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(mddev_t *mddev)
380 INIT_WORK(&mddev->flush_work, md_submit_flush_data);
381 atomic_set(&mddev->flush_pending, 1);
383 list_for_each_entry_rcu(rdev, &mddev->disks, same_set)
384 if (rdev->raid_disk >= 0 &&
385 !test_bit(Faulty, &rdev->flags)) {
386 /* Take two references, one is dropped
387 * when request finishes, one after
388 * we reclaim rcu_read_lock
391 atomic_inc(&rdev->nr_pending);
392 atomic_inc(&rdev->nr_pending);
394 bi = bio_alloc_mddev(GFP_KERNEL, 0, mddev);
395 bi->bi_end_io = md_end_flush;
396 bi->bi_private = rdev;
397 bi->bi_bdev = rdev->bdev;
398 atomic_inc(&mddev->flush_pending);
399 submit_bio(WRITE_FLUSH, bi);
401 rdev_dec_pending(rdev, mddev);
404 if (atomic_dec_and_test(&mddev->flush_pending))
405 queue_work(md_wq, &mddev->flush_work);
408 static void md_submit_flush_data(struct work_struct *ws)
410 mddev_t *mddev = container_of(ws, mddev_t, flush_work);
411 struct bio *bio = mddev->flush_bio;
413 if (bio->bi_size == 0)
414 /* an empty barrier - all done */
417 bio->bi_rw &= ~REQ_FLUSH;
418 if (mddev->pers->make_request(mddev, bio))
419 generic_make_request(bio);
422 mddev->flush_bio = NULL;
423 wake_up(&mddev->sb_wait);
426 void md_flush_request(mddev_t *mddev, struct bio *bio)
428 spin_lock_irq(&mddev->write_lock);
429 wait_event_lock_irq(mddev->sb_wait,
431 mddev->write_lock, /*nothing*/);
432 mddev->flush_bio = bio;
433 spin_unlock_irq(&mddev->write_lock);
435 submit_flushes(mddev);
437 EXPORT_SYMBOL(md_flush_request);
439 /* Support for plugging.
440 * This mirrors the plugging support in request_queue, but does not
441 * require having a whole queue
443 static void plugger_work(struct work_struct *work)
445 struct plug_handle *plug =
446 container_of(work, struct plug_handle, unplug_work);
447 plug->unplug_fn(plug);
449 static void plugger_timeout(unsigned long data)
451 struct plug_handle *plug = (void *)data;
452 kblockd_schedule_work(NULL, &plug->unplug_work);
454 void plugger_init(struct plug_handle *plug,
455 void (*unplug_fn)(struct plug_handle *))
457 plug->unplug_flag = 0;
458 plug->unplug_fn = unplug_fn;
459 init_timer(&plug->unplug_timer);
460 plug->unplug_timer.function = plugger_timeout;
461 plug->unplug_timer.data = (unsigned long)plug;
462 INIT_WORK(&plug->unplug_work, plugger_work);
464 EXPORT_SYMBOL_GPL(plugger_init);
466 void plugger_set_plug(struct plug_handle *plug)
468 if (!test_and_set_bit(PLUGGED_FLAG, &plug->unplug_flag))
469 mod_timer(&plug->unplug_timer, jiffies + msecs_to_jiffies(3)+1);
471 EXPORT_SYMBOL_GPL(plugger_set_plug);
473 int plugger_remove_plug(struct plug_handle *plug)
475 if (test_and_clear_bit(PLUGGED_FLAG, &plug->unplug_flag)) {
476 del_timer(&plug->unplug_timer);
481 EXPORT_SYMBOL_GPL(plugger_remove_plug);
484 static inline mddev_t *mddev_get(mddev_t *mddev)
486 atomic_inc(&mddev->active);
490 static void mddev_delayed_delete(struct work_struct *ws);
492 static void mddev_put(mddev_t *mddev)
494 struct bio_set *bs = NULL;
496 if (!atomic_dec_and_lock(&mddev->active, &all_mddevs_lock))
498 if (!mddev->raid_disks && list_empty(&mddev->disks) &&
499 mddev->ctime == 0 && !mddev->hold_active) {
500 /* Array is not configured at all, and not held active,
502 list_del(&mddev->all_mddevs);
504 mddev->bio_set = NULL;
505 if (mddev->gendisk) {
506 /* We did a probe so need to clean up. Call
507 * queue_work inside the spinlock so that
508 * flush_workqueue() after mddev_find will
509 * succeed in waiting for the work to be done.
511 INIT_WORK(&mddev->del_work, mddev_delayed_delete);
512 queue_work(md_misc_wq, &mddev->del_work);
516 spin_unlock(&all_mddevs_lock);
521 void mddev_init(mddev_t *mddev)
523 mutex_init(&mddev->open_mutex);
524 mutex_init(&mddev->reconfig_mutex);
525 mutex_init(&mddev->bitmap_info.mutex);
526 INIT_LIST_HEAD(&mddev->disks);
527 INIT_LIST_HEAD(&mddev->all_mddevs);
528 init_timer(&mddev->safemode_timer);
529 atomic_set(&mddev->active, 1);
530 atomic_set(&mddev->openers, 0);
531 atomic_set(&mddev->active_io, 0);
532 spin_lock_init(&mddev->write_lock);
533 atomic_set(&mddev->flush_pending, 0);
534 init_waitqueue_head(&mddev->sb_wait);
535 init_waitqueue_head(&mddev->recovery_wait);
536 mddev->reshape_position = MaxSector;
537 mddev->resync_min = 0;
538 mddev->resync_max = MaxSector;
539 mddev->level = LEVEL_NONE;
541 EXPORT_SYMBOL_GPL(mddev_init);
543 static mddev_t * mddev_find(dev_t unit)
545 mddev_t *mddev, *new = NULL;
548 spin_lock(&all_mddevs_lock);
551 list_for_each_entry(mddev, &all_mddevs, all_mddevs)
552 if (mddev->unit == unit) {
554 spin_unlock(&all_mddevs_lock);
560 list_add(&new->all_mddevs, &all_mddevs);
561 spin_unlock(&all_mddevs_lock);
562 new->hold_active = UNTIL_IOCTL;
566 /* find an unused unit number */
567 static int next_minor = 512;
568 int start = next_minor;
572 dev = MKDEV(MD_MAJOR, next_minor);
574 if (next_minor > MINORMASK)
576 if (next_minor == start) {
577 /* Oh dear, all in use. */
578 spin_unlock(&all_mddevs_lock);
584 list_for_each_entry(mddev, &all_mddevs, all_mddevs)
585 if (mddev->unit == dev) {
591 new->md_minor = MINOR(dev);
592 new->hold_active = UNTIL_STOP;
593 list_add(&new->all_mddevs, &all_mddevs);
594 spin_unlock(&all_mddevs_lock);
597 spin_unlock(&all_mddevs_lock);
599 new = kzalloc(sizeof(*new), GFP_KERNEL);
604 if (MAJOR(unit) == MD_MAJOR)
605 new->md_minor = MINOR(unit);
607 new->md_minor = MINOR(unit) >> MdpMinorShift;
614 static inline int mddev_lock(mddev_t * mddev)
616 return mutex_lock_interruptible(&mddev->reconfig_mutex);
619 static inline int mddev_is_locked(mddev_t *mddev)
621 return mutex_is_locked(&mddev->reconfig_mutex);
624 static inline int mddev_trylock(mddev_t * mddev)
626 return mutex_trylock(&mddev->reconfig_mutex);
629 static struct attribute_group md_redundancy_group;
631 static void mddev_unlock(mddev_t * mddev)
633 if (mddev->to_remove) {
634 /* These cannot be removed under reconfig_mutex as
635 * an access to the files will try to take reconfig_mutex
636 * while holding the file unremovable, which leads to
638 * So hold set sysfs_active while the remove in happeing,
639 * and anything else which might set ->to_remove or my
640 * otherwise change the sysfs namespace will fail with
641 * -EBUSY if sysfs_active is still set.
642 * We set sysfs_active under reconfig_mutex and elsewhere
643 * test it under the same mutex to ensure its correct value
646 struct attribute_group *to_remove = mddev->to_remove;
647 mddev->to_remove = NULL;
648 mddev->sysfs_active = 1;
649 mutex_unlock(&mddev->reconfig_mutex);
651 if (mddev->kobj.sd) {
652 if (to_remove != &md_redundancy_group)
653 sysfs_remove_group(&mddev->kobj, to_remove);
654 if (mddev->pers == NULL ||
655 mddev->pers->sync_request == NULL) {
656 sysfs_remove_group(&mddev->kobj, &md_redundancy_group);
657 if (mddev->sysfs_action)
658 sysfs_put(mddev->sysfs_action);
659 mddev->sysfs_action = NULL;
662 mddev->sysfs_active = 0;
664 mutex_unlock(&mddev->reconfig_mutex);
666 md_wakeup_thread(mddev->thread);
669 static mdk_rdev_t * find_rdev_nr(mddev_t *mddev, int nr)
673 list_for_each_entry(rdev, &mddev->disks, same_set)
674 if (rdev->desc_nr == nr)
680 static mdk_rdev_t * find_rdev(mddev_t * mddev, dev_t dev)
684 list_for_each_entry(rdev, &mddev->disks, same_set)
685 if (rdev->bdev->bd_dev == dev)
691 static struct mdk_personality *find_pers(int level, char *clevel)
693 struct mdk_personality *pers;
694 list_for_each_entry(pers, &pers_list, list) {
695 if (level != LEVEL_NONE && pers->level == level)
697 if (strcmp(pers->name, clevel)==0)
703 /* return the offset of the super block in 512byte sectors */
704 static inline sector_t calc_dev_sboffset(struct block_device *bdev)
706 sector_t num_sectors = i_size_read(bdev->bd_inode) / 512;
707 return MD_NEW_SIZE_SECTORS(num_sectors);
710 static int alloc_disk_sb(mdk_rdev_t * rdev)
715 rdev->sb_page = alloc_page(GFP_KERNEL);
716 if (!rdev->sb_page) {
717 printk(KERN_ALERT "md: out of memory.\n");
724 static void free_disk_sb(mdk_rdev_t * rdev)
727 put_page(rdev->sb_page);
729 rdev->sb_page = NULL;
736 static void super_written(struct bio *bio, int error)
738 mdk_rdev_t *rdev = bio->bi_private;
739 mddev_t *mddev = rdev->mddev;
741 if (error || !test_bit(BIO_UPTODATE, &bio->bi_flags)) {
742 printk("md: super_written gets error=%d, uptodate=%d\n",
743 error, test_bit(BIO_UPTODATE, &bio->bi_flags));
744 WARN_ON(test_bit(BIO_UPTODATE, &bio->bi_flags));
745 md_error(mddev, rdev);
748 if (atomic_dec_and_test(&mddev->pending_writes))
749 wake_up(&mddev->sb_wait);
753 void md_super_write(mddev_t *mddev, mdk_rdev_t *rdev,
754 sector_t sector, int size, struct page *page)
756 /* write first size bytes of page to sector of rdev
757 * Increment mddev->pending_writes before returning
758 * and decrement it on completion, waking up sb_wait
759 * if zero is reached.
760 * If an error occurred, call md_error
762 struct bio *bio = bio_alloc_mddev(GFP_NOIO, 1, mddev);
764 bio->bi_bdev = rdev->bdev;
765 bio->bi_sector = sector;
766 bio_add_page(bio, page, size, 0);
767 bio->bi_private = rdev;
768 bio->bi_end_io = super_written;
770 atomic_inc(&mddev->pending_writes);
771 submit_bio(REQ_WRITE | REQ_SYNC | REQ_UNPLUG | REQ_FLUSH | REQ_FUA,
775 void md_super_wait(mddev_t *mddev)
777 /* wait for all superblock writes that were scheduled to complete */
780 prepare_to_wait(&mddev->sb_wait, &wq, TASK_UNINTERRUPTIBLE);
781 if (atomic_read(&mddev->pending_writes)==0)
785 finish_wait(&mddev->sb_wait, &wq);
788 static void bi_complete(struct bio *bio, int error)
790 complete((struct completion*)bio->bi_private);
793 int sync_page_io(mdk_rdev_t *rdev, sector_t sector, int size,
794 struct page *page, int rw)
796 struct bio *bio = bio_alloc_mddev(GFP_NOIO, 1, rdev->mddev);
797 struct completion event;
800 rw |= REQ_SYNC | REQ_UNPLUG;
802 bio->bi_bdev = rdev->bdev;
803 bio->bi_sector = sector;
804 bio_add_page(bio, page, size, 0);
805 init_completion(&event);
806 bio->bi_private = &event;
807 bio->bi_end_io = bi_complete;
809 wait_for_completion(&event);
811 ret = test_bit(BIO_UPTODATE, &bio->bi_flags);
815 EXPORT_SYMBOL_GPL(sync_page_io);
817 static int read_disk_sb(mdk_rdev_t * rdev, int size)
819 char b[BDEVNAME_SIZE];
820 if (!rdev->sb_page) {
828 if (!sync_page_io(rdev, rdev->sb_start, size, rdev->sb_page, READ))
834 printk(KERN_WARNING "md: disabled device %s, could not read superblock.\n",
835 bdevname(rdev->bdev,b));
839 static int uuid_equal(mdp_super_t *sb1, mdp_super_t *sb2)
841 return sb1->set_uuid0 == sb2->set_uuid0 &&
842 sb1->set_uuid1 == sb2->set_uuid1 &&
843 sb1->set_uuid2 == sb2->set_uuid2 &&
844 sb1->set_uuid3 == sb2->set_uuid3;
847 static int sb_equal(mdp_super_t *sb1, mdp_super_t *sb2)
850 mdp_super_t *tmp1, *tmp2;
852 tmp1 = kmalloc(sizeof(*tmp1),GFP_KERNEL);
853 tmp2 = kmalloc(sizeof(*tmp2),GFP_KERNEL);
855 if (!tmp1 || !tmp2) {
857 printk(KERN_INFO "md.c sb_equal(): failed to allocate memory!\n");
865 * nr_disks is not constant
870 ret = (memcmp(tmp1, tmp2, MD_SB_GENERIC_CONSTANT_WORDS * 4) == 0);
878 static u32 md_csum_fold(u32 csum)
880 csum = (csum & 0xffff) + (csum >> 16);
881 return (csum & 0xffff) + (csum >> 16);
884 static unsigned int calc_sb_csum(mdp_super_t * sb)
887 u32 *sb32 = (u32*)sb;
889 unsigned int disk_csum, csum;
891 disk_csum = sb->sb_csum;
894 for (i = 0; i < MD_SB_BYTES/4 ; i++)
896 csum = (newcsum & 0xffffffff) + (newcsum>>32);
900 /* This used to use csum_partial, which was wrong for several
901 * reasons including that different results are returned on
902 * different architectures. It isn't critical that we get exactly
903 * the same return value as before (we always csum_fold before
904 * testing, and that removes any differences). However as we
905 * know that csum_partial always returned a 16bit value on
906 * alphas, do a fold to maximise conformity to previous behaviour.
908 sb->sb_csum = md_csum_fold(disk_csum);
910 sb->sb_csum = disk_csum;
917 * Handle superblock details.
918 * We want to be able to handle multiple superblock formats
919 * so we have a common interface to them all, and an array of
920 * different handlers.
921 * We rely on user-space to write the initial superblock, and support
922 * reading and updating of superblocks.
923 * Interface methods are:
924 * int load_super(mdk_rdev_t *dev, mdk_rdev_t *refdev, int minor_version)
925 * loads and validates a superblock on dev.
926 * if refdev != NULL, compare superblocks on both devices
928 * 0 - dev has a superblock that is compatible with refdev
929 * 1 - dev has a superblock that is compatible and newer than refdev
930 * so dev should be used as the refdev in future
931 * -EINVAL superblock incompatible or invalid
932 * -othererror e.g. -EIO
934 * int validate_super(mddev_t *mddev, mdk_rdev_t *dev)
935 * Verify that dev is acceptable into mddev.
936 * The first time, mddev->raid_disks will be 0, and data from
937 * dev should be merged in. Subsequent calls check that dev
938 * is new enough. Return 0 or -EINVAL
940 * void sync_super(mddev_t *mddev, mdk_rdev_t *dev)
941 * Update the superblock for rdev with data in mddev
942 * This does not write to disc.
948 struct module *owner;
949 int (*load_super)(mdk_rdev_t *rdev, mdk_rdev_t *refdev,
951 int (*validate_super)(mddev_t *mddev, mdk_rdev_t *rdev);
952 void (*sync_super)(mddev_t *mddev, mdk_rdev_t *rdev);
953 unsigned long long (*rdev_size_change)(mdk_rdev_t *rdev,
954 sector_t num_sectors);
958 * Check that the given mddev has no bitmap.
960 * This function is called from the run method of all personalities that do not
961 * support bitmaps. It prints an error message and returns non-zero if mddev
962 * has a bitmap. Otherwise, it returns 0.
965 int md_check_no_bitmap(mddev_t *mddev)
967 if (!mddev->bitmap_info.file && !mddev->bitmap_info.offset)
969 printk(KERN_ERR "%s: bitmaps are not supported for %s\n",
970 mdname(mddev), mddev->pers->name);
973 EXPORT_SYMBOL(md_check_no_bitmap);
976 * load_super for 0.90.0
978 static int super_90_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
980 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
985 * Calculate the position of the superblock (512byte sectors),
986 * it's at the end of the disk.
988 * It also happens to be a multiple of 4Kb.
990 rdev->sb_start = calc_dev_sboffset(rdev->bdev);
992 ret = read_disk_sb(rdev, MD_SB_BYTES);
997 bdevname(rdev->bdev, b);
998 sb = (mdp_super_t*)page_address(rdev->sb_page);
1000 if (sb->md_magic != MD_SB_MAGIC) {
1001 printk(KERN_ERR "md: invalid raid superblock magic on %s\n",
1006 if (sb->major_version != 0 ||
1007 sb->minor_version < 90 ||
1008 sb->minor_version > 91) {
1009 printk(KERN_WARNING "Bad version number %d.%d on %s\n",
1010 sb->major_version, sb->minor_version,
1015 if (sb->raid_disks <= 0)
1018 if (md_csum_fold(calc_sb_csum(sb)) != md_csum_fold(sb->sb_csum)) {
1019 printk(KERN_WARNING "md: invalid superblock checksum on %s\n",
1024 rdev->preferred_minor = sb->md_minor;
1025 rdev->data_offset = 0;
1026 rdev->sb_size = MD_SB_BYTES;
1028 if (sb->level == LEVEL_MULTIPATH)
1031 rdev->desc_nr = sb->this_disk.number;
1037 mdp_super_t *refsb = (mdp_super_t*)page_address(refdev->sb_page);
1038 if (!uuid_equal(refsb, sb)) {
1039 printk(KERN_WARNING "md: %s has different UUID to %s\n",
1040 b, bdevname(refdev->bdev,b2));
1043 if (!sb_equal(refsb, sb)) {
1044 printk(KERN_WARNING "md: %s has same UUID"
1045 " but different superblock to %s\n",
1046 b, bdevname(refdev->bdev, b2));
1050 ev2 = md_event(refsb);
1056 rdev->sectors = rdev->sb_start;
1058 if (rdev->sectors < sb->size * 2 && sb->level > 1)
1059 /* "this cannot possibly happen" ... */
1067 * validate_super for 0.90.0
1069 static int super_90_validate(mddev_t *mddev, mdk_rdev_t *rdev)
1072 mdp_super_t *sb = (mdp_super_t *)page_address(rdev->sb_page);
1073 __u64 ev1 = md_event(sb);
1075 rdev->raid_disk = -1;
1076 clear_bit(Faulty, &rdev->flags);
1077 clear_bit(In_sync, &rdev->flags);
1078 clear_bit(WriteMostly, &rdev->flags);
1080 if (mddev->raid_disks == 0) {
1081 mddev->major_version = 0;
1082 mddev->minor_version = sb->minor_version;
1083 mddev->patch_version = sb->patch_version;
1084 mddev->external = 0;
1085 mddev->chunk_sectors = sb->chunk_size >> 9;
1086 mddev->ctime = sb->ctime;
1087 mddev->utime = sb->utime;
1088 mddev->level = sb->level;
1089 mddev->clevel[0] = 0;
1090 mddev->layout = sb->layout;
1091 mddev->raid_disks = sb->raid_disks;
1092 mddev->dev_sectors = sb->size * 2;
1093 mddev->events = ev1;
1094 mddev->bitmap_info.offset = 0;
1095 mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9;
1097 if (mddev->minor_version >= 91) {
1098 mddev->reshape_position = sb->reshape_position;
1099 mddev->delta_disks = sb->delta_disks;
1100 mddev->new_level = sb->new_level;
1101 mddev->new_layout = sb->new_layout;
1102 mddev->new_chunk_sectors = sb->new_chunk >> 9;
1104 mddev->reshape_position = MaxSector;
1105 mddev->delta_disks = 0;
1106 mddev->new_level = mddev->level;
1107 mddev->new_layout = mddev->layout;
1108 mddev->new_chunk_sectors = mddev->chunk_sectors;
1111 if (sb->state & (1<<MD_SB_CLEAN))
1112 mddev->recovery_cp = MaxSector;
1114 if (sb->events_hi == sb->cp_events_hi &&
1115 sb->events_lo == sb->cp_events_lo) {
1116 mddev->recovery_cp = sb->recovery_cp;
1118 mddev->recovery_cp = 0;
1121 memcpy(mddev->uuid+0, &sb->set_uuid0, 4);
1122 memcpy(mddev->uuid+4, &sb->set_uuid1, 4);
1123 memcpy(mddev->uuid+8, &sb->set_uuid2, 4);
1124 memcpy(mddev->uuid+12,&sb->set_uuid3, 4);
1126 mddev->max_disks = MD_SB_DISKS;
1128 if (sb->state & (1<<MD_SB_BITMAP_PRESENT) &&
1129 mddev->bitmap_info.file == NULL)
1130 mddev->bitmap_info.offset =
1131 mddev->bitmap_info.default_offset;
1133 } else if (mddev->pers == NULL) {
1134 /* Insist on good event counter while assembling, except
1135 * for spares (which don't need an event count) */
1137 if (sb->disks[rdev->desc_nr].state & (
1138 (1<<MD_DISK_SYNC) | (1 << MD_DISK_ACTIVE)))
1139 if (ev1 < mddev->events)
1141 } else if (mddev->bitmap) {
1142 /* if adding to array with a bitmap, then we can accept an
1143 * older device ... but not too old.
1145 if (ev1 < mddev->bitmap->events_cleared)
1148 if (ev1 < mddev->events)
1149 /* just a hot-add of a new device, leave raid_disk at -1 */
1153 if (mddev->level != LEVEL_MULTIPATH) {
1154 desc = sb->disks + rdev->desc_nr;
1156 if (desc->state & (1<<MD_DISK_FAULTY))
1157 set_bit(Faulty, &rdev->flags);
1158 else if (desc->state & (1<<MD_DISK_SYNC) /* &&
1159 desc->raid_disk < mddev->raid_disks */) {
1160 set_bit(In_sync, &rdev->flags);
1161 rdev->raid_disk = desc->raid_disk;
1162 } else if (desc->state & (1<<MD_DISK_ACTIVE)) {
1163 /* active but not in sync implies recovery up to
1164 * reshape position. We don't know exactly where
1165 * that is, so set to zero for now */
1166 if (mddev->minor_version >= 91) {
1167 rdev->recovery_offset = 0;
1168 rdev->raid_disk = desc->raid_disk;
1171 if (desc->state & (1<<MD_DISK_WRITEMOSTLY))
1172 set_bit(WriteMostly, &rdev->flags);
1173 } else /* MULTIPATH are always insync */
1174 set_bit(In_sync, &rdev->flags);
1179 * sync_super for 0.90.0
1181 static void super_90_sync(mddev_t *mddev, mdk_rdev_t *rdev)
1185 int next_spare = mddev->raid_disks;
1188 /* make rdev->sb match mddev data..
1191 * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
1192 * 3/ any empty disks < next_spare become removed
1194 * disks[0] gets initialised to REMOVED because
1195 * we cannot be sure from other fields if it has
1196 * been initialised or not.
1199 int active=0, working=0,failed=0,spare=0,nr_disks=0;
1201 rdev->sb_size = MD_SB_BYTES;
1203 sb = (mdp_super_t*)page_address(rdev->sb_page);
1205 memset(sb, 0, sizeof(*sb));
1207 sb->md_magic = MD_SB_MAGIC;
1208 sb->major_version = mddev->major_version;
1209 sb->patch_version = mddev->patch_version;
1210 sb->gvalid_words = 0; /* ignored */
1211 memcpy(&sb->set_uuid0, mddev->uuid+0, 4);
1212 memcpy(&sb->set_uuid1, mddev->uuid+4, 4);
1213 memcpy(&sb->set_uuid2, mddev->uuid+8, 4);
1214 memcpy(&sb->set_uuid3, mddev->uuid+12,4);
1216 sb->ctime = mddev->ctime;
1217 sb->level = mddev->level;
1218 sb->size = mddev->dev_sectors / 2;
1219 sb->raid_disks = mddev->raid_disks;
1220 sb->md_minor = mddev->md_minor;
1221 sb->not_persistent = 0;
1222 sb->utime = mddev->utime;
1224 sb->events_hi = (mddev->events>>32);
1225 sb->events_lo = (u32)mddev->events;
1227 if (mddev->reshape_position == MaxSector)
1228 sb->minor_version = 90;
1230 sb->minor_version = 91;
1231 sb->reshape_position = mddev->reshape_position;
1232 sb->new_level = mddev->new_level;
1233 sb->delta_disks = mddev->delta_disks;
1234 sb->new_layout = mddev->new_layout;
1235 sb->new_chunk = mddev->new_chunk_sectors << 9;
1237 mddev->minor_version = sb->minor_version;
1240 sb->recovery_cp = mddev->recovery_cp;
1241 sb->cp_events_hi = (mddev->events>>32);
1242 sb->cp_events_lo = (u32)mddev->events;
1243 if (mddev->recovery_cp == MaxSector)
1244 sb->state = (1<< MD_SB_CLEAN);
1246 sb->recovery_cp = 0;
1248 sb->layout = mddev->layout;
1249 sb->chunk_size = mddev->chunk_sectors << 9;
1251 if (mddev->bitmap && mddev->bitmap_info.file == NULL)
1252 sb->state |= (1<<MD_SB_BITMAP_PRESENT);
1254 sb->disks[0].state = (1<<MD_DISK_REMOVED);
1255 list_for_each_entry(rdev2, &mddev->disks, same_set) {
1258 int is_active = test_bit(In_sync, &rdev2->flags);
1260 if (rdev2->raid_disk >= 0 &&
1261 sb->minor_version >= 91)
1262 /* we have nowhere to store the recovery_offset,
1263 * but if it is not below the reshape_position,
1264 * we can piggy-back on that.
1267 if (rdev2->raid_disk < 0 ||
1268 test_bit(Faulty, &rdev2->flags))
1271 desc_nr = rdev2->raid_disk;
1273 desc_nr = next_spare++;
1274 rdev2->desc_nr = desc_nr;
1275 d = &sb->disks[rdev2->desc_nr];
1277 d->number = rdev2->desc_nr;
1278 d->major = MAJOR(rdev2->bdev->bd_dev);
1279 d->minor = MINOR(rdev2->bdev->bd_dev);
1281 d->raid_disk = rdev2->raid_disk;
1283 d->raid_disk = rdev2->desc_nr; /* compatibility */
1284 if (test_bit(Faulty, &rdev2->flags))
1285 d->state = (1<<MD_DISK_FAULTY);
1286 else if (is_active) {
1287 d->state = (1<<MD_DISK_ACTIVE);
1288 if (test_bit(In_sync, &rdev2->flags))
1289 d->state |= (1<<MD_DISK_SYNC);
1297 if (test_bit(WriteMostly, &rdev2->flags))
1298 d->state |= (1<<MD_DISK_WRITEMOSTLY);
1300 /* now set the "removed" and "faulty" bits on any missing devices */
1301 for (i=0 ; i < mddev->raid_disks ; i++) {
1302 mdp_disk_t *d = &sb->disks[i];
1303 if (d->state == 0 && d->number == 0) {
1306 d->state = (1<<MD_DISK_REMOVED);
1307 d->state |= (1<<MD_DISK_FAULTY);
1311 sb->nr_disks = nr_disks;
1312 sb->active_disks = active;
1313 sb->working_disks = working;
1314 sb->failed_disks = failed;
1315 sb->spare_disks = spare;
1317 sb->this_disk = sb->disks[rdev->desc_nr];
1318 sb->sb_csum = calc_sb_csum(sb);
1322 * rdev_size_change for 0.90.0
1324 static unsigned long long
1325 super_90_rdev_size_change(mdk_rdev_t *rdev, sector_t num_sectors)
1327 if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
1328 return 0; /* component must fit device */
1329 if (rdev->mddev->bitmap_info.offset)
1330 return 0; /* can't move bitmap */
1331 rdev->sb_start = calc_dev_sboffset(rdev->bdev);
1332 if (!num_sectors || num_sectors > rdev->sb_start)
1333 num_sectors = rdev->sb_start;
1334 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
1336 md_super_wait(rdev->mddev);
1342 * version 1 superblock
1345 static __le32 calc_sb_1_csum(struct mdp_superblock_1 * sb)
1349 unsigned long long newcsum;
1350 int size = 256 + le32_to_cpu(sb->max_dev)*2;
1351 __le32 *isuper = (__le32*)sb;
1354 disk_csum = sb->sb_csum;
1357 for (i=0; size>=4; size -= 4 )
1358 newcsum += le32_to_cpu(*isuper++);
1361 newcsum += le16_to_cpu(*(__le16*) isuper);
1363 csum = (newcsum & 0xffffffff) + (newcsum >> 32);
1364 sb->sb_csum = disk_csum;
1365 return cpu_to_le32(csum);
1368 static int super_1_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
1370 struct mdp_superblock_1 *sb;
1373 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
1377 * Calculate the position of the superblock in 512byte sectors.
1378 * It is always aligned to a 4K boundary and
1379 * depeding on minor_version, it can be:
1380 * 0: At least 8K, but less than 12K, from end of device
1381 * 1: At start of device
1382 * 2: 4K from start of device.
1384 switch(minor_version) {
1386 sb_start = i_size_read(rdev->bdev->bd_inode) >> 9;
1388 sb_start &= ~(sector_t)(4*2-1);
1399 rdev->sb_start = sb_start;
1401 /* superblock is rarely larger than 1K, but it can be larger,
1402 * and it is safe to read 4k, so we do that
1404 ret = read_disk_sb(rdev, 4096);
1405 if (ret) return ret;
1408 sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1410 if (sb->magic != cpu_to_le32(MD_SB_MAGIC) ||
1411 sb->major_version != cpu_to_le32(1) ||
1412 le32_to_cpu(sb->max_dev) > (4096-256)/2 ||
1413 le64_to_cpu(sb->super_offset) != rdev->sb_start ||
1414 (le32_to_cpu(sb->feature_map) & ~MD_FEATURE_ALL) != 0)
1417 if (calc_sb_1_csum(sb) != sb->sb_csum) {
1418 printk("md: invalid superblock checksum on %s\n",
1419 bdevname(rdev->bdev,b));
1422 if (le64_to_cpu(sb->data_size) < 10) {
1423 printk("md: data_size too small on %s\n",
1424 bdevname(rdev->bdev,b));
1428 rdev->preferred_minor = 0xffff;
1429 rdev->data_offset = le64_to_cpu(sb->data_offset);
1430 atomic_set(&rdev->corrected_errors, le32_to_cpu(sb->cnt_corrected_read));
1432 rdev->sb_size = le32_to_cpu(sb->max_dev) * 2 + 256;
1433 bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
1434 if (rdev->sb_size & bmask)
1435 rdev->sb_size = (rdev->sb_size | bmask) + 1;
1438 && rdev->data_offset < sb_start + (rdev->sb_size/512))
1441 if (sb->level == cpu_to_le32(LEVEL_MULTIPATH))
1444 rdev->desc_nr = le32_to_cpu(sb->dev_number);
1450 struct mdp_superblock_1 *refsb =
1451 (struct mdp_superblock_1*)page_address(refdev->sb_page);
1453 if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 ||
1454 sb->level != refsb->level ||
1455 sb->layout != refsb->layout ||
1456 sb->chunksize != refsb->chunksize) {
1457 printk(KERN_WARNING "md: %s has strangely different"
1458 " superblock to %s\n",
1459 bdevname(rdev->bdev,b),
1460 bdevname(refdev->bdev,b2));
1463 ev1 = le64_to_cpu(sb->events);
1464 ev2 = le64_to_cpu(refsb->events);
1472 rdev->sectors = (i_size_read(rdev->bdev->bd_inode) >> 9) -
1473 le64_to_cpu(sb->data_offset);
1475 rdev->sectors = rdev->sb_start;
1476 if (rdev->sectors < le64_to_cpu(sb->data_size))
1478 rdev->sectors = le64_to_cpu(sb->data_size);
1479 if (le64_to_cpu(sb->size) > rdev->sectors)
1484 static int super_1_validate(mddev_t *mddev, mdk_rdev_t *rdev)
1486 struct mdp_superblock_1 *sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1487 __u64 ev1 = le64_to_cpu(sb->events);
1489 rdev->raid_disk = -1;
1490 clear_bit(Faulty, &rdev->flags);
1491 clear_bit(In_sync, &rdev->flags);
1492 clear_bit(WriteMostly, &rdev->flags);
1494 if (mddev->raid_disks == 0) {
1495 mddev->major_version = 1;
1496 mddev->patch_version = 0;
1497 mddev->external = 0;
1498 mddev->chunk_sectors = le32_to_cpu(sb->chunksize);
1499 mddev->ctime = le64_to_cpu(sb->ctime) & ((1ULL << 32)-1);
1500 mddev->utime = le64_to_cpu(sb->utime) & ((1ULL << 32)-1);
1501 mddev->level = le32_to_cpu(sb->level);
1502 mddev->clevel[0] = 0;
1503 mddev->layout = le32_to_cpu(sb->layout);
1504 mddev->raid_disks = le32_to_cpu(sb->raid_disks);
1505 mddev->dev_sectors = le64_to_cpu(sb->size);
1506 mddev->events = ev1;
1507 mddev->bitmap_info.offset = 0;
1508 mddev->bitmap_info.default_offset = 1024 >> 9;
1510 mddev->recovery_cp = le64_to_cpu(sb->resync_offset);
1511 memcpy(mddev->uuid, sb->set_uuid, 16);
1513 mddev->max_disks = (4096-256)/2;
1515 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) &&
1516 mddev->bitmap_info.file == NULL )
1517 mddev->bitmap_info.offset =
1518 (__s32)le32_to_cpu(sb->bitmap_offset);
1520 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) {
1521 mddev->reshape_position = le64_to_cpu(sb->reshape_position);
1522 mddev->delta_disks = le32_to_cpu(sb->delta_disks);
1523 mddev->new_level = le32_to_cpu(sb->new_level);
1524 mddev->new_layout = le32_to_cpu(sb->new_layout);
1525 mddev->new_chunk_sectors = le32_to_cpu(sb->new_chunk);
1527 mddev->reshape_position = MaxSector;
1528 mddev->delta_disks = 0;
1529 mddev->new_level = mddev->level;
1530 mddev->new_layout = mddev->layout;
1531 mddev->new_chunk_sectors = mddev->chunk_sectors;
1534 } else if (mddev->pers == NULL) {
1535 /* Insist of good event counter while assembling, except for
1536 * spares (which don't need an event count) */
1538 if (rdev->desc_nr >= 0 &&
1539 rdev->desc_nr < le32_to_cpu(sb->max_dev) &&
1540 le16_to_cpu(sb->dev_roles[rdev->desc_nr]) < 0xfffe)
1541 if (ev1 < mddev->events)
1543 } else if (mddev->bitmap) {
1544 /* If adding to array with a bitmap, then we can accept an
1545 * older device, but not too old.
1547 if (ev1 < mddev->bitmap->events_cleared)
1550 if (ev1 < mddev->events)
1551 /* just a hot-add of a new device, leave raid_disk at -1 */
1554 if (mddev->level != LEVEL_MULTIPATH) {
1556 if (rdev->desc_nr < 0 ||
1557 rdev->desc_nr >= le32_to_cpu(sb->max_dev)) {
1561 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
1563 case 0xffff: /* spare */
1565 case 0xfffe: /* faulty */
1566 set_bit(Faulty, &rdev->flags);
1569 if ((le32_to_cpu(sb->feature_map) &
1570 MD_FEATURE_RECOVERY_OFFSET))
1571 rdev->recovery_offset = le64_to_cpu(sb->recovery_offset);
1573 set_bit(In_sync, &rdev->flags);
1574 rdev->raid_disk = role;
1577 if (sb->devflags & WriteMostly1)
1578 set_bit(WriteMostly, &rdev->flags);
1579 } else /* MULTIPATH are always insync */
1580 set_bit(In_sync, &rdev->flags);
1585 static void super_1_sync(mddev_t *mddev, mdk_rdev_t *rdev)
1587 struct mdp_superblock_1 *sb;
1590 /* make rdev->sb match mddev and rdev data. */
1592 sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1594 sb->feature_map = 0;
1596 sb->recovery_offset = cpu_to_le64(0);
1597 memset(sb->pad1, 0, sizeof(sb->pad1));
1598 memset(sb->pad2, 0, sizeof(sb->pad2));
1599 memset(sb->pad3, 0, sizeof(sb->pad3));
1601 sb->utime = cpu_to_le64((__u64)mddev->utime);
1602 sb->events = cpu_to_le64(mddev->events);
1604 sb->resync_offset = cpu_to_le64(mddev->recovery_cp);
1606 sb->resync_offset = cpu_to_le64(0);
1608 sb->cnt_corrected_read = cpu_to_le32(atomic_read(&rdev->corrected_errors));
1610 sb->raid_disks = cpu_to_le32(mddev->raid_disks);
1611 sb->size = cpu_to_le64(mddev->dev_sectors);
1612 sb->chunksize = cpu_to_le32(mddev->chunk_sectors);
1613 sb->level = cpu_to_le32(mddev->level);
1614 sb->layout = cpu_to_le32(mddev->layout);
1616 if (mddev->bitmap && mddev->bitmap_info.file == NULL) {
1617 sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_info.offset);
1618 sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET);
1621 if (rdev->raid_disk >= 0 &&
1622 !test_bit(In_sync, &rdev->flags)) {
1624 cpu_to_le32(MD_FEATURE_RECOVERY_OFFSET);
1625 sb->recovery_offset =
1626 cpu_to_le64(rdev->recovery_offset);
1629 if (mddev->reshape_position != MaxSector) {
1630 sb->feature_map |= cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE);
1631 sb->reshape_position = cpu_to_le64(mddev->reshape_position);
1632 sb->new_layout = cpu_to_le32(mddev->new_layout);
1633 sb->delta_disks = cpu_to_le32(mddev->delta_disks);
1634 sb->new_level = cpu_to_le32(mddev->new_level);
1635 sb->new_chunk = cpu_to_le32(mddev->new_chunk_sectors);
1639 list_for_each_entry(rdev2, &mddev->disks, same_set)
1640 if (rdev2->desc_nr+1 > max_dev)
1641 max_dev = rdev2->desc_nr+1;
1643 if (max_dev > le32_to_cpu(sb->max_dev)) {
1645 sb->max_dev = cpu_to_le32(max_dev);
1646 rdev->sb_size = max_dev * 2 + 256;
1647 bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
1648 if (rdev->sb_size & bmask)
1649 rdev->sb_size = (rdev->sb_size | bmask) + 1;
1651 max_dev = le32_to_cpu(sb->max_dev);
1653 for (i=0; i<max_dev;i++)
1654 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1656 list_for_each_entry(rdev2, &mddev->disks, same_set) {
1658 if (test_bit(Faulty, &rdev2->flags))
1659 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1660 else if (test_bit(In_sync, &rdev2->flags))
1661 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1662 else if (rdev2->raid_disk >= 0)
1663 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1665 sb->dev_roles[i] = cpu_to_le16(0xffff);
1668 sb->sb_csum = calc_sb_1_csum(sb);
1671 static unsigned long long
1672 super_1_rdev_size_change(mdk_rdev_t *rdev, sector_t num_sectors)
1674 struct mdp_superblock_1 *sb;
1675 sector_t max_sectors;
1676 if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
1677 return 0; /* component must fit device */
1678 if (rdev->sb_start < rdev->data_offset) {
1679 /* minor versions 1 and 2; superblock before data */
1680 max_sectors = i_size_read(rdev->bdev->bd_inode) >> 9;
1681 max_sectors -= rdev->data_offset;
1682 if (!num_sectors || num_sectors > max_sectors)
1683 num_sectors = max_sectors;
1684 } else if (rdev->mddev->bitmap_info.offset) {
1685 /* minor version 0 with bitmap we can't move */
1688 /* minor version 0; superblock after data */
1690 sb_start = (i_size_read(rdev->bdev->bd_inode) >> 9) - 8*2;
1691 sb_start &= ~(sector_t)(4*2 - 1);
1692 max_sectors = rdev->sectors + sb_start - rdev->sb_start;
1693 if (!num_sectors || num_sectors > max_sectors)
1694 num_sectors = max_sectors;
1695 rdev->sb_start = sb_start;
1697 sb = (struct mdp_superblock_1 *) page_address(rdev->sb_page);
1698 sb->data_size = cpu_to_le64(num_sectors);
1699 sb->super_offset = rdev->sb_start;
1700 sb->sb_csum = calc_sb_1_csum(sb);
1701 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
1703 md_super_wait(rdev->mddev);
1707 static struct super_type super_types[] = {
1710 .owner = THIS_MODULE,
1711 .load_super = super_90_load,
1712 .validate_super = super_90_validate,
1713 .sync_super = super_90_sync,
1714 .rdev_size_change = super_90_rdev_size_change,
1718 .owner = THIS_MODULE,
1719 .load_super = super_1_load,
1720 .validate_super = super_1_validate,
1721 .sync_super = super_1_sync,
1722 .rdev_size_change = super_1_rdev_size_change,
1726 static int match_mddev_units(mddev_t *mddev1, mddev_t *mddev2)
1728 mdk_rdev_t *rdev, *rdev2;
1731 rdev_for_each_rcu(rdev, mddev1)
1732 rdev_for_each_rcu(rdev2, mddev2)
1733 if (rdev->bdev->bd_contains ==
1734 rdev2->bdev->bd_contains) {
1742 static LIST_HEAD(pending_raid_disks);
1745 * Try to register data integrity profile for an mddev
1747 * This is called when an array is started and after a disk has been kicked
1748 * from the array. It only succeeds if all working and active component devices
1749 * are integrity capable with matching profiles.
1751 int md_integrity_register(mddev_t *mddev)
1753 mdk_rdev_t *rdev, *reference = NULL;
1755 if (list_empty(&mddev->disks))
1756 return 0; /* nothing to do */
1757 if (blk_get_integrity(mddev->gendisk))
1758 return 0; /* already registered */
1759 list_for_each_entry(rdev, &mddev->disks, same_set) {
1760 /* skip spares and non-functional disks */
1761 if (test_bit(Faulty, &rdev->flags))
1763 if (rdev->raid_disk < 0)
1766 * If at least one rdev is not integrity capable, we can not
1767 * enable data integrity for the md device.
1769 if (!bdev_get_integrity(rdev->bdev))
1772 /* Use the first rdev as the reference */
1776 /* does this rdev's profile match the reference profile? */
1777 if (blk_integrity_compare(reference->bdev->bd_disk,
1778 rdev->bdev->bd_disk) < 0)
1782 * All component devices are integrity capable and have matching
1783 * profiles, register the common profile for the md device.
1785 if (blk_integrity_register(mddev->gendisk,
1786 bdev_get_integrity(reference->bdev)) != 0) {
1787 printk(KERN_ERR "md: failed to register integrity for %s\n",
1791 printk(KERN_NOTICE "md: data integrity on %s enabled\n",
1795 EXPORT_SYMBOL(md_integrity_register);
1797 /* Disable data integrity if non-capable/non-matching disk is being added */
1798 void md_integrity_add_rdev(mdk_rdev_t *rdev, mddev_t *mddev)
1800 struct blk_integrity *bi_rdev = bdev_get_integrity(rdev->bdev);
1801 struct blk_integrity *bi_mddev = blk_get_integrity(mddev->gendisk);
1803 if (!bi_mddev) /* nothing to do */
1805 if (rdev->raid_disk < 0) /* skip spares */
1807 if (bi_rdev && blk_integrity_compare(mddev->gendisk,
1808 rdev->bdev->bd_disk) >= 0)
1810 printk(KERN_NOTICE "disabling data integrity on %s\n", mdname(mddev));
1811 blk_integrity_unregister(mddev->gendisk);
1813 EXPORT_SYMBOL(md_integrity_add_rdev);
1815 static int bind_rdev_to_array(mdk_rdev_t * rdev, mddev_t * mddev)
1817 char b[BDEVNAME_SIZE];
1827 /* prevent duplicates */
1828 if (find_rdev(mddev, rdev->bdev->bd_dev))
1831 /* make sure rdev->sectors exceeds mddev->dev_sectors */
1832 if (rdev->sectors && (mddev->dev_sectors == 0 ||
1833 rdev->sectors < mddev->dev_sectors)) {
1835 /* Cannot change size, so fail
1836 * If mddev->level <= 0, then we don't care
1837 * about aligning sizes (e.g. linear)
1839 if (mddev->level > 0)
1842 mddev->dev_sectors = rdev->sectors;
1845 /* Verify rdev->desc_nr is unique.
1846 * If it is -1, assign a free number, else
1847 * check number is not in use
1849 if (rdev->desc_nr < 0) {
1851 if (mddev->pers) choice = mddev->raid_disks;
1852 while (find_rdev_nr(mddev, choice))
1854 rdev->desc_nr = choice;
1856 if (find_rdev_nr(mddev, rdev->desc_nr))
1859 if (mddev->max_disks && rdev->desc_nr >= mddev->max_disks) {
1860 printk(KERN_WARNING "md: %s: array is limited to %d devices\n",
1861 mdname(mddev), mddev->max_disks);
1864 bdevname(rdev->bdev,b);
1865 while ( (s=strchr(b, '/')) != NULL)
1868 rdev->mddev = mddev;
1869 printk(KERN_INFO "md: bind<%s>\n", b);
1871 if ((err = kobject_add(&rdev->kobj, &mddev->kobj, "dev-%s", b)))
1874 ko = &part_to_dev(rdev->bdev->bd_part)->kobj;
1875 if (sysfs_create_link(&rdev->kobj, ko, "block"))
1876 /* failure here is OK */;
1877 rdev->sysfs_state = sysfs_get_dirent_safe(rdev->kobj.sd, "state");
1879 list_add_rcu(&rdev->same_set, &mddev->disks);
1880 bd_claim_by_disk(rdev->bdev, rdev->bdev->bd_holder, mddev->gendisk);
1882 /* May as well allow recovery to be retried once */
1883 mddev->recovery_disabled = 0;
1888 printk(KERN_WARNING "md: failed to register dev-%s for %s\n",
1893 static void md_delayed_delete(struct work_struct *ws)
1895 mdk_rdev_t *rdev = container_of(ws, mdk_rdev_t, del_work);
1896 kobject_del(&rdev->kobj);
1897 kobject_put(&rdev->kobj);
1900 static void unbind_rdev_from_array(mdk_rdev_t * rdev)
1902 char b[BDEVNAME_SIZE];
1907 bd_release_from_disk(rdev->bdev, rdev->mddev->gendisk);
1908 list_del_rcu(&rdev->same_set);
1909 printk(KERN_INFO "md: unbind<%s>\n", bdevname(rdev->bdev,b));
1911 sysfs_remove_link(&rdev->kobj, "block");
1912 sysfs_put(rdev->sysfs_state);
1913 rdev->sysfs_state = NULL;
1914 /* We need to delay this, otherwise we can deadlock when
1915 * writing to 'remove' to "dev/state". We also need
1916 * to delay it due to rcu usage.
1919 INIT_WORK(&rdev->del_work, md_delayed_delete);
1920 kobject_get(&rdev->kobj);
1921 queue_work(md_misc_wq, &rdev->del_work);
1925 * prevent the device from being mounted, repartitioned or
1926 * otherwise reused by a RAID array (or any other kernel
1927 * subsystem), by bd_claiming the device.
1929 static int lock_rdev(mdk_rdev_t *rdev, dev_t dev, int shared)
1932 struct block_device *bdev;
1933 char b[BDEVNAME_SIZE];
1935 bdev = open_by_devnum(dev, FMODE_READ|FMODE_WRITE);
1937 printk(KERN_ERR "md: could not open %s.\n",
1938 __bdevname(dev, b));
1939 return PTR_ERR(bdev);
1941 err = bd_claim(bdev, shared ? (mdk_rdev_t *)lock_rdev : rdev);
1943 printk(KERN_ERR "md: could not bd_claim %s.\n",
1945 blkdev_put(bdev, FMODE_READ|FMODE_WRITE);
1949 set_bit(AllReserved, &rdev->flags);
1954 static void unlock_rdev(mdk_rdev_t *rdev)
1956 struct block_device *bdev = rdev->bdev;
1961 blkdev_put(bdev, FMODE_READ|FMODE_WRITE);
1964 void md_autodetect_dev(dev_t dev);
1966 static void export_rdev(mdk_rdev_t * rdev)
1968 char b[BDEVNAME_SIZE];
1969 printk(KERN_INFO "md: export_rdev(%s)\n",
1970 bdevname(rdev->bdev,b));
1975 if (test_bit(AutoDetected, &rdev->flags))
1976 md_autodetect_dev(rdev->bdev->bd_dev);
1979 kobject_put(&rdev->kobj);
1982 static void kick_rdev_from_array(mdk_rdev_t * rdev)
1984 unbind_rdev_from_array(rdev);
1988 static void export_array(mddev_t *mddev)
1990 mdk_rdev_t *rdev, *tmp;
1992 rdev_for_each(rdev, tmp, mddev) {
1997 kick_rdev_from_array(rdev);
1999 if (!list_empty(&mddev->disks))
2001 mddev->raid_disks = 0;
2002 mddev->major_version = 0;
2005 static void print_desc(mdp_disk_t *desc)
2007 printk(" DISK<N:%d,(%d,%d),R:%d,S:%d>\n", desc->number,
2008 desc->major,desc->minor,desc->raid_disk,desc->state);
2011 static void print_sb_90(mdp_super_t *sb)
2016 "md: SB: (V:%d.%d.%d) ID:<%08x.%08x.%08x.%08x> CT:%08x\n",
2017 sb->major_version, sb->minor_version, sb->patch_version,
2018 sb->set_uuid0, sb->set_uuid1, sb->set_uuid2, sb->set_uuid3,
2020 printk(KERN_INFO "md: L%d S%08d ND:%d RD:%d md%d LO:%d CS:%d\n",
2021 sb->level, sb->size, sb->nr_disks, sb->raid_disks,
2022 sb->md_minor, sb->layout, sb->chunk_size);
2023 printk(KERN_INFO "md: UT:%08x ST:%d AD:%d WD:%d"
2024 " FD:%d SD:%d CSUM:%08x E:%08lx\n",
2025 sb->utime, sb->state, sb->active_disks, sb->working_disks,
2026 sb->failed_disks, sb->spare_disks,
2027 sb->sb_csum, (unsigned long)sb->events_lo);
2030 for (i = 0; i < MD_SB_DISKS; i++) {
2033 desc = sb->disks + i;
2034 if (desc->number || desc->major || desc->minor ||
2035 desc->raid_disk || (desc->state && (desc->state != 4))) {
2036 printk(" D %2d: ", i);
2040 printk(KERN_INFO "md: THIS: ");
2041 print_desc(&sb->this_disk);
2044 static void print_sb_1(struct mdp_superblock_1 *sb)
2048 uuid = sb->set_uuid;
2050 "md: SB: (V:%u) (F:0x%08x) Array-ID:<%pU>\n"
2051 "md: Name: \"%s\" CT:%llu\n",
2052 le32_to_cpu(sb->major_version),
2053 le32_to_cpu(sb->feature_map),
2056 (unsigned long long)le64_to_cpu(sb->ctime)
2057 & MD_SUPERBLOCK_1_TIME_SEC_MASK);
2059 uuid = sb->device_uuid;
2061 "md: L%u SZ%llu RD:%u LO:%u CS:%u DO:%llu DS:%llu SO:%llu"
2063 "md: Dev:%08x UUID: %pU\n"
2064 "md: (F:0x%08x) UT:%llu Events:%llu ResyncOffset:%llu CSUM:0x%08x\n"
2065 "md: (MaxDev:%u) \n",
2066 le32_to_cpu(sb->level),
2067 (unsigned long long)le64_to_cpu(sb->size),
2068 le32_to_cpu(sb->raid_disks),
2069 le32_to_cpu(sb->layout),
2070 le32_to_cpu(sb->chunksize),
2071 (unsigned long long)le64_to_cpu(sb->data_offset),
2072 (unsigned long long)le64_to_cpu(sb->data_size),
2073 (unsigned long long)le64_to_cpu(sb->super_offset),
2074 (unsigned long long)le64_to_cpu(sb->recovery_offset),
2075 le32_to_cpu(sb->dev_number),
2078 (unsigned long long)le64_to_cpu(sb->utime) & MD_SUPERBLOCK_1_TIME_SEC_MASK,
2079 (unsigned long long)le64_to_cpu(sb->events),
2080 (unsigned long long)le64_to_cpu(sb->resync_offset),
2081 le32_to_cpu(sb->sb_csum),
2082 le32_to_cpu(sb->max_dev)
2086 static void print_rdev(mdk_rdev_t *rdev, int major_version)
2088 char b[BDEVNAME_SIZE];
2089 printk(KERN_INFO "md: rdev %s, Sect:%08llu F:%d S:%d DN:%u\n",
2090 bdevname(rdev->bdev, b), (unsigned long long)rdev->sectors,
2091 test_bit(Faulty, &rdev->flags), test_bit(In_sync, &rdev->flags),
2093 if (rdev->sb_loaded) {
2094 printk(KERN_INFO "md: rdev superblock (MJ:%d):\n", major_version);
2095 switch (major_version) {
2097 print_sb_90((mdp_super_t*)page_address(rdev->sb_page));
2100 print_sb_1((struct mdp_superblock_1 *)page_address(rdev->sb_page));
2104 printk(KERN_INFO "md: no rdev superblock!\n");
2107 static void md_print_devices(void)
2109 struct list_head *tmp;
2112 char b[BDEVNAME_SIZE];
2115 printk("md: **********************************\n");
2116 printk("md: * <COMPLETE RAID STATE PRINTOUT> *\n");
2117 printk("md: **********************************\n");
2118 for_each_mddev(mddev, tmp) {
2121 bitmap_print_sb(mddev->bitmap);
2123 printk("%s: ", mdname(mddev));
2124 list_for_each_entry(rdev, &mddev->disks, same_set)
2125 printk("<%s>", bdevname(rdev->bdev,b));
2128 list_for_each_entry(rdev, &mddev->disks, same_set)
2129 print_rdev(rdev, mddev->major_version);
2131 printk("md: **********************************\n");
2136 static void sync_sbs(mddev_t * mddev, int nospares)
2138 /* Update each superblock (in-memory image), but
2139 * if we are allowed to, skip spares which already
2140 * have the right event counter, or have one earlier
2141 * (which would mean they aren't being marked as dirty
2142 * with the rest of the array)
2145 list_for_each_entry(rdev, &mddev->disks, same_set) {
2146 if (rdev->sb_events == mddev->events ||
2148 rdev->raid_disk < 0 &&
2149 rdev->sb_events+1 == mddev->events)) {
2150 /* Don't update this superblock */
2151 rdev->sb_loaded = 2;
2153 super_types[mddev->major_version].
2154 sync_super(mddev, rdev);
2155 rdev->sb_loaded = 1;
2160 static void md_update_sb(mddev_t * mddev, int force_change)
2167 /* First make sure individual recovery_offsets are correct */
2168 list_for_each_entry(rdev, &mddev->disks, same_set) {
2169 if (rdev->raid_disk >= 0 &&
2170 mddev->delta_disks >= 0 &&
2171 !test_bit(In_sync, &rdev->flags) &&
2172 mddev->curr_resync_completed > rdev->recovery_offset)
2173 rdev->recovery_offset = mddev->curr_resync_completed;
2176 if (!mddev->persistent) {
2177 clear_bit(MD_CHANGE_CLEAN, &mddev->flags);
2178 clear_bit(MD_CHANGE_DEVS, &mddev->flags);
2179 if (!mddev->external)
2180 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
2181 wake_up(&mddev->sb_wait);
2185 spin_lock_irq(&mddev->write_lock);
2187 mddev->utime = get_seconds();
2189 if (test_and_clear_bit(MD_CHANGE_DEVS, &mddev->flags))
2191 if (test_and_clear_bit(MD_CHANGE_CLEAN, &mddev->flags))
2192 /* just a clean<-> dirty transition, possibly leave spares alone,
2193 * though if events isn't the right even/odd, we will have to do
2199 if (mddev->degraded)
2200 /* If the array is degraded, then skipping spares is both
2201 * dangerous and fairly pointless.
2202 * Dangerous because a device that was removed from the array
2203 * might have a event_count that still looks up-to-date,
2204 * so it can be re-added without a resync.
2205 * Pointless because if there are any spares to skip,
2206 * then a recovery will happen and soon that array won't
2207 * be degraded any more and the spare can go back to sleep then.
2211 sync_req = mddev->in_sync;
2213 /* If this is just a dirty<->clean transition, and the array is clean
2214 * and 'events' is odd, we can roll back to the previous clean state */
2216 && (mddev->in_sync && mddev->recovery_cp == MaxSector)
2217 && mddev->can_decrease_events
2218 && mddev->events != 1) {
2220 mddev->can_decrease_events = 0;
2222 /* otherwise we have to go forward and ... */
2224 mddev->can_decrease_events = nospares;
2227 if (!mddev->events) {
2229 * oops, this 64-bit counter should never wrap.
2230 * Either we are in around ~1 trillion A.C., assuming
2231 * 1 reboot per second, or we have a bug:
2236 sync_sbs(mddev, nospares);
2237 spin_unlock_irq(&mddev->write_lock);
2240 "md: updating %s RAID superblock on device (in sync %d)\n",
2241 mdname(mddev),mddev->in_sync);
2243 bitmap_update_sb(mddev->bitmap);
2244 list_for_each_entry(rdev, &mddev->disks, same_set) {
2245 char b[BDEVNAME_SIZE];
2246 dprintk(KERN_INFO "md: ");
2247 if (rdev->sb_loaded != 1)
2248 continue; /* no noise on spare devices */
2249 if (test_bit(Faulty, &rdev->flags))
2250 dprintk("(skipping faulty ");
2252 dprintk("%s ", bdevname(rdev->bdev,b));
2253 if (!test_bit(Faulty, &rdev->flags)) {
2254 md_super_write(mddev,rdev,
2255 rdev->sb_start, rdev->sb_size,
2257 dprintk(KERN_INFO "(write) %s's sb offset: %llu\n",
2258 bdevname(rdev->bdev,b),
2259 (unsigned long long)rdev->sb_start);
2260 rdev->sb_events = mddev->events;
2264 if (mddev->level == LEVEL_MULTIPATH)
2265 /* only need to write one superblock... */
2268 md_super_wait(mddev);
2269 /* if there was a failure, MD_CHANGE_DEVS was set, and we re-write super */
2271 spin_lock_irq(&mddev->write_lock);
2272 if (mddev->in_sync != sync_req ||
2273 test_bit(MD_CHANGE_DEVS, &mddev->flags)) {
2274 /* have to write it out again */
2275 spin_unlock_irq(&mddev->write_lock);
2278 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
2279 spin_unlock_irq(&mddev->write_lock);
2280 wake_up(&mddev->sb_wait);
2281 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
2282 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
2286 /* words written to sysfs files may, or may not, be \n terminated.
2287 * We want to accept with case. For this we use cmd_match.
2289 static int cmd_match(const char *cmd, const char *str)
2291 /* See if cmd, written into a sysfs file, matches
2292 * str. They must either be the same, or cmd can
2293 * have a trailing newline
2295 while (*cmd && *str && *cmd == *str) {
2306 struct rdev_sysfs_entry {
2307 struct attribute attr;
2308 ssize_t (*show)(mdk_rdev_t *, char *);
2309 ssize_t (*store)(mdk_rdev_t *, const char *, size_t);
2313 state_show(mdk_rdev_t *rdev, char *page)
2318 if (test_bit(Faulty, &rdev->flags)) {
2319 len+= sprintf(page+len, "%sfaulty",sep);
2322 if (test_bit(In_sync, &rdev->flags)) {
2323 len += sprintf(page+len, "%sin_sync",sep);
2326 if (test_bit(WriteMostly, &rdev->flags)) {
2327 len += sprintf(page+len, "%swrite_mostly",sep);
2330 if (test_bit(Blocked, &rdev->flags)) {
2331 len += sprintf(page+len, "%sblocked", sep);
2334 if (!test_bit(Faulty, &rdev->flags) &&
2335 !test_bit(In_sync, &rdev->flags)) {
2336 len += sprintf(page+len, "%sspare", sep);
2339 return len+sprintf(page+len, "\n");
2343 state_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2346 * faulty - simulates and error
2347 * remove - disconnects the device
2348 * writemostly - sets write_mostly
2349 * -writemostly - clears write_mostly
2350 * blocked - sets the Blocked flag
2351 * -blocked - clears the Blocked flag
2352 * insync - sets Insync providing device isn't active
2355 if (cmd_match(buf, "faulty") && rdev->mddev->pers) {
2356 md_error(rdev->mddev, rdev);
2358 } else if (cmd_match(buf, "remove")) {
2359 if (rdev->raid_disk >= 0)
2362 mddev_t *mddev = rdev->mddev;
2363 kick_rdev_from_array(rdev);
2365 md_update_sb(mddev, 1);
2366 md_new_event(mddev);
2369 } else if (cmd_match(buf, "writemostly")) {
2370 set_bit(WriteMostly, &rdev->flags);
2372 } else if (cmd_match(buf, "-writemostly")) {
2373 clear_bit(WriteMostly, &rdev->flags);
2375 } else if (cmd_match(buf, "blocked")) {
2376 set_bit(Blocked, &rdev->flags);
2378 } else if (cmd_match(buf, "-blocked")) {
2379 clear_bit(Blocked, &rdev->flags);
2380 wake_up(&rdev->blocked_wait);
2381 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2382 md_wakeup_thread(rdev->mddev->thread);
2385 } else if (cmd_match(buf, "insync") && rdev->raid_disk == -1) {
2386 set_bit(In_sync, &rdev->flags);
2390 sysfs_notify_dirent_safe(rdev->sysfs_state);
2391 return err ? err : len;
2393 static struct rdev_sysfs_entry rdev_state =
2394 __ATTR(state, S_IRUGO|S_IWUSR, state_show, state_store);
2397 errors_show(mdk_rdev_t *rdev, char *page)
2399 return sprintf(page, "%d\n", atomic_read(&rdev->corrected_errors));
2403 errors_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2406 unsigned long n = simple_strtoul(buf, &e, 10);
2407 if (*buf && (*e == 0 || *e == '\n')) {
2408 atomic_set(&rdev->corrected_errors, n);
2413 static struct rdev_sysfs_entry rdev_errors =
2414 __ATTR(errors, S_IRUGO|S_IWUSR, errors_show, errors_store);
2417 slot_show(mdk_rdev_t *rdev, char *page)
2419 if (rdev->raid_disk < 0)
2420 return sprintf(page, "none\n");
2422 return sprintf(page, "%d\n", rdev->raid_disk);
2426 slot_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2431 int slot = simple_strtoul(buf, &e, 10);
2432 if (strncmp(buf, "none", 4)==0)
2434 else if (e==buf || (*e && *e!= '\n'))
2436 if (rdev->mddev->pers && slot == -1) {
2437 /* Setting 'slot' on an active array requires also
2438 * updating the 'rd%d' link, and communicating
2439 * with the personality with ->hot_*_disk.
2440 * For now we only support removing
2441 * failed/spare devices. This normally happens automatically,
2442 * but not when the metadata is externally managed.
2444 if (rdev->raid_disk == -1)
2446 /* personality does all needed checks */
2447 if (rdev->mddev->pers->hot_add_disk == NULL)
2449 err = rdev->mddev->pers->
2450 hot_remove_disk(rdev->mddev, rdev->raid_disk);
2453 sprintf(nm, "rd%d", rdev->raid_disk);
2454 sysfs_remove_link(&rdev->mddev->kobj, nm);
2455 rdev->raid_disk = -1;
2456 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2457 md_wakeup_thread(rdev->mddev->thread);
2458 } else if (rdev->mddev->pers) {
2460 /* Activating a spare .. or possibly reactivating
2461 * if we ever get bitmaps working here.
2464 if (rdev->raid_disk != -1)
2467 if (rdev->mddev->pers->hot_add_disk == NULL)
2470 list_for_each_entry(rdev2, &rdev->mddev->disks, same_set)
2471 if (rdev2->raid_disk == slot)
2474 rdev->raid_disk = slot;
2475 if (test_bit(In_sync, &rdev->flags))
2476 rdev->saved_raid_disk = slot;
2478 rdev->saved_raid_disk = -1;
2479 err = rdev->mddev->pers->
2480 hot_add_disk(rdev->mddev, rdev);
2482 rdev->raid_disk = -1;
2485 sysfs_notify_dirent_safe(rdev->sysfs_state);
2486 sprintf(nm, "rd%d", rdev->raid_disk);
2487 if (sysfs_create_link(&rdev->mddev->kobj, &rdev->kobj, nm))
2488 /* failure here is OK */;
2489 /* don't wakeup anyone, leave that to userspace. */
2491 if (slot >= rdev->mddev->raid_disks)
2493 rdev->raid_disk = slot;
2494 /* assume it is working */
2495 clear_bit(Faulty, &rdev->flags);
2496 clear_bit(WriteMostly, &rdev->flags);
2497 set_bit(In_sync, &rdev->flags);
2498 sysfs_notify_dirent_safe(rdev->sysfs_state);
2504 static struct rdev_sysfs_entry rdev_slot =
2505 __ATTR(slot, S_IRUGO|S_IWUSR, slot_show, slot_store);
2508 offset_show(mdk_rdev_t *rdev, char *page)
2510 return sprintf(page, "%llu\n", (unsigned long long)rdev->data_offset);
2514 offset_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2517 unsigned long long offset = simple_strtoull(buf, &e, 10);
2518 if (e==buf || (*e && *e != '\n'))
2520 if (rdev->mddev->pers && rdev->raid_disk >= 0)
2522 if (rdev->sectors && rdev->mddev->external)
2523 /* Must set offset before size, so overlap checks
2526 rdev->data_offset = offset;
2530 static struct rdev_sysfs_entry rdev_offset =
2531 __ATTR(offset, S_IRUGO|S_IWUSR, offset_show, offset_store);
2534 rdev_size_show(mdk_rdev_t *rdev, char *page)
2536 return sprintf(page, "%llu\n", (unsigned long long)rdev->sectors / 2);
2539 static int overlaps(sector_t s1, sector_t l1, sector_t s2, sector_t l2)
2541 /* check if two start/length pairs overlap */
2549 static int strict_blocks_to_sectors(const char *buf, sector_t *sectors)
2551 unsigned long long blocks;
2554 if (strict_strtoull(buf, 10, &blocks) < 0)
2557 if (blocks & 1ULL << (8 * sizeof(blocks) - 1))
2558 return -EINVAL; /* sector conversion overflow */
2561 if (new != blocks * 2)
2562 return -EINVAL; /* unsigned long long to sector_t overflow */
2569 rdev_size_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2571 mddev_t *my_mddev = rdev->mddev;
2572 sector_t oldsectors = rdev->sectors;
2575 if (strict_blocks_to_sectors(buf, §ors) < 0)
2577 if (my_mddev->pers && rdev->raid_disk >= 0) {
2578 if (my_mddev->persistent) {
2579 sectors = super_types[my_mddev->major_version].
2580 rdev_size_change(rdev, sectors);
2583 } else if (!sectors)
2584 sectors = (i_size_read(rdev->bdev->bd_inode) >> 9) -
2587 if (sectors < my_mddev->dev_sectors)
2588 return -EINVAL; /* component must fit device */
2590 rdev->sectors = sectors;
2591 if (sectors > oldsectors && my_mddev->external) {
2592 /* need to check that all other rdevs with the same ->bdev
2593 * do not overlap. We need to unlock the mddev to avoid
2594 * a deadlock. We have already changed rdev->sectors, and if
2595 * we have to change it back, we will have the lock again.
2599 struct list_head *tmp;
2601 mddev_unlock(my_mddev);
2602 for_each_mddev(mddev, tmp) {
2606 list_for_each_entry(rdev2, &mddev->disks, same_set)
2607 if (test_bit(AllReserved, &rdev2->flags) ||
2608 (rdev->bdev == rdev2->bdev &&
2610 overlaps(rdev->data_offset, rdev->sectors,
2616 mddev_unlock(mddev);
2622 mddev_lock(my_mddev);
2624 /* Someone else could have slipped in a size
2625 * change here, but doing so is just silly.
2626 * We put oldsectors back because we *know* it is
2627 * safe, and trust userspace not to race with
2630 rdev->sectors = oldsectors;
2637 static struct rdev_sysfs_entry rdev_size =
2638 __ATTR(size, S_IRUGO|S_IWUSR, rdev_size_show, rdev_size_store);
2641 static ssize_t recovery_start_show(mdk_rdev_t *rdev, char *page)
2643 unsigned long long recovery_start = rdev->recovery_offset;
2645 if (test_bit(In_sync, &rdev->flags) ||
2646 recovery_start == MaxSector)
2647 return sprintf(page, "none\n");
2649 return sprintf(page, "%llu\n", recovery_start);
2652 static ssize_t recovery_start_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2654 unsigned long long recovery_start;
2656 if (cmd_match(buf, "none"))
2657 recovery_start = MaxSector;
2658 else if (strict_strtoull(buf, 10, &recovery_start))
2661 if (rdev->mddev->pers &&
2662 rdev->raid_disk >= 0)
2665 rdev->recovery_offset = recovery_start;
2666 if (recovery_start == MaxSector)
2667 set_bit(In_sync, &rdev->flags);
2669 clear_bit(In_sync, &rdev->flags);
2673 static struct rdev_sysfs_entry rdev_recovery_start =
2674 __ATTR(recovery_start, S_IRUGO|S_IWUSR, recovery_start_show, recovery_start_store);
2676 static struct attribute *rdev_default_attrs[] = {
2682 &rdev_recovery_start.attr,
2686 rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
2688 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
2689 mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
2690 mddev_t *mddev = rdev->mddev;
2696 rv = mddev ? mddev_lock(mddev) : -EBUSY;
2698 if (rdev->mddev == NULL)
2701 rv = entry->show(rdev, page);
2702 mddev_unlock(mddev);
2708 rdev_attr_store(struct kobject *kobj, struct attribute *attr,
2709 const char *page, size_t length)
2711 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
2712 mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
2714 mddev_t *mddev = rdev->mddev;
2718 if (!capable(CAP_SYS_ADMIN))
2720 rv = mddev ? mddev_lock(mddev): -EBUSY;
2722 if (rdev->mddev == NULL)
2725 rv = entry->store(rdev, page, length);
2726 mddev_unlock(mddev);
2731 static void rdev_free(struct kobject *ko)
2733 mdk_rdev_t *rdev = container_of(ko, mdk_rdev_t, kobj);
2736 static const struct sysfs_ops rdev_sysfs_ops = {
2737 .show = rdev_attr_show,
2738 .store = rdev_attr_store,
2740 static struct kobj_type rdev_ktype = {
2741 .release = rdev_free,
2742 .sysfs_ops = &rdev_sysfs_ops,
2743 .default_attrs = rdev_default_attrs,
2746 void md_rdev_init(mdk_rdev_t *rdev)
2749 rdev->saved_raid_disk = -1;
2750 rdev->raid_disk = -1;
2752 rdev->data_offset = 0;
2753 rdev->sb_events = 0;
2754 rdev->last_read_error.tv_sec = 0;
2755 rdev->last_read_error.tv_nsec = 0;
2756 atomic_set(&rdev->nr_pending, 0);
2757 atomic_set(&rdev->read_errors, 0);
2758 atomic_set(&rdev->corrected_errors, 0);
2760 INIT_LIST_HEAD(&rdev->same_set);
2761 init_waitqueue_head(&rdev->blocked_wait);
2763 EXPORT_SYMBOL_GPL(md_rdev_init);
2765 * Import a device. If 'super_format' >= 0, then sanity check the superblock
2767 * mark the device faulty if:
2769 * - the device is nonexistent (zero size)
2770 * - the device has no valid superblock
2772 * a faulty rdev _never_ has rdev->sb set.
2774 static mdk_rdev_t *md_import_device(dev_t newdev, int super_format, int super_minor)
2776 char b[BDEVNAME_SIZE];
2781 rdev = kzalloc(sizeof(*rdev), GFP_KERNEL);
2783 printk(KERN_ERR "md: could not alloc mem for new device!\n");
2784 return ERR_PTR(-ENOMEM);
2788 if ((err = alloc_disk_sb(rdev)))
2791 err = lock_rdev(rdev, newdev, super_format == -2);
2795 kobject_init(&rdev->kobj, &rdev_ktype);
2797 size = i_size_read(rdev->bdev->bd_inode) >> BLOCK_SIZE_BITS;
2800 "md: %s has zero or unknown size, marking faulty!\n",
2801 bdevname(rdev->bdev,b));
2806 if (super_format >= 0) {
2807 err = super_types[super_format].
2808 load_super(rdev, NULL, super_minor);
2809 if (err == -EINVAL) {
2811 "md: %s does not have a valid v%d.%d "
2812 "superblock, not importing!\n",
2813 bdevname(rdev->bdev,b),
2814 super_format, super_minor);
2819 "md: could not read %s's sb, not importing!\n",
2820 bdevname(rdev->bdev,b));
2828 if (rdev->sb_page) {
2834 return ERR_PTR(err);
2838 * Check a full RAID array for plausibility
2842 static void analyze_sbs(mddev_t * mddev)
2845 mdk_rdev_t *rdev, *freshest, *tmp;
2846 char b[BDEVNAME_SIZE];
2849 rdev_for_each(rdev, tmp, mddev)
2850 switch (super_types[mddev->major_version].
2851 load_super(rdev, freshest, mddev->minor_version)) {
2859 "md: fatal superblock inconsistency in %s"
2860 " -- removing from array\n",
2861 bdevname(rdev->bdev,b));
2862 kick_rdev_from_array(rdev);
2866 super_types[mddev->major_version].
2867 validate_super(mddev, freshest);
2870 rdev_for_each(rdev, tmp, mddev) {
2871 if (mddev->max_disks &&
2872 (rdev->desc_nr >= mddev->max_disks ||
2873 i > mddev->max_disks)) {
2875 "md: %s: %s: only %d devices permitted\n",
2876 mdname(mddev), bdevname(rdev->bdev, b),
2878 kick_rdev_from_array(rdev);
2881 if (rdev != freshest)
2882 if (super_types[mddev->major_version].
2883 validate_super(mddev, rdev)) {
2884 printk(KERN_WARNING "md: kicking non-fresh %s"
2886 bdevname(rdev->bdev,b));
2887 kick_rdev_from_array(rdev);
2890 if (mddev->level == LEVEL_MULTIPATH) {
2891 rdev->desc_nr = i++;
2892 rdev->raid_disk = rdev->desc_nr;
2893 set_bit(In_sync, &rdev->flags);
2894 } else if (rdev->raid_disk >= (mddev->raid_disks - min(0, mddev->delta_disks))) {
2895 rdev->raid_disk = -1;
2896 clear_bit(In_sync, &rdev->flags);
2901 /* Read a fixed-point number.
2902 * Numbers in sysfs attributes should be in "standard" units where
2903 * possible, so time should be in seconds.
2904 * However we internally use a a much smaller unit such as
2905 * milliseconds or jiffies.
2906 * This function takes a decimal number with a possible fractional
2907 * component, and produces an integer which is the result of
2908 * multiplying that number by 10^'scale'.
2909 * all without any floating-point arithmetic.
2911 int strict_strtoul_scaled(const char *cp, unsigned long *res, int scale)
2913 unsigned long result = 0;
2915 while (isdigit(*cp) || (*cp == '.' && decimals < 0)) {
2918 else if (decimals < scale) {
2921 result = result * 10 + value;
2933 while (decimals < scale) {
2942 static void md_safemode_timeout(unsigned long data);
2945 safe_delay_show(mddev_t *mddev, char *page)
2947 int msec = (mddev->safemode_delay*1000)/HZ;
2948 return sprintf(page, "%d.%03d\n", msec/1000, msec%1000);
2951 safe_delay_store(mddev_t *mddev, const char *cbuf, size_t len)
2955 if (strict_strtoul_scaled(cbuf, &msec, 3) < 0)
2958 mddev->safemode_delay = 0;
2960 unsigned long old_delay = mddev->safemode_delay;
2961 mddev->safemode_delay = (msec*HZ)/1000;
2962 if (mddev->safemode_delay == 0)
2963 mddev->safemode_delay = 1;
2964 if (mddev->safemode_delay < old_delay)
2965 md_safemode_timeout((unsigned long)mddev);
2969 static struct md_sysfs_entry md_safe_delay =
2970 __ATTR(safe_mode_delay, S_IRUGO|S_IWUSR,safe_delay_show, safe_delay_store);
2973 level_show(mddev_t *mddev, char *page)
2975 struct mdk_personality *p = mddev->pers;
2977 return sprintf(page, "%s\n", p->name);
2978 else if (mddev->clevel[0])
2979 return sprintf(page, "%s\n", mddev->clevel);
2980 else if (mddev->level != LEVEL_NONE)
2981 return sprintf(page, "%d\n", mddev->level);
2987 level_store(mddev_t *mddev, const char *buf, size_t len)
2991 struct mdk_personality *pers;
2996 if (mddev->pers == NULL) {
2999 if (len >= sizeof(mddev->clevel))
3001 strncpy(mddev->clevel, buf, len);
3002 if (mddev->clevel[len-1] == '\n')
3004 mddev->clevel[len] = 0;
3005 mddev->level = LEVEL_NONE;
3009 /* request to change the personality. Need to ensure:
3010 * - array is not engaged in resync/recovery/reshape
3011 * - old personality can be suspended
3012 * - new personality will access other array.
3015 if (mddev->sync_thread ||
3016 mddev->reshape_position != MaxSector ||
3017 mddev->sysfs_active)
3020 if (!mddev->pers->quiesce) {
3021 printk(KERN_WARNING "md: %s: %s does not support online personality change\n",
3022 mdname(mddev), mddev->pers->name);
3026 /* Now find the new personality */
3027 if (len == 0 || len >= sizeof(clevel))
3029 strncpy(clevel, buf, len);
3030 if (clevel[len-1] == '\n')
3033 if (strict_strtol(clevel, 10, &level))
3036 if (request_module("md-%s", clevel) != 0)
3037 request_module("md-level-%s", clevel);
3038 spin_lock(&pers_lock);
3039 pers = find_pers(level, clevel);
3040 if (!pers || !try_module_get(pers->owner)) {
3041 spin_unlock(&pers_lock);
3042 printk(KERN_WARNING "md: personality %s not loaded\n", clevel);
3045 spin_unlock(&pers_lock);
3047 if (pers == mddev->pers) {
3048 /* Nothing to do! */
3049 module_put(pers->owner);
3052 if (!pers->takeover) {
3053 module_put(pers->owner);
3054 printk(KERN_WARNING "md: %s: %s does not support personality takeover\n",
3055 mdname(mddev), clevel);
3059 list_for_each_entry(rdev, &mddev->disks, same_set)
3060 rdev->new_raid_disk = rdev->raid_disk;
3062 /* ->takeover must set new_* and/or delta_disks
3063 * if it succeeds, and may set them when it fails.
3065 priv = pers->takeover(mddev);
3067 mddev->new_level = mddev->level;
3068 mddev->new_layout = mddev->layout;
3069 mddev->new_chunk_sectors = mddev->chunk_sectors;
3070 mddev->raid_disks -= mddev->delta_disks;
3071 mddev->delta_disks = 0;
3072 module_put(pers->owner);
3073 printk(KERN_WARNING "md: %s: %s would not accept array\n",
3074 mdname(mddev), clevel);
3075 return PTR_ERR(priv);
3078 /* Looks like we have a winner */
3079 mddev_suspend(mddev);
3080 mddev->pers->stop(mddev);
3082 if (mddev->pers->sync_request == NULL &&
3083 pers->sync_request != NULL) {
3084 /* need to add the md_redundancy_group */
3085 if (sysfs_create_group(&mddev->kobj, &md_redundancy_group))
3087 "md: cannot register extra attributes for %s\n",
3089 mddev->sysfs_action = sysfs_get_dirent(mddev->kobj.sd, NULL, "sync_action");
3091 if (mddev->pers->sync_request != NULL &&
3092 pers->sync_request == NULL) {
3093 /* need to remove the md_redundancy_group */
3094 if (mddev->to_remove == NULL)
3095 mddev->to_remove = &md_redundancy_group;
3098 if (mddev->pers->sync_request == NULL &&
3100 /* We are converting from a no-redundancy array
3101 * to a redundancy array and metadata is managed
3102 * externally so we need to be sure that writes
3103 * won't block due to a need to transition
3105 * until external management is started.
3108 mddev->safemode_delay = 0;
3109 mddev->safemode = 0;
3112 list_for_each_entry(rdev, &mddev->disks, same_set) {
3114 if (rdev->raid_disk < 0)
3116 if (rdev->new_raid_disk > mddev->raid_disks)
3117 rdev->new_raid_disk = -1;
3118 if (rdev->new_raid_disk == rdev->raid_disk)
3120 sprintf(nm, "rd%d", rdev->raid_disk);
3121 sysfs_remove_link(&mddev->kobj, nm);
3123 list_for_each_entry(rdev, &mddev->disks, same_set) {
3124 if (rdev->raid_disk < 0)
3126 if (rdev->new_raid_disk == rdev->raid_disk)
3128 rdev->raid_disk = rdev->new_raid_disk;
3129 if (rdev->raid_disk < 0)
3130 clear_bit(In_sync, &rdev->flags);
3133 sprintf(nm, "rd%d", rdev->raid_disk);
3134 if(sysfs_create_link(&mddev->kobj, &rdev->kobj, nm))
3135 printk("md: cannot register %s for %s after level change\n",
3140 module_put(mddev->pers->owner);
3142 mddev->private = priv;
3143 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
3144 mddev->level = mddev->new_level;
3145 mddev->layout = mddev->new_layout;
3146 mddev->chunk_sectors = mddev->new_chunk_sectors;
3147 mddev->delta_disks = 0;
3148 if (mddev->pers->sync_request == NULL) {
3149 /* this is now an array without redundancy, so
3150 * it must always be in_sync
3153 del_timer_sync(&mddev->safemode_timer);
3156 mddev_resume(mddev);
3157 set_bit(MD_CHANGE_DEVS, &mddev->flags);
3158 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3159 md_wakeup_thread(mddev->thread);
3160 sysfs_notify(&mddev->kobj, NULL, "level");
3161 md_new_event(mddev);
3165 static struct md_sysfs_entry md_level =
3166 __ATTR(level, S_IRUGO|S_IWUSR, level_show, level_store);
3170 layout_show(mddev_t *mddev, char *page)
3172 /* just a number, not meaningful for all levels */
3173 if (mddev->reshape_position != MaxSector &&
3174 mddev->layout != mddev->new_layout)
3175 return sprintf(page, "%d (%d)\n",
3176 mddev->new_layout, mddev->layout);
3177 return sprintf(page, "%d\n", mddev->layout);
3181 layout_store(mddev_t *mddev, const char *buf, size_t len)
3184 unsigned long n = simple_strtoul(buf, &e, 10);
3186 if (!*buf || (*e && *e != '\n'))
3191 if (mddev->pers->check_reshape == NULL)
3193 mddev->new_layout = n;
3194 err = mddev->pers->check_reshape(mddev);
3196 mddev->new_layout = mddev->layout;
3200 mddev->new_layout = n;
3201 if (mddev->reshape_position == MaxSector)
3206 static struct md_sysfs_entry md_layout =
3207 __ATTR(layout, S_IRUGO|S_IWUSR, layout_show, layout_store);
3211 raid_disks_show(mddev_t *mddev, char *page)
3213 if (mddev->raid_disks == 0)
3215 if (mddev->reshape_position != MaxSector &&
3216 mddev->delta_disks != 0)
3217 return sprintf(page, "%d (%d)\n", mddev->raid_disks,
3218 mddev->raid_disks - mddev->delta_disks);
3219 return sprintf(page, "%d\n", mddev->raid_disks);
3222 static int update_raid_disks(mddev_t *mddev, int raid_disks);
3225 raid_disks_store(mddev_t *mddev, const char *buf, size_t len)
3229 unsigned long n = simple_strtoul(buf, &e, 10);
3231 if (!*buf || (*e && *e != '\n'))
3235 rv = update_raid_disks(mddev, n);
3236 else if (mddev->reshape_position != MaxSector) {
3237 int olddisks = mddev->raid_disks - mddev->delta_disks;
3238 mddev->delta_disks = n - olddisks;
3239 mddev->raid_disks = n;
3241 mddev->raid_disks = n;
3242 return rv ? rv : len;
3244 static struct md_sysfs_entry md_raid_disks =
3245 __ATTR(raid_disks, S_IRUGO|S_IWUSR, raid_disks_show, raid_disks_store);
3248 chunk_size_show(mddev_t *mddev, char *page)
3250 if (mddev->reshape_position != MaxSector &&
3251 mddev->chunk_sectors != mddev->new_chunk_sectors)
3252 return sprintf(page, "%d (%d)\n",
3253 mddev->new_chunk_sectors << 9,
3254 mddev->chunk_sectors << 9);
3255 return sprintf(page, "%d\n", mddev->chunk_sectors << 9);
3259 chunk_size_store(mddev_t *mddev, const char *buf, size_t len)
3262 unsigned long n = simple_strtoul(buf, &e, 10);
3264 if (!*buf || (*e && *e != '\n'))
3269 if (mddev->pers->check_reshape == NULL)
3271 mddev->new_chunk_sectors = n >> 9;
3272 err = mddev->pers->check_reshape(mddev);
3274 mddev->new_chunk_sectors = mddev->chunk_sectors;
3278 mddev->new_chunk_sectors = n >> 9;
3279 if (mddev->reshape_position == MaxSector)
3280 mddev->chunk_sectors = n >> 9;
3284 static struct md_sysfs_entry md_chunk_size =
3285 __ATTR(chunk_size, S_IRUGO|S_IWUSR, chunk_size_show, chunk_size_store);
3288 resync_start_show(mddev_t *mddev, char *page)
3290 if (mddev->recovery_cp == MaxSector)
3291 return sprintf(page, "none\n");
3292 return sprintf(page, "%llu\n", (unsigned long long)mddev->recovery_cp);
3296 resync_start_store(mddev_t *mddev, const char *buf, size_t len)
3299 unsigned long long n = simple_strtoull(buf, &e, 10);
3303 if (cmd_match(buf, "none"))
3305 else if (!*buf || (*e && *e != '\n'))
3308 mddev->recovery_cp = n;
3311 static struct md_sysfs_entry md_resync_start =
3312 __ATTR(resync_start, S_IRUGO|S_IWUSR, resync_start_show, resync_start_store);
3315 * The array state can be:
3318 * No devices, no size, no level
3319 * Equivalent to STOP_ARRAY ioctl
3321 * May have some settings, but array is not active
3322 * all IO results in error
3323 * When written, doesn't tear down array, but just stops it
3324 * suspended (not supported yet)
3325 * All IO requests will block. The array can be reconfigured.
3326 * Writing this, if accepted, will block until array is quiescent
3328 * no resync can happen. no superblocks get written.
3329 * write requests fail
3331 * like readonly, but behaves like 'clean' on a write request.
3333 * clean - no pending writes, but otherwise active.
3334 * When written to inactive array, starts without resync
3335 * If a write request arrives then
3336 * if metadata is known, mark 'dirty' and switch to 'active'.
3337 * if not known, block and switch to write-pending
3338 * If written to an active array that has pending writes, then fails.
3340 * fully active: IO and resync can be happening.
3341 * When written to inactive array, starts with resync
3344 * clean, but writes are blocked waiting for 'active' to be written.
3347 * like active, but no writes have been seen for a while (100msec).
3350 enum array_state { clear, inactive, suspended, readonly, read_auto, clean, active,
3351 write_pending, active_idle, bad_word};
3352 static char *array_states[] = {
3353 "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active",
3354 "write-pending", "active-idle", NULL };
3356 static int match_word(const char *word, char **list)
3359 for (n=0; list[n]; n++)
3360 if (cmd_match(word, list[n]))
3366 array_state_show(mddev_t *mddev, char *page)
3368 enum array_state st = inactive;
3381 else if (test_bit(MD_CHANGE_PENDING, &mddev->flags))
3383 else if (mddev->safemode)
3389 if (list_empty(&mddev->disks) &&
3390 mddev->raid_disks == 0 &&
3391 mddev->dev_sectors == 0)
3396 return sprintf(page, "%s\n", array_states[st]);
3399 static int do_md_stop(mddev_t * mddev, int ro, int is_open);
3400 static int md_set_readonly(mddev_t * mddev, int is_open);
3401 static int do_md_run(mddev_t * mddev);
3402 static int restart_array(mddev_t *mddev);
3405 array_state_store(mddev_t *mddev, const char *buf, size_t len)
3408 enum array_state st = match_word(buf, array_states);
3413 /* stopping an active array */
3414 if (atomic_read(&mddev->openers) > 0)
3416 err = do_md_stop(mddev, 0, 0);
3419 /* stopping an active array */
3421 if (atomic_read(&mddev->openers) > 0)
3423 err = do_md_stop(mddev, 2, 0);
3425 err = 0; /* already inactive */
3428 break; /* not supported yet */
3431 err = md_set_readonly(mddev, 0);
3434 set_disk_ro(mddev->gendisk, 1);
3435 err = do_md_run(mddev);
3441 err = md_set_readonly(mddev, 0);
3442 else if (mddev->ro == 1)
3443 err = restart_array(mddev);
3446 set_disk_ro(mddev->gendisk, 0);
3450 err = do_md_run(mddev);
3455 restart_array(mddev);
3456 spin_lock_irq(&mddev->write_lock);
3457 if (atomic_read(&mddev->writes_pending) == 0) {
3458 if (mddev->in_sync == 0) {
3460 if (mddev->safemode == 1)
3461 mddev->safemode = 0;
3462 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
3467 spin_unlock_irq(&mddev->write_lock);
3473 restart_array(mddev);
3474 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
3475 wake_up(&mddev->sb_wait);
3479 set_disk_ro(mddev->gendisk, 0);
3480 err = do_md_run(mddev);
3485 /* these cannot be set */
3491 sysfs_notify_dirent_safe(mddev->sysfs_state);
3495 static struct md_sysfs_entry md_array_state =
3496 __ATTR(array_state, S_IRUGO|S_IWUSR, array_state_show, array_state_store);
3499 max_corrected_read_errors_show(mddev_t *mddev, char *page) {
3500 return sprintf(page, "%d\n",
3501 atomic_read(&mddev->max_corr_read_errors));
3505 max_corrected_read_errors_store(mddev_t *mddev, const char *buf, size_t len)
3508 unsigned long n = simple_strtoul(buf, &e, 10);
3510 if (*buf && (*e == 0 || *e == '\n')) {
3511 atomic_set(&mddev->max_corr_read_errors, n);
3517 static struct md_sysfs_entry max_corr_read_errors =
3518 __ATTR(max_read_errors, S_IRUGO|S_IWUSR, max_corrected_read_errors_show,
3519 max_corrected_read_errors_store);
3522 null_show(mddev_t *mddev, char *page)
3528 new_dev_store(mddev_t *mddev, const char *buf, size_t len)
3530 /* buf must be %d:%d\n? giving major and minor numbers */
3531 /* The new device is added to the array.
3532 * If the array has a persistent superblock, we read the
3533 * superblock to initialise info and check validity.
3534 * Otherwise, only checking done is that in bind_rdev_to_array,
3535 * which mainly checks size.
3538 int major = simple_strtoul(buf, &e, 10);
3544 if (!*buf || *e != ':' || !e[1] || e[1] == '\n')
3546 minor = simple_strtoul(e+1, &e, 10);
3547 if (*e && *e != '\n')
3549 dev = MKDEV(major, minor);
3550 if (major != MAJOR(dev) ||
3551 minor != MINOR(dev))
3555 if (mddev->persistent) {
3556 rdev = md_import_device(dev, mddev->major_version,
3557 mddev->minor_version);
3558 if (!IS_ERR(rdev) && !list_empty(&mddev->disks)) {
3559 mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
3560 mdk_rdev_t, same_set);
3561 err = super_types[mddev->major_version]
3562 .load_super(rdev, rdev0, mddev->minor_version);
3566 } else if (mddev->external)
3567 rdev = md_import_device(dev, -2, -1);
3569 rdev = md_import_device(dev, -1, -1);
3572 return PTR_ERR(rdev);
3573 err = bind_rdev_to_array(rdev, mddev);
3577 return err ? err : len;
3580 static struct md_sysfs_entry md_new_device =
3581 __ATTR(new_dev, S_IWUSR, null_show, new_dev_store);
3584 bitmap_store(mddev_t *mddev, const char *buf, size_t len)
3587 unsigned long chunk, end_chunk;
3591 /* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */
3593 chunk = end_chunk = simple_strtoul(buf, &end, 0);
3594 if (buf == end) break;
3595 if (*end == '-') { /* range */
3597 end_chunk = simple_strtoul(buf, &end, 0);
3598 if (buf == end) break;
3600 if (*end && !isspace(*end)) break;
3601 bitmap_dirty_bits(mddev->bitmap, chunk, end_chunk);
3602 buf = skip_spaces(end);
3604 bitmap_unplug(mddev->bitmap); /* flush the bits to disk */
3609 static struct md_sysfs_entry md_bitmap =
3610 __ATTR(bitmap_set_bits, S_IWUSR, null_show, bitmap_store);
3613 size_show(mddev_t *mddev, char *page)
3615 return sprintf(page, "%llu\n",
3616 (unsigned long long)mddev->dev_sectors / 2);
3619 static int update_size(mddev_t *mddev, sector_t num_sectors);
3622 size_store(mddev_t *mddev, const char *buf, size_t len)
3624 /* If array is inactive, we can reduce the component size, but
3625 * not increase it (except from 0).
3626 * If array is active, we can try an on-line resize
3629 int err = strict_blocks_to_sectors(buf, §ors);
3634 err = update_size(mddev, sectors);
3635 md_update_sb(mddev, 1);
3637 if (mddev->dev_sectors == 0 ||
3638 mddev->dev_sectors > sectors)
3639 mddev->dev_sectors = sectors;
3643 return err ? err : len;
3646 static struct md_sysfs_entry md_size =
3647 __ATTR(component_size, S_IRUGO|S_IWUSR, size_show, size_store);
3652 * 'none' for arrays with no metadata (good luck...)
3653 * 'external' for arrays with externally managed metadata,
3654 * or N.M for internally known formats
3657 metadata_show(mddev_t *mddev, char *page)
3659 if (mddev->persistent)
3660 return sprintf(page, "%d.%d\n",
3661 mddev->major_version, mddev->minor_version);
3662 else if (mddev->external)
3663 return sprintf(page, "external:%s\n", mddev->metadata_type);
3665 return sprintf(page, "none\n");
3669 metadata_store(mddev_t *mddev, const char *buf, size_t len)
3673 /* Changing the details of 'external' metadata is
3674 * always permitted. Otherwise there must be
3675 * no devices attached to the array.
3677 if (mddev->external && strncmp(buf, "external:", 9) == 0)
3679 else if (!list_empty(&mddev->disks))
3682 if (cmd_match(buf, "none")) {
3683 mddev->persistent = 0;
3684 mddev->external = 0;
3685 mddev->major_version = 0;
3686 mddev->minor_version = 90;
3689 if (strncmp(buf, "external:", 9) == 0) {
3690 size_t namelen = len-9;
3691 if (namelen >= sizeof(mddev->metadata_type))
3692 namelen = sizeof(mddev->metadata_type)-1;
3693 strncpy(mddev->metadata_type, buf+9, namelen);
3694 mddev->metadata_type[namelen] = 0;
3695 if (namelen && mddev->metadata_type[namelen-1] == '\n')
3696 mddev->metadata_type[--namelen] = 0;
3697 mddev->persistent = 0;
3698 mddev->external = 1;
3699 mddev->major_version = 0;
3700 mddev->minor_version = 90;
3703 major = simple_strtoul(buf, &e, 10);
3704 if (e==buf || *e != '.')
3707 minor = simple_strtoul(buf, &e, 10);
3708 if (e==buf || (*e && *e != '\n') )
3710 if (major >= ARRAY_SIZE(super_types) || super_types[major].name == NULL)
3712 mddev->major_version = major;
3713 mddev->minor_version = minor;
3714 mddev->persistent = 1;
3715 mddev->external = 0;
3719 static struct md_sysfs_entry md_metadata =
3720 __ATTR(metadata_version, S_IRUGO|S_IWUSR, metadata_show, metadata_store);
3723 action_show(mddev_t *mddev, char *page)
3725 char *type = "idle";
3726 if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
3728 else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
3729 (!mddev->ro && test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))) {
3730 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
3732 else if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
3733 if (!test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
3735 else if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
3739 } else if (test_bit(MD_RECOVERY_RECOVER, &mddev->recovery))
3742 return sprintf(page, "%s\n", type);
3746 action_store(mddev_t *mddev, const char *page, size_t len)
3748 if (!mddev->pers || !mddev->pers->sync_request)
3751 if (cmd_match(page, "frozen"))
3752 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
3754 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
3756 if (cmd_match(page, "idle") || cmd_match(page, "frozen")) {
3757 if (mddev->sync_thread) {
3758 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
3759 md_unregister_thread(mddev->sync_thread);
3760 mddev->sync_thread = NULL;
3761 mddev->recovery = 0;
3763 } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
3764 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
3766 else if (cmd_match(page, "resync"))
3767 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3768 else if (cmd_match(page, "recover")) {
3769 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
3770 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3771 } else if (cmd_match(page, "reshape")) {
3773 if (mddev->pers->start_reshape == NULL)
3775 err = mddev->pers->start_reshape(mddev);
3778 sysfs_notify(&mddev->kobj, NULL, "degraded");
3780 if (cmd_match(page, "check"))
3781 set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
3782 else if (!cmd_match(page, "repair"))
3784 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
3785 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
3787 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3788 md_wakeup_thread(mddev->thread);
3789 sysfs_notify_dirent_safe(mddev->sysfs_action);
3794 mismatch_cnt_show(mddev_t *mddev, char *page)
3796 return sprintf(page, "%llu\n",
3797 (unsigned long long) mddev->resync_mismatches);
3800 static struct md_sysfs_entry md_scan_mode =
3801 __ATTR(sync_action, S_IRUGO|S_IWUSR, action_show, action_store);
3804 static struct md_sysfs_entry md_mismatches = __ATTR_RO(mismatch_cnt);
3807 sync_min_show(mddev_t *mddev, char *page)
3809 return sprintf(page, "%d (%s)\n", speed_min(mddev),
3810 mddev->sync_speed_min ? "local": "system");
3814 sync_min_store(mddev_t *mddev, const char *buf, size_t len)
3818 if (strncmp(buf, "system", 6)==0) {
3819 mddev->sync_speed_min = 0;
3822 min = simple_strtoul(buf, &e, 10);
3823 if (buf == e || (*e && *e != '\n') || min <= 0)
3825 mddev->sync_speed_min = min;
3829 static struct md_sysfs_entry md_sync_min =
3830 __ATTR(sync_speed_min, S_IRUGO|S_IWUSR, sync_min_show, sync_min_store);
3833 sync_max_show(mddev_t *mddev, char *page)
3835 return sprintf(page, "%d (%s)\n", speed_max(mddev),
3836 mddev->sync_speed_max ? "local": "system");
3840 sync_max_store(mddev_t *mddev, const char *buf, size_t len)
3844 if (strncmp(buf, "system", 6)==0) {
3845 mddev->sync_speed_max = 0;
3848 max = simple_strtoul(buf, &e, 10);
3849 if (buf == e || (*e && *e != '\n') || max <= 0)
3851 mddev->sync_speed_max = max;
3855 static struct md_sysfs_entry md_sync_max =
3856 __ATTR(sync_speed_max, S_IRUGO|S_IWUSR, sync_max_show, sync_max_store);
3859 degraded_show(mddev_t *mddev, char *page)
3861 return sprintf(page, "%d\n", mddev->degraded);
3863 static struct md_sysfs_entry md_degraded = __ATTR_RO(degraded);
3866 sync_force_parallel_show(mddev_t *mddev, char *page)
3868 return sprintf(page, "%d\n", mddev->parallel_resync);
3872 sync_force_parallel_store(mddev_t *mddev, const char *buf, size_t len)
3876 if (strict_strtol(buf, 10, &n))
3879 if (n != 0 && n != 1)
3882 mddev->parallel_resync = n;
3884 if (mddev->sync_thread)
3885 wake_up(&resync_wait);
3890 /* force parallel resync, even with shared block devices */
3891 static struct md_sysfs_entry md_sync_force_parallel =
3892 __ATTR(sync_force_parallel, S_IRUGO|S_IWUSR,
3893 sync_force_parallel_show, sync_force_parallel_store);
3896 sync_speed_show(mddev_t *mddev, char *page)
3898 unsigned long resync, dt, db;
3899 if (mddev->curr_resync == 0)
3900 return sprintf(page, "none\n");
3901 resync = mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active);
3902 dt = (jiffies - mddev->resync_mark) / HZ;
3904 db = resync - mddev->resync_mark_cnt;
3905 return sprintf(page, "%lu\n", db/dt/2); /* K/sec */
3908 static struct md_sysfs_entry md_sync_speed = __ATTR_RO(sync_speed);
3911 sync_completed_show(mddev_t *mddev, char *page)
3913 unsigned long max_sectors, resync;
3915 if (!test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
3916 return sprintf(page, "none\n");
3918 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
3919 max_sectors = mddev->resync_max_sectors;
3921 max_sectors = mddev->dev_sectors;
3923 resync = mddev->curr_resync_completed;
3924 return sprintf(page, "%lu / %lu\n", resync, max_sectors);
3927 static struct md_sysfs_entry md_sync_completed = __ATTR_RO(sync_completed);
3930 min_sync_show(mddev_t *mddev, char *page)
3932 return sprintf(page, "%llu\n",
3933 (unsigned long long)mddev->resync_min);
3936 min_sync_store(mddev_t *mddev, const char *buf, size_t len)
3938 unsigned long long min;
3939 if (strict_strtoull(buf, 10, &min))
3941 if (min > mddev->resync_max)
3943 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
3946 /* Must be a multiple of chunk_size */
3947 if (mddev->chunk_sectors) {
3948 sector_t temp = min;
3949 if (sector_div(temp, mddev->chunk_sectors))
3952 mddev->resync_min = min;
3957 static struct md_sysfs_entry md_min_sync =
3958 __ATTR(sync_min, S_IRUGO|S_IWUSR, min_sync_show, min_sync_store);
3961 max_sync_show(mddev_t *mddev, char *page)
3963 if (mddev->resync_max == MaxSector)
3964 return sprintf(page, "max\n");
3966 return sprintf(page, "%llu\n",
3967 (unsigned long long)mddev->resync_max);
3970 max_sync_store(mddev_t *mddev, const char *buf, size_t len)
3972 if (strncmp(buf, "max", 3) == 0)
3973 mddev->resync_max = MaxSector;
3975 unsigned long long max;
3976 if (strict_strtoull(buf, 10, &max))
3978 if (max < mddev->resync_min)
3980 if (max < mddev->resync_max &&
3982 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
3985 /* Must be a multiple of chunk_size */
3986 if (mddev->chunk_sectors) {
3987 sector_t temp = max;
3988 if (sector_div(temp, mddev->chunk_sectors))
3991 mddev->resync_max = max;
3993 wake_up(&mddev->recovery_wait);
3997 static struct md_sysfs_entry md_max_sync =
3998 __ATTR(sync_max, S_IRUGO|S_IWUSR, max_sync_show, max_sync_store);
4001 suspend_lo_show(mddev_t *mddev, char *page)
4003 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_lo);
4007 suspend_lo_store(mddev_t *mddev, const char *buf, size_t len)
4010 unsigned long long new = simple_strtoull(buf, &e, 10);
4012 if (mddev->pers == NULL ||
4013 mddev->pers->quiesce == NULL)
4015 if (buf == e || (*e && *e != '\n'))
4017 if (new >= mddev->suspend_hi ||
4018 (new > mddev->suspend_lo && new < mddev->suspend_hi)) {
4019 mddev->suspend_lo = new;
4020 mddev->pers->quiesce(mddev, 2);
4025 static struct md_sysfs_entry md_suspend_lo =
4026 __ATTR(suspend_lo, S_IRUGO|S_IWUSR, suspend_lo_show, suspend_lo_store);
4030 suspend_hi_show(mddev_t *mddev, char *page)
4032 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_hi);
4036 suspend_hi_store(mddev_t *mddev, const char *buf, size_t len)
4039 unsigned long long new = simple_strtoull(buf, &e, 10);
4041 if (mddev->pers == NULL ||
4042 mddev->pers->quiesce == NULL)
4044 if (buf == e || (*e && *e != '\n'))
4046 if ((new <= mddev->suspend_lo && mddev->suspend_lo >= mddev->suspend_hi) ||
4047 (new > mddev->suspend_lo && new > mddev->suspend_hi)) {
4048 mddev->suspend_hi = new;
4049 mddev->pers->quiesce(mddev, 1);
4050 mddev->pers->quiesce(mddev, 0);
4055 static struct md_sysfs_entry md_suspend_hi =
4056 __ATTR(suspend_hi, S_IRUGO|S_IWUSR, suspend_hi_show, suspend_hi_store);
4059 reshape_position_show(mddev_t *mddev, char *page)
4061 if (mddev->reshape_position != MaxSector)
4062 return sprintf(page, "%llu\n",
4063 (unsigned long long)mddev->reshape_position);
4064 strcpy(page, "none\n");
4069 reshape_position_store(mddev_t *mddev, const char *buf, size_t len)
4072 unsigned long long new = simple_strtoull(buf, &e, 10);
4075 if (buf == e || (*e && *e != '\n'))
4077 mddev->reshape_position = new;
4078 mddev->delta_disks = 0;
4079 mddev->new_level = mddev->level;
4080 mddev->new_layout = mddev->layout;
4081 mddev->new_chunk_sectors = mddev->chunk_sectors;
4085 static struct md_sysfs_entry md_reshape_position =
4086 __ATTR(reshape_position, S_IRUGO|S_IWUSR, reshape_position_show,
4087 reshape_position_store);
4090 array_size_show(mddev_t *mddev, char *page)
4092 if (mddev->external_size)
4093 return sprintf(page, "%llu\n",
4094 (unsigned long long)mddev->array_sectors/2);
4096 return sprintf(page, "default\n");
4100 array_size_store(mddev_t *mddev, const char *buf, size_t len)
4104 if (strncmp(buf, "default", 7) == 0) {
4106 sectors = mddev->pers->size(mddev, 0, 0);
4108 sectors = mddev->array_sectors;
4110 mddev->external_size = 0;
4112 if (strict_blocks_to_sectors(buf, §ors) < 0)
4114 if (mddev->pers && mddev->pers->size(mddev, 0, 0) < sectors)
4117 mddev->external_size = 1;
4120 mddev->array_sectors = sectors;
4121 set_capacity(mddev->gendisk, mddev->array_sectors);
4123 revalidate_disk(mddev->gendisk);
4128 static struct md_sysfs_entry md_array_size =
4129 __ATTR(array_size, S_IRUGO|S_IWUSR, array_size_show,
4132 static struct attribute *md_default_attrs[] = {
4135 &md_raid_disks.attr,
4136 &md_chunk_size.attr,
4138 &md_resync_start.attr,
4140 &md_new_device.attr,
4141 &md_safe_delay.attr,
4142 &md_array_state.attr,
4143 &md_reshape_position.attr,
4144 &md_array_size.attr,
4145 &max_corr_read_errors.attr,
4149 static struct attribute *md_redundancy_attrs[] = {
4151 &md_mismatches.attr,
4154 &md_sync_speed.attr,
4155 &md_sync_force_parallel.attr,
4156 &md_sync_completed.attr,
4159 &md_suspend_lo.attr,
4160 &md_suspend_hi.attr,
4165 static struct attribute_group md_redundancy_group = {
4167 .attrs = md_redundancy_attrs,
4172 md_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
4174 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
4175 mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
4180 rv = mddev_lock(mddev);
4182 rv = entry->show(mddev, page);
4183 mddev_unlock(mddev);
4189 md_attr_store(struct kobject *kobj, struct attribute *attr,
4190 const char *page, size_t length)
4192 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
4193 mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
4198 if (!capable(CAP_SYS_ADMIN))
4200 rv = mddev_lock(mddev);
4201 if (mddev->hold_active == UNTIL_IOCTL)
4202 mddev->hold_active = 0;
4204 rv = entry->store(mddev, page, length);
4205 mddev_unlock(mddev);
4210 static void md_free(struct kobject *ko)
4212 mddev_t *mddev = container_of(ko, mddev_t, kobj);
4214 if (mddev->sysfs_state)
4215 sysfs_put(mddev->sysfs_state);
4217 if (mddev->gendisk) {
4218 del_gendisk(mddev->gendisk);
4219 put_disk(mddev->gendisk);
4222 blk_cleanup_queue(mddev->queue);
4227 static const struct sysfs_ops md_sysfs_ops = {
4228 .show = md_attr_show,
4229 .store = md_attr_store,
4231 static struct kobj_type md_ktype = {
4233 .sysfs_ops = &md_sysfs_ops,
4234 .default_attrs = md_default_attrs,
4239 static void mddev_delayed_delete(struct work_struct *ws)
4241 mddev_t *mddev = container_of(ws, mddev_t, del_work);
4243 sysfs_remove_group(&mddev->kobj, &md_bitmap_group);
4244 kobject_del(&mddev->kobj);
4245 kobject_put(&mddev->kobj);
4248 static int md_alloc(dev_t dev, char *name)
4250 static DEFINE_MUTEX(disks_mutex);
4251 mddev_t *mddev = mddev_find(dev);
4252 struct gendisk *disk;
4261 partitioned = (MAJOR(mddev->unit) != MD_MAJOR);
4262 shift = partitioned ? MdpMinorShift : 0;
4263 unit = MINOR(mddev->unit) >> shift;
4265 /* wait for any previous instance of this device to be
4266 * completely removed (mddev_delayed_delete).
4268 flush_workqueue(md_misc_wq);
4270 mutex_lock(&disks_mutex);
4276 /* Need to ensure that 'name' is not a duplicate.
4279 spin_lock(&all_mddevs_lock);
4281 list_for_each_entry(mddev2, &all_mddevs, all_mddevs)
4282 if (mddev2->gendisk &&
4283 strcmp(mddev2->gendisk->disk_name, name) == 0) {
4284 spin_unlock(&all_mddevs_lock);
4287 spin_unlock(&all_mddevs_lock);
4291 mddev->queue = blk_alloc_queue(GFP_KERNEL);
4294 mddev->queue->queuedata = mddev;
4296 /* Can be unlocked because the queue is new: no concurrency */
4297 queue_flag_set_unlocked(QUEUE_FLAG_CLUSTER, mddev->queue);
4299 blk_queue_make_request(mddev->queue, md_make_request);
4301 disk = alloc_disk(1 << shift);
4303 blk_cleanup_queue(mddev->queue);
4304 mddev->queue = NULL;
4307 disk->major = MAJOR(mddev->unit);
4308 disk->first_minor = unit << shift;
4310 strcpy(disk->disk_name, name);
4311 else if (partitioned)
4312 sprintf(disk->disk_name, "md_d%d", unit);
4314 sprintf(disk->disk_name, "md%d", unit);
4315 disk->fops = &md_fops;
4316 disk->private_data = mddev;
4317 disk->queue = mddev->queue;
4318 /* Allow extended partitions. This makes the
4319 * 'mdp' device redundant, but we can't really
4322 disk->flags |= GENHD_FL_EXT_DEVT;
4324 mddev->gendisk = disk;
4325 error = kobject_init_and_add(&mddev->kobj, &md_ktype,
4326 &disk_to_dev(disk)->kobj, "%s", "md");
4328 /* This isn't possible, but as kobject_init_and_add is marked
4329 * __must_check, we must do something with the result
4331 printk(KERN_WARNING "md: cannot register %s/md - name in use\n",
4335 if (mddev->kobj.sd &&
4336 sysfs_create_group(&mddev->kobj, &md_bitmap_group))
4337 printk(KERN_DEBUG "pointless warning\n");
4339 blk_queue_flush(mddev->queue, REQ_FLUSH | REQ_FUA);
4341 mutex_unlock(&disks_mutex);
4342 if (!error && mddev->kobj.sd) {
4343 kobject_uevent(&mddev->kobj, KOBJ_ADD);
4344 mddev->sysfs_state = sysfs_get_dirent_safe(mddev->kobj.sd, "array_state");
4350 static struct kobject *md_probe(dev_t dev, int *part, void *data)
4352 md_alloc(dev, NULL);
4356 static int add_named_array(const char *val, struct kernel_param *kp)
4358 /* val must be "md_*" where * is not all digits.
4359 * We allocate an array with a large free minor number, and
4360 * set the name to val. val must not already be an active name.
4362 int len = strlen(val);
4363 char buf[DISK_NAME_LEN];
4365 while (len && val[len-1] == '\n')
4367 if (len >= DISK_NAME_LEN)
4369 strlcpy(buf, val, len+1);
4370 if (strncmp(buf, "md_", 3) != 0)
4372 return md_alloc(0, buf);
4375 static void md_safemode_timeout(unsigned long data)
4377 mddev_t *mddev = (mddev_t *) data;
4379 if (!atomic_read(&mddev->writes_pending)) {
4380 mddev->safemode = 1;
4381 if (mddev->external)
4382 sysfs_notify_dirent_safe(mddev->sysfs_state);
4384 md_wakeup_thread(mddev->thread);
4387 static int start_dirty_degraded;
4389 int md_run(mddev_t *mddev)
4393 struct mdk_personality *pers;
4395 if (list_empty(&mddev->disks))
4396 /* cannot run an array with no devices.. */
4401 /* Cannot run until previous stop completes properly */
4402 if (mddev->sysfs_active)
4406 * Analyze all RAID superblock(s)
4408 if (!mddev->raid_disks) {
4409 if (!mddev->persistent)
4414 if (mddev->level != LEVEL_NONE)
4415 request_module("md-level-%d", mddev->level);
4416 else if (mddev->clevel[0])
4417 request_module("md-%s", mddev->clevel);
4420 * Drop all container device buffers, from now on
4421 * the only valid external interface is through the md
4424 list_for_each_entry(rdev, &mddev->disks, same_set) {
4425 if (test_bit(Faulty, &rdev->flags))
4427 sync_blockdev(rdev->bdev);
4428 invalidate_bdev(rdev->bdev);
4430 /* perform some consistency tests on the device.
4431 * We don't want the data to overlap the metadata,
4432 * Internal Bitmap issues have been handled elsewhere.
4434 if (rdev->data_offset < rdev->sb_start) {
4435 if (mddev->dev_sectors &&
4436 rdev->data_offset + mddev->dev_sectors
4438 printk("md: %s: data overlaps metadata\n",
4443 if (rdev->sb_start + rdev->sb_size/512
4444 > rdev->data_offset) {
4445 printk("md: %s: metadata overlaps data\n",
4450 sysfs_notify_dirent_safe(rdev->sysfs_state);
4453 if (mddev->bio_set == NULL)
4454 mddev->bio_set = bioset_create(BIO_POOL_SIZE, sizeof(mddev));
4456 spin_lock(&pers_lock);
4457 pers = find_pers(mddev->level, mddev->clevel);
4458 if (!pers || !try_module_get(pers->owner)) {
4459 spin_unlock(&pers_lock);
4460 if (mddev->level != LEVEL_NONE)
4461 printk(KERN_WARNING "md: personality for level %d is not loaded!\n",
4464 printk(KERN_WARNING "md: personality for level %s is not loaded!\n",
4469 spin_unlock(&pers_lock);
4470 if (mddev->level != pers->level) {
4471 mddev->level = pers->level;
4472 mddev->new_level = pers->level;
4474 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
4476 if (mddev->reshape_position != MaxSector &&
4477 pers->start_reshape == NULL) {
4478 /* This personality cannot handle reshaping... */
4480 module_put(pers->owner);
4484 if (pers->sync_request) {
4485 /* Warn if this is a potentially silly
4488 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
4492 list_for_each_entry(rdev, &mddev->disks, same_set)
4493 list_for_each_entry(rdev2, &mddev->disks, same_set) {
4495 rdev->bdev->bd_contains ==
4496 rdev2->bdev->bd_contains) {
4498 "%s: WARNING: %s appears to be"
4499 " on the same physical disk as"
4502 bdevname(rdev->bdev,b),
4503 bdevname(rdev2->bdev,b2));
4510 "True protection against single-disk"
4511 " failure might be compromised.\n");
4514 mddev->recovery = 0;
4515 /* may be over-ridden by personality */
4516 mddev->resync_max_sectors = mddev->dev_sectors;
4518 mddev->ok_start_degraded = start_dirty_degraded;
4520 if (start_readonly && mddev->ro == 0)
4521 mddev->ro = 2; /* read-only, but switch on first write */
4523 err = mddev->pers->run(mddev);
4525 printk(KERN_ERR "md: pers->run() failed ...\n");
4526 else if (mddev->pers->size(mddev, 0, 0) < mddev->array_sectors) {
4527 WARN_ONCE(!mddev->external_size, "%s: default size too small,"
4528 " but 'external_size' not in effect?\n", __func__);
4530 "md: invalid array_size %llu > default size %llu\n",
4531 (unsigned long long)mddev->array_sectors / 2,
4532 (unsigned long long)mddev->pers->size(mddev, 0, 0) / 2);
4534 mddev->pers->stop(mddev);
4536 if (err == 0 && mddev->pers->sync_request) {
4537 err = bitmap_create(mddev);
4539 printk(KERN_ERR "%s: failed to create bitmap (%d)\n",
4540 mdname(mddev), err);
4541 mddev->pers->stop(mddev);
4545 module_put(mddev->pers->owner);
4547 bitmap_destroy(mddev);
4550 if (mddev->pers->sync_request) {
4551 if (mddev->kobj.sd &&
4552 sysfs_create_group(&mddev->kobj, &md_redundancy_group))
4554 "md: cannot register extra attributes for %s\n",
4556 mddev->sysfs_action = sysfs_get_dirent_safe(mddev->kobj.sd, "sync_action");
4557 } else if (mddev->ro == 2) /* auto-readonly not meaningful */
4560 atomic_set(&mddev->writes_pending,0);
4561 atomic_set(&mddev->max_corr_read_errors,
4562 MD_DEFAULT_MAX_CORRECTED_READ_ERRORS);
4563 mddev->safemode = 0;
4564 mddev->safemode_timer.function = md_safemode_timeout;
4565 mddev->safemode_timer.data = (unsigned long) mddev;
4566 mddev->safemode_delay = (200 * HZ)/1000 +1; /* 200 msec delay */
4569 list_for_each_entry(rdev, &mddev->disks, same_set)
4570 if (rdev->raid_disk >= 0) {
4572 sprintf(nm, "rd%d", rdev->raid_disk);
4573 if (sysfs_create_link(&mddev->kobj, &rdev->kobj, nm))
4574 /* failure here is OK */;
4577 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4580 md_update_sb(mddev, 0);
4582 md_wakeup_thread(mddev->thread);
4583 md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
4585 md_new_event(mddev);
4586 sysfs_notify_dirent_safe(mddev->sysfs_state);
4587 sysfs_notify_dirent_safe(mddev->sysfs_action);
4588 sysfs_notify(&mddev->kobj, NULL, "degraded");
4591 EXPORT_SYMBOL_GPL(md_run);
4593 static int do_md_run(mddev_t *mddev)
4597 err = md_run(mddev);
4600 err = bitmap_load(mddev);
4602 bitmap_destroy(mddev);
4605 set_capacity(mddev->gendisk, mddev->array_sectors);
4606 revalidate_disk(mddev->gendisk);
4607 kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
4612 static int restart_array(mddev_t *mddev)
4614 struct gendisk *disk = mddev->gendisk;
4616 /* Complain if it has no devices */
4617 if (list_empty(&mddev->disks))
4623 mddev->safemode = 0;
4625 set_disk_ro(disk, 0);
4626 printk(KERN_INFO "md: %s switched to read-write mode.\n",
4628 /* Kick recovery or resync if necessary */
4629 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4630 md_wakeup_thread(mddev->thread);
4631 md_wakeup_thread(mddev->sync_thread);
4632 sysfs_notify_dirent_safe(mddev->sysfs_state);
4636 /* similar to deny_write_access, but accounts for our holding a reference
4637 * to the file ourselves */
4638 static int deny_bitmap_write_access(struct file * file)
4640 struct inode *inode = file->f_mapping->host;
4642 spin_lock(&inode->i_lock);
4643 if (atomic_read(&inode->i_writecount) > 1) {
4644 spin_unlock(&inode->i_lock);
4647 atomic_set(&inode->i_writecount, -1);
4648 spin_unlock(&inode->i_lock);
4653 void restore_bitmap_write_access(struct file *file)
4655 struct inode *inode = file->f_mapping->host;
4657 spin_lock(&inode->i_lock);
4658 atomic_set(&inode->i_writecount, 1);
4659 spin_unlock(&inode->i_lock);
4662 static void md_clean(mddev_t *mddev)
4664 mddev->array_sectors = 0;
4665 mddev->external_size = 0;
4666 mddev->dev_sectors = 0;
4667 mddev->raid_disks = 0;
4668 mddev->recovery_cp = 0;
4669 mddev->resync_min = 0;
4670 mddev->resync_max = MaxSector;
4671 mddev->reshape_position = MaxSector;
4672 mddev->external = 0;
4673 mddev->persistent = 0;
4674 mddev->level = LEVEL_NONE;
4675 mddev->clevel[0] = 0;
4678 mddev->metadata_type[0] = 0;
4679 mddev->chunk_sectors = 0;
4680 mddev->ctime = mddev->utime = 0;
4682 mddev->max_disks = 0;
4684 mddev->can_decrease_events = 0;
4685 mddev->delta_disks = 0;
4686 mddev->new_level = LEVEL_NONE;
4687 mddev->new_layout = 0;
4688 mddev->new_chunk_sectors = 0;
4689 mddev->curr_resync = 0;
4690 mddev->resync_mismatches = 0;
4691 mddev->suspend_lo = mddev->suspend_hi = 0;
4692 mddev->sync_speed_min = mddev->sync_speed_max = 0;
4693 mddev->recovery = 0;
4695 mddev->degraded = 0;
4696 mddev->safemode = 0;
4697 mddev->bitmap_info.offset = 0;
4698 mddev->bitmap_info.default_offset = 0;
4699 mddev->bitmap_info.chunksize = 0;
4700 mddev->bitmap_info.daemon_sleep = 0;
4701 mddev->bitmap_info.max_write_behind = 0;
4705 void md_stop_writes(mddev_t *mddev)
4707 if (mddev->sync_thread) {
4708 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4709 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
4710 md_unregister_thread(mddev->sync_thread);
4711 mddev->sync_thread = NULL;
4714 del_timer_sync(&mddev->safemode_timer);
4716 bitmap_flush(mddev);
4717 md_super_wait(mddev);
4719 if (!mddev->in_sync || mddev->flags) {
4720 /* mark array as shutdown cleanly */
4722 md_update_sb(mddev, 1);
4725 EXPORT_SYMBOL_GPL(md_stop_writes);
4727 void md_stop(mddev_t *mddev)
4729 mddev->pers->stop(mddev);
4730 if (mddev->pers->sync_request && mddev->to_remove == NULL)
4731 mddev->to_remove = &md_redundancy_group;
4732 module_put(mddev->pers->owner);
4734 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4736 EXPORT_SYMBOL_GPL(md_stop);
4738 static int md_set_readonly(mddev_t *mddev, int is_open)
4741 mutex_lock(&mddev->open_mutex);
4742 if (atomic_read(&mddev->openers) > is_open) {
4743 printk("md: %s still in use.\n",mdname(mddev));
4748 md_stop_writes(mddev);
4754 set_disk_ro(mddev->gendisk, 1);
4755 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4756 sysfs_notify_dirent_safe(mddev->sysfs_state);
4760 mutex_unlock(&mddev->open_mutex);
4765 * 0 - completely stop and dis-assemble array
4766 * 2 - stop but do not disassemble array
4768 static int do_md_stop(mddev_t * mddev, int mode, int is_open)
4770 struct gendisk *disk = mddev->gendisk;
4773 mutex_lock(&mddev->open_mutex);
4774 if (atomic_read(&mddev->openers) > is_open ||
4775 mddev->sysfs_active) {
4776 printk("md: %s still in use.\n",mdname(mddev));
4777 mutex_unlock(&mddev->open_mutex);
4783 set_disk_ro(disk, 0);
4785 md_stop_writes(mddev);
4787 mddev->queue->merge_bvec_fn = NULL;
4788 mddev->queue->unplug_fn = NULL;
4789 mddev->queue->backing_dev_info.congested_fn = NULL;
4791 /* tell userspace to handle 'inactive' */
4792 sysfs_notify_dirent_safe(mddev->sysfs_state);
4794 list_for_each_entry(rdev, &mddev->disks, same_set)
4795 if (rdev->raid_disk >= 0) {
4797 sprintf(nm, "rd%d", rdev->raid_disk);
4798 sysfs_remove_link(&mddev->kobj, nm);
4801 set_capacity(disk, 0);
4802 mutex_unlock(&mddev->open_mutex);
4803 revalidate_disk(disk);
4808 mutex_unlock(&mddev->open_mutex);
4810 * Free resources if final stop
4813 printk(KERN_INFO "md: %s stopped.\n", mdname(mddev));
4815 bitmap_destroy(mddev);
4816 if (mddev->bitmap_info.file) {
4817 restore_bitmap_write_access(mddev->bitmap_info.file);
4818 fput(mddev->bitmap_info.file);
4819 mddev->bitmap_info.file = NULL;
4821 mddev->bitmap_info.offset = 0;
4823 export_array(mddev);
4826 kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
4827 if (mddev->hold_active == UNTIL_STOP)
4828 mddev->hold_active = 0;
4830 blk_integrity_unregister(disk);
4831 md_new_event(mddev);
4832 sysfs_notify_dirent_safe(mddev->sysfs_state);
4837 static void autorun_array(mddev_t *mddev)
4842 if (list_empty(&mddev->disks))
4845 printk(KERN_INFO "md: running: ");
4847 list_for_each_entry(rdev, &mddev->disks, same_set) {
4848 char b[BDEVNAME_SIZE];
4849 printk("<%s>", bdevname(rdev->bdev,b));
4853 err = do_md_run(mddev);
4855 printk(KERN_WARNING "md: do_md_run() returned %d\n", err);
4856 do_md_stop(mddev, 0, 0);
4861 * lets try to run arrays based on all disks that have arrived
4862 * until now. (those are in pending_raid_disks)
4864 * the method: pick the first pending disk, collect all disks with
4865 * the same UUID, remove all from the pending list and put them into
4866 * the 'same_array' list. Then order this list based on superblock
4867 * update time (freshest comes first), kick out 'old' disks and
4868 * compare superblocks. If everything's fine then run it.
4870 * If "unit" is allocated, then bump its reference count
4872 static void autorun_devices(int part)
4874 mdk_rdev_t *rdev0, *rdev, *tmp;
4876 char b[BDEVNAME_SIZE];
4878 printk(KERN_INFO "md: autorun ...\n");
4879 while (!list_empty(&pending_raid_disks)) {
4882 LIST_HEAD(candidates);
4883 rdev0 = list_entry(pending_raid_disks.next,
4884 mdk_rdev_t, same_set);
4886 printk(KERN_INFO "md: considering %s ...\n",
4887 bdevname(rdev0->bdev,b));
4888 INIT_LIST_HEAD(&candidates);
4889 rdev_for_each_list(rdev, tmp, &pending_raid_disks)
4890 if (super_90_load(rdev, rdev0, 0) >= 0) {
4891 printk(KERN_INFO "md: adding %s ...\n",
4892 bdevname(rdev->bdev,b));
4893 list_move(&rdev->same_set, &candidates);
4896 * now we have a set of devices, with all of them having
4897 * mostly sane superblocks. It's time to allocate the
4901 dev = MKDEV(mdp_major,
4902 rdev0->preferred_minor << MdpMinorShift);
4903 unit = MINOR(dev) >> MdpMinorShift;
4905 dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
4908 if (rdev0->preferred_minor != unit) {
4909 printk(KERN_INFO "md: unit number in %s is bad: %d\n",
4910 bdevname(rdev0->bdev, b), rdev0->preferred_minor);
4914 md_probe(dev, NULL, NULL);
4915 mddev = mddev_find(dev);
4916 if (!mddev || !mddev->gendisk) {
4920 "md: cannot allocate memory for md drive.\n");
4923 if (mddev_lock(mddev))
4924 printk(KERN_WARNING "md: %s locked, cannot run\n",
4926 else if (mddev->raid_disks || mddev->major_version
4927 || !list_empty(&mddev->disks)) {
4929 "md: %s already running, cannot run %s\n",
4930 mdname(mddev), bdevname(rdev0->bdev,b));
4931 mddev_unlock(mddev);
4933 printk(KERN_INFO "md: created %s\n", mdname(mddev));
4934 mddev->persistent = 1;
4935 rdev_for_each_list(rdev, tmp, &candidates) {
4936 list_del_init(&rdev->same_set);
4937 if (bind_rdev_to_array(rdev, mddev))
4940 autorun_array(mddev);
4941 mddev_unlock(mddev);
4943 /* on success, candidates will be empty, on error
4946 rdev_for_each_list(rdev, tmp, &candidates) {
4947 list_del_init(&rdev->same_set);
4952 printk(KERN_INFO "md: ... autorun DONE.\n");
4954 #endif /* !MODULE */
4956 static int get_version(void __user * arg)
4960 ver.major = MD_MAJOR_VERSION;
4961 ver.minor = MD_MINOR_VERSION;
4962 ver.patchlevel = MD_PATCHLEVEL_VERSION;
4964 if (copy_to_user(arg, &ver, sizeof(ver)))
4970 static int get_array_info(mddev_t * mddev, void __user * arg)
4972 mdu_array_info_t info;
4973 int nr,working,insync,failed,spare;
4976 nr=working=insync=failed=spare=0;
4977 list_for_each_entry(rdev, &mddev->disks, same_set) {
4979 if (test_bit(Faulty, &rdev->flags))
4983 if (test_bit(In_sync, &rdev->flags))
4990 info.major_version = mddev->major_version;
4991 info.minor_version = mddev->minor_version;
4992 info.patch_version = MD_PATCHLEVEL_VERSION;
4993 info.ctime = mddev->ctime;
4994 info.level = mddev->level;
4995 info.size = mddev->dev_sectors / 2;
4996 if (info.size != mddev->dev_sectors / 2) /* overflow */
4999 info.raid_disks = mddev->raid_disks;
5000 info.md_minor = mddev->md_minor;
5001 info.not_persistent= !mddev->persistent;
5003 info.utime = mddev->utime;
5006 info.state = (1<<MD_SB_CLEAN);
5007 if (mddev->bitmap && mddev->bitmap_info.offset)
5008 info.state = (1<<MD_SB_BITMAP_PRESENT);
5009 info.active_disks = insync;
5010 info.working_disks = working;
5011 info.failed_disks = failed;
5012 info.spare_disks = spare;
5014 info.layout = mddev->layout;
5015 info.chunk_size = mddev->chunk_sectors << 9;
5017 if (copy_to_user(arg, &info, sizeof(info)))
5023 static int get_bitmap_file(mddev_t * mddev, void __user * arg)
5025 mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */
5026 char *ptr, *buf = NULL;
5029 if (md_allow_write(mddev))
5030 file = kmalloc(sizeof(*file), GFP_NOIO);
5032 file = kmalloc(sizeof(*file), GFP_KERNEL);
5037 /* bitmap disabled, zero the first byte and copy out */
5038 if (!mddev->bitmap || !mddev->bitmap->file) {
5039 file->pathname[0] = '\0';
5043 buf = kmalloc(sizeof(file->pathname), GFP_KERNEL);
5047 ptr = d_path(&mddev->bitmap->file->f_path, buf, sizeof(file->pathname));
5051 strcpy(file->pathname, ptr);
5055 if (copy_to_user(arg, file, sizeof(*file)))
5063 static int get_disk_info(mddev_t * mddev, void __user * arg)
5065 mdu_disk_info_t info;
5068 if (copy_from_user(&info, arg, sizeof(info)))
5071 rdev = find_rdev_nr(mddev, info.number);
5073 info.major = MAJOR(rdev->bdev->bd_dev);
5074 info.minor = MINOR(rdev->bdev->bd_dev);
5075 info.raid_disk = rdev->raid_disk;
5077 if (test_bit(Faulty, &rdev->flags))
5078 info.state |= (1<<MD_DISK_FAULTY);
5079 else if (test_bit(In_sync, &rdev->flags)) {
5080 info.state |= (1<<MD_DISK_ACTIVE);
5081 info.state |= (1<<MD_DISK_SYNC);
5083 if (test_bit(WriteMostly, &rdev->flags))
5084 info.state |= (1<<MD_DISK_WRITEMOSTLY);
5086 info.major = info.minor = 0;
5087 info.raid_disk = -1;
5088 info.state = (1<<MD_DISK_REMOVED);
5091 if (copy_to_user(arg, &info, sizeof(info)))
5097 static int add_new_disk(mddev_t * mddev, mdu_disk_info_t *info)
5099 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
5101 dev_t dev = MKDEV(info->major,info->minor);
5103 if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
5106 if (!mddev->raid_disks) {
5108 /* expecting a device which has a superblock */
5109 rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
5112 "md: md_import_device returned %ld\n",
5114 return PTR_ERR(rdev);
5116 if (!list_empty(&mddev->disks)) {
5117 mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
5118 mdk_rdev_t, same_set);
5119 err = super_types[mddev->major_version]
5120 .load_super(rdev, rdev0, mddev->minor_version);
5123 "md: %s has different UUID to %s\n",
5124 bdevname(rdev->bdev,b),
5125 bdevname(rdev0->bdev,b2));
5130 err = bind_rdev_to_array(rdev, mddev);
5137 * add_new_disk can be used once the array is assembled
5138 * to add "hot spares". They must already have a superblock
5143 if (!mddev->pers->hot_add_disk) {
5145 "%s: personality does not support diskops!\n",
5149 if (mddev->persistent)
5150 rdev = md_import_device(dev, mddev->major_version,
5151 mddev->minor_version);
5153 rdev = md_import_device(dev, -1, -1);
5156 "md: md_import_device returned %ld\n",
5158 return PTR_ERR(rdev);
5160 /* set save_raid_disk if appropriate */
5161 if (!mddev->persistent) {
5162 if (info->state & (1<<MD_DISK_SYNC) &&
5163 info->raid_disk < mddev->raid_disks)
5164 rdev->raid_disk = info->raid_disk;
5166 rdev->raid_disk = -1;
5168 super_types[mddev->major_version].
5169 validate_super(mddev, rdev);
5170 rdev->saved_raid_disk = rdev->raid_disk;
5172 clear_bit(In_sync, &rdev->flags); /* just to be sure */
5173 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
5174 set_bit(WriteMostly, &rdev->flags);
5176 clear_bit(WriteMostly, &rdev->flags);
5178 rdev->raid_disk = -1;
5179 err = bind_rdev_to_array(rdev, mddev);
5180 if (!err && !mddev->pers->hot_remove_disk) {
5181 /* If there is hot_add_disk but no hot_remove_disk
5182 * then added disks for geometry changes,
5183 * and should be added immediately.
5185 super_types[mddev->major_version].
5186 validate_super(mddev, rdev);
5187 err = mddev->pers->hot_add_disk(mddev, rdev);
5189 unbind_rdev_from_array(rdev);
5194 sysfs_notify_dirent_safe(rdev->sysfs_state);
5196 md_update_sb(mddev, 1);
5197 if (mddev->degraded)
5198 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
5199 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5200 md_wakeup_thread(mddev->thread);
5204 /* otherwise, add_new_disk is only allowed
5205 * for major_version==0 superblocks
5207 if (mddev->major_version != 0) {
5208 printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n",
5213 if (!(info->state & (1<<MD_DISK_FAULTY))) {
5215 rdev = md_import_device(dev, -1, 0);
5218 "md: error, md_import_device() returned %ld\n",
5220 return PTR_ERR(rdev);
5222 rdev->desc_nr = info->number;
5223 if (info->raid_disk < mddev->raid_disks)
5224 rdev->raid_disk = info->raid_disk;
5226 rdev->raid_disk = -1;
5228 if (rdev->raid_disk < mddev->raid_disks)
5229 if (info->state & (1<<MD_DISK_SYNC))
5230 set_bit(In_sync, &rdev->flags);
5232 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
5233 set_bit(WriteMostly, &rdev->flags);
5235 if (!mddev->persistent) {
5236 printk(KERN_INFO "md: nonpersistent superblock ...\n");
5237 rdev->sb_start = i_size_read(rdev->bdev->bd_inode) / 512;
5239 rdev->sb_start = calc_dev_sboffset(rdev->bdev);
5240 rdev->sectors = rdev->sb_start;
5242 err = bind_rdev_to_array(rdev, mddev);
5252 static int hot_remove_disk(mddev_t * mddev, dev_t dev)
5254 char b[BDEVNAME_SIZE];
5257 rdev = find_rdev(mddev, dev);
5261 if (rdev->raid_disk >= 0)
5264 kick_rdev_from_array(rdev);
5265 md_update_sb(mddev, 1);
5266 md_new_event(mddev);
5270 printk(KERN_WARNING "md: cannot remove active disk %s from %s ...\n",
5271 bdevname(rdev->bdev,b), mdname(mddev));
5275 static int hot_add_disk(mddev_t * mddev, dev_t dev)
5277 char b[BDEVNAME_SIZE];
5284 if (mddev->major_version != 0) {
5285 printk(KERN_WARNING "%s: HOT_ADD may only be used with"
5286 " version-0 superblocks.\n",
5290 if (!mddev->pers->hot_add_disk) {
5292 "%s: personality does not support diskops!\n",
5297 rdev = md_import_device(dev, -1, 0);
5300 "md: error, md_import_device() returned %ld\n",
5305 if (mddev->persistent)
5306 rdev->sb_start = calc_dev_sboffset(rdev->bdev);
5308 rdev->sb_start = i_size_read(rdev->bdev->bd_inode) / 512;
5310 rdev->sectors = rdev->sb_start;
5312 if (test_bit(Faulty, &rdev->flags)) {
5314 "md: can not hot-add faulty %s disk to %s!\n",
5315 bdevname(rdev->bdev,b), mdname(mddev));
5319 clear_bit(In_sync, &rdev->flags);
5321 rdev->saved_raid_disk = -1;
5322 err = bind_rdev_to_array(rdev, mddev);
5327 * The rest should better be atomic, we can have disk failures
5328 * noticed in interrupt contexts ...
5331 rdev->raid_disk = -1;
5333 md_update_sb(mddev, 1);
5336 * Kick recovery, maybe this spare has to be added to the
5337 * array immediately.
5339 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5340 md_wakeup_thread(mddev->thread);
5341 md_new_event(mddev);
5349 static int set_bitmap_file(mddev_t *mddev, int fd)
5354 if (!mddev->pers->quiesce)
5356 if (mddev->recovery || mddev->sync_thread)
5358 /* we should be able to change the bitmap.. */
5364 return -EEXIST; /* cannot add when bitmap is present */
5365 mddev->bitmap_info.file = fget(fd);
5367 if (mddev->bitmap_info.file == NULL) {
5368 printk(KERN_ERR "%s: error: failed to get bitmap file\n",
5373 err = deny_bitmap_write_access(mddev->bitmap_info.file);
5375 printk(KERN_ERR "%s: error: bitmap file is already in use\n",
5377 fput(mddev->bitmap_info.file);
5378 mddev->bitmap_info.file = NULL;
5381 mddev->bitmap_info.offset = 0; /* file overrides offset */
5382 } else if (mddev->bitmap == NULL)
5383 return -ENOENT; /* cannot remove what isn't there */
5386 mddev->pers->quiesce(mddev, 1);
5388 err = bitmap_create(mddev);
5390 err = bitmap_load(mddev);
5392 if (fd < 0 || err) {
5393 bitmap_destroy(mddev);
5394 fd = -1; /* make sure to put the file */
5396 mddev->pers->quiesce(mddev, 0);
5399 if (mddev->bitmap_info.file) {
5400 restore_bitmap_write_access(mddev->bitmap_info.file);
5401 fput(mddev->bitmap_info.file);
5403 mddev->bitmap_info.file = NULL;
5410 * set_array_info is used two different ways
5411 * The original usage is when creating a new array.
5412 * In this usage, raid_disks is > 0 and it together with
5413 * level, size, not_persistent,layout,chunksize determine the
5414 * shape of the array.
5415 * This will always create an array with a type-0.90.0 superblock.
5416 * The newer usage is when assembling an array.
5417 * In this case raid_disks will be 0, and the major_version field is
5418 * use to determine which style super-blocks are to be found on the devices.
5419 * The minor and patch _version numbers are also kept incase the
5420 * super_block handler wishes to interpret them.
5422 static int set_array_info(mddev_t * mddev, mdu_array_info_t *info)
5425 if (info->raid_disks == 0) {
5426 /* just setting version number for superblock loading */
5427 if (info->major_version < 0 ||
5428 info->major_version >= ARRAY_SIZE(super_types) ||
5429 super_types[info->major_version].name == NULL) {
5430 /* maybe try to auto-load a module? */
5432 "md: superblock version %d not known\n",
5433 info->major_version);
5436 mddev->major_version = info->major_version;
5437 mddev->minor_version = info->minor_version;
5438 mddev->patch_version = info->patch_version;
5439 mddev->persistent = !info->not_persistent;
5440 /* ensure mddev_put doesn't delete this now that there
5441 * is some minimal configuration.
5443 mddev->ctime = get_seconds();
5446 mddev->major_version = MD_MAJOR_VERSION;
5447 mddev->minor_version = MD_MINOR_VERSION;
5448 mddev->patch_version = MD_PATCHLEVEL_VERSION;
5449 mddev->ctime = get_seconds();
5451 mddev->level = info->level;
5452 mddev->clevel[0] = 0;
5453 mddev->dev_sectors = 2 * (sector_t)info->size;
5454 mddev->raid_disks = info->raid_disks;
5455 /* don't set md_minor, it is determined by which /dev/md* was
5458 if (info->state & (1<<MD_SB_CLEAN))
5459 mddev->recovery_cp = MaxSector;
5461 mddev->recovery_cp = 0;
5462 mddev->persistent = ! info->not_persistent;
5463 mddev->external = 0;
5465 mddev->layout = info->layout;
5466 mddev->chunk_sectors = info->chunk_size >> 9;
5468 mddev->max_disks = MD_SB_DISKS;
5470 if (mddev->persistent)
5472 set_bit(MD_CHANGE_DEVS, &mddev->flags);
5474 mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9;
5475 mddev->bitmap_info.offset = 0;
5477 mddev->reshape_position = MaxSector;
5480 * Generate a 128 bit UUID
5482 get_random_bytes(mddev->uuid, 16);
5484 mddev->new_level = mddev->level;
5485 mddev->new_chunk_sectors = mddev->chunk_sectors;
5486 mddev->new_layout = mddev->layout;
5487 mddev->delta_disks = 0;
5492 void md_set_array_sectors(mddev_t *mddev, sector_t array_sectors)
5494 WARN(!mddev_is_locked(mddev), "%s: unlocked mddev!\n", __func__);
5496 if (mddev->external_size)
5499 mddev->array_sectors = array_sectors;
5501 EXPORT_SYMBOL(md_set_array_sectors);
5503 static int update_size(mddev_t *mddev, sector_t num_sectors)
5507 int fit = (num_sectors == 0);
5509 if (mddev->pers->resize == NULL)
5511 /* The "num_sectors" is the number of sectors of each device that
5512 * is used. This can only make sense for arrays with redundancy.
5513 * linear and raid0 always use whatever space is available. We can only
5514 * consider changing this number if no resync or reconstruction is
5515 * happening, and if the new size is acceptable. It must fit before the
5516 * sb_start or, if that is <data_offset, it must fit before the size
5517 * of each device. If num_sectors is zero, we find the largest size
5521 if (mddev->sync_thread)
5524 /* Sorry, cannot grow a bitmap yet, just remove it,
5528 list_for_each_entry(rdev, &mddev->disks, same_set) {
5529 sector_t avail = rdev->sectors;
5531 if (fit && (num_sectors == 0 || num_sectors > avail))
5532 num_sectors = avail;
5533 if (avail < num_sectors)
5536 rv = mddev->pers->resize(mddev, num_sectors);
5538 revalidate_disk(mddev->gendisk);
5542 static int update_raid_disks(mddev_t *mddev, int raid_disks)
5545 /* change the number of raid disks */
5546 if (mddev->pers->check_reshape == NULL)
5548 if (raid_disks <= 0 ||
5549 (mddev->max_disks && raid_disks >= mddev->max_disks))
5551 if (mddev->sync_thread || mddev->reshape_position != MaxSector)
5553 mddev->delta_disks = raid_disks - mddev->raid_disks;
5555 rv = mddev->pers->check_reshape(mddev);
5561 * update_array_info is used to change the configuration of an
5563 * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
5564 * fields in the info are checked against the array.
5565 * Any differences that cannot be handled will cause an error.
5566 * Normally, only one change can be managed at a time.
5568 static int update_array_info(mddev_t *mddev, mdu_array_info_t *info)
5574 /* calculate expected state,ignoring low bits */
5575 if (mddev->bitmap && mddev->bitmap_info.offset)
5576 state |= (1 << MD_SB_BITMAP_PRESENT);
5578 if (mddev->major_version != info->major_version ||
5579 mddev->minor_version != info->minor_version ||
5580 /* mddev->patch_version != info->patch_version || */
5581 mddev->ctime != info->ctime ||
5582 mddev->level != info->level ||
5583 /* mddev->layout != info->layout || */
5584 !mddev->persistent != info->not_persistent||
5585 mddev->chunk_sectors != info->chunk_size >> 9 ||
5586 /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
5587 ((state^info->state) & 0xfffffe00)
5590 /* Check there is only one change */
5591 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
5593 if (mddev->raid_disks != info->raid_disks)
5595 if (mddev->layout != info->layout)
5597 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT))
5604 if (mddev->layout != info->layout) {
5606 * we don't need to do anything at the md level, the
5607 * personality will take care of it all.
5609 if (mddev->pers->check_reshape == NULL)
5612 mddev->new_layout = info->layout;
5613 rv = mddev->pers->check_reshape(mddev);
5615 mddev->new_layout = mddev->layout;
5619 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
5620 rv = update_size(mddev, (sector_t)info->size * 2);
5622 if (mddev->raid_disks != info->raid_disks)
5623 rv = update_raid_disks(mddev, info->raid_disks);
5625 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) {
5626 if (mddev->pers->quiesce == NULL)
5628 if (mddev->recovery || mddev->sync_thread)
5630 if (info->state & (1<<MD_SB_BITMAP_PRESENT)) {
5631 /* add the bitmap */
5634 if (mddev->bitmap_info.default_offset == 0)
5636 mddev->bitmap_info.offset =
5637 mddev->bitmap_info.default_offset;
5638 mddev->pers->quiesce(mddev, 1);
5639 rv = bitmap_create(mddev);
5641 rv = bitmap_load(mddev);
5643 bitmap_destroy(mddev);
5644 mddev->pers->quiesce(mddev, 0);
5646 /* remove the bitmap */
5649 if (mddev->bitmap->file)
5651 mddev->pers->quiesce(mddev, 1);
5652 bitmap_destroy(mddev);
5653 mddev->pers->quiesce(mddev, 0);
5654 mddev->bitmap_info.offset = 0;
5657 md_update_sb(mddev, 1);
5661 static int set_disk_faulty(mddev_t *mddev, dev_t dev)
5665 if (mddev->pers == NULL)
5668 rdev = find_rdev(mddev, dev);
5672 md_error(mddev, rdev);
5677 * We have a problem here : there is no easy way to give a CHS
5678 * virtual geometry. We currently pretend that we have a 2 heads
5679 * 4 sectors (with a BIG number of cylinders...). This drives
5680 * dosfs just mad... ;-)
5682 static int md_getgeo(struct block_device *bdev, struct hd_geometry *geo)
5684 mddev_t *mddev = bdev->bd_disk->private_data;
5688 geo->cylinders = mddev->array_sectors / 8;
5692 static int md_ioctl(struct block_device *bdev, fmode_t mode,
5693 unsigned int cmd, unsigned long arg)
5696 void __user *argp = (void __user *)arg;
5697 mddev_t *mddev = NULL;
5700 if (!capable(CAP_SYS_ADMIN))
5704 * Commands dealing with the RAID driver but not any
5710 err = get_version(argp);
5713 case PRINT_RAID_DEBUG:
5721 autostart_arrays(arg);
5728 * Commands creating/starting a new array:
5731 mddev = bdev->bd_disk->private_data;
5738 err = mddev_lock(mddev);
5741 "md: ioctl lock interrupted, reason %d, cmd %d\n",
5748 case SET_ARRAY_INFO:
5750 mdu_array_info_t info;
5752 memset(&info, 0, sizeof(info));
5753 else if (copy_from_user(&info, argp, sizeof(info))) {
5758 err = update_array_info(mddev, &info);
5760 printk(KERN_WARNING "md: couldn't update"
5761 " array info. %d\n", err);
5766 if (!list_empty(&mddev->disks)) {
5768 "md: array %s already has disks!\n",
5773 if (mddev->raid_disks) {
5775 "md: array %s already initialised!\n",
5780 err = set_array_info(mddev, &info);
5782 printk(KERN_WARNING "md: couldn't set"
5783 " array info. %d\n", err);
5793 * Commands querying/configuring an existing array:
5795 /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
5796 * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */
5797 if ((!mddev->raid_disks && !mddev->external)
5798 && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY
5799 && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE
5800 && cmd != GET_BITMAP_FILE) {
5806 * Commands even a read-only array can execute:
5810 case GET_ARRAY_INFO:
5811 err = get_array_info(mddev, argp);
5814 case GET_BITMAP_FILE:
5815 err = get_bitmap_file(mddev, argp);
5819 err = get_disk_info(mddev, argp);
5822 case RESTART_ARRAY_RW:
5823 err = restart_array(mddev);
5827 err = do_md_stop(mddev, 0, 1);
5831 err = md_set_readonly(mddev, 1);
5835 if (get_user(ro, (int __user *)(arg))) {
5841 /* if the bdev is going readonly the value of mddev->ro
5842 * does not matter, no writes are coming
5847 /* are we are already prepared for writes? */
5851 /* transitioning to readauto need only happen for
5852 * arrays that call md_write_start
5855 err = restart_array(mddev);
5858 set_disk_ro(mddev->gendisk, 0);
5865 * The remaining ioctls are changing the state of the
5866 * superblock, so we do not allow them on read-only arrays.
5867 * However non-MD ioctls (e.g. get-size) will still come through
5868 * here and hit the 'default' below, so only disallow
5869 * 'md' ioctls, and switch to rw mode if started auto-readonly.
5871 if (_IOC_TYPE(cmd) == MD_MAJOR && mddev->ro && mddev->pers) {
5872 if (mddev->ro == 2) {
5874 sysfs_notify_dirent_safe(mddev->sysfs_state);
5875 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5876 md_wakeup_thread(mddev->thread);
5887 mdu_disk_info_t info;
5888 if (copy_from_user(&info, argp, sizeof(info)))
5891 err = add_new_disk(mddev, &info);
5895 case HOT_REMOVE_DISK:
5896 err = hot_remove_disk(mddev, new_decode_dev(arg));
5900 err = hot_add_disk(mddev, new_decode_dev(arg));
5903 case SET_DISK_FAULTY:
5904 err = set_disk_faulty(mddev, new_decode_dev(arg));
5908 err = do_md_run(mddev);
5911 case SET_BITMAP_FILE:
5912 err = set_bitmap_file(mddev, (int)arg);
5922 if (mddev->hold_active == UNTIL_IOCTL &&
5924 mddev->hold_active = 0;
5925 mddev_unlock(mddev);
5934 #ifdef CONFIG_COMPAT
5935 static int md_compat_ioctl(struct block_device *bdev, fmode_t mode,
5936 unsigned int cmd, unsigned long arg)
5939 case HOT_REMOVE_DISK:
5941 case SET_DISK_FAULTY:
5942 case SET_BITMAP_FILE:
5943 /* These take in integer arg, do not convert */
5946 arg = (unsigned long)compat_ptr(arg);
5950 return md_ioctl(bdev, mode, cmd, arg);
5952 #endif /* CONFIG_COMPAT */
5954 static int md_open(struct block_device *bdev, fmode_t mode)
5957 * Succeed if we can lock the mddev, which confirms that
5958 * it isn't being stopped right now.
5960 mddev_t *mddev = mddev_find(bdev->bd_dev);
5963 if (mddev->gendisk != bdev->bd_disk) {
5964 /* we are racing with mddev_put which is discarding this
5968 /* Wait until bdev->bd_disk is definitely gone */
5969 flush_workqueue(md_misc_wq);
5970 /* Then retry the open from the top */
5971 return -ERESTARTSYS;
5973 BUG_ON(mddev != bdev->bd_disk->private_data);
5975 if ((err = mutex_lock_interruptible(&mddev->open_mutex)))
5979 atomic_inc(&mddev->openers);
5980 mutex_unlock(&mddev->open_mutex);
5982 check_disk_size_change(mddev->gendisk, bdev);
5987 static int md_release(struct gendisk *disk, fmode_t mode)
5989 mddev_t *mddev = disk->private_data;
5992 atomic_dec(&mddev->openers);
5997 static const struct block_device_operations md_fops =
5999 .owner = THIS_MODULE,
6001 .release = md_release,
6003 #ifdef CONFIG_COMPAT
6004 .compat_ioctl = md_compat_ioctl,
6006 .getgeo = md_getgeo,
6009 static int md_thread(void * arg)
6011 mdk_thread_t *thread = arg;
6014 * md_thread is a 'system-thread', it's priority should be very
6015 * high. We avoid resource deadlocks individually in each
6016 * raid personality. (RAID5 does preallocation) We also use RR and
6017 * the very same RT priority as kswapd, thus we will never get
6018 * into a priority inversion deadlock.
6020 * we definitely have to have equal or higher priority than
6021 * bdflush, otherwise bdflush will deadlock if there are too
6022 * many dirty RAID5 blocks.
6025 allow_signal(SIGKILL);
6026 while (!kthread_should_stop()) {
6028 /* We need to wait INTERRUPTIBLE so that
6029 * we don't add to the load-average.
6030 * That means we need to be sure no signals are
6033 if (signal_pending(current))
6034 flush_signals(current);
6036 wait_event_interruptible_timeout
6038 test_bit(THREAD_WAKEUP, &thread->flags)
6039 || kthread_should_stop(),
6042 clear_bit(THREAD_WAKEUP, &thread->flags);
6044 thread->run(thread->mddev);
6050 void md_wakeup_thread(mdk_thread_t *thread)
6053 dprintk("md: waking up MD thread %s.\n", thread->tsk->comm);
6054 set_bit(THREAD_WAKEUP, &thread->flags);
6055 wake_up(&thread->wqueue);
6059 mdk_thread_t *md_register_thread(void (*run) (mddev_t *), mddev_t *mddev,
6062 mdk_thread_t *thread;
6064 thread = kzalloc(sizeof(mdk_thread_t), GFP_KERNEL);
6068 init_waitqueue_head(&thread->wqueue);
6071 thread->mddev = mddev;
6072 thread->timeout = MAX_SCHEDULE_TIMEOUT;
6073 thread->tsk = kthread_run(md_thread, thread,
6075 mdname(thread->mddev),
6076 name ?: mddev->pers->name);
6077 if (IS_ERR(thread->tsk)) {
6084 void md_unregister_thread(mdk_thread_t *thread)
6088 dprintk("interrupting MD-thread pid %d\n", task_pid_nr(thread->tsk));
6090 kthread_stop(thread->tsk);
6094 void md_error(mddev_t *mddev, mdk_rdev_t *rdev)
6101 if (!rdev || test_bit(Faulty, &rdev->flags))
6104 if (mddev->external)
6105 set_bit(Blocked, &rdev->flags);
6107 dprintk("md_error dev:%s, rdev:(%d:%d), (caller: %p,%p,%p,%p).\n",
6109 MAJOR(rdev->bdev->bd_dev), MINOR(rdev->bdev->bd_dev),
6110 __builtin_return_address(0),__builtin_return_address(1),
6111 __builtin_return_address(2),__builtin_return_address(3));
6115 if (!mddev->pers->error_handler)
6117 mddev->pers->error_handler(mddev,rdev);
6118 if (mddev->degraded)
6119 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
6120 sysfs_notify_dirent_safe(rdev->sysfs_state);
6121 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6122 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6123 md_wakeup_thread(mddev->thread);
6124 if (mddev->event_work.func)
6125 queue_work(md_misc_wq, &mddev->event_work);
6126 md_new_event_inintr(mddev);
6129 /* seq_file implementation /proc/mdstat */
6131 static void status_unused(struct seq_file *seq)
6136 seq_printf(seq, "unused devices: ");
6138 list_for_each_entry(rdev, &pending_raid_disks, same_set) {
6139 char b[BDEVNAME_SIZE];
6141 seq_printf(seq, "%s ",
6142 bdevname(rdev->bdev,b));
6145 seq_printf(seq, "<none>");
6147 seq_printf(seq, "\n");
6151 static void status_resync(struct seq_file *seq, mddev_t * mddev)
6153 sector_t max_sectors, resync, res;
6154 unsigned long dt, db;
6157 unsigned int per_milli;
6159 resync = mddev->curr_resync - atomic_read(&mddev->recovery_active);
6161 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
6162 max_sectors = mddev->resync_max_sectors;
6164 max_sectors = mddev->dev_sectors;
6167 * Should not happen.
6173 /* Pick 'scale' such that (resync>>scale)*1000 will fit
6174 * in a sector_t, and (max_sectors>>scale) will fit in a
6175 * u32, as those are the requirements for sector_div.
6176 * Thus 'scale' must be at least 10
6179 if (sizeof(sector_t) > sizeof(unsigned long)) {
6180 while ( max_sectors/2 > (1ULL<<(scale+32)))
6183 res = (resync>>scale)*1000;
6184 sector_div(res, (u32)((max_sectors>>scale)+1));
6188 int i, x = per_milli/50, y = 20-x;
6189 seq_printf(seq, "[");
6190 for (i = 0; i < x; i++)
6191 seq_printf(seq, "=");
6192 seq_printf(seq, ">");
6193 for (i = 0; i < y; i++)
6194 seq_printf(seq, ".");
6195 seq_printf(seq, "] ");
6197 seq_printf(seq, " %s =%3u.%u%% (%llu/%llu)",
6198 (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)?
6200 (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)?
6202 (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
6203 "resync" : "recovery"))),
6204 per_milli/10, per_milli % 10,
6205 (unsigned long long) resync/2,
6206 (unsigned long long) max_sectors/2);
6209 * dt: time from mark until now
6210 * db: blocks written from mark until now
6211 * rt: remaining time
6213 * rt is a sector_t, so could be 32bit or 64bit.
6214 * So we divide before multiply in case it is 32bit and close
6216 * We scale the divisor (db) by 32 to avoid loosing precision
6217 * near the end of resync when the number of remaining sectors
6219 * We then divide rt by 32 after multiplying by db to compensate.
6220 * The '+1' avoids division by zero if db is very small.
6222 dt = ((jiffies - mddev->resync_mark) / HZ);
6224 db = (mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active))
6225 - mddev->resync_mark_cnt;
6227 rt = max_sectors - resync; /* number of remaining sectors */
6228 sector_div(rt, db/32+1);
6232 seq_printf(seq, " finish=%lu.%lumin", (unsigned long)rt / 60,
6233 ((unsigned long)rt % 60)/6);
6235 seq_printf(seq, " speed=%ldK/sec", db/2/dt);
6238 static void *md_seq_start(struct seq_file *seq, loff_t *pos)
6240 struct list_head *tmp;
6250 spin_lock(&all_mddevs_lock);
6251 list_for_each(tmp,&all_mddevs)
6253 mddev = list_entry(tmp, mddev_t, all_mddevs);
6255 spin_unlock(&all_mddevs_lock);
6258 spin_unlock(&all_mddevs_lock);
6260 return (void*)2;/* tail */
6264 static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
6266 struct list_head *tmp;
6267 mddev_t *next_mddev, *mddev = v;
6273 spin_lock(&all_mddevs_lock);
6275 tmp = all_mddevs.next;
6277 tmp = mddev->all_mddevs.next;
6278 if (tmp != &all_mddevs)
6279 next_mddev = mddev_get(list_entry(tmp,mddev_t,all_mddevs));
6281 next_mddev = (void*)2;
6284 spin_unlock(&all_mddevs_lock);
6292 static void md_seq_stop(struct seq_file *seq, void *v)
6296 if (mddev && v != (void*)1 && v != (void*)2)
6300 struct mdstat_info {
6304 static int md_seq_show(struct seq_file *seq, void *v)
6309 struct mdstat_info *mi = seq->private;
6310 struct bitmap *bitmap;
6312 if (v == (void*)1) {
6313 struct mdk_personality *pers;
6314 seq_printf(seq, "Personalities : ");
6315 spin_lock(&pers_lock);
6316 list_for_each_entry(pers, &pers_list, list)
6317 seq_printf(seq, "[%s] ", pers->name);
6319 spin_unlock(&pers_lock);
6320 seq_printf(seq, "\n");
6321 mi->event = atomic_read(&md_event_count);
6324 if (v == (void*)2) {
6329 if (mddev_lock(mddev) < 0)
6332 if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
6333 seq_printf(seq, "%s : %sactive", mdname(mddev),
6334 mddev->pers ? "" : "in");
6337 seq_printf(seq, " (read-only)");
6339 seq_printf(seq, " (auto-read-only)");
6340 seq_printf(seq, " %s", mddev->pers->name);
6344 list_for_each_entry(rdev, &mddev->disks, same_set) {
6345 char b[BDEVNAME_SIZE];
6346 seq_printf(seq, " %s[%d]",
6347 bdevname(rdev->bdev,b), rdev->desc_nr);
6348 if (test_bit(WriteMostly, &rdev->flags))
6349 seq_printf(seq, "(W)");
6350 if (test_bit(Faulty, &rdev->flags)) {
6351 seq_printf(seq, "(F)");
6353 } else if (rdev->raid_disk < 0)
6354 seq_printf(seq, "(S)"); /* spare */
6355 sectors += rdev->sectors;
6358 if (!list_empty(&mddev->disks)) {
6360 seq_printf(seq, "\n %llu blocks",
6361 (unsigned long long)
6362 mddev->array_sectors / 2);
6364 seq_printf(seq, "\n %llu blocks",
6365 (unsigned long long)sectors / 2);
6367 if (mddev->persistent) {
6368 if (mddev->major_version != 0 ||
6369 mddev->minor_version != 90) {
6370 seq_printf(seq," super %d.%d",
6371 mddev->major_version,
6372 mddev->minor_version);
6374 } else if (mddev->external)
6375 seq_printf(seq, " super external:%s",
6376 mddev->metadata_type);
6378 seq_printf(seq, " super non-persistent");
6381 mddev->pers->status(seq, mddev);
6382 seq_printf(seq, "\n ");
6383 if (mddev->pers->sync_request) {
6384 if (mddev->curr_resync > 2) {
6385 status_resync(seq, mddev);
6386 seq_printf(seq, "\n ");
6387 } else if (mddev->curr_resync == 1 || mddev->curr_resync == 2)
6388 seq_printf(seq, "\tresync=DELAYED\n ");
6389 else if (mddev->recovery_cp < MaxSector)
6390 seq_printf(seq, "\tresync=PENDING\n ");
6393 seq_printf(seq, "\n ");
6395 if ((bitmap = mddev->bitmap)) {
6396 unsigned long chunk_kb;
6397 unsigned long flags;
6398 spin_lock_irqsave(&bitmap->lock, flags);
6399 chunk_kb = mddev->bitmap_info.chunksize >> 10;
6400 seq_printf(seq, "bitmap: %lu/%lu pages [%luKB], "
6402 bitmap->pages - bitmap->missing_pages,
6404 (bitmap->pages - bitmap->missing_pages)
6405 << (PAGE_SHIFT - 10),
6406 chunk_kb ? chunk_kb : mddev->bitmap_info.chunksize,
6407 chunk_kb ? "KB" : "B");
6409 seq_printf(seq, ", file: ");
6410 seq_path(seq, &bitmap->file->f_path, " \t\n");
6413 seq_printf(seq, "\n");
6414 spin_unlock_irqrestore(&bitmap->lock, flags);
6417 seq_printf(seq, "\n");
6419 mddev_unlock(mddev);
6424 static const struct seq_operations md_seq_ops = {
6425 .start = md_seq_start,
6426 .next = md_seq_next,
6427 .stop = md_seq_stop,
6428 .show = md_seq_show,
6431 static int md_seq_open(struct inode *inode, struct file *file)
6434 struct mdstat_info *mi = kmalloc(sizeof(*mi), GFP_KERNEL);
6438 error = seq_open(file, &md_seq_ops);
6442 struct seq_file *p = file->private_data;
6444 mi->event = atomic_read(&md_event_count);
6449 static unsigned int mdstat_poll(struct file *filp, poll_table *wait)
6451 struct seq_file *m = filp->private_data;
6452 struct mdstat_info *mi = m->private;
6455 poll_wait(filp, &md_event_waiters, wait);
6457 /* always allow read */
6458 mask = POLLIN | POLLRDNORM;
6460 if (mi->event != atomic_read(&md_event_count))
6461 mask |= POLLERR | POLLPRI;
6465 static const struct file_operations md_seq_fops = {
6466 .owner = THIS_MODULE,
6467 .open = md_seq_open,
6469 .llseek = seq_lseek,
6470 .release = seq_release_private,
6471 .poll = mdstat_poll,
6474 int register_md_personality(struct mdk_personality *p)
6476 spin_lock(&pers_lock);
6477 list_add_tail(&p->list, &pers_list);
6478 printk(KERN_INFO "md: %s personality registered for level %d\n", p->name, p->level);
6479 spin_unlock(&pers_lock);
6483 int unregister_md_personality(struct mdk_personality *p)
6485 printk(KERN_INFO "md: %s personality unregistered\n", p->name);
6486 spin_lock(&pers_lock);
6487 list_del_init(&p->list);
6488 spin_unlock(&pers_lock);
6492 static int is_mddev_idle(mddev_t *mddev, int init)
6500 rdev_for_each_rcu(rdev, mddev) {
6501 struct gendisk *disk = rdev->bdev->bd_contains->bd_disk;
6502 curr_events = (int)part_stat_read(&disk->part0, sectors[0]) +
6503 (int)part_stat_read(&disk->part0, sectors[1]) -
6504 atomic_read(&disk->sync_io);
6505 /* sync IO will cause sync_io to increase before the disk_stats
6506 * as sync_io is counted when a request starts, and
6507 * disk_stats is counted when it completes.
6508 * So resync activity will cause curr_events to be smaller than
6509 * when there was no such activity.
6510 * non-sync IO will cause disk_stat to increase without
6511 * increasing sync_io so curr_events will (eventually)
6512 * be larger than it was before. Once it becomes
6513 * substantially larger, the test below will cause
6514 * the array to appear non-idle, and resync will slow
6516 * If there is a lot of outstanding resync activity when
6517 * we set last_event to curr_events, then all that activity
6518 * completing might cause the array to appear non-idle
6519 * and resync will be slowed down even though there might
6520 * not have been non-resync activity. This will only
6521 * happen once though. 'last_events' will soon reflect
6522 * the state where there is little or no outstanding
6523 * resync requests, and further resync activity will
6524 * always make curr_events less than last_events.
6527 if (init || curr_events - rdev->last_events > 64) {
6528 rdev->last_events = curr_events;
6536 void md_done_sync(mddev_t *mddev, int blocks, int ok)
6538 /* another "blocks" (512byte) blocks have been synced */
6539 atomic_sub(blocks, &mddev->recovery_active);
6540 wake_up(&mddev->recovery_wait);
6542 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6543 md_wakeup_thread(mddev->thread);
6544 // stop recovery, signal do_sync ....
6549 /* md_write_start(mddev, bi)
6550 * If we need to update some array metadata (e.g. 'active' flag
6551 * in superblock) before writing, schedule a superblock update
6552 * and wait for it to complete.
6554 void md_write_start(mddev_t *mddev, struct bio *bi)
6557 if (bio_data_dir(bi) != WRITE)
6560 BUG_ON(mddev->ro == 1);
6561 if (mddev->ro == 2) {
6562 /* need to switch to read/write */
6564 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6565 md_wakeup_thread(mddev->thread);
6566 md_wakeup_thread(mddev->sync_thread);
6569 atomic_inc(&mddev->writes_pending);
6570 if (mddev->safemode == 1)
6571 mddev->safemode = 0;
6572 if (mddev->in_sync) {
6573 spin_lock_irq(&mddev->write_lock);
6574 if (mddev->in_sync) {
6576 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
6577 set_bit(MD_CHANGE_PENDING, &mddev->flags);
6578 md_wakeup_thread(mddev->thread);
6581 spin_unlock_irq(&mddev->write_lock);
6584 sysfs_notify_dirent_safe(mddev->sysfs_state);
6585 wait_event(mddev->sb_wait,
6586 !test_bit(MD_CHANGE_PENDING, &mddev->flags));
6589 void md_write_end(mddev_t *mddev)
6591 if (atomic_dec_and_test(&mddev->writes_pending)) {
6592 if (mddev->safemode == 2)
6593 md_wakeup_thread(mddev->thread);
6594 else if (mddev->safemode_delay)
6595 mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay);
6599 /* md_allow_write(mddev)
6600 * Calling this ensures that the array is marked 'active' so that writes
6601 * may proceed without blocking. It is important to call this before
6602 * attempting a GFP_KERNEL allocation while holding the mddev lock.
6603 * Must be called with mddev_lock held.
6605 * In the ->external case MD_CHANGE_CLEAN can not be cleared until mddev->lock
6606 * is dropped, so return -EAGAIN after notifying userspace.
6608 int md_allow_write(mddev_t *mddev)
6614 if (!mddev->pers->sync_request)
6617 spin_lock_irq(&mddev->write_lock);
6618 if (mddev->in_sync) {
6620 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
6621 set_bit(MD_CHANGE_PENDING, &mddev->flags);
6622 if (mddev->safemode_delay &&
6623 mddev->safemode == 0)
6624 mddev->safemode = 1;
6625 spin_unlock_irq(&mddev->write_lock);
6626 md_update_sb(mddev, 0);
6627 sysfs_notify_dirent_safe(mddev->sysfs_state);
6629 spin_unlock_irq(&mddev->write_lock);
6631 if (test_bit(MD_CHANGE_PENDING, &mddev->flags))
6636 EXPORT_SYMBOL_GPL(md_allow_write);
6638 void md_unplug(mddev_t *mddev)
6641 blk_unplug(mddev->queue);
6643 mddev->plug->unplug_fn(mddev->plug);
6646 #define SYNC_MARKS 10
6647 #define SYNC_MARK_STEP (3*HZ)
6648 void md_do_sync(mddev_t *mddev)
6651 unsigned int currspeed = 0,
6653 sector_t max_sectors,j, io_sectors;
6654 unsigned long mark[SYNC_MARKS];
6655 sector_t mark_cnt[SYNC_MARKS];
6657 struct list_head *tmp;
6658 sector_t last_check;
6663 /* just incase thread restarts... */
6664 if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
6666 if (mddev->ro) /* never try to sync a read-only array */
6669 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
6670 if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
6671 desc = "data-check";
6672 else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
6673 desc = "requested-resync";
6676 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
6681 /* we overload curr_resync somewhat here.
6682 * 0 == not engaged in resync at all
6683 * 2 == checking that there is no conflict with another sync
6684 * 1 == like 2, but have yielded to allow conflicting resync to
6686 * other == active in resync - this many blocks
6688 * Before starting a resync we must have set curr_resync to
6689 * 2, and then checked that every "conflicting" array has curr_resync
6690 * less than ours. When we find one that is the same or higher
6691 * we wait on resync_wait. To avoid deadlock, we reduce curr_resync
6692 * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
6693 * This will mean we have to start checking from the beginning again.
6698 mddev->curr_resync = 2;
6701 if (kthread_should_stop())
6702 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6704 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
6706 for_each_mddev(mddev2, tmp) {
6707 if (mddev2 == mddev)
6709 if (!mddev->parallel_resync
6710 && mddev2->curr_resync
6711 && match_mddev_units(mddev, mddev2)) {
6713 if (mddev < mddev2 && mddev->curr_resync == 2) {
6714 /* arbitrarily yield */
6715 mddev->curr_resync = 1;
6716 wake_up(&resync_wait);
6718 if (mddev > mddev2 && mddev->curr_resync == 1)
6719 /* no need to wait here, we can wait the next
6720 * time 'round when curr_resync == 2
6723 /* We need to wait 'interruptible' so as not to
6724 * contribute to the load average, and not to
6725 * be caught by 'softlockup'
6727 prepare_to_wait(&resync_wait, &wq, TASK_INTERRUPTIBLE);
6728 if (!kthread_should_stop() &&
6729 mddev2->curr_resync >= mddev->curr_resync) {
6730 printk(KERN_INFO "md: delaying %s of %s"
6731 " until %s has finished (they"
6732 " share one or more physical units)\n",
6733 desc, mdname(mddev), mdname(mddev2));
6735 if (signal_pending(current))
6736 flush_signals(current);
6738 finish_wait(&resync_wait, &wq);
6741 finish_wait(&resync_wait, &wq);
6744 } while (mddev->curr_resync < 2);
6747 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
6748 /* resync follows the size requested by the personality,
6749 * which defaults to physical size, but can be virtual size
6751 max_sectors = mddev->resync_max_sectors;
6752 mddev->resync_mismatches = 0;
6753 /* we don't use the checkpoint if there's a bitmap */
6754 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
6755 j = mddev->resync_min;
6756 else if (!mddev->bitmap)
6757 j = mddev->recovery_cp;
6759 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
6760 max_sectors = mddev->dev_sectors;
6762 /* recovery follows the physical size of devices */
6763 max_sectors = mddev->dev_sectors;
6766 list_for_each_entry_rcu(rdev, &mddev->disks, same_set)
6767 if (rdev->raid_disk >= 0 &&
6768 !test_bit(Faulty, &rdev->flags) &&
6769 !test_bit(In_sync, &rdev->flags) &&
6770 rdev->recovery_offset < j)
6771 j = rdev->recovery_offset;
6775 printk(KERN_INFO "md: %s of RAID array %s\n", desc, mdname(mddev));
6776 printk(KERN_INFO "md: minimum _guaranteed_ speed:"
6777 " %d KB/sec/disk.\n", speed_min(mddev));
6778 printk(KERN_INFO "md: using maximum available idle IO bandwidth "
6779 "(but not more than %d KB/sec) for %s.\n",
6780 speed_max(mddev), desc);
6782 is_mddev_idle(mddev, 1); /* this initializes IO event counters */
6785 for (m = 0; m < SYNC_MARKS; m++) {
6787 mark_cnt[m] = io_sectors;
6790 mddev->resync_mark = mark[last_mark];
6791 mddev->resync_mark_cnt = mark_cnt[last_mark];
6794 * Tune reconstruction:
6796 window = 32*(PAGE_SIZE/512);
6797 printk(KERN_INFO "md: using %dk window, over a total of %llu blocks.\n",
6798 window/2,(unsigned long long) max_sectors/2);
6800 atomic_set(&mddev->recovery_active, 0);
6805 "md: resuming %s of %s from checkpoint.\n",
6806 desc, mdname(mddev));
6807 mddev->curr_resync = j;
6809 mddev->curr_resync_completed = mddev->curr_resync;
6811 while (j < max_sectors) {
6816 if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
6817 ((mddev->curr_resync > mddev->curr_resync_completed &&
6818 (mddev->curr_resync - mddev->curr_resync_completed)
6819 > (max_sectors >> 4)) ||
6820 (j - mddev->curr_resync_completed)*2
6821 >= mddev->resync_max - mddev->curr_resync_completed
6823 /* time to update curr_resync_completed */
6825 wait_event(mddev->recovery_wait,
6826 atomic_read(&mddev->recovery_active) == 0);
6827 mddev->curr_resync_completed =
6829 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
6830 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
6833 while (j >= mddev->resync_max && !kthread_should_stop()) {
6834 /* As this condition is controlled by user-space,
6835 * we can block indefinitely, so use '_interruptible'
6836 * to avoid triggering warnings.
6838 flush_signals(current); /* just in case */
6839 wait_event_interruptible(mddev->recovery_wait,
6840 mddev->resync_max > j
6841 || kthread_should_stop());
6844 if (kthread_should_stop())
6847 sectors = mddev->pers->sync_request(mddev, j, &skipped,
6848 currspeed < speed_min(mddev));
6850 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6854 if (!skipped) { /* actual IO requested */
6855 io_sectors += sectors;
6856 atomic_add(sectors, &mddev->recovery_active);
6860 if (j>1) mddev->curr_resync = j;
6861 mddev->curr_mark_cnt = io_sectors;
6862 if (last_check == 0)
6863 /* this is the earliers that rebuilt will be
6864 * visible in /proc/mdstat
6866 md_new_event(mddev);
6868 if (last_check + window > io_sectors || j == max_sectors)
6871 last_check = io_sectors;
6873 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
6877 if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) {
6879 int next = (last_mark+1) % SYNC_MARKS;
6881 mddev->resync_mark = mark[next];
6882 mddev->resync_mark_cnt = mark_cnt[next];
6883 mark[next] = jiffies;
6884 mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active);
6889 if (kthread_should_stop())
6894 * this loop exits only if either when we are slower than
6895 * the 'hard' speed limit, or the system was IO-idle for
6897 * the system might be non-idle CPU-wise, but we only care
6898 * about not overloading the IO subsystem. (things like an
6899 * e2fsck being done on the RAID array should execute fast)
6904 currspeed = ((unsigned long)(io_sectors-mddev->resync_mark_cnt))/2
6905 /((jiffies-mddev->resync_mark)/HZ +1) +1;
6907 if (currspeed > speed_min(mddev)) {
6908 if ((currspeed > speed_max(mddev)) ||
6909 !is_mddev_idle(mddev, 0)) {
6915 printk(KERN_INFO "md: %s: %s done.\n",mdname(mddev), desc);
6917 * this also signals 'finished resyncing' to md_stop
6922 wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
6924 /* tell personality that we are finished */
6925 mddev->pers->sync_request(mddev, max_sectors, &skipped, 1);
6927 if (!test_bit(MD_RECOVERY_CHECK, &mddev->recovery) &&
6928 mddev->curr_resync > 2) {
6929 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
6930 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
6931 if (mddev->curr_resync >= mddev->recovery_cp) {
6933 "md: checkpointing %s of %s.\n",
6934 desc, mdname(mddev));
6935 mddev->recovery_cp = mddev->curr_resync;
6938 mddev->recovery_cp = MaxSector;
6940 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery))
6941 mddev->curr_resync = MaxSector;
6943 list_for_each_entry_rcu(rdev, &mddev->disks, same_set)
6944 if (rdev->raid_disk >= 0 &&
6945 mddev->delta_disks >= 0 &&
6946 !test_bit(Faulty, &rdev->flags) &&
6947 !test_bit(In_sync, &rdev->flags) &&
6948 rdev->recovery_offset < mddev->curr_resync)
6949 rdev->recovery_offset = mddev->curr_resync;
6953 set_bit(MD_CHANGE_DEVS, &mddev->flags);
6956 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
6957 /* We completed so min/max setting can be forgotten if used. */
6958 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
6959 mddev->resync_min = 0;
6960 mddev->resync_max = MaxSector;
6961 } else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
6962 mddev->resync_min = mddev->curr_resync_completed;
6963 mddev->curr_resync = 0;
6964 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery))
6965 mddev->curr_resync_completed = 0;
6966 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
6967 wake_up(&resync_wait);
6968 set_bit(MD_RECOVERY_DONE, &mddev->recovery);
6969 md_wakeup_thread(mddev->thread);
6974 * got a signal, exit.
6977 "md: md_do_sync() got signal ... exiting\n");
6978 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6982 EXPORT_SYMBOL_GPL(md_do_sync);
6985 static int remove_and_add_spares(mddev_t *mddev)
6990 mddev->curr_resync_completed = 0;
6992 list_for_each_entry(rdev, &mddev->disks, same_set)
6993 if (rdev->raid_disk >= 0 &&
6994 !test_bit(Blocked, &rdev->flags) &&
6995 (test_bit(Faulty, &rdev->flags) ||
6996 ! test_bit(In_sync, &rdev->flags)) &&
6997 atomic_read(&rdev->nr_pending)==0) {
6998 if (mddev->pers->hot_remove_disk(
6999 mddev, rdev->raid_disk)==0) {
7001 sprintf(nm,"rd%d", rdev->raid_disk);
7002 sysfs_remove_link(&mddev->kobj, nm);
7003 rdev->raid_disk = -1;
7007 if (mddev->degraded && ! mddev->ro && !mddev->recovery_disabled) {
7008 list_for_each_entry(rdev, &mddev->disks, same_set) {
7009 if (rdev->raid_disk >= 0 &&
7010 !test_bit(In_sync, &rdev->flags) &&
7011 !test_bit(Blocked, &rdev->flags))
7013 if (rdev->raid_disk < 0
7014 && !test_bit(Faulty, &rdev->flags)) {
7015 rdev->recovery_offset = 0;
7017 hot_add_disk(mddev, rdev) == 0) {
7019 sprintf(nm, "rd%d", rdev->raid_disk);
7020 if (sysfs_create_link(&mddev->kobj,
7022 /* failure here is OK */;
7024 md_new_event(mddev);
7025 set_bit(MD_CHANGE_DEVS, &mddev->flags);
7034 * This routine is regularly called by all per-raid-array threads to
7035 * deal with generic issues like resync and super-block update.
7036 * Raid personalities that don't have a thread (linear/raid0) do not
7037 * need this as they never do any recovery or update the superblock.
7039 * It does not do any resync itself, but rather "forks" off other threads
7040 * to do that as needed.
7041 * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
7042 * "->recovery" and create a thread at ->sync_thread.
7043 * When the thread finishes it sets MD_RECOVERY_DONE
7044 * and wakeups up this thread which will reap the thread and finish up.
7045 * This thread also removes any faulty devices (with nr_pending == 0).
7047 * The overall approach is:
7048 * 1/ if the superblock needs updating, update it.
7049 * 2/ If a recovery thread is running, don't do anything else.
7050 * 3/ If recovery has finished, clean up, possibly marking spares active.
7051 * 4/ If there are any faulty devices, remove them.
7052 * 5/ If array is degraded, try to add spares devices
7053 * 6/ If array has spares or is not in-sync, start a resync thread.
7055 void md_check_recovery(mddev_t *mddev)
7061 bitmap_daemon_work(mddev);
7066 if (signal_pending(current)) {
7067 if (mddev->pers->sync_request && !mddev->external) {
7068 printk(KERN_INFO "md: %s in immediate safe mode\n",
7070 mddev->safemode = 2;
7072 flush_signals(current);
7075 if (mddev->ro && !test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
7078 (mddev->flags & ~ (1<<MD_CHANGE_PENDING)) ||
7079 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
7080 test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
7081 (mddev->external == 0 && mddev->safemode == 1) ||
7082 (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending)
7083 && !mddev->in_sync && mddev->recovery_cp == MaxSector)
7087 if (mddev_trylock(mddev)) {
7091 /* Only thing we do on a ro array is remove
7094 remove_and_add_spares(mddev);
7095 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7099 if (!mddev->external) {
7101 spin_lock_irq(&mddev->write_lock);
7102 if (mddev->safemode &&
7103 !atomic_read(&mddev->writes_pending) &&
7105 mddev->recovery_cp == MaxSector) {
7108 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
7110 if (mddev->safemode == 1)
7111 mddev->safemode = 0;
7112 spin_unlock_irq(&mddev->write_lock);
7114 sysfs_notify_dirent_safe(mddev->sysfs_state);
7118 md_update_sb(mddev, 0);
7120 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
7121 !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
7122 /* resync/recovery still happening */
7123 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7126 if (mddev->sync_thread) {
7127 /* resync has finished, collect result */
7128 md_unregister_thread(mddev->sync_thread);
7129 mddev->sync_thread = NULL;
7130 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
7131 !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
7133 /* activate any spares */
7134 if (mddev->pers->spare_active(mddev))
7135 sysfs_notify(&mddev->kobj, NULL,
7138 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
7139 mddev->pers->finish_reshape)
7140 mddev->pers->finish_reshape(mddev);
7141 md_update_sb(mddev, 1);
7143 /* if array is no-longer degraded, then any saved_raid_disk
7144 * information must be scrapped
7146 if (!mddev->degraded)
7147 list_for_each_entry(rdev, &mddev->disks, same_set)
7148 rdev->saved_raid_disk = -1;
7150 mddev->recovery = 0;
7151 /* flag recovery needed just to double check */
7152 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7153 sysfs_notify_dirent_safe(mddev->sysfs_action);
7154 md_new_event(mddev);
7157 /* Set RUNNING before clearing NEEDED to avoid
7158 * any transients in the value of "sync_action".
7160 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
7161 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7162 /* Clear some bits that don't mean anything, but
7165 clear_bit(MD_RECOVERY_INTR, &mddev->recovery);
7166 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
7168 if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
7170 /* no recovery is running.
7171 * remove any failed drives, then
7172 * add spares if possible.
7173 * Spare are also removed and re-added, to allow
7174 * the personality to fail the re-add.
7177 if (mddev->reshape_position != MaxSector) {
7178 if (mddev->pers->check_reshape == NULL ||
7179 mddev->pers->check_reshape(mddev) != 0)
7180 /* Cannot proceed */
7182 set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
7183 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
7184 } else if ((spares = remove_and_add_spares(mddev))) {
7185 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
7186 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
7187 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
7188 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
7189 } else if (mddev->recovery_cp < MaxSector) {
7190 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
7191 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
7192 } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
7193 /* nothing to be done ... */
7196 if (mddev->pers->sync_request) {
7197 if (spares && mddev->bitmap && ! mddev->bitmap->file) {
7198 /* We are adding a device or devices to an array
7199 * which has the bitmap stored on all devices.
7200 * So make sure all bitmap pages get written
7202 bitmap_write_all(mddev->bitmap);
7204 mddev->sync_thread = md_register_thread(md_do_sync,
7207 if (!mddev->sync_thread) {
7208 printk(KERN_ERR "%s: could not start resync"
7211 /* leave the spares where they are, it shouldn't hurt */
7212 mddev->recovery = 0;
7214 md_wakeup_thread(mddev->sync_thread);
7215 sysfs_notify_dirent_safe(mddev->sysfs_action);
7216 md_new_event(mddev);
7219 if (!mddev->sync_thread) {
7220 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
7221 if (test_and_clear_bit(MD_RECOVERY_RECOVER,
7223 if (mddev->sysfs_action)
7224 sysfs_notify_dirent_safe(mddev->sysfs_action);
7226 mddev_unlock(mddev);
7230 void md_wait_for_blocked_rdev(mdk_rdev_t *rdev, mddev_t *mddev)
7232 sysfs_notify_dirent_safe(rdev->sysfs_state);
7233 wait_event_timeout(rdev->blocked_wait,
7234 !test_bit(Blocked, &rdev->flags),
7235 msecs_to_jiffies(5000));
7236 rdev_dec_pending(rdev, mddev);
7238 EXPORT_SYMBOL(md_wait_for_blocked_rdev);
7240 static int md_notify_reboot(struct notifier_block *this,
7241 unsigned long code, void *x)
7243 struct list_head *tmp;
7246 if ((code == SYS_DOWN) || (code == SYS_HALT) || (code == SYS_POWER_OFF)) {
7248 printk(KERN_INFO "md: stopping all md devices.\n");
7250 for_each_mddev(mddev, tmp)
7251 if (mddev_trylock(mddev)) {
7252 /* Force a switch to readonly even array
7253 * appears to still be in use. Hence
7256 md_set_readonly(mddev, 100);
7257 mddev_unlock(mddev);
7260 * certain more exotic SCSI devices are known to be
7261 * volatile wrt too early system reboots. While the
7262 * right place to handle this issue is the given
7263 * driver, we do want to have a safe RAID driver ...
7270 static struct notifier_block md_notifier = {
7271 .notifier_call = md_notify_reboot,
7273 .priority = INT_MAX, /* before any real devices */
7276 static void md_geninit(void)
7278 dprintk("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
7280 proc_create("mdstat", S_IRUGO, NULL, &md_seq_fops);
7283 static int __init md_init(void)
7287 md_wq = alloc_workqueue("md", WQ_RESCUER, 0);
7291 md_misc_wq = alloc_workqueue("md_misc", 0, 0);
7295 if ((ret = register_blkdev(MD_MAJOR, "md")) < 0)
7298 if ((ret = register_blkdev(0, "mdp")) < 0)
7302 blk_register_region(MKDEV(MD_MAJOR, 0), 1UL<<MINORBITS, THIS_MODULE,
7303 md_probe, NULL, NULL);
7304 blk_register_region(MKDEV(mdp_major, 0), 1UL<<MINORBITS, THIS_MODULE,
7305 md_probe, NULL, NULL);
7307 register_reboot_notifier(&md_notifier);
7308 raid_table_header = register_sysctl_table(raid_root_table);
7314 unregister_blkdev(MD_MAJOR, "md");
7316 destroy_workqueue(md_misc_wq);
7318 destroy_workqueue(md_wq);
7326 * Searches all registered partitions for autorun RAID arrays
7330 static LIST_HEAD(all_detected_devices);
7331 struct detected_devices_node {
7332 struct list_head list;
7336 void md_autodetect_dev(dev_t dev)
7338 struct detected_devices_node *node_detected_dev;
7340 node_detected_dev = kzalloc(sizeof(*node_detected_dev), GFP_KERNEL);
7341 if (node_detected_dev) {
7342 node_detected_dev->dev = dev;
7343 list_add_tail(&node_detected_dev->list, &all_detected_devices);
7345 printk(KERN_CRIT "md: md_autodetect_dev: kzalloc failed"
7346 ", skipping dev(%d,%d)\n", MAJOR(dev), MINOR(dev));
7351 static void autostart_arrays(int part)
7354 struct detected_devices_node *node_detected_dev;
7356 int i_scanned, i_passed;
7361 printk(KERN_INFO "md: Autodetecting RAID arrays.\n");
7363 while (!list_empty(&all_detected_devices) && i_scanned < INT_MAX) {
7365 node_detected_dev = list_entry(all_detected_devices.next,
7366 struct detected_devices_node, list);
7367 list_del(&node_detected_dev->list);
7368 dev = node_detected_dev->dev;
7369 kfree(node_detected_dev);
7370 rdev = md_import_device(dev,0, 90);
7374 if (test_bit(Faulty, &rdev->flags)) {
7378 set_bit(AutoDetected, &rdev->flags);
7379 list_add(&rdev->same_set, &pending_raid_disks);
7383 printk(KERN_INFO "md: Scanned %d and added %d devices.\n",
7384 i_scanned, i_passed);
7386 autorun_devices(part);
7389 #endif /* !MODULE */
7391 static __exit void md_exit(void)
7394 struct list_head *tmp;
7396 blk_unregister_region(MKDEV(MD_MAJOR,0), 1U << MINORBITS);
7397 blk_unregister_region(MKDEV(mdp_major,0), 1U << MINORBITS);
7399 unregister_blkdev(MD_MAJOR,"md");
7400 unregister_blkdev(mdp_major, "mdp");
7401 unregister_reboot_notifier(&md_notifier);
7402 unregister_sysctl_table(raid_table_header);
7403 remove_proc_entry("mdstat", NULL);
7404 for_each_mddev(mddev, tmp) {
7405 export_array(mddev);
7406 mddev->hold_active = 0;
7408 destroy_workqueue(md_misc_wq);
7409 destroy_workqueue(md_wq);
7412 subsys_initcall(md_init);
7413 module_exit(md_exit)
7415 static int get_ro(char *buffer, struct kernel_param *kp)
7417 return sprintf(buffer, "%d", start_readonly);
7419 static int set_ro(const char *val, struct kernel_param *kp)
7422 int num = simple_strtoul(val, &e, 10);
7423 if (*val && (*e == '\0' || *e == '\n')) {
7424 start_readonly = num;
7430 module_param_call(start_ro, set_ro, get_ro, NULL, S_IRUSR|S_IWUSR);
7431 module_param(start_dirty_degraded, int, S_IRUGO|S_IWUSR);
7433 module_param_call(new_array, add_named_array, NULL, NULL, S_IWUSR);
7435 EXPORT_SYMBOL(register_md_personality);
7436 EXPORT_SYMBOL(unregister_md_personality);
7437 EXPORT_SYMBOL(md_error);
7438 EXPORT_SYMBOL(md_done_sync);
7439 EXPORT_SYMBOL(md_write_start);
7440 EXPORT_SYMBOL(md_write_end);
7441 EXPORT_SYMBOL(md_register_thread);
7442 EXPORT_SYMBOL(md_unregister_thread);
7443 EXPORT_SYMBOL(md_wakeup_thread);
7444 EXPORT_SYMBOL(md_check_recovery);
7445 MODULE_LICENSE("GPL");
7446 MODULE_DESCRIPTION("MD RAID framework");
7448 MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR);