len)
2721 rv = kstrtouint(buf, 10, &n);
2724 atomic_set(&rdev->corrected_errors, n);
2727 static struct rdev_sysfs_entry rdev_errors =
2728 __ATTR(errors, S_IRUGO|S_IWUSR, errors_show, errors_store);
2731 slot_show(struct md_rdev *rdev, char *page)
2733 if (test_bit(Journal, &rdev->flags))
2734 return sprintf(page, "journal\n");
2735 else if (rdev->raid_disk < 0)
2736 return sprintf(page, "none\n");
2738 return sprintf(page, "%d\n", rdev->raid_disk);
2742 slot_store(struct md_rdev *rdev, const char *buf, size_t len)
2747 if (test_bit(Journal, &rdev->flags))
2749 if (strncmp(buf, "none", 4)==0)
2752 err = kstrtouint(buf, 10, (unsigned int *)&slot);
2756 if (rdev->mddev->pers && slot == -1) {
2757 /* Setting 'slot' on an active array requires also
2758 * updating the 'rd%d' link, and communicating
2759 * with the personality with ->hot_*_disk.
2760 * For now we only support removing
2761 * failed/spare devices. This normally happens automatically,
2762 * but not when the metadata is externally managed.
2764 if (rdev->raid_disk == -1)
2766 /* personality does all needed checks */
2767 if (rdev->mddev->pers->hot_remove_disk == NULL)
2769 clear_bit(Blocked, &rdev->flags);
2770 remove_and_add_spares(rdev->mddev, rdev);
2771 if (rdev->raid_disk >= 0)
2773 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2774 md_wakeup_thread(rdev->mddev->thread);
2775 } else if (rdev->mddev->pers) {
2776 /* Activating a spare .. or possibly reactivating
2777 * if we ever get bitmaps working here.
2781 if (rdev->raid_disk != -1)
2784 if (test_bit(MD_RECOVERY_RUNNING, &rdev->mddev->recovery))
2787 if (rdev->mddev->pers->hot_add_disk == NULL)
2790 if (slot >= rdev->mddev->raid_disks &&
2791 slot >= rdev->mddev->raid_disks + rdev->mddev->delta_disks)
2794 rdev->raid_disk = slot;
2795 if (test_bit(In_sync, &rdev->flags))
2796 rdev->saved_raid_disk = slot;
2798 rdev->saved_raid_disk = -1;
2799 clear_bit(In_sync, &rdev->flags);
2800 clear_bit(Bitmap_sync, &rdev->flags);
2801 err = rdev->mddev->pers->
2802 hot_add_disk(rdev->mddev, rdev);
2804 rdev->raid_disk = -1;
2807 sysfs_notify_dirent_safe(rdev->sysfs_state);
2808 if (sysfs_link_rdev(rdev->mddev, rdev))
2809 /* failure here is OK */;
2810 /* don't wakeup anyone, leave that to userspace. */
2812 if (slot >= rdev->mddev->raid_disks &&
2813 slot >= rdev->mddev->raid_disks + rdev->mddev->delta_disks)
2815 rdev->raid_disk = slot;
2816 /* assume it is working */
2817 clear_bit(Faulty, &rdev->flags);
2818 clear_bit(WriteMostly, &rdev->flags);
2819 set_bit(In_sync, &rdev->flags);
2820 sysfs_notify_dirent_safe(rdev->sysfs_state);
2825 static struct rdev_sysfs_entry rdev_slot =
2826 __ATTR(slot, S_IRUGO|S_IWUSR, slot_show, slot_store);
2829 offset_show(struct md_rdev *rdev, char *page)
2831 return sprintf(page, "%llu\n", (unsigned long long)rdev->data_offset);
2835 offset_store(struct md_rdev *rdev, const char *buf, size_t len)
2837 unsigned long long offset;
2838 if (kstrtoull(buf, 10, &offset) < 0)
2840 if (rdev->mddev->pers && rdev->raid_disk >= 0)
2842 if (rdev->sectors && rdev->mddev->external)
2843 /* Must set offset before size, so overlap checks
2846 rdev->data_offset = offset;
2847 rdev->new_data_offset = offset;
2851 static struct rdev_sysfs_entry rdev_offset =
2852 __ATTR(offset, S_IRUGO|S_IWUSR, offset_show, offset_store);
2854 static ssize_t new_offset_show(struct md_rdev *rdev, char *page)
2856 return sprintf(page, "%llu\n",
2857 (unsigned long long)rdev->new_data_offset);
2860 static ssize_t new_offset_store(struct md_rdev *rdev,
2861 const char *buf, size_t len)
2863 unsigned long long new_offset;
2864 struct mddev *mddev = rdev->mddev;
2866 if (kstrtoull(buf, 10, &new_offset) < 0)
2869 if (mddev->sync_thread ||
2870 test_bit(MD_RECOVERY_RUNNING,&mddev->recovery))
2872 if (new_offset == rdev->data_offset)
2873 /* reset is always permitted */
2875 else if (new_offset > rdev->data_offset) {
2876 /* must not push array size beyond rdev_sectors */
2877 if (new_offset - rdev->data_offset
2878 + mddev->dev_sectors > rdev->sectors)
2881 /* Metadata worries about other space details. */
2883 /* decreasing the offset is inconsistent with a backwards
2886 if (new_offset < rdev->data_offset &&
2887 mddev->reshape_backwards)
2889 /* Increasing offset is inconsistent with forwards
2890 * reshape. reshape_direction should be set to
2891 * 'backwards' first.
2893 if (new_offset > rdev->data_offset &&
2894 !mddev->reshape_backwards)
2897 if (mddev->pers && mddev->persistent &&
2898 !super_types[mddev->major_version]
2899 .allow_new_offset(rdev, new_offset))
2901 rdev->new_data_offset = new_offset;
2902 if (new_offset > rdev->data_offset)
2903 mddev->reshape_backwards = 1;
2904 else if (new_offset < rdev->data_offset)
2905 mddev->reshape_backwards = 0;
2909 static struct rdev_sysfs_entry rdev_new_offset =
2910 __ATTR(new_offset, S_IRUGO|S_IWUSR, new_offset_show, new_offset_store);
2913 rdev_size_show(struct md_rdev *rdev, char *page)
2915 return sprintf(page, "%llu\n", (unsigned long long)rdev->sectors / 2);
2918 static int overlaps(sector_t s1, sector_t l1, sector_t s2, sector_t l2)
2920 /* check if two start/length pairs overlap */
2928 static int strict_blocks_to_sectors(const char *buf, sector_t *sectors)
2930 unsigned long long blocks;
2933 if (kstrtoull(buf, 10, &blocks) < 0)
2936 if (blocks & 1ULL << (8 * sizeof(blocks) - 1))
2937 return -EINVAL; /* sector conversion overflow */
2940 if (new != blocks * 2)
2941 return -EINVAL; /* unsigned long long to sector_t overflow */
2948 rdev_size_store(struct md_rdev *rdev, const char *buf, size_t len)
2950 struct mddev *my_mddev = rdev->mddev;
2951 sector_t oldsectors = rdev->sectors;
2954 if (test_bit(Journal, &rdev->flags))
2956 if (strict_blocks_to_sectors(buf, §ors) < 0)
2958 if (rdev->data_offset != rdev->new_data_offset)
2959 return -EINVAL; /* too confusing */
2960 if (my_mddev->pers && rdev->raid_disk >= 0) {
2961 if (my_mddev->persistent) {
2962 sectors = super_types[my_mddev->major_version].
2963 rdev_size_change(rdev, sectors);
2966 } else if (!sectors)
2967 sectors = (i_size_read(rdev->bdev->bd_inode) >> 9) -
2969 if (!my_mddev->pers->resize)
2970 /* Cannot change size for RAID0 or Linear etc */
2973 if (sectors < my_mddev->dev_sectors)
2974 return -EINVAL; /* component must fit device */
2976 rdev->sectors = sectors;
2977 if (sectors > oldsectors && my_mddev->external) {
2978 /* Need to check that all other rdevs with the same
2979 * ->bdev do not overlap. 'rcu' is sufficient to walk
2980 * the rdev lists safely.
2981 * This check does not provide a hard guarantee, it
2982 * just helps avoid dangerous mistakes.
2984 struct mddev *mddev;
2986 struct list_head *tmp;
2989 for_each_mddev(mddev, tmp) {
2990 struct md_rdev *rdev2;
2992 rdev_for_each(rdev2, mddev)
2993 if (rdev->bdev == rdev2->bdev &&
2995 overlaps(rdev->data_offset, rdev->sectors,
3008 /* Someone else could have slipped in a size
3009 * change here, but doing so is just silly.
3010 * We put oldsectors back because we *know* it is
3011 * safe, and trust userspace not to race with
3014 rdev->sectors = oldsectors;
3021 static struct rdev_sysfs_entry rdev_size =
3022 __ATTR(size, S_IRUGO|S_IWUSR, rdev_size_show, rdev_size_store);
3024 static ssize_t recovery_start_show(struct md_rdev *rdev, char *page)
3026 unsigned long long recovery_start = rdev->recovery_offset;
3028 if (test_bit(In_sync, &rdev->flags) ||
3029 recovery_start == MaxSector)
3030 return sprintf(page, "none\n");
3032 return sprintf(page, "%llu\n", recovery_start);
3035 static ssize_t recovery_start_store(struct md_rdev *rdev, const char *buf, size_t len)
3037 unsigned long long recovery_start;
3039 if (cmd_match(buf, "none"))
3040 recovery_start = MaxSector;
3041 else if (kstrtoull(buf, 10, &recovery_start))
3044 if (rdev->mddev->pers &&
3045 rdev->raid_disk >= 0)
3048 rdev->recovery_offset = recovery_start;
3049 if (recovery_start == MaxSector)
3050 set_bit(In_sync, &rdev->flags);
3052 clear_bit(In_sync, &rdev->flags);
3056 static struct rdev_sysfs_entry rdev_recovery_start =
3057 __ATTR(recovery_start, S_IRUGO|S_IWUSR, recovery_start_show, recovery_start_store);
3059 /* sysfs access to bad-blocks list.
3060 * We present two files.
3061 * 'bad-blocks' lists sector numbers and lengths of ranges that
3062 * are recorded as bad. The list is truncated to fit within
3063 * the one-page limit of sysfs.
3064 * Writing "sector length" to this file adds an acknowledged
3066 * 'unacknowledged-bad-blocks' lists bad blocks that have not yet
3067 * been acknowledged. Writing to this file adds bad blocks
3068 * without acknowledging them. This is largely for testing.
3070 static ssize_t bb_show(struct md_rdev *rdev, char *page)
3072 return badblocks_show(&rdev->badblocks, page, 0);
3074 static ssize_t bb_store(struct md_rdev *rdev, const char *page, size_t len)
3076 int rv = badblocks_store(&rdev->badblocks, page, len, 0);
3077 /* Maybe that ack was all we needed */
3078 if (test_and_clear_bit(BlockedBadBlocks, &rdev->flags))
3079 wake_up(&rdev->blocked_wait);
3082 static struct rdev_sysfs_entry rdev_bad_blocks =
3083 __ATTR(bad_blocks, S_IRUGO|S_IWUSR, bb_show, bb_store);
3085 static ssize_t ubb_show(struct md_rdev *rdev, char *page)
3087 return badblocks_show(&rdev->badblocks, page, 1);
3089 static ssize_t ubb_store(struct md_rdev *rdev, const char *page, size_t len)
3091 return badblocks_store(&rdev->badblocks, page, len, 1);
3093 static struct rdev_sysfs_entry rdev_unack_bad_blocks =
3094 __ATTR(unacknowledged_bad_blocks, S_IRUGO|S_IWUSR, ubb_show, ubb_store);
3096 static struct attribute *rdev_default_attrs[] = {
3101 &rdev_new_offset.attr,
3103 &rdev_recovery_start.attr,
3104 &rdev_bad_blocks.attr,
3105 &rdev_unack_bad_blocks.attr,
3109 rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
3111 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
3112 struct md_rdev *rdev = container_of(kobj, struct md_rdev, kobj);
3118 return entry->show(rdev, page);
3122 rdev_attr_store(struct kobject *kobj, struct attribute *attr,
3123 const char *page, size_t length)
3125 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
3126 struct md_rdev *rdev = container_of(kobj, struct md_rdev, kobj);
3128 struct mddev *mddev = rdev->mddev;
3132 if (!capable(CAP_SYS_ADMIN))
3134 rv = mddev ? mddev_lock(mddev): -EBUSY;
3136 if (rdev->mddev == NULL)
3139 rv = entry->store(rdev, page, length);
3140 mddev_unlock(mddev);
3145 static void rdev_free(struct kobject *ko)
3147 struct md_rdev *rdev = container_of(ko, struct md_rdev, kobj);
3150 static const struct sysfs_ops rdev_sysfs_ops = {
3151 .show = rdev_attr_show,
3152 .store = rdev_attr_store,
3154 static struct kobj_type rdev_ktype = {
3155 .release = rdev_free,
3156 .sysfs_ops = &rdev_sysfs_ops,
3157 .default_attrs = rdev_default_attrs,
3160 int md_rdev_init(struct md_rdev *rdev)
3163 rdev->saved_raid_disk = -1;
3164 rdev->raid_disk = -1;
3166 rdev->data_offset = 0;
3167 rdev->new_data_offset = 0;
3168 rdev->sb_events = 0;
3169 rdev->last_read_error.tv_sec = 0;
3170 rdev->last_read_error.tv_nsec = 0;
3171 rdev->sb_loaded = 0;
3172 rdev->bb_page = NULL;
3173 atomic_set(&rdev->nr_pending, 0);
3174 atomic_set(&rdev->read_errors, 0);
3175 atomic_set(&rdev->corrected_errors, 0);
3177 INIT_LIST_HEAD(&rdev->same_set);
3178 init_waitqueue_head(&rdev->blocked_wait);
3180 /* Add space to store bad block list.
3181 * This reserves the space even on arrays where it cannot
3182 * be used - I wonder if that matters
3184 return badblocks_init(&rdev->badblocks, 0);
3186 EXPORT_SYMBOL_GPL(md_rdev_init);
3188 * Import a device. If 'super_format' >= 0, then sanity check the superblock
3190 * mark the device faulty if:
3192 * - the device is nonexistent (zero size)
3193 * - the device has no valid superblock
3195 * a faulty rdev _never_ has rdev->sb set.
3197 static struct md_rdev *md_import_device(dev_t newdev, int super_format, int super_minor)
3199 char b[BDEVNAME_SIZE];
3201 struct md_rdev *rdev;
3204 rdev = kzalloc(sizeof(*rdev), GFP_KERNEL);
3206 printk(KERN_ERR "md: could not alloc mem for new device!\n");
3207 return ERR_PTR(-ENOMEM);
3210 err = md_rdev_init(rdev);
3213 err = alloc_disk_sb(rdev);
3217 err = lock_rdev(rdev, newdev, super_format == -2);
3221 kobject_init(&rdev->kobj, &rdev_ktype);
3223 size = i_size_read(rdev->bdev->bd_inode) >> BLOCK_SIZE_BITS;
3226 "md: %s has zero or unknown size, marking faulty!\n",
3227 bdevname(rdev->bdev,b));
3232 if (super_format >= 0) {
3233 err = super_types[super_format].
3234 load_super(rdev, NULL, super_minor);
3235 if (err == -EINVAL) {
3237 "md: %s does not have a valid v%d.%d "
3238 "superblock, not importing!\n",
3239 bdevname(rdev->bdev,b),
3240 super_format, super_minor);
3245 "md: could not read %s's sb, not importing!\n",
3246 bdevname(rdev->bdev,b));
3256 md_rdev_clear(rdev);
3258 return ERR_PTR(err);
3262 * Check a full RAID array for plausibility
3265 static void analyze_sbs(struct mddev *mddev)
3268 struct md_rdev *rdev, *freshest, *tmp;
3269 char b[BDEVNAME_SIZE];
3272 rdev_for_each_safe(rdev, tmp, mddev)
3273 switch (super_types[mddev->major_version].
3274 load_super(rdev, freshest, mddev->minor_version)) {
3282 "md: fatal superblock inconsistency in %s"
3283 " -- removing from array\n",
3284 bdevname(rdev->bdev,b));
3285 md_kick_rdev_from_array(rdev);
3288 super_types[mddev->major_version].
3289 validate_super(mddev, freshest);
3292 rdev_for_each_safe(rdev, tmp, mddev) {
3293 if (mddev->max_disks &&
3294 (rdev->desc_nr >= mddev->max_disks ||
3295 i > mddev->max_disks)) {
3297 "md: %s: %s: only %d devices permitted\n",
3298 mdname(mddev), bdevname(rdev->bdev, b),
3300 md_kick_rdev_from_array(rdev);
3303 if (rdev != freshest) {
3304 if (super_types[mddev->major_version].
3305 validate_super(mddev, rdev)) {
3306 printk(KERN_WARNING "md: kicking non-fresh %s"
3308 bdevname(rdev->bdev,b));
3309 md_kick_rdev_from_array(rdev);
3313 if (mddev->level == LEVEL_MULTIPATH) {
3314 rdev->desc_nr = i++;
3315 rdev->raid_disk = rdev->desc_nr;
3316 set_bit(In_sync, &rdev->flags);
3317 } else if (rdev->raid_disk >=
3318 (mddev->raid_disks - min(0, mddev->delta_disks)) &&
3319 !test_bit(Journal, &rdev->flags)) {
3320 rdev->raid_disk = -1;
3321 clear_bit(In_sync, &rdev->flags);
3326 /* Read a fixed-point number.
3327 * Numbers in sysfs attributes should be in "standard" units where
3328 * possible, so time should be in seconds.
3329 * However we internally use a a much smaller unit such as
3330 * milliseconds or jiffies.
3331 * This function takes a decimal number with a possible fractional
3332 * component, and produces an integer which is the result of
3333 * multiplying that number by 10^'scale'.
3334 * all without any floating-point arithmetic.
3336 int strict_strtoul_scaled(const char *cp, unsigned long *res, int scale)
3338 unsigned long result = 0;
3340 while (isdigit(*cp) || (*cp == '.' && decimals < 0)) {
3343 else if (decimals < scale) {
3346 result = result * 10 + value;
3358 while (decimals < scale) {
3367 safe_delay_show(struct mddev *mddev, char *page)
3369 int msec = (mddev->safemode_delay*1000)/HZ;
3370 return sprintf(page, "%d.%03d\n", msec/1000, msec%1000);
3373 safe_delay_store(struct mddev *mddev, const char *cbuf, size_t len)
3377 if (mddev_is_clustered(mddev)) {
3378 pr_info("md: Safemode is disabled for clustered mode\n");
3382 if (strict_strtoul_scaled(cbuf, &msec, 3) < 0)
3385 mddev->safemode_delay = 0;
3387 unsigned long old_delay = mddev->safemode_delay;
3388 unsigned long new_delay = (msec*HZ)/1000;
3392 mddev->safemode_delay = new_delay;
3393 if (new_delay < old_delay || old_delay == 0)
3394 mod_timer(&mddev->safemode_timer, jiffies+1);
3398 static struct md_sysfs_entry md_safe_delay =
3399 __ATTR(safe_mode_delay, S_IRUGO|S_IWUSR,safe_delay_show, safe_delay_store);
3402 level_show(struct mddev *mddev, char *page)
3404 struct md_personality *p;
3406 spin_lock(&mddev->lock);
3409 ret = sprintf(page, "%s\n", p->name);
3410 else if (mddev->clevel[0])
3411 ret = sprintf(page, "%s\n", mddev->clevel);
3412 else if (mddev->level != LEVEL_NONE)
3413 ret = sprintf(page, "%d\n", mddev->level);
3416 spin_unlock(&mddev->lock);
3421 level_store(struct mddev *mddev, const char *buf, size_t len)
3426 struct md_personality *pers, *oldpers;
3428 void *priv, *oldpriv;
3429 struct md_rdev *rdev;
3431 if (slen == 0 || slen >= sizeof(clevel))
3434 rv = mddev_lock(mddev);
3438 if (mddev->pers == NULL) {
3439 strncpy(mddev->clevel, buf, slen);
3440 if (mddev->clevel[slen-1] == '\n')
3442 mddev->clevel[slen] = 0;
3443 mddev->level = LEVEL_NONE;
3451 /* request to change the personality. Need to ensure:
3452 * - array is not engaged in resync/recovery/reshape
3453 * - old personality can be suspended
3454 * - new personality will access other array.
3458 if (mddev->sync_thread ||
3459 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
3460 mddev->reshape_position != MaxSector ||
3461 mddev->sysfs_active)
3465 if (!mddev->pers->quiesce) {
3466 printk(KERN_WARNING "md: %s: %s does not support online personality change\n",
3467 mdname(mddev), mddev->pers->name);
3471 /* Now find the new personality */
3472 strncpy(clevel, buf, slen);
3473 if (clevel[slen-1] == '\n')
3476 if (kstrtol(clevel, 10, &level))
3479 if (request_module("md-%s", clevel) != 0)
3480 request_module("md-level-%s", clevel);
3481 spin_lock(&pers_lock);
3482 pers = find_pers(level, clevel);
3483 if (!pers || !try_module_get(pers->owner)) {
3484 spin_unlock(&pers_lock);
3485 printk(KERN_WARNING "md: personality %s not loaded\n", clevel);
3489 spin_unlock(&pers_lock);
3491 if (pers == mddev->pers) {
3492 /* Nothing to do! */
3493 module_put(pers->owner);
3497 if (!pers->takeover) {
3498 module_put(pers->owner);
3499 printk(KERN_WARNING "md: %s: %s does not support personality takeover\n",
3500 mdname(mddev), clevel);
3505 rdev_for_each(rdev, mddev)
3506 rdev->new_raid_disk = rdev->raid_disk;
3508 /* ->takeover must set new_* and/or delta_disks
3509 * if it succeeds, and may set them when it fails.
3511 priv = pers->takeover(mddev);
3513 mddev->new_level = mddev->level;
3514 mddev->new_layout = mddev->layout;
3515 mddev->new_chunk_sectors = mddev->chunk_sectors;
3516 mddev->raid_disks -= mddev->delta_disks;
3517 mddev->delta_disks = 0;
3518 mddev->reshape_backwards = 0;
3519 module_put(pers->owner);
3520 printk(KERN_WARNING "md: %s: %s would not accept array\n",
3521 mdname(mddev), clevel);
3526 /* Looks like we have a winner */
3527 mddev_suspend(mddev);
3528 mddev_detach(mddev);
3530 spin_lock(&mddev->lock);
3531 oldpers = mddev->pers;
3532 oldpriv = mddev->private;
3534 mddev->private = priv;
3535 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
3536 mddev->level = mddev->new_level;
3537 mddev->layout = mddev->new_layout;
3538 mddev->chunk_sectors = mddev->new_chunk_sectors;
3539 mddev->delta_disks = 0;
3540 mddev->reshape_backwards = 0;
3541 mddev->degraded = 0;
3542 spin_unlock(&mddev->lock);
3544 if (oldpers->sync_request == NULL &&
3546 /* We are converting from a no-redundancy array
3547 * to a redundancy array and metadata is managed
3548 * externally so we need to be sure that writes
3549 * won't block due to a need to transition
3551 * until external management is started.
3554 mddev->safemode_delay = 0;
3555 mddev->safemode = 0;
3558 oldpers->free(mddev, oldpriv);
3560 if (oldpers->sync_request == NULL &&
3561 pers->sync_request != NULL) {
3562 /* need to add the md_redundancy_group */
3563 if (sysfs_create_group(&mddev->kobj, &md_redundancy_group))
3565 "md: cannot register extra attributes for %s\n",
3567 mddev->sysfs_action = sysfs_get_dirent(mddev->kobj.sd, "sync_action");
3569 if (oldpers->sync_request != NULL &&
3570 pers->sync_request == NULL) {
3571 /* need to remove the md_redundancy_group */
3572 if (mddev->to_remove == NULL)
3573 mddev->to_remove = &md_redundancy_group;
3576 rdev_for_each(rdev, mddev) {
3577 if (rdev->raid_disk < 0)
3579 if (rdev->new_raid_disk >= mddev->raid_disks)
3580 rdev->new_raid_disk = -1;
3581 if (rdev->new_raid_disk == rdev->raid_disk)
3583 sysfs_unlink_rdev(mddev, rdev);
3585 rdev_for_each(rdev, mddev) {
3586 if (rdev->raid_disk < 0)
3588 if (rdev->new_raid_disk == rdev->raid_disk)
3590 rdev->raid_disk = rdev->new_raid_disk;
3591 if (rdev->raid_disk < 0)
3592 clear_bit(In_sync, &rdev->flags);
3594 if (sysfs_link_rdev(mddev, rdev))
3595 printk(KERN_WARNING "md: cannot register rd%d"
3596 " for %s after level change\n",
3597 rdev->raid_disk, mdname(mddev));
3601 if (pers->sync_request == NULL) {
3602 /* this is now an array without redundancy, so
3603 * it must always be in_sync
3606 del_timer_sync(&mddev->safemode_timer);
3608 blk_set_stacking_limits(&mddev->queue->limits);
3610 set_bit(MD_CHANGE_DEVS, &mddev->flags);
3611 mddev_resume(mddev);
3613 md_update_sb(mddev, 1);
3614 sysfs_notify(&mddev->kobj, NULL, "level");
3615 md_new_event(mddev);
3618 mddev_unlock(mddev);
3622 static struct md_sysfs_entry md_level =
3623 __ATTR(level, S_IRUGO|S_IWUSR, level_show, level_store);
3626 layout_show(struct mddev *mddev, char *page)
3628 /* just a number, not meaningful for all levels */
3629 if (mddev->reshape_position != MaxSector &&
3630 mddev->layout != mddev->new_layout)
3631 return sprintf(page, "%d (%d)\n",
3632 mddev->new_layout, mddev->layout);
3633 return sprintf(page, "%d\n", mddev->layout);
3637 layout_store(struct mddev *mddev, const char *buf, size_t len)
3642 err = kstrtouint(buf, 10, &n);
3645 err = mddev_lock(mddev);
3650 if (mddev->pers->check_reshape == NULL)
3655 mddev->new_layout = n;
3656 err = mddev->pers->check_reshape(mddev);
3658 mddev->new_layout = mddev->layout;
3661 mddev->new_layout = n;
3662 if (mddev->reshape_position == MaxSector)
3665 mddev_unlock(mddev);
3668 static struct md_sysfs_entry md_layout =
3669 __ATTR(layout, S_IRUGO|S_IWUSR, layout_show, layout_store);
3672 raid_disks_show(struct mddev *mddev, char *page)
3674 if (mddev->raid_disks == 0)
3676 if (mddev->reshape_position != MaxSector &&
3677 mddev->delta_disks != 0)
3678 return sprintf(page, "%d (%d)\n", mddev->raid_disks,
3679 mddev->raid_disks - mddev->delta_disks);
3680 return sprintf(page, "%d\n", mddev->raid_disks);
3683 static int update_raid_disks(struct mddev *mddev, int raid_disks);
3686 raid_disks_store(struct mddev *mddev, const char *buf, size_t len)
3691 err = kstrtouint(buf, 10, &n);
3695 err = mddev_lock(mddev);
3699 err = update_raid_disks(mddev, n);
3700 else if (mddev->reshape_position != MaxSector) {
3701 struct md_rdev *rdev;
3702 int olddisks = mddev->raid_disks - mddev->delta_disks;
3705 rdev_for_each(rdev, mddev) {
3707 rdev->data_offset < rdev->new_data_offset)
3710 rdev->data_offset > rdev->new_data_offset)
3714 mddev->delta_disks = n - olddisks;
3715 mddev->raid_disks = n;
3716 mddev->reshape_backwards = (mddev->delta_disks < 0);
3718 mddev->raid_disks = n;
3720 mddev_unlock(mddev);
3721 return err ? err : len;
3723 static struct md_sysfs_entry md_raid_disks =
3724 __ATTR(raid_disks, S_IRUGO|S_IWUSR, raid_disks_show, raid_disks_store);
3727 chunk_size_show(struct mddev *mddev, char *page)
3729 if (mddev->reshape_position != MaxSector &&
3730 mddev->chunk_sectors != mddev->new_chunk_sectors)
3731 return sprintf(page, "%d (%d)\n",
3732 mddev->new_chunk_sectors << 9,
3733 mddev->chunk_sectors << 9);
3734 return sprintf(page, "%d\n", mddev->chunk_sectors << 9);
3738 chunk_size_store(struct mddev *mddev, const char *buf, size_t len)
3743 err = kstrtoul(buf, 10, &n);
3747 err = mddev_lock(mddev);
3751 if (mddev->pers->check_reshape == NULL)
3756 mddev->new_chunk_sectors = n >> 9;
3757 err = mddev->pers->check_reshape(mddev);
3759 mddev->new_chunk_sectors = mddev->chunk_sectors;
3762 mddev->new_chunk_sectors = n >> 9;
3763 if (mddev->reshape_position == MaxSector)
3764 mddev->chunk_sectors = n >> 9;
3766 mddev_unlock(mddev);
3769 static struct md_sysfs_entry md_chunk_size =
3770 __ATTR(chunk_size, S_IRUGO|S_IWUSR, chunk_size_show, chunk_size_store);
3773 resync_start_show(struct mddev *mddev, char *page)
3775 if (mddev->recovery_cp == MaxSector)
3776 return sprintf(page, "none\n");
3777 return sprintf(page, "%llu\n", (unsigned long long)mddev->recovery_cp);
3781 resync_start_store(struct mddev *mddev, const char *buf, size_t len)
3783 unsigned long long n;
3786 if (cmd_match(buf, "none"))
3789 err = kstrtoull(buf, 10, &n);
3792 if (n != (sector_t)n)
3796 err = mddev_lock(mddev);
3799 if (mddev->pers && !test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
3803 mddev->recovery_cp = n;
3805 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
3807 mddev_unlock(mddev);
3810 static struct md_sysfs_entry md_resync_start =
3811 __ATTR_PREALLOC(resync_start, S_IRUGO|S_IWUSR,
3812 resync_start_show, resync_start_store);
3815 * The array state can be:
3818 * No devices, no size, no level
3819 * Equivalent to STOP_ARRAY ioctl
3821 * May have some settings, but array is not active
3822 * all IO results in error
3823 * When written, doesn't tear down array, but just stops it
3824 * suspended (not supported yet)
3825 * All IO requests will block. The array can be reconfigured.
3826 * Writing this, if accepted, will block until array is quiescent
3828 * no resync can happen. no superblocks get written.
3829 * write requests fail
3831 * like readonly, but behaves like 'clean' on a write request.
3833 * clean - no pending writes, but otherwise active.
3834 * When written to inactive array, starts without resync
3835 * If a write request arrives then
3836 * if metadata is known, mark 'dirty' and switch to 'active'.
3837 * if not known, block and switch to write-pending
3838 * If written to an active array that has pending writes, then fails.
3840 * fully active: IO and resync can be happening.
3841 * When written to inactive array, starts with resync
3844 * clean, but writes are blocked waiting for 'active' to be written.
3847 * like active, but no writes have been seen for a while (100msec).
3850 enum array_state { clear, inactive, suspended, readonly, read_auto, clean, active,
3851 write_pending, active_idle, bad_word};
3852 static char *array_states[] = {
3853 "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active",
3854 "write-pending", "active-idle", NULL };
3856 static int match_word(const char *word, char **list)
3859 for (n=0; list[n]; n++)
3860 if (cmd_match(word, list[n]))
3866 array_state_show(struct mddev *mddev, char *page)
3868 enum array_state st = inactive;
3881 else if (test_bit(MD_CHANGE_PENDING, &mddev->flags))
3883 else if (mddev->safemode)
3889 if (list_empty(&mddev->disks) &&
3890 mddev->raid_disks == 0 &&
3891 mddev->dev_sectors == 0)
3896 return sprintf(page, "%s\n", array_states[st]);
3899 static int do_md_stop(struct mddev *mddev, int ro, struct block_device *bdev);
3900 static int md_set_readonly(struct mddev *mddev, struct block_device *bdev);
3901 static int do_md_run(struct mddev *mddev);
3902 static int restart_array(struct mddev *mddev);
3905 array_state_store(struct mddev *mddev, const char *buf, size_t len)
3908 enum array_state st = match_word(buf, array_states);
3910 if (mddev->pers && (st == active || st == clean) && mddev->ro != 1) {
3911 /* don't take reconfig_mutex when toggling between
3914 spin_lock(&mddev->lock);
3916 restart_array(mddev);
3917 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
3918 wake_up(&mddev->sb_wait);
3920 } else /* st == clean */ {
3921 restart_array(mddev);
3922 if (atomic_read(&mddev->writes_pending) == 0) {
3923 if (mddev->in_sync == 0) {
3925 if (mddev->safemode == 1)
3926 mddev->safemode = 0;
3927 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
3933 spin_unlock(&mddev->lock);
3936 err = mddev_lock(mddev);
3944 /* stopping an active array */
3945 err = do_md_stop(mddev, 0, NULL);
3948 /* stopping an active array */
3950 err = do_md_stop(mddev, 2, NULL);
3952 err = 0; /* already inactive */
3955 break; /* not supported yet */
3958 err = md_set_readonly(mddev, NULL);
3961 set_disk_ro(mddev->gendisk, 1);
3962 err = do_md_run(mddev);
3968 err = md_set_readonly(mddev, NULL);
3969 else if (mddev->ro == 1)
3970 err = restart_array(mddev);
3973 set_disk_ro(mddev->gendisk, 0);
3977 err = do_md_run(mddev);
3982 err = restart_array(mddev);
3985 spin_lock(&mddev->lock);
3986 if (atomic_read(&mddev->writes_pending) == 0) {
3987 if (mddev->in_sync == 0) {
3989 if (mddev->safemode == 1)
3990 mddev->safemode = 0;
3991 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
3996 spin_unlock(&mddev->lock);
4002 err = restart_array(mddev);
4005 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
4006 wake_up(&mddev->sb_wait);
4010 set_disk_ro(mddev->gendisk, 0);
4011 err = do_md_run(mddev);
4016 /* these cannot be set */
4021 if (mddev->hold_active == UNTIL_IOCTL)
4022 mddev->hold_active = 0;
4023 sysfs_notify_dirent_safe(mddev->sysfs_state);
4025 mddev_unlock(mddev);
4028 static struct md_sysfs_entry md_array_state =
4029 __ATTR_PREALLOC(array_state, S_IRUGO|S_IWUSR, array_state_show, array_state_store);
4032 max_corrected_read_errors_show(struct mddev *mddev, char *page) {
4033 return sprintf(page, "%d\n",
4034 atomic_read(&mddev->max_corr_read_errors));
4038 max_corrected_read_errors_store(struct mddev *mddev, const char *buf, size_t len)
4043 rv = kstrtouint(buf, 10, &n);
4046 atomic_set(&mddev->max_corr_read_errors, n);
4050 static struct md_sysfs_entry max_corr_read_errors =
4051 __ATTR(max_read_errors, S_IRUGO|S_IWUSR, max_corrected_read_errors_show,
4052 max_corrected_read_errors_store);
4055 null_show(struct mddev *mddev, char *page)
4061 new_dev_store(struct mddev *mddev, const char *buf, size_t len)
4063 /* buf must be %d:%d\n? giving major and minor numbers */
4064 /* The new device is added to the array.
4065 * If the array has a persistent superblock, we read the
4066 * superblock to initialise info and check validity.
4067 * Otherwise, only checking done is that in bind_rdev_to_array,
4068 * which mainly checks size.
4071 int major = simple_strtoul(buf, &e, 10);
4074 struct md_rdev *rdev;
4077 if (!*buf || *e != ':' || !e[1] || e[1] == '\n')
4079 minor = simple_strtoul(e+1, &e, 10);
4080 if (*e && *e != '\n')
4082 dev = MKDEV(major, minor);
4083 if (major != MAJOR(dev) ||
4084 minor != MINOR(dev))
4087 flush_workqueue(md_misc_wq);
4089 err = mddev_lock(mddev);
4092 if (mddev->persistent) {
4093 rdev = md_import_device(dev, mddev->major_version,
4094 mddev->minor_version);
4095 if (!IS_ERR(rdev) && !list_empty(&mddev->disks)) {
4096 struct md_rdev *rdev0
4097 = list_entry(mddev->disks.next,
4098 struct md_rdev, same_set);
4099 err = super_types[mddev->major_version]
4100 .load_super(rdev, rdev0, mddev->minor_version);
4104 } else if (mddev->external)
4105 rdev = md_import_device(dev, -2, -1);
4107 rdev = md_import_device(dev, -1, -1);
4110 mddev_unlock(mddev);
4111 return PTR_ERR(rdev);
4113 err = bind_rdev_to_array(rdev, mddev);
4117 mddev_unlock(mddev);
4118 return err ? err : len;
4121 static struct md_sysfs_entry md_new_device =
4122 __ATTR(new_dev, S_IWUSR, null_show, new_dev_store);
4125 bitmap_store(struct mddev *mddev, const char *buf, size_t len)
4128 unsigned long chunk, end_chunk;
4131 err = mddev_lock(mddev);
4136 /* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */
4138 chunk = end_chunk = simple_strtoul(buf, &end, 0);
4139 if (buf == end) break;
4140 if (*end == '-') { /* range */
4142 end_chunk = simple_strtoul(buf, &end, 0);
4143 if (buf == end) break;
4145 if (*end && !isspace(*end)) break;
4146 bitmap_dirty_bits(mddev->bitmap, chunk, end_chunk);
4147 buf = skip_spaces(end);
4149 bitmap_unplug(mddev->bitmap); /* flush the bits to disk */
4151 mddev_unlock(mddev);
4155 static struct md_sysfs_entry md_bitmap =
4156 __ATTR(bitmap_set_bits, S_IWUSR, null_show, bitmap_store);
4159 size_show(struct mddev *mddev, char *page)
4161 return sprintf(page, "%llu\n",
4162 (unsigned long long)mddev->dev_sectors / 2);
4165 static int update_size(struct mddev *mddev, sector_t num_sectors);
4168 size_store(struct mddev *mddev, const char *buf, size_t len)
4170 /* If array is inactive, we can reduce the component size, but
4171 * not increase it (except from 0).
4172 * If array is active, we can try an on-line resize
4175 int err = strict_blocks_to_sectors(buf, §ors);
4179 err = mddev_lock(mddev);
4183 err = update_size(mddev, sectors);
4184 md_update_sb(mddev, 1);
4186 if (mddev->dev_sectors == 0 ||
4187 mddev->dev_sectors > sectors)
4188 mddev->dev_sectors = sectors;
4192 mddev_unlock(mddev);
4193 return err ? err : len;
4196 static struct md_sysfs_entry md_size =
4197 __ATTR(component_size, S_IRUGO|S_IWUSR, size_show, size_store);
4199 /* Metadata version.
4201 * 'none' for arrays with no metadata (good luck...)
4202 * 'external' for arrays with externally managed metadata,
4203 * or N.M for internally known formats
4206 metadata_show(struct mddev *mddev, char *page)
4208 if (mddev->persistent)
4209 return sprintf(page, "%d.%d\n",
4210 mddev->major_version, mddev->minor_version);
4211 else if (mddev->external)
4212 return sprintf(page, "external:%s\n", mddev->metadata_type);
4214 return sprintf(page, "none\n");
4218 metadata_store(struct mddev *mddev, const char *buf, size_t len)
4223 /* Changing the details of 'external' metadata is
4224 * always permitted. Otherwise there must be
4225 * no devices attached to the array.
4228 err = mddev_lock(mddev);
4232 if (mddev->external && strncmp(buf, "external:", 9) == 0)
4234 else if (!list_empty(&mddev->disks))
4238 if (cmd_match(buf, "none")) {
4239 mddev->persistent = 0;
4240 mddev->external = 0;
4241 mddev->major_version = 0;
4242 mddev->minor_version = 90;
4245 if (strncmp(buf, "external:", 9) == 0) {
4246 size_t namelen = len-9;
4247 if (namelen >= sizeof(mddev->metadata_type))
4248 namelen = sizeof(mddev->metadata_type)-1;
4249 strncpy(mddev->metadata_type, buf+9, namelen);
4250 mddev->metadata_type[namelen] = 0;
4251 if (namelen && mddev->metadata_type[namelen-1] == '\n')
4252 mddev->metadata_type[--namelen] = 0;
4253 mddev->persistent = 0;
4254 mddev->external = 1;
4255 mddev->major_version = 0;
4256 mddev->minor_version = 90;
4259 major = simple_strtoul(buf, &e, 10);
4261 if (e==buf || *e != '.')
4264 minor = simple_strtoul(buf, &e, 10);
4265 if (e==buf || (*e && *e != '\n') )
4268 if (major >= ARRAY_SIZE(super_types) || super_types[major].name == NULL)
4270 mddev->major_version = major;
4271 mddev->minor_version = minor;
4272 mddev->persistent = 1;
4273 mddev->external = 0;
4276 mddev_unlock(mddev);
4280 static struct md_sysfs_entry md_metadata =
4281 __ATTR_PREALLOC(metadata_version, S_IRUGO|S_IWUSR, metadata_show, metadata_store);
4284 action_show(struct mddev *mddev, char *page)
4286 char *type = "idle";
4287 unsigned long recovery = mddev->recovery;
4288 if (test_bit(MD_RECOVERY_FROZEN, &recovery))
4290 else if (test_bit(MD_RECOVERY_RUNNING, &recovery) ||
4291 (!mddev->ro && test_bit(MD_RECOVERY_NEEDED, &recovery))) {
4292 if (test_bit(MD_RECOVERY_RESHAPE, &recovery))
4294 else if (test_bit(MD_RECOVERY_SYNC, &recovery)) {
4295 if (!test_bit(MD_RECOVERY_REQUESTED, &recovery))
4297 else if (test_bit(MD_RECOVERY_CHECK, &recovery))
4301 } else if (test_bit(MD_RECOVERY_RECOVER, &recovery))
4303 else if (mddev->reshape_position != MaxSector)
4306 return sprintf(page, "%s\n", type);
4310 action_store(struct mddev *mddev, const char *page, size_t len)
4312 if (!mddev->pers || !mddev->pers->sync_request)
4316 if (cmd_match(page, "idle") || cmd_match(page, "frozen")) {
4317 if (cmd_match(page, "frozen"))
4318 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4320 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4321 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
4322 mddev_lock(mddev) == 0) {
4323 flush_workqueue(md_misc_wq);
4324 if (mddev->sync_thread) {
4325 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
4326 md_reap_sync_thread(mddev);
4328 mddev_unlock(mddev);
4330 } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4332 else if (cmd_match(page, "resync"))
4333 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4334 else if (cmd_match(page, "recover")) {
4335 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4336 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
4337 } else if (cmd_match(page, "reshape")) {
4339 if (mddev->pers->start_reshape == NULL)
4341 err = mddev_lock(mddev);
4343 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4346 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4347 err = mddev->pers->start_reshape(mddev);
4349 mddev_unlock(mddev);
4353 sysfs_notify(&mddev->kobj, NULL, "degraded");
4355 if (cmd_match(page, "check"))
4356 set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
4357 else if (!cmd_match(page, "repair"))
4359 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4360 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
4361 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
4363 if (mddev->ro == 2) {
4364 /* A write to sync_action is enough to justify
4365 * canceling read-auto mode
4368 md_wakeup_thread(mddev->sync_thread);
4370 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4371 md_wakeup_thread(mddev->thread);
4372 sysfs_notify_dirent_safe(mddev->sysfs_action);
4376 static struct md_sysfs_entry md_scan_mode =
4377 __ATTR_PREALLOC(sync_action, S_IRUGO|S_IWUSR, action_show, action_store);
4380 last_sync_action_show(struct mddev *mddev, char *page)
4382 return sprintf(page, "%s\n", mddev->last_sync_action);
4385 static struct md_sysfs_entry md_last_scan_mode = __ATTR_RO(last_sync_action);
4388 mismatch_cnt_show(struct mddev *mddev, char *page)
4390 return sprintf(page, "%llu\n",
4391 (unsigned long long)
4392 atomic64_read(&mddev->resync_mismatches));
4395 static struct md_sysfs_entry md_mismatches = __ATTR_RO(mismatch_cnt);
4398 sync_min_show(struct mddev *mddev, char *page)
4400 return sprintf(page, "%d (%s)\n", speed_min(mddev),
4401 mddev->sync_speed_min ? "local": "system");
4405 sync_min_store(struct mddev *mddev, const char *buf, size_t len)
4410 if (strncmp(buf, "system", 6)==0) {
4413 rv = kstrtouint(buf, 10, &min);
4419 mddev->sync_speed_min = min;
4423 static struct md_sysfs_entry md_sync_min =
4424 __ATTR(sync_speed_min, S_IRUGO|S_IWUSR, sync_min_show, sync_min_store);
4427 sync_max_show(struct mddev *mddev, char *page)
4429 return sprintf(page, "%d (%s)\n", speed_max(mddev),
4430 mddev->sync_speed_max ? "local": "system");
4434 sync_max_store(struct mddev *mddev, const char *buf, size_t len)
4439 if (strncmp(buf, "system", 6)==0) {
4442 rv = kstrtouint(buf, 10, &max);
4448 mddev->sync_speed_max = max;
4452 static struct md_sysfs_entry md_sync_max =
4453 __ATTR(sync_speed_max, S_IRUGO|S_IWUSR, sync_max_show, sync_max_store);
4456 degraded_show(struct mddev *mddev, char *page)
4458 return sprintf(page, "%d\n", mddev->degraded);
4460 static struct md_sysfs_entry md_degraded = __ATTR_RO(degraded);
4463 sync_force_parallel_show(struct mddev *mddev, char *page)
4465 return sprintf(page, "%d\n", mddev->parallel_resync);
4469 sync_force_parallel_store(struct mddev *mddev, const char *buf, size_t len)
4473 if (kstrtol(buf, 10, &n))
4476 if (n != 0 && n != 1)
4479 mddev->parallel_resync = n;
4481 if (mddev->sync_thread)
4482 wake_up(&resync_wait);
4487 /* force parallel resync, even with shared block devices */
4488 static struct md_sysfs_entry md_sync_force_parallel =
4489 __ATTR(sync_force_parallel, S_IRUGO|S_IWUSR,
4490 sync_force_parallel_show, sync_force_parallel_store);
4493 sync_speed_show(struct mddev *mddev, char *page)
4495 unsigned long resync, dt, db;
4496 if (mddev->curr_resync == 0)
4497 return sprintf(page, "none\n");
4498 resync = mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active);
4499 dt = (jiffies - mddev->resync_mark) / HZ;
4501 db = resync - mddev->resync_mark_cnt;
4502 return sprintf(page, "%lu\n", db/dt/2); /* K/sec */
4505 static struct md_sysfs_entry md_sync_speed = __ATTR_RO(sync_speed);
4508 sync_completed_show(struct mddev *mddev, char *page)
4510 unsigned long long max_sectors, resync;
4512 if (!test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4513 return sprintf(page, "none\n");
4515 if (mddev->curr_resync == 1 ||
4516 mddev->curr_resync == 2)
4517 return sprintf(page, "delayed\n");
4519 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ||
4520 test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
4521 max_sectors = mddev->resync_max_sectors;
4523 max_sectors = mddev->dev_sectors;
4525 resync = mddev->curr_resync_completed;
4526 return sprintf(page, "%llu / %llu\n", resync, max_sectors);
4529 static struct md_sysfs_entry md_sync_completed =
4530 __ATTR_PREALLOC(sync_completed, S_IRUGO, sync_completed_show, NULL);
4533 min_sync_show(struct mddev *mddev, char *page)
4535 return sprintf(page, "%llu\n",
4536 (unsigned long long)mddev->resync_min);
4539 min_sync_store(struct mddev *mddev, const char *buf, size_t len)
4541 unsigned long long min;
4544 if (kstrtoull(buf, 10, &min))
4547 spin_lock(&mddev->lock);
4549 if (min > mddev->resync_max)
4553 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4556 /* Round down to multiple of 4K for safety */
4557 mddev->resync_min = round_down(min, 8);
4561 spin_unlock(&mddev->lock);
4565 static struct md_sysfs_entry md_min_sync =
4566 __ATTR(sync_min, S_IRUGO|S_IWUSR, min_sync_show, min_sync_store);
4569 max_sync_show(struct mddev *mddev, char *page)
4571 if (mddev->resync_max == MaxSector)
4572 return sprintf(page, "max\n");
4574 return sprintf(page, "%llu\n",
4575 (unsigned long long)mddev->resync_max);
4578 max_sync_store(struct mddev *mddev, const char *buf, size_t len)
4581 spin_lock(&mddev->lock);
4582 if (strncmp(buf, "max", 3) == 0)
4583 mddev->resync_max = MaxSector;
4585 unsigned long long max;
4589 if (kstrtoull(buf, 10, &max))
4591 if (max < mddev->resync_min)
4595 if (max < mddev->resync_max &&
4597 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4600 /* Must be a multiple of chunk_size */
4601 chunk = mddev->chunk_sectors;
4603 sector_t temp = max;
4606 if (sector_div(temp, chunk))
4609 mddev->resync_max = max;
4611 wake_up(&mddev->recovery_wait);
4614 spin_unlock(&mddev->lock);
4618 static struct md_sysfs_entry md_max_sync =
4619 __ATTR(sync_max, S_IRUGO|S_IWUSR, max_sync_show, max_sync_store);
4622 suspend_lo_show(struct mddev *mddev, char *page)
4624 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_lo);
4628 suspend_lo_store(struct mddev *mddev, const char *buf, size_t len)
4630 unsigned long long old, new;
4633 err = kstrtoull(buf, 10, &new);
4636 if (new != (sector_t)new)
4639 err = mddev_lock(mddev);
4643 if (mddev->pers == NULL ||
4644 mddev->pers->quiesce == NULL)
4646 old = mddev->suspend_lo;
4647 mddev->suspend_lo = new;
4649 /* Shrinking suspended region */
4650 mddev->pers->quiesce(mddev, 2);
4652 /* Expanding suspended region - need to wait */
4653 mddev->pers->quiesce(mddev, 1);
4654 mddev->pers->quiesce(mddev, 0);
4658 mddev_unlock(mddev);
4661 static struct md_sysfs_entry md_suspend_lo =
4662 __ATTR(suspend_lo, S_IRUGO|S_IWUSR, suspend_lo_show, suspend_lo_store);
4665 suspend_hi_show(struct mddev *mddev, char *page)
4667 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_hi);
4671 suspend_hi_store(struct mddev *mddev, const char *buf, size_t len)
4673 unsigned long long old, new;
4676 err = kstrtoull(buf, 10, &new);
4679 if (new != (sector_t)new)
4682 err = mddev_lock(mddev);
4686 if (mddev->pers == NULL ||
4687 mddev->pers->quiesce == NULL)
4689 old = mddev->suspend_hi;
4690 mddev->suspend_hi = new;
4692 /* Shrinking suspended region */
4693 mddev->pers->quiesce(mddev, 2);
4695 /* Expanding suspended region - need to wait */
4696 mddev->pers->quiesce(mddev, 1);
4697 mddev->pers->quiesce(mddev, 0);
4701 mddev_unlock(mddev);
4704 static struct md_sysfs_entry md_suspend_hi =
4705 __ATTR(suspend_hi, S_IRUGO|S_IWUSR, suspend_hi_show, suspend_hi_store);
4708 reshape_position_show(struct mddev *mddev, char *page)
4710 if (mddev->reshape_position != MaxSector)
4711 return sprintf(page, "%llu\n",
4712 (unsigned long long)mddev->reshape_position);
4713 strcpy(page, "none\n");
4718 reshape_position_store(struct mddev *mddev, const char *buf, size_t len)
4720 struct md_rdev *rdev;
4721 unsigned long long new;
4724 err = kstrtoull(buf, 10, &new);
4727 if (new != (sector_t)new)
4729 err = mddev_lock(mddev);
4735 mddev->reshape_position = new;
4736 mddev->delta_disks = 0;
4737 mddev->reshape_backwards = 0;
4738 mddev->new_level = mddev->level;
4739 mddev->new_layout = mddev->layout;
4740 mddev->new_chunk_sectors = mddev->chunk_sectors;
4741 rdev_for_each(rdev, mddev)
4742 rdev->new_data_offset = rdev->data_offset;
4745 mddev_unlock(mddev);
4749 static struct md_sysfs_entry md_reshape_position =
4750 __ATTR(reshape_position, S_IRUGO|S_IWUSR, reshape_position_show,
4751 reshape_position_store);
4754 reshape_direction_show(struct mddev *mddev, char *page)
4756 return sprintf(page, "%s\n",
4757 mddev->reshape_backwards ? "backwards" : "forwards");
4761 reshape_direction_store(struct mddev *mddev, const char *buf, size_t len)
4766 if (cmd_match(buf, "forwards"))
4768 else if (cmd_match(buf, "backwards"))
4772 if (mddev->reshape_backwards == backwards)
4775 err = mddev_lock(mddev);
4778 /* check if we are allowed to change */
4779 if (mddev->delta_disks)
4781 else if (mddev->persistent &&
4782 mddev->major_version == 0)
4785 mddev->reshape_backwards = backwards;
4786 mddev_unlock(mddev);
4790 static struct md_sysfs_entry md_reshape_direction =
4791 __ATTR(reshape_direction, S_IRUGO|S_IWUSR, reshape_direction_show,
4792 reshape_direction_store);
4795 array_size_show(struct mddev *mddev, char *page)
4797 if (mddev->external_size)
4798 return sprintf(page, "%llu\n",
4799 (unsigned long long)mddev->array_sectors/2);
4801 return sprintf(page, "default\n");
4805 array_size_store(struct mddev *mddev, const char *buf, size_t len)
4810 err = mddev_lock(mddev);
4814 if (strncmp(buf, "default", 7) == 0) {
4816 sectors = mddev->pers->size(mddev, 0, 0);
4818 sectors = mddev->array_sectors;
4820 mddev->external_size = 0;
4822 if (strict_blocks_to_sectors(buf, §ors) < 0)
4824 else if (mddev->pers && mddev->pers->size(mddev, 0, 0) < sectors)
4827 mddev->external_size = 1;
4831 mddev->array_sectors = sectors;
4833 set_capacity(mddev->gendisk, mddev->array_sectors);
4834 revalidate_disk(mddev->gendisk);
4837 mddev_unlock(mddev);
4841 static struct md_sysfs_entry md_array_size =
4842 __ATTR(array_size, S_IRUGO|S_IWUSR, array_size_show,
4845 static struct attribute *md_default_attrs[] = {
4848 &md_raid_disks.attr,
4849 &md_chunk_size.attr,
4851 &md_resync_start.attr,
4853 &md_new_device.attr,
4854 &md_safe_delay.attr,
4855 &md_array_state.attr,
4856 &md_reshape_position.attr,
4857 &md_reshape_direction.attr,
4858 &md_array_size.attr,
4859 &max_corr_read_errors.attr,
4863 static struct attribute *md_redundancy_attrs[] = {
4865 &md_last_scan_mode.attr,
4866 &md_mismatches.attr,
4869 &md_sync_speed.attr,
4870 &md_sync_force_parallel.attr,
4871 &md_sync_completed.attr,
4874 &md_suspend_lo.attr,
4875 &md_suspend_hi.attr,
4880 static struct attribute_group md_redundancy_group = {
4882 .attrs = md_redundancy_attrs,
4886 md_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
4888 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
4889 struct mddev *mddev = container_of(kobj, struct mddev, kobj);
4894 spin_lock(&all_mddevs_lock);
4895 if (list_empty(&mddev->all_mddevs)) {
4896 spin_unlock(&all_mddevs_lock);
4900 spin_unlock(&all_mddevs_lock);
4902 rv = entry->show(mddev, page);
4908 md_attr_store(struct kobject *kobj, struct attribute *attr,
4909 const char *page, size_t length)
4911 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
4912 struct mddev *mddev = container_of(kobj, struct mddev, kobj);
4917 if (!capable(CAP_SYS_ADMIN))
4919 spin_lock(&all_mddevs_lock);
4920 if (list_empty(&mddev->all_mddevs)) {
4921 spin_unlock(&all_mddevs_lock);
4925 spin_unlock(&all_mddevs_lock);
4926 rv = entry->store(mddev, page, length);
4931 static void md_free(struct kobject *ko)
4933 struct mddev *mddev = container_of(ko, struct mddev, kobj);
4935 if (mddev->sysfs_state)
4936 sysfs_put(mddev->sysfs_state);
4939 blk_cleanup_queue(mddev->queue);
4940 if (mddev->gendisk) {
4941 del_gendisk(mddev->gendisk);
4942 put_disk(mddev->gendisk);
4948 static const struct sysfs_ops md_sysfs_ops = {
4949 .show = md_attr_show,
4950 .store = md_attr_store,
4952 static struct kobj_type md_ktype = {
4954 .sysfs_ops = &md_sysfs_ops,
4955 .default_attrs = md_default_attrs,
4960 static void mddev_delayed_delete(struct work_struct *ws)
4962 struct mddev *mddev = container_of(ws, struct mddev, del_work);
4964 sysfs_remove_group(&mddev->kobj, &md_bitmap_group);
4965 kobject_del(&mddev->kobj);
4966 kobject_put(&mddev->kobj);
4969 static int md_alloc(dev_t dev, char *name)
4971 static DEFINE_MUTEX(disks_mutex);
4972 struct mddev *mddev = mddev_find(dev);
4973 struct gendisk *disk;
4982 partitioned = (MAJOR(mddev->unit) != MD_MAJOR);
4983 shift = partitioned ? MdpMinorShift : 0;
4984 unit = MINOR(mddev->unit) >> shift;
4986 /* wait for any previous instance of this device to be
4987 * completely removed (mddev_delayed_delete).
4989 flush_workqueue(md_misc_wq);
4991 mutex_lock(&disks_mutex);
4997 /* Need to ensure that 'name' is not a duplicate.
4999 struct mddev *mddev2;
5000 spin_lock(&all_mddevs_lock);
5002 list_for_each_entry(mddev2, &all_mddevs, all_mddevs)
5003 if (mddev2->gendisk &&
5004 strcmp(mddev2->gendisk->disk_name, name) == 0) {
5005 spin_unlock(&all_mddevs_lock);
5008 spin_unlock(&all_mddevs_lock);
5012 mddev->queue = blk_alloc_queue(GFP_KERNEL);
5015 mddev->queue->queuedata = mddev;
5017 blk_queue_make_request(mddev->queue, md_make_request);
5018 blk_set_stacking_limits(&mddev->queue->limits);
5020 disk = alloc_disk(1 << shift);
5022 blk_cleanup_queue(mddev->queue);
5023 mddev->queue = NULL;
5026 disk->major = MAJOR(mddev->unit);
5027 disk->first_minor = unit << shift;
5029 strcpy(disk->disk_name, name);
5030 else if (partitioned)
5031 sprintf(disk->disk_name, "md_d%d", unit);
5033 sprintf(disk->disk_name, "md%d", unit);
5034 disk->fops = &md_fops;
5035 disk->private_data = mddev;
5036 disk->queue = mddev->queue;
5037 blk_queue_flush(mddev->queue, REQ_FLUSH | REQ_FUA);
5038 /* Allow extended partitions. This makes the
5039 * 'mdp' device redundant, but we can't really
5042 disk->flags |= GENHD_FL_EXT_DEVT;
5043 mddev->gendisk = disk;
5044 /* As soon as we call add_disk(), another thread could get
5045 * through to md_open, so make sure it doesn't get too far
5047 mutex_lock(&mddev->open_mutex);
5050 error = kobject_init_and_add(&mddev->kobj, &md_ktype,
5051 &disk_to_dev(disk)->kobj, "%s", "md");
5053 /* This isn't possible, but as kobject_init_and_add is marked
5054 * __must_check, we must do something with the result
5056 printk(KERN_WARNING "md: cannot register %s/md - name in use\n",
5060 if (mddev->kobj.sd &&
5061 sysfs_create_group(&mddev->kobj, &md_bitmap_group))
5062 printk(KERN_DEBUG "pointless warning\n");
5063 mutex_unlock(&mddev->open_mutex);
5065 mutex_unlock(&disks_mutex);
5066 if (!error && mddev->kobj.sd) {
5067 kobject_uevent(&mddev->kobj, KOBJ_ADD);
5068 mddev->sysfs_state = sysfs_get_dirent_safe(mddev->kobj.sd, "array_state");
5074 static struct kobject *md_probe(dev_t dev, int *part, void *data)
5076 md_alloc(dev, NULL);
5080 static int add_named_array(const char *val, struct kernel_param *kp)
5082 /* val must be "md_*" where * is not all digits.
5083 * We allocate an array with a large free minor number, and
5084 * set the name to val. val must not already be an active name.
5086 int len = strlen(val);
5087 char buf[DISK_NAME_LEN];
5089 while (len && val[len-1] == '\n')
5091 if (len >= DISK_NAME_LEN)
5093 strlcpy(buf, val, len+1);
5094 if (strncmp(buf, "md_", 3) != 0)
5096 return md_alloc(0, buf);
5099 static void md_safemode_timeout(unsigned long data)
5101 struct mddev *mddev = (struct mddev *) data;
5103 if (!atomic_read(&mddev->writes_pending)) {
5104 mddev->safemode = 1;
5105 if (mddev->external)
5106 sysfs_notify_dirent_safe(mddev->sysfs_state);
5108 md_wakeup_thread(mddev->thread);
5111 static int start_dirty_degraded;
5113 int md_run(struct mddev *mddev)
5116 struct md_rdev *rdev;
5117 struct md_personality *pers;
5119 if (list_empty(&mddev->disks))
5120 /* cannot run an array with no devices.. */
5125 /* Cannot run until previous stop completes properly */
5126 if (mddev->sysfs_active)
5130 * Analyze all RAID superblock(s)
5132 if (!mddev->raid_disks) {
5133 if (!mddev->persistent)
5138 if (mddev->level != LEVEL_NONE)
5139 request_module("md-level-%d", mddev->level);
5140 else if (mddev->clevel[0])
5141 request_module("md-%s", mddev->clevel);
5144 * Drop all container device buffers, from now on
5145 * the only valid external interface is through the md
5148 rdev_for_each(rdev, mddev) {
5149 if (test_bit(Faulty, &rdev->flags))
5151 sync_blockdev(rdev->bdev);
5152 invalidate_bdev(rdev->bdev);
5154 /* perform some consistency tests on the device.
5155 * We don't want the data to overlap the metadata,
5156 * Internal Bitmap issues have been handled elsewhere.
5158 if (rdev->meta_bdev) {
5159 /* Nothing to check */;
5160 } else if (rdev->data_offset < rdev->sb_start) {
5161 if (mddev->dev_sectors &&
5162 rdev->data_offset + mddev->dev_sectors
5164 printk("md: %s: data overlaps metadata\n",
5169 if (rdev->sb_start + rdev->sb_size/512
5170 > rdev->data_offset) {
5171 printk("md: %s: metadata overlaps data\n",
5176 sysfs_notify_dirent_safe(rdev->sysfs_state);
5179 if (mddev->bio_set == NULL)
5180 mddev->bio_set = bioset_create(BIO_POOL_SIZE, 0);
5182 spin_lock(&pers_lock);
5183 pers = find_pers(mddev->level, mddev->clevel);
5184 if (!pers || !try_module_get(pers->owner)) {
5185 spin_unlock(&pers_lock);
5186 if (mddev->level != LEVEL_NONE)
5187 printk(KERN_WARNING "md: personality for level %d is not loaded!\n",
5190 printk(KERN_WARNING "md: personality for level %s is not loaded!\n",
5194 spin_unlock(&pers_lock);
5195 if (mddev->level != pers->level) {
5196 mddev->level = pers->level;
5197 mddev->new_level = pers->level;
5199 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
5201 if (mddev->reshape_position != MaxSector &&
5202 pers->start_reshape == NULL) {
5203 /* This personality cannot handle reshaping... */
5204 module_put(pers->owner);
5208 if (pers->sync_request) {
5209 /* Warn if this is a potentially silly
5212 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
5213 struct md_rdev *rdev2;
5216 rdev_for_each(rdev, mddev)
5217 rdev_for_each(rdev2, mddev) {
5219 rdev->bdev->bd_contains ==
5220 rdev2->bdev->bd_contains) {
5222 "%s: WARNING: %s appears to be"
5223 " on the same physical disk as"
5226 bdevname(rdev->bdev,b),
5227 bdevname(rdev2->bdev,b2));
5234 "True protection against single-disk"
5235 " failure might be compromised.\n");
5238 mddev->recovery = 0;
5239 /* may be over-ridden by personality */
5240 mddev->resync_max_sectors = mddev->dev_sectors;
5242 mddev->ok_start_degraded = start_dirty_degraded;
5244 if (start_readonly && mddev->ro == 0)
5245 mddev->ro = 2; /* read-only, but switch on first write */
5247 err = pers->run(mddev);
5249 printk(KERN_ERR "md: pers->run() failed ...\n");
5250 else if (pers->size(mddev, 0, 0) < mddev->array_sectors) {
5251 WARN_ONCE(!mddev->external_size, "%s: default size too small,"
5252 " but 'external_size' not in effect?\n", __func__);
5254 "md: invalid array_size %llu > default size %llu\n",
5255 (unsigned long long)mddev->array_sectors / 2,
5256 (unsigned long long)pers->size(mddev, 0, 0) / 2);
5259 if (err == 0 && pers->sync_request &&
5260 (mddev->bitmap_info.file || mddev->bitmap_info.offset)) {
5261 struct bitmap *bitmap;
5263 bitmap = bitmap_create(mddev, -1);
5264 if (IS_ERR(bitmap)) {
5265 err = PTR_ERR(bitmap);
5266 printk(KERN_ERR "%s: failed to create bitmap (%d)\n",
5267 mdname(mddev), err);
5269 mddev->bitmap = bitmap;
5273 mddev_detach(mddev);
5275 pers->free(mddev, mddev->private);
5276 mddev->private = NULL;
5277 module_put(pers->owner);
5278 bitmap_destroy(mddev);
5282 mddev->queue->backing_dev_info.congested_data = mddev;
5283 mddev->queue->backing_dev_info.congested_fn = md_congested;
5285 if (pers->sync_request) {
5286 if (mddev->kobj.sd &&
5287 sysfs_create_group(&mddev->kobj, &md_redundancy_group))
5289 "md: cannot register extra attributes for %s\n",
5291 mddev->sysfs_action = sysfs_get_dirent_safe(mddev->kobj.sd, "sync_action");
5292 } else if (mddev->ro == 2) /* auto-readonly not meaningful */
5295 atomic_set(&mddev->writes_pending,0);
5296 atomic_set(&mddev->max_corr_read_errors,
5297 MD_DEFAULT_MAX_CORRECTED_READ_ERRORS);
5298 mddev->safemode = 0;
5299 if (mddev_is_clustered(mddev))
5300 mddev->safemode_delay = 0;
5302 mddev->safemode_delay = (200 * HZ)/1000 +1; /* 200 msec delay */
5305 spin_lock(&mddev->lock);
5307 spin_unlock(&mddev->lock);
5308 rdev_for_each(rdev, mddev)
5309 if (rdev->raid_disk >= 0)
5310 if (sysfs_link_rdev(mddev, rdev))
5311 /* failure here is OK */;
5313 if (mddev->degraded && !mddev->ro)
5314 /* This ensures that recovering status is reported immediately
5315 * via sysfs - until a lack of spares is confirmed.
5317 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
5318 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5320 if (mddev->flags & MD_UPDATE_SB_FLAGS)
5321 md_update_sb(mddev, 0);
5323 md_new_event(mddev);
5324 sysfs_notify_dirent_safe(mddev->sysfs_state);
5325 sysfs_notify_dirent_safe(mddev->sysfs_action);
5326 sysfs_notify(&mddev->kobj, NULL, "degraded");
5329 EXPORT_SYMBOL_GPL(md_run);
5331 static int do_md_run(struct mddev *mddev)
5335 err = md_run(mddev);
5338 err = bitmap_load(mddev);
5340 bitmap_destroy(mddev);
5344 if (mddev_is_clustered(mddev))
5345 md_allow_write(mddev);
5347 md_wakeup_thread(mddev->thread);
5348 md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
5350 set_capacity(mddev->gendisk, mddev->array_sectors);
5351 revalidate_disk(mddev->gendisk);
5353 kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
5358 static int restart_array(struct mddev *mddev)
5360 struct gendisk *disk = mddev->gendisk;
5362 /* Complain if it has no devices */
5363 if (list_empty(&mddev->disks))
5369 if (test_bit(MD_HAS_JOURNAL, &mddev->flags)) {
5370 struct md_rdev *rdev;
5371 bool has_journal = false;
5374 rdev_for_each_rcu(rdev, mddev) {
5375 if (test_bit(Journal, &rdev->flags) &&
5376 !test_bit(Faulty, &rdev->flags)) {
5383 /* Don't restart rw with journal missing/faulty */
5388 mddev->safemode = 0;
5390 set_disk_ro(disk, 0);
5391 printk(KERN_INFO "md: %s switched to read-write mode.\n",
5393 /* Kick recovery or resync if necessary */
5394 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5395 md_wakeup_thread(mddev->thread);
5396 md_wakeup_thread(mddev->sync_thread);
5397 sysfs_notify_dirent_safe(mddev->sysfs_state);
5401 static void md_clean(struct mddev *mddev)
5403 mddev->array_sectors = 0;
5404 mddev->external_size = 0;
5405 mddev->dev_sectors = 0;
5406 mddev->raid_disks = 0;
5407 mddev->recovery_cp = 0;
5408 mddev->resync_min = 0;
5409 mddev->resync_max = MaxSector;
5410 mddev->reshape_position = MaxSector;
5411 mddev->external = 0;
5412 mddev->persistent = 0;
5413 mddev->level = LEVEL_NONE;
5414 mddev->clevel[0] = 0;
5417 mddev->metadata_type[0] = 0;
5418 mddev->chunk_sectors = 0;
5419 mddev->ctime = mddev->utime = 0;
5421 mddev->max_disks = 0;
5423 mddev->can_decrease_events = 0;
5424 mddev->delta_disks = 0;
5425 mddev->reshape_backwards = 0;
5426 mddev->new_level = LEVEL_NONE;
5427 mddev->new_layout = 0;
5428 mddev->new_chunk_sectors = 0;
5429 mddev->curr_resync = 0;
5430 atomic64_set(&mddev->resync_mismatches, 0);
5431 mddev->suspend_lo = mddev->suspend_hi = 0;
5432 mddev->sync_speed_min = mddev->sync_speed_max = 0;
5433 mddev->recovery = 0;
5436 mddev->degraded = 0;
5437 mddev->safemode = 0;
5438 mddev->private = NULL;
5439 mddev->bitmap_info.offset = 0;
5440 mddev->bitmap_info.default_offset = 0;
5441 mddev->bitmap_info.default_space = 0;
5442 mddev->bitmap_info.chunksize = 0;
5443 mddev->bitmap_info.daemon_sleep = 0;
5444 mddev->bitmap_info.max_write_behind = 0;
5447 static void __md_stop_writes(struct mddev *mddev)
5449 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5450 flush_workqueue(md_misc_wq);
5451 if (mddev->sync_thread) {
5452 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5453 md_reap_sync_thread(mddev);
5456 del_timer_sync(&mddev->safemode_timer);
5458 bitmap_flush(mddev);
5459 md_super_wait(mddev);
5461 if (mddev->ro == 0 &&
5462 ((!mddev->in_sync && !mddev_is_clustered(mddev)) ||
5463 (mddev->flags & MD_UPDATE_SB_FLAGS))) {
5464 /* mark array as shutdown cleanly */
5465 if (!mddev_is_clustered(mddev))
5467 md_update_sb(mddev, 1);
5471 void md_stop_writes(struct mddev *mddev)
5473 mddev_lock_nointr(mddev);
5474 __md_stop_writes(mddev);
5475 mddev_unlock(mddev);
5477 EXPORT_SYMBOL_GPL(md_stop_writes);
5479 static void mddev_detach(struct mddev *mddev)
5481 struct bitmap *bitmap = mddev->bitmap;
5482 /* wait for behind writes to complete */
5483 if (bitmap && atomic_read(&bitmap->behind_writes) > 0) {
5484 printk(KERN_INFO "md:%s: behind writes in progress - waiting to stop.\n",
5486 /* need to kick something here to make sure I/O goes? */
5487 wait_event(bitmap->behind_wait,
5488 atomic_read(&bitmap->behind_writes) == 0);
5490 if (mddev->pers && mddev->pers->quiesce) {
5491 mddev->pers->quiesce(mddev, 1);
5492 mddev->pers->quiesce(mddev, 0);
5494 md_unregister_thread(&mddev->thread);
5496 blk_sync_queue(mddev->queue); /* the unplug fn references 'conf'*/
5499 static void __md_stop(struct mddev *mddev)
5501 struct md_personality *pers = mddev->pers;
5502 mddev_detach(mddev);
5503 /* Ensure ->event_work is done */
5504 flush_workqueue(md_misc_wq);
5505 spin_lock(&mddev->lock);
5507 spin_unlock(&mddev->lock);
5508 pers->free(mddev, mddev->private);
5509 mddev->private = NULL;
5510 if (pers->sync_request && mddev->to_remove == NULL)
5511 mddev->to_remove = &md_redundancy_group;
5512 module_put(pers->owner);
5513 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5516 void md_stop(struct mddev *mddev)
5518 /* stop the array and free an attached data structures.
5519 * This is called from dm-raid
5522 bitmap_destroy(mddev);
5524 bioset_free(mddev->bio_set);
5527 EXPORT_SYMBOL_GPL(md_stop);
5529 static int md_set_readonly(struct mddev *mddev, struct block_device *bdev)
5534 if (!test_bit(MD_RECOVERY_FROZEN, &mddev->recovery)) {
5536 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5537 md_wakeup_thread(mddev->thread);
5539 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
5540 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5541 if (mddev->sync_thread)
5542 /* Thread might be blocked waiting for metadata update
5543 * which will now never happen */
5544 wake_up_process(mddev->sync_thread->tsk);
5546 if (mddev->external && test_bit(MD_CHANGE_PENDING, &mddev->flags))
5548 mddev_unlock(mddev);
5549 wait_event(resync_wait, !test_bit(MD_RECOVERY_RUNNING,
5551 wait_event(mddev->sb_wait,
5552 !test_bit(MD_CHANGE_PENDING, &mddev->flags));
5553 mddev_lock_nointr(mddev);
5555 mutex_lock(&mddev->open_mutex);
5556 if ((mddev->pers && atomic_read(&mddev->openers) > !!bdev) ||
5557 mddev->sync_thread ||
5558 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
5559 (bdev && !test_bit(MD_STILL_CLOSED, &mddev->flags))) {
5560 printk("md: %s still in use.\n",mdname(mddev));
5562 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5563 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5564 md_wakeup_thread(mddev->thread);
5570 __md_stop_writes(mddev);
5576 set_disk_ro(mddev->gendisk, 1);
5577 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5578 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5579 md_wakeup_thread(mddev->thread);
5580 sysfs_notify_dirent_safe(mddev->sysfs_state);
5584 mutex_unlock(&mddev->open_mutex);
5589 * 0 - completely stop and dis-assemble array
5590 * 2 - stop but do not disassemble array
5592 static int do_md_stop(struct mddev *mddev, int mode,
5593 struct block_device *bdev)
5595 struct gendisk *disk = mddev->gendisk;
5596 struct md_rdev *rdev;
5599 if (!test_bit(MD_RECOVERY_FROZEN, &mddev->recovery)) {
5601 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5602 md_wakeup_thread(mddev->thread);
5604 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
5605 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5606 if (mddev->sync_thread)
5607 /* Thread might be blocked waiting for metadata update
5608 * which will now never happen */
5609 wake_up_process(mddev->sync_thread->tsk);
5611 mddev_unlock(mddev);
5612 wait_event(resync_wait, (mddev->sync_thread == NULL &&
5613 !test_bit(MD_RECOVERY_RUNNING,
5614 &mddev->recovery)));
5615 mddev_lock_nointr(mddev);
5617 mutex_lock(&mddev->open_mutex);
5618 if ((mddev->pers && atomic_read(&mddev->openers) > !!bdev) ||
5619 mddev->sysfs_active ||
5620 mddev->sync_thread ||
5621 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
5622 (bdev && !test_bit(MD_STILL_CLOSED, &mddev->flags))) {
5623 printk("md: %s still in use.\n",mdname(mddev));
5624 mutex_unlock(&mddev->open_mutex);
5626 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5627 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5628 md_wakeup_thread(mddev->thread);
5634 set_disk_ro(disk, 0);
5636 __md_stop_writes(mddev);
5638 mddev->queue->backing_dev_info.congested_fn = NULL;
5640 /* tell userspace to handle 'inactive' */
5641 sysfs_notify_dirent_safe(mddev->sysfs_state);
5643 rdev_for_each(rdev, mddev)
5644 if (rdev->raid_disk >= 0)
5645 sysfs_unlink_rdev(mddev, rdev);
5647 set_capacity(disk, 0);
5648 mutex_unlock(&mddev->open_mutex);
5650 revalidate_disk(disk);
5655 mutex_unlock(&mddev->open_mutex);
5657 * Free resources if final stop
5660 printk(KERN_INFO "md: %s stopped.\n", mdname(mddev));
5662 bitmap_destroy(mddev);
5663 if (mddev->bitmap_info.file) {
5664 struct file *f = mddev->bitmap_info.file;
5665 spin_lock(&mddev->lock);
5666 mddev->bitmap_info.file = NULL;
5667 spin_unlock(&mddev->lock);
5670 mddev->bitmap_info.offset = 0;
5672 export_array(mddev);
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);