2 md.c : Multiple Devices driver for Linux
3 Copyright (C) 1998, 1999, 2000 Ingo Molnar
5 completely rewritten, based on the MD driver code from Marc Zyngier
9 - RAID-1/RAID-5 extensions by Miguel de Icaza, Gadi Oxman, Ingo Molnar
10 - RAID-6 extensions by H. Peter Anvin <hpa@zytor.com>
11 - boot support for linear and striped mode by Harald Hoyer <HarryH@Royal.Net>
12 - kerneld support by Boris Tobotras <boris@xtalk.msk.su>
13 - kmod support by: Cyrus Durgin
14 - RAID0 bugfixes: Mark Anthony Lisher <markal@iname.com>
15 - Devfs support by Richard Gooch <rgooch@atnf.csiro.au>
17 - lots of fixes and improvements to the RAID1/RAID5 and generic
18 RAID code (such as request based resynchronization):
20 Neil Brown <neilb@cse.unsw.edu.au>.
22 - persistent bitmap code
23 Copyright (C) 2003-2004, Paul Clements, SteelEye Technology, Inc.
25 This program is free software; you can redistribute it and/or modify
26 it under the terms of the GNU General Public License as published by
27 the Free Software Foundation; either version 2, or (at your option)
30 You should have received a copy of the GNU General Public License
31 (for example /usr/src/linux/COPYING); if not, write to the Free
32 Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
35 #include <linux/kthread.h>
36 #include <linux/blkdev.h>
37 #include <linux/sysctl.h>
38 #include <linux/seq_file.h>
39 #include <linux/mutex.h>
40 #include <linux/buffer_head.h> /* for invalidate_bdev */
41 #include <linux/poll.h>
42 #include <linux/ctype.h>
43 #include <linux/string.h>
44 #include <linux/hdreg.h>
45 #include <linux/proc_fs.h>
46 #include <linux/random.h>
47 #include <linux/reboot.h>
48 #include <linux/file.h>
49 #include <linux/compat.h>
50 #include <linux/delay.h>
51 #include <linux/raid/md_p.h>
52 #include <linux/raid/md_u.h>
53 #include <linux/slab.h>
58 #define dprintk(x...) ((void)(DEBUG && printk(x)))
61 static void autostart_arrays(int part);
64 static LIST_HEAD(pers_list);
65 static DEFINE_SPINLOCK(pers_lock);
67 static void md_print_devices(void);
69 static DECLARE_WAIT_QUEUE_HEAD(resync_wait);
70 static struct workqueue_struct *md_wq;
71 static struct workqueue_struct *md_misc_wq;
73 #define MD_BUG(x...) { printk("md: bug in file %s, line %d\n", __FILE__, __LINE__); md_print_devices(); }
76 * Default number of read corrections we'll attempt on an rdev
77 * before ejecting it from the array. We divide the read error
78 * count by 2 for every hour elapsed between read errors.
80 #define MD_DEFAULT_MAX_CORRECTED_READ_ERRORS 20
82 * Current RAID-1,4,5 parallel reconstruction 'guaranteed speed limit'
83 * is 1000 KB/sec, so the extra system load does not show up that much.
84 * Increase it if you want to have more _guaranteed_ speed. Note that
85 * the RAID driver will use the maximum available bandwidth if the IO
86 * subsystem is idle. There is also an 'absolute maximum' reconstruction
87 * speed limit - in case reconstruction slows down your system despite
90 * you can change it via /proc/sys/dev/raid/speed_limit_min and _max.
91 * or /sys/block/mdX/md/sync_speed_{min,max}
94 static int sysctl_speed_limit_min = 1000;
95 static int sysctl_speed_limit_max = 200000;
96 static inline int speed_min(mddev_t *mddev)
98 return mddev->sync_speed_min ?
99 mddev->sync_speed_min : sysctl_speed_limit_min;
102 static inline int speed_max(mddev_t *mddev)
104 return mddev->sync_speed_max ?
105 mddev->sync_speed_max : sysctl_speed_limit_max;
108 static struct ctl_table_header *raid_table_header;
110 static ctl_table raid_table[] = {
112 .procname = "speed_limit_min",
113 .data = &sysctl_speed_limit_min,
114 .maxlen = sizeof(int),
115 .mode = S_IRUGO|S_IWUSR,
116 .proc_handler = proc_dointvec,
119 .procname = "speed_limit_max",
120 .data = &sysctl_speed_limit_max,
121 .maxlen = sizeof(int),
122 .mode = S_IRUGO|S_IWUSR,
123 .proc_handler = proc_dointvec,
128 static ctl_table raid_dir_table[] = {
132 .mode = S_IRUGO|S_IXUGO,
138 static ctl_table raid_root_table[] = {
143 .child = raid_dir_table,
148 static const struct block_device_operations md_fops;
150 static int start_readonly;
153 * like bio_clone, but with a local bio set
156 static void mddev_bio_destructor(struct bio *bio)
158 mddev_t *mddev, **mddevp;
163 bio_free(bio, mddev->bio_set);
166 struct bio *bio_alloc_mddev(gfp_t gfp_mask, int nr_iovecs,
172 if (!mddev || !mddev->bio_set)
173 return bio_alloc(gfp_mask, nr_iovecs);
175 b = bio_alloc_bioset(gfp_mask, nr_iovecs,
181 b->bi_destructor = mddev_bio_destructor;
184 EXPORT_SYMBOL_GPL(bio_alloc_mddev);
186 struct bio *bio_clone_mddev(struct bio *bio, gfp_t gfp_mask,
192 if (!mddev || !mddev->bio_set)
193 return bio_clone(bio, gfp_mask);
195 b = bio_alloc_bioset(gfp_mask, bio->bi_max_vecs,
201 b->bi_destructor = mddev_bio_destructor;
203 if (bio_integrity(bio)) {
206 ret = bio_integrity_clone(b, bio, gfp_mask, mddev->bio_set);
216 EXPORT_SYMBOL_GPL(bio_clone_mddev);
219 * We have a system wide 'event count' that is incremented
220 * on any 'interesting' event, and readers of /proc/mdstat
221 * can use 'poll' or 'select' to find out when the event
225 * start array, stop array, error, add device, remove device,
226 * start build, activate spare
228 static DECLARE_WAIT_QUEUE_HEAD(md_event_waiters);
229 static atomic_t md_event_count;
230 void md_new_event(mddev_t *mddev)
232 atomic_inc(&md_event_count);
233 wake_up(&md_event_waiters);
235 EXPORT_SYMBOL_GPL(md_new_event);
237 /* Alternate version that can be called from interrupts
238 * when calling sysfs_notify isn't needed.
240 static void md_new_event_inintr(mddev_t *mddev)
242 atomic_inc(&md_event_count);
243 wake_up(&md_event_waiters);
247 * Enables to iterate over all existing md arrays
248 * all_mddevs_lock protects this list.
250 static LIST_HEAD(all_mddevs);
251 static DEFINE_SPINLOCK(all_mddevs_lock);
255 * iterates through all used mddevs in the system.
256 * We take care to grab the all_mddevs_lock whenever navigating
257 * the list, and to always hold a refcount when unlocked.
258 * Any code which breaks out of this loop while own
259 * a reference to the current mddev and must mddev_put it.
261 #define for_each_mddev(mddev,tmp) \
263 for (({ spin_lock(&all_mddevs_lock); \
264 tmp = all_mddevs.next; \
266 ({ if (tmp != &all_mddevs) \
267 mddev_get(list_entry(tmp, mddev_t, all_mddevs));\
268 spin_unlock(&all_mddevs_lock); \
269 if (mddev) mddev_put(mddev); \
270 mddev = list_entry(tmp, mddev_t, all_mddevs); \
271 tmp != &all_mddevs;}); \
272 ({ spin_lock(&all_mddevs_lock); \
277 /* Rather than calling directly into the personality make_request function,
278 * IO requests come here first so that we can check if the device is
279 * being suspended pending a reconfiguration.
280 * We hold a refcount over the call to ->make_request. By the time that
281 * call has finished, the bio has been linked into some internal structure
282 * and so is visible to ->quiesce(), so we don't need the refcount any more.
284 static int md_make_request(struct request_queue *q, struct bio *bio)
286 const int rw = bio_data_dir(bio);
287 mddev_t *mddev = q->queuedata;
291 if (mddev == NULL || mddev->pers == NULL) {
296 if (mddev->suspended) {
299 prepare_to_wait(&mddev->sb_wait, &__wait,
300 TASK_UNINTERRUPTIBLE);
301 if (!mddev->suspended)
307 finish_wait(&mddev->sb_wait, &__wait);
309 atomic_inc(&mddev->active_io);
312 rv = mddev->pers->make_request(mddev, bio);
314 cpu = part_stat_lock();
315 part_stat_inc(cpu, &mddev->gendisk->part0, ios[rw]);
316 part_stat_add(cpu, &mddev->gendisk->part0, sectors[rw],
320 if (atomic_dec_and_test(&mddev->active_io) && mddev->suspended)
321 wake_up(&mddev->sb_wait);
326 /* mddev_suspend makes sure no new requests are submitted
327 * to the device, and that any requests that have been submitted
328 * are completely handled.
329 * Once ->stop is called and completes, the module will be completely
332 void mddev_suspend(mddev_t *mddev)
334 BUG_ON(mddev->suspended);
335 mddev->suspended = 1;
337 wait_event(mddev->sb_wait, atomic_read(&mddev->active_io) == 0);
338 mddev->pers->quiesce(mddev, 1);
340 EXPORT_SYMBOL_GPL(mddev_suspend);
342 void mddev_resume(mddev_t *mddev)
344 mddev->suspended = 0;
345 wake_up(&mddev->sb_wait);
346 mddev->pers->quiesce(mddev, 0);
348 EXPORT_SYMBOL_GPL(mddev_resume);
350 int mddev_congested(mddev_t *mddev, int bits)
352 return mddev->suspended;
354 EXPORT_SYMBOL(mddev_congested);
357 * Generic flush handling for md
360 static void md_end_flush(struct bio *bio, int err)
362 mdk_rdev_t *rdev = bio->bi_private;
363 mddev_t *mddev = rdev->mddev;
365 rdev_dec_pending(rdev, mddev);
367 if (atomic_dec_and_test(&mddev->flush_pending)) {
368 /* The pre-request flush has finished */
369 queue_work(md_wq, &mddev->flush_work);
374 static void md_submit_flush_data(struct work_struct *ws);
376 static void submit_flushes(struct work_struct *ws)
378 mddev_t *mddev = container_of(ws, mddev_t, flush_work);
381 INIT_WORK(&mddev->flush_work, md_submit_flush_data);
382 atomic_set(&mddev->flush_pending, 1);
384 list_for_each_entry_rcu(rdev, &mddev->disks, same_set)
385 if (rdev->raid_disk >= 0 &&
386 !test_bit(Faulty, &rdev->flags)) {
387 /* Take two references, one is dropped
388 * when request finishes, one after
389 * we reclaim rcu_read_lock
392 atomic_inc(&rdev->nr_pending);
393 atomic_inc(&rdev->nr_pending);
395 bi = bio_alloc_mddev(GFP_KERNEL, 0, mddev);
396 bi->bi_end_io = md_end_flush;
397 bi->bi_private = rdev;
398 bi->bi_bdev = rdev->bdev;
399 atomic_inc(&mddev->flush_pending);
400 submit_bio(WRITE_FLUSH, bi);
402 rdev_dec_pending(rdev, mddev);
405 if (atomic_dec_and_test(&mddev->flush_pending))
406 queue_work(md_wq, &mddev->flush_work);
409 static void md_submit_flush_data(struct work_struct *ws)
411 mddev_t *mddev = container_of(ws, mddev_t, flush_work);
412 struct bio *bio = mddev->flush_bio;
414 if (bio->bi_size == 0)
415 /* an empty barrier - all done */
418 bio->bi_rw &= ~REQ_FLUSH;
419 if (mddev->pers->make_request(mddev, bio))
420 generic_make_request(bio);
423 mddev->flush_bio = NULL;
424 wake_up(&mddev->sb_wait);
427 void md_flush_request(mddev_t *mddev, struct bio *bio)
429 spin_lock_irq(&mddev->write_lock);
430 wait_event_lock_irq(mddev->sb_wait,
432 mddev->write_lock, /*nothing*/);
433 mddev->flush_bio = bio;
434 spin_unlock_irq(&mddev->write_lock);
436 INIT_WORK(&mddev->flush_work, submit_flushes);
437 queue_work(md_wq, &mddev->flush_work);
439 EXPORT_SYMBOL(md_flush_request);
441 /* Support for plugging.
442 * This mirrors the plugging support in request_queue, but does not
443 * require having a whole queue
445 static void plugger_work(struct work_struct *work)
447 struct plug_handle *plug =
448 container_of(work, struct plug_handle, unplug_work);
449 plug->unplug_fn(plug);
451 static void plugger_timeout(unsigned long data)
453 struct plug_handle *plug = (void *)data;
454 kblockd_schedule_work(NULL, &plug->unplug_work);
456 void plugger_init(struct plug_handle *plug,
457 void (*unplug_fn)(struct plug_handle *))
459 plug->unplug_flag = 0;
460 plug->unplug_fn = unplug_fn;
461 init_timer(&plug->unplug_timer);
462 plug->unplug_timer.function = plugger_timeout;
463 plug->unplug_timer.data = (unsigned long)plug;
464 INIT_WORK(&plug->unplug_work, plugger_work);
466 EXPORT_SYMBOL_GPL(plugger_init);
468 void plugger_set_plug(struct plug_handle *plug)
470 if (!test_and_set_bit(PLUGGED_FLAG, &plug->unplug_flag))
471 mod_timer(&plug->unplug_timer, jiffies + msecs_to_jiffies(3)+1);
473 EXPORT_SYMBOL_GPL(plugger_set_plug);
475 int plugger_remove_plug(struct plug_handle *plug)
477 if (test_and_clear_bit(PLUGGED_FLAG, &plug->unplug_flag)) {
478 del_timer(&plug->unplug_timer);
483 EXPORT_SYMBOL_GPL(plugger_remove_plug);
486 static inline mddev_t *mddev_get(mddev_t *mddev)
488 atomic_inc(&mddev->active);
492 static void mddev_delayed_delete(struct work_struct *ws);
494 static void mddev_put(mddev_t *mddev)
496 struct bio_set *bs = NULL;
498 if (!atomic_dec_and_lock(&mddev->active, &all_mddevs_lock))
500 if (!mddev->raid_disks && list_empty(&mddev->disks) &&
501 mddev->ctime == 0 && !mddev->hold_active) {
502 /* Array is not configured at all, and not held active,
504 list_del(&mddev->all_mddevs);
506 mddev->bio_set = NULL;
507 if (mddev->gendisk) {
508 /* We did a probe so need to clean up. Call
509 * queue_work inside the spinlock so that
510 * flush_workqueue() after mddev_find will
511 * succeed in waiting for the work to be done.
513 INIT_WORK(&mddev->del_work, mddev_delayed_delete);
514 queue_work(md_misc_wq, &mddev->del_work);
518 spin_unlock(&all_mddevs_lock);
523 void mddev_init(mddev_t *mddev)
525 mutex_init(&mddev->open_mutex);
526 mutex_init(&mddev->reconfig_mutex);
527 mutex_init(&mddev->bitmap_info.mutex);
528 INIT_LIST_HEAD(&mddev->disks);
529 INIT_LIST_HEAD(&mddev->all_mddevs);
530 init_timer(&mddev->safemode_timer);
531 atomic_set(&mddev->active, 1);
532 atomic_set(&mddev->openers, 0);
533 atomic_set(&mddev->active_io, 0);
534 spin_lock_init(&mddev->write_lock);
535 atomic_set(&mddev->flush_pending, 0);
536 init_waitqueue_head(&mddev->sb_wait);
537 init_waitqueue_head(&mddev->recovery_wait);
538 mddev->reshape_position = MaxSector;
539 mddev->resync_min = 0;
540 mddev->resync_max = MaxSector;
541 mddev->level = LEVEL_NONE;
543 EXPORT_SYMBOL_GPL(mddev_init);
545 static mddev_t * mddev_find(dev_t unit)
547 mddev_t *mddev, *new = NULL;
550 spin_lock(&all_mddevs_lock);
553 list_for_each_entry(mddev, &all_mddevs, all_mddevs)
554 if (mddev->unit == unit) {
556 spin_unlock(&all_mddevs_lock);
562 list_add(&new->all_mddevs, &all_mddevs);
563 spin_unlock(&all_mddevs_lock);
564 new->hold_active = UNTIL_IOCTL;
568 /* find an unused unit number */
569 static int next_minor = 512;
570 int start = next_minor;
574 dev = MKDEV(MD_MAJOR, next_minor);
576 if (next_minor > MINORMASK)
578 if (next_minor == start) {
579 /* Oh dear, all in use. */
580 spin_unlock(&all_mddevs_lock);
586 list_for_each_entry(mddev, &all_mddevs, all_mddevs)
587 if (mddev->unit == dev) {
593 new->md_minor = MINOR(dev);
594 new->hold_active = UNTIL_STOP;
595 list_add(&new->all_mddevs, &all_mddevs);
596 spin_unlock(&all_mddevs_lock);
599 spin_unlock(&all_mddevs_lock);
601 new = kzalloc(sizeof(*new), GFP_KERNEL);
606 if (MAJOR(unit) == MD_MAJOR)
607 new->md_minor = MINOR(unit);
609 new->md_minor = MINOR(unit) >> MdpMinorShift;
616 static inline int mddev_lock(mddev_t * mddev)
618 return mutex_lock_interruptible(&mddev->reconfig_mutex);
621 static inline int mddev_is_locked(mddev_t *mddev)
623 return mutex_is_locked(&mddev->reconfig_mutex);
626 static inline int mddev_trylock(mddev_t * mddev)
628 return mutex_trylock(&mddev->reconfig_mutex);
631 static struct attribute_group md_redundancy_group;
633 static void mddev_unlock(mddev_t * mddev)
635 if (mddev->to_remove) {
636 /* These cannot be removed under reconfig_mutex as
637 * an access to the files will try to take reconfig_mutex
638 * while holding the file unremovable, which leads to
640 * So hold set sysfs_active while the remove in happeing,
641 * and anything else which might set ->to_remove or my
642 * otherwise change the sysfs namespace will fail with
643 * -EBUSY if sysfs_active is still set.
644 * We set sysfs_active under reconfig_mutex and elsewhere
645 * test it under the same mutex to ensure its correct value
648 struct attribute_group *to_remove = mddev->to_remove;
649 mddev->to_remove = NULL;
650 mddev->sysfs_active = 1;
651 mutex_unlock(&mddev->reconfig_mutex);
653 if (mddev->kobj.sd) {
654 if (to_remove != &md_redundancy_group)
655 sysfs_remove_group(&mddev->kobj, to_remove);
656 if (mddev->pers == NULL ||
657 mddev->pers->sync_request == NULL) {
658 sysfs_remove_group(&mddev->kobj, &md_redundancy_group);
659 if (mddev->sysfs_action)
660 sysfs_put(mddev->sysfs_action);
661 mddev->sysfs_action = NULL;
664 mddev->sysfs_active = 0;
666 mutex_unlock(&mddev->reconfig_mutex);
668 md_wakeup_thread(mddev->thread);
671 static mdk_rdev_t * find_rdev_nr(mddev_t *mddev, int nr)
675 list_for_each_entry(rdev, &mddev->disks, same_set)
676 if (rdev->desc_nr == nr)
682 static mdk_rdev_t * find_rdev(mddev_t * mddev, dev_t dev)
686 list_for_each_entry(rdev, &mddev->disks, same_set)
687 if (rdev->bdev->bd_dev == dev)
693 static struct mdk_personality *find_pers(int level, char *clevel)
695 struct mdk_personality *pers;
696 list_for_each_entry(pers, &pers_list, list) {
697 if (level != LEVEL_NONE && pers->level == level)
699 if (strcmp(pers->name, clevel)==0)
705 /* return the offset of the super block in 512byte sectors */
706 static inline sector_t calc_dev_sboffset(struct block_device *bdev)
708 sector_t num_sectors = i_size_read(bdev->bd_inode) / 512;
709 return MD_NEW_SIZE_SECTORS(num_sectors);
712 static int alloc_disk_sb(mdk_rdev_t * rdev)
717 rdev->sb_page = alloc_page(GFP_KERNEL);
718 if (!rdev->sb_page) {
719 printk(KERN_ALERT "md: out of memory.\n");
726 static void free_disk_sb(mdk_rdev_t * rdev)
729 put_page(rdev->sb_page);
731 rdev->sb_page = NULL;
738 static void super_written(struct bio *bio, int error)
740 mdk_rdev_t *rdev = bio->bi_private;
741 mddev_t *mddev = rdev->mddev;
743 if (error || !test_bit(BIO_UPTODATE, &bio->bi_flags)) {
744 printk("md: super_written gets error=%d, uptodate=%d\n",
745 error, test_bit(BIO_UPTODATE, &bio->bi_flags));
746 WARN_ON(test_bit(BIO_UPTODATE, &bio->bi_flags));
747 md_error(mddev, rdev);
750 if (atomic_dec_and_test(&mddev->pending_writes))
751 wake_up(&mddev->sb_wait);
755 void md_super_write(mddev_t *mddev, mdk_rdev_t *rdev,
756 sector_t sector, int size, struct page *page)
758 /* write first size bytes of page to sector of rdev
759 * Increment mddev->pending_writes before returning
760 * and decrement it on completion, waking up sb_wait
761 * if zero is reached.
762 * If an error occurred, call md_error
764 struct bio *bio = bio_alloc_mddev(GFP_NOIO, 1, mddev);
766 bio->bi_bdev = rdev->bdev;
767 bio->bi_sector = sector;
768 bio_add_page(bio, page, size, 0);
769 bio->bi_private = rdev;
770 bio->bi_end_io = super_written;
772 atomic_inc(&mddev->pending_writes);
773 submit_bio(REQ_WRITE | REQ_SYNC | REQ_UNPLUG | REQ_FLUSH | REQ_FUA,
777 void md_super_wait(mddev_t *mddev)
779 /* wait for all superblock writes that were scheduled to complete */
782 prepare_to_wait(&mddev->sb_wait, &wq, TASK_UNINTERRUPTIBLE);
783 if (atomic_read(&mddev->pending_writes)==0)
787 finish_wait(&mddev->sb_wait, &wq);
790 static void bi_complete(struct bio *bio, int error)
792 complete((struct completion*)bio->bi_private);
795 int sync_page_io(mdk_rdev_t *rdev, sector_t sector, int size,
796 struct page *page, int rw)
798 struct bio *bio = bio_alloc_mddev(GFP_NOIO, 1, rdev->mddev);
799 struct completion event;
802 rw |= REQ_SYNC | REQ_UNPLUG;
804 bio->bi_bdev = rdev->bdev;
805 bio->bi_sector = sector;
806 bio_add_page(bio, page, size, 0);
807 init_completion(&event);
808 bio->bi_private = &event;
809 bio->bi_end_io = bi_complete;
811 wait_for_completion(&event);
813 ret = test_bit(BIO_UPTODATE, &bio->bi_flags);
817 EXPORT_SYMBOL_GPL(sync_page_io);
819 static int read_disk_sb(mdk_rdev_t * rdev, int size)
821 char b[BDEVNAME_SIZE];
822 if (!rdev->sb_page) {
830 if (!sync_page_io(rdev, rdev->sb_start, size, rdev->sb_page, READ))
836 printk(KERN_WARNING "md: disabled device %s, could not read superblock.\n",
837 bdevname(rdev->bdev,b));
841 static int uuid_equal(mdp_super_t *sb1, mdp_super_t *sb2)
843 return sb1->set_uuid0 == sb2->set_uuid0 &&
844 sb1->set_uuid1 == sb2->set_uuid1 &&
845 sb1->set_uuid2 == sb2->set_uuid2 &&
846 sb1->set_uuid3 == sb2->set_uuid3;
849 static int sb_equal(mdp_super_t *sb1, mdp_super_t *sb2)
852 mdp_super_t *tmp1, *tmp2;
854 tmp1 = kmalloc(sizeof(*tmp1),GFP_KERNEL);
855 tmp2 = kmalloc(sizeof(*tmp2),GFP_KERNEL);
857 if (!tmp1 || !tmp2) {
859 printk(KERN_INFO "md.c sb_equal(): failed to allocate memory!\n");
867 * nr_disks is not constant
872 ret = (memcmp(tmp1, tmp2, MD_SB_GENERIC_CONSTANT_WORDS * 4) == 0);
880 static u32 md_csum_fold(u32 csum)
882 csum = (csum & 0xffff) + (csum >> 16);
883 return (csum & 0xffff) + (csum >> 16);
886 static unsigned int calc_sb_csum(mdp_super_t * sb)
889 u32 *sb32 = (u32*)sb;
891 unsigned int disk_csum, csum;
893 disk_csum = sb->sb_csum;
896 for (i = 0; i < MD_SB_BYTES/4 ; i++)
898 csum = (newcsum & 0xffffffff) + (newcsum>>32);
902 /* This used to use csum_partial, which was wrong for several
903 * reasons including that different results are returned on
904 * different architectures. It isn't critical that we get exactly
905 * the same return value as before (we always csum_fold before
906 * testing, and that removes any differences). However as we
907 * know that csum_partial always returned a 16bit value on
908 * alphas, do a fold to maximise conformity to previous behaviour.
910 sb->sb_csum = md_csum_fold(disk_csum);
912 sb->sb_csum = disk_csum;
919 * Handle superblock details.
920 * We want to be able to handle multiple superblock formats
921 * so we have a common interface to them all, and an array of
922 * different handlers.
923 * We rely on user-space to write the initial superblock, and support
924 * reading and updating of superblocks.
925 * Interface methods are:
926 * int load_super(mdk_rdev_t *dev, mdk_rdev_t *refdev, int minor_version)
927 * loads and validates a superblock on dev.
928 * if refdev != NULL, compare superblocks on both devices
930 * 0 - dev has a superblock that is compatible with refdev
931 * 1 - dev has a superblock that is compatible and newer than refdev
932 * so dev should be used as the refdev in future
933 * -EINVAL superblock incompatible or invalid
934 * -othererror e.g. -EIO
936 * int validate_super(mddev_t *mddev, mdk_rdev_t *dev)
937 * Verify that dev is acceptable into mddev.
938 * The first time, mddev->raid_disks will be 0, and data from
939 * dev should be merged in. Subsequent calls check that dev
940 * is new enough. Return 0 or -EINVAL
942 * void sync_super(mddev_t *mddev, mdk_rdev_t *dev)
943 * Update the superblock for rdev with data in mddev
944 * This does not write to disc.
950 struct module *owner;
951 int (*load_super)(mdk_rdev_t *rdev, mdk_rdev_t *refdev,
953 int (*validate_super)(mddev_t *mddev, mdk_rdev_t *rdev);
954 void (*sync_super)(mddev_t *mddev, mdk_rdev_t *rdev);
955 unsigned long long (*rdev_size_change)(mdk_rdev_t *rdev,
956 sector_t num_sectors);
960 * Check that the given mddev has no bitmap.
962 * This function is called from the run method of all personalities that do not
963 * support bitmaps. It prints an error message and returns non-zero if mddev
964 * has a bitmap. Otherwise, it returns 0.
967 int md_check_no_bitmap(mddev_t *mddev)
969 if (!mddev->bitmap_info.file && !mddev->bitmap_info.offset)
971 printk(KERN_ERR "%s: bitmaps are not supported for %s\n",
972 mdname(mddev), mddev->pers->name);
975 EXPORT_SYMBOL(md_check_no_bitmap);
978 * load_super for 0.90.0
980 static int super_90_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
982 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
987 * Calculate the position of the superblock (512byte sectors),
988 * it's at the end of the disk.
990 * It also happens to be a multiple of 4Kb.
992 rdev->sb_start = calc_dev_sboffset(rdev->bdev);
994 ret = read_disk_sb(rdev, MD_SB_BYTES);
999 bdevname(rdev->bdev, b);
1000 sb = (mdp_super_t*)page_address(rdev->sb_page);
1002 if (sb->md_magic != MD_SB_MAGIC) {
1003 printk(KERN_ERR "md: invalid raid superblock magic on %s\n",
1008 if (sb->major_version != 0 ||
1009 sb->minor_version < 90 ||
1010 sb->minor_version > 91) {
1011 printk(KERN_WARNING "Bad version number %d.%d on %s\n",
1012 sb->major_version, sb->minor_version,
1017 if (sb->raid_disks <= 0)
1020 if (md_csum_fold(calc_sb_csum(sb)) != md_csum_fold(sb->sb_csum)) {
1021 printk(KERN_WARNING "md: invalid superblock checksum on %s\n",
1026 rdev->preferred_minor = sb->md_minor;
1027 rdev->data_offset = 0;
1028 rdev->sb_size = MD_SB_BYTES;
1030 if (sb->level == LEVEL_MULTIPATH)
1033 rdev->desc_nr = sb->this_disk.number;
1039 mdp_super_t *refsb = (mdp_super_t*)page_address(refdev->sb_page);
1040 if (!uuid_equal(refsb, sb)) {
1041 printk(KERN_WARNING "md: %s has different UUID to %s\n",
1042 b, bdevname(refdev->bdev,b2));
1045 if (!sb_equal(refsb, sb)) {
1046 printk(KERN_WARNING "md: %s has same UUID"
1047 " but different superblock to %s\n",
1048 b, bdevname(refdev->bdev, b2));
1052 ev2 = md_event(refsb);
1058 rdev->sectors = rdev->sb_start;
1060 if (rdev->sectors < sb->size * 2 && sb->level > 1)
1061 /* "this cannot possibly happen" ... */
1069 * validate_super for 0.90.0
1071 static int super_90_validate(mddev_t *mddev, mdk_rdev_t *rdev)
1074 mdp_super_t *sb = (mdp_super_t *)page_address(rdev->sb_page);
1075 __u64 ev1 = md_event(sb);
1077 rdev->raid_disk = -1;
1078 clear_bit(Faulty, &rdev->flags);
1079 clear_bit(In_sync, &rdev->flags);
1080 clear_bit(WriteMostly, &rdev->flags);
1082 if (mddev->raid_disks == 0) {
1083 mddev->major_version = 0;
1084 mddev->minor_version = sb->minor_version;
1085 mddev->patch_version = sb->patch_version;
1086 mddev->external = 0;
1087 mddev->chunk_sectors = sb->chunk_size >> 9;
1088 mddev->ctime = sb->ctime;
1089 mddev->utime = sb->utime;
1090 mddev->level = sb->level;
1091 mddev->clevel[0] = 0;
1092 mddev->layout = sb->layout;
1093 mddev->raid_disks = sb->raid_disks;
1094 mddev->dev_sectors = sb->size * 2;
1095 mddev->events = ev1;
1096 mddev->bitmap_info.offset = 0;
1097 mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9;
1099 if (mddev->minor_version >= 91) {
1100 mddev->reshape_position = sb->reshape_position;
1101 mddev->delta_disks = sb->delta_disks;
1102 mddev->new_level = sb->new_level;
1103 mddev->new_layout = sb->new_layout;
1104 mddev->new_chunk_sectors = sb->new_chunk >> 9;
1106 mddev->reshape_position = MaxSector;
1107 mddev->delta_disks = 0;
1108 mddev->new_level = mddev->level;
1109 mddev->new_layout = mddev->layout;
1110 mddev->new_chunk_sectors = mddev->chunk_sectors;
1113 if (sb->state & (1<<MD_SB_CLEAN))
1114 mddev->recovery_cp = MaxSector;
1116 if (sb->events_hi == sb->cp_events_hi &&
1117 sb->events_lo == sb->cp_events_lo) {
1118 mddev->recovery_cp = sb->recovery_cp;
1120 mddev->recovery_cp = 0;
1123 memcpy(mddev->uuid+0, &sb->set_uuid0, 4);
1124 memcpy(mddev->uuid+4, &sb->set_uuid1, 4);
1125 memcpy(mddev->uuid+8, &sb->set_uuid2, 4);
1126 memcpy(mddev->uuid+12,&sb->set_uuid3, 4);
1128 mddev->max_disks = MD_SB_DISKS;
1130 if (sb->state & (1<<MD_SB_BITMAP_PRESENT) &&
1131 mddev->bitmap_info.file == NULL)
1132 mddev->bitmap_info.offset =
1133 mddev->bitmap_info.default_offset;
1135 } else if (mddev->pers == NULL) {
1136 /* Insist on good event counter while assembling, except
1137 * for spares (which don't need an event count) */
1139 if (sb->disks[rdev->desc_nr].state & (
1140 (1<<MD_DISK_SYNC) | (1 << MD_DISK_ACTIVE)))
1141 if (ev1 < mddev->events)
1143 } else if (mddev->bitmap) {
1144 /* if adding to array with a bitmap, then we can accept an
1145 * older device ... but not too old.
1147 if (ev1 < mddev->bitmap->events_cleared)
1150 if (ev1 < mddev->events)
1151 /* just a hot-add of a new device, leave raid_disk at -1 */
1155 if (mddev->level != LEVEL_MULTIPATH) {
1156 desc = sb->disks + rdev->desc_nr;
1158 if (desc->state & (1<<MD_DISK_FAULTY))
1159 set_bit(Faulty, &rdev->flags);
1160 else if (desc->state & (1<<MD_DISK_SYNC) /* &&
1161 desc->raid_disk < mddev->raid_disks */) {
1162 set_bit(In_sync, &rdev->flags);
1163 rdev->raid_disk = desc->raid_disk;
1164 } else if (desc->state & (1<<MD_DISK_ACTIVE)) {
1165 /* active but not in sync implies recovery up to
1166 * reshape position. We don't know exactly where
1167 * that is, so set to zero for now */
1168 if (mddev->minor_version >= 91) {
1169 rdev->recovery_offset = 0;
1170 rdev->raid_disk = desc->raid_disk;
1173 if (desc->state & (1<<MD_DISK_WRITEMOSTLY))
1174 set_bit(WriteMostly, &rdev->flags);
1175 } else /* MULTIPATH are always insync */
1176 set_bit(In_sync, &rdev->flags);
1181 * sync_super for 0.90.0
1183 static void super_90_sync(mddev_t *mddev, mdk_rdev_t *rdev)
1187 int next_spare = mddev->raid_disks;
1190 /* make rdev->sb match mddev data..
1193 * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
1194 * 3/ any empty disks < next_spare become removed
1196 * disks[0] gets initialised to REMOVED because
1197 * we cannot be sure from other fields if it has
1198 * been initialised or not.
1201 int active=0, working=0,failed=0,spare=0,nr_disks=0;
1203 rdev->sb_size = MD_SB_BYTES;
1205 sb = (mdp_super_t*)page_address(rdev->sb_page);
1207 memset(sb, 0, sizeof(*sb));
1209 sb->md_magic = MD_SB_MAGIC;
1210 sb->major_version = mddev->major_version;
1211 sb->patch_version = mddev->patch_version;
1212 sb->gvalid_words = 0; /* ignored */
1213 memcpy(&sb->set_uuid0, mddev->uuid+0, 4);
1214 memcpy(&sb->set_uuid1, mddev->uuid+4, 4);
1215 memcpy(&sb->set_uuid2, mddev->uuid+8, 4);
1216 memcpy(&sb->set_uuid3, mddev->uuid+12,4);
1218 sb->ctime = mddev->ctime;
1219 sb->level = mddev->level;
1220 sb->size = mddev->dev_sectors / 2;
1221 sb->raid_disks = mddev->raid_disks;
1222 sb->md_minor = mddev->md_minor;
1223 sb->not_persistent = 0;
1224 sb->utime = mddev->utime;
1226 sb->events_hi = (mddev->events>>32);
1227 sb->events_lo = (u32)mddev->events;
1229 if (mddev->reshape_position == MaxSector)
1230 sb->minor_version = 90;
1232 sb->minor_version = 91;
1233 sb->reshape_position = mddev->reshape_position;
1234 sb->new_level = mddev->new_level;
1235 sb->delta_disks = mddev->delta_disks;
1236 sb->new_layout = mddev->new_layout;
1237 sb->new_chunk = mddev->new_chunk_sectors << 9;
1239 mddev->minor_version = sb->minor_version;
1242 sb->recovery_cp = mddev->recovery_cp;
1243 sb->cp_events_hi = (mddev->events>>32);
1244 sb->cp_events_lo = (u32)mddev->events;
1245 if (mddev->recovery_cp == MaxSector)
1246 sb->state = (1<< MD_SB_CLEAN);
1248 sb->recovery_cp = 0;
1250 sb->layout = mddev->layout;
1251 sb->chunk_size = mddev->chunk_sectors << 9;
1253 if (mddev->bitmap && mddev->bitmap_info.file == NULL)
1254 sb->state |= (1<<MD_SB_BITMAP_PRESENT);
1256 sb->disks[0].state = (1<<MD_DISK_REMOVED);
1257 list_for_each_entry(rdev2, &mddev->disks, same_set) {
1260 int is_active = test_bit(In_sync, &rdev2->flags);
1262 if (rdev2->raid_disk >= 0 &&
1263 sb->minor_version >= 91)
1264 /* we have nowhere to store the recovery_offset,
1265 * but if it is not below the reshape_position,
1266 * we can piggy-back on that.
1269 if (rdev2->raid_disk < 0 ||
1270 test_bit(Faulty, &rdev2->flags))
1273 desc_nr = rdev2->raid_disk;
1275 desc_nr = next_spare++;
1276 rdev2->desc_nr = desc_nr;
1277 d = &sb->disks[rdev2->desc_nr];
1279 d->number = rdev2->desc_nr;
1280 d->major = MAJOR(rdev2->bdev->bd_dev);
1281 d->minor = MINOR(rdev2->bdev->bd_dev);
1283 d->raid_disk = rdev2->raid_disk;
1285 d->raid_disk = rdev2->desc_nr; /* compatibility */
1286 if (test_bit(Faulty, &rdev2->flags))
1287 d->state = (1<<MD_DISK_FAULTY);
1288 else if (is_active) {
1289 d->state = (1<<MD_DISK_ACTIVE);
1290 if (test_bit(In_sync, &rdev2->flags))
1291 d->state |= (1<<MD_DISK_SYNC);
1299 if (test_bit(WriteMostly, &rdev2->flags))
1300 d->state |= (1<<MD_DISK_WRITEMOSTLY);
1302 /* now set the "removed" and "faulty" bits on any missing devices */
1303 for (i=0 ; i < mddev->raid_disks ; i++) {
1304 mdp_disk_t *d = &sb->disks[i];
1305 if (d->state == 0 && d->number == 0) {
1308 d->state = (1<<MD_DISK_REMOVED);
1309 d->state |= (1<<MD_DISK_FAULTY);
1313 sb->nr_disks = nr_disks;
1314 sb->active_disks = active;
1315 sb->working_disks = working;
1316 sb->failed_disks = failed;
1317 sb->spare_disks = spare;
1319 sb->this_disk = sb->disks[rdev->desc_nr];
1320 sb->sb_csum = calc_sb_csum(sb);
1324 * rdev_size_change for 0.90.0
1326 static unsigned long long
1327 super_90_rdev_size_change(mdk_rdev_t *rdev, sector_t num_sectors)
1329 if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
1330 return 0; /* component must fit device */
1331 if (rdev->mddev->bitmap_info.offset)
1332 return 0; /* can't move bitmap */
1333 rdev->sb_start = calc_dev_sboffset(rdev->bdev);
1334 if (!num_sectors || num_sectors > rdev->sb_start)
1335 num_sectors = rdev->sb_start;
1336 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
1338 md_super_wait(rdev->mddev);
1344 * version 1 superblock
1347 static __le32 calc_sb_1_csum(struct mdp_superblock_1 * sb)
1351 unsigned long long newcsum;
1352 int size = 256 + le32_to_cpu(sb->max_dev)*2;
1353 __le32 *isuper = (__le32*)sb;
1356 disk_csum = sb->sb_csum;
1359 for (i=0; size>=4; size -= 4 )
1360 newcsum += le32_to_cpu(*isuper++);
1363 newcsum += le16_to_cpu(*(__le16*) isuper);
1365 csum = (newcsum & 0xffffffff) + (newcsum >> 32);
1366 sb->sb_csum = disk_csum;
1367 return cpu_to_le32(csum);
1370 static int super_1_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
1372 struct mdp_superblock_1 *sb;
1375 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
1379 * Calculate the position of the superblock in 512byte sectors.
1380 * It is always aligned to a 4K boundary and
1381 * depeding on minor_version, it can be:
1382 * 0: At least 8K, but less than 12K, from end of device
1383 * 1: At start of device
1384 * 2: 4K from start of device.
1386 switch(minor_version) {
1388 sb_start = i_size_read(rdev->bdev->bd_inode) >> 9;
1390 sb_start &= ~(sector_t)(4*2-1);
1401 rdev->sb_start = sb_start;
1403 /* superblock is rarely larger than 1K, but it can be larger,
1404 * and it is safe to read 4k, so we do that
1406 ret = read_disk_sb(rdev, 4096);
1407 if (ret) return ret;
1410 sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1412 if (sb->magic != cpu_to_le32(MD_SB_MAGIC) ||
1413 sb->major_version != cpu_to_le32(1) ||
1414 le32_to_cpu(sb->max_dev) > (4096-256)/2 ||
1415 le64_to_cpu(sb->super_offset) != rdev->sb_start ||
1416 (le32_to_cpu(sb->feature_map) & ~MD_FEATURE_ALL) != 0)
1419 if (calc_sb_1_csum(sb) != sb->sb_csum) {
1420 printk("md: invalid superblock checksum on %s\n",
1421 bdevname(rdev->bdev,b));
1424 if (le64_to_cpu(sb->data_size) < 10) {
1425 printk("md: data_size too small on %s\n",
1426 bdevname(rdev->bdev,b));
1430 rdev->preferred_minor = 0xffff;
1431 rdev->data_offset = le64_to_cpu(sb->data_offset);
1432 atomic_set(&rdev->corrected_errors, le32_to_cpu(sb->cnt_corrected_read));
1434 rdev->sb_size = le32_to_cpu(sb->max_dev) * 2 + 256;
1435 bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
1436 if (rdev->sb_size & bmask)
1437 rdev->sb_size = (rdev->sb_size | bmask) + 1;
1440 && rdev->data_offset < sb_start + (rdev->sb_size/512))
1443 if (sb->level == cpu_to_le32(LEVEL_MULTIPATH))
1446 rdev->desc_nr = le32_to_cpu(sb->dev_number);
1452 struct mdp_superblock_1 *refsb =
1453 (struct mdp_superblock_1*)page_address(refdev->sb_page);
1455 if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 ||
1456 sb->level != refsb->level ||
1457 sb->layout != refsb->layout ||
1458 sb->chunksize != refsb->chunksize) {
1459 printk(KERN_WARNING "md: %s has strangely different"
1460 " superblock to %s\n",
1461 bdevname(rdev->bdev,b),
1462 bdevname(refdev->bdev,b2));
1465 ev1 = le64_to_cpu(sb->events);
1466 ev2 = le64_to_cpu(refsb->events);
1474 rdev->sectors = (i_size_read(rdev->bdev->bd_inode) >> 9) -
1475 le64_to_cpu(sb->data_offset);
1477 rdev->sectors = rdev->sb_start;
1478 if (rdev->sectors < le64_to_cpu(sb->data_size))
1480 rdev->sectors = le64_to_cpu(sb->data_size);
1481 if (le64_to_cpu(sb->size) > rdev->sectors)
1486 static int super_1_validate(mddev_t *mddev, mdk_rdev_t *rdev)
1488 struct mdp_superblock_1 *sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1489 __u64 ev1 = le64_to_cpu(sb->events);
1491 rdev->raid_disk = -1;
1492 clear_bit(Faulty, &rdev->flags);
1493 clear_bit(In_sync, &rdev->flags);
1494 clear_bit(WriteMostly, &rdev->flags);
1496 if (mddev->raid_disks == 0) {
1497 mddev->major_version = 1;
1498 mddev->patch_version = 0;
1499 mddev->external = 0;
1500 mddev->chunk_sectors = le32_to_cpu(sb->chunksize);
1501 mddev->ctime = le64_to_cpu(sb->ctime) & ((1ULL << 32)-1);
1502 mddev->utime = le64_to_cpu(sb->utime) & ((1ULL << 32)-1);
1503 mddev->level = le32_to_cpu(sb->level);
1504 mddev->clevel[0] = 0;
1505 mddev->layout = le32_to_cpu(sb->layout);
1506 mddev->raid_disks = le32_to_cpu(sb->raid_disks);
1507 mddev->dev_sectors = le64_to_cpu(sb->size);
1508 mddev->events = ev1;
1509 mddev->bitmap_info.offset = 0;
1510 mddev->bitmap_info.default_offset = 1024 >> 9;
1512 mddev->recovery_cp = le64_to_cpu(sb->resync_offset);
1513 memcpy(mddev->uuid, sb->set_uuid, 16);
1515 mddev->max_disks = (4096-256)/2;
1517 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) &&
1518 mddev->bitmap_info.file == NULL )
1519 mddev->bitmap_info.offset =
1520 (__s32)le32_to_cpu(sb->bitmap_offset);
1522 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) {
1523 mddev->reshape_position = le64_to_cpu(sb->reshape_position);
1524 mddev->delta_disks = le32_to_cpu(sb->delta_disks);
1525 mddev->new_level = le32_to_cpu(sb->new_level);
1526 mddev->new_layout = le32_to_cpu(sb->new_layout);
1527 mddev->new_chunk_sectors = le32_to_cpu(sb->new_chunk);
1529 mddev->reshape_position = MaxSector;
1530 mddev->delta_disks = 0;
1531 mddev->new_level = mddev->level;
1532 mddev->new_layout = mddev->layout;
1533 mddev->new_chunk_sectors = mddev->chunk_sectors;
1536 } else if (mddev->pers == NULL) {
1537 /* Insist of good event counter while assembling, except for
1538 * spares (which don't need an event count) */
1540 if (rdev->desc_nr >= 0 &&
1541 rdev->desc_nr < le32_to_cpu(sb->max_dev) &&
1542 le16_to_cpu(sb->dev_roles[rdev->desc_nr]) < 0xfffe)
1543 if (ev1 < mddev->events)
1545 } else if (mddev->bitmap) {
1546 /* If adding to array with a bitmap, then we can accept an
1547 * older device, but not too old.
1549 if (ev1 < mddev->bitmap->events_cleared)
1552 if (ev1 < mddev->events)
1553 /* just a hot-add of a new device, leave raid_disk at -1 */
1556 if (mddev->level != LEVEL_MULTIPATH) {
1558 if (rdev->desc_nr < 0 ||
1559 rdev->desc_nr >= le32_to_cpu(sb->max_dev)) {
1563 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
1565 case 0xffff: /* spare */
1567 case 0xfffe: /* faulty */
1568 set_bit(Faulty, &rdev->flags);
1571 if ((le32_to_cpu(sb->feature_map) &
1572 MD_FEATURE_RECOVERY_OFFSET))
1573 rdev->recovery_offset = le64_to_cpu(sb->recovery_offset);
1575 set_bit(In_sync, &rdev->flags);
1576 rdev->raid_disk = role;
1579 if (sb->devflags & WriteMostly1)
1580 set_bit(WriteMostly, &rdev->flags);
1581 } else /* MULTIPATH are always insync */
1582 set_bit(In_sync, &rdev->flags);
1587 static void super_1_sync(mddev_t *mddev, mdk_rdev_t *rdev)
1589 struct mdp_superblock_1 *sb;
1592 /* make rdev->sb match mddev and rdev data. */
1594 sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1596 sb->feature_map = 0;
1598 sb->recovery_offset = cpu_to_le64(0);
1599 memset(sb->pad1, 0, sizeof(sb->pad1));
1600 memset(sb->pad2, 0, sizeof(sb->pad2));
1601 memset(sb->pad3, 0, sizeof(sb->pad3));
1603 sb->utime = cpu_to_le64((__u64)mddev->utime);
1604 sb->events = cpu_to_le64(mddev->events);
1606 sb->resync_offset = cpu_to_le64(mddev->recovery_cp);
1608 sb->resync_offset = cpu_to_le64(0);
1610 sb->cnt_corrected_read = cpu_to_le32(atomic_read(&rdev->corrected_errors));
1612 sb->raid_disks = cpu_to_le32(mddev->raid_disks);
1613 sb->size = cpu_to_le64(mddev->dev_sectors);
1614 sb->chunksize = cpu_to_le32(mddev->chunk_sectors);
1615 sb->level = cpu_to_le32(mddev->level);
1616 sb->layout = cpu_to_le32(mddev->layout);
1618 if (mddev->bitmap && mddev->bitmap_info.file == NULL) {
1619 sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_info.offset);
1620 sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET);
1623 if (rdev->raid_disk >= 0 &&
1624 !test_bit(In_sync, &rdev->flags)) {
1626 cpu_to_le32(MD_FEATURE_RECOVERY_OFFSET);
1627 sb->recovery_offset =
1628 cpu_to_le64(rdev->recovery_offset);
1631 if (mddev->reshape_position != MaxSector) {
1632 sb->feature_map |= cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE);
1633 sb->reshape_position = cpu_to_le64(mddev->reshape_position);
1634 sb->new_layout = cpu_to_le32(mddev->new_layout);
1635 sb->delta_disks = cpu_to_le32(mddev->delta_disks);
1636 sb->new_level = cpu_to_le32(mddev->new_level);
1637 sb->new_chunk = cpu_to_le32(mddev->new_chunk_sectors);
1641 list_for_each_entry(rdev2, &mddev->disks, same_set)
1642 if (rdev2->desc_nr+1 > max_dev)
1643 max_dev = rdev2->desc_nr+1;
1645 if (max_dev > le32_to_cpu(sb->max_dev)) {
1647 sb->max_dev = cpu_to_le32(max_dev);
1648 rdev->sb_size = max_dev * 2 + 256;
1649 bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
1650 if (rdev->sb_size & bmask)
1651 rdev->sb_size = (rdev->sb_size | bmask) + 1;
1653 max_dev = le32_to_cpu(sb->max_dev);
1655 for (i=0; i<max_dev;i++)
1656 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1658 list_for_each_entry(rdev2, &mddev->disks, same_set) {
1660 if (test_bit(Faulty, &rdev2->flags))
1661 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1662 else if (test_bit(In_sync, &rdev2->flags))
1663 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1664 else if (rdev2->raid_disk >= 0)
1665 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1667 sb->dev_roles[i] = cpu_to_le16(0xffff);
1670 sb->sb_csum = calc_sb_1_csum(sb);
1673 static unsigned long long
1674 super_1_rdev_size_change(mdk_rdev_t *rdev, sector_t num_sectors)
1676 struct mdp_superblock_1 *sb;
1677 sector_t max_sectors;
1678 if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
1679 return 0; /* component must fit device */
1680 if (rdev->sb_start < rdev->data_offset) {
1681 /* minor versions 1 and 2; superblock before data */
1682 max_sectors = i_size_read(rdev->bdev->bd_inode) >> 9;
1683 max_sectors -= rdev->data_offset;
1684 if (!num_sectors || num_sectors > max_sectors)
1685 num_sectors = max_sectors;
1686 } else if (rdev->mddev->bitmap_info.offset) {
1687 /* minor version 0 with bitmap we can't move */
1690 /* minor version 0; superblock after data */
1692 sb_start = (i_size_read(rdev->bdev->bd_inode) >> 9) - 8*2;
1693 sb_start &= ~(sector_t)(4*2 - 1);
1694 max_sectors = rdev->sectors + sb_start - rdev->sb_start;
1695 if (!num_sectors || num_sectors > max_sectors)
1696 num_sectors = max_sectors;
1697 rdev->sb_start = sb_start;
1699 sb = (struct mdp_superblock_1 *) page_address(rdev->sb_page);
1700 sb->data_size = cpu_to_le64(num_sectors);
1701 sb->super_offset = rdev->sb_start;
1702 sb->sb_csum = calc_sb_1_csum(sb);
1703 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
1705 md_super_wait(rdev->mddev);
1709 static struct super_type super_types[] = {
1712 .owner = THIS_MODULE,
1713 .load_super = super_90_load,
1714 .validate_super = super_90_validate,
1715 .sync_super = super_90_sync,
1716 .rdev_size_change = super_90_rdev_size_change,
1720 .owner = THIS_MODULE,
1721 .load_super = super_1_load,
1722 .validate_super = super_1_validate,
1723 .sync_super = super_1_sync,
1724 .rdev_size_change = super_1_rdev_size_change,
1728 static int match_mddev_units(mddev_t *mddev1, mddev_t *mddev2)
1730 mdk_rdev_t *rdev, *rdev2;
1733 rdev_for_each_rcu(rdev, mddev1)
1734 rdev_for_each_rcu(rdev2, mddev2)
1735 if (rdev->bdev->bd_contains ==
1736 rdev2->bdev->bd_contains) {
1744 static LIST_HEAD(pending_raid_disks);
1747 * Try to register data integrity profile for an mddev
1749 * This is called when an array is started and after a disk has been kicked
1750 * from the array. It only succeeds if all working and active component devices
1751 * are integrity capable with matching profiles.
1753 int md_integrity_register(mddev_t *mddev)
1755 mdk_rdev_t *rdev, *reference = NULL;
1757 if (list_empty(&mddev->disks))
1758 return 0; /* nothing to do */
1759 if (blk_get_integrity(mddev->gendisk))
1760 return 0; /* already registered */
1761 list_for_each_entry(rdev, &mddev->disks, same_set) {
1762 /* skip spares and non-functional disks */
1763 if (test_bit(Faulty, &rdev->flags))
1765 if (rdev->raid_disk < 0)
1768 * If at least one rdev is not integrity capable, we can not
1769 * enable data integrity for the md device.
1771 if (!bdev_get_integrity(rdev->bdev))
1774 /* Use the first rdev as the reference */
1778 /* does this rdev's profile match the reference profile? */
1779 if (blk_integrity_compare(reference->bdev->bd_disk,
1780 rdev->bdev->bd_disk) < 0)
1784 * All component devices are integrity capable and have matching
1785 * profiles, register the common profile for the md device.
1787 if (blk_integrity_register(mddev->gendisk,
1788 bdev_get_integrity(reference->bdev)) != 0) {
1789 printk(KERN_ERR "md: failed to register integrity for %s\n",
1793 printk(KERN_NOTICE "md: data integrity on %s enabled\n",
1797 EXPORT_SYMBOL(md_integrity_register);
1799 /* Disable data integrity if non-capable/non-matching disk is being added */
1800 void md_integrity_add_rdev(mdk_rdev_t *rdev, mddev_t *mddev)
1802 struct blk_integrity *bi_rdev = bdev_get_integrity(rdev->bdev);
1803 struct blk_integrity *bi_mddev = blk_get_integrity(mddev->gendisk);
1805 if (!bi_mddev) /* nothing to do */
1807 if (rdev->raid_disk < 0) /* skip spares */
1809 if (bi_rdev && blk_integrity_compare(mddev->gendisk,
1810 rdev->bdev->bd_disk) >= 0)
1812 printk(KERN_NOTICE "disabling data integrity on %s\n", mdname(mddev));
1813 blk_integrity_unregister(mddev->gendisk);
1815 EXPORT_SYMBOL(md_integrity_add_rdev);
1817 static int bind_rdev_to_array(mdk_rdev_t * rdev, mddev_t * mddev)
1819 char b[BDEVNAME_SIZE];
1829 /* prevent duplicates */
1830 if (find_rdev(mddev, rdev->bdev->bd_dev))
1833 /* make sure rdev->sectors exceeds mddev->dev_sectors */
1834 if (rdev->sectors && (mddev->dev_sectors == 0 ||
1835 rdev->sectors < mddev->dev_sectors)) {
1837 /* Cannot change size, so fail
1838 * If mddev->level <= 0, then we don't care
1839 * about aligning sizes (e.g. linear)
1841 if (mddev->level > 0)
1844 mddev->dev_sectors = rdev->sectors;
1847 /* Verify rdev->desc_nr is unique.
1848 * If it is -1, assign a free number, else
1849 * check number is not in use
1851 if (rdev->desc_nr < 0) {
1853 if (mddev->pers) choice = mddev->raid_disks;
1854 while (find_rdev_nr(mddev, choice))
1856 rdev->desc_nr = choice;
1858 if (find_rdev_nr(mddev, rdev->desc_nr))
1861 if (mddev->max_disks && rdev->desc_nr >= mddev->max_disks) {
1862 printk(KERN_WARNING "md: %s: array is limited to %d devices\n",
1863 mdname(mddev), mddev->max_disks);
1866 bdevname(rdev->bdev,b);
1867 while ( (s=strchr(b, '/')) != NULL)
1870 rdev->mddev = mddev;
1871 printk(KERN_INFO "md: bind<%s>\n", b);
1873 if ((err = kobject_add(&rdev->kobj, &mddev->kobj, "dev-%s", b)))
1876 ko = &part_to_dev(rdev->bdev->bd_part)->kobj;
1877 if (sysfs_create_link(&rdev->kobj, ko, "block"))
1878 /* failure here is OK */;
1879 rdev->sysfs_state = sysfs_get_dirent_safe(rdev->kobj.sd, "state");
1881 list_add_rcu(&rdev->same_set, &mddev->disks);
1882 bd_claim_by_disk(rdev->bdev, rdev->bdev->bd_holder, mddev->gendisk);
1884 /* May as well allow recovery to be retried once */
1885 mddev->recovery_disabled = 0;
1890 printk(KERN_WARNING "md: failed to register dev-%s for %s\n",
1895 static void md_delayed_delete(struct work_struct *ws)
1897 mdk_rdev_t *rdev = container_of(ws, mdk_rdev_t, del_work);
1898 kobject_del(&rdev->kobj);
1899 kobject_put(&rdev->kobj);
1902 static void unbind_rdev_from_array(mdk_rdev_t * rdev)
1904 char b[BDEVNAME_SIZE];
1909 bd_release_from_disk(rdev->bdev, rdev->mddev->gendisk);
1910 list_del_rcu(&rdev->same_set);
1911 printk(KERN_INFO "md: unbind<%s>\n", bdevname(rdev->bdev,b));
1913 sysfs_remove_link(&rdev->kobj, "block");
1914 sysfs_put(rdev->sysfs_state);
1915 rdev->sysfs_state = NULL;
1916 /* We need to delay this, otherwise we can deadlock when
1917 * writing to 'remove' to "dev/state". We also need
1918 * to delay it due to rcu usage.
1921 INIT_WORK(&rdev->del_work, md_delayed_delete);
1922 kobject_get(&rdev->kobj);
1923 queue_work(md_misc_wq, &rdev->del_work);
1927 * prevent the device from being mounted, repartitioned or
1928 * otherwise reused by a RAID array (or any other kernel
1929 * subsystem), by bd_claiming the device.
1931 static int lock_rdev(mdk_rdev_t *rdev, dev_t dev, int shared)
1934 struct block_device *bdev;
1935 char b[BDEVNAME_SIZE];
1937 bdev = open_by_devnum(dev, FMODE_READ|FMODE_WRITE);
1939 printk(KERN_ERR "md: could not open %s.\n",
1940 __bdevname(dev, b));
1941 return PTR_ERR(bdev);
1943 err = bd_claim(bdev, shared ? (mdk_rdev_t *)lock_rdev : rdev);
1945 printk(KERN_ERR "md: could not bd_claim %s.\n",
1947 blkdev_put(bdev, FMODE_READ|FMODE_WRITE);
1951 set_bit(AllReserved, &rdev->flags);
1956 static void unlock_rdev(mdk_rdev_t *rdev)
1958 struct block_device *bdev = rdev->bdev;
1963 blkdev_put(bdev, FMODE_READ|FMODE_WRITE);
1966 void md_autodetect_dev(dev_t dev);
1968 static void export_rdev(mdk_rdev_t * rdev)
1970 char b[BDEVNAME_SIZE];
1971 printk(KERN_INFO "md: export_rdev(%s)\n",
1972 bdevname(rdev->bdev,b));
1977 if (test_bit(AutoDetected, &rdev->flags))
1978 md_autodetect_dev(rdev->bdev->bd_dev);
1981 kobject_put(&rdev->kobj);
1984 static void kick_rdev_from_array(mdk_rdev_t * rdev)
1986 unbind_rdev_from_array(rdev);
1990 static void export_array(mddev_t *mddev)
1992 mdk_rdev_t *rdev, *tmp;
1994 rdev_for_each(rdev, tmp, mddev) {
1999 kick_rdev_from_array(rdev);
2001 if (!list_empty(&mddev->disks))
2003 mddev->raid_disks = 0;
2004 mddev->major_version = 0;
2007 static void print_desc(mdp_disk_t *desc)
2009 printk(" DISK<N:%d,(%d,%d),R:%d,S:%d>\n", desc->number,
2010 desc->major,desc->minor,desc->raid_disk,desc->state);
2013 static void print_sb_90(mdp_super_t *sb)
2018 "md: SB: (V:%d.%d.%d) ID:<%08x.%08x.%08x.%08x> CT:%08x\n",
2019 sb->major_version, sb->minor_version, sb->patch_version,
2020 sb->set_uuid0, sb->set_uuid1, sb->set_uuid2, sb->set_uuid3,
2022 printk(KERN_INFO "md: L%d S%08d ND:%d RD:%d md%d LO:%d CS:%d\n",
2023 sb->level, sb->size, sb->nr_disks, sb->raid_disks,
2024 sb->md_minor, sb->layout, sb->chunk_size);
2025 printk(KERN_INFO "md: UT:%08x ST:%d AD:%d WD:%d"
2026 " FD:%d SD:%d CSUM:%08x E:%08lx\n",
2027 sb->utime, sb->state, sb->active_disks, sb->working_disks,
2028 sb->failed_disks, sb->spare_disks,
2029 sb->sb_csum, (unsigned long)sb->events_lo);
2032 for (i = 0; i < MD_SB_DISKS; i++) {
2035 desc = sb->disks + i;
2036 if (desc->number || desc->major || desc->minor ||
2037 desc->raid_disk || (desc->state && (desc->state != 4))) {
2038 printk(" D %2d: ", i);
2042 printk(KERN_INFO "md: THIS: ");
2043 print_desc(&sb->this_disk);
2046 static void print_sb_1(struct mdp_superblock_1 *sb)
2050 uuid = sb->set_uuid;
2052 "md: SB: (V:%u) (F:0x%08x) Array-ID:<%pU>\n"
2053 "md: Name: \"%s\" CT:%llu\n",
2054 le32_to_cpu(sb->major_version),
2055 le32_to_cpu(sb->feature_map),
2058 (unsigned long long)le64_to_cpu(sb->ctime)
2059 & MD_SUPERBLOCK_1_TIME_SEC_MASK);
2061 uuid = sb->device_uuid;
2063 "md: L%u SZ%llu RD:%u LO:%u CS:%u DO:%llu DS:%llu SO:%llu"
2065 "md: Dev:%08x UUID: %pU\n"
2066 "md: (F:0x%08x) UT:%llu Events:%llu ResyncOffset:%llu CSUM:0x%08x\n"
2067 "md: (MaxDev:%u) \n",
2068 le32_to_cpu(sb->level),
2069 (unsigned long long)le64_to_cpu(sb->size),
2070 le32_to_cpu(sb->raid_disks),
2071 le32_to_cpu(sb->layout),
2072 le32_to_cpu(sb->chunksize),
2073 (unsigned long long)le64_to_cpu(sb->data_offset),
2074 (unsigned long long)le64_to_cpu(sb->data_size),
2075 (unsigned long long)le64_to_cpu(sb->super_offset),
2076 (unsigned long long)le64_to_cpu(sb->recovery_offset),
2077 le32_to_cpu(sb->dev_number),
2080 (unsigned long long)le64_to_cpu(sb->utime) & MD_SUPERBLOCK_1_TIME_SEC_MASK,
2081 (unsigned long long)le64_to_cpu(sb->events),
2082 (unsigned long long)le64_to_cpu(sb->resync_offset),
2083 le32_to_cpu(sb->sb_csum),
2084 le32_to_cpu(sb->max_dev)
2088 static void print_rdev(mdk_rdev_t *rdev, int major_version)
2090 char b[BDEVNAME_SIZE];
2091 printk(KERN_INFO "md: rdev %s, Sect:%08llu F:%d S:%d DN:%u\n",
2092 bdevname(rdev->bdev, b), (unsigned long long)rdev->sectors,
2093 test_bit(Faulty, &rdev->flags), test_bit(In_sync, &rdev->flags),
2095 if (rdev->sb_loaded) {
2096 printk(KERN_INFO "md: rdev superblock (MJ:%d):\n", major_version);
2097 switch (major_version) {
2099 print_sb_90((mdp_super_t*)page_address(rdev->sb_page));
2102 print_sb_1((struct mdp_superblock_1 *)page_address(rdev->sb_page));
2106 printk(KERN_INFO "md: no rdev superblock!\n");
2109 static void md_print_devices(void)
2111 struct list_head *tmp;
2114 char b[BDEVNAME_SIZE];
2117 printk("md: **********************************\n");
2118 printk("md: * <COMPLETE RAID STATE PRINTOUT> *\n");
2119 printk("md: **********************************\n");
2120 for_each_mddev(mddev, tmp) {
2123 bitmap_print_sb(mddev->bitmap);
2125 printk("%s: ", mdname(mddev));
2126 list_for_each_entry(rdev, &mddev->disks, same_set)
2127 printk("<%s>", bdevname(rdev->bdev,b));
2130 list_for_each_entry(rdev, &mddev->disks, same_set)
2131 print_rdev(rdev, mddev->major_version);
2133 printk("md: **********************************\n");
2138 static void sync_sbs(mddev_t * mddev, int nospares)
2140 /* Update each superblock (in-memory image), but
2141 * if we are allowed to, skip spares which already
2142 * have the right event counter, or have one earlier
2143 * (which would mean they aren't being marked as dirty
2144 * with the rest of the array)
2147 list_for_each_entry(rdev, &mddev->disks, same_set) {
2148 if (rdev->sb_events == mddev->events ||
2150 rdev->raid_disk < 0 &&
2151 rdev->sb_events+1 == mddev->events)) {
2152 /* Don't update this superblock */
2153 rdev->sb_loaded = 2;
2155 super_types[mddev->major_version].
2156 sync_super(mddev, rdev);
2157 rdev->sb_loaded = 1;
2162 static void md_update_sb(mddev_t * mddev, int force_change)
2169 /* First make sure individual recovery_offsets are correct */
2170 list_for_each_entry(rdev, &mddev->disks, same_set) {
2171 if (rdev->raid_disk >= 0 &&
2172 mddev->delta_disks >= 0 &&
2173 !test_bit(In_sync, &rdev->flags) &&
2174 mddev->curr_resync_completed > rdev->recovery_offset)
2175 rdev->recovery_offset = mddev->curr_resync_completed;
2178 if (!mddev->persistent) {
2179 clear_bit(MD_CHANGE_CLEAN, &mddev->flags);
2180 clear_bit(MD_CHANGE_DEVS, &mddev->flags);
2181 if (!mddev->external)
2182 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
2183 wake_up(&mddev->sb_wait);
2187 spin_lock_irq(&mddev->write_lock);
2189 mddev->utime = get_seconds();
2191 if (test_and_clear_bit(MD_CHANGE_DEVS, &mddev->flags))
2193 if (test_and_clear_bit(MD_CHANGE_CLEAN, &mddev->flags))
2194 /* just a clean<-> dirty transition, possibly leave spares alone,
2195 * though if events isn't the right even/odd, we will have to do
2201 if (mddev->degraded)
2202 /* If the array is degraded, then skipping spares is both
2203 * dangerous and fairly pointless.
2204 * Dangerous because a device that was removed from the array
2205 * might have a event_count that still looks up-to-date,
2206 * so it can be re-added without a resync.
2207 * Pointless because if there are any spares to skip,
2208 * then a recovery will happen and soon that array won't
2209 * be degraded any more and the spare can go back to sleep then.
2213 sync_req = mddev->in_sync;
2215 /* If this is just a dirty<->clean transition, and the array is clean
2216 * and 'events' is odd, we can roll back to the previous clean state */
2218 && (mddev->in_sync && mddev->recovery_cp == MaxSector)
2219 && mddev->can_decrease_events
2220 && mddev->events != 1) {
2222 mddev->can_decrease_events = 0;
2224 /* otherwise we have to go forward and ... */
2226 mddev->can_decrease_events = nospares;
2229 if (!mddev->events) {
2231 * oops, this 64-bit counter should never wrap.
2232 * Either we are in around ~1 trillion A.C., assuming
2233 * 1 reboot per second, or we have a bug:
2238 sync_sbs(mddev, nospares);
2239 spin_unlock_irq(&mddev->write_lock);
2242 "md: updating %s RAID superblock on device (in sync %d)\n",
2243 mdname(mddev),mddev->in_sync);
2245 bitmap_update_sb(mddev->bitmap);
2246 list_for_each_entry(rdev, &mddev->disks, same_set) {
2247 char b[BDEVNAME_SIZE];
2248 dprintk(KERN_INFO "md: ");
2249 if (rdev->sb_loaded != 1)
2250 continue; /* no noise on spare devices */
2251 if (test_bit(Faulty, &rdev->flags))
2252 dprintk("(skipping faulty ");
2254 dprintk("%s ", bdevname(rdev->bdev,b));
2255 if (!test_bit(Faulty, &rdev->flags)) {
2256 md_super_write(mddev,rdev,
2257 rdev->sb_start, rdev->sb_size,
2259 dprintk(KERN_INFO "(write) %s's sb offset: %llu\n",
2260 bdevname(rdev->bdev,b),
2261 (unsigned long long)rdev->sb_start);
2262 rdev->sb_events = mddev->events;
2266 if (mddev->level == LEVEL_MULTIPATH)
2267 /* only need to write one superblock... */
2270 md_super_wait(mddev);
2271 /* if there was a failure, MD_CHANGE_DEVS was set, and we re-write super */
2273 spin_lock_irq(&mddev->write_lock);
2274 if (mddev->in_sync != sync_req ||
2275 test_bit(MD_CHANGE_DEVS, &mddev->flags)) {
2276 /* have to write it out again */
2277 spin_unlock_irq(&mddev->write_lock);
2280 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
2281 spin_unlock_irq(&mddev->write_lock);
2282 wake_up(&mddev->sb_wait);
2283 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
2284 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
2288 /* words written to sysfs files may, or may not, be \n terminated.
2289 * We want to accept with case. For this we use cmd_match.
2291 static int cmd_match(const char *cmd, const char *str)
2293 /* See if cmd, written into a sysfs file, matches
2294 * str. They must either be the same, or cmd can
2295 * have a trailing newline
2297 while (*cmd && *str && *cmd == *str) {
2308 struct rdev_sysfs_entry {
2309 struct attribute attr;
2310 ssize_t (*show)(mdk_rdev_t *, char *);
2311 ssize_t (*store)(mdk_rdev_t *, const char *, size_t);
2315 state_show(mdk_rdev_t *rdev, char *page)
2320 if (test_bit(Faulty, &rdev->flags)) {
2321 len+= sprintf(page+len, "%sfaulty",sep);
2324 if (test_bit(In_sync, &rdev->flags)) {
2325 len += sprintf(page+len, "%sin_sync",sep);
2328 if (test_bit(WriteMostly, &rdev->flags)) {
2329 len += sprintf(page+len, "%swrite_mostly",sep);
2332 if (test_bit(Blocked, &rdev->flags)) {
2333 len += sprintf(page+len, "%sblocked", sep);
2336 if (!test_bit(Faulty, &rdev->flags) &&
2337 !test_bit(In_sync, &rdev->flags)) {
2338 len += sprintf(page+len, "%sspare", sep);
2341 return len+sprintf(page+len, "\n");
2345 state_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2348 * faulty - simulates and error
2349 * remove - disconnects the device
2350 * writemostly - sets write_mostly
2351 * -writemostly - clears write_mostly
2352 * blocked - sets the Blocked flag
2353 * -blocked - clears the Blocked flag
2354 * insync - sets Insync providing device isn't active
2357 if (cmd_match(buf, "faulty") && rdev->mddev->pers) {
2358 md_error(rdev->mddev, rdev);
2360 } else if (cmd_match(buf, "remove")) {
2361 if (rdev->raid_disk >= 0)
2364 mddev_t *mddev = rdev->mddev;
2365 kick_rdev_from_array(rdev);
2367 md_update_sb(mddev, 1);
2368 md_new_event(mddev);
2371 } else if (cmd_match(buf, "writemostly")) {
2372 set_bit(WriteMostly, &rdev->flags);
2374 } else if (cmd_match(buf, "-writemostly")) {
2375 clear_bit(WriteMostly, &rdev->flags);
2377 } else if (cmd_match(buf, "blocked")) {
2378 set_bit(Blocked, &rdev->flags);
2380 } else if (cmd_match(buf, "-blocked")) {
2381 clear_bit(Blocked, &rdev->flags);
2382 wake_up(&rdev->blocked_wait);
2383 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2384 md_wakeup_thread(rdev->mddev->thread);
2387 } else if (cmd_match(buf, "insync") && rdev->raid_disk == -1) {
2388 set_bit(In_sync, &rdev->flags);
2392 sysfs_notify_dirent_safe(rdev->sysfs_state);
2393 return err ? err : len;
2395 static struct rdev_sysfs_entry rdev_state =
2396 __ATTR(state, S_IRUGO|S_IWUSR, state_show, state_store);
2399 errors_show(mdk_rdev_t *rdev, char *page)
2401 return sprintf(page, "%d\n", atomic_read(&rdev->corrected_errors));
2405 errors_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2408 unsigned long n = simple_strtoul(buf, &e, 10);
2409 if (*buf && (*e == 0 || *e == '\n')) {
2410 atomic_set(&rdev->corrected_errors, n);
2415 static struct rdev_sysfs_entry rdev_errors =
2416 __ATTR(errors, S_IRUGO|S_IWUSR, errors_show, errors_store);
2419 slot_show(mdk_rdev_t *rdev, char *page)
2421 if (rdev->raid_disk < 0)
2422 return sprintf(page, "none\n");
2424 return sprintf(page, "%d\n", rdev->raid_disk);
2428 slot_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2433 int slot = simple_strtoul(buf, &e, 10);
2434 if (strncmp(buf, "none", 4)==0)
2436 else if (e==buf || (*e && *e!= '\n'))
2438 if (rdev->mddev->pers && slot == -1) {
2439 /* Setting 'slot' on an active array requires also
2440 * updating the 'rd%d' link, and communicating
2441 * with the personality with ->hot_*_disk.
2442 * For now we only support removing
2443 * failed/spare devices. This normally happens automatically,
2444 * but not when the metadata is externally managed.
2446 if (rdev->raid_disk == -1)
2448 /* personality does all needed checks */
2449 if (rdev->mddev->pers->hot_add_disk == NULL)
2451 err = rdev->mddev->pers->
2452 hot_remove_disk(rdev->mddev, rdev->raid_disk);
2455 sprintf(nm, "rd%d", rdev->raid_disk);
2456 sysfs_remove_link(&rdev->mddev->kobj, nm);
2457 rdev->raid_disk = -1;
2458 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2459 md_wakeup_thread(rdev->mddev->thread);
2460 } else if (rdev->mddev->pers) {
2462 /* Activating a spare .. or possibly reactivating
2463 * if we ever get bitmaps working here.
2466 if (rdev->raid_disk != -1)
2469 if (rdev->mddev->pers->hot_add_disk == NULL)
2472 list_for_each_entry(rdev2, &rdev->mddev->disks, same_set)
2473 if (rdev2->raid_disk == slot)
2476 rdev->raid_disk = slot;
2477 if (test_bit(In_sync, &rdev->flags))
2478 rdev->saved_raid_disk = slot;
2480 rdev->saved_raid_disk = -1;
2481 err = rdev->mddev->pers->
2482 hot_add_disk(rdev->mddev, rdev);
2484 rdev->raid_disk = -1;
2487 sysfs_notify_dirent_safe(rdev->sysfs_state);
2488 sprintf(nm, "rd%d", rdev->raid_disk);
2489 if (sysfs_create_link(&rdev->mddev->kobj, &rdev->kobj, nm))
2490 /* failure here is OK */;
2491 /* don't wakeup anyone, leave that to userspace. */
2493 if (slot >= rdev->mddev->raid_disks)
2495 rdev->raid_disk = slot;
2496 /* assume it is working */
2497 clear_bit(Faulty, &rdev->flags);
2498 clear_bit(WriteMostly, &rdev->flags);
2499 set_bit(In_sync, &rdev->flags);
2500 sysfs_notify_dirent_safe(rdev->sysfs_state);
2506 static struct rdev_sysfs_entry rdev_slot =
2507 __ATTR(slot, S_IRUGO|S_IWUSR, slot_show, slot_store);
2510 offset_show(mdk_rdev_t *rdev, char *page)
2512 return sprintf(page, "%llu\n", (unsigned long long)rdev->data_offset);
2516 offset_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2519 unsigned long long offset = simple_strtoull(buf, &e, 10);
2520 if (e==buf || (*e && *e != '\n'))
2522 if (rdev->mddev->pers && rdev->raid_disk >= 0)
2524 if (rdev->sectors && rdev->mddev->external)
2525 /* Must set offset before size, so overlap checks
2528 rdev->data_offset = offset;
2532 static struct rdev_sysfs_entry rdev_offset =
2533 __ATTR(offset, S_IRUGO|S_IWUSR, offset_show, offset_store);
2536 rdev_size_show(mdk_rdev_t *rdev, char *page)
2538 return sprintf(page, "%llu\n", (unsigned long long)rdev->sectors / 2);
2541 static int overlaps(sector_t s1, sector_t l1, sector_t s2, sector_t l2)
2543 /* check if two start/length pairs overlap */
2551 static int strict_blocks_to_sectors(const char *buf, sector_t *sectors)
2553 unsigned long long blocks;
2556 if (strict_strtoull(buf, 10, &blocks) < 0)
2559 if (blocks & 1ULL << (8 * sizeof(blocks) - 1))
2560 return -EINVAL; /* sector conversion overflow */
2563 if (new != blocks * 2)
2564 return -EINVAL; /* unsigned long long to sector_t overflow */
2571 rdev_size_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2573 mddev_t *my_mddev = rdev->mddev;
2574 sector_t oldsectors = rdev->sectors;
2577 if (strict_blocks_to_sectors(buf, §ors) < 0)
2579 if (my_mddev->pers && rdev->raid_disk >= 0) {
2580 if (my_mddev->persistent) {
2581 sectors = super_types[my_mddev->major_version].
2582 rdev_size_change(rdev, sectors);
2585 } else if (!sectors)
2586 sectors = (i_size_read(rdev->bdev->bd_inode) >> 9) -
2589 if (sectors < my_mddev->dev_sectors)
2590 return -EINVAL; /* component must fit device */
2592 rdev->sectors = sectors;
2593 if (sectors > oldsectors && my_mddev->external) {
2594 /* need to check that all other rdevs with the same ->bdev
2595 * do not overlap. We need to unlock the mddev to avoid
2596 * a deadlock. We have already changed rdev->sectors, and if
2597 * we have to change it back, we will have the lock again.
2601 struct list_head *tmp;
2603 mddev_unlock(my_mddev);
2604 for_each_mddev(mddev, tmp) {
2608 list_for_each_entry(rdev2, &mddev->disks, same_set)
2609 if (test_bit(AllReserved, &rdev2->flags) ||
2610 (rdev->bdev == rdev2->bdev &&
2612 overlaps(rdev->data_offset, rdev->sectors,
2618 mddev_unlock(mddev);
2624 mddev_lock(my_mddev);
2626 /* Someone else could have slipped in a size
2627 * change here, but doing so is just silly.
2628 * We put oldsectors back because we *know* it is
2629 * safe, and trust userspace not to race with
2632 rdev->sectors = oldsectors;
2639 static struct rdev_sysfs_entry rdev_size =
2640 __ATTR(size, S_IRUGO|S_IWUSR, rdev_size_show, rdev_size_store);
2643 static ssize_t recovery_start_show(mdk_rdev_t *rdev, char *page)
2645 unsigned long long recovery_start = rdev->recovery_offset;
2647 if (test_bit(In_sync, &rdev->flags) ||
2648 recovery_start == MaxSector)
2649 return sprintf(page, "none\n");
2651 return sprintf(page, "%llu\n", recovery_start);
2654 static ssize_t recovery_start_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2656 unsigned long long recovery_start;
2658 if (cmd_match(buf, "none"))
2659 recovery_start = MaxSector;
2660 else if (strict_strtoull(buf, 10, &recovery_start))
2663 if (rdev->mddev->pers &&
2664 rdev->raid_disk >= 0)
2667 rdev->recovery_offset = recovery_start;
2668 if (recovery_start == MaxSector)
2669 set_bit(In_sync, &rdev->flags);
2671 clear_bit(In_sync, &rdev->flags);
2675 static struct rdev_sysfs_entry rdev_recovery_start =
2676 __ATTR(recovery_start, S_IRUGO|S_IWUSR, recovery_start_show, recovery_start_store);
2678 static struct attribute *rdev_default_attrs[] = {
2684 &rdev_recovery_start.attr,
2688 rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
2690 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
2691 mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
2692 mddev_t *mddev = rdev->mddev;
2698 rv = mddev ? mddev_lock(mddev) : -EBUSY;
2700 if (rdev->mddev == NULL)
2703 rv = entry->show(rdev, page);
2704 mddev_unlock(mddev);
2710 rdev_attr_store(struct kobject *kobj, struct attribute *attr,
2711 const char *page, size_t length)
2713 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
2714 mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
2716 mddev_t *mddev = rdev->mddev;
2720 if (!capable(CAP_SYS_ADMIN))
2722 rv = mddev ? mddev_lock(mddev): -EBUSY;
2724 if (rdev->mddev == NULL)
2727 rv = entry->store(rdev, page, length);
2728 mddev_unlock(mddev);
2733 static void rdev_free(struct kobject *ko)
2735 mdk_rdev_t *rdev = container_of(ko, mdk_rdev_t, kobj);
2738 static const struct sysfs_ops rdev_sysfs_ops = {
2739 .show = rdev_attr_show,
2740 .store = rdev_attr_store,
2742 static struct kobj_type rdev_ktype = {
2743 .release = rdev_free,
2744 .sysfs_ops = &rdev_sysfs_ops,
2745 .default_attrs = rdev_default_attrs,
2748 void md_rdev_init(mdk_rdev_t *rdev)
2751 rdev->saved_raid_disk = -1;
2752 rdev->raid_disk = -1;
2754 rdev->data_offset = 0;
2755 rdev->sb_events = 0;
2756 rdev->last_read_error.tv_sec = 0;
2757 rdev->last_read_error.tv_nsec = 0;
2758 atomic_set(&rdev->nr_pending, 0);
2759 atomic_set(&rdev->read_errors, 0);
2760 atomic_set(&rdev->corrected_errors, 0);
2762 INIT_LIST_HEAD(&rdev->same_set);
2763 init_waitqueue_head(&rdev->blocked_wait);
2765 EXPORT_SYMBOL_GPL(md_rdev_init);
2767 * Import a device. If 'super_format' >= 0, then sanity check the superblock
2769 * mark the device faulty if:
2771 * - the device is nonexistent (zero size)
2772 * - the device has no valid superblock
2774 * a faulty rdev _never_ has rdev->sb set.
2776 static mdk_rdev_t *md_import_device(dev_t newdev, int super_format, int super_minor)
2778 char b[BDEVNAME_SIZE];
2783 rdev = kzalloc(sizeof(*rdev), GFP_KERNEL);
2785 printk(KERN_ERR "md: could not alloc mem for new device!\n");
2786 return ERR_PTR(-ENOMEM);
2790 if ((err = alloc_disk_sb(rdev)))
2793 err = lock_rdev(rdev, newdev, super_format == -2);
2797 kobject_init(&rdev->kobj, &rdev_ktype);
2799 size = i_size_read(rdev->bdev->bd_inode) >> BLOCK_SIZE_BITS;
2802 "md: %s has zero or unknown size, marking faulty!\n",
2803 bdevname(rdev->bdev,b));
2808 if (super_format >= 0) {
2809 err = super_types[super_format].
2810 load_super(rdev, NULL, super_minor);
2811 if (err == -EINVAL) {
2813 "md: %s does not have a valid v%d.%d "
2814 "superblock, not importing!\n",
2815 bdevname(rdev->bdev,b),
2816 super_format, super_minor);
2821 "md: could not read %s's sb, not importing!\n",
2822 bdevname(rdev->bdev,b));
2830 if (rdev->sb_page) {
2836 return ERR_PTR(err);
2840 * Check a full RAID array for plausibility
2844 static void analyze_sbs(mddev_t * mddev)
2847 mdk_rdev_t *rdev, *freshest, *tmp;
2848 char b[BDEVNAME_SIZE];
2851 rdev_for_each(rdev, tmp, mddev)
2852 switch (super_types[mddev->major_version].
2853 load_super(rdev, freshest, mddev->minor_version)) {
2861 "md: fatal superblock inconsistency in %s"
2862 " -- removing from array\n",
2863 bdevname(rdev->bdev,b));
2864 kick_rdev_from_array(rdev);
2868 super_types[mddev->major_version].
2869 validate_super(mddev, freshest);
2872 rdev_for_each(rdev, tmp, mddev) {
2873 if (mddev->max_disks &&
2874 (rdev->desc_nr >= mddev->max_disks ||
2875 i > mddev->max_disks)) {
2877 "md: %s: %s: only %d devices permitted\n",
2878 mdname(mddev), bdevname(rdev->bdev, b),
2880 kick_rdev_from_array(rdev);
2883 if (rdev != freshest)
2884 if (super_types[mddev->major_version].
2885 validate_super(mddev, rdev)) {
2886 printk(KERN_WARNING "md: kicking non-fresh %s"
2888 bdevname(rdev->bdev,b));
2889 kick_rdev_from_array(rdev);
2892 if (mddev->level == LEVEL_MULTIPATH) {
2893 rdev->desc_nr = i++;
2894 rdev->raid_disk = rdev->desc_nr;
2895 set_bit(In_sync, &rdev->flags);
2896 } else if (rdev->raid_disk >= (mddev->raid_disks - min(0, mddev->delta_disks))) {
2897 rdev->raid_disk = -1;
2898 clear_bit(In_sync, &rdev->flags);
2903 /* Read a fixed-point number.
2904 * Numbers in sysfs attributes should be in "standard" units where
2905 * possible, so time should be in seconds.
2906 * However we internally use a a much smaller unit such as
2907 * milliseconds or jiffies.
2908 * This function takes a decimal number with a possible fractional
2909 * component, and produces an integer which is the result of
2910 * multiplying that number by 10^'scale'.
2911 * all without any floating-point arithmetic.
2913 int strict_strtoul_scaled(const char *cp, unsigned long *res, int scale)
2915 unsigned long result = 0;
2917 while (isdigit(*cp) || (*cp == '.' && decimals < 0)) {
2920 else if (decimals < scale) {
2923 result = result * 10 + value;
2935 while (decimals < scale) {
2944 static void md_safemode_timeout(unsigned long data);
2947 safe_delay_show(mddev_t *mddev, char *page)
2949 int msec = (mddev->safemode_delay*1000)/HZ;
2950 return sprintf(page, "%d.%03d\n", msec/1000, msec%1000);
2953 safe_delay_store(mddev_t *mddev, const char *cbuf, size_t len)
2957 if (strict_strtoul_scaled(cbuf, &msec, 3) < 0)
2960 mddev->safemode_delay = 0;
2962 unsigned long old_delay = mddev->safemode_delay;
2963 mddev->safemode_delay = (msec*HZ)/1000;
2964 if (mddev->safemode_delay == 0)
2965 mddev->safemode_delay = 1;
2966 if (mddev->safemode_delay < old_delay)
2967 md_safemode_timeout((unsigned long)mddev);
2971 static struct md_sysfs_entry md_safe_delay =
2972 __ATTR(safe_mode_delay, S_IRUGO|S_IWUSR,safe_delay_show, safe_delay_store);
2975 level_show(mddev_t *mddev, char *page)
2977 struct mdk_personality *p = mddev->pers;
2979 return sprintf(page, "%s\n", p->name);
2980 else if (mddev->clevel[0])
2981 return sprintf(page, "%s\n", mddev->clevel);
2982 else if (mddev->level != LEVEL_NONE)
2983 return sprintf(page, "%d\n", mddev->level);
2989 level_store(mddev_t *mddev, const char *buf, size_t len)
2993 struct mdk_personality *pers;
2998 if (mddev->pers == NULL) {
3001 if (len >= sizeof(mddev->clevel))
3003 strncpy(mddev->clevel, buf, len);
3004 if (mddev->clevel[len-1] == '\n')
3006 mddev->clevel[len] = 0;
3007 mddev->level = LEVEL_NONE;
3011 /* request to change the personality. Need to ensure:
3012 * - array is not engaged in resync/recovery/reshape
3013 * - old personality can be suspended
3014 * - new personality will access other array.
3017 if (mddev->sync_thread ||
3018 mddev->reshape_position != MaxSector ||
3019 mddev->sysfs_active)
3022 if (!mddev->pers->quiesce) {
3023 printk(KERN_WARNING "md: %s: %s does not support online personality change\n",
3024 mdname(mddev), mddev->pers->name);
3028 /* Now find the new personality */
3029 if (len == 0 || len >= sizeof(clevel))
3031 strncpy(clevel, buf, len);
3032 if (clevel[len-1] == '\n')
3035 if (strict_strtol(clevel, 10, &level))
3038 if (request_module("md-%s", clevel) != 0)
3039 request_module("md-level-%s", clevel);
3040 spin_lock(&pers_lock);
3041 pers = find_pers(level, clevel);
3042 if (!pers || !try_module_get(pers->owner)) {
3043 spin_unlock(&pers_lock);
3044 printk(KERN_WARNING "md: personality %s not loaded\n", clevel);
3047 spin_unlock(&pers_lock);
3049 if (pers == mddev->pers) {
3050 /* Nothing to do! */
3051 module_put(pers->owner);
3054 if (!pers->takeover) {
3055 module_put(pers->owner);
3056 printk(KERN_WARNING "md: %s: %s does not support personality takeover\n",
3057 mdname(mddev), clevel);
3061 list_for_each_entry(rdev, &mddev->disks, same_set)
3062 rdev->new_raid_disk = rdev->raid_disk;
3064 /* ->takeover must set new_* and/or delta_disks
3065 * if it succeeds, and may set them when it fails.
3067 priv = pers->takeover(mddev);
3069 mddev->new_level = mddev->level;
3070 mddev->new_layout = mddev->layout;
3071 mddev->new_chunk_sectors = mddev->chunk_sectors;
3072 mddev->raid_disks -= mddev->delta_disks;
3073 mddev->delta_disks = 0;
3074 module_put(pers->owner);
3075 printk(KERN_WARNING "md: %s: %s would not accept array\n",
3076 mdname(mddev), clevel);
3077 return PTR_ERR(priv);
3080 /* Looks like we have a winner */
3081 mddev_suspend(mddev);
3082 mddev->pers->stop(mddev);
3084 if (mddev->pers->sync_request == NULL &&
3085 pers->sync_request != NULL) {
3086 /* need to add the md_redundancy_group */
3087 if (sysfs_create_group(&mddev->kobj, &md_redundancy_group))
3089 "md: cannot register extra attributes for %s\n",
3091 mddev->sysfs_action = sysfs_get_dirent(mddev->kobj.sd, NULL, "sync_action");
3093 if (mddev->pers->sync_request != NULL &&
3094 pers->sync_request == NULL) {
3095 /* need to remove the md_redundancy_group */
3096 if (mddev->to_remove == NULL)
3097 mddev->to_remove = &md_redundancy_group;
3100 if (mddev->pers->sync_request == NULL &&
3102 /* We are converting from a no-redundancy array
3103 * to a redundancy array and metadata is managed
3104 * externally so we need to be sure that writes
3105 * won't block due to a need to transition
3107 * until external management is started.
3110 mddev->safemode_delay = 0;
3111 mddev->safemode = 0;
3114 list_for_each_entry(rdev, &mddev->disks, same_set) {
3116 if (rdev->raid_disk < 0)
3118 if (rdev->new_raid_disk > mddev->raid_disks)
3119 rdev->new_raid_disk = -1;
3120 if (rdev->new_raid_disk == rdev->raid_disk)
3122 sprintf(nm, "rd%d", rdev->raid_disk);
3123 sysfs_remove_link(&mddev->kobj, nm);
3125 list_for_each_entry(rdev, &mddev->disks, same_set) {
3126 if (rdev->raid_disk < 0)
3128 if (rdev->new_raid_disk == rdev->raid_disk)
3130 rdev->raid_disk = rdev->new_raid_disk;
3131 if (rdev->raid_disk < 0)
3132 clear_bit(In_sync, &rdev->flags);
3135 sprintf(nm, "rd%d", rdev->raid_disk);
3136 if(sysfs_create_link(&mddev->kobj, &rdev->kobj, nm))
3137 printk("md: cannot register %s for %s after level change\n",
3142 module_put(mddev->pers->owner);
3144 mddev->private = priv;
3145 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
3146 mddev->level = mddev->new_level;
3147 mddev->layout = mddev->new_layout;
3148 mddev->chunk_sectors = mddev->new_chunk_sectors;
3149 mddev->delta_disks = 0;
3150 if (mddev->pers->sync_request == NULL) {
3151 /* this is now an array without redundancy, so
3152 * it must always be in_sync
3155 del_timer_sync(&mddev->safemode_timer);
3158 mddev_resume(mddev);
3159 set_bit(MD_CHANGE_DEVS, &mddev->flags);
3160 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3161 md_wakeup_thread(mddev->thread);
3162 sysfs_notify(&mddev->kobj, NULL, "level");
3163 md_new_event(mddev);
3167 static struct md_sysfs_entry md_level =
3168 __ATTR(level, S_IRUGO|S_IWUSR, level_show, level_store);
3172 layout_show(mddev_t *mddev, char *page)
3174 /* just a number, not meaningful for all levels */
3175 if (mddev->reshape_position != MaxSector &&
3176 mddev->layout != mddev->new_layout)
3177 return sprintf(page, "%d (%d)\n",
3178 mddev->new_layout, mddev->layout);
3179 return sprintf(page, "%d\n", mddev->layout);
3183 layout_store(mddev_t *mddev, const char *buf, size_t len)
3186 unsigned long n = simple_strtoul(buf, &e, 10);
3188 if (!*buf || (*e && *e != '\n'))
3193 if (mddev->pers->check_reshape == NULL)
3195 mddev->new_layout = n;
3196 err = mddev->pers->check_reshape(mddev);
3198 mddev->new_layout = mddev->layout;
3202 mddev->new_layout = n;
3203 if (mddev->reshape_position == MaxSector)
3208 static struct md_sysfs_entry md_layout =
3209 __ATTR(layout, S_IRUGO|S_IWUSR, layout_show, layout_store);
3213 raid_disks_show(mddev_t *mddev, char *page)
3215 if (mddev->raid_disks == 0)
3217 if (mddev->reshape_position != MaxSector &&
3218 mddev->delta_disks != 0)
3219 return sprintf(page, "%d (%d)\n", mddev->raid_disks,
3220 mddev->raid_disks - mddev->delta_disks);
3221 return sprintf(page, "%d\n", mddev->raid_disks);
3224 static int update_raid_disks(mddev_t *mddev, int raid_disks);
3227 raid_disks_store(mddev_t *mddev, const char *buf, size_t len)
3231 unsigned long n = simple_strtoul(buf, &e, 10);
3233 if (!*buf || (*e && *e != '\n'))
3237 rv = update_raid_disks(mddev, n);
3238 else if (mddev->reshape_position != MaxSector) {
3239 int olddisks = mddev->raid_disks - mddev->delta_disks;
3240 mddev->delta_disks = n - olddisks;
3241 mddev->raid_disks = n;
3243 mddev->raid_disks = n;
3244 return rv ? rv : len;
3246 static struct md_sysfs_entry md_raid_disks =
3247 __ATTR(raid_disks, S_IRUGO|S_IWUSR, raid_disks_show, raid_disks_store);
3250 chunk_size_show(mddev_t *mddev, char *page)
3252 if (mddev->reshape_position != MaxSector &&
3253 mddev->chunk_sectors != mddev->new_chunk_sectors)
3254 return sprintf(page, "%d (%d)\n",
3255 mddev->new_chunk_sectors << 9,
3256 mddev->chunk_sectors << 9);
3257 return sprintf(page, "%d\n", mddev->chunk_sectors << 9);
3261 chunk_size_store(mddev_t *mddev, const char *buf, size_t len)
3264 unsigned long n = simple_strtoul(buf, &e, 10);
3266 if (!*buf || (*e && *e != '\n'))
3271 if (mddev->pers->check_reshape == NULL)
3273 mddev->new_chunk_sectors = n >> 9;
3274 err = mddev->pers->check_reshape(mddev);
3276 mddev->new_chunk_sectors = mddev->chunk_sectors;
3280 mddev->new_chunk_sectors = n >> 9;
3281 if (mddev->reshape_position == MaxSector)
3282 mddev->chunk_sectors = n >> 9;
3286 static struct md_sysfs_entry md_chunk_size =
3287 __ATTR(chunk_size, S_IRUGO|S_IWUSR, chunk_size_show, chunk_size_store);
3290 resync_start_show(mddev_t *mddev, char *page)
3292 if (mddev->recovery_cp == MaxSector)
3293 return sprintf(page, "none\n");
3294 return sprintf(page, "%llu\n", (unsigned long long)mddev->recovery_cp);
3298 resync_start_store(mddev_t *mddev, const char *buf, size_t len)
3301 unsigned long long n = simple_strtoull(buf, &e, 10);
3305 if (cmd_match(buf, "none"))
3307 else if (!*buf || (*e && *e != '\n'))
3310 mddev->recovery_cp = n;
3313 static struct md_sysfs_entry md_resync_start =
3314 __ATTR(resync_start, S_IRUGO|S_IWUSR, resync_start_show, resync_start_store);
3317 * The array state can be:
3320 * No devices, no size, no level
3321 * Equivalent to STOP_ARRAY ioctl
3323 * May have some settings, but array is not active
3324 * all IO results in error
3325 * When written, doesn't tear down array, but just stops it
3326 * suspended (not supported yet)
3327 * All IO requests will block. The array can be reconfigured.
3328 * Writing this, if accepted, will block until array is quiescent
3330 * no resync can happen. no superblocks get written.
3331 * write requests fail
3333 * like readonly, but behaves like 'clean' on a write request.
3335 * clean - no pending writes, but otherwise active.
3336 * When written to inactive array, starts without resync
3337 * If a write request arrives then
3338 * if metadata is known, mark 'dirty' and switch to 'active'.
3339 * if not known, block and switch to write-pending
3340 * If written to an active array that has pending writes, then fails.
3342 * fully active: IO and resync can be happening.
3343 * When written to inactive array, starts with resync
3346 * clean, but writes are blocked waiting for 'active' to be written.
3349 * like active, but no writes have been seen for a while (100msec).
3352 enum array_state { clear, inactive, suspended, readonly, read_auto, clean, active,
3353 write_pending, active_idle, bad_word};
3354 static char *array_states[] = {
3355 "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active",
3356 "write-pending", "active-idle", NULL };
3358 static int match_word(const char *word, char **list)
3361 for (n=0; list[n]; n++)
3362 if (cmd_match(word, list[n]))
3368 array_state_show(mddev_t *mddev, char *page)
3370 enum array_state st = inactive;
3383 else if (test_bit(MD_CHANGE_PENDING, &mddev->flags))
3385 else if (mddev->safemode)
3391 if (list_empty(&mddev->disks) &&
3392 mddev->raid_disks == 0 &&
3393 mddev->dev_sectors == 0)
3398 return sprintf(page, "%s\n", array_states[st]);
3401 static int do_md_stop(mddev_t * mddev, int ro, int is_open);
3402 static int md_set_readonly(mddev_t * mddev, int is_open);
3403 static int do_md_run(mddev_t * mddev);
3404 static int restart_array(mddev_t *mddev);
3407 array_state_store(mddev_t *mddev, const char *buf, size_t len)
3410 enum array_state st = match_word(buf, array_states);
3415 /* stopping an active array */
3416 if (atomic_read(&mddev->openers) > 0)
3418 err = do_md_stop(mddev, 0, 0);
3421 /* stopping an active array */
3423 if (atomic_read(&mddev->openers) > 0)
3425 err = do_md_stop(mddev, 2, 0);
3427 err = 0; /* already inactive */
3430 break; /* not supported yet */
3433 err = md_set_readonly(mddev, 0);
3436 set_disk_ro(mddev->gendisk, 1);
3437 err = do_md_run(mddev);
3443 err = md_set_readonly(mddev, 0);
3444 else if (mddev->ro == 1)
3445 err = restart_array(mddev);
3448 set_disk_ro(mddev->gendisk, 0);
3452 err = do_md_run(mddev);
3457 restart_array(mddev);
3458 spin_lock_irq(&mddev->write_lock);
3459 if (atomic_read(&mddev->writes_pending) == 0) {
3460 if (mddev->in_sync == 0) {
3462 if (mddev->safemode == 1)
3463 mddev->safemode = 0;
3464 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
3469 spin_unlock_irq(&mddev->write_lock);
3475 restart_array(mddev);
3476 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
3477 wake_up(&mddev->sb_wait);
3481 set_disk_ro(mddev->gendisk, 0);
3482 err = do_md_run(mddev);
3487 /* these cannot be set */
3493 sysfs_notify_dirent_safe(mddev->sysfs_state);
3497 static struct md_sysfs_entry md_array_state =
3498 __ATTR(array_state, S_IRUGO|S_IWUSR, array_state_show, array_state_store);
3501 max_corrected_read_errors_show(mddev_t *mddev, char *page) {
3502 return sprintf(page, "%d\n",
3503 atomic_read(&mddev->max_corr_read_errors));
3507 max_corrected_read_errors_store(mddev_t *mddev, const char *buf, size_t len)
3510 unsigned long n = simple_strtoul(buf, &e, 10);
3512 if (*buf && (*e == 0 || *e == '\n')) {
3513 atomic_set(&mddev->max_corr_read_errors, n);
3519 static struct md_sysfs_entry max_corr_read_errors =
3520 __ATTR(max_read_errors, S_IRUGO|S_IWUSR, max_corrected_read_errors_show,
3521 max_corrected_read_errors_store);
3524 null_show(mddev_t *mddev, char *page)
3530 new_dev_store(mddev_t *mddev, const char *buf, size_t len)
3532 /* buf must be %d:%d\n? giving major and minor numbers */
3533 /* The new device is added to the array.
3534 * If the array has a persistent superblock, we read the
3535 * superblock to initialise info and check validity.
3536 * Otherwise, only checking done is that in bind_rdev_to_array,
3537 * which mainly checks size.
3540 int major = simple_strtoul(buf, &e, 10);
3546 if (!*buf || *e != ':' || !e[1] || e[1] == '\n')
3548 minor = simple_strtoul(e+1, &e, 10);
3549 if (*e && *e != '\n')
3551 dev = MKDEV(major, minor);
3552 if (major != MAJOR(dev) ||
3553 minor != MINOR(dev))
3557 if (mddev->persistent) {
3558 rdev = md_import_device(dev, mddev->major_version,
3559 mddev->minor_version);
3560 if (!IS_ERR(rdev) && !list_empty(&mddev->disks)) {
3561 mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
3562 mdk_rdev_t, same_set);
3563 err = super_types[mddev->major_version]
3564 .load_super(rdev, rdev0, mddev->minor_version);
3568 } else if (mddev->external)
3569 rdev = md_import_device(dev, -2, -1);
3571 rdev = md_import_device(dev, -1, -1);
3574 return PTR_ERR(rdev);
3575 err = bind_rdev_to_array(rdev, mddev);
3579 return err ? err : len;
3582 static struct md_sysfs_entry md_new_device =
3583 __ATTR(new_dev, S_IWUSR, null_show, new_dev_store);
3586 bitmap_store(mddev_t *mddev, const char *buf, size_t len)
3589 unsigned long chunk, end_chunk;
3593 /* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */
3595 chunk = end_chunk = simple_strtoul(buf, &end, 0);
3596 if (buf == end) break;
3597 if (*end == '-') { /* range */
3599 end_chunk = simple_strtoul(buf, &end, 0);
3600 if (buf == end) break;
3602 if (*end && !isspace(*end)) break;
3603 bitmap_dirty_bits(mddev->bitmap, chunk, end_chunk);
3604 buf = skip_spaces(end);
3606 bitmap_unplug(mddev->bitmap); /* flush the bits to disk */
3611 static struct md_sysfs_entry md_bitmap =
3612 __ATTR(bitmap_set_bits, S_IWUSR, null_show, bitmap_store);
3615 size_show(mddev_t *mddev, char *page)
3617 return sprintf(page, "%llu\n",
3618 (unsigned long long)mddev->dev_sectors / 2);
3621 static int update_size(mddev_t *mddev, sector_t num_sectors);
3624 size_store(mddev_t *mddev, const char *buf, size_t len)
3626 /* If array is inactive, we can reduce the component size, but
3627 * not increase it (except from 0).
3628 * If array is active, we can try an on-line resize
3631 int err = strict_blocks_to_sectors(buf, §ors);
3636 err = update_size(mddev, sectors);
3637 md_update_sb(mddev, 1);
3639 if (mddev->dev_sectors == 0 ||
3640 mddev->dev_sectors > sectors)
3641 mddev->dev_sectors = sectors;
3645 return err ? err : len;
3648 static struct md_sysfs_entry md_size =
3649 __ATTR(component_size, S_IRUGO|S_IWUSR, size_show, size_store);
3654 * 'none' for arrays with no metadata (good luck...)
3655 * 'external' for arrays with externally managed metadata,
3656 * or N.M for internally known formats
3659 metadata_show(mddev_t *mddev, char *page)
3661 if (mddev->persistent)
3662 return sprintf(page, "%d.%d\n",
3663 mddev->major_version, mddev->minor_version);
3664 else if (mddev->external)
3665 return sprintf(page, "external:%s\n", mddev->metadata_type);
3667 return sprintf(page, "none\n");
3671 metadata_store(mddev_t *mddev, const char *buf, size_t len)
3675 /* Changing the details of 'external' metadata is
3676 * always permitted. Otherwise there must be
3677 * no devices attached to the array.
3679 if (mddev->external && strncmp(buf, "external:", 9) == 0)
3681 else if (!list_empty(&mddev->disks))
3684 if (cmd_match(buf, "none")) {
3685 mddev->persistent = 0;
3686 mddev->external = 0;
3687 mddev->major_version = 0;
3688 mddev->minor_version = 90;
3691 if (strncmp(buf, "external:", 9) == 0) {
3692 size_t namelen = len-9;
3693 if (namelen >= sizeof(mddev->metadata_type))
3694 namelen = sizeof(mddev->metadata_type)-1;
3695 strncpy(mddev->metadata_type, buf+9, namelen);
3696 mddev->metadata_type[namelen] = 0;
3697 if (namelen && mddev->metadata_type[namelen-1] == '\n')
3698 mddev->metadata_type[--namelen] = 0;
3699 mddev->persistent = 0;
3700 mddev->external = 1;
3701 mddev->major_version = 0;
3702 mddev->minor_version = 90;
3705 major = simple_strtoul(buf, &e, 10);
3706 if (e==buf || *e != '.')
3709 minor = simple_strtoul(buf, &e, 10);
3710 if (e==buf || (*e && *e != '\n') )
3712 if (major >= ARRAY_SIZE(super_types) || super_types[major].name == NULL)
3714 mddev->major_version = major;
3715 mddev->minor_version = minor;
3716 mddev->persistent = 1;
3717 mddev->external = 0;
3721 static struct md_sysfs_entry md_metadata =
3722 __ATTR(metadata_version, S_IRUGO|S_IWUSR, metadata_show, metadata_store);
3725 action_show(mddev_t *mddev, char *page)
3727 char *type = "idle";
3728 if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
3730 else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
3731 (!mddev->ro && test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))) {
3732 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
3734 else if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
3735 if (!test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
3737 else if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
3741 } else if (test_bit(MD_RECOVERY_RECOVER, &mddev->recovery))
3744 return sprintf(page, "%s\n", type);
3748 action_store(mddev_t *mddev, const char *page, size_t len)
3750 if (!mddev->pers || !mddev->pers->sync_request)
3753 if (cmd_match(page, "frozen"))
3754 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
3756 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
3758 if (cmd_match(page, "idle") || cmd_match(page, "frozen")) {
3759 if (mddev->sync_thread) {
3760 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
3761 md_unregister_thread(mddev->sync_thread);
3762 mddev->sync_thread = NULL;
3763 mddev->recovery = 0;
3765 } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
3766 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
3768 else if (cmd_match(page, "resync"))
3769 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3770 else if (cmd_match(page, "recover")) {
3771 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
3772 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3773 } else if (cmd_match(page, "reshape")) {
3775 if (mddev->pers->start_reshape == NULL)
3777 err = mddev->pers->start_reshape(mddev);
3780 sysfs_notify(&mddev->kobj, NULL, "degraded");
3782 if (cmd_match(page, "check"))
3783 set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
3784 else if (!cmd_match(page, "repair"))
3786 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
3787 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
3789 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3790 md_wakeup_thread(mddev->thread);
3791 sysfs_notify_dirent_safe(mddev->sysfs_action);
3796 mismatch_cnt_show(mddev_t *mddev, char *page)
3798 return sprintf(page, "%llu\n",
3799 (unsigned long long) mddev->resync_mismatches);
3802 static struct md_sysfs_entry md_scan_mode =
3803 __ATTR(sync_action, S_IRUGO|S_IWUSR, action_show, action_store);
3806 static struct md_sysfs_entry md_mismatches = __ATTR_RO(mismatch_cnt);
3809 sync_min_show(mddev_t *mddev, char *page)
3811 return sprintf(page, "%d (%s)\n", speed_min(mddev),
3812 mddev->sync_speed_min ? "local": "system");
3816 sync_min_store(mddev_t *mddev, const char *buf, size_t len)
3820 if (strncmp(buf, "system", 6)==0) {
3821 mddev->sync_speed_min = 0;
3824 min = simple_strtoul(buf, &e, 10);
3825 if (buf == e || (*e && *e != '\n') || min <= 0)
3827 mddev->sync_speed_min = min;
3831 static struct md_sysfs_entry md_sync_min =
3832 __ATTR(sync_speed_min, S_IRUGO|S_IWUSR, sync_min_show, sync_min_store);
3835 sync_max_show(mddev_t *mddev, char *page)
3837 return sprintf(page, "%d (%s)\n", speed_max(mddev),
3838 mddev->sync_speed_max ? "local": "system");
3842 sync_max_store(mddev_t *mddev, const char *buf, size_t len)
3846 if (strncmp(buf, "system", 6)==0) {
3847 mddev->sync_speed_max = 0;
3850 max = simple_strtoul(buf, &e, 10);
3851 if (buf == e || (*e && *e != '\n') || max <= 0)
3853 mddev->sync_speed_max = max;
3857 static struct md_sysfs_entry md_sync_max =
3858 __ATTR(sync_speed_max, S_IRUGO|S_IWUSR, sync_max_show, sync_max_store);
3861 degraded_show(mddev_t *mddev, char *page)
3863 return sprintf(page, "%d\n", mddev->degraded);
3865 static struct md_sysfs_entry md_degraded = __ATTR_RO(degraded);
3868 sync_force_parallel_show(mddev_t *mddev, char *page)
3870 return sprintf(page, "%d\n", mddev->parallel_resync);
3874 sync_force_parallel_store(mddev_t *mddev, const char *buf, size_t len)
3878 if (strict_strtol(buf, 10, &n))
3881 if (n != 0 && n != 1)
3884 mddev->parallel_resync = n;
3886 if (mddev->sync_thread)
3887 wake_up(&resync_wait);
3892 /* force parallel resync, even with shared block devices */
3893 static struct md_sysfs_entry md_sync_force_parallel =
3894 __ATTR(sync_force_parallel, S_IRUGO|S_IWUSR,
3895 sync_force_parallel_show, sync_force_parallel_store);
3898 sync_speed_show(mddev_t *mddev, char *page)
3900 unsigned long resync, dt, db;
3901 if (mddev->curr_resync == 0)
3902 return sprintf(page, "none\n");
3903 resync = mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active);
3904 dt = (jiffies - mddev->resync_mark) / HZ;
3906 db = resync - mddev->resync_mark_cnt;
3907 return sprintf(page, "%lu\n", db/dt/2); /* K/sec */
3910 static struct md_sysfs_entry md_sync_speed = __ATTR_RO(sync_speed);
3913 sync_completed_show(mddev_t *mddev, char *page)
3915 unsigned long max_sectors, resync;
3917 if (!test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
3918 return sprintf(page, "none\n");
3920 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
3921 max_sectors = mddev->resync_max_sectors;
3923 max_sectors = mddev->dev_sectors;
3925 resync = mddev->curr_resync_completed;
3926 return sprintf(page, "%lu / %lu\n", resync, max_sectors);
3929 static struct md_sysfs_entry md_sync_completed = __ATTR_RO(sync_completed);
3932 min_sync_show(mddev_t *mddev, char *page)
3934 return sprintf(page, "%llu\n",
3935 (unsigned long long)mddev->resync_min);
3938 min_sync_store(mddev_t *mddev, const char *buf, size_t len)
3940 unsigned long long min;
3941 if (strict_strtoull(buf, 10, &min))
3943 if (min > mddev->resync_max)
3945 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
3948 /* Must be a multiple of chunk_size */
3949 if (mddev->chunk_sectors) {
3950 sector_t temp = min;
3951 if (sector_div(temp, mddev->chunk_sectors))
3954 mddev->resync_min = min;
3959 static struct md_sysfs_entry md_min_sync =
3960 __ATTR(sync_min, S_IRUGO|S_IWUSR, min_sync_show, min_sync_store);
3963 max_sync_show(mddev_t *mddev, char *page)
3965 if (mddev->resync_max == MaxSector)
3966 return sprintf(page, "max\n");
3968 return sprintf(page, "%llu\n",
3969 (unsigned long long)mddev->resync_max);
3972 max_sync_store(mddev_t *mddev, const char *buf, size_t len)
3974 if (strncmp(buf, "max", 3) == 0)
3975 mddev->resync_max = MaxSector;
3977 unsigned long long max;
3978 if (strict_strtoull(buf, 10, &max))
3980 if (max < mddev->resync_min)
3982 if (max < mddev->resync_max &&
3984 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
3987 /* Must be a multiple of chunk_size */
3988 if (mddev->chunk_sectors) {
3989 sector_t temp = max;
3990 if (sector_div(temp, mddev->chunk_sectors))
3993 mddev->resync_max = max;
3995 wake_up(&mddev->recovery_wait);
3999 static struct md_sysfs_entry md_max_sync =
4000 __ATTR(sync_max, S_IRUGO|S_IWUSR, max_sync_show, max_sync_store);
4003 suspend_lo_show(mddev_t *mddev, char *page)
4005 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_lo);
4009 suspend_lo_store(mddev_t *mddev, const char *buf, size_t len)
4012 unsigned long long new = simple_strtoull(buf, &e, 10);
4014 if (mddev->pers == NULL ||
4015 mddev->pers->quiesce == NULL)
4017 if (buf == e || (*e && *e != '\n'))
4019 if (new >= mddev->suspend_hi ||
4020 (new > mddev->suspend_lo && new < mddev->suspend_hi)) {
4021 mddev->suspend_lo = new;
4022 mddev->pers->quiesce(mddev, 2);
4027 static struct md_sysfs_entry md_suspend_lo =
4028 __ATTR(suspend_lo, S_IRUGO|S_IWUSR, suspend_lo_show, suspend_lo_store);
4032 suspend_hi_show(mddev_t *mddev, char *page)
4034 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_hi);
4038 suspend_hi_store(mddev_t *mddev, const char *buf, size_t len)
4041 unsigned long long new = simple_strtoull(buf, &e, 10);
4043 if (mddev->pers == NULL ||
4044 mddev->pers->quiesce == NULL)
4046 if (buf == e || (*e && *e != '\n'))
4048 if ((new <= mddev->suspend_lo && mddev->suspend_lo >= mddev->suspend_hi) ||
4049 (new > mddev->suspend_lo && new > mddev->suspend_hi)) {
4050 mddev->suspend_hi = new;
4051 mddev->pers->quiesce(mddev, 1);
4052 mddev->pers->quiesce(mddev, 0);
4057 static struct md_sysfs_entry md_suspend_hi =
4058 __ATTR(suspend_hi, S_IRUGO|S_IWUSR, suspend_hi_show, suspend_hi_store);
4061 reshape_position_show(mddev_t *mddev, char *page)
4063 if (mddev->reshape_position != MaxSector)
4064 return sprintf(page, "%llu\n",
4065 (unsigned long long)mddev->reshape_position);
4066 strcpy(page, "none\n");
4071 reshape_position_store(mddev_t *mddev, const char *buf, size_t len)
4074 unsigned long long new = simple_strtoull(buf, &e, 10);
4077 if (buf == e || (*e && *e != '\n'))
4079 mddev->reshape_position = new;
4080 mddev->delta_disks = 0;
4081 mddev->new_level = mddev->level;
4082 mddev->new_layout = mddev->layout;
4083 mddev->new_chunk_sectors = mddev->chunk_sectors;
4087 static struct md_sysfs_entry md_reshape_position =
4088 __ATTR(reshape_position, S_IRUGO|S_IWUSR, reshape_position_show,
4089 reshape_position_store);
4092 array_size_show(mddev_t *mddev, char *page)
4094 if (mddev->external_size)
4095 return sprintf(page, "%llu\n",
4096 (unsigned long long)mddev->array_sectors/2);
4098 return sprintf(page, "default\n");
4102 array_size_store(mddev_t *mddev, const char *buf, size_t len)
4106 if (strncmp(buf, "default", 7) == 0) {
4108 sectors = mddev->pers->size(mddev, 0, 0);
4110 sectors = mddev->array_sectors;
4112 mddev->external_size = 0;
4114 if (strict_blocks_to_sectors(buf, §ors) < 0)
4116 if (mddev->pers && mddev->pers->size(mddev, 0, 0) < sectors)
4119 mddev->external_size = 1;
4122 mddev->array_sectors = sectors;
4123 set_capacity(mddev->gendisk, mddev->array_sectors);
4125 revalidate_disk(mddev->gendisk);
4130 static struct md_sysfs_entry md_array_size =
4131 __ATTR(array_size, S_IRUGO|S_IWUSR, array_size_show,
4134 static struct attribute *md_default_attrs[] = {
4137 &md_raid_disks.attr,
4138 &md_chunk_size.attr,
4140 &md_resync_start.attr,
4142 &md_new_device.attr,
4143 &md_safe_delay.attr,
4144 &md_array_state.attr,
4145 &md_reshape_position.attr,
4146 &md_array_size.attr,
4147 &max_corr_read_errors.attr,
4151 static struct attribute *md_redundancy_attrs[] = {
4153 &md_mismatches.attr,
4156 &md_sync_speed.attr,
4157 &md_sync_force_parallel.attr,
4158 &md_sync_completed.attr,
4161 &md_suspend_lo.attr,
4162 &md_suspend_hi.attr,
4167 static struct attribute_group md_redundancy_group = {
4169 .attrs = md_redundancy_attrs,
4174 md_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
4176 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
4177 mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
4182 rv = mddev_lock(mddev);
4184 rv = entry->show(mddev, page);
4185 mddev_unlock(mddev);
4191 md_attr_store(struct kobject *kobj, struct attribute *attr,
4192 const char *page, size_t length)
4194 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
4195 mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
4200 if (!capable(CAP_SYS_ADMIN))
4202 rv = mddev_lock(mddev);
4203 if (mddev->hold_active == UNTIL_IOCTL)
4204 mddev->hold_active = 0;
4206 rv = entry->store(mddev, page, length);
4207 mddev_unlock(mddev);
4212 static void md_free(struct kobject *ko)
4214 mddev_t *mddev = container_of(ko, mddev_t, kobj);
4216 if (mddev->sysfs_state)
4217 sysfs_put(mddev->sysfs_state);
4219 if (mddev->gendisk) {
4220 del_gendisk(mddev->gendisk);
4221 put_disk(mddev->gendisk);
4224 blk_cleanup_queue(mddev->queue);
4229 static const struct sysfs_ops md_sysfs_ops = {
4230 .show = md_attr_show,
4231 .store = md_attr_store,
4233 static struct kobj_type md_ktype = {
4235 .sysfs_ops = &md_sysfs_ops,
4236 .default_attrs = md_default_attrs,
4241 static void mddev_delayed_delete(struct work_struct *ws)
4243 mddev_t *mddev = container_of(ws, mddev_t, del_work);
4245 sysfs_remove_group(&mddev->kobj, &md_bitmap_group);
4246 kobject_del(&mddev->kobj);
4247 kobject_put(&mddev->kobj);
4250 static int md_alloc(dev_t dev, char *name)
4252 static DEFINE_MUTEX(disks_mutex);
4253 mddev_t *mddev = mddev_find(dev);
4254 struct gendisk *disk;
4263 partitioned = (MAJOR(mddev->unit) != MD_MAJOR);
4264 shift = partitioned ? MdpMinorShift : 0;
4265 unit = MINOR(mddev->unit) >> shift;
4267 /* wait for any previous instance of this device to be
4268 * completely removed (mddev_delayed_delete).
4270 flush_workqueue(md_misc_wq);
4272 mutex_lock(&disks_mutex);
4278 /* Need to ensure that 'name' is not a duplicate.
4281 spin_lock(&all_mddevs_lock);
4283 list_for_each_entry(mddev2, &all_mddevs, all_mddevs)
4284 if (mddev2->gendisk &&
4285 strcmp(mddev2->gendisk->disk_name, name) == 0) {
4286 spin_unlock(&all_mddevs_lock);
4289 spin_unlock(&all_mddevs_lock);
4293 mddev->queue = blk_alloc_queue(GFP_KERNEL);
4296 mddev->queue->queuedata = mddev;
4298 /* Can be unlocked because the queue is new: no concurrency */
4299 queue_flag_set_unlocked(QUEUE_FLAG_CLUSTER, mddev->queue);
4301 blk_queue_make_request(mddev->queue, md_make_request);
4303 disk = alloc_disk(1 << shift);
4305 blk_cleanup_queue(mddev->queue);
4306 mddev->queue = NULL;
4309 disk->major = MAJOR(mddev->unit);
4310 disk->first_minor = unit << shift;
4312 strcpy(disk->disk_name, name);
4313 else if (partitioned)
4314 sprintf(disk->disk_name, "md_d%d", unit);
4316 sprintf(disk->disk_name, "md%d", unit);
4317 disk->fops = &md_fops;
4318 disk->private_data = mddev;
4319 disk->queue = mddev->queue;
4320 /* Allow extended partitions. This makes the
4321 * 'mdp' device redundant, but we can't really
4324 disk->flags |= GENHD_FL_EXT_DEVT;
4326 mddev->gendisk = disk;
4327 error = kobject_init_and_add(&mddev->kobj, &md_ktype,
4328 &disk_to_dev(disk)->kobj, "%s", "md");
4330 /* This isn't possible, but as kobject_init_and_add is marked
4331 * __must_check, we must do something with the result
4333 printk(KERN_WARNING "md: cannot register %s/md - name in use\n",
4337 if (mddev->kobj.sd &&
4338 sysfs_create_group(&mddev->kobj, &md_bitmap_group))
4339 printk(KERN_DEBUG "pointless warning\n");
4341 blk_queue_flush(mddev->queue, REQ_FLUSH | REQ_FUA);
4343 mutex_unlock(&disks_mutex);
4344 if (!error && mddev->kobj.sd) {
4345 kobject_uevent(&mddev->kobj, KOBJ_ADD);
4346 mddev->sysfs_state = sysfs_get_dirent_safe(mddev->kobj.sd, "array_state");
4352 static struct kobject *md_probe(dev_t dev, int *part, void *data)
4354 md_alloc(dev, NULL);
4358 static int add_named_array(const char *val, struct kernel_param *kp)
4360 /* val must be "md_*" where * is not all digits.
4361 * We allocate an array with a large free minor number, and
4362 * set the name to val. val must not already be an active name.
4364 int len = strlen(val);
4365 char buf[DISK_NAME_LEN];
4367 while (len && val[len-1] == '\n')
4369 if (len >= DISK_NAME_LEN)
4371 strlcpy(buf, val, len+1);
4372 if (strncmp(buf, "md_", 3) != 0)
4374 return md_alloc(0, buf);
4377 static void md_safemode_timeout(unsigned long data)
4379 mddev_t *mddev = (mddev_t *) data;
4381 if (!atomic_read(&mddev->writes_pending)) {
4382 mddev->safemode = 1;
4383 if (mddev->external)
4384 sysfs_notify_dirent_safe(mddev->sysfs_state);
4386 md_wakeup_thread(mddev->thread);
4389 static int start_dirty_degraded;
4391 int md_run(mddev_t *mddev)
4395 struct mdk_personality *pers;
4397 if (list_empty(&mddev->disks))
4398 /* cannot run an array with no devices.. */
4403 /* Cannot run until previous stop completes properly */
4404 if (mddev->sysfs_active)
4408 * Analyze all RAID superblock(s)
4410 if (!mddev->raid_disks) {
4411 if (!mddev->persistent)
4416 if (mddev->level != LEVEL_NONE)
4417 request_module("md-level-%d", mddev->level);
4418 else if (mddev->clevel[0])
4419 request_module("md-%s", mddev->clevel);
4422 * Drop all container device buffers, from now on
4423 * the only valid external interface is through the md
4426 list_for_each_entry(rdev, &mddev->disks, same_set) {
4427 if (test_bit(Faulty, &rdev->flags))
4429 sync_blockdev(rdev->bdev);
4430 invalidate_bdev(rdev->bdev);
4432 /* perform some consistency tests on the device.
4433 * We don't want the data to overlap the metadata,
4434 * Internal Bitmap issues have been handled elsewhere.
4436 if (rdev->data_offset < rdev->sb_start) {
4437 if (mddev->dev_sectors &&
4438 rdev->data_offset + mddev->dev_sectors
4440 printk("md: %s: data overlaps metadata\n",
4445 if (rdev->sb_start + rdev->sb_size/512
4446 > rdev->data_offset) {
4447 printk("md: %s: metadata overlaps data\n",
4452 sysfs_notify_dirent_safe(rdev->sysfs_state);
4455 if (mddev->bio_set == NULL)
4456 mddev->bio_set = bioset_create(BIO_POOL_SIZE, sizeof(mddev));
4458 spin_lock(&pers_lock);
4459 pers = find_pers(mddev->level, mddev->clevel);
4460 if (!pers || !try_module_get(pers->owner)) {
4461 spin_unlock(&pers_lock);
4462 if (mddev->level != LEVEL_NONE)
4463 printk(KERN_WARNING "md: personality for level %d is not loaded!\n",
4466 printk(KERN_WARNING "md: personality for level %s is not loaded!\n",
4471 spin_unlock(&pers_lock);
4472 if (mddev->level != pers->level) {
4473 mddev->level = pers->level;
4474 mddev->new_level = pers->level;
4476 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
4478 if (mddev->reshape_position != MaxSector &&
4479 pers->start_reshape == NULL) {
4480 /* This personality cannot handle reshaping... */
4482 module_put(pers->owner);
4486 if (pers->sync_request) {
4487 /* Warn if this is a potentially silly
4490 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
4494 list_for_each_entry(rdev, &mddev->disks, same_set)
4495 list_for_each_entry(rdev2, &mddev->disks, same_set) {
4497 rdev->bdev->bd_contains ==
4498 rdev2->bdev->bd_contains) {
4500 "%s: WARNING: %s appears to be"
4501 " on the same physical disk as"
4504 bdevname(rdev->bdev,b),
4505 bdevname(rdev2->bdev,b2));
4512 "True protection against single-disk"
4513 " failure might be compromised.\n");
4516 mddev->recovery = 0;
4517 /* may be over-ridden by personality */
4518 mddev->resync_max_sectors = mddev->dev_sectors;
4520 mddev->ok_start_degraded = start_dirty_degraded;
4522 if (start_readonly && mddev->ro == 0)
4523 mddev->ro = 2; /* read-only, but switch on first write */
4525 err = mddev->pers->run(mddev);
4527 printk(KERN_ERR "md: pers->run() failed ...\n");
4528 else if (mddev->pers->size(mddev, 0, 0) < mddev->array_sectors) {
4529 WARN_ONCE(!mddev->external_size, "%s: default size too small,"
4530 " but 'external_size' not in effect?\n", __func__);
4532 "md: invalid array_size %llu > default size %llu\n",
4533 (unsigned long long)mddev->array_sectors / 2,
4534 (unsigned long long)mddev->pers->size(mddev, 0, 0) / 2);
4536 mddev->pers->stop(mddev);
4538 if (err == 0 && mddev->pers->sync_request) {
4539 err = bitmap_create(mddev);
4541 printk(KERN_ERR "%s: failed to create bitmap (%d)\n",
4542 mdname(mddev), err);
4543 mddev->pers->stop(mddev);
4547 module_put(mddev->pers->owner);
4549 bitmap_destroy(mddev);
4552 if (mddev->pers->sync_request) {
4553 if (mddev->kobj.sd &&
4554 sysfs_create_group(&mddev->kobj, &md_redundancy_group))
4556 "md: cannot register extra attributes for %s\n",
4558 mddev->sysfs_action = sysfs_get_dirent_safe(mddev->kobj.sd, "sync_action");
4559 } else if (mddev->ro == 2) /* auto-readonly not meaningful */
4562 atomic_set(&mddev->writes_pending,0);
4563 atomic_set(&mddev->max_corr_read_errors,
4564 MD_DEFAULT_MAX_CORRECTED_READ_ERRORS);
4565 mddev->safemode = 0;
4566 mddev->safemode_timer.function = md_safemode_timeout;
4567 mddev->safemode_timer.data = (unsigned long) mddev;
4568 mddev->safemode_delay = (200 * HZ)/1000 +1; /* 200 msec delay */
4571 list_for_each_entry(rdev, &mddev->disks, same_set)
4572 if (rdev->raid_disk >= 0) {
4574 sprintf(nm, "rd%d", rdev->raid_disk);
4575 if (sysfs_create_link(&mddev->kobj, &rdev->kobj, nm))
4576 /* failure here is OK */;
4579 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4582 md_update_sb(mddev, 0);
4584 md_wakeup_thread(mddev->thread);
4585 md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
4587 md_new_event(mddev);
4588 sysfs_notify_dirent_safe(mddev->sysfs_state);
4589 sysfs_notify_dirent_safe(mddev->sysfs_action);
4590 sysfs_notify(&mddev->kobj, NULL, "degraded");
4593 EXPORT_SYMBOL_GPL(md_run);
4595 static int do_md_run(mddev_t *mddev)
4599 err = md_run(mddev);
4602 err = bitmap_load(mddev);
4604 bitmap_destroy(mddev);
4607 set_capacity(mddev->gendisk, mddev->array_sectors);
4608 revalidate_disk(mddev->gendisk);
4609 kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
4614 static int restart_array(mddev_t *mddev)
4616 struct gendisk *disk = mddev->gendisk;
4618 /* Complain if it has no devices */
4619 if (list_empty(&mddev->disks))
4625 mddev->safemode = 0;
4627 set_disk_ro(disk, 0);
4628 printk(KERN_INFO "md: %s switched to read-write mode.\n",
4630 /* Kick recovery or resync if necessary */
4631 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4632 md_wakeup_thread(mddev->thread);
4633 md_wakeup_thread(mddev->sync_thread);
4634 sysfs_notify_dirent_safe(mddev->sysfs_state);
4638 /* similar to deny_write_access, but accounts for our holding a reference
4639 * to the file ourselves */
4640 static int deny_bitmap_write_access(struct file * file)
4642 struct inode *inode = file->f_mapping->host;
4644 spin_lock(&inode->i_lock);
4645 if (atomic_read(&inode->i_writecount) > 1) {
4646 spin_unlock(&inode->i_lock);
4649 atomic_set(&inode->i_writecount, -1);
4650 spin_unlock(&inode->i_lock);
4655 void restore_bitmap_write_access(struct file *file)
4657 struct inode *inode = file->f_mapping->host;
4659 spin_lock(&inode->i_lock);
4660 atomic_set(&inode->i_writecount, 1);
4661 spin_unlock(&inode->i_lock);
4664 static void md_clean(mddev_t *mddev)
4666 mddev->array_sectors = 0;
4667 mddev->external_size = 0;
4668 mddev->dev_sectors = 0;
4669 mddev->raid_disks = 0;
4670 mddev->recovery_cp = 0;
4671 mddev->resync_min = 0;
4672 mddev->resync_max = MaxSector;
4673 mddev->reshape_position = MaxSector;
4674 mddev->external = 0;
4675 mddev->persistent = 0;
4676 mddev->level = LEVEL_NONE;
4677 mddev->clevel[0] = 0;
4680 mddev->metadata_type[0] = 0;
4681 mddev->chunk_sectors = 0;
4682 mddev->ctime = mddev->utime = 0;
4684 mddev->max_disks = 0;
4686 mddev->can_decrease_events = 0;
4687 mddev->delta_disks = 0;
4688 mddev->new_level = LEVEL_NONE;
4689 mddev->new_layout = 0;
4690 mddev->new_chunk_sectors = 0;
4691 mddev->curr_resync = 0;
4692 mddev->resync_mismatches = 0;
4693 mddev->suspend_lo = mddev->suspend_hi = 0;
4694 mddev->sync_speed_min = mddev->sync_speed_max = 0;
4695 mddev->recovery = 0;
4697 mddev->degraded = 0;
4698 mddev->safemode = 0;
4699 mddev->bitmap_info.offset = 0;
4700 mddev->bitmap_info.default_offset = 0;
4701 mddev->bitmap_info.chunksize = 0;
4702 mddev->bitmap_info.daemon_sleep = 0;
4703 mddev->bitmap_info.max_write_behind = 0;
4707 void md_stop_writes(mddev_t *mddev)
4709 if (mddev->sync_thread) {
4710 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4711 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
4712 md_unregister_thread(mddev->sync_thread);
4713 mddev->sync_thread = NULL;
4716 del_timer_sync(&mddev->safemode_timer);
4718 bitmap_flush(mddev);
4719 md_super_wait(mddev);
4721 if (!mddev->in_sync || mddev->flags) {
4722 /* mark array as shutdown cleanly */
4724 md_update_sb(mddev, 1);
4727 EXPORT_SYMBOL_GPL(md_stop_writes);
4729 void md_stop(mddev_t *mddev)
4731 mddev->pers->stop(mddev);
4732 if (mddev->pers->sync_request && mddev->to_remove == NULL)
4733 mddev->to_remove = &md_redundancy_group;
4734 module_put(mddev->pers->owner);
4736 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4738 EXPORT_SYMBOL_GPL(md_stop);
4740 static int md_set_readonly(mddev_t *mddev, int is_open)
4743 mutex_lock(&mddev->open_mutex);
4744 if (atomic_read(&mddev->openers) > is_open) {
4745 printk("md: %s still in use.\n",mdname(mddev));
4750 md_stop_writes(mddev);
4756 set_disk_ro(mddev->gendisk, 1);
4757 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4758 sysfs_notify_dirent_safe(mddev->sysfs_state);
4762 mutex_unlock(&mddev->open_mutex);
4767 * 0 - completely stop and dis-assemble array
4768 * 2 - stop but do not disassemble array
4770 static int do_md_stop(mddev_t * mddev, int mode, int is_open)
4772 struct gendisk *disk = mddev->gendisk;
4775 mutex_lock(&mddev->open_mutex);
4776 if (atomic_read(&mddev->openers) > is_open ||
4777 mddev->sysfs_active) {
4778 printk("md: %s still in use.\n",mdname(mddev));
4779 mutex_unlock(&mddev->open_mutex);
4785 set_disk_ro(disk, 0);
4787 md_stop_writes(mddev);
4789 mddev->queue->merge_bvec_fn = NULL;
4790 mddev->queue->unplug_fn = NULL;
4791 mddev->queue->backing_dev_info.congested_fn = NULL;
4793 /* tell userspace to handle 'inactive' */
4794 sysfs_notify_dirent_safe(mddev->sysfs_state);
4796 list_for_each_entry(rdev, &mddev->disks, same_set)
4797 if (rdev->raid_disk >= 0) {
4799 sprintf(nm, "rd%d", rdev->raid_disk);
4800 sysfs_remove_link(&mddev->kobj, nm);
4803 set_capacity(disk, 0);
4804 mutex_unlock(&mddev->open_mutex);
4805 revalidate_disk(disk);
4810 mutex_unlock(&mddev->open_mutex);
4812 * Free resources if final stop
4815 printk(KERN_INFO "md: %s stopped.\n", mdname(mddev));
4817 bitmap_destroy(mddev);
4818 if (mddev->bitmap_info.file) {
4819 restore_bitmap_write_access(mddev->bitmap_info.file);
4820 fput(mddev->bitmap_info.file);
4821 mddev->bitmap_info.file = NULL;
4823 mddev->bitmap_info.offset = 0;
4825 export_array(mddev);
4828 kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
4829 if (mddev->hold_active == UNTIL_STOP)
4830 mddev->hold_active = 0;
4832 blk_integrity_unregister(disk);
4833 md_new_event(mddev);
4834 sysfs_notify_dirent_safe(mddev->sysfs_state);
4839 static void autorun_array(mddev_t *mddev)
4844 if (list_empty(&mddev->disks))
4847 printk(KERN_INFO "md: running: ");
4849 list_for_each_entry(rdev, &mddev->disks, same_set) {
4850 char b[BDEVNAME_SIZE];
4851 printk("<%s>", bdevname(rdev->bdev,b));
4855 err = do_md_run(mddev);
4857 printk(KERN_WARNING "md: do_md_run() returned %d\n", err);
4858 do_md_stop(mddev, 0, 0);
4863 * lets try to run arrays based on all disks that have arrived
4864 * until now. (those are in pending_raid_disks)
4866 * the method: pick the first pending disk, collect all disks with
4867 * the same UUID, remove all from the pending list and put them into
4868 * the 'same_array' list. Then order this list based on superblock
4869 * update time (freshest comes first), kick out 'old' disks and
4870 * compare superblocks. If everything's fine then run it.
4872 * If "unit" is allocated, then bump its reference count
4874 static void autorun_devices(int part)
4876 mdk_rdev_t *rdev0, *rdev, *tmp;
4878 char b[BDEVNAME_SIZE];
4880 printk(KERN_INFO "md: autorun ...\n");
4881 while (!list_empty(&pending_raid_disks)) {
4884 LIST_HEAD(candidates);
4885 rdev0 = list_entry(pending_raid_disks.next,
4886 mdk_rdev_t, same_set);
4888 printk(KERN_INFO "md: considering %s ...\n",
4889 bdevname(rdev0->bdev,b));
4890 INIT_LIST_HEAD(&candidates);
4891 rdev_for_each_list(rdev, tmp, &pending_raid_disks)
4892 if (super_90_load(rdev, rdev0, 0) >= 0) {
4893 printk(KERN_INFO "md: adding %s ...\n",
4894 bdevname(rdev->bdev,b));
4895 list_move(&rdev->same_set, &candidates);
4898 * now we have a set of devices, with all of them having
4899 * mostly sane superblocks. It's time to allocate the
4903 dev = MKDEV(mdp_major,
4904 rdev0->preferred_minor << MdpMinorShift);
4905 unit = MINOR(dev) >> MdpMinorShift;
4907 dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
4910 if (rdev0->preferred_minor != unit) {
4911 printk(KERN_INFO "md: unit number in %s is bad: %d\n",
4912 bdevname(rdev0->bdev, b), rdev0->preferred_minor);
4916 md_probe(dev, NULL, NULL);
4917 mddev = mddev_find(dev);
4918 if (!mddev || !mddev->gendisk) {
4922 "md: cannot allocate memory for md drive.\n");
4925 if (mddev_lock(mddev))
4926 printk(KERN_WARNING "md: %s locked, cannot run\n",
4928 else if (mddev->raid_disks || mddev->major_version
4929 || !list_empty(&mddev->disks)) {
4931 "md: %s already running, cannot run %s\n",
4932 mdname(mddev), bdevname(rdev0->bdev,b));
4933 mddev_unlock(mddev);
4935 printk(KERN_INFO "md: created %s\n", mdname(mddev));
4936 mddev->persistent = 1;
4937 rdev_for_each_list(rdev, tmp, &candidates) {
4938 list_del_init(&rdev->same_set);
4939 if (bind_rdev_to_array(rdev, mddev))
4942 autorun_array(mddev);
4943 mddev_unlock(mddev);
4945 /* on success, candidates will be empty, on error
4948 rdev_for_each_list(rdev, tmp, &candidates) {
4949 list_del_init(&rdev->same_set);
4954 printk(KERN_INFO "md: ... autorun DONE.\n");
4956 #endif /* !MODULE */
4958 static int get_version(void __user * arg)
4962 ver.major = MD_MAJOR_VERSION;
4963 ver.minor = MD_MINOR_VERSION;
4964 ver.patchlevel = MD_PATCHLEVEL_VERSION;
4966 if (copy_to_user(arg, &ver, sizeof(ver)))
4972 static int get_array_info(mddev_t * mddev, void __user * arg)
4974 mdu_array_info_t info;
4975 int nr,working,insync,failed,spare;
4978 nr=working=insync=failed=spare=0;
4979 list_for_each_entry(rdev, &mddev->disks, same_set) {
4981 if (test_bit(Faulty, &rdev->flags))
4985 if (test_bit(In_sync, &rdev->flags))
4992 info.major_version = mddev->major_version;
4993 info.minor_version = mddev->minor_version;
4994 info.patch_version = MD_PATCHLEVEL_VERSION;
4995 info.ctime = mddev->ctime;
4996 info.level = mddev->level;
4997 info.size = mddev->dev_sectors / 2;
4998 if (info.size != mddev->dev_sectors / 2) /* overflow */
5001 info.raid_disks = mddev->raid_disks;
5002 info.md_minor = mddev->md_minor;
5003 info.not_persistent= !mddev->persistent;
5005 info.utime = mddev->utime;
5008 info.state = (1<<MD_SB_CLEAN);
5009 if (mddev->bitmap && mddev->bitmap_info.offset)
5010 info.state = (1<<MD_SB_BITMAP_PRESENT);
5011 info.active_disks = insync;
5012 info.working_disks = working;
5013 info.failed_disks = failed;
5014 info.spare_disks = spare;
5016 info.layout = mddev->layout;
5017 info.chunk_size = mddev->chunk_sectors << 9;
5019 if (copy_to_user(arg, &info, sizeof(info)))
5025 static int get_bitmap_file(mddev_t * mddev, void __user * arg)
5027 mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */
5028 char *ptr, *buf = NULL;
5031 if (md_allow_write(mddev))
5032 file = kmalloc(sizeof(*file), GFP_NOIO);
5034 file = kmalloc(sizeof(*file), GFP_KERNEL);
5039 /* bitmap disabled, zero the first byte and copy out */
5040 if (!mddev->bitmap || !mddev->bitmap->file) {
5041 file->pathname[0] = '\0';
5045 buf = kmalloc(sizeof(file->pathname), GFP_KERNEL);
5049 ptr = d_path(&mddev->bitmap->file->f_path, buf, sizeof(file->pathname));
5053 strcpy(file->pathname, ptr);
5057 if (copy_to_user(arg, file, sizeof(*file)))
5065 static int get_disk_info(mddev_t * mddev, void __user * arg)
5067 mdu_disk_info_t info;
5070 if (copy_from_user(&info, arg, sizeof(info)))
5073 rdev = find_rdev_nr(mddev, info.number);
5075 info.major = MAJOR(rdev->bdev->bd_dev);
5076 info.minor = MINOR(rdev->bdev->bd_dev);
5077 info.raid_disk = rdev->raid_disk;
5079 if (test_bit(Faulty, &rdev->flags))
5080 info.state |= (1<<MD_DISK_FAULTY);
5081 else if (test_bit(In_sync, &rdev->flags)) {
5082 info.state |= (1<<MD_DISK_ACTIVE);
5083 info.state |= (1<<MD_DISK_SYNC);
5085 if (test_bit(WriteMostly, &rdev->flags))
5086 info.state |= (1<<MD_DISK_WRITEMOSTLY);
5088 info.major = info.minor = 0;
5089 info.raid_disk = -1;
5090 info.state = (1<<MD_DISK_REMOVED);
5093 if (copy_to_user(arg, &info, sizeof(info)))
5099 static int add_new_disk(mddev_t * mddev, mdu_disk_info_t *info)
5101 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
5103 dev_t dev = MKDEV(info->major,info->minor);
5105 if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
5108 if (!mddev->raid_disks) {
5110 /* expecting a device which has a superblock */
5111 rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
5114 "md: md_import_device returned %ld\n",
5116 return PTR_ERR(rdev);
5118 if (!list_empty(&mddev->disks)) {
5119 mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
5120 mdk_rdev_t, same_set);
5121 err = super_types[mddev->major_version]
5122 .load_super(rdev, rdev0, mddev->minor_version);
5125 "md: %s has different UUID to %s\n",
5126 bdevname(rdev->bdev,b),
5127 bdevname(rdev0->bdev,b2));
5132 err = bind_rdev_to_array(rdev, mddev);
5139 * add_new_disk can be used once the array is assembled
5140 * to add "hot spares". They must already have a superblock
5145 if (!mddev->pers->hot_add_disk) {
5147 "%s: personality does not support diskops!\n",
5151 if (mddev->persistent)
5152 rdev = md_import_device(dev, mddev->major_version,
5153 mddev->minor_version);
5155 rdev = md_import_device(dev, -1, -1);
5158 "md: md_import_device returned %ld\n",
5160 return PTR_ERR(rdev);
5162 /* set saved_raid_disk if appropriate */
5163 if (!mddev->persistent) {
5164 if (info->state & (1<<MD_DISK_SYNC) &&
5165 info->raid_disk < mddev->raid_disks)
5166 rdev->raid_disk = info->raid_disk;
5168 rdev->raid_disk = -1;
5170 super_types[mddev->major_version].
5171 validate_super(mddev, rdev);
5172 if (test_bit(In_sync, &rdev->flags))
5173 rdev->saved_raid_disk = rdev->raid_disk;
5175 rdev->saved_raid_disk = -1;
5177 clear_bit(In_sync, &rdev->flags); /* just to be sure */
5178 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
5179 set_bit(WriteMostly, &rdev->flags);
5181 clear_bit(WriteMostly, &rdev->flags);
5183 rdev->raid_disk = -1;
5184 err = bind_rdev_to_array(rdev, mddev);
5185 if (!err && !mddev->pers->hot_remove_disk) {
5186 /* If there is hot_add_disk but no hot_remove_disk
5187 * then added disks for geometry changes,
5188 * and should be added immediately.
5190 super_types[mddev->major_version].
5191 validate_super(mddev, rdev);
5192 err = mddev->pers->hot_add_disk(mddev, rdev);
5194 unbind_rdev_from_array(rdev);
5199 sysfs_notify_dirent_safe(rdev->sysfs_state);
5201 md_update_sb(mddev, 1);
5202 if (mddev->degraded)
5203 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
5204 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5205 md_wakeup_thread(mddev->thread);
5209 /* otherwise, add_new_disk is only allowed
5210 * for major_version==0 superblocks
5212 if (mddev->major_version != 0) {
5213 printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n",
5218 if (!(info->state & (1<<MD_DISK_FAULTY))) {
5220 rdev = md_import_device(dev, -1, 0);
5223 "md: error, md_import_device() returned %ld\n",
5225 return PTR_ERR(rdev);
5227 rdev->desc_nr = info->number;
5228 if (info->raid_disk < mddev->raid_disks)
5229 rdev->raid_disk = info->raid_disk;
5231 rdev->raid_disk = -1;
5233 if (rdev->raid_disk < mddev->raid_disks)
5234 if (info->state & (1<<MD_DISK_SYNC))
5235 set_bit(In_sync, &rdev->flags);
5237 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
5238 set_bit(WriteMostly, &rdev->flags);
5240 if (!mddev->persistent) {
5241 printk(KERN_INFO "md: nonpersistent superblock ...\n");
5242 rdev->sb_start = i_size_read(rdev->bdev->bd_inode) / 512;
5244 rdev->sb_start = calc_dev_sboffset(rdev->bdev);
5245 rdev->sectors = rdev->sb_start;
5247 err = bind_rdev_to_array(rdev, mddev);
5257 static int hot_remove_disk(mddev_t * mddev, dev_t dev)
5259 char b[BDEVNAME_SIZE];
5262 rdev = find_rdev(mddev, dev);
5266 if (rdev->raid_disk >= 0)
5269 kick_rdev_from_array(rdev);
5270 md_update_sb(mddev, 1);
5271 md_new_event(mddev);
5275 printk(KERN_WARNING "md: cannot remove active disk %s from %s ...\n",
5276 bdevname(rdev->bdev,b), mdname(mddev));
5280 static int hot_add_disk(mddev_t * mddev, dev_t dev)
5282 char b[BDEVNAME_SIZE];
5289 if (mddev->major_version != 0) {
5290 printk(KERN_WARNING "%s: HOT_ADD may only be used with"
5291 " version-0 superblocks.\n",
5295 if (!mddev->pers->hot_add_disk) {
5297 "%s: personality does not support diskops!\n",
5302 rdev = md_import_device(dev, -1, 0);
5305 "md: error, md_import_device() returned %ld\n",
5310 if (mddev->persistent)
5311 rdev->sb_start = calc_dev_sboffset(rdev->bdev);
5313 rdev->sb_start = i_size_read(rdev->bdev->bd_inode) / 512;
5315 rdev->sectors = rdev->sb_start;
5317 if (test_bit(Faulty, &rdev->flags)) {
5319 "md: can not hot-add faulty %s disk to %s!\n",
5320 bdevname(rdev->bdev,b), mdname(mddev));
5324 clear_bit(In_sync, &rdev->flags);
5326 rdev->saved_raid_disk = -1;
5327 err = bind_rdev_to_array(rdev, mddev);
5332 * The rest should better be atomic, we can have disk failures
5333 * noticed in interrupt contexts ...
5336 rdev->raid_disk = -1;
5338 md_update_sb(mddev, 1);
5341 * Kick recovery, maybe this spare has to be added to the
5342 * array immediately.
5344 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5345 md_wakeup_thread(mddev->thread);
5346 md_new_event(mddev);
5354 static int set_bitmap_file(mddev_t *mddev, int fd)
5359 if (!mddev->pers->quiesce)
5361 if (mddev->recovery || mddev->sync_thread)
5363 /* we should be able to change the bitmap.. */
5369 return -EEXIST; /* cannot add when bitmap is present */
5370 mddev->bitmap_info.file = fget(fd);
5372 if (mddev->bitmap_info.file == NULL) {
5373 printk(KERN_ERR "%s: error: failed to get bitmap file\n",
5378 err = deny_bitmap_write_access(mddev->bitmap_info.file);
5380 printk(KERN_ERR "%s: error: bitmap file is already in use\n",
5382 fput(mddev->bitmap_info.file);
5383 mddev->bitmap_info.file = NULL;
5386 mddev->bitmap_info.offset = 0; /* file overrides offset */
5387 } else if (mddev->bitmap == NULL)
5388 return -ENOENT; /* cannot remove what isn't there */
5391 mddev->pers->quiesce(mddev, 1);
5393 err = bitmap_create(mddev);
5395 err = bitmap_load(mddev);
5397 if (fd < 0 || err) {
5398 bitmap_destroy(mddev);
5399 fd = -1; /* make sure to put the file */
5401 mddev->pers->quiesce(mddev, 0);
5404 if (mddev->bitmap_info.file) {
5405 restore_bitmap_write_access(mddev->bitmap_info.file);
5406 fput(mddev->bitmap_info.file);
5408 mddev->bitmap_info.file = NULL;
5415 * set_array_info is used two different ways
5416 * The original usage is when creating a new array.
5417 * In this usage, raid_disks is > 0 and it together with
5418 * level, size, not_persistent,layout,chunksize determine the
5419 * shape of the array.
5420 * This will always create an array with a type-0.90.0 superblock.
5421 * The newer usage is when assembling an array.
5422 * In this case raid_disks will be 0, and the major_version field is
5423 * use to determine which style super-blocks are to be found on the devices.
5424 * The minor and patch _version numbers are also kept incase the
5425 * super_block handler wishes to interpret them.
5427 static int set_array_info(mddev_t * mddev, mdu_array_info_t *info)
5430 if (info->raid_disks == 0) {
5431 /* just setting version number for superblock loading */
5432 if (info->major_version < 0 ||
5433 info->major_version >= ARRAY_SIZE(super_types) ||
5434 super_types[info->major_version].name == NULL) {
5435 /* maybe try to auto-load a module? */
5437 "md: superblock version %d not known\n",
5438 info->major_version);
5441 mddev->major_version = info->major_version;
5442 mddev->minor_version = info->minor_version;
5443 mddev->patch_version = info->patch_version;
5444 mddev->persistent = !info->not_persistent;
5445 /* ensure mddev_put doesn't delete this now that there
5446 * is some minimal configuration.
5448 mddev->ctime = get_seconds();
5451 mddev->major_version = MD_MAJOR_VERSION;
5452 mddev->minor_version = MD_MINOR_VERSION;
5453 mddev->patch_version = MD_PATCHLEVEL_VERSION;
5454 mddev->ctime = get_seconds();
5456 mddev->level = info->level;
5457 mddev->clevel[0] = 0;
5458 mddev->dev_sectors = 2 * (sector_t)info->size;
5459 mddev->raid_disks = info->raid_disks;
5460 /* don't set md_minor, it is determined by which /dev/md* was
5463 if (info->state & (1<<MD_SB_CLEAN))
5464 mddev->recovery_cp = MaxSector;
5466 mddev->recovery_cp = 0;
5467 mddev->persistent = ! info->not_persistent;
5468 mddev->external = 0;
5470 mddev->layout = info->layout;
5471 mddev->chunk_sectors = info->chunk_size >> 9;
5473 mddev->max_disks = MD_SB_DISKS;
5475 if (mddev->persistent)
5477 set_bit(MD_CHANGE_DEVS, &mddev->flags);
5479 mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9;
5480 mddev->bitmap_info.offset = 0;
5482 mddev->reshape_position = MaxSector;
5485 * Generate a 128 bit UUID
5487 get_random_bytes(mddev->uuid, 16);
5489 mddev->new_level = mddev->level;
5490 mddev->new_chunk_sectors = mddev->chunk_sectors;
5491 mddev->new_layout = mddev->layout;
5492 mddev->delta_disks = 0;
5497 void md_set_array_sectors(mddev_t *mddev, sector_t array_sectors)
5499 WARN(!mddev_is_locked(mddev), "%s: unlocked mddev!\n", __func__);
5501 if (mddev->external_size)
5504 mddev->array_sectors = array_sectors;
5506 EXPORT_SYMBOL(md_set_array_sectors);
5508 static int update_size(mddev_t *mddev, sector_t num_sectors)
5512 int fit = (num_sectors == 0);
5514 if (mddev->pers->resize == NULL)
5516 /* The "num_sectors" is the number of sectors of each device that
5517 * is used. This can only make sense for arrays with redundancy.
5518 * linear and raid0 always use whatever space is available. We can only
5519 * consider changing this number if no resync or reconstruction is
5520 * happening, and if the new size is acceptable. It must fit before the
5521 * sb_start or, if that is <data_offset, it must fit before the size
5522 * of each device. If num_sectors is zero, we find the largest size
5526 if (mddev->sync_thread)
5529 /* Sorry, cannot grow a bitmap yet, just remove it,
5533 list_for_each_entry(rdev, &mddev->disks, same_set) {
5534 sector_t avail = rdev->sectors;
5536 if (fit && (num_sectors == 0 || num_sectors > avail))
5537 num_sectors = avail;
5538 if (avail < num_sectors)
5541 rv = mddev->pers->resize(mddev, num_sectors);
5543 revalidate_disk(mddev->gendisk);
5547 static int update_raid_disks(mddev_t *mddev, int raid_disks)
5550 /* change the number of raid disks */
5551 if (mddev->pers->check_reshape == NULL)
5553 if (raid_disks <= 0 ||
5554 (mddev->max_disks && raid_disks >= mddev->max_disks))
5556 if (mddev->sync_thread || mddev->reshape_position != MaxSector)
5558 mddev->delta_disks = raid_disks - mddev->raid_disks;
5560 rv = mddev->pers->check_reshape(mddev);
5566 * update_array_info is used to change the configuration of an
5568 * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
5569 * fields in the info are checked against the array.
5570 * Any differences that cannot be handled will cause an error.
5571 * Normally, only one change can be managed at a time.
5573 static int update_array_info(mddev_t *mddev, mdu_array_info_t *info)
5579 /* calculate expected state,ignoring low bits */
5580 if (mddev->bitmap && mddev->bitmap_info.offset)
5581 state |= (1 << MD_SB_BITMAP_PRESENT);
5583 if (mddev->major_version != info->major_version ||
5584 mddev->minor_version != info->minor_version ||
5585 /* mddev->patch_version != info->patch_version || */
5586 mddev->ctime != info->ctime ||
5587 mddev->level != info->level ||
5588 /* mddev->layout != info->layout || */
5589 !mddev->persistent != info->not_persistent||
5590 mddev->chunk_sectors != info->chunk_size >> 9 ||
5591 /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
5592 ((state^info->state) & 0xfffffe00)
5595 /* Check there is only one change */
5596 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
5598 if (mddev->raid_disks != info->raid_disks)
5600 if (mddev->layout != info->layout)
5602 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT))
5609 if (mddev->layout != info->layout) {
5611 * we don't need to do anything at the md level, the
5612 * personality will take care of it all.
5614 if (mddev->pers->check_reshape == NULL)
5617 mddev->new_layout = info->layout;
5618 rv = mddev->pers->check_reshape(mddev);
5620 mddev->new_layout = mddev->layout;
5624 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
5625 rv = update_size(mddev, (sector_t)info->size * 2);
5627 if (mddev->raid_disks != info->raid_disks)
5628 rv = update_raid_disks(mddev, info->raid_disks);
5630 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) {
5631 if (mddev->pers->quiesce == NULL)
5633 if (mddev->recovery || mddev->sync_thread)
5635 if (info->state & (1<<MD_SB_BITMAP_PRESENT)) {
5636 /* add the bitmap */
5639 if (mddev->bitmap_info.default_offset == 0)
5641 mddev->bitmap_info.offset =
5642 mddev->bitmap_info.default_offset;
5643 mddev->pers->quiesce(mddev, 1);
5644 rv = bitmap_create(mddev);
5646 rv = bitmap_load(mddev);
5648 bitmap_destroy(mddev);
5649 mddev->pers->quiesce(mddev, 0);
5651 /* remove the bitmap */
5654 if (mddev->bitmap->file)
5656 mddev->pers->quiesce(mddev, 1);
5657 bitmap_destroy(mddev);
5658 mddev->pers->quiesce(mddev, 0);
5659 mddev->bitmap_info.offset = 0;
5662 md_update_sb(mddev, 1);
5666 static int set_disk_faulty(mddev_t *mddev, dev_t dev)
5670 if (mddev->pers == NULL)
5673 rdev = find_rdev(mddev, dev);
5677 md_error(mddev, rdev);
5682 * We have a problem here : there is no easy way to give a CHS
5683 * virtual geometry. We currently pretend that we have a 2 heads
5684 * 4 sectors (with a BIG number of cylinders...). This drives
5685 * dosfs just mad... ;-)
5687 static int md_getgeo(struct block_device *bdev, struct hd_geometry *geo)
5689 mddev_t *mddev = bdev->bd_disk->private_data;
5693 geo->cylinders = mddev->array_sectors / 8;
5697 static int md_ioctl(struct block_device *bdev, fmode_t mode,
5698 unsigned int cmd, unsigned long arg)
5701 void __user *argp = (void __user *)arg;
5702 mddev_t *mddev = NULL;
5705 if (!capable(CAP_SYS_ADMIN))
5709 * Commands dealing with the RAID driver but not any
5715 err = get_version(argp);
5718 case PRINT_RAID_DEBUG:
5726 autostart_arrays(arg);
5733 * Commands creating/starting a new array:
5736 mddev = bdev->bd_disk->private_data;
5743 err = mddev_lock(mddev);
5746 "md: ioctl lock interrupted, reason %d, cmd %d\n",
5753 case SET_ARRAY_INFO:
5755 mdu_array_info_t info;
5757 memset(&info, 0, sizeof(info));
5758 else if (copy_from_user(&info, argp, sizeof(info))) {
5763 err = update_array_info(mddev, &info);
5765 printk(KERN_WARNING "md: couldn't update"
5766 " array info. %d\n", err);
5771 if (!list_empty(&mddev->disks)) {
5773 "md: array %s already has disks!\n",
5778 if (mddev->raid_disks) {
5780 "md: array %s already initialised!\n",
5785 err = set_array_info(mddev, &info);
5787 printk(KERN_WARNING "md: couldn't set"
5788 " array info. %d\n", err);
5798 * Commands querying/configuring an existing array:
5800 /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
5801 * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */
5802 if ((!mddev->raid_disks && !mddev->external)
5803 && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY
5804 && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE
5805 && cmd != GET_BITMAP_FILE) {
5811 * Commands even a read-only array can execute:
5815 case GET_ARRAY_INFO:
5816 err = get_array_info(mddev, argp);
5819 case GET_BITMAP_FILE:
5820 err = get_bitmap_file(mddev, argp);
5824 err = get_disk_info(mddev, argp);
5827 case RESTART_ARRAY_RW:
5828 err = restart_array(mddev);
5832 err = do_md_stop(mddev, 0, 1);
5836 err = md_set_readonly(mddev, 1);
5840 if (get_user(ro, (int __user *)(arg))) {
5846 /* if the bdev is going readonly the value of mddev->ro
5847 * does not matter, no writes are coming
5852 /* are we are already prepared for writes? */
5856 /* transitioning to readauto need only happen for
5857 * arrays that call md_write_start
5860 err = restart_array(mddev);
5863 set_disk_ro(mddev->gendisk, 0);
5870 * The remaining ioctls are changing the state of the
5871 * superblock, so we do not allow them on read-only arrays.
5872 * However non-MD ioctls (e.g. get-size) will still come through
5873 * here and hit the 'default' below, so only disallow
5874 * 'md' ioctls, and switch to rw mode if started auto-readonly.
5876 if (_IOC_TYPE(cmd) == MD_MAJOR && mddev->ro && mddev->pers) {
5877 if (mddev->ro == 2) {
5879 sysfs_notify_dirent_safe(mddev->sysfs_state);
5880 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5881 md_wakeup_thread(mddev->thread);
5892 mdu_disk_info_t info;
5893 if (copy_from_user(&info, argp, sizeof(info)))
5896 err = add_new_disk(mddev, &info);
5900 case HOT_REMOVE_DISK:
5901 err = hot_remove_disk(mddev, new_decode_dev(arg));
5905 err = hot_add_disk(mddev, new_decode_dev(arg));
5908 case SET_DISK_FAULTY:
5909 err = set_disk_faulty(mddev, new_decode_dev(arg));
5913 err = do_md_run(mddev);
5916 case SET_BITMAP_FILE:
5917 err = set_bitmap_file(mddev, (int)arg);
5927 if (mddev->hold_active == UNTIL_IOCTL &&
5929 mddev->hold_active = 0;
5930 mddev_unlock(mddev);
5939 #ifdef CONFIG_COMPAT
5940 static int md_compat_ioctl(struct block_device *bdev, fmode_t mode,
5941 unsigned int cmd, unsigned long arg)
5944 case HOT_REMOVE_DISK:
5946 case SET_DISK_FAULTY:
5947 case SET_BITMAP_FILE:
5948 /* These take in integer arg, do not convert */
5951 arg = (unsigned long)compat_ptr(arg);
5955 return md_ioctl(bdev, mode, cmd, arg);
5957 #endif /* CONFIG_COMPAT */
5959 static int md_open(struct block_device *bdev, fmode_t mode)
5962 * Succeed if we can lock the mddev, which confirms that
5963 * it isn't being stopped right now.
5965 mddev_t *mddev = mddev_find(bdev->bd_dev);
5968 if (mddev->gendisk != bdev->bd_disk) {
5969 /* we are racing with mddev_put which is discarding this
5973 /* Wait until bdev->bd_disk is definitely gone */
5974 flush_workqueue(md_misc_wq);
5975 /* Then retry the open from the top */
5976 return -ERESTARTSYS;
5978 BUG_ON(mddev != bdev->bd_disk->private_data);
5980 if ((err = mutex_lock_interruptible(&mddev->open_mutex)))
5984 atomic_inc(&mddev->openers);
5985 mutex_unlock(&mddev->open_mutex);
5987 check_disk_size_change(mddev->gendisk, bdev);
5992 static int md_release(struct gendisk *disk, fmode_t mode)
5994 mddev_t *mddev = disk->private_data;
5997 atomic_dec(&mddev->openers);
6002 static const struct block_device_operations md_fops =
6004 .owner = THIS_MODULE,
6006 .release = md_release,
6008 #ifdef CONFIG_COMPAT
6009 .compat_ioctl = md_compat_ioctl,
6011 .getgeo = md_getgeo,
6014 static int md_thread(void * arg)
6016 mdk_thread_t *thread = arg;
6019 * md_thread is a 'system-thread', it's priority should be very
6020 * high. We avoid resource deadlocks individually in each
6021 * raid personality. (RAID5 does preallocation) We also use RR and
6022 * the very same RT priority as kswapd, thus we will never get
6023 * into a priority inversion deadlock.
6025 * we definitely have to have equal or higher priority than
6026 * bdflush, otherwise bdflush will deadlock if there are too
6027 * many dirty RAID5 blocks.
6030 allow_signal(SIGKILL);
6031 while (!kthread_should_stop()) {
6033 /* We need to wait INTERRUPTIBLE so that
6034 * we don't add to the load-average.
6035 * That means we need to be sure no signals are
6038 if (signal_pending(current))
6039 flush_signals(current);
6041 wait_event_interruptible_timeout
6043 test_bit(THREAD_WAKEUP, &thread->flags)
6044 || kthread_should_stop(),
6047 if (test_and_clear_bit(THREAD_WAKEUP, &thread->flags))
6048 thread->run(thread->mddev);
6054 void md_wakeup_thread(mdk_thread_t *thread)
6057 dprintk("md: waking up MD thread %s.\n", thread->tsk->comm);
6058 set_bit(THREAD_WAKEUP, &thread->flags);
6059 wake_up(&thread->wqueue);
6063 mdk_thread_t *md_register_thread(void (*run) (mddev_t *), mddev_t *mddev,
6066 mdk_thread_t *thread;
6068 thread = kzalloc(sizeof(mdk_thread_t), GFP_KERNEL);
6072 init_waitqueue_head(&thread->wqueue);
6075 thread->mddev = mddev;
6076 thread->timeout = MAX_SCHEDULE_TIMEOUT;
6077 thread->tsk = kthread_run(md_thread, thread,
6079 mdname(thread->mddev),
6080 name ?: mddev->pers->name);
6081 if (IS_ERR(thread->tsk)) {
6088 void md_unregister_thread(mdk_thread_t *thread)
6092 dprintk("interrupting MD-thread pid %d\n", task_pid_nr(thread->tsk));
6094 kthread_stop(thread->tsk);
6098 void md_error(mddev_t *mddev, mdk_rdev_t *rdev)
6105 if (!rdev || test_bit(Faulty, &rdev->flags))
6108 if (mddev->external)
6109 set_bit(Blocked, &rdev->flags);
6111 dprintk("md_error dev:%s, rdev:(%d:%d), (caller: %p,%p,%p,%p).\n",
6113 MAJOR(rdev->bdev->bd_dev), MINOR(rdev->bdev->bd_dev),
6114 __builtin_return_address(0),__builtin_return_address(1),
6115 __builtin_return_address(2),__builtin_return_address(3));
6119 if (!mddev->pers->error_handler)
6121 mddev->pers->error_handler(mddev,rdev);
6122 if (mddev->degraded)
6123 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
6124 sysfs_notify_dirent_safe(rdev->sysfs_state);
6125 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6126 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6127 md_wakeup_thread(mddev->thread);
6128 if (mddev->event_work.func)
6129 queue_work(md_misc_wq, &mddev->event_work);
6130 md_new_event_inintr(mddev);
6133 /* seq_file implementation /proc/mdstat */
6135 static void status_unused(struct seq_file *seq)
6140 seq_printf(seq, "unused devices: ");
6142 list_for_each_entry(rdev, &pending_raid_disks, same_set) {
6143 char b[BDEVNAME_SIZE];
6145 seq_printf(seq, "%s ",
6146 bdevname(rdev->bdev,b));
6149 seq_printf(seq, "<none>");
6151 seq_printf(seq, "\n");
6155 static void status_resync(struct seq_file *seq, mddev_t * mddev)
6157 sector_t max_sectors, resync, res;
6158 unsigned long dt, db;
6161 unsigned int per_milli;
6163 resync = mddev->curr_resync - atomic_read(&mddev->recovery_active);
6165 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
6166 max_sectors = mddev->resync_max_sectors;
6168 max_sectors = mddev->dev_sectors;
6171 * Should not happen.
6177 /* Pick 'scale' such that (resync>>scale)*1000 will fit
6178 * in a sector_t, and (max_sectors>>scale) will fit in a
6179 * u32, as those are the requirements for sector_div.
6180 * Thus 'scale' must be at least 10
6183 if (sizeof(sector_t) > sizeof(unsigned long)) {
6184 while ( max_sectors/2 > (1ULL<<(scale+32)))
6187 res = (resync>>scale)*1000;
6188 sector_div(res, (u32)((max_sectors>>scale)+1));
6192 int i, x = per_milli/50, y = 20-x;
6193 seq_printf(seq, "[");
6194 for (i = 0; i < x; i++)
6195 seq_printf(seq, "=");
6196 seq_printf(seq, ">");
6197 for (i = 0; i < y; i++)
6198 seq_printf(seq, ".");
6199 seq_printf(seq, "] ");
6201 seq_printf(seq, " %s =%3u.%u%% (%llu/%llu)",
6202 (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)?
6204 (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)?
6206 (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
6207 "resync" : "recovery"))),
6208 per_milli/10, per_milli % 10,
6209 (unsigned long long) resync/2,
6210 (unsigned long long) max_sectors/2);
6213 * dt: time from mark until now
6214 * db: blocks written from mark until now
6215 * rt: remaining time
6217 * rt is a sector_t, so could be 32bit or 64bit.
6218 * So we divide before multiply in case it is 32bit and close
6220 * We scale the divisor (db) by 32 to avoid loosing precision
6221 * near the end of resync when the number of remaining sectors
6223 * We then divide rt by 32 after multiplying by db to compensate.
6224 * The '+1' avoids division by zero if db is very small.
6226 dt = ((jiffies - mddev->resync_mark) / HZ);
6228 db = (mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active))
6229 - mddev->resync_mark_cnt;
6231 rt = max_sectors - resync; /* number of remaining sectors */
6232 sector_div(rt, db/32+1);
6236 seq_printf(seq, " finish=%lu.%lumin", (unsigned long)rt / 60,
6237 ((unsigned long)rt % 60)/6);
6239 seq_printf(seq, " speed=%ldK/sec", db/2/dt);
6242 static void *md_seq_start(struct seq_file *seq, loff_t *pos)
6244 struct list_head *tmp;
6254 spin_lock(&all_mddevs_lock);
6255 list_for_each(tmp,&all_mddevs)
6257 mddev = list_entry(tmp, mddev_t, all_mddevs);
6259 spin_unlock(&all_mddevs_lock);
6262 spin_unlock(&all_mddevs_lock);
6264 return (void*)2;/* tail */
6268 static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
6270 struct list_head *tmp;
6271 mddev_t *next_mddev, *mddev = v;
6277 spin_lock(&all_mddevs_lock);
6279 tmp = all_mddevs.next;
6281 tmp = mddev->all_mddevs.next;
6282 if (tmp != &all_mddevs)
6283 next_mddev = mddev_get(list_entry(tmp,mddev_t,all_mddevs));
6285 next_mddev = (void*)2;
6288 spin_unlock(&all_mddevs_lock);
6296 static void md_seq_stop(struct seq_file *seq, void *v)
6300 if (mddev && v != (void*)1 && v != (void*)2)
6304 struct mdstat_info {
6308 static int md_seq_show(struct seq_file *seq, void *v)
6313 struct mdstat_info *mi = seq->private;
6314 struct bitmap *bitmap;
6316 if (v == (void*)1) {
6317 struct mdk_personality *pers;
6318 seq_printf(seq, "Personalities : ");
6319 spin_lock(&pers_lock);
6320 list_for_each_entry(pers, &pers_list, list)
6321 seq_printf(seq, "[%s] ", pers->name);
6323 spin_unlock(&pers_lock);
6324 seq_printf(seq, "\n");
6325 mi->event = atomic_read(&md_event_count);
6328 if (v == (void*)2) {
6333 if (mddev_lock(mddev) < 0)
6336 if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
6337 seq_printf(seq, "%s : %sactive", mdname(mddev),
6338 mddev->pers ? "" : "in");
6341 seq_printf(seq, " (read-only)");
6343 seq_printf(seq, " (auto-read-only)");
6344 seq_printf(seq, " %s", mddev->pers->name);
6348 list_for_each_entry(rdev, &mddev->disks, same_set) {
6349 char b[BDEVNAME_SIZE];
6350 seq_printf(seq, " %s[%d]",
6351 bdevname(rdev->bdev,b), rdev->desc_nr);
6352 if (test_bit(WriteMostly, &rdev->flags))
6353 seq_printf(seq, "(W)");
6354 if (test_bit(Faulty, &rdev->flags)) {
6355 seq_printf(seq, "(F)");
6357 } else if (rdev->raid_disk < 0)
6358 seq_printf(seq, "(S)"); /* spare */
6359 sectors += rdev->sectors;
6362 if (!list_empty(&mddev->disks)) {
6364 seq_printf(seq, "\n %llu blocks",
6365 (unsigned long long)
6366 mddev->array_sectors / 2);
6368 seq_printf(seq, "\n %llu blocks",
6369 (unsigned long long)sectors / 2);
6371 if (mddev->persistent) {
6372 if (mddev->major_version != 0 ||
6373 mddev->minor_version != 90) {
6374 seq_printf(seq," super %d.%d",
6375 mddev->major_version,
6376 mddev->minor_version);
6378 } else if (mddev->external)
6379 seq_printf(seq, " super external:%s",
6380 mddev->metadata_type);
6382 seq_printf(seq, " super non-persistent");
6385 mddev->pers->status(seq, mddev);
6386 seq_printf(seq, "\n ");
6387 if (mddev->pers->sync_request) {
6388 if (mddev->curr_resync > 2) {
6389 status_resync(seq, mddev);
6390 seq_printf(seq, "\n ");
6391 } else if (mddev->curr_resync == 1 || mddev->curr_resync == 2)
6392 seq_printf(seq, "\tresync=DELAYED\n ");
6393 else if (mddev->recovery_cp < MaxSector)
6394 seq_printf(seq, "\tresync=PENDING\n ");
6397 seq_printf(seq, "\n ");
6399 if ((bitmap = mddev->bitmap)) {
6400 unsigned long chunk_kb;
6401 unsigned long flags;
6402 spin_lock_irqsave(&bitmap->lock, flags);
6403 chunk_kb = mddev->bitmap_info.chunksize >> 10;
6404 seq_printf(seq, "bitmap: %lu/%lu pages [%luKB], "
6406 bitmap->pages - bitmap->missing_pages,
6408 (bitmap->pages - bitmap->missing_pages)
6409 << (PAGE_SHIFT - 10),
6410 chunk_kb ? chunk_kb : mddev->bitmap_info.chunksize,
6411 chunk_kb ? "KB" : "B");
6413 seq_printf(seq, ", file: ");
6414 seq_path(seq, &bitmap->file->f_path, " \t\n");
6417 seq_printf(seq, "\n");
6418 spin_unlock_irqrestore(&bitmap->lock, flags);
6421 seq_printf(seq, "\n");
6423 mddev_unlock(mddev);
6428 static const struct seq_operations md_seq_ops = {
6429 .start = md_seq_start,
6430 .next = md_seq_next,
6431 .stop = md_seq_stop,
6432 .show = md_seq_show,
6435 static int md_seq_open(struct inode *inode, struct file *file)
6438 struct mdstat_info *mi = kmalloc(sizeof(*mi), GFP_KERNEL);
6442 error = seq_open(file, &md_seq_ops);
6446 struct seq_file *p = file->private_data;
6448 mi->event = atomic_read(&md_event_count);
6453 static unsigned int mdstat_poll(struct file *filp, poll_table *wait)
6455 struct seq_file *m = filp->private_data;
6456 struct mdstat_info *mi = m->private;
6459 poll_wait(filp, &md_event_waiters, wait);
6461 /* always allow read */
6462 mask = POLLIN | POLLRDNORM;
6464 if (mi->event != atomic_read(&md_event_count))
6465 mask |= POLLERR | POLLPRI;
6469 static const struct file_operations md_seq_fops = {
6470 .owner = THIS_MODULE,
6471 .open = md_seq_open,
6473 .llseek = seq_lseek,
6474 .release = seq_release_private,
6475 .poll = mdstat_poll,
6478 int register_md_personality(struct mdk_personality *p)
6480 spin_lock(&pers_lock);
6481 list_add_tail(&p->list, &pers_list);
6482 printk(KERN_INFO "md: %s personality registered for level %d\n", p->name, p->level);
6483 spin_unlock(&pers_lock);
6487 int unregister_md_personality(struct mdk_personality *p)
6489 printk(KERN_INFO "md: %s personality unregistered\n", p->name);
6490 spin_lock(&pers_lock);
6491 list_del_init(&p->list);
6492 spin_unlock(&pers_lock);
6496 static int is_mddev_idle(mddev_t *mddev, int init)
6504 rdev_for_each_rcu(rdev, mddev) {
6505 struct gendisk *disk = rdev->bdev->bd_contains->bd_disk;
6506 curr_events = (int)part_stat_read(&disk->part0, sectors[0]) +
6507 (int)part_stat_read(&disk->part0, sectors[1]) -
6508 atomic_read(&disk->sync_io);
6509 /* sync IO will cause sync_io to increase before the disk_stats
6510 * as sync_io is counted when a request starts, and
6511 * disk_stats is counted when it completes.
6512 * So resync activity will cause curr_events to be smaller than
6513 * when there was no such activity.
6514 * non-sync IO will cause disk_stat to increase without
6515 * increasing sync_io so curr_events will (eventually)
6516 * be larger than it was before. Once it becomes
6517 * substantially larger, the test below will cause
6518 * the array to appear non-idle, and resync will slow
6520 * If there is a lot of outstanding resync activity when
6521 * we set last_event to curr_events, then all that activity
6522 * completing might cause the array to appear non-idle
6523 * and resync will be slowed down even though there might
6524 * not have been non-resync activity. This will only
6525 * happen once though. 'last_events' will soon reflect
6526 * the state where there is little or no outstanding
6527 * resync requests, and further resync activity will
6528 * always make curr_events less than last_events.
6531 if (init || curr_events - rdev->last_events > 64) {
6532 rdev->last_events = curr_events;
6540 void md_done_sync(mddev_t *mddev, int blocks, int ok)
6542 /* another "blocks" (512byte) blocks have been synced */
6543 atomic_sub(blocks, &mddev->recovery_active);
6544 wake_up(&mddev->recovery_wait);
6546 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6547 md_wakeup_thread(mddev->thread);
6548 // stop recovery, signal do_sync ....
6553 /* md_write_start(mddev, bi)
6554 * If we need to update some array metadata (e.g. 'active' flag
6555 * in superblock) before writing, schedule a superblock update
6556 * and wait for it to complete.
6558 void md_write_start(mddev_t *mddev, struct bio *bi)
6561 if (bio_data_dir(bi) != WRITE)
6564 BUG_ON(mddev->ro == 1);
6565 if (mddev->ro == 2) {
6566 /* need to switch to read/write */
6568 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6569 md_wakeup_thread(mddev->thread);
6570 md_wakeup_thread(mddev->sync_thread);
6573 atomic_inc(&mddev->writes_pending);
6574 if (mddev->safemode == 1)
6575 mddev->safemode = 0;
6576 if (mddev->in_sync) {
6577 spin_lock_irq(&mddev->write_lock);
6578 if (mddev->in_sync) {
6580 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
6581 set_bit(MD_CHANGE_PENDING, &mddev->flags);
6582 md_wakeup_thread(mddev->thread);
6585 spin_unlock_irq(&mddev->write_lock);
6588 sysfs_notify_dirent_safe(mddev->sysfs_state);
6589 wait_event(mddev->sb_wait,
6590 !test_bit(MD_CHANGE_PENDING, &mddev->flags));
6593 void md_write_end(mddev_t *mddev)
6595 if (atomic_dec_and_test(&mddev->writes_pending)) {
6596 if (mddev->safemode == 2)
6597 md_wakeup_thread(mddev->thread);
6598 else if (mddev->safemode_delay)
6599 mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay);
6603 /* md_allow_write(mddev)
6604 * Calling this ensures that the array is marked 'active' so that writes
6605 * may proceed without blocking. It is important to call this before
6606 * attempting a GFP_KERNEL allocation while holding the mddev lock.
6607 * Must be called with mddev_lock held.
6609 * In the ->external case MD_CHANGE_CLEAN can not be cleared until mddev->lock
6610 * is dropped, so return -EAGAIN after notifying userspace.
6612 int md_allow_write(mddev_t *mddev)
6618 if (!mddev->pers->sync_request)
6621 spin_lock_irq(&mddev->write_lock);
6622 if (mddev->in_sync) {
6624 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
6625 set_bit(MD_CHANGE_PENDING, &mddev->flags);
6626 if (mddev->safemode_delay &&
6627 mddev->safemode == 0)
6628 mddev->safemode = 1;
6629 spin_unlock_irq(&mddev->write_lock);
6630 md_update_sb(mddev, 0);
6631 sysfs_notify_dirent_safe(mddev->sysfs_state);
6633 spin_unlock_irq(&mddev->write_lock);
6635 if (test_bit(MD_CHANGE_PENDING, &mddev->flags))
6640 EXPORT_SYMBOL_GPL(md_allow_write);
6642 void md_unplug(mddev_t *mddev)
6645 blk_unplug(mddev->queue);
6647 mddev->plug->unplug_fn(mddev->plug);
6650 #define SYNC_MARKS 10
6651 #define SYNC_MARK_STEP (3*HZ)
6652 void md_do_sync(mddev_t *mddev)
6655 unsigned int currspeed = 0,
6657 sector_t max_sectors,j, io_sectors;
6658 unsigned long mark[SYNC_MARKS];
6659 sector_t mark_cnt[SYNC_MARKS];
6661 struct list_head *tmp;
6662 sector_t last_check;
6667 /* just incase thread restarts... */
6668 if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
6670 if (mddev->ro) /* never try to sync a read-only array */
6673 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
6674 if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
6675 desc = "data-check";
6676 else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
6677 desc = "requested-resync";
6680 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
6685 /* we overload curr_resync somewhat here.
6686 * 0 == not engaged in resync at all
6687 * 2 == checking that there is no conflict with another sync
6688 * 1 == like 2, but have yielded to allow conflicting resync to
6690 * other == active in resync - this many blocks
6692 * Before starting a resync we must have set curr_resync to
6693 * 2, and then checked that every "conflicting" array has curr_resync
6694 * less than ours. When we find one that is the same or higher
6695 * we wait on resync_wait. To avoid deadlock, we reduce curr_resync
6696 * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
6697 * This will mean we have to start checking from the beginning again.
6702 mddev->curr_resync = 2;
6705 if (kthread_should_stop())
6706 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6708 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
6710 for_each_mddev(mddev2, tmp) {
6711 if (mddev2 == mddev)
6713 if (!mddev->parallel_resync
6714 && mddev2->curr_resync
6715 && match_mddev_units(mddev, mddev2)) {
6717 if (mddev < mddev2 && mddev->curr_resync == 2) {
6718 /* arbitrarily yield */
6719 mddev->curr_resync = 1;
6720 wake_up(&resync_wait);
6722 if (mddev > mddev2 && mddev->curr_resync == 1)
6723 /* no need to wait here, we can wait the next
6724 * time 'round when curr_resync == 2
6727 /* We need to wait 'interruptible' so as not to
6728 * contribute to the load average, and not to
6729 * be caught by 'softlockup'
6731 prepare_to_wait(&resync_wait, &wq, TASK_INTERRUPTIBLE);
6732 if (!kthread_should_stop() &&
6733 mddev2->curr_resync >= mddev->curr_resync) {
6734 printk(KERN_INFO "md: delaying %s of %s"
6735 " until %s has finished (they"
6736 " share one or more physical units)\n",
6737 desc, mdname(mddev), mdname(mddev2));
6739 if (signal_pending(current))
6740 flush_signals(current);
6742 finish_wait(&resync_wait, &wq);
6745 finish_wait(&resync_wait, &wq);
6748 } while (mddev->curr_resync < 2);
6751 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
6752 /* resync follows the size requested by the personality,
6753 * which defaults to physical size, but can be virtual size
6755 max_sectors = mddev->resync_max_sectors;
6756 mddev->resync_mismatches = 0;
6757 /* we don't use the checkpoint if there's a bitmap */
6758 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
6759 j = mddev->resync_min;
6760 else if (!mddev->bitmap)
6761 j = mddev->recovery_cp;
6763 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
6764 max_sectors = mddev->dev_sectors;
6766 /* recovery follows the physical size of devices */
6767 max_sectors = mddev->dev_sectors;
6770 list_for_each_entry_rcu(rdev, &mddev->disks, same_set)
6771 if (rdev->raid_disk >= 0 &&
6772 !test_bit(Faulty, &rdev->flags) &&
6773 !test_bit(In_sync, &rdev->flags) &&
6774 rdev->recovery_offset < j)
6775 j = rdev->recovery_offset;
6779 printk(KERN_INFO "md: %s of RAID array %s\n", desc, mdname(mddev));
6780 printk(KERN_INFO "md: minimum _guaranteed_ speed:"
6781 " %d KB/sec/disk.\n", speed_min(mddev));
6782 printk(KERN_INFO "md: using maximum available idle IO bandwidth "
6783 "(but not more than %d KB/sec) for %s.\n",
6784 speed_max(mddev), desc);
6786 is_mddev_idle(mddev, 1); /* this initializes IO event counters */
6789 for (m = 0; m < SYNC_MARKS; m++) {
6791 mark_cnt[m] = io_sectors;
6794 mddev->resync_mark = mark[last_mark];
6795 mddev->resync_mark_cnt = mark_cnt[last_mark];
6798 * Tune reconstruction:
6800 window = 32*(PAGE_SIZE/512);
6801 printk(KERN_INFO "md: using %dk window, over a total of %llu blocks.\n",
6802 window/2,(unsigned long long) max_sectors/2);
6804 atomic_set(&mddev->recovery_active, 0);
6809 "md: resuming %s of %s from checkpoint.\n",
6810 desc, mdname(mddev));
6811 mddev->curr_resync = j;
6813 mddev->curr_resync_completed = mddev->curr_resync;
6815 while (j < max_sectors) {
6820 if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
6821 ((mddev->curr_resync > mddev->curr_resync_completed &&
6822 (mddev->curr_resync - mddev->curr_resync_completed)
6823 > (max_sectors >> 4)) ||
6824 (j - mddev->curr_resync_completed)*2
6825 >= mddev->resync_max - mddev->curr_resync_completed
6827 /* time to update curr_resync_completed */
6829 wait_event(mddev->recovery_wait,
6830 atomic_read(&mddev->recovery_active) == 0);
6831 mddev->curr_resync_completed =
6833 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
6834 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
6837 while (j >= mddev->resync_max && !kthread_should_stop()) {
6838 /* As this condition is controlled by user-space,
6839 * we can block indefinitely, so use '_interruptible'
6840 * to avoid triggering warnings.
6842 flush_signals(current); /* just in case */
6843 wait_event_interruptible(mddev->recovery_wait,
6844 mddev->resync_max > j
6845 || kthread_should_stop());
6848 if (kthread_should_stop())
6851 sectors = mddev->pers->sync_request(mddev, j, &skipped,
6852 currspeed < speed_min(mddev));
6854 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6858 if (!skipped) { /* actual IO requested */
6859 io_sectors += sectors;
6860 atomic_add(sectors, &mddev->recovery_active);
6864 if (j>1) mddev->curr_resync = j;
6865 mddev->curr_mark_cnt = io_sectors;
6866 if (last_check == 0)
6867 /* this is the earliers that rebuilt will be
6868 * visible in /proc/mdstat
6870 md_new_event(mddev);
6872 if (last_check + window > io_sectors || j == max_sectors)
6875 last_check = io_sectors;
6877 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
6881 if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) {
6883 int next = (last_mark+1) % SYNC_MARKS;
6885 mddev->resync_mark = mark[next];
6886 mddev->resync_mark_cnt = mark_cnt[next];
6887 mark[next] = jiffies;
6888 mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active);
6893 if (kthread_should_stop())
6898 * this loop exits only if either when we are slower than
6899 * the 'hard' speed limit, or the system was IO-idle for
6901 * the system might be non-idle CPU-wise, but we only care
6902 * about not overloading the IO subsystem. (things like an
6903 * e2fsck being done on the RAID array should execute fast)
6908 currspeed = ((unsigned long)(io_sectors-mddev->resync_mark_cnt))/2
6909 /((jiffies-mddev->resync_mark)/HZ +1) +1;
6911 if (currspeed > speed_min(mddev)) {
6912 if ((currspeed > speed_max(mddev)) ||
6913 !is_mddev_idle(mddev, 0)) {
6919 printk(KERN_INFO "md: %s: %s done.\n",mdname(mddev), desc);
6921 * this also signals 'finished resyncing' to md_stop
6926 wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
6928 /* tell personality that we are finished */
6929 mddev->pers->sync_request(mddev, max_sectors, &skipped, 1);
6931 if (!test_bit(MD_RECOVERY_CHECK, &mddev->recovery) &&
6932 mddev->curr_resync > 2) {
6933 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
6934 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
6935 if (mddev->curr_resync >= mddev->recovery_cp) {
6937 "md: checkpointing %s of %s.\n",
6938 desc, mdname(mddev));
6939 mddev->recovery_cp = mddev->curr_resync;
6942 mddev->recovery_cp = MaxSector;
6944 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery))
6945 mddev->curr_resync = MaxSector;
6947 list_for_each_entry_rcu(rdev, &mddev->disks, same_set)
6948 if (rdev->raid_disk >= 0 &&
6949 mddev->delta_disks >= 0 &&
6950 !test_bit(Faulty, &rdev->flags) &&
6951 !test_bit(In_sync, &rdev->flags) &&
6952 rdev->recovery_offset < mddev->curr_resync)
6953 rdev->recovery_offset = mddev->curr_resync;
6957 set_bit(MD_CHANGE_DEVS, &mddev->flags);
6960 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
6961 /* We completed so min/max setting can be forgotten if used. */
6962 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
6963 mddev->resync_min = 0;
6964 mddev->resync_max = MaxSector;
6965 } else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
6966 mddev->resync_min = mddev->curr_resync_completed;
6967 mddev->curr_resync = 0;
6968 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery))
6969 mddev->curr_resync_completed = 0;
6970 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
6971 wake_up(&resync_wait);
6972 set_bit(MD_RECOVERY_DONE, &mddev->recovery);
6973 md_wakeup_thread(mddev->thread);
6978 * got a signal, exit.
6981 "md: md_do_sync() got signal ... exiting\n");
6982 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6986 EXPORT_SYMBOL_GPL(md_do_sync);
6989 static int remove_and_add_spares(mddev_t *mddev)
6994 mddev->curr_resync_completed = 0;
6996 list_for_each_entry(rdev, &mddev->disks, same_set)
6997 if (rdev->raid_disk >= 0 &&
6998 !test_bit(Blocked, &rdev->flags) &&
6999 (test_bit(Faulty, &rdev->flags) ||
7000 ! test_bit(In_sync, &rdev->flags)) &&
7001 atomic_read(&rdev->nr_pending)==0) {
7002 if (mddev->pers->hot_remove_disk(
7003 mddev, rdev->raid_disk)==0) {
7005 sprintf(nm,"rd%d", rdev->raid_disk);
7006 sysfs_remove_link(&mddev->kobj, nm);
7007 rdev->raid_disk = -1;
7011 if (mddev->degraded && ! mddev->ro && !mddev->recovery_disabled) {
7012 list_for_each_entry(rdev, &mddev->disks, same_set) {
7013 if (rdev->raid_disk >= 0 &&
7014 !test_bit(In_sync, &rdev->flags) &&
7015 !test_bit(Blocked, &rdev->flags))
7017 if (rdev->raid_disk < 0
7018 && !test_bit(Faulty, &rdev->flags)) {
7019 rdev->recovery_offset = 0;
7021 hot_add_disk(mddev, rdev) == 0) {
7023 sprintf(nm, "rd%d", rdev->raid_disk);
7024 if (sysfs_create_link(&mddev->kobj,
7026 /* failure here is OK */;
7028 md_new_event(mddev);
7029 set_bit(MD_CHANGE_DEVS, &mddev->flags);
7038 * This routine is regularly called by all per-raid-array threads to
7039 * deal with generic issues like resync and super-block update.
7040 * Raid personalities that don't have a thread (linear/raid0) do not
7041 * need this as they never do any recovery or update the superblock.
7043 * It does not do any resync itself, but rather "forks" off other threads
7044 * to do that as needed.
7045 * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
7046 * "->recovery" and create a thread at ->sync_thread.
7047 * When the thread finishes it sets MD_RECOVERY_DONE
7048 * and wakeups up this thread which will reap the thread and finish up.
7049 * This thread also removes any faulty devices (with nr_pending == 0).
7051 * The overall approach is:
7052 * 1/ if the superblock needs updating, update it.
7053 * 2/ If a recovery thread is running, don't do anything else.
7054 * 3/ If recovery has finished, clean up, possibly marking spares active.
7055 * 4/ If there are any faulty devices, remove them.
7056 * 5/ If array is degraded, try to add spares devices
7057 * 6/ If array has spares or is not in-sync, start a resync thread.
7059 void md_check_recovery(mddev_t *mddev)
7065 bitmap_daemon_work(mddev);
7070 if (signal_pending(current)) {
7071 if (mddev->pers->sync_request && !mddev->external) {
7072 printk(KERN_INFO "md: %s in immediate safe mode\n",
7074 mddev->safemode = 2;
7076 flush_signals(current);
7079 if (mddev->ro && !test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
7082 (mddev->flags & ~ (1<<MD_CHANGE_PENDING)) ||
7083 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
7084 test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
7085 (mddev->external == 0 && mddev->safemode == 1) ||
7086 (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending)
7087 && !mddev->in_sync && mddev->recovery_cp == MaxSector)
7091 if (mddev_trylock(mddev)) {
7095 /* Only thing we do on a ro array is remove
7098 remove_and_add_spares(mddev);
7099 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7103 if (!mddev->external) {
7105 spin_lock_irq(&mddev->write_lock);
7106 if (mddev->safemode &&
7107 !atomic_read(&mddev->writes_pending) &&
7109 mddev->recovery_cp == MaxSector) {
7112 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
7114 if (mddev->safemode == 1)
7115 mddev->safemode = 0;
7116 spin_unlock_irq(&mddev->write_lock);
7118 sysfs_notify_dirent_safe(mddev->sysfs_state);
7122 md_update_sb(mddev, 0);
7124 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
7125 !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
7126 /* resync/recovery still happening */
7127 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7130 if (mddev->sync_thread) {
7131 /* resync has finished, collect result */
7132 md_unregister_thread(mddev->sync_thread);
7133 mddev->sync_thread = NULL;
7134 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
7135 !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
7137 /* activate any spares */
7138 if (mddev->pers->spare_active(mddev))
7139 sysfs_notify(&mddev->kobj, NULL,
7142 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
7143 mddev->pers->finish_reshape)
7144 mddev->pers->finish_reshape(mddev);
7145 md_update_sb(mddev, 1);
7147 /* if array is no-longer degraded, then any saved_raid_disk
7148 * information must be scrapped
7150 if (!mddev->degraded)
7151 list_for_each_entry(rdev, &mddev->disks, same_set)
7152 rdev->saved_raid_disk = -1;
7154 mddev->recovery = 0;
7155 /* flag recovery needed just to double check */
7156 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7157 sysfs_notify_dirent_safe(mddev->sysfs_action);
7158 md_new_event(mddev);
7161 /* Set RUNNING before clearing NEEDED to avoid
7162 * any transients in the value of "sync_action".
7164 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
7165 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7166 /* Clear some bits that don't mean anything, but
7169 clear_bit(MD_RECOVERY_INTR, &mddev->recovery);
7170 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
7172 if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
7174 /* no recovery is running.
7175 * remove any failed drives, then
7176 * add spares if possible.
7177 * Spare are also removed and re-added, to allow
7178 * the personality to fail the re-add.
7181 if (mddev->reshape_position != MaxSector) {
7182 if (mddev->pers->check_reshape == NULL ||
7183 mddev->pers->check_reshape(mddev) != 0)
7184 /* Cannot proceed */
7186 set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
7187 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
7188 } else if ((spares = remove_and_add_spares(mddev))) {
7189 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
7190 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
7191 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
7192 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
7193 } else if (mddev->recovery_cp < MaxSector) {
7194 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
7195 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
7196 } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
7197 /* nothing to be done ... */
7200 if (mddev->pers->sync_request) {
7201 if (spares && mddev->bitmap && ! mddev->bitmap->file) {
7202 /* We are adding a device or devices to an array
7203 * which has the bitmap stored on all devices.
7204 * So make sure all bitmap pages get written
7206 bitmap_write_all(mddev->bitmap);
7208 mddev->sync_thread = md_register_thread(md_do_sync,
7211 if (!mddev->sync_thread) {
7212 printk(KERN_ERR "%s: could not start resync"
7215 /* leave the spares where they are, it shouldn't hurt */
7216 mddev->recovery = 0;
7218 md_wakeup_thread(mddev->sync_thread);
7219 sysfs_notify_dirent_safe(mddev->sysfs_action);
7220 md_new_event(mddev);
7223 if (!mddev->sync_thread) {
7224 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
7225 if (test_and_clear_bit(MD_RECOVERY_RECOVER,
7227 if (mddev->sysfs_action)
7228 sysfs_notify_dirent_safe(mddev->sysfs_action);
7230 mddev_unlock(mddev);
7234 void md_wait_for_blocked_rdev(mdk_rdev_t *rdev, mddev_t *mddev)
7236 sysfs_notify_dirent_safe(rdev->sysfs_state);
7237 wait_event_timeout(rdev->blocked_wait,
7238 !test_bit(Blocked, &rdev->flags),
7239 msecs_to_jiffies(5000));
7240 rdev_dec_pending(rdev, mddev);
7242 EXPORT_SYMBOL(md_wait_for_blocked_rdev);
7244 static int md_notify_reboot(struct notifier_block *this,
7245 unsigned long code, void *x)
7247 struct list_head *tmp;
7250 if ((code == SYS_DOWN) || (code == SYS_HALT) || (code == SYS_POWER_OFF)) {
7252 printk(KERN_INFO "md: stopping all md devices.\n");
7254 for_each_mddev(mddev, tmp)
7255 if (mddev_trylock(mddev)) {
7256 /* Force a switch to readonly even array
7257 * appears to still be in use. Hence
7260 md_set_readonly(mddev, 100);
7261 mddev_unlock(mddev);
7264 * certain more exotic SCSI devices are known to be
7265 * volatile wrt too early system reboots. While the
7266 * right place to handle this issue is the given
7267 * driver, we do want to have a safe RAID driver ...
7274 static struct notifier_block md_notifier = {
7275 .notifier_call = md_notify_reboot,
7277 .priority = INT_MAX, /* before any real devices */
7280 static void md_geninit(void)
7282 dprintk("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
7284 proc_create("mdstat", S_IRUGO, NULL, &md_seq_fops);
7287 static int __init md_init(void)
7291 md_wq = alloc_workqueue("md", WQ_RESCUER, 0);
7295 md_misc_wq = alloc_workqueue("md_misc", 0, 0);
7299 if ((ret = register_blkdev(MD_MAJOR, "md")) < 0)
7302 if ((ret = register_blkdev(0, "mdp")) < 0)
7306 blk_register_region(MKDEV(MD_MAJOR, 0), 1UL<<MINORBITS, THIS_MODULE,
7307 md_probe, NULL, NULL);
7308 blk_register_region(MKDEV(mdp_major, 0), 1UL<<MINORBITS, THIS_MODULE,
7309 md_probe, NULL, NULL);
7311 register_reboot_notifier(&md_notifier);
7312 raid_table_header = register_sysctl_table(raid_root_table);
7318 unregister_blkdev(MD_MAJOR, "md");
7320 destroy_workqueue(md_misc_wq);
7322 destroy_workqueue(md_wq);
7330 * Searches all registered partitions for autorun RAID arrays
7334 static LIST_HEAD(all_detected_devices);
7335 struct detected_devices_node {
7336 struct list_head list;
7340 void md_autodetect_dev(dev_t dev)
7342 struct detected_devices_node *node_detected_dev;
7344 node_detected_dev = kzalloc(sizeof(*node_detected_dev), GFP_KERNEL);
7345 if (node_detected_dev) {
7346 node_detected_dev->dev = dev;
7347 list_add_tail(&node_detected_dev->list, &all_detected_devices);
7349 printk(KERN_CRIT "md: md_autodetect_dev: kzalloc failed"
7350 ", skipping dev(%d,%d)\n", MAJOR(dev), MINOR(dev));
7355 static void autostart_arrays(int part)
7358 struct detected_devices_node *node_detected_dev;
7360 int i_scanned, i_passed;
7365 printk(KERN_INFO "md: Autodetecting RAID arrays.\n");
7367 while (!list_empty(&all_detected_devices) && i_scanned < INT_MAX) {
7369 node_detected_dev = list_entry(all_detected_devices.next,
7370 struct detected_devices_node, list);
7371 list_del(&node_detected_dev->list);
7372 dev = node_detected_dev->dev;
7373 kfree(node_detected_dev);
7374 rdev = md_import_device(dev,0, 90);
7378 if (test_bit(Faulty, &rdev->flags)) {
7382 set_bit(AutoDetected, &rdev->flags);
7383 list_add(&rdev->same_set, &pending_raid_disks);
7387 printk(KERN_INFO "md: Scanned %d and added %d devices.\n",
7388 i_scanned, i_passed);
7390 autorun_devices(part);
7393 #endif /* !MODULE */
7395 static __exit void md_exit(void)
7398 struct list_head *tmp;
7400 blk_unregister_region(MKDEV(MD_MAJOR,0), 1U << MINORBITS);
7401 blk_unregister_region(MKDEV(mdp_major,0), 1U << MINORBITS);
7403 unregister_blkdev(MD_MAJOR,"md");
7404 unregister_blkdev(mdp_major, "mdp");
7405 unregister_reboot_notifier(&md_notifier);
7406 unregister_sysctl_table(raid_table_header);
7407 remove_proc_entry("mdstat", NULL);
7408 for_each_mddev(mddev, tmp) {
7409 export_array(mddev);
7410 mddev->hold_active = 0;
7412 destroy_workqueue(md_misc_wq);
7413 destroy_workqueue(md_wq);
7416 subsys_initcall(md_init);
7417 module_exit(md_exit)
7419 static int get_ro(char *buffer, struct kernel_param *kp)
7421 return sprintf(buffer, "%d", start_readonly);
7423 static int set_ro(const char *val, struct kernel_param *kp)
7426 int num = simple_strtoul(val, &e, 10);
7427 if (*val && (*e == '\0' || *e == '\n')) {
7428 start_readonly = num;
7434 module_param_call(start_ro, set_ro, get_ro, NULL, S_IRUSR|S_IWUSR);
7435 module_param(start_dirty_degraded, int, S_IRUGO|S_IWUSR);
7437 module_param_call(new_array, add_named_array, NULL, NULL, S_IWUSR);
7439 EXPORT_SYMBOL(register_md_personality);
7440 EXPORT_SYMBOL(unregister_md_personality);
7441 EXPORT_SYMBOL(md_error);
7442 EXPORT_SYMBOL(md_done_sync);
7443 EXPORT_SYMBOL(md_write_start);
7444 EXPORT_SYMBOL(md_write_end);
7445 EXPORT_SYMBOL(md_register_thread);
7446 EXPORT_SYMBOL(md_unregister_thread);
7447 EXPORT_SYMBOL(md_wakeup_thread);
7448 EXPORT_SYMBOL(md_check_recovery);
7449 MODULE_LICENSE("GPL");
7450 MODULE_DESCRIPTION("MD RAID framework");
7452 MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR);