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/smp_lock.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)))
62 static void autostart_arrays(int part);
65 static LIST_HEAD(pers_list);
66 static DEFINE_SPINLOCK(pers_lock);
68 static void md_print_devices(void);
70 static DECLARE_WAIT_QUEUE_HEAD(resync_wait);
72 #define MD_BUG(x...) { printk("md: bug in file %s, line %d\n", __FILE__, __LINE__); md_print_devices(); }
75 * Default number of read corrections we'll attempt on an rdev
76 * before ejecting it from the array. We divide the read error
77 * count by 2 for every hour elapsed between read errors.
79 #define MD_DEFAULT_MAX_CORRECTED_READ_ERRORS 20
81 * Current RAID-1,4,5 parallel reconstruction 'guaranteed speed limit'
82 * is 1000 KB/sec, so the extra system load does not show up that much.
83 * Increase it if you want to have more _guaranteed_ speed. Note that
84 * the RAID driver will use the maximum available bandwidth if the IO
85 * subsystem is idle. There is also an 'absolute maximum' reconstruction
86 * speed limit - in case reconstruction slows down your system despite
89 * you can change it via /proc/sys/dev/raid/speed_limit_min and _max.
90 * or /sys/block/mdX/md/sync_speed_{min,max}
93 static int sysctl_speed_limit_min = 1000;
94 static int sysctl_speed_limit_max = 200000;
95 static inline int speed_min(mddev_t *mddev)
97 return mddev->sync_speed_min ?
98 mddev->sync_speed_min : sysctl_speed_limit_min;
101 static inline int speed_max(mddev_t *mddev)
103 return mddev->sync_speed_max ?
104 mddev->sync_speed_max : sysctl_speed_limit_max;
107 static struct ctl_table_header *raid_table_header;
109 static ctl_table raid_table[] = {
111 .procname = "speed_limit_min",
112 .data = &sysctl_speed_limit_min,
113 .maxlen = sizeof(int),
114 .mode = S_IRUGO|S_IWUSR,
115 .proc_handler = proc_dointvec,
118 .procname = "speed_limit_max",
119 .data = &sysctl_speed_limit_max,
120 .maxlen = sizeof(int),
121 .mode = S_IRUGO|S_IWUSR,
122 .proc_handler = proc_dointvec,
127 static ctl_table raid_dir_table[] = {
131 .mode = S_IRUGO|S_IXUGO,
137 static ctl_table raid_root_table[] = {
142 .child = raid_dir_table,
147 static const struct block_device_operations md_fops;
149 static int start_readonly;
152 * We have a system wide 'event count' that is incremented
153 * on any 'interesting' event, and readers of /proc/mdstat
154 * can use 'poll' or 'select' to find out when the event
158 * start array, stop array, error, add device, remove device,
159 * start build, activate spare
161 static DECLARE_WAIT_QUEUE_HEAD(md_event_waiters);
162 static atomic_t md_event_count;
163 void md_new_event(mddev_t *mddev)
165 atomic_inc(&md_event_count);
166 wake_up(&md_event_waiters);
168 EXPORT_SYMBOL_GPL(md_new_event);
170 /* Alternate version that can be called from interrupts
171 * when calling sysfs_notify isn't needed.
173 static void md_new_event_inintr(mddev_t *mddev)
175 atomic_inc(&md_event_count);
176 wake_up(&md_event_waiters);
180 * Enables to iterate over all existing md arrays
181 * all_mddevs_lock protects this list.
183 static LIST_HEAD(all_mddevs);
184 static DEFINE_SPINLOCK(all_mddevs_lock);
188 * iterates through all used mddevs in the system.
189 * We take care to grab the all_mddevs_lock whenever navigating
190 * the list, and to always hold a refcount when unlocked.
191 * Any code which breaks out of this loop while own
192 * a reference to the current mddev and must mddev_put it.
194 #define for_each_mddev(mddev,tmp) \
196 for (({ spin_lock(&all_mddevs_lock); \
197 tmp = all_mddevs.next; \
199 ({ if (tmp != &all_mddevs) \
200 mddev_get(list_entry(tmp, mddev_t, all_mddevs));\
201 spin_unlock(&all_mddevs_lock); \
202 if (mddev) mddev_put(mddev); \
203 mddev = list_entry(tmp, mddev_t, all_mddevs); \
204 tmp != &all_mddevs;}); \
205 ({ spin_lock(&all_mddevs_lock); \
210 /* Rather than calling directly into the personality make_request function,
211 * IO requests come here first so that we can check if the device is
212 * being suspended pending a reconfiguration.
213 * We hold a refcount over the call to ->make_request. By the time that
214 * call has finished, the bio has been linked into some internal structure
215 * and so is visible to ->quiesce(), so we don't need the refcount any more.
217 static int md_make_request(struct request_queue *q, struct bio *bio)
219 const int rw = bio_data_dir(bio);
220 mddev_t *mddev = q->queuedata;
224 if (mddev == NULL || mddev->pers == NULL) {
229 if (mddev->suspended || mddev->barrier) {
232 prepare_to_wait(&mddev->sb_wait, &__wait,
233 TASK_UNINTERRUPTIBLE);
234 if (!mddev->suspended && !mddev->barrier)
240 finish_wait(&mddev->sb_wait, &__wait);
242 atomic_inc(&mddev->active_io);
245 rv = mddev->pers->make_request(mddev, bio);
247 cpu = part_stat_lock();
248 part_stat_inc(cpu, &mddev->gendisk->part0, ios[rw]);
249 part_stat_add(cpu, &mddev->gendisk->part0, sectors[rw],
253 if (atomic_dec_and_test(&mddev->active_io) && mddev->suspended)
254 wake_up(&mddev->sb_wait);
259 /* mddev_suspend makes sure no new requests are submitted
260 * to the device, and that any requests that have been submitted
261 * are completely handled.
262 * Once ->stop is called and completes, the module will be completely
265 void mddev_suspend(mddev_t *mddev)
267 BUG_ON(mddev->suspended);
268 mddev->suspended = 1;
270 wait_event(mddev->sb_wait, atomic_read(&mddev->active_io) == 0);
271 mddev->pers->quiesce(mddev, 1);
273 EXPORT_SYMBOL_GPL(mddev_suspend);
275 void mddev_resume(mddev_t *mddev)
277 mddev->suspended = 0;
278 wake_up(&mddev->sb_wait);
279 mddev->pers->quiesce(mddev, 0);
281 EXPORT_SYMBOL_GPL(mddev_resume);
283 int mddev_congested(mddev_t *mddev, int bits)
287 return mddev->suspended;
289 EXPORT_SYMBOL(mddev_congested);
292 * Generic barrier handling for md
295 #define POST_REQUEST_BARRIER ((void*)1)
297 static void md_end_barrier(struct bio *bio, int err)
299 mdk_rdev_t *rdev = bio->bi_private;
300 mddev_t *mddev = rdev->mddev;
301 if (err == -EOPNOTSUPP && mddev->barrier != POST_REQUEST_BARRIER)
302 set_bit(BIO_EOPNOTSUPP, &mddev->barrier->bi_flags);
304 rdev_dec_pending(rdev, mddev);
306 if (atomic_dec_and_test(&mddev->flush_pending)) {
307 if (mddev->barrier == POST_REQUEST_BARRIER) {
308 /* This was a post-request barrier */
309 mddev->barrier = NULL;
310 wake_up(&mddev->sb_wait);
312 /* The pre-request barrier has finished */
313 schedule_work(&mddev->barrier_work);
318 static void submit_barriers(mddev_t *mddev)
323 list_for_each_entry_rcu(rdev, &mddev->disks, same_set)
324 if (rdev->raid_disk >= 0 &&
325 !test_bit(Faulty, &rdev->flags)) {
326 /* Take two references, one is dropped
327 * when request finishes, one after
328 * we reclaim rcu_read_lock
331 atomic_inc(&rdev->nr_pending);
332 atomic_inc(&rdev->nr_pending);
334 bi = bio_alloc(GFP_KERNEL, 0);
335 bi->bi_end_io = md_end_barrier;
336 bi->bi_private = rdev;
337 bi->bi_bdev = rdev->bdev;
338 atomic_inc(&mddev->flush_pending);
339 submit_bio(WRITE_BARRIER, bi);
341 rdev_dec_pending(rdev, mddev);
346 static void md_submit_barrier(struct work_struct *ws)
348 mddev_t *mddev = container_of(ws, mddev_t, barrier_work);
349 struct bio *bio = mddev->barrier;
351 atomic_set(&mddev->flush_pending, 1);
353 if (test_bit(BIO_EOPNOTSUPP, &bio->bi_flags))
354 bio_endio(bio, -EOPNOTSUPP);
355 else if (bio->bi_size == 0)
356 /* an empty barrier - all done */
359 bio->bi_rw &= ~REQ_HARDBARRIER;
360 if (mddev->pers->make_request(mddev, bio))
361 generic_make_request(bio);
362 mddev->barrier = POST_REQUEST_BARRIER;
363 submit_barriers(mddev);
365 if (atomic_dec_and_test(&mddev->flush_pending)) {
366 mddev->barrier = NULL;
367 wake_up(&mddev->sb_wait);
371 void md_barrier_request(mddev_t *mddev, struct bio *bio)
373 spin_lock_irq(&mddev->write_lock);
374 wait_event_lock_irq(mddev->sb_wait,
376 mddev->write_lock, /*nothing*/);
377 mddev->barrier = bio;
378 spin_unlock_irq(&mddev->write_lock);
380 atomic_set(&mddev->flush_pending, 1);
381 INIT_WORK(&mddev->barrier_work, md_submit_barrier);
383 submit_barriers(mddev);
385 if (atomic_dec_and_test(&mddev->flush_pending))
386 schedule_work(&mddev->barrier_work);
388 EXPORT_SYMBOL(md_barrier_request);
390 /* Support for plugging.
391 * This mirrors the plugging support in request_queue, but does not
392 * require having a whole queue
394 static void plugger_work(struct work_struct *work)
396 struct plug_handle *plug =
397 container_of(work, struct plug_handle, unplug_work);
398 plug->unplug_fn(plug);
400 static void plugger_timeout(unsigned long data)
402 struct plug_handle *plug = (void *)data;
403 kblockd_schedule_work(NULL, &plug->unplug_work);
405 void plugger_init(struct plug_handle *plug,
406 void (*unplug_fn)(struct plug_handle *))
408 plug->unplug_flag = 0;
409 plug->unplug_fn = unplug_fn;
410 init_timer(&plug->unplug_timer);
411 plug->unplug_timer.function = plugger_timeout;
412 plug->unplug_timer.data = (unsigned long)plug;
413 INIT_WORK(&plug->unplug_work, plugger_work);
415 EXPORT_SYMBOL_GPL(plugger_init);
417 void plugger_set_plug(struct plug_handle *plug)
419 if (!test_and_set_bit(PLUGGED_FLAG, &plug->unplug_flag))
420 mod_timer(&plug->unplug_timer, jiffies + msecs_to_jiffies(3)+1);
422 EXPORT_SYMBOL_GPL(plugger_set_plug);
424 int plugger_remove_plug(struct plug_handle *plug)
426 if (test_and_clear_bit(PLUGGED_FLAG, &plug->unplug_flag)) {
427 del_timer(&plug->unplug_timer);
432 EXPORT_SYMBOL_GPL(plugger_remove_plug);
435 static inline mddev_t *mddev_get(mddev_t *mddev)
437 atomic_inc(&mddev->active);
441 static void mddev_delayed_delete(struct work_struct *ws);
443 static void mddev_put(mddev_t *mddev)
445 if (!atomic_dec_and_lock(&mddev->active, &all_mddevs_lock))
447 if (!mddev->raid_disks && list_empty(&mddev->disks) &&
448 mddev->ctime == 0 && !mddev->hold_active) {
449 /* Array is not configured at all, and not held active,
451 list_del(&mddev->all_mddevs);
452 if (mddev->gendisk) {
453 /* we did a probe so need to clean up.
454 * Call schedule_work inside the spinlock
455 * so that flush_scheduled_work() after
456 * mddev_find will succeed in waiting for the
459 INIT_WORK(&mddev->del_work, mddev_delayed_delete);
460 schedule_work(&mddev->del_work);
464 spin_unlock(&all_mddevs_lock);
467 void mddev_init(mddev_t *mddev)
469 mutex_init(&mddev->open_mutex);
470 mutex_init(&mddev->reconfig_mutex);
471 mutex_init(&mddev->bitmap_info.mutex);
472 INIT_LIST_HEAD(&mddev->disks);
473 INIT_LIST_HEAD(&mddev->all_mddevs);
474 init_timer(&mddev->safemode_timer);
475 atomic_set(&mddev->active, 1);
476 atomic_set(&mddev->openers, 0);
477 atomic_set(&mddev->active_io, 0);
478 spin_lock_init(&mddev->write_lock);
479 atomic_set(&mddev->flush_pending, 0);
480 init_waitqueue_head(&mddev->sb_wait);
481 init_waitqueue_head(&mddev->recovery_wait);
482 mddev->reshape_position = MaxSector;
483 mddev->resync_min = 0;
484 mddev->resync_max = MaxSector;
485 mddev->level = LEVEL_NONE;
487 EXPORT_SYMBOL_GPL(mddev_init);
489 static mddev_t * mddev_find(dev_t unit)
491 mddev_t *mddev, *new = NULL;
494 spin_lock(&all_mddevs_lock);
497 list_for_each_entry(mddev, &all_mddevs, all_mddevs)
498 if (mddev->unit == unit) {
500 spin_unlock(&all_mddevs_lock);
506 list_add(&new->all_mddevs, &all_mddevs);
507 spin_unlock(&all_mddevs_lock);
508 new->hold_active = UNTIL_IOCTL;
512 /* find an unused unit number */
513 static int next_minor = 512;
514 int start = next_minor;
518 dev = MKDEV(MD_MAJOR, next_minor);
520 if (next_minor > MINORMASK)
522 if (next_minor == start) {
523 /* Oh dear, all in use. */
524 spin_unlock(&all_mddevs_lock);
530 list_for_each_entry(mddev, &all_mddevs, all_mddevs)
531 if (mddev->unit == dev) {
537 new->md_minor = MINOR(dev);
538 new->hold_active = UNTIL_STOP;
539 list_add(&new->all_mddevs, &all_mddevs);
540 spin_unlock(&all_mddevs_lock);
543 spin_unlock(&all_mddevs_lock);
545 new = kzalloc(sizeof(*new), GFP_KERNEL);
550 if (MAJOR(unit) == MD_MAJOR)
551 new->md_minor = MINOR(unit);
553 new->md_minor = MINOR(unit) >> MdpMinorShift;
560 static inline int mddev_lock(mddev_t * mddev)
562 return mutex_lock_interruptible(&mddev->reconfig_mutex);
565 static inline int mddev_is_locked(mddev_t *mddev)
567 return mutex_is_locked(&mddev->reconfig_mutex);
570 static inline int mddev_trylock(mddev_t * mddev)
572 return mutex_trylock(&mddev->reconfig_mutex);
575 static struct attribute_group md_redundancy_group;
577 static void mddev_unlock(mddev_t * mddev)
579 if (mddev->to_remove) {
580 /* These cannot be removed under reconfig_mutex as
581 * an access to the files will try to take reconfig_mutex
582 * while holding the file unremovable, which leads to
584 * So hold set sysfs_active while the remove in happeing,
585 * and anything else which might set ->to_remove or my
586 * otherwise change the sysfs namespace will fail with
587 * -EBUSY if sysfs_active is still set.
588 * We set sysfs_active under reconfig_mutex and elsewhere
589 * test it under the same mutex to ensure its correct value
592 struct attribute_group *to_remove = mddev->to_remove;
593 mddev->to_remove = NULL;
594 mddev->sysfs_active = 1;
595 mutex_unlock(&mddev->reconfig_mutex);
597 if (mddev->kobj.sd) {
598 if (to_remove != &md_redundancy_group)
599 sysfs_remove_group(&mddev->kobj, to_remove);
600 if (mddev->pers == NULL ||
601 mddev->pers->sync_request == NULL) {
602 sysfs_remove_group(&mddev->kobj, &md_redundancy_group);
603 if (mddev->sysfs_action)
604 sysfs_put(mddev->sysfs_action);
605 mddev->sysfs_action = NULL;
608 mddev->sysfs_active = 0;
610 mutex_unlock(&mddev->reconfig_mutex);
612 md_wakeup_thread(mddev->thread);
615 static mdk_rdev_t * find_rdev_nr(mddev_t *mddev, int nr)
619 list_for_each_entry(rdev, &mddev->disks, same_set)
620 if (rdev->desc_nr == nr)
626 static mdk_rdev_t * find_rdev(mddev_t * mddev, dev_t dev)
630 list_for_each_entry(rdev, &mddev->disks, same_set)
631 if (rdev->bdev->bd_dev == dev)
637 static struct mdk_personality *find_pers(int level, char *clevel)
639 struct mdk_personality *pers;
640 list_for_each_entry(pers, &pers_list, list) {
641 if (level != LEVEL_NONE && pers->level == level)
643 if (strcmp(pers->name, clevel)==0)
649 /* return the offset of the super block in 512byte sectors */
650 static inline sector_t calc_dev_sboffset(struct block_device *bdev)
652 sector_t num_sectors = bdev->bd_inode->i_size / 512;
653 return MD_NEW_SIZE_SECTORS(num_sectors);
656 static int alloc_disk_sb(mdk_rdev_t * rdev)
661 rdev->sb_page = alloc_page(GFP_KERNEL);
662 if (!rdev->sb_page) {
663 printk(KERN_ALERT "md: out of memory.\n");
670 static void free_disk_sb(mdk_rdev_t * rdev)
673 put_page(rdev->sb_page);
675 rdev->sb_page = NULL;
682 static void super_written(struct bio *bio, int error)
684 mdk_rdev_t *rdev = bio->bi_private;
685 mddev_t *mddev = rdev->mddev;
687 if (error || !test_bit(BIO_UPTODATE, &bio->bi_flags)) {
688 printk("md: super_written gets error=%d, uptodate=%d\n",
689 error, test_bit(BIO_UPTODATE, &bio->bi_flags));
690 WARN_ON(test_bit(BIO_UPTODATE, &bio->bi_flags));
691 md_error(mddev, rdev);
694 if (atomic_dec_and_test(&mddev->pending_writes))
695 wake_up(&mddev->sb_wait);
699 static void super_written_barrier(struct bio *bio, int error)
701 struct bio *bio2 = bio->bi_private;
702 mdk_rdev_t *rdev = bio2->bi_private;
703 mddev_t *mddev = rdev->mddev;
705 if (!test_bit(BIO_UPTODATE, &bio->bi_flags) &&
706 error == -EOPNOTSUPP) {
708 /* barriers don't appear to be supported :-( */
709 set_bit(BarriersNotsupp, &rdev->flags);
710 mddev->barriers_work = 0;
711 spin_lock_irqsave(&mddev->write_lock, flags);
712 bio2->bi_next = mddev->biolist;
713 mddev->biolist = bio2;
714 spin_unlock_irqrestore(&mddev->write_lock, flags);
715 wake_up(&mddev->sb_wait);
719 bio->bi_private = rdev;
720 super_written(bio, error);
724 void md_super_write(mddev_t *mddev, mdk_rdev_t *rdev,
725 sector_t sector, int size, struct page *page)
727 /* write first size bytes of page to sector of rdev
728 * Increment mddev->pending_writes before returning
729 * and decrement it on completion, waking up sb_wait
730 * if zero is reached.
731 * If an error occurred, call md_error
733 * As we might need to resubmit the request if REQ_HARDBARRIER
734 * causes ENOTSUPP, we allocate a spare bio...
736 struct bio *bio = bio_alloc(GFP_NOIO, 1);
737 int rw = REQ_WRITE | REQ_SYNC | REQ_UNPLUG;
739 bio->bi_bdev = rdev->bdev;
740 bio->bi_sector = sector;
741 bio_add_page(bio, page, size, 0);
742 bio->bi_private = rdev;
743 bio->bi_end_io = super_written;
746 atomic_inc(&mddev->pending_writes);
747 if (!test_bit(BarriersNotsupp, &rdev->flags)) {
749 rw |= REQ_HARDBARRIER;
750 rbio = bio_clone(bio, GFP_NOIO);
751 rbio->bi_private = bio;
752 rbio->bi_end_io = super_written_barrier;
753 submit_bio(rw, rbio);
758 void md_super_wait(mddev_t *mddev)
760 /* wait for all superblock writes that were scheduled to complete.
761 * if any had to be retried (due to BARRIER problems), retry them
765 prepare_to_wait(&mddev->sb_wait, &wq, TASK_UNINTERRUPTIBLE);
766 if (atomic_read(&mddev->pending_writes)==0)
768 while (mddev->biolist) {
770 spin_lock_irq(&mddev->write_lock);
771 bio = mddev->biolist;
772 mddev->biolist = bio->bi_next ;
774 spin_unlock_irq(&mddev->write_lock);
775 submit_bio(bio->bi_rw, bio);
779 finish_wait(&mddev->sb_wait, &wq);
782 static void bi_complete(struct bio *bio, int error)
784 complete((struct completion*)bio->bi_private);
787 int sync_page_io(struct block_device *bdev, sector_t sector, int size,
788 struct page *page, int rw)
790 struct bio *bio = bio_alloc(GFP_NOIO, 1);
791 struct completion event;
794 rw |= REQ_SYNC | REQ_UNPLUG;
797 bio->bi_sector = sector;
798 bio_add_page(bio, page, size, 0);
799 init_completion(&event);
800 bio->bi_private = &event;
801 bio->bi_end_io = bi_complete;
803 wait_for_completion(&event);
805 ret = test_bit(BIO_UPTODATE, &bio->bi_flags);
809 EXPORT_SYMBOL_GPL(sync_page_io);
811 static int read_disk_sb(mdk_rdev_t * rdev, int size)
813 char b[BDEVNAME_SIZE];
814 if (!rdev->sb_page) {
822 if (!sync_page_io(rdev->bdev, rdev->sb_start, size, rdev->sb_page, READ))
828 printk(KERN_WARNING "md: disabled device %s, could not read superblock.\n",
829 bdevname(rdev->bdev,b));
833 static int uuid_equal(mdp_super_t *sb1, mdp_super_t *sb2)
835 return sb1->set_uuid0 == sb2->set_uuid0 &&
836 sb1->set_uuid1 == sb2->set_uuid1 &&
837 sb1->set_uuid2 == sb2->set_uuid2 &&
838 sb1->set_uuid3 == sb2->set_uuid3;
841 static int sb_equal(mdp_super_t *sb1, mdp_super_t *sb2)
844 mdp_super_t *tmp1, *tmp2;
846 tmp1 = kmalloc(sizeof(*tmp1),GFP_KERNEL);
847 tmp2 = kmalloc(sizeof(*tmp2),GFP_KERNEL);
849 if (!tmp1 || !tmp2) {
851 printk(KERN_INFO "md.c sb_equal(): failed to allocate memory!\n");
859 * nr_disks is not constant
864 ret = (memcmp(tmp1, tmp2, MD_SB_GENERIC_CONSTANT_WORDS * 4) == 0);
872 static u32 md_csum_fold(u32 csum)
874 csum = (csum & 0xffff) + (csum >> 16);
875 return (csum & 0xffff) + (csum >> 16);
878 static unsigned int calc_sb_csum(mdp_super_t * sb)
881 u32 *sb32 = (u32*)sb;
883 unsigned int disk_csum, csum;
885 disk_csum = sb->sb_csum;
888 for (i = 0; i < MD_SB_BYTES/4 ; i++)
890 csum = (newcsum & 0xffffffff) + (newcsum>>32);
894 /* This used to use csum_partial, which was wrong for several
895 * reasons including that different results are returned on
896 * different architectures. It isn't critical that we get exactly
897 * the same return value as before (we always csum_fold before
898 * testing, and that removes any differences). However as we
899 * know that csum_partial always returned a 16bit value on
900 * alphas, do a fold to maximise conformity to previous behaviour.
902 sb->sb_csum = md_csum_fold(disk_csum);
904 sb->sb_csum = disk_csum;
911 * Handle superblock details.
912 * We want to be able to handle multiple superblock formats
913 * so we have a common interface to them all, and an array of
914 * different handlers.
915 * We rely on user-space to write the initial superblock, and support
916 * reading and updating of superblocks.
917 * Interface methods are:
918 * int load_super(mdk_rdev_t *dev, mdk_rdev_t *refdev, int minor_version)
919 * loads and validates a superblock on dev.
920 * if refdev != NULL, compare superblocks on both devices
922 * 0 - dev has a superblock that is compatible with refdev
923 * 1 - dev has a superblock that is compatible and newer than refdev
924 * so dev should be used as the refdev in future
925 * -EINVAL superblock incompatible or invalid
926 * -othererror e.g. -EIO
928 * int validate_super(mddev_t *mddev, mdk_rdev_t *dev)
929 * Verify that dev is acceptable into mddev.
930 * The first time, mddev->raid_disks will be 0, and data from
931 * dev should be merged in. Subsequent calls check that dev
932 * is new enough. Return 0 or -EINVAL
934 * void sync_super(mddev_t *mddev, mdk_rdev_t *dev)
935 * Update the superblock for rdev with data in mddev
936 * This does not write to disc.
942 struct module *owner;
943 int (*load_super)(mdk_rdev_t *rdev, mdk_rdev_t *refdev,
945 int (*validate_super)(mddev_t *mddev, mdk_rdev_t *rdev);
946 void (*sync_super)(mddev_t *mddev, mdk_rdev_t *rdev);
947 unsigned long long (*rdev_size_change)(mdk_rdev_t *rdev,
948 sector_t num_sectors);
952 * Check that the given mddev has no bitmap.
954 * This function is called from the run method of all personalities that do not
955 * support bitmaps. It prints an error message and returns non-zero if mddev
956 * has a bitmap. Otherwise, it returns 0.
959 int md_check_no_bitmap(mddev_t *mddev)
961 if (!mddev->bitmap_info.file && !mddev->bitmap_info.offset)
963 printk(KERN_ERR "%s: bitmaps are not supported for %s\n",
964 mdname(mddev), mddev->pers->name);
967 EXPORT_SYMBOL(md_check_no_bitmap);
970 * load_super for 0.90.0
972 static int super_90_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
974 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
979 * Calculate the position of the superblock (512byte sectors),
980 * it's at the end of the disk.
982 * It also happens to be a multiple of 4Kb.
984 rdev->sb_start = calc_dev_sboffset(rdev->bdev);
986 ret = read_disk_sb(rdev, MD_SB_BYTES);
991 bdevname(rdev->bdev, b);
992 sb = (mdp_super_t*)page_address(rdev->sb_page);
994 if (sb->md_magic != MD_SB_MAGIC) {
995 printk(KERN_ERR "md: invalid raid superblock magic on %s\n",
1000 if (sb->major_version != 0 ||
1001 sb->minor_version < 90 ||
1002 sb->minor_version > 91) {
1003 printk(KERN_WARNING "Bad version number %d.%d on %s\n",
1004 sb->major_version, sb->minor_version,
1009 if (sb->raid_disks <= 0)
1012 if (md_csum_fold(calc_sb_csum(sb)) != md_csum_fold(sb->sb_csum)) {
1013 printk(KERN_WARNING "md: invalid superblock checksum on %s\n",
1018 rdev->preferred_minor = sb->md_minor;
1019 rdev->data_offset = 0;
1020 rdev->sb_size = MD_SB_BYTES;
1022 if (sb->level == LEVEL_MULTIPATH)
1025 rdev->desc_nr = sb->this_disk.number;
1031 mdp_super_t *refsb = (mdp_super_t*)page_address(refdev->sb_page);
1032 if (!uuid_equal(refsb, sb)) {
1033 printk(KERN_WARNING "md: %s has different UUID to %s\n",
1034 b, bdevname(refdev->bdev,b2));
1037 if (!sb_equal(refsb, sb)) {
1038 printk(KERN_WARNING "md: %s has same UUID"
1039 " but different superblock to %s\n",
1040 b, bdevname(refdev->bdev, b2));
1044 ev2 = md_event(refsb);
1050 rdev->sectors = rdev->sb_start;
1052 if (rdev->sectors < sb->size * 2 && sb->level > 1)
1053 /* "this cannot possibly happen" ... */
1061 * validate_super for 0.90.0
1063 static int super_90_validate(mddev_t *mddev, mdk_rdev_t *rdev)
1066 mdp_super_t *sb = (mdp_super_t *)page_address(rdev->sb_page);
1067 __u64 ev1 = md_event(sb);
1069 rdev->raid_disk = -1;
1070 clear_bit(Faulty, &rdev->flags);
1071 clear_bit(In_sync, &rdev->flags);
1072 clear_bit(WriteMostly, &rdev->flags);
1073 clear_bit(BarriersNotsupp, &rdev->flags);
1075 if (mddev->raid_disks == 0) {
1076 mddev->major_version = 0;
1077 mddev->minor_version = sb->minor_version;
1078 mddev->patch_version = sb->patch_version;
1079 mddev->external = 0;
1080 mddev->chunk_sectors = sb->chunk_size >> 9;
1081 mddev->ctime = sb->ctime;
1082 mddev->utime = sb->utime;
1083 mddev->level = sb->level;
1084 mddev->clevel[0] = 0;
1085 mddev->layout = sb->layout;
1086 mddev->raid_disks = sb->raid_disks;
1087 mddev->dev_sectors = sb->size * 2;
1088 mddev->events = ev1;
1089 mddev->bitmap_info.offset = 0;
1090 mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9;
1092 if (mddev->minor_version >= 91) {
1093 mddev->reshape_position = sb->reshape_position;
1094 mddev->delta_disks = sb->delta_disks;
1095 mddev->new_level = sb->new_level;
1096 mddev->new_layout = sb->new_layout;
1097 mddev->new_chunk_sectors = sb->new_chunk >> 9;
1099 mddev->reshape_position = MaxSector;
1100 mddev->delta_disks = 0;
1101 mddev->new_level = mddev->level;
1102 mddev->new_layout = mddev->layout;
1103 mddev->new_chunk_sectors = mddev->chunk_sectors;
1106 if (sb->state & (1<<MD_SB_CLEAN))
1107 mddev->recovery_cp = MaxSector;
1109 if (sb->events_hi == sb->cp_events_hi &&
1110 sb->events_lo == sb->cp_events_lo) {
1111 mddev->recovery_cp = sb->recovery_cp;
1113 mddev->recovery_cp = 0;
1116 memcpy(mddev->uuid+0, &sb->set_uuid0, 4);
1117 memcpy(mddev->uuid+4, &sb->set_uuid1, 4);
1118 memcpy(mddev->uuid+8, &sb->set_uuid2, 4);
1119 memcpy(mddev->uuid+12,&sb->set_uuid3, 4);
1121 mddev->max_disks = MD_SB_DISKS;
1123 if (sb->state & (1<<MD_SB_BITMAP_PRESENT) &&
1124 mddev->bitmap_info.file == NULL)
1125 mddev->bitmap_info.offset =
1126 mddev->bitmap_info.default_offset;
1128 } else if (mddev->pers == NULL) {
1129 /* Insist on good event counter while assembling, except
1130 * for spares (which don't need an event count) */
1132 if (sb->disks[rdev->desc_nr].state & (
1133 (1<<MD_DISK_SYNC) | (1 << MD_DISK_ACTIVE)))
1134 if (ev1 < mddev->events)
1136 } else if (mddev->bitmap) {
1137 /* if adding to array with a bitmap, then we can accept an
1138 * older device ... but not too old.
1140 if (ev1 < mddev->bitmap->events_cleared)
1143 if (ev1 < mddev->events)
1144 /* just a hot-add of a new device, leave raid_disk at -1 */
1148 if (mddev->level != LEVEL_MULTIPATH) {
1149 desc = sb->disks + rdev->desc_nr;
1151 if (desc->state & (1<<MD_DISK_FAULTY))
1152 set_bit(Faulty, &rdev->flags);
1153 else if (desc->state & (1<<MD_DISK_SYNC) /* &&
1154 desc->raid_disk < mddev->raid_disks */) {
1155 set_bit(In_sync, &rdev->flags);
1156 rdev->raid_disk = desc->raid_disk;
1157 } else if (desc->state & (1<<MD_DISK_ACTIVE)) {
1158 /* active but not in sync implies recovery up to
1159 * reshape position. We don't know exactly where
1160 * that is, so set to zero for now */
1161 if (mddev->minor_version >= 91) {
1162 rdev->recovery_offset = 0;
1163 rdev->raid_disk = desc->raid_disk;
1166 if (desc->state & (1<<MD_DISK_WRITEMOSTLY))
1167 set_bit(WriteMostly, &rdev->flags);
1168 } else /* MULTIPATH are always insync */
1169 set_bit(In_sync, &rdev->flags);
1174 * sync_super for 0.90.0
1176 static void super_90_sync(mddev_t *mddev, mdk_rdev_t *rdev)
1180 int next_spare = mddev->raid_disks;
1183 /* make rdev->sb match mddev data..
1186 * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
1187 * 3/ any empty disks < next_spare become removed
1189 * disks[0] gets initialised to REMOVED because
1190 * we cannot be sure from other fields if it has
1191 * been initialised or not.
1194 int active=0, working=0,failed=0,spare=0,nr_disks=0;
1196 rdev->sb_size = MD_SB_BYTES;
1198 sb = (mdp_super_t*)page_address(rdev->sb_page);
1200 memset(sb, 0, sizeof(*sb));
1202 sb->md_magic = MD_SB_MAGIC;
1203 sb->major_version = mddev->major_version;
1204 sb->patch_version = mddev->patch_version;
1205 sb->gvalid_words = 0; /* ignored */
1206 memcpy(&sb->set_uuid0, mddev->uuid+0, 4);
1207 memcpy(&sb->set_uuid1, mddev->uuid+4, 4);
1208 memcpy(&sb->set_uuid2, mddev->uuid+8, 4);
1209 memcpy(&sb->set_uuid3, mddev->uuid+12,4);
1211 sb->ctime = mddev->ctime;
1212 sb->level = mddev->level;
1213 sb->size = mddev->dev_sectors / 2;
1214 sb->raid_disks = mddev->raid_disks;
1215 sb->md_minor = mddev->md_minor;
1216 sb->not_persistent = 0;
1217 sb->utime = mddev->utime;
1219 sb->events_hi = (mddev->events>>32);
1220 sb->events_lo = (u32)mddev->events;
1222 if (mddev->reshape_position == MaxSector)
1223 sb->minor_version = 90;
1225 sb->minor_version = 91;
1226 sb->reshape_position = mddev->reshape_position;
1227 sb->new_level = mddev->new_level;
1228 sb->delta_disks = mddev->delta_disks;
1229 sb->new_layout = mddev->new_layout;
1230 sb->new_chunk = mddev->new_chunk_sectors << 9;
1232 mddev->minor_version = sb->minor_version;
1235 sb->recovery_cp = mddev->recovery_cp;
1236 sb->cp_events_hi = (mddev->events>>32);
1237 sb->cp_events_lo = (u32)mddev->events;
1238 if (mddev->recovery_cp == MaxSector)
1239 sb->state = (1<< MD_SB_CLEAN);
1241 sb->recovery_cp = 0;
1243 sb->layout = mddev->layout;
1244 sb->chunk_size = mddev->chunk_sectors << 9;
1246 if (mddev->bitmap && mddev->bitmap_info.file == NULL)
1247 sb->state |= (1<<MD_SB_BITMAP_PRESENT);
1249 sb->disks[0].state = (1<<MD_DISK_REMOVED);
1250 list_for_each_entry(rdev2, &mddev->disks, same_set) {
1253 int is_active = test_bit(In_sync, &rdev2->flags);
1255 if (rdev2->raid_disk >= 0 &&
1256 sb->minor_version >= 91)
1257 /* we have nowhere to store the recovery_offset,
1258 * but if it is not below the reshape_position,
1259 * we can piggy-back on that.
1262 if (rdev2->raid_disk < 0 ||
1263 test_bit(Faulty, &rdev2->flags))
1266 desc_nr = rdev2->raid_disk;
1268 desc_nr = next_spare++;
1269 rdev2->desc_nr = desc_nr;
1270 d = &sb->disks[rdev2->desc_nr];
1272 d->number = rdev2->desc_nr;
1273 d->major = MAJOR(rdev2->bdev->bd_dev);
1274 d->minor = MINOR(rdev2->bdev->bd_dev);
1276 d->raid_disk = rdev2->raid_disk;
1278 d->raid_disk = rdev2->desc_nr; /* compatibility */
1279 if (test_bit(Faulty, &rdev2->flags))
1280 d->state = (1<<MD_DISK_FAULTY);
1281 else if (is_active) {
1282 d->state = (1<<MD_DISK_ACTIVE);
1283 if (test_bit(In_sync, &rdev2->flags))
1284 d->state |= (1<<MD_DISK_SYNC);
1292 if (test_bit(WriteMostly, &rdev2->flags))
1293 d->state |= (1<<MD_DISK_WRITEMOSTLY);
1295 /* now set the "removed" and "faulty" bits on any missing devices */
1296 for (i=0 ; i < mddev->raid_disks ; i++) {
1297 mdp_disk_t *d = &sb->disks[i];
1298 if (d->state == 0 && d->number == 0) {
1301 d->state = (1<<MD_DISK_REMOVED);
1302 d->state |= (1<<MD_DISK_FAULTY);
1306 sb->nr_disks = nr_disks;
1307 sb->active_disks = active;
1308 sb->working_disks = working;
1309 sb->failed_disks = failed;
1310 sb->spare_disks = spare;
1312 sb->this_disk = sb->disks[rdev->desc_nr];
1313 sb->sb_csum = calc_sb_csum(sb);
1317 * rdev_size_change for 0.90.0
1319 static unsigned long long
1320 super_90_rdev_size_change(mdk_rdev_t *rdev, sector_t num_sectors)
1322 if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
1323 return 0; /* component must fit device */
1324 if (rdev->mddev->bitmap_info.offset)
1325 return 0; /* can't move bitmap */
1326 rdev->sb_start = calc_dev_sboffset(rdev->bdev);
1327 if (!num_sectors || num_sectors > rdev->sb_start)
1328 num_sectors = rdev->sb_start;
1329 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
1331 md_super_wait(rdev->mddev);
1332 return num_sectors / 2; /* kB for sysfs */
1337 * version 1 superblock
1340 static __le32 calc_sb_1_csum(struct mdp_superblock_1 * sb)
1344 unsigned long long newcsum;
1345 int size = 256 + le32_to_cpu(sb->max_dev)*2;
1346 __le32 *isuper = (__le32*)sb;
1349 disk_csum = sb->sb_csum;
1352 for (i=0; size>=4; size -= 4 )
1353 newcsum += le32_to_cpu(*isuper++);
1356 newcsum += le16_to_cpu(*(__le16*) isuper);
1358 csum = (newcsum & 0xffffffff) + (newcsum >> 32);
1359 sb->sb_csum = disk_csum;
1360 return cpu_to_le32(csum);
1363 static int super_1_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
1365 struct mdp_superblock_1 *sb;
1368 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
1372 * Calculate the position of the superblock in 512byte sectors.
1373 * It is always aligned to a 4K boundary and
1374 * depeding on minor_version, it can be:
1375 * 0: At least 8K, but less than 12K, from end of device
1376 * 1: At start of device
1377 * 2: 4K from start of device.
1379 switch(minor_version) {
1381 sb_start = rdev->bdev->bd_inode->i_size >> 9;
1383 sb_start &= ~(sector_t)(4*2-1);
1394 rdev->sb_start = sb_start;
1396 /* superblock is rarely larger than 1K, but it can be larger,
1397 * and it is safe to read 4k, so we do that
1399 ret = read_disk_sb(rdev, 4096);
1400 if (ret) return ret;
1403 sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1405 if (sb->magic != cpu_to_le32(MD_SB_MAGIC) ||
1406 sb->major_version != cpu_to_le32(1) ||
1407 le32_to_cpu(sb->max_dev) > (4096-256)/2 ||
1408 le64_to_cpu(sb->super_offset) != rdev->sb_start ||
1409 (le32_to_cpu(sb->feature_map) & ~MD_FEATURE_ALL) != 0)
1412 if (calc_sb_1_csum(sb) != sb->sb_csum) {
1413 printk("md: invalid superblock checksum on %s\n",
1414 bdevname(rdev->bdev,b));
1417 if (le64_to_cpu(sb->data_size) < 10) {
1418 printk("md: data_size too small on %s\n",
1419 bdevname(rdev->bdev,b));
1423 rdev->preferred_minor = 0xffff;
1424 rdev->data_offset = le64_to_cpu(sb->data_offset);
1425 atomic_set(&rdev->corrected_errors, le32_to_cpu(sb->cnt_corrected_read));
1427 rdev->sb_size = le32_to_cpu(sb->max_dev) * 2 + 256;
1428 bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
1429 if (rdev->sb_size & bmask)
1430 rdev->sb_size = (rdev->sb_size | bmask) + 1;
1433 && rdev->data_offset < sb_start + (rdev->sb_size/512))
1436 if (sb->level == cpu_to_le32(LEVEL_MULTIPATH))
1439 rdev->desc_nr = le32_to_cpu(sb->dev_number);
1445 struct mdp_superblock_1 *refsb =
1446 (struct mdp_superblock_1*)page_address(refdev->sb_page);
1448 if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 ||
1449 sb->level != refsb->level ||
1450 sb->layout != refsb->layout ||
1451 sb->chunksize != refsb->chunksize) {
1452 printk(KERN_WARNING "md: %s has strangely different"
1453 " superblock to %s\n",
1454 bdevname(rdev->bdev,b),
1455 bdevname(refdev->bdev,b2));
1458 ev1 = le64_to_cpu(sb->events);
1459 ev2 = le64_to_cpu(refsb->events);
1467 rdev->sectors = (rdev->bdev->bd_inode->i_size >> 9) -
1468 le64_to_cpu(sb->data_offset);
1470 rdev->sectors = rdev->sb_start;
1471 if (rdev->sectors < le64_to_cpu(sb->data_size))
1473 rdev->sectors = le64_to_cpu(sb->data_size);
1474 if (le64_to_cpu(sb->size) > rdev->sectors)
1479 static int super_1_validate(mddev_t *mddev, mdk_rdev_t *rdev)
1481 struct mdp_superblock_1 *sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1482 __u64 ev1 = le64_to_cpu(sb->events);
1484 rdev->raid_disk = -1;
1485 clear_bit(Faulty, &rdev->flags);
1486 clear_bit(In_sync, &rdev->flags);
1487 clear_bit(WriteMostly, &rdev->flags);
1488 clear_bit(BarriersNotsupp, &rdev->flags);
1490 if (mddev->raid_disks == 0) {
1491 mddev->major_version = 1;
1492 mddev->patch_version = 0;
1493 mddev->external = 0;
1494 mddev->chunk_sectors = le32_to_cpu(sb->chunksize);
1495 mddev->ctime = le64_to_cpu(sb->ctime) & ((1ULL << 32)-1);
1496 mddev->utime = le64_to_cpu(sb->utime) & ((1ULL << 32)-1);
1497 mddev->level = le32_to_cpu(sb->level);
1498 mddev->clevel[0] = 0;
1499 mddev->layout = le32_to_cpu(sb->layout);
1500 mddev->raid_disks = le32_to_cpu(sb->raid_disks);
1501 mddev->dev_sectors = le64_to_cpu(sb->size);
1502 mddev->events = ev1;
1503 mddev->bitmap_info.offset = 0;
1504 mddev->bitmap_info.default_offset = 1024 >> 9;
1506 mddev->recovery_cp = le64_to_cpu(sb->resync_offset);
1507 memcpy(mddev->uuid, sb->set_uuid, 16);
1509 mddev->max_disks = (4096-256)/2;
1511 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) &&
1512 mddev->bitmap_info.file == NULL )
1513 mddev->bitmap_info.offset =
1514 (__s32)le32_to_cpu(sb->bitmap_offset);
1516 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) {
1517 mddev->reshape_position = le64_to_cpu(sb->reshape_position);
1518 mddev->delta_disks = le32_to_cpu(sb->delta_disks);
1519 mddev->new_level = le32_to_cpu(sb->new_level);
1520 mddev->new_layout = le32_to_cpu(sb->new_layout);
1521 mddev->new_chunk_sectors = le32_to_cpu(sb->new_chunk);
1523 mddev->reshape_position = MaxSector;
1524 mddev->delta_disks = 0;
1525 mddev->new_level = mddev->level;
1526 mddev->new_layout = mddev->layout;
1527 mddev->new_chunk_sectors = mddev->chunk_sectors;
1530 } else if (mddev->pers == NULL) {
1531 /* Insist of good event counter while assembling, except for
1532 * spares (which don't need an event count) */
1534 if (rdev->desc_nr >= 0 &&
1535 rdev->desc_nr < le32_to_cpu(sb->max_dev) &&
1536 le16_to_cpu(sb->dev_roles[rdev->desc_nr]) < 0xfffe)
1537 if (ev1 < mddev->events)
1539 } else if (mddev->bitmap) {
1540 /* If adding to array with a bitmap, then we can accept an
1541 * older device, but not too old.
1543 if (ev1 < mddev->bitmap->events_cleared)
1546 if (ev1 < mddev->events)
1547 /* just a hot-add of a new device, leave raid_disk at -1 */
1550 if (mddev->level != LEVEL_MULTIPATH) {
1552 if (rdev->desc_nr < 0 ||
1553 rdev->desc_nr >= le32_to_cpu(sb->max_dev)) {
1557 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
1559 case 0xffff: /* spare */
1561 case 0xfffe: /* faulty */
1562 set_bit(Faulty, &rdev->flags);
1565 if ((le32_to_cpu(sb->feature_map) &
1566 MD_FEATURE_RECOVERY_OFFSET))
1567 rdev->recovery_offset = le64_to_cpu(sb->recovery_offset);
1569 set_bit(In_sync, &rdev->flags);
1570 rdev->raid_disk = role;
1573 if (sb->devflags & WriteMostly1)
1574 set_bit(WriteMostly, &rdev->flags);
1575 } else /* MULTIPATH are always insync */
1576 set_bit(In_sync, &rdev->flags);
1581 static void super_1_sync(mddev_t *mddev, mdk_rdev_t *rdev)
1583 struct mdp_superblock_1 *sb;
1586 /* make rdev->sb match mddev and rdev data. */
1588 sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1590 sb->feature_map = 0;
1592 sb->recovery_offset = cpu_to_le64(0);
1593 memset(sb->pad1, 0, sizeof(sb->pad1));
1594 memset(sb->pad2, 0, sizeof(sb->pad2));
1595 memset(sb->pad3, 0, sizeof(sb->pad3));
1597 sb->utime = cpu_to_le64((__u64)mddev->utime);
1598 sb->events = cpu_to_le64(mddev->events);
1600 sb->resync_offset = cpu_to_le64(mddev->recovery_cp);
1602 sb->resync_offset = cpu_to_le64(0);
1604 sb->cnt_corrected_read = cpu_to_le32(atomic_read(&rdev->corrected_errors));
1606 sb->raid_disks = cpu_to_le32(mddev->raid_disks);
1607 sb->size = cpu_to_le64(mddev->dev_sectors);
1608 sb->chunksize = cpu_to_le32(mddev->chunk_sectors);
1609 sb->level = cpu_to_le32(mddev->level);
1610 sb->layout = cpu_to_le32(mddev->layout);
1612 if (mddev->bitmap && mddev->bitmap_info.file == NULL) {
1613 sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_info.offset);
1614 sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET);
1617 if (rdev->raid_disk >= 0 &&
1618 !test_bit(In_sync, &rdev->flags)) {
1620 cpu_to_le32(MD_FEATURE_RECOVERY_OFFSET);
1621 sb->recovery_offset =
1622 cpu_to_le64(rdev->recovery_offset);
1625 if (mddev->reshape_position != MaxSector) {
1626 sb->feature_map |= cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE);
1627 sb->reshape_position = cpu_to_le64(mddev->reshape_position);
1628 sb->new_layout = cpu_to_le32(mddev->new_layout);
1629 sb->delta_disks = cpu_to_le32(mddev->delta_disks);
1630 sb->new_level = cpu_to_le32(mddev->new_level);
1631 sb->new_chunk = cpu_to_le32(mddev->new_chunk_sectors);
1635 list_for_each_entry(rdev2, &mddev->disks, same_set)
1636 if (rdev2->desc_nr+1 > max_dev)
1637 max_dev = rdev2->desc_nr+1;
1639 if (max_dev > le32_to_cpu(sb->max_dev)) {
1641 sb->max_dev = cpu_to_le32(max_dev);
1642 rdev->sb_size = max_dev * 2 + 256;
1643 bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
1644 if (rdev->sb_size & bmask)
1645 rdev->sb_size = (rdev->sb_size | bmask) + 1;
1647 for (i=0; i<max_dev;i++)
1648 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1650 list_for_each_entry(rdev2, &mddev->disks, same_set) {
1652 if (test_bit(Faulty, &rdev2->flags))
1653 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1654 else if (test_bit(In_sync, &rdev2->flags))
1655 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1656 else if (rdev2->raid_disk >= 0)
1657 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1659 sb->dev_roles[i] = cpu_to_le16(0xffff);
1662 sb->sb_csum = calc_sb_1_csum(sb);
1665 static unsigned long long
1666 super_1_rdev_size_change(mdk_rdev_t *rdev, sector_t num_sectors)
1668 struct mdp_superblock_1 *sb;
1669 sector_t max_sectors;
1670 if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
1671 return 0; /* component must fit device */
1672 if (rdev->sb_start < rdev->data_offset) {
1673 /* minor versions 1 and 2; superblock before data */
1674 max_sectors = rdev->bdev->bd_inode->i_size >> 9;
1675 max_sectors -= rdev->data_offset;
1676 if (!num_sectors || num_sectors > max_sectors)
1677 num_sectors = max_sectors;
1678 } else if (rdev->mddev->bitmap_info.offset) {
1679 /* minor version 0 with bitmap we can't move */
1682 /* minor version 0; superblock after data */
1684 sb_start = (rdev->bdev->bd_inode->i_size >> 9) - 8*2;
1685 sb_start &= ~(sector_t)(4*2 - 1);
1686 max_sectors = rdev->sectors + sb_start - rdev->sb_start;
1687 if (!num_sectors || num_sectors > max_sectors)
1688 num_sectors = max_sectors;
1689 rdev->sb_start = sb_start;
1691 sb = (struct mdp_superblock_1 *) page_address(rdev->sb_page);
1692 sb->data_size = cpu_to_le64(num_sectors);
1693 sb->super_offset = rdev->sb_start;
1694 sb->sb_csum = calc_sb_1_csum(sb);
1695 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
1697 md_super_wait(rdev->mddev);
1698 return num_sectors / 2; /* kB for sysfs */
1701 static struct super_type super_types[] = {
1704 .owner = THIS_MODULE,
1705 .load_super = super_90_load,
1706 .validate_super = super_90_validate,
1707 .sync_super = super_90_sync,
1708 .rdev_size_change = super_90_rdev_size_change,
1712 .owner = THIS_MODULE,
1713 .load_super = super_1_load,
1714 .validate_super = super_1_validate,
1715 .sync_super = super_1_sync,
1716 .rdev_size_change = super_1_rdev_size_change,
1720 static int match_mddev_units(mddev_t *mddev1, mddev_t *mddev2)
1722 mdk_rdev_t *rdev, *rdev2;
1725 rdev_for_each_rcu(rdev, mddev1)
1726 rdev_for_each_rcu(rdev2, mddev2)
1727 if (rdev->bdev->bd_contains ==
1728 rdev2->bdev->bd_contains) {
1736 static LIST_HEAD(pending_raid_disks);
1739 * Try to register data integrity profile for an mddev
1741 * This is called when an array is started and after a disk has been kicked
1742 * from the array. It only succeeds if all working and active component devices
1743 * are integrity capable with matching profiles.
1745 int md_integrity_register(mddev_t *mddev)
1747 mdk_rdev_t *rdev, *reference = NULL;
1749 if (list_empty(&mddev->disks))
1750 return 0; /* nothing to do */
1751 if (blk_get_integrity(mddev->gendisk))
1752 return 0; /* already registered */
1753 list_for_each_entry(rdev, &mddev->disks, same_set) {
1754 /* skip spares and non-functional disks */
1755 if (test_bit(Faulty, &rdev->flags))
1757 if (rdev->raid_disk < 0)
1760 * If at least one rdev is not integrity capable, we can not
1761 * enable data integrity for the md device.
1763 if (!bdev_get_integrity(rdev->bdev))
1766 /* Use the first rdev as the reference */
1770 /* does this rdev's profile match the reference profile? */
1771 if (blk_integrity_compare(reference->bdev->bd_disk,
1772 rdev->bdev->bd_disk) < 0)
1776 * All component devices are integrity capable and have matching
1777 * profiles, register the common profile for the md device.
1779 if (blk_integrity_register(mddev->gendisk,
1780 bdev_get_integrity(reference->bdev)) != 0) {
1781 printk(KERN_ERR "md: failed to register integrity for %s\n",
1785 printk(KERN_NOTICE "md: data integrity on %s enabled\n",
1789 EXPORT_SYMBOL(md_integrity_register);
1791 /* Disable data integrity if non-capable/non-matching disk is being added */
1792 void md_integrity_add_rdev(mdk_rdev_t *rdev, mddev_t *mddev)
1794 struct blk_integrity *bi_rdev = bdev_get_integrity(rdev->bdev);
1795 struct blk_integrity *bi_mddev = blk_get_integrity(mddev->gendisk);
1797 if (!bi_mddev) /* nothing to do */
1799 if (rdev->raid_disk < 0) /* skip spares */
1801 if (bi_rdev && blk_integrity_compare(mddev->gendisk,
1802 rdev->bdev->bd_disk) >= 0)
1804 printk(KERN_NOTICE "disabling data integrity on %s\n", mdname(mddev));
1805 blk_integrity_unregister(mddev->gendisk);
1807 EXPORT_SYMBOL(md_integrity_add_rdev);
1809 static int bind_rdev_to_array(mdk_rdev_t * rdev, mddev_t * mddev)
1811 char b[BDEVNAME_SIZE];
1821 /* prevent duplicates */
1822 if (find_rdev(mddev, rdev->bdev->bd_dev))
1825 /* make sure rdev->sectors exceeds mddev->dev_sectors */
1826 if (rdev->sectors && (mddev->dev_sectors == 0 ||
1827 rdev->sectors < mddev->dev_sectors)) {
1829 /* Cannot change size, so fail
1830 * If mddev->level <= 0, then we don't care
1831 * about aligning sizes (e.g. linear)
1833 if (mddev->level > 0)
1836 mddev->dev_sectors = rdev->sectors;
1839 /* Verify rdev->desc_nr is unique.
1840 * If it is -1, assign a free number, else
1841 * check number is not in use
1843 if (rdev->desc_nr < 0) {
1845 if (mddev->pers) choice = mddev->raid_disks;
1846 while (find_rdev_nr(mddev, choice))
1848 rdev->desc_nr = choice;
1850 if (find_rdev_nr(mddev, rdev->desc_nr))
1853 if (mddev->max_disks && rdev->desc_nr >= mddev->max_disks) {
1854 printk(KERN_WARNING "md: %s: array is limited to %d devices\n",
1855 mdname(mddev), mddev->max_disks);
1858 bdevname(rdev->bdev,b);
1859 while ( (s=strchr(b, '/')) != NULL)
1862 rdev->mddev = mddev;
1863 printk(KERN_INFO "md: bind<%s>\n", b);
1865 if ((err = kobject_add(&rdev->kobj, &mddev->kobj, "dev-%s", b)))
1868 ko = &part_to_dev(rdev->bdev->bd_part)->kobj;
1869 if (sysfs_create_link(&rdev->kobj, ko, "block"))
1870 /* failure here is OK */;
1871 rdev->sysfs_state = sysfs_get_dirent_safe(rdev->kobj.sd, "state");
1873 list_add_rcu(&rdev->same_set, &mddev->disks);
1874 bd_claim_by_disk(rdev->bdev, rdev->bdev->bd_holder, mddev->gendisk);
1876 /* May as well allow recovery to be retried once */
1877 mddev->recovery_disabled = 0;
1882 printk(KERN_WARNING "md: failed to register dev-%s for %s\n",
1887 static void md_delayed_delete(struct work_struct *ws)
1889 mdk_rdev_t *rdev = container_of(ws, mdk_rdev_t, del_work);
1890 kobject_del(&rdev->kobj);
1891 kobject_put(&rdev->kobj);
1894 static void unbind_rdev_from_array(mdk_rdev_t * rdev)
1896 char b[BDEVNAME_SIZE];
1901 bd_release_from_disk(rdev->bdev, rdev->mddev->gendisk);
1902 list_del_rcu(&rdev->same_set);
1903 printk(KERN_INFO "md: unbind<%s>\n", bdevname(rdev->bdev,b));
1905 sysfs_remove_link(&rdev->kobj, "block");
1906 sysfs_put(rdev->sysfs_state);
1907 rdev->sysfs_state = NULL;
1908 /* We need to delay this, otherwise we can deadlock when
1909 * writing to 'remove' to "dev/state". We also need
1910 * to delay it due to rcu usage.
1913 INIT_WORK(&rdev->del_work, md_delayed_delete);
1914 kobject_get(&rdev->kobj);
1915 schedule_work(&rdev->del_work);
1919 * prevent the device from being mounted, repartitioned or
1920 * otherwise reused by a RAID array (or any other kernel
1921 * subsystem), by bd_claiming the device.
1923 static int lock_rdev(mdk_rdev_t *rdev, dev_t dev, int shared)
1926 struct block_device *bdev;
1927 char b[BDEVNAME_SIZE];
1929 bdev = open_by_devnum(dev, FMODE_READ|FMODE_WRITE);
1931 printk(KERN_ERR "md: could not open %s.\n",
1932 __bdevname(dev, b));
1933 return PTR_ERR(bdev);
1935 err = bd_claim(bdev, shared ? (mdk_rdev_t *)lock_rdev : rdev);
1937 printk(KERN_ERR "md: could not bd_claim %s.\n",
1939 blkdev_put(bdev, FMODE_READ|FMODE_WRITE);
1943 set_bit(AllReserved, &rdev->flags);
1948 static void unlock_rdev(mdk_rdev_t *rdev)
1950 struct block_device *bdev = rdev->bdev;
1955 blkdev_put(bdev, FMODE_READ|FMODE_WRITE);
1958 void md_autodetect_dev(dev_t dev);
1960 static void export_rdev(mdk_rdev_t * rdev)
1962 char b[BDEVNAME_SIZE];
1963 printk(KERN_INFO "md: export_rdev(%s)\n",
1964 bdevname(rdev->bdev,b));
1969 if (test_bit(AutoDetected, &rdev->flags))
1970 md_autodetect_dev(rdev->bdev->bd_dev);
1973 kobject_put(&rdev->kobj);
1976 static void kick_rdev_from_array(mdk_rdev_t * rdev)
1978 unbind_rdev_from_array(rdev);
1982 static void export_array(mddev_t *mddev)
1984 mdk_rdev_t *rdev, *tmp;
1986 rdev_for_each(rdev, tmp, mddev) {
1991 kick_rdev_from_array(rdev);
1993 if (!list_empty(&mddev->disks))
1995 mddev->raid_disks = 0;
1996 mddev->major_version = 0;
1999 static void print_desc(mdp_disk_t *desc)
2001 printk(" DISK<N:%d,(%d,%d),R:%d,S:%d>\n", desc->number,
2002 desc->major,desc->minor,desc->raid_disk,desc->state);
2005 static void print_sb_90(mdp_super_t *sb)
2010 "md: SB: (V:%d.%d.%d) ID:<%08x.%08x.%08x.%08x> CT:%08x\n",
2011 sb->major_version, sb->minor_version, sb->patch_version,
2012 sb->set_uuid0, sb->set_uuid1, sb->set_uuid2, sb->set_uuid3,
2014 printk(KERN_INFO "md: L%d S%08d ND:%d RD:%d md%d LO:%d CS:%d\n",
2015 sb->level, sb->size, sb->nr_disks, sb->raid_disks,
2016 sb->md_minor, sb->layout, sb->chunk_size);
2017 printk(KERN_INFO "md: UT:%08x ST:%d AD:%d WD:%d"
2018 " FD:%d SD:%d CSUM:%08x E:%08lx\n",
2019 sb->utime, sb->state, sb->active_disks, sb->working_disks,
2020 sb->failed_disks, sb->spare_disks,
2021 sb->sb_csum, (unsigned long)sb->events_lo);
2024 for (i = 0; i < MD_SB_DISKS; i++) {
2027 desc = sb->disks + i;
2028 if (desc->number || desc->major || desc->minor ||
2029 desc->raid_disk || (desc->state && (desc->state != 4))) {
2030 printk(" D %2d: ", i);
2034 printk(KERN_INFO "md: THIS: ");
2035 print_desc(&sb->this_disk);
2038 static void print_sb_1(struct mdp_superblock_1 *sb)
2042 uuid = sb->set_uuid;
2044 "md: SB: (V:%u) (F:0x%08x) Array-ID:<%pU>\n"
2045 "md: Name: \"%s\" CT:%llu\n",
2046 le32_to_cpu(sb->major_version),
2047 le32_to_cpu(sb->feature_map),
2050 (unsigned long long)le64_to_cpu(sb->ctime)
2051 & MD_SUPERBLOCK_1_TIME_SEC_MASK);
2053 uuid = sb->device_uuid;
2055 "md: L%u SZ%llu RD:%u LO:%u CS:%u DO:%llu DS:%llu SO:%llu"
2057 "md: Dev:%08x UUID: %pU\n"
2058 "md: (F:0x%08x) UT:%llu Events:%llu ResyncOffset:%llu CSUM:0x%08x\n"
2059 "md: (MaxDev:%u) \n",
2060 le32_to_cpu(sb->level),
2061 (unsigned long long)le64_to_cpu(sb->size),
2062 le32_to_cpu(sb->raid_disks),
2063 le32_to_cpu(sb->layout),
2064 le32_to_cpu(sb->chunksize),
2065 (unsigned long long)le64_to_cpu(sb->data_offset),
2066 (unsigned long long)le64_to_cpu(sb->data_size),
2067 (unsigned long long)le64_to_cpu(sb->super_offset),
2068 (unsigned long long)le64_to_cpu(sb->recovery_offset),
2069 le32_to_cpu(sb->dev_number),
2072 (unsigned long long)le64_to_cpu(sb->utime) & MD_SUPERBLOCK_1_TIME_SEC_MASK,
2073 (unsigned long long)le64_to_cpu(sb->events),
2074 (unsigned long long)le64_to_cpu(sb->resync_offset),
2075 le32_to_cpu(sb->sb_csum),
2076 le32_to_cpu(sb->max_dev)
2080 static void print_rdev(mdk_rdev_t *rdev, int major_version)
2082 char b[BDEVNAME_SIZE];
2083 printk(KERN_INFO "md: rdev %s, Sect:%08llu F:%d S:%d DN:%u\n",
2084 bdevname(rdev->bdev, b), (unsigned long long)rdev->sectors,
2085 test_bit(Faulty, &rdev->flags), test_bit(In_sync, &rdev->flags),
2087 if (rdev->sb_loaded) {
2088 printk(KERN_INFO "md: rdev superblock (MJ:%d):\n", major_version);
2089 switch (major_version) {
2091 print_sb_90((mdp_super_t*)page_address(rdev->sb_page));
2094 print_sb_1((struct mdp_superblock_1 *)page_address(rdev->sb_page));
2098 printk(KERN_INFO "md: no rdev superblock!\n");
2101 static void md_print_devices(void)
2103 struct list_head *tmp;
2106 char b[BDEVNAME_SIZE];
2109 printk("md: **********************************\n");
2110 printk("md: * <COMPLETE RAID STATE PRINTOUT> *\n");
2111 printk("md: **********************************\n");
2112 for_each_mddev(mddev, tmp) {
2115 bitmap_print_sb(mddev->bitmap);
2117 printk("%s: ", mdname(mddev));
2118 list_for_each_entry(rdev, &mddev->disks, same_set)
2119 printk("<%s>", bdevname(rdev->bdev,b));
2122 list_for_each_entry(rdev, &mddev->disks, same_set)
2123 print_rdev(rdev, mddev->major_version);
2125 printk("md: **********************************\n");
2130 static void sync_sbs(mddev_t * mddev, int nospares)
2132 /* Update each superblock (in-memory image), but
2133 * if we are allowed to, skip spares which already
2134 * have the right event counter, or have one earlier
2135 * (which would mean they aren't being marked as dirty
2136 * with the rest of the array)
2140 /* First make sure individual recovery_offsets are correct */
2141 list_for_each_entry(rdev, &mddev->disks, same_set) {
2142 if (rdev->raid_disk >= 0 &&
2143 mddev->delta_disks >= 0 &&
2144 !test_bit(In_sync, &rdev->flags) &&
2145 mddev->curr_resync_completed > rdev->recovery_offset)
2146 rdev->recovery_offset = mddev->curr_resync_completed;
2149 list_for_each_entry(rdev, &mddev->disks, same_set) {
2150 if (rdev->sb_events == mddev->events ||
2152 rdev->raid_disk < 0 &&
2153 rdev->sb_events+1 == mddev->events)) {
2154 /* Don't update this superblock */
2155 rdev->sb_loaded = 2;
2157 super_types[mddev->major_version].
2158 sync_super(mddev, rdev);
2159 rdev->sb_loaded = 1;
2164 static void md_update_sb(mddev_t * mddev, int force_change)
2170 mddev->utime = get_seconds();
2171 if (mddev->external)
2174 spin_lock_irq(&mddev->write_lock);
2176 set_bit(MD_CHANGE_PENDING, &mddev->flags);
2177 if (test_and_clear_bit(MD_CHANGE_DEVS, &mddev->flags))
2179 if (test_and_clear_bit(MD_CHANGE_CLEAN, &mddev->flags))
2180 /* just a clean<-> dirty transition, possibly leave spares alone,
2181 * though if events isn't the right even/odd, we will have to do
2187 if (mddev->degraded)
2188 /* If the array is degraded, then skipping spares is both
2189 * dangerous and fairly pointless.
2190 * Dangerous because a device that was removed from the array
2191 * might have a event_count that still looks up-to-date,
2192 * so it can be re-added without a resync.
2193 * Pointless because if there are any spares to skip,
2194 * then a recovery will happen and soon that array won't
2195 * be degraded any more and the spare can go back to sleep then.
2199 sync_req = mddev->in_sync;
2201 /* If this is just a dirty<->clean transition, and the array is clean
2202 * and 'events' is odd, we can roll back to the previous clean state */
2204 && (mddev->in_sync && mddev->recovery_cp == MaxSector)
2205 && mddev->can_decrease_events
2206 && mddev->events != 1) {
2208 mddev->can_decrease_events = 0;
2210 /* otherwise we have to go forward and ... */
2212 mddev->can_decrease_events = nospares;
2215 if (!mddev->events) {
2217 * oops, this 64-bit counter should never wrap.
2218 * Either we are in around ~1 trillion A.C., assuming
2219 * 1 reboot per second, or we have a bug:
2226 * do not write anything to disk if using
2227 * nonpersistent superblocks
2229 if (!mddev->persistent) {
2230 if (!mddev->external)
2231 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
2233 spin_unlock_irq(&mddev->write_lock);
2234 wake_up(&mddev->sb_wait);
2237 sync_sbs(mddev, nospares);
2238 spin_unlock_irq(&mddev->write_lock);
2241 "md: updating %s RAID superblock on device (in sync %d)\n",
2242 mdname(mddev),mddev->in_sync);
2244 bitmap_update_sb(mddev->bitmap);
2245 list_for_each_entry(rdev, &mddev->disks, same_set) {
2246 char b[BDEVNAME_SIZE];
2247 dprintk(KERN_INFO "md: ");
2248 if (rdev->sb_loaded != 1)
2249 continue; /* no noise on spare devices */
2250 if (test_bit(Faulty, &rdev->flags))
2251 dprintk("(skipping faulty ");
2253 dprintk("%s ", bdevname(rdev->bdev,b));
2254 if (!test_bit(Faulty, &rdev->flags)) {
2255 md_super_write(mddev,rdev,
2256 rdev->sb_start, rdev->sb_size,
2258 dprintk(KERN_INFO "(write) %s's sb offset: %llu\n",
2259 bdevname(rdev->bdev,b),
2260 (unsigned long long)rdev->sb_start);
2261 rdev->sb_events = mddev->events;
2265 if (mddev->level == LEVEL_MULTIPATH)
2266 /* only need to write one superblock... */
2269 md_super_wait(mddev);
2270 /* if there was a failure, MD_CHANGE_DEVS was set, and we re-write super */
2272 spin_lock_irq(&mddev->write_lock);
2273 if (mddev->in_sync != sync_req ||
2274 test_bit(MD_CHANGE_DEVS, &mddev->flags)) {
2275 /* have to write it out again */
2276 spin_unlock_irq(&mddev->write_lock);
2279 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
2280 spin_unlock_irq(&mddev->write_lock);
2281 wake_up(&mddev->sb_wait);
2282 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
2283 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
2287 /* words written to sysfs files may, or may not, be \n terminated.
2288 * We want to accept with case. For this we use cmd_match.
2290 static int cmd_match(const char *cmd, const char *str)
2292 /* See if cmd, written into a sysfs file, matches
2293 * str. They must either be the same, or cmd can
2294 * have a trailing newline
2296 while (*cmd && *str && *cmd == *str) {
2307 struct rdev_sysfs_entry {
2308 struct attribute attr;
2309 ssize_t (*show)(mdk_rdev_t *, char *);
2310 ssize_t (*store)(mdk_rdev_t *, const char *, size_t);
2314 state_show(mdk_rdev_t *rdev, char *page)
2319 if (test_bit(Faulty, &rdev->flags)) {
2320 len+= sprintf(page+len, "%sfaulty",sep);
2323 if (test_bit(In_sync, &rdev->flags)) {
2324 len += sprintf(page+len, "%sin_sync",sep);
2327 if (test_bit(WriteMostly, &rdev->flags)) {
2328 len += sprintf(page+len, "%swrite_mostly",sep);
2331 if (test_bit(Blocked, &rdev->flags)) {
2332 len += sprintf(page+len, "%sblocked", sep);
2335 if (!test_bit(Faulty, &rdev->flags) &&
2336 !test_bit(In_sync, &rdev->flags)) {
2337 len += sprintf(page+len, "%sspare", sep);
2340 return len+sprintf(page+len, "\n");
2344 state_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2347 * faulty - simulates and error
2348 * remove - disconnects the device
2349 * writemostly - sets write_mostly
2350 * -writemostly - clears write_mostly
2351 * blocked - sets the Blocked flag
2352 * -blocked - clears the Blocked flag
2353 * insync - sets Insync providing device isn't active
2356 if (cmd_match(buf, "faulty") && rdev->mddev->pers) {
2357 md_error(rdev->mddev, rdev);
2359 } else if (cmd_match(buf, "remove")) {
2360 if (rdev->raid_disk >= 0)
2363 mddev_t *mddev = rdev->mddev;
2364 kick_rdev_from_array(rdev);
2366 md_update_sb(mddev, 1);
2367 md_new_event(mddev);
2370 } else if (cmd_match(buf, "writemostly")) {
2371 set_bit(WriteMostly, &rdev->flags);
2373 } else if (cmd_match(buf, "-writemostly")) {
2374 clear_bit(WriteMostly, &rdev->flags);
2376 } else if (cmd_match(buf, "blocked")) {
2377 set_bit(Blocked, &rdev->flags);
2379 } else if (cmd_match(buf, "-blocked")) {
2380 clear_bit(Blocked, &rdev->flags);
2381 wake_up(&rdev->blocked_wait);
2382 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2383 md_wakeup_thread(rdev->mddev->thread);
2386 } else if (cmd_match(buf, "insync") && rdev->raid_disk == -1) {
2387 set_bit(In_sync, &rdev->flags);
2391 sysfs_notify_dirent_safe(rdev->sysfs_state);
2392 return err ? err : len;
2394 static struct rdev_sysfs_entry rdev_state =
2395 __ATTR(state, S_IRUGO|S_IWUSR, state_show, state_store);
2398 errors_show(mdk_rdev_t *rdev, char *page)
2400 return sprintf(page, "%d\n", atomic_read(&rdev->corrected_errors));
2404 errors_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2407 unsigned long n = simple_strtoul(buf, &e, 10);
2408 if (*buf && (*e == 0 || *e == '\n')) {
2409 atomic_set(&rdev->corrected_errors, n);
2414 static struct rdev_sysfs_entry rdev_errors =
2415 __ATTR(errors, S_IRUGO|S_IWUSR, errors_show, errors_store);
2418 slot_show(mdk_rdev_t *rdev, char *page)
2420 if (rdev->raid_disk < 0)
2421 return sprintf(page, "none\n");
2423 return sprintf(page, "%d\n", rdev->raid_disk);
2427 slot_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2432 int slot = simple_strtoul(buf, &e, 10);
2433 if (strncmp(buf, "none", 4)==0)
2435 else if (e==buf || (*e && *e!= '\n'))
2437 if (rdev->mddev->pers && slot == -1) {
2438 /* Setting 'slot' on an active array requires also
2439 * updating the 'rd%d' link, and communicating
2440 * with the personality with ->hot_*_disk.
2441 * For now we only support removing
2442 * failed/spare devices. This normally happens automatically,
2443 * but not when the metadata is externally managed.
2445 if (rdev->raid_disk == -1)
2447 /* personality does all needed checks */
2448 if (rdev->mddev->pers->hot_add_disk == NULL)
2450 err = rdev->mddev->pers->
2451 hot_remove_disk(rdev->mddev, rdev->raid_disk);
2454 sprintf(nm, "rd%d", rdev->raid_disk);
2455 sysfs_remove_link(&rdev->mddev->kobj, nm);
2456 rdev->raid_disk = -1;
2457 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2458 md_wakeup_thread(rdev->mddev->thread);
2459 } else if (rdev->mddev->pers) {
2461 /* Activating a spare .. or possibly reactivating
2462 * if we ever get bitmaps working here.
2465 if (rdev->raid_disk != -1)
2468 if (rdev->mddev->pers->hot_add_disk == NULL)
2471 list_for_each_entry(rdev2, &rdev->mddev->disks, same_set)
2472 if (rdev2->raid_disk == slot)
2475 rdev->raid_disk = slot;
2476 if (test_bit(In_sync, &rdev->flags))
2477 rdev->saved_raid_disk = slot;
2479 rdev->saved_raid_disk = -1;
2480 err = rdev->mddev->pers->
2481 hot_add_disk(rdev->mddev, rdev);
2483 rdev->raid_disk = -1;
2486 sysfs_notify_dirent_safe(rdev->sysfs_state);
2487 sprintf(nm, "rd%d", rdev->raid_disk);
2488 if (sysfs_create_link(&rdev->mddev->kobj, &rdev->kobj, nm))
2489 /* failure here is OK */;
2490 /* don't wakeup anyone, leave that to userspace. */
2492 if (slot >= rdev->mddev->raid_disks)
2494 rdev->raid_disk = slot;
2495 /* assume it is working */
2496 clear_bit(Faulty, &rdev->flags);
2497 clear_bit(WriteMostly, &rdev->flags);
2498 set_bit(In_sync, &rdev->flags);
2499 sysfs_notify_dirent_safe(rdev->sysfs_state);
2505 static struct rdev_sysfs_entry rdev_slot =
2506 __ATTR(slot, S_IRUGO|S_IWUSR, slot_show, slot_store);
2509 offset_show(mdk_rdev_t *rdev, char *page)
2511 return sprintf(page, "%llu\n", (unsigned long long)rdev->data_offset);
2515 offset_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2518 unsigned long long offset = simple_strtoull(buf, &e, 10);
2519 if (e==buf || (*e && *e != '\n'))
2521 if (rdev->mddev->pers && rdev->raid_disk >= 0)
2523 if (rdev->sectors && rdev->mddev->external)
2524 /* Must set offset before size, so overlap checks
2527 rdev->data_offset = offset;
2531 static struct rdev_sysfs_entry rdev_offset =
2532 __ATTR(offset, S_IRUGO|S_IWUSR, offset_show, offset_store);
2535 rdev_size_show(mdk_rdev_t *rdev, char *page)
2537 return sprintf(page, "%llu\n", (unsigned long long)rdev->sectors / 2);
2540 static int overlaps(sector_t s1, sector_t l1, sector_t s2, sector_t l2)
2542 /* check if two start/length pairs overlap */
2550 static int strict_blocks_to_sectors(const char *buf, sector_t *sectors)
2552 unsigned long long blocks;
2555 if (strict_strtoull(buf, 10, &blocks) < 0)
2558 if (blocks & 1ULL << (8 * sizeof(blocks) - 1))
2559 return -EINVAL; /* sector conversion overflow */
2562 if (new != blocks * 2)
2563 return -EINVAL; /* unsigned long long to sector_t overflow */
2570 rdev_size_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2572 mddev_t *my_mddev = rdev->mddev;
2573 sector_t oldsectors = rdev->sectors;
2576 if (strict_blocks_to_sectors(buf, §ors) < 0)
2578 if (my_mddev->pers && rdev->raid_disk >= 0) {
2579 if (my_mddev->persistent) {
2580 sectors = super_types[my_mddev->major_version].
2581 rdev_size_change(rdev, sectors);
2584 } else if (!sectors)
2585 sectors = (rdev->bdev->bd_inode->i_size >> 9) -
2588 if (sectors < my_mddev->dev_sectors)
2589 return -EINVAL; /* component must fit device */
2591 rdev->sectors = sectors;
2592 if (sectors > oldsectors && my_mddev->external) {
2593 /* need to check that all other rdevs with the same ->bdev
2594 * do not overlap. We need to unlock the mddev to avoid
2595 * a deadlock. We have already changed rdev->sectors, and if
2596 * we have to change it back, we will have the lock again.
2600 struct list_head *tmp;
2602 mddev_unlock(my_mddev);
2603 for_each_mddev(mddev, tmp) {
2607 list_for_each_entry(rdev2, &mddev->disks, same_set)
2608 if (test_bit(AllReserved, &rdev2->flags) ||
2609 (rdev->bdev == rdev2->bdev &&
2611 overlaps(rdev->data_offset, rdev->sectors,
2617 mddev_unlock(mddev);
2623 mddev_lock(my_mddev);
2625 /* Someone else could have slipped in a size
2626 * change here, but doing so is just silly.
2627 * We put oldsectors back because we *know* it is
2628 * safe, and trust userspace not to race with
2631 rdev->sectors = oldsectors;
2638 static struct rdev_sysfs_entry rdev_size =
2639 __ATTR(size, S_IRUGO|S_IWUSR, rdev_size_show, rdev_size_store);
2642 static ssize_t recovery_start_show(mdk_rdev_t *rdev, char *page)
2644 unsigned long long recovery_start = rdev->recovery_offset;
2646 if (test_bit(In_sync, &rdev->flags) ||
2647 recovery_start == MaxSector)
2648 return sprintf(page, "none\n");
2650 return sprintf(page, "%llu\n", recovery_start);
2653 static ssize_t recovery_start_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2655 unsigned long long recovery_start;
2657 if (cmd_match(buf, "none"))
2658 recovery_start = MaxSector;
2659 else if (strict_strtoull(buf, 10, &recovery_start))
2662 if (rdev->mddev->pers &&
2663 rdev->raid_disk >= 0)
2666 rdev->recovery_offset = recovery_start;
2667 if (recovery_start == MaxSector)
2668 set_bit(In_sync, &rdev->flags);
2670 clear_bit(In_sync, &rdev->flags);
2674 static struct rdev_sysfs_entry rdev_recovery_start =
2675 __ATTR(recovery_start, S_IRUGO|S_IWUSR, recovery_start_show, recovery_start_store);
2677 static struct attribute *rdev_default_attrs[] = {
2683 &rdev_recovery_start.attr,
2687 rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
2689 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
2690 mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
2691 mddev_t *mddev = rdev->mddev;
2697 rv = mddev ? mddev_lock(mddev) : -EBUSY;
2699 if (rdev->mddev == NULL)
2702 rv = entry->show(rdev, page);
2703 mddev_unlock(mddev);
2709 rdev_attr_store(struct kobject *kobj, struct attribute *attr,
2710 const char *page, size_t length)
2712 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
2713 mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
2715 mddev_t *mddev = rdev->mddev;
2719 if (!capable(CAP_SYS_ADMIN))
2721 rv = mddev ? mddev_lock(mddev): -EBUSY;
2723 if (rdev->mddev == NULL)
2726 rv = entry->store(rdev, page, length);
2727 mddev_unlock(mddev);
2732 static void rdev_free(struct kobject *ko)
2734 mdk_rdev_t *rdev = container_of(ko, mdk_rdev_t, kobj);
2737 static const struct sysfs_ops rdev_sysfs_ops = {
2738 .show = rdev_attr_show,
2739 .store = rdev_attr_store,
2741 static struct kobj_type rdev_ktype = {
2742 .release = rdev_free,
2743 .sysfs_ops = &rdev_sysfs_ops,
2744 .default_attrs = rdev_default_attrs,
2747 void md_rdev_init(mdk_rdev_t *rdev)
2750 rdev->saved_raid_disk = -1;
2751 rdev->raid_disk = -1;
2753 rdev->data_offset = 0;
2754 rdev->sb_events = 0;
2755 rdev->last_read_error.tv_sec = 0;
2756 rdev->last_read_error.tv_nsec = 0;
2757 atomic_set(&rdev->nr_pending, 0);
2758 atomic_set(&rdev->read_errors, 0);
2759 atomic_set(&rdev->corrected_errors, 0);
2761 INIT_LIST_HEAD(&rdev->same_set);
2762 init_waitqueue_head(&rdev->blocked_wait);
2764 EXPORT_SYMBOL_GPL(md_rdev_init);
2766 * Import a device. If 'super_format' >= 0, then sanity check the superblock
2768 * mark the device faulty if:
2770 * - the device is nonexistent (zero size)
2771 * - the device has no valid superblock
2773 * a faulty rdev _never_ has rdev->sb set.
2775 static mdk_rdev_t *md_import_device(dev_t newdev, int super_format, int super_minor)
2777 char b[BDEVNAME_SIZE];
2782 rdev = kzalloc(sizeof(*rdev), GFP_KERNEL);
2784 printk(KERN_ERR "md: could not alloc mem for new device!\n");
2785 return ERR_PTR(-ENOMEM);
2789 if ((err = alloc_disk_sb(rdev)))
2792 err = lock_rdev(rdev, newdev, super_format == -2);
2796 kobject_init(&rdev->kobj, &rdev_ktype);
2798 size = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
2801 "md: %s has zero or unknown size, marking faulty!\n",
2802 bdevname(rdev->bdev,b));
2807 if (super_format >= 0) {
2808 err = super_types[super_format].
2809 load_super(rdev, NULL, super_minor);
2810 if (err == -EINVAL) {
2812 "md: %s does not have a valid v%d.%d "
2813 "superblock, not importing!\n",
2814 bdevname(rdev->bdev,b),
2815 super_format, super_minor);
2820 "md: could not read %s's sb, not importing!\n",
2821 bdevname(rdev->bdev,b));
2829 if (rdev->sb_page) {
2835 return ERR_PTR(err);
2839 * Check a full RAID array for plausibility
2843 static void analyze_sbs(mddev_t * mddev)
2846 mdk_rdev_t *rdev, *freshest, *tmp;
2847 char b[BDEVNAME_SIZE];
2850 rdev_for_each(rdev, tmp, mddev)
2851 switch (super_types[mddev->major_version].
2852 load_super(rdev, freshest, mddev->minor_version)) {
2860 "md: fatal superblock inconsistency in %s"
2861 " -- removing from array\n",
2862 bdevname(rdev->bdev,b));
2863 kick_rdev_from_array(rdev);
2867 super_types[mddev->major_version].
2868 validate_super(mddev, freshest);
2871 rdev_for_each(rdev, tmp, mddev) {
2872 if (mddev->max_disks &&
2873 (rdev->desc_nr >= mddev->max_disks ||
2874 i > mddev->max_disks)) {
2876 "md: %s: %s: only %d devices permitted\n",
2877 mdname(mddev), bdevname(rdev->bdev, b),
2879 kick_rdev_from_array(rdev);
2882 if (rdev != freshest)
2883 if (super_types[mddev->major_version].
2884 validate_super(mddev, rdev)) {
2885 printk(KERN_WARNING "md: kicking non-fresh %s"
2887 bdevname(rdev->bdev,b));
2888 kick_rdev_from_array(rdev);
2891 if (mddev->level == LEVEL_MULTIPATH) {
2892 rdev->desc_nr = i++;
2893 rdev->raid_disk = rdev->desc_nr;
2894 set_bit(In_sync, &rdev->flags);
2895 } else if (rdev->raid_disk >= (mddev->raid_disks - min(0, mddev->delta_disks))) {
2896 rdev->raid_disk = -1;
2897 clear_bit(In_sync, &rdev->flags);
2902 /* Read a fixed-point number.
2903 * Numbers in sysfs attributes should be in "standard" units where
2904 * possible, so time should be in seconds.
2905 * However we internally use a a much smaller unit such as
2906 * milliseconds or jiffies.
2907 * This function takes a decimal number with a possible fractional
2908 * component, and produces an integer which is the result of
2909 * multiplying that number by 10^'scale'.
2910 * all without any floating-point arithmetic.
2912 int strict_strtoul_scaled(const char *cp, unsigned long *res, int scale)
2914 unsigned long result = 0;
2916 while (isdigit(*cp) || (*cp == '.' && decimals < 0)) {
2919 else if (decimals < scale) {
2922 result = result * 10 + value;
2934 while (decimals < scale) {
2943 static void md_safemode_timeout(unsigned long data);
2946 safe_delay_show(mddev_t *mddev, char *page)
2948 int msec = (mddev->safemode_delay*1000)/HZ;
2949 return sprintf(page, "%d.%03d\n", msec/1000, msec%1000);
2952 safe_delay_store(mddev_t *mddev, const char *cbuf, size_t len)
2956 if (strict_strtoul_scaled(cbuf, &msec, 3) < 0)
2959 mddev->safemode_delay = 0;
2961 unsigned long old_delay = mddev->safemode_delay;
2962 mddev->safemode_delay = (msec*HZ)/1000;
2963 if (mddev->safemode_delay == 0)
2964 mddev->safemode_delay = 1;
2965 if (mddev->safemode_delay < old_delay)
2966 md_safemode_timeout((unsigned long)mddev);
2970 static struct md_sysfs_entry md_safe_delay =
2971 __ATTR(safe_mode_delay, S_IRUGO|S_IWUSR,safe_delay_show, safe_delay_store);
2974 level_show(mddev_t *mddev, char *page)
2976 struct mdk_personality *p = mddev->pers;
2978 return sprintf(page, "%s\n", p->name);
2979 else if (mddev->clevel[0])
2980 return sprintf(page, "%s\n", mddev->clevel);
2981 else if (mddev->level != LEVEL_NONE)
2982 return sprintf(page, "%d\n", mddev->level);
2988 level_store(mddev_t *mddev, const char *buf, size_t len)
2992 struct mdk_personality *pers;
2997 if (mddev->pers == NULL) {
3000 if (len >= sizeof(mddev->clevel))
3002 strncpy(mddev->clevel, buf, len);
3003 if (mddev->clevel[len-1] == '\n')
3005 mddev->clevel[len] = 0;
3006 mddev->level = LEVEL_NONE;
3010 /* request to change the personality. Need to ensure:
3011 * - array is not engaged in resync/recovery/reshape
3012 * - old personality can be suspended
3013 * - new personality will access other array.
3016 if (mddev->sync_thread ||
3017 mddev->reshape_position != MaxSector ||
3018 mddev->sysfs_active)
3021 if (!mddev->pers->quiesce) {
3022 printk(KERN_WARNING "md: %s: %s does not support online personality change\n",
3023 mdname(mddev), mddev->pers->name);
3027 /* Now find the new personality */
3028 if (len == 0 || len >= sizeof(clevel))
3030 strncpy(clevel, buf, len);
3031 if (clevel[len-1] == '\n')
3034 if (strict_strtol(clevel, 10, &level))
3037 if (request_module("md-%s", clevel) != 0)
3038 request_module("md-level-%s", clevel);
3039 spin_lock(&pers_lock);
3040 pers = find_pers(level, clevel);
3041 if (!pers || !try_module_get(pers->owner)) {
3042 spin_unlock(&pers_lock);
3043 printk(KERN_WARNING "md: personality %s not loaded\n", clevel);
3046 spin_unlock(&pers_lock);
3048 if (pers == mddev->pers) {
3049 /* Nothing to do! */
3050 module_put(pers->owner);
3053 if (!pers->takeover) {
3054 module_put(pers->owner);
3055 printk(KERN_WARNING "md: %s: %s does not support personality takeover\n",
3056 mdname(mddev), clevel);
3060 list_for_each_entry(rdev, &mddev->disks, same_set)
3061 rdev->new_raid_disk = rdev->raid_disk;
3063 /* ->takeover must set new_* and/or delta_disks
3064 * if it succeeds, and may set them when it fails.
3066 priv = pers->takeover(mddev);
3068 mddev->new_level = mddev->level;
3069 mddev->new_layout = mddev->layout;
3070 mddev->new_chunk_sectors = mddev->chunk_sectors;
3071 mddev->raid_disks -= mddev->delta_disks;
3072 mddev->delta_disks = 0;
3073 module_put(pers->owner);
3074 printk(KERN_WARNING "md: %s: %s would not accept array\n",
3075 mdname(mddev), clevel);
3076 return PTR_ERR(priv);
3079 /* Looks like we have a winner */
3080 mddev_suspend(mddev);
3081 mddev->pers->stop(mddev);
3083 if (mddev->pers->sync_request == NULL &&
3084 pers->sync_request != NULL) {
3085 /* need to add the md_redundancy_group */
3086 if (sysfs_create_group(&mddev->kobj, &md_redundancy_group))
3088 "md: cannot register extra attributes for %s\n",
3090 mddev->sysfs_action = sysfs_get_dirent(mddev->kobj.sd, NULL, "sync_action");
3092 if (mddev->pers->sync_request != NULL &&
3093 pers->sync_request == NULL) {
3094 /* need to remove the md_redundancy_group */
3095 if (mddev->to_remove == NULL)
3096 mddev->to_remove = &md_redundancy_group;
3099 if (mddev->pers->sync_request == NULL &&
3101 /* We are converting from a no-redundancy array
3102 * to a redundancy array and metadata is managed
3103 * externally so we need to be sure that writes
3104 * won't block due to a need to transition
3106 * until external management is started.
3109 mddev->safemode_delay = 0;
3110 mddev->safemode = 0;
3113 list_for_each_entry(rdev, &mddev->disks, same_set) {
3115 if (rdev->raid_disk < 0)
3117 if (rdev->new_raid_disk > mddev->raid_disks)
3118 rdev->new_raid_disk = -1;
3119 if (rdev->new_raid_disk == rdev->raid_disk)
3121 sprintf(nm, "rd%d", rdev->raid_disk);
3122 sysfs_remove_link(&mddev->kobj, nm);
3124 list_for_each_entry(rdev, &mddev->disks, same_set) {
3125 if (rdev->raid_disk < 0)
3127 if (rdev->new_raid_disk == rdev->raid_disk)
3129 rdev->raid_disk = rdev->new_raid_disk;
3130 if (rdev->raid_disk < 0)
3131 clear_bit(In_sync, &rdev->flags);
3134 sprintf(nm, "rd%d", rdev->raid_disk);
3135 if(sysfs_create_link(&mddev->kobj, &rdev->kobj, nm))
3136 printk("md: cannot register %s for %s after level change\n",
3141 module_put(mddev->pers->owner);
3143 mddev->private = priv;
3144 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
3145 mddev->level = mddev->new_level;
3146 mddev->layout = mddev->new_layout;
3147 mddev->chunk_sectors = mddev->new_chunk_sectors;
3148 mddev->delta_disks = 0;
3149 if (mddev->pers->sync_request == NULL) {
3150 /* this is now an array without redundancy, so
3151 * it must always be in_sync
3154 del_timer_sync(&mddev->safemode_timer);
3157 mddev_resume(mddev);
3158 set_bit(MD_CHANGE_DEVS, &mddev->flags);
3159 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3160 md_wakeup_thread(mddev->thread);
3161 sysfs_notify(&mddev->kobj, NULL, "level");
3162 md_new_event(mddev);
3166 static struct md_sysfs_entry md_level =
3167 __ATTR(level, S_IRUGO|S_IWUSR, level_show, level_store);
3171 layout_show(mddev_t *mddev, char *page)
3173 /* just a number, not meaningful for all levels */
3174 if (mddev->reshape_position != MaxSector &&
3175 mddev->layout != mddev->new_layout)
3176 return sprintf(page, "%d (%d)\n",
3177 mddev->new_layout, mddev->layout);
3178 return sprintf(page, "%d\n", mddev->layout);
3182 layout_store(mddev_t *mddev, const char *buf, size_t len)
3185 unsigned long n = simple_strtoul(buf, &e, 10);
3187 if (!*buf || (*e && *e != '\n'))
3192 if (mddev->pers->check_reshape == NULL)
3194 mddev->new_layout = n;
3195 err = mddev->pers->check_reshape(mddev);
3197 mddev->new_layout = mddev->layout;
3201 mddev->new_layout = n;
3202 if (mddev->reshape_position == MaxSector)
3207 static struct md_sysfs_entry md_layout =
3208 __ATTR(layout, S_IRUGO|S_IWUSR, layout_show, layout_store);
3212 raid_disks_show(mddev_t *mddev, char *page)
3214 if (mddev->raid_disks == 0)
3216 if (mddev->reshape_position != MaxSector &&
3217 mddev->delta_disks != 0)
3218 return sprintf(page, "%d (%d)\n", mddev->raid_disks,
3219 mddev->raid_disks - mddev->delta_disks);
3220 return sprintf(page, "%d\n", mddev->raid_disks);
3223 static int update_raid_disks(mddev_t *mddev, int raid_disks);
3226 raid_disks_store(mddev_t *mddev, const char *buf, size_t len)
3230 unsigned long n = simple_strtoul(buf, &e, 10);
3232 if (!*buf || (*e && *e != '\n'))
3236 rv = update_raid_disks(mddev, n);
3237 else if (mddev->reshape_position != MaxSector) {
3238 int olddisks = mddev->raid_disks - mddev->delta_disks;
3239 mddev->delta_disks = n - olddisks;
3240 mddev->raid_disks = n;
3242 mddev->raid_disks = n;
3243 return rv ? rv : len;
3245 static struct md_sysfs_entry md_raid_disks =
3246 __ATTR(raid_disks, S_IRUGO|S_IWUSR, raid_disks_show, raid_disks_store);
3249 chunk_size_show(mddev_t *mddev, char *page)
3251 if (mddev->reshape_position != MaxSector &&
3252 mddev->chunk_sectors != mddev->new_chunk_sectors)
3253 return sprintf(page, "%d (%d)\n",
3254 mddev->new_chunk_sectors << 9,
3255 mddev->chunk_sectors << 9);
3256 return sprintf(page, "%d\n", mddev->chunk_sectors << 9);
3260 chunk_size_store(mddev_t *mddev, const char *buf, size_t len)
3263 unsigned long n = simple_strtoul(buf, &e, 10);
3265 if (!*buf || (*e && *e != '\n'))
3270 if (mddev->pers->check_reshape == NULL)
3272 mddev->new_chunk_sectors = n >> 9;
3273 err = mddev->pers->check_reshape(mddev);
3275 mddev->new_chunk_sectors = mddev->chunk_sectors;
3279 mddev->new_chunk_sectors = n >> 9;
3280 if (mddev->reshape_position == MaxSector)
3281 mddev->chunk_sectors = n >> 9;
3285 static struct md_sysfs_entry md_chunk_size =
3286 __ATTR(chunk_size, S_IRUGO|S_IWUSR, chunk_size_show, chunk_size_store);
3289 resync_start_show(mddev_t *mddev, char *page)
3291 if (mddev->recovery_cp == MaxSector)
3292 return sprintf(page, "none\n");
3293 return sprintf(page, "%llu\n", (unsigned long long)mddev->recovery_cp);
3297 resync_start_store(mddev_t *mddev, const char *buf, size_t len)
3300 unsigned long long n = simple_strtoull(buf, &e, 10);
3304 if (cmd_match(buf, "none"))
3306 else if (!*buf || (*e && *e != '\n'))
3309 mddev->recovery_cp = n;
3312 static struct md_sysfs_entry md_resync_start =
3313 __ATTR(resync_start, S_IRUGO|S_IWUSR, resync_start_show, resync_start_store);
3316 * The array state can be:
3319 * No devices, no size, no level
3320 * Equivalent to STOP_ARRAY ioctl
3322 * May have some settings, but array is not active
3323 * all IO results in error
3324 * When written, doesn't tear down array, but just stops it
3325 * suspended (not supported yet)
3326 * All IO requests will block. The array can be reconfigured.
3327 * Writing this, if accepted, will block until array is quiescent
3329 * no resync can happen. no superblocks get written.
3330 * write requests fail
3332 * like readonly, but behaves like 'clean' on a write request.
3334 * clean - no pending writes, but otherwise active.
3335 * When written to inactive array, starts without resync
3336 * If a write request arrives then
3337 * if metadata is known, mark 'dirty' and switch to 'active'.
3338 * if not known, block and switch to write-pending
3339 * If written to an active array that has pending writes, then fails.
3341 * fully active: IO and resync can be happening.
3342 * When written to inactive array, starts with resync
3345 * clean, but writes are blocked waiting for 'active' to be written.
3348 * like active, but no writes have been seen for a while (100msec).
3351 enum array_state { clear, inactive, suspended, readonly, read_auto, clean, active,
3352 write_pending, active_idle, bad_word};
3353 static char *array_states[] = {
3354 "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active",
3355 "write-pending", "active-idle", NULL };
3357 static int match_word(const char *word, char **list)
3360 for (n=0; list[n]; n++)
3361 if (cmd_match(word, list[n]))
3367 array_state_show(mddev_t *mddev, char *page)
3369 enum array_state st = inactive;
3382 else if (test_bit(MD_CHANGE_CLEAN, &mddev->flags))
3384 else if (mddev->safemode)
3390 if (list_empty(&mddev->disks) &&
3391 mddev->raid_disks == 0 &&
3392 mddev->dev_sectors == 0)
3397 return sprintf(page, "%s\n", array_states[st]);
3400 static int do_md_stop(mddev_t * mddev, int ro, int is_open);
3401 static int md_set_readonly(mddev_t * mddev, int is_open);
3402 static int do_md_run(mddev_t * mddev);
3403 static int restart_array(mddev_t *mddev);
3406 array_state_store(mddev_t *mddev, const char *buf, size_t len)
3409 enum array_state st = match_word(buf, array_states);
3414 /* stopping an active array */
3415 if (atomic_read(&mddev->openers) > 0)
3417 err = do_md_stop(mddev, 0, 0);
3420 /* stopping an active array */
3422 if (atomic_read(&mddev->openers) > 0)
3424 err = do_md_stop(mddev, 2, 0);
3426 err = 0; /* already inactive */
3429 break; /* not supported yet */
3432 err = md_set_readonly(mddev, 0);
3435 set_disk_ro(mddev->gendisk, 1);
3436 err = do_md_run(mddev);
3442 err = md_set_readonly(mddev, 0);
3443 else if (mddev->ro == 1)
3444 err = restart_array(mddev);
3447 set_disk_ro(mddev->gendisk, 0);
3451 err = do_md_run(mddev);
3456 restart_array(mddev);
3457 spin_lock_irq(&mddev->write_lock);
3458 if (atomic_read(&mddev->writes_pending) == 0) {
3459 if (mddev->in_sync == 0) {
3461 if (mddev->safemode == 1)
3462 mddev->safemode = 0;
3463 if (mddev->persistent)
3464 set_bit(MD_CHANGE_CLEAN,
3470 spin_unlock_irq(&mddev->write_lock);
3476 restart_array(mddev);
3477 if (mddev->external)
3478 clear_bit(MD_CHANGE_CLEAN, &mddev->flags);
3479 wake_up(&mddev->sb_wait);
3483 set_disk_ro(mddev->gendisk, 0);
3484 err = do_md_run(mddev);
3489 /* these cannot be set */
3495 sysfs_notify_dirent_safe(mddev->sysfs_state);
3499 static struct md_sysfs_entry md_array_state =
3500 __ATTR(array_state, S_IRUGO|S_IWUSR, array_state_show, array_state_store);
3503 max_corrected_read_errors_show(mddev_t *mddev, char *page) {
3504 return sprintf(page, "%d\n",
3505 atomic_read(&mddev->max_corr_read_errors));
3509 max_corrected_read_errors_store(mddev_t *mddev, const char *buf, size_t len)
3512 unsigned long n = simple_strtoul(buf, &e, 10);
3514 if (*buf && (*e == 0 || *e == '\n')) {
3515 atomic_set(&mddev->max_corr_read_errors, n);
3521 static struct md_sysfs_entry max_corr_read_errors =
3522 __ATTR(max_read_errors, S_IRUGO|S_IWUSR, max_corrected_read_errors_show,
3523 max_corrected_read_errors_store);
3526 null_show(mddev_t *mddev, char *page)
3532 new_dev_store(mddev_t *mddev, const char *buf, size_t len)
3534 /* buf must be %d:%d\n? giving major and minor numbers */
3535 /* The new device is added to the array.
3536 * If the array has a persistent superblock, we read the
3537 * superblock to initialise info and check validity.
3538 * Otherwise, only checking done is that in bind_rdev_to_array,
3539 * which mainly checks size.
3542 int major = simple_strtoul(buf, &e, 10);
3548 if (!*buf || *e != ':' || !e[1] || e[1] == '\n')
3550 minor = simple_strtoul(e+1, &e, 10);
3551 if (*e && *e != '\n')
3553 dev = MKDEV(major, minor);
3554 if (major != MAJOR(dev) ||
3555 minor != MINOR(dev))
3559 if (mddev->persistent) {
3560 rdev = md_import_device(dev, mddev->major_version,
3561 mddev->minor_version);
3562 if (!IS_ERR(rdev) && !list_empty(&mddev->disks)) {
3563 mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
3564 mdk_rdev_t, same_set);
3565 err = super_types[mddev->major_version]
3566 .load_super(rdev, rdev0, mddev->minor_version);
3570 } else if (mddev->external)
3571 rdev = md_import_device(dev, -2, -1);
3573 rdev = md_import_device(dev, -1, -1);
3576 return PTR_ERR(rdev);
3577 err = bind_rdev_to_array(rdev, mddev);
3581 return err ? err : len;
3584 static struct md_sysfs_entry md_new_device =
3585 __ATTR(new_dev, S_IWUSR, null_show, new_dev_store);
3588 bitmap_store(mddev_t *mddev, const char *buf, size_t len)
3591 unsigned long chunk, end_chunk;
3595 /* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */
3597 chunk = end_chunk = simple_strtoul(buf, &end, 0);
3598 if (buf == end) break;
3599 if (*end == '-') { /* range */
3601 end_chunk = simple_strtoul(buf, &end, 0);
3602 if (buf == end) break;
3604 if (*end && !isspace(*end)) break;
3605 bitmap_dirty_bits(mddev->bitmap, chunk, end_chunk);
3606 buf = skip_spaces(end);
3608 bitmap_unplug(mddev->bitmap); /* flush the bits to disk */
3613 static struct md_sysfs_entry md_bitmap =
3614 __ATTR(bitmap_set_bits, S_IWUSR, null_show, bitmap_store);
3617 size_show(mddev_t *mddev, char *page)
3619 return sprintf(page, "%llu\n",
3620 (unsigned long long)mddev->dev_sectors / 2);
3623 static int update_size(mddev_t *mddev, sector_t num_sectors);
3626 size_store(mddev_t *mddev, const char *buf, size_t len)
3628 /* If array is inactive, we can reduce the component size, but
3629 * not increase it (except from 0).
3630 * If array is active, we can try an on-line resize
3633 int err = strict_blocks_to_sectors(buf, §ors);
3638 err = update_size(mddev, sectors);
3639 md_update_sb(mddev, 1);
3641 if (mddev->dev_sectors == 0 ||
3642 mddev->dev_sectors > sectors)
3643 mddev->dev_sectors = sectors;
3647 return err ? err : len;
3650 static struct md_sysfs_entry md_size =
3651 __ATTR(component_size, S_IRUGO|S_IWUSR, size_show, size_store);
3656 * 'none' for arrays with no metadata (good luck...)
3657 * 'external' for arrays with externally managed metadata,
3658 * or N.M for internally known formats
3661 metadata_show(mddev_t *mddev, char *page)
3663 if (mddev->persistent)
3664 return sprintf(page, "%d.%d\n",
3665 mddev->major_version, mddev->minor_version);
3666 else if (mddev->external)
3667 return sprintf(page, "external:%s\n", mddev->metadata_type);
3669 return sprintf(page, "none\n");
3673 metadata_store(mddev_t *mddev, const char *buf, size_t len)
3677 /* Changing the details of 'external' metadata is
3678 * always permitted. Otherwise there must be
3679 * no devices attached to the array.
3681 if (mddev->external && strncmp(buf, "external:", 9) == 0)
3683 else if (!list_empty(&mddev->disks))
3686 if (cmd_match(buf, "none")) {
3687 mddev->persistent = 0;
3688 mddev->external = 0;
3689 mddev->major_version = 0;
3690 mddev->minor_version = 90;
3693 if (strncmp(buf, "external:", 9) == 0) {
3694 size_t namelen = len-9;
3695 if (namelen >= sizeof(mddev->metadata_type))
3696 namelen = sizeof(mddev->metadata_type)-1;
3697 strncpy(mddev->metadata_type, buf+9, namelen);
3698 mddev->metadata_type[namelen] = 0;
3699 if (namelen && mddev->metadata_type[namelen-1] == '\n')
3700 mddev->metadata_type[--namelen] = 0;
3701 mddev->persistent = 0;
3702 mddev->external = 1;
3703 mddev->major_version = 0;
3704 mddev->minor_version = 90;
3707 major = simple_strtoul(buf, &e, 10);
3708 if (e==buf || *e != '.')
3711 minor = simple_strtoul(buf, &e, 10);
3712 if (e==buf || (*e && *e != '\n') )
3714 if (major >= ARRAY_SIZE(super_types) || super_types[major].name == NULL)
3716 mddev->major_version = major;
3717 mddev->minor_version = minor;
3718 mddev->persistent = 1;
3719 mddev->external = 0;
3723 static struct md_sysfs_entry md_metadata =
3724 __ATTR(metadata_version, S_IRUGO|S_IWUSR, metadata_show, metadata_store);
3727 action_show(mddev_t *mddev, char *page)
3729 char *type = "idle";
3730 if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
3732 else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
3733 (!mddev->ro && test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))) {
3734 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
3736 else if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
3737 if (!test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
3739 else if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
3743 } else if (test_bit(MD_RECOVERY_RECOVER, &mddev->recovery))
3746 return sprintf(page, "%s\n", type);
3750 action_store(mddev_t *mddev, const char *page, size_t len)
3752 if (!mddev->pers || !mddev->pers->sync_request)
3755 if (cmd_match(page, "frozen"))
3756 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
3758 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
3760 if (cmd_match(page, "idle") || cmd_match(page, "frozen")) {
3761 if (mddev->sync_thread) {
3762 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
3763 md_unregister_thread(mddev->sync_thread);
3764 mddev->sync_thread = NULL;
3765 mddev->recovery = 0;
3767 } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
3768 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
3770 else if (cmd_match(page, "resync"))
3771 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3772 else if (cmd_match(page, "recover")) {
3773 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
3774 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3775 } else if (cmd_match(page, "reshape")) {
3777 if (mddev->pers->start_reshape == NULL)
3779 err = mddev->pers->start_reshape(mddev);
3782 sysfs_notify(&mddev->kobj, NULL, "degraded");
3784 if (cmd_match(page, "check"))
3785 set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
3786 else if (!cmd_match(page, "repair"))
3788 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
3789 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
3791 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3792 md_wakeup_thread(mddev->thread);
3793 sysfs_notify_dirent_safe(mddev->sysfs_action);
3798 mismatch_cnt_show(mddev_t *mddev, char *page)
3800 return sprintf(page, "%llu\n",
3801 (unsigned long long) mddev->resync_mismatches);
3804 static struct md_sysfs_entry md_scan_mode =
3805 __ATTR(sync_action, S_IRUGO|S_IWUSR, action_show, action_store);
3808 static struct md_sysfs_entry md_mismatches = __ATTR_RO(mismatch_cnt);
3811 sync_min_show(mddev_t *mddev, char *page)
3813 return sprintf(page, "%d (%s)\n", speed_min(mddev),
3814 mddev->sync_speed_min ? "local": "system");
3818 sync_min_store(mddev_t *mddev, const char *buf, size_t len)
3822 if (strncmp(buf, "system", 6)==0) {
3823 mddev->sync_speed_min = 0;
3826 min = simple_strtoul(buf, &e, 10);
3827 if (buf == e || (*e && *e != '\n') || min <= 0)
3829 mddev->sync_speed_min = min;
3833 static struct md_sysfs_entry md_sync_min =
3834 __ATTR(sync_speed_min, S_IRUGO|S_IWUSR, sync_min_show, sync_min_store);
3837 sync_max_show(mddev_t *mddev, char *page)
3839 return sprintf(page, "%d (%s)\n", speed_max(mddev),
3840 mddev->sync_speed_max ? "local": "system");
3844 sync_max_store(mddev_t *mddev, const char *buf, size_t len)
3848 if (strncmp(buf, "system", 6)==0) {
3849 mddev->sync_speed_max = 0;
3852 max = simple_strtoul(buf, &e, 10);
3853 if (buf == e || (*e && *e != '\n') || max <= 0)
3855 mddev->sync_speed_max = max;
3859 static struct md_sysfs_entry md_sync_max =
3860 __ATTR(sync_speed_max, S_IRUGO|S_IWUSR, sync_max_show, sync_max_store);
3863 degraded_show(mddev_t *mddev, char *page)
3865 return sprintf(page, "%d\n", mddev->degraded);
3867 static struct md_sysfs_entry md_degraded = __ATTR_RO(degraded);
3870 sync_force_parallel_show(mddev_t *mddev, char *page)
3872 return sprintf(page, "%d\n", mddev->parallel_resync);
3876 sync_force_parallel_store(mddev_t *mddev, const char *buf, size_t len)
3880 if (strict_strtol(buf, 10, &n))
3883 if (n != 0 && n != 1)
3886 mddev->parallel_resync = n;
3888 if (mddev->sync_thread)
3889 wake_up(&resync_wait);
3894 /* force parallel resync, even with shared block devices */
3895 static struct md_sysfs_entry md_sync_force_parallel =
3896 __ATTR(sync_force_parallel, S_IRUGO|S_IWUSR,
3897 sync_force_parallel_show, sync_force_parallel_store);
3900 sync_speed_show(mddev_t *mddev, char *page)
3902 unsigned long resync, dt, db;
3903 if (mddev->curr_resync == 0)
3904 return sprintf(page, "none\n");
3905 resync = mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active);
3906 dt = (jiffies - mddev->resync_mark) / HZ;
3908 db = resync - mddev->resync_mark_cnt;
3909 return sprintf(page, "%lu\n", db/dt/2); /* K/sec */
3912 static struct md_sysfs_entry md_sync_speed = __ATTR_RO(sync_speed);
3915 sync_completed_show(mddev_t *mddev, char *page)
3917 unsigned long max_sectors, resync;
3919 if (!test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
3920 return sprintf(page, "none\n");
3922 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
3923 max_sectors = mddev->resync_max_sectors;
3925 max_sectors = mddev->dev_sectors;
3927 resync = mddev->curr_resync_completed;
3928 return sprintf(page, "%lu / %lu\n", resync, max_sectors);
3931 static struct md_sysfs_entry md_sync_completed = __ATTR_RO(sync_completed);
3934 min_sync_show(mddev_t *mddev, char *page)
3936 return sprintf(page, "%llu\n",
3937 (unsigned long long)mddev->resync_min);
3940 min_sync_store(mddev_t *mddev, const char *buf, size_t len)
3942 unsigned long long min;
3943 if (strict_strtoull(buf, 10, &min))
3945 if (min > mddev->resync_max)
3947 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
3950 /* Must be a multiple of chunk_size */
3951 if (mddev->chunk_sectors) {
3952 sector_t temp = min;
3953 if (sector_div(temp, mddev->chunk_sectors))
3956 mddev->resync_min = min;
3961 static struct md_sysfs_entry md_min_sync =
3962 __ATTR(sync_min, S_IRUGO|S_IWUSR, min_sync_show, min_sync_store);
3965 max_sync_show(mddev_t *mddev, char *page)
3967 if (mddev->resync_max == MaxSector)
3968 return sprintf(page, "max\n");
3970 return sprintf(page, "%llu\n",
3971 (unsigned long long)mddev->resync_max);
3974 max_sync_store(mddev_t *mddev, const char *buf, size_t len)
3976 if (strncmp(buf, "max", 3) == 0)
3977 mddev->resync_max = MaxSector;
3979 unsigned long long max;
3980 if (strict_strtoull(buf, 10, &max))
3982 if (max < mddev->resync_min)
3984 if (max < mddev->resync_max &&
3986 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
3989 /* Must be a multiple of chunk_size */
3990 if (mddev->chunk_sectors) {
3991 sector_t temp = max;
3992 if (sector_div(temp, mddev->chunk_sectors))
3995 mddev->resync_max = max;
3997 wake_up(&mddev->recovery_wait);
4001 static struct md_sysfs_entry md_max_sync =
4002 __ATTR(sync_max, S_IRUGO|S_IWUSR, max_sync_show, max_sync_store);
4005 suspend_lo_show(mddev_t *mddev, char *page)
4007 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_lo);
4011 suspend_lo_store(mddev_t *mddev, const char *buf, size_t len)
4014 unsigned long long new = simple_strtoull(buf, &e, 10);
4016 if (mddev->pers == NULL ||
4017 mddev->pers->quiesce == NULL)
4019 if (buf == e || (*e && *e != '\n'))
4021 if (new >= mddev->suspend_hi ||
4022 (new > mddev->suspend_lo && new < mddev->suspend_hi)) {
4023 mddev->suspend_lo = new;
4024 mddev->pers->quiesce(mddev, 2);
4029 static struct md_sysfs_entry md_suspend_lo =
4030 __ATTR(suspend_lo, S_IRUGO|S_IWUSR, suspend_lo_show, suspend_lo_store);
4034 suspend_hi_show(mddev_t *mddev, char *page)
4036 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_hi);
4040 suspend_hi_store(mddev_t *mddev, const char *buf, size_t len)
4043 unsigned long long new = simple_strtoull(buf, &e, 10);
4045 if (mddev->pers == NULL ||
4046 mddev->pers->quiesce == NULL)
4048 if (buf == e || (*e && *e != '\n'))
4050 if ((new <= mddev->suspend_lo && mddev->suspend_lo >= mddev->suspend_hi) ||
4051 (new > mddev->suspend_lo && new > mddev->suspend_hi)) {
4052 mddev->suspend_hi = new;
4053 mddev->pers->quiesce(mddev, 1);
4054 mddev->pers->quiesce(mddev, 0);
4059 static struct md_sysfs_entry md_suspend_hi =
4060 __ATTR(suspend_hi, S_IRUGO|S_IWUSR, suspend_hi_show, suspend_hi_store);
4063 reshape_position_show(mddev_t *mddev, char *page)
4065 if (mddev->reshape_position != MaxSector)
4066 return sprintf(page, "%llu\n",
4067 (unsigned long long)mddev->reshape_position);
4068 strcpy(page, "none\n");
4073 reshape_position_store(mddev_t *mddev, const char *buf, size_t len)
4076 unsigned long long new = simple_strtoull(buf, &e, 10);
4079 if (buf == e || (*e && *e != '\n'))
4081 mddev->reshape_position = new;
4082 mddev->delta_disks = 0;
4083 mddev->new_level = mddev->level;
4084 mddev->new_layout = mddev->layout;
4085 mddev->new_chunk_sectors = mddev->chunk_sectors;
4089 static struct md_sysfs_entry md_reshape_position =
4090 __ATTR(reshape_position, S_IRUGO|S_IWUSR, reshape_position_show,
4091 reshape_position_store);
4094 array_size_show(mddev_t *mddev, char *page)
4096 if (mddev->external_size)
4097 return sprintf(page, "%llu\n",
4098 (unsigned long long)mddev->array_sectors/2);
4100 return sprintf(page, "default\n");
4104 array_size_store(mddev_t *mddev, const char *buf, size_t len)
4108 if (strncmp(buf, "default", 7) == 0) {
4110 sectors = mddev->pers->size(mddev, 0, 0);
4112 sectors = mddev->array_sectors;
4114 mddev->external_size = 0;
4116 if (strict_blocks_to_sectors(buf, §ors) < 0)
4118 if (mddev->pers && mddev->pers->size(mddev, 0, 0) < sectors)
4121 mddev->external_size = 1;
4124 mddev->array_sectors = sectors;
4125 set_capacity(mddev->gendisk, mddev->array_sectors);
4127 revalidate_disk(mddev->gendisk);
4132 static struct md_sysfs_entry md_array_size =
4133 __ATTR(array_size, S_IRUGO|S_IWUSR, array_size_show,
4136 static struct attribute *md_default_attrs[] = {
4139 &md_raid_disks.attr,
4140 &md_chunk_size.attr,
4142 &md_resync_start.attr,
4144 &md_new_device.attr,
4145 &md_safe_delay.attr,
4146 &md_array_state.attr,
4147 &md_reshape_position.attr,
4148 &md_array_size.attr,
4149 &max_corr_read_errors.attr,
4153 static struct attribute *md_redundancy_attrs[] = {
4155 &md_mismatches.attr,
4158 &md_sync_speed.attr,
4159 &md_sync_force_parallel.attr,
4160 &md_sync_completed.attr,
4163 &md_suspend_lo.attr,
4164 &md_suspend_hi.attr,
4169 static struct attribute_group md_redundancy_group = {
4171 .attrs = md_redundancy_attrs,
4176 md_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
4178 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
4179 mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
4184 rv = mddev_lock(mddev);
4186 rv = entry->show(mddev, page);
4187 mddev_unlock(mddev);
4193 md_attr_store(struct kobject *kobj, struct attribute *attr,
4194 const char *page, size_t length)
4196 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
4197 mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
4202 if (!capable(CAP_SYS_ADMIN))
4204 rv = mddev_lock(mddev);
4205 if (mddev->hold_active == UNTIL_IOCTL)
4206 mddev->hold_active = 0;
4208 rv = entry->store(mddev, page, length);
4209 mddev_unlock(mddev);
4214 static void md_free(struct kobject *ko)
4216 mddev_t *mddev = container_of(ko, mddev_t, kobj);
4218 if (mddev->sysfs_state)
4219 sysfs_put(mddev->sysfs_state);
4221 if (mddev->gendisk) {
4222 del_gendisk(mddev->gendisk);
4223 put_disk(mddev->gendisk);
4226 blk_cleanup_queue(mddev->queue);
4231 static const struct sysfs_ops md_sysfs_ops = {
4232 .show = md_attr_show,
4233 .store = md_attr_store,
4235 static struct kobj_type md_ktype = {
4237 .sysfs_ops = &md_sysfs_ops,
4238 .default_attrs = md_default_attrs,
4243 static void mddev_delayed_delete(struct work_struct *ws)
4245 mddev_t *mddev = container_of(ws, mddev_t, del_work);
4247 sysfs_remove_group(&mddev->kobj, &md_bitmap_group);
4248 kobject_del(&mddev->kobj);
4249 kobject_put(&mddev->kobj);
4252 static int md_alloc(dev_t dev, char *name)
4254 static DEFINE_MUTEX(disks_mutex);
4255 mddev_t *mddev = mddev_find(dev);
4256 struct gendisk *disk;
4265 partitioned = (MAJOR(mddev->unit) != MD_MAJOR);
4266 shift = partitioned ? MdpMinorShift : 0;
4267 unit = MINOR(mddev->unit) >> shift;
4269 /* wait for any previous instance if this device
4270 * to be completed removed (mddev_delayed_delete).
4272 flush_scheduled_work();
4274 mutex_lock(&disks_mutex);
4280 /* Need to ensure that 'name' is not a duplicate.
4283 spin_lock(&all_mddevs_lock);
4285 list_for_each_entry(mddev2, &all_mddevs, all_mddevs)
4286 if (mddev2->gendisk &&
4287 strcmp(mddev2->gendisk->disk_name, name) == 0) {
4288 spin_unlock(&all_mddevs_lock);
4291 spin_unlock(&all_mddevs_lock);
4295 mddev->queue = blk_alloc_queue(GFP_KERNEL);
4298 mddev->queue->queuedata = mddev;
4300 /* Can be unlocked because the queue is new: no concurrency */
4301 queue_flag_set_unlocked(QUEUE_FLAG_CLUSTER, mddev->queue);
4303 blk_queue_make_request(mddev->queue, md_make_request);
4305 disk = alloc_disk(1 << shift);
4307 blk_cleanup_queue(mddev->queue);
4308 mddev->queue = NULL;
4311 disk->major = MAJOR(mddev->unit);
4312 disk->first_minor = unit << shift;
4314 strcpy(disk->disk_name, name);
4315 else if (partitioned)
4316 sprintf(disk->disk_name, "md_d%d", unit);
4318 sprintf(disk->disk_name, "md%d", unit);
4319 disk->fops = &md_fops;
4320 disk->private_data = mddev;
4321 disk->queue = mddev->queue;
4322 /* Allow extended partitions. This makes the
4323 * 'mdp' device redundant, but we can't really
4326 disk->flags |= GENHD_FL_EXT_DEVT;
4328 mddev->gendisk = disk;
4329 error = kobject_init_and_add(&mddev->kobj, &md_ktype,
4330 &disk_to_dev(disk)->kobj, "%s", "md");
4332 /* This isn't possible, but as kobject_init_and_add is marked
4333 * __must_check, we must do something with the result
4335 printk(KERN_WARNING "md: cannot register %s/md - name in use\n",
4339 if (mddev->kobj.sd &&
4340 sysfs_create_group(&mddev->kobj, &md_bitmap_group))
4341 printk(KERN_DEBUG "pointless warning\n");
4343 mutex_unlock(&disks_mutex);
4344 if (!error && mddev->kobj.sd) {
4345 kobject_uevent(&mddev->kobj, KOBJ_ADD);
4346 mddev->sysfs_state = sysfs_get_dirent_safe(mddev->kobj.sd, "array_state");
4352 static struct kobject *md_probe(dev_t dev, int *part, void *data)
4354 md_alloc(dev, NULL);
4358 static int add_named_array(const char *val, struct kernel_param *kp)
4360 /* val must be "md_*" where * is not all digits.
4361 * We allocate an array with a large free minor number, and
4362 * set the name to val. val must not already be an active name.
4364 int len = strlen(val);
4365 char buf[DISK_NAME_LEN];
4367 while (len && val[len-1] == '\n')
4369 if (len >= DISK_NAME_LEN)
4371 strlcpy(buf, val, len+1);
4372 if (strncmp(buf, "md_", 3) != 0)
4374 return md_alloc(0, buf);
4377 static void md_safemode_timeout(unsigned long data)
4379 mddev_t *mddev = (mddev_t *) data;
4381 if (!atomic_read(&mddev->writes_pending)) {
4382 mddev->safemode = 1;
4383 if (mddev->external)
4384 sysfs_notify_dirent_safe(mddev->sysfs_state);
4386 md_wakeup_thread(mddev->thread);
4389 static int start_dirty_degraded;
4391 int md_run(mddev_t *mddev)
4395 struct mdk_personality *pers;
4397 if (list_empty(&mddev->disks))
4398 /* cannot run an array with no devices.. */
4403 /* Cannot run until previous stop completes properly */
4404 if (mddev->sysfs_active)
4408 * Analyze all RAID superblock(s)
4410 if (!mddev->raid_disks) {
4411 if (!mddev->persistent)
4416 if (mddev->level != LEVEL_NONE)
4417 request_module("md-level-%d", mddev->level);
4418 else if (mddev->clevel[0])
4419 request_module("md-%s", mddev->clevel);
4422 * Drop all container device buffers, from now on
4423 * the only valid external interface is through the md
4426 list_for_each_entry(rdev, &mddev->disks, same_set) {
4427 if (test_bit(Faulty, &rdev->flags))
4429 sync_blockdev(rdev->bdev);
4430 invalidate_bdev(rdev->bdev);
4432 /* perform some consistency tests on the device.
4433 * We don't want the data to overlap the metadata,
4434 * Internal Bitmap issues have been handled elsewhere.
4436 if (rdev->data_offset < rdev->sb_start) {
4437 if (mddev->dev_sectors &&
4438 rdev->data_offset + mddev->dev_sectors
4440 printk("md: %s: data overlaps metadata\n",
4445 if (rdev->sb_start + rdev->sb_size/512
4446 > rdev->data_offset) {
4447 printk("md: %s: metadata overlaps data\n",
4452 sysfs_notify_dirent_safe(rdev->sysfs_state);
4455 spin_lock(&pers_lock);
4456 pers = find_pers(mddev->level, mddev->clevel);
4457 if (!pers || !try_module_get(pers->owner)) {
4458 spin_unlock(&pers_lock);
4459 if (mddev->level != LEVEL_NONE)
4460 printk(KERN_WARNING "md: personality for level %d is not loaded!\n",
4463 printk(KERN_WARNING "md: personality for level %s is not loaded!\n",
4468 spin_unlock(&pers_lock);
4469 if (mddev->level != pers->level) {
4470 mddev->level = pers->level;
4471 mddev->new_level = pers->level;
4473 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
4475 if (mddev->reshape_position != MaxSector &&
4476 pers->start_reshape == NULL) {
4477 /* This personality cannot handle reshaping... */
4479 module_put(pers->owner);
4483 if (pers->sync_request) {
4484 /* Warn if this is a potentially silly
4487 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
4491 list_for_each_entry(rdev, &mddev->disks, same_set)
4492 list_for_each_entry(rdev2, &mddev->disks, same_set) {
4494 rdev->bdev->bd_contains ==
4495 rdev2->bdev->bd_contains) {
4497 "%s: WARNING: %s appears to be"
4498 " on the same physical disk as"
4501 bdevname(rdev->bdev,b),
4502 bdevname(rdev2->bdev,b2));
4509 "True protection against single-disk"
4510 " failure might be compromised.\n");
4513 mddev->recovery = 0;
4514 /* may be over-ridden by personality */
4515 mddev->resync_max_sectors = mddev->dev_sectors;
4517 mddev->barriers_work = 1;
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->barriers_work = 0;
4697 mddev->safemode = 0;
4698 mddev->bitmap_info.offset = 0;
4699 mddev->bitmap_info.default_offset = 0;
4700 mddev->bitmap_info.chunksize = 0;
4701 mddev->bitmap_info.daemon_sleep = 0;
4702 mddev->bitmap_info.max_write_behind = 0;
4706 void md_stop_writes(mddev_t *mddev)
4708 if (mddev->sync_thread) {
4709 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4710 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
4711 md_unregister_thread(mddev->sync_thread);
4712 mddev->sync_thread = NULL;
4715 del_timer_sync(&mddev->safemode_timer);
4717 bitmap_flush(mddev);
4718 md_super_wait(mddev);
4720 if (!mddev->in_sync || mddev->flags) {
4721 /* mark array as shutdown cleanly */
4723 md_update_sb(mddev, 1);
4726 EXPORT_SYMBOL_GPL(md_stop_writes);
4728 void md_stop(mddev_t *mddev)
4730 mddev->pers->stop(mddev);
4731 if (mddev->pers->sync_request && mddev->to_remove == NULL)
4732 mddev->to_remove = &md_redundancy_group;
4733 module_put(mddev->pers->owner);
4735 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4737 EXPORT_SYMBOL_GPL(md_stop);
4739 static int md_set_readonly(mddev_t *mddev, int is_open)
4742 mutex_lock(&mddev->open_mutex);
4743 if (atomic_read(&mddev->openers) > is_open) {
4744 printk("md: %s still in use.\n",mdname(mddev));
4749 md_stop_writes(mddev);
4755 set_disk_ro(mddev->gendisk, 1);
4756 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4757 sysfs_notify_dirent_safe(mddev->sysfs_state);
4761 mutex_unlock(&mddev->open_mutex);
4766 * 0 - completely stop and dis-assemble array
4767 * 2 - stop but do not disassemble array
4769 static int do_md_stop(mddev_t * mddev, int mode, int is_open)
4771 struct gendisk *disk = mddev->gendisk;
4774 mutex_lock(&mddev->open_mutex);
4775 if (atomic_read(&mddev->openers) > is_open ||
4776 mddev->sysfs_active) {
4777 printk("md: %s still in use.\n",mdname(mddev));
4778 mutex_unlock(&mddev->open_mutex);
4784 set_disk_ro(disk, 0);
4786 md_stop_writes(mddev);
4788 mddev->queue->merge_bvec_fn = NULL;
4789 mddev->queue->unplug_fn = NULL;
4790 mddev->queue->backing_dev_info.congested_fn = NULL;
4792 /* tell userspace to handle 'inactive' */
4793 sysfs_notify_dirent_safe(mddev->sysfs_state);
4795 list_for_each_entry(rdev, &mddev->disks, same_set)
4796 if (rdev->raid_disk >= 0) {
4798 sprintf(nm, "rd%d", rdev->raid_disk);
4799 sysfs_remove_link(&mddev->kobj, nm);
4802 set_capacity(disk, 0);
4803 mutex_unlock(&mddev->open_mutex);
4804 revalidate_disk(disk);
4809 mutex_unlock(&mddev->open_mutex);
4811 * Free resources if final stop
4814 printk(KERN_INFO "md: %s stopped.\n", mdname(mddev));
4816 bitmap_destroy(mddev);
4817 if (mddev->bitmap_info.file) {
4818 restore_bitmap_write_access(mddev->bitmap_info.file);
4819 fput(mddev->bitmap_info.file);
4820 mddev->bitmap_info.file = NULL;
4822 mddev->bitmap_info.offset = 0;
4824 export_array(mddev);
4827 kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
4828 if (mddev->hold_active == UNTIL_STOP)
4829 mddev->hold_active = 0;
4831 blk_integrity_unregister(disk);
4832 md_new_event(mddev);
4833 sysfs_notify_dirent_safe(mddev->sysfs_state);
4838 static void autorun_array(mddev_t *mddev)
4843 if (list_empty(&mddev->disks))
4846 printk(KERN_INFO "md: running: ");
4848 list_for_each_entry(rdev, &mddev->disks, same_set) {
4849 char b[BDEVNAME_SIZE];
4850 printk("<%s>", bdevname(rdev->bdev,b));
4854 err = do_md_run(mddev);
4856 printk(KERN_WARNING "md: do_md_run() returned %d\n", err);
4857 do_md_stop(mddev, 0, 0);
4862 * lets try to run arrays based on all disks that have arrived
4863 * until now. (those are in pending_raid_disks)
4865 * the method: pick the first pending disk, collect all disks with
4866 * the same UUID, remove all from the pending list and put them into
4867 * the 'same_array' list. Then order this list based on superblock
4868 * update time (freshest comes first), kick out 'old' disks and
4869 * compare superblocks. If everything's fine then run it.
4871 * If "unit" is allocated, then bump its reference count
4873 static void autorun_devices(int part)
4875 mdk_rdev_t *rdev0, *rdev, *tmp;
4877 char b[BDEVNAME_SIZE];
4879 printk(KERN_INFO "md: autorun ...\n");
4880 while (!list_empty(&pending_raid_disks)) {
4883 LIST_HEAD(candidates);
4884 rdev0 = list_entry(pending_raid_disks.next,
4885 mdk_rdev_t, same_set);
4887 printk(KERN_INFO "md: considering %s ...\n",
4888 bdevname(rdev0->bdev,b));
4889 INIT_LIST_HEAD(&candidates);
4890 rdev_for_each_list(rdev, tmp, &pending_raid_disks)
4891 if (super_90_load(rdev, rdev0, 0) >= 0) {
4892 printk(KERN_INFO "md: adding %s ...\n",
4893 bdevname(rdev->bdev,b));
4894 list_move(&rdev->same_set, &candidates);
4897 * now we have a set of devices, with all of them having
4898 * mostly sane superblocks. It's time to allocate the
4902 dev = MKDEV(mdp_major,
4903 rdev0->preferred_minor << MdpMinorShift);
4904 unit = MINOR(dev) >> MdpMinorShift;
4906 dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
4909 if (rdev0->preferred_minor != unit) {
4910 printk(KERN_INFO "md: unit number in %s is bad: %d\n",
4911 bdevname(rdev0->bdev, b), rdev0->preferred_minor);
4915 md_probe(dev, NULL, NULL);
4916 mddev = mddev_find(dev);
4917 if (!mddev || !mddev->gendisk) {
4921 "md: cannot allocate memory for md drive.\n");
4924 if (mddev_lock(mddev))
4925 printk(KERN_WARNING "md: %s locked, cannot run\n",
4927 else if (mddev->raid_disks || mddev->major_version
4928 || !list_empty(&mddev->disks)) {
4930 "md: %s already running, cannot run %s\n",
4931 mdname(mddev), bdevname(rdev0->bdev,b));
4932 mddev_unlock(mddev);
4934 printk(KERN_INFO "md: created %s\n", mdname(mddev));
4935 mddev->persistent = 1;
4936 rdev_for_each_list(rdev, tmp, &candidates) {
4937 list_del_init(&rdev->same_set);
4938 if (bind_rdev_to_array(rdev, mddev))
4941 autorun_array(mddev);
4942 mddev_unlock(mddev);
4944 /* on success, candidates will be empty, on error
4947 rdev_for_each_list(rdev, tmp, &candidates) {
4948 list_del_init(&rdev->same_set);
4953 printk(KERN_INFO "md: ... autorun DONE.\n");
4955 #endif /* !MODULE */
4957 static int get_version(void __user * arg)
4961 ver.major = MD_MAJOR_VERSION;
4962 ver.minor = MD_MINOR_VERSION;
4963 ver.patchlevel = MD_PATCHLEVEL_VERSION;
4965 if (copy_to_user(arg, &ver, sizeof(ver)))
4971 static int get_array_info(mddev_t * mddev, void __user * arg)
4973 mdu_array_info_t info;
4974 int nr,working,insync,failed,spare;
4977 nr=working=insync=failed=spare=0;
4978 list_for_each_entry(rdev, &mddev->disks, same_set) {
4980 if (test_bit(Faulty, &rdev->flags))
4984 if (test_bit(In_sync, &rdev->flags))
4991 info.major_version = mddev->major_version;
4992 info.minor_version = mddev->minor_version;
4993 info.patch_version = MD_PATCHLEVEL_VERSION;
4994 info.ctime = mddev->ctime;
4995 info.level = mddev->level;
4996 info.size = mddev->dev_sectors / 2;
4997 if (info.size != mddev->dev_sectors / 2) /* overflow */
5000 info.raid_disks = mddev->raid_disks;
5001 info.md_minor = mddev->md_minor;
5002 info.not_persistent= !mddev->persistent;
5004 info.utime = mddev->utime;
5007 info.state = (1<<MD_SB_CLEAN);
5008 if (mddev->bitmap && mddev->bitmap_info.offset)
5009 info.state = (1<<MD_SB_BITMAP_PRESENT);
5010 info.active_disks = insync;
5011 info.working_disks = working;
5012 info.failed_disks = failed;
5013 info.spare_disks = spare;
5015 info.layout = mddev->layout;
5016 info.chunk_size = mddev->chunk_sectors << 9;
5018 if (copy_to_user(arg, &info, sizeof(info)))
5024 static int get_bitmap_file(mddev_t * mddev, void __user * arg)
5026 mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */
5027 char *ptr, *buf = NULL;
5030 if (md_allow_write(mddev))
5031 file = kmalloc(sizeof(*file), GFP_NOIO);
5033 file = kmalloc(sizeof(*file), GFP_KERNEL);
5038 /* bitmap disabled, zero the first byte and copy out */
5039 if (!mddev->bitmap || !mddev->bitmap->file) {
5040 file->pathname[0] = '\0';
5044 buf = kmalloc(sizeof(file->pathname), GFP_KERNEL);
5048 ptr = d_path(&mddev->bitmap->file->f_path, buf, sizeof(file->pathname));
5052 strcpy(file->pathname, ptr);
5056 if (copy_to_user(arg, file, sizeof(*file)))
5064 static int get_disk_info(mddev_t * mddev, void __user * arg)
5066 mdu_disk_info_t info;
5069 if (copy_from_user(&info, arg, sizeof(info)))
5072 rdev = find_rdev_nr(mddev, info.number);
5074 info.major = MAJOR(rdev->bdev->bd_dev);
5075 info.minor = MINOR(rdev->bdev->bd_dev);
5076 info.raid_disk = rdev->raid_disk;
5078 if (test_bit(Faulty, &rdev->flags))
5079 info.state |= (1<<MD_DISK_FAULTY);
5080 else if (test_bit(In_sync, &rdev->flags)) {
5081 info.state |= (1<<MD_DISK_ACTIVE);
5082 info.state |= (1<<MD_DISK_SYNC);
5084 if (test_bit(WriteMostly, &rdev->flags))
5085 info.state |= (1<<MD_DISK_WRITEMOSTLY);
5087 info.major = info.minor = 0;
5088 info.raid_disk = -1;
5089 info.state = (1<<MD_DISK_REMOVED);
5092 if (copy_to_user(arg, &info, sizeof(info)))
5098 static int add_new_disk(mddev_t * mddev, mdu_disk_info_t *info)
5100 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
5102 dev_t dev = MKDEV(info->major,info->minor);
5104 if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
5107 if (!mddev->raid_disks) {
5109 /* expecting a device which has a superblock */
5110 rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
5113 "md: md_import_device returned %ld\n",
5115 return PTR_ERR(rdev);
5117 if (!list_empty(&mddev->disks)) {
5118 mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
5119 mdk_rdev_t, same_set);
5120 err = super_types[mddev->major_version]
5121 .load_super(rdev, rdev0, mddev->minor_version);
5124 "md: %s has different UUID to %s\n",
5125 bdevname(rdev->bdev,b),
5126 bdevname(rdev0->bdev,b2));
5131 err = bind_rdev_to_array(rdev, mddev);
5138 * add_new_disk can be used once the array is assembled
5139 * to add "hot spares". They must already have a superblock
5144 if (!mddev->pers->hot_add_disk) {
5146 "%s: personality does not support diskops!\n",
5150 if (mddev->persistent)
5151 rdev = md_import_device(dev, mddev->major_version,
5152 mddev->minor_version);
5154 rdev = md_import_device(dev, -1, -1);
5157 "md: md_import_device returned %ld\n",
5159 return PTR_ERR(rdev);
5161 /* set save_raid_disk if appropriate */
5162 if (!mddev->persistent) {
5163 if (info->state & (1<<MD_DISK_SYNC) &&
5164 info->raid_disk < mddev->raid_disks)
5165 rdev->raid_disk = info->raid_disk;
5167 rdev->raid_disk = -1;
5169 super_types[mddev->major_version].
5170 validate_super(mddev, rdev);
5171 rdev->saved_raid_disk = rdev->raid_disk;
5173 clear_bit(In_sync, &rdev->flags); /* just to be sure */
5174 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
5175 set_bit(WriteMostly, &rdev->flags);
5177 clear_bit(WriteMostly, &rdev->flags);
5179 rdev->raid_disk = -1;
5180 err = bind_rdev_to_array(rdev, mddev);
5181 if (!err && !mddev->pers->hot_remove_disk) {
5182 /* If there is hot_add_disk but no hot_remove_disk
5183 * then added disks for geometry changes,
5184 * and should be added immediately.
5186 super_types[mddev->major_version].
5187 validate_super(mddev, rdev);
5188 err = mddev->pers->hot_add_disk(mddev, rdev);
5190 unbind_rdev_from_array(rdev);
5195 sysfs_notify_dirent_safe(rdev->sysfs_state);
5197 md_update_sb(mddev, 1);
5198 if (mddev->degraded)
5199 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
5200 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5201 md_wakeup_thread(mddev->thread);
5205 /* otherwise, add_new_disk is only allowed
5206 * for major_version==0 superblocks
5208 if (mddev->major_version != 0) {
5209 printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n",
5214 if (!(info->state & (1<<MD_DISK_FAULTY))) {
5216 rdev = md_import_device(dev, -1, 0);
5219 "md: error, md_import_device() returned %ld\n",
5221 return PTR_ERR(rdev);
5223 rdev->desc_nr = info->number;
5224 if (info->raid_disk < mddev->raid_disks)
5225 rdev->raid_disk = info->raid_disk;
5227 rdev->raid_disk = -1;
5229 if (rdev->raid_disk < mddev->raid_disks)
5230 if (info->state & (1<<MD_DISK_SYNC))
5231 set_bit(In_sync, &rdev->flags);
5233 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
5234 set_bit(WriteMostly, &rdev->flags);
5236 if (!mddev->persistent) {
5237 printk(KERN_INFO "md: nonpersistent superblock ...\n");
5238 rdev->sb_start = rdev->bdev->bd_inode->i_size / 512;
5240 rdev->sb_start = calc_dev_sboffset(rdev->bdev);
5241 rdev->sectors = rdev->sb_start;
5243 err = bind_rdev_to_array(rdev, mddev);
5253 static int hot_remove_disk(mddev_t * mddev, dev_t dev)
5255 char b[BDEVNAME_SIZE];
5258 rdev = find_rdev(mddev, dev);
5262 if (rdev->raid_disk >= 0)
5265 kick_rdev_from_array(rdev);
5266 md_update_sb(mddev, 1);
5267 md_new_event(mddev);
5271 printk(KERN_WARNING "md: cannot remove active disk %s from %s ...\n",
5272 bdevname(rdev->bdev,b), mdname(mddev));
5276 static int hot_add_disk(mddev_t * mddev, dev_t dev)
5278 char b[BDEVNAME_SIZE];
5285 if (mddev->major_version != 0) {
5286 printk(KERN_WARNING "%s: HOT_ADD may only be used with"
5287 " version-0 superblocks.\n",
5291 if (!mddev->pers->hot_add_disk) {
5293 "%s: personality does not support diskops!\n",
5298 rdev = md_import_device(dev, -1, 0);
5301 "md: error, md_import_device() returned %ld\n",
5306 if (mddev->persistent)
5307 rdev->sb_start = calc_dev_sboffset(rdev->bdev);
5309 rdev->sb_start = rdev->bdev->bd_inode->i_size / 512;
5311 rdev->sectors = rdev->sb_start;
5313 if (test_bit(Faulty, &rdev->flags)) {
5315 "md: can not hot-add faulty %s disk to %s!\n",
5316 bdevname(rdev->bdev,b), mdname(mddev));
5320 clear_bit(In_sync, &rdev->flags);
5322 rdev->saved_raid_disk = -1;
5323 err = bind_rdev_to_array(rdev, mddev);
5328 * The rest should better be atomic, we can have disk failures
5329 * noticed in interrupt contexts ...
5332 rdev->raid_disk = -1;
5334 md_update_sb(mddev, 1);
5337 * Kick recovery, maybe this spare has to be added to the
5338 * array immediately.
5340 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5341 md_wakeup_thread(mddev->thread);
5342 md_new_event(mddev);
5350 static int set_bitmap_file(mddev_t *mddev, int fd)
5355 if (!mddev->pers->quiesce)
5357 if (mddev->recovery || mddev->sync_thread)
5359 /* we should be able to change the bitmap.. */
5365 return -EEXIST; /* cannot add when bitmap is present */
5366 mddev->bitmap_info.file = fget(fd);
5368 if (mddev->bitmap_info.file == NULL) {
5369 printk(KERN_ERR "%s: error: failed to get bitmap file\n",
5374 err = deny_bitmap_write_access(mddev->bitmap_info.file);
5376 printk(KERN_ERR "%s: error: bitmap file is already in use\n",
5378 fput(mddev->bitmap_info.file);
5379 mddev->bitmap_info.file = NULL;
5382 mddev->bitmap_info.offset = 0; /* file overrides offset */
5383 } else if (mddev->bitmap == NULL)
5384 return -ENOENT; /* cannot remove what isn't there */
5387 mddev->pers->quiesce(mddev, 1);
5389 err = bitmap_create(mddev);
5391 err = bitmap_load(mddev);
5393 if (fd < 0 || err) {
5394 bitmap_destroy(mddev);
5395 fd = -1; /* make sure to put the file */
5397 mddev->pers->quiesce(mddev, 0);
5400 if (mddev->bitmap_info.file) {
5401 restore_bitmap_write_access(mddev->bitmap_info.file);
5402 fput(mddev->bitmap_info.file);
5404 mddev->bitmap_info.file = NULL;
5411 * set_array_info is used two different ways
5412 * The original usage is when creating a new array.
5413 * In this usage, raid_disks is > 0 and it together with
5414 * level, size, not_persistent,layout,chunksize determine the
5415 * shape of the array.
5416 * This will always create an array with a type-0.90.0 superblock.
5417 * The newer usage is when assembling an array.
5418 * In this case raid_disks will be 0, and the major_version field is
5419 * use to determine which style super-blocks are to be found on the devices.
5420 * The minor and patch _version numbers are also kept incase the
5421 * super_block handler wishes to interpret them.
5423 static int set_array_info(mddev_t * mddev, mdu_array_info_t *info)
5426 if (info->raid_disks == 0) {
5427 /* just setting version number for superblock loading */
5428 if (info->major_version < 0 ||
5429 info->major_version >= ARRAY_SIZE(super_types) ||
5430 super_types[info->major_version].name == NULL) {
5431 /* maybe try to auto-load a module? */
5433 "md: superblock version %d not known\n",
5434 info->major_version);
5437 mddev->major_version = info->major_version;
5438 mddev->minor_version = info->minor_version;
5439 mddev->patch_version = info->patch_version;
5440 mddev->persistent = !info->not_persistent;
5441 /* ensure mddev_put doesn't delete this now that there
5442 * is some minimal configuration.
5444 mddev->ctime = get_seconds();
5447 mddev->major_version = MD_MAJOR_VERSION;
5448 mddev->minor_version = MD_MINOR_VERSION;
5449 mddev->patch_version = MD_PATCHLEVEL_VERSION;
5450 mddev->ctime = get_seconds();
5452 mddev->level = info->level;
5453 mddev->clevel[0] = 0;
5454 mddev->dev_sectors = 2 * (sector_t)info->size;
5455 mddev->raid_disks = info->raid_disks;
5456 /* don't set md_minor, it is determined by which /dev/md* was
5459 if (info->state & (1<<MD_SB_CLEAN))
5460 mddev->recovery_cp = MaxSector;
5462 mddev->recovery_cp = 0;
5463 mddev->persistent = ! info->not_persistent;
5464 mddev->external = 0;
5466 mddev->layout = info->layout;
5467 mddev->chunk_sectors = info->chunk_size >> 9;
5469 mddev->max_disks = MD_SB_DISKS;
5471 if (mddev->persistent)
5473 set_bit(MD_CHANGE_DEVS, &mddev->flags);
5475 mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9;
5476 mddev->bitmap_info.offset = 0;
5478 mddev->reshape_position = MaxSector;
5481 * Generate a 128 bit UUID
5483 get_random_bytes(mddev->uuid, 16);
5485 mddev->new_level = mddev->level;
5486 mddev->new_chunk_sectors = mddev->chunk_sectors;
5487 mddev->new_layout = mddev->layout;
5488 mddev->delta_disks = 0;
5493 void md_set_array_sectors(mddev_t *mddev, sector_t array_sectors)
5495 WARN(!mddev_is_locked(mddev), "%s: unlocked mddev!\n", __func__);
5497 if (mddev->external_size)
5500 mddev->array_sectors = array_sectors;
5502 EXPORT_SYMBOL(md_set_array_sectors);
5504 static int update_size(mddev_t *mddev, sector_t num_sectors)
5508 int fit = (num_sectors == 0);
5510 if (mddev->pers->resize == NULL)
5512 /* The "num_sectors" is the number of sectors of each device that
5513 * is used. This can only make sense for arrays with redundancy.
5514 * linear and raid0 always use whatever space is available. We can only
5515 * consider changing this number if no resync or reconstruction is
5516 * happening, and if the new size is acceptable. It must fit before the
5517 * sb_start or, if that is <data_offset, it must fit before the size
5518 * of each device. If num_sectors is zero, we find the largest size
5522 if (mddev->sync_thread)
5525 /* Sorry, cannot grow a bitmap yet, just remove it,
5529 list_for_each_entry(rdev, &mddev->disks, same_set) {
5530 sector_t avail = rdev->sectors;
5532 if (fit && (num_sectors == 0 || num_sectors > avail))
5533 num_sectors = avail;
5534 if (avail < num_sectors)
5537 rv = mddev->pers->resize(mddev, num_sectors);
5539 revalidate_disk(mddev->gendisk);
5543 static int update_raid_disks(mddev_t *mddev, int raid_disks)
5546 /* change the number of raid disks */
5547 if (mddev->pers->check_reshape == NULL)
5549 if (raid_disks <= 0 ||
5550 (mddev->max_disks && raid_disks >= mddev->max_disks))
5552 if (mddev->sync_thread || mddev->reshape_position != MaxSector)
5554 mddev->delta_disks = raid_disks - mddev->raid_disks;
5556 rv = mddev->pers->check_reshape(mddev);
5562 * update_array_info is used to change the configuration of an
5564 * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
5565 * fields in the info are checked against the array.
5566 * Any differences that cannot be handled will cause an error.
5567 * Normally, only one change can be managed at a time.
5569 static int update_array_info(mddev_t *mddev, mdu_array_info_t *info)
5575 /* calculate expected state,ignoring low bits */
5576 if (mddev->bitmap && mddev->bitmap_info.offset)
5577 state |= (1 << MD_SB_BITMAP_PRESENT);
5579 if (mddev->major_version != info->major_version ||
5580 mddev->minor_version != info->minor_version ||
5581 /* mddev->patch_version != info->patch_version || */
5582 mddev->ctime != info->ctime ||
5583 mddev->level != info->level ||
5584 /* mddev->layout != info->layout || */
5585 !mddev->persistent != info->not_persistent||
5586 mddev->chunk_sectors != info->chunk_size >> 9 ||
5587 /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
5588 ((state^info->state) & 0xfffffe00)
5591 /* Check there is only one change */
5592 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
5594 if (mddev->raid_disks != info->raid_disks)
5596 if (mddev->layout != info->layout)
5598 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT))
5605 if (mddev->layout != info->layout) {
5607 * we don't need to do anything at the md level, the
5608 * personality will take care of it all.
5610 if (mddev->pers->check_reshape == NULL)
5613 mddev->new_layout = info->layout;
5614 rv = mddev->pers->check_reshape(mddev);
5616 mddev->new_layout = mddev->layout;
5620 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
5621 rv = update_size(mddev, (sector_t)info->size * 2);
5623 if (mddev->raid_disks != info->raid_disks)
5624 rv = update_raid_disks(mddev, info->raid_disks);
5626 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) {
5627 if (mddev->pers->quiesce == NULL)
5629 if (mddev->recovery || mddev->sync_thread)
5631 if (info->state & (1<<MD_SB_BITMAP_PRESENT)) {
5632 /* add the bitmap */
5635 if (mddev->bitmap_info.default_offset == 0)
5637 mddev->bitmap_info.offset =
5638 mddev->bitmap_info.default_offset;
5639 mddev->pers->quiesce(mddev, 1);
5640 rv = bitmap_create(mddev);
5642 rv = bitmap_load(mddev);
5644 bitmap_destroy(mddev);
5645 mddev->pers->quiesce(mddev, 0);
5647 /* remove the bitmap */
5650 if (mddev->bitmap->file)
5652 mddev->pers->quiesce(mddev, 1);
5653 bitmap_destroy(mddev);
5654 mddev->pers->quiesce(mddev, 0);
5655 mddev->bitmap_info.offset = 0;
5658 md_update_sb(mddev, 1);
5662 static int set_disk_faulty(mddev_t *mddev, dev_t dev)
5666 if (mddev->pers == NULL)
5669 rdev = find_rdev(mddev, dev);
5673 md_error(mddev, rdev);
5678 * We have a problem here : there is no easy way to give a CHS
5679 * virtual geometry. We currently pretend that we have a 2 heads
5680 * 4 sectors (with a BIG number of cylinders...). This drives
5681 * dosfs just mad... ;-)
5683 static int md_getgeo(struct block_device *bdev, struct hd_geometry *geo)
5685 mddev_t *mddev = bdev->bd_disk->private_data;
5689 geo->cylinders = mddev->array_sectors / 8;
5693 static int md_ioctl(struct block_device *bdev, fmode_t mode,
5694 unsigned int cmd, unsigned long arg)
5697 void __user *argp = (void __user *)arg;
5698 mddev_t *mddev = NULL;
5701 if (!capable(CAP_SYS_ADMIN))
5705 * Commands dealing with the RAID driver but not any
5711 err = get_version(argp);
5714 case PRINT_RAID_DEBUG:
5722 autostart_arrays(arg);
5729 * Commands creating/starting a new array:
5732 mddev = bdev->bd_disk->private_data;
5739 err = mddev_lock(mddev);
5742 "md: ioctl lock interrupted, reason %d, cmd %d\n",
5749 case SET_ARRAY_INFO:
5751 mdu_array_info_t info;
5753 memset(&info, 0, sizeof(info));
5754 else if (copy_from_user(&info, argp, sizeof(info))) {
5759 err = update_array_info(mddev, &info);
5761 printk(KERN_WARNING "md: couldn't update"
5762 " array info. %d\n", err);
5767 if (!list_empty(&mddev->disks)) {
5769 "md: array %s already has disks!\n",
5774 if (mddev->raid_disks) {
5776 "md: array %s already initialised!\n",
5781 err = set_array_info(mddev, &info);
5783 printk(KERN_WARNING "md: couldn't set"
5784 " array info. %d\n", err);
5794 * Commands querying/configuring an existing array:
5796 /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
5797 * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */
5798 if ((!mddev->raid_disks && !mddev->external)
5799 && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY
5800 && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE
5801 && cmd != GET_BITMAP_FILE) {
5807 * Commands even a read-only array can execute:
5811 case GET_ARRAY_INFO:
5812 err = get_array_info(mddev, argp);
5815 case GET_BITMAP_FILE:
5816 err = get_bitmap_file(mddev, argp);
5820 err = get_disk_info(mddev, argp);
5823 case RESTART_ARRAY_RW:
5824 err = restart_array(mddev);
5828 err = do_md_stop(mddev, 0, 1);
5832 err = md_set_readonly(mddev, 1);
5836 if (get_user(ro, (int __user *)(arg))) {
5842 /* if the bdev is going readonly the value of mddev->ro
5843 * does not matter, no writes are coming
5848 /* are we are already prepared for writes? */
5852 /* transitioning to readauto need only happen for
5853 * arrays that call md_write_start
5856 err = restart_array(mddev);
5859 set_disk_ro(mddev->gendisk, 0);
5866 * The remaining ioctls are changing the state of the
5867 * superblock, so we do not allow them on read-only arrays.
5868 * However non-MD ioctls (e.g. get-size) will still come through
5869 * here and hit the 'default' below, so only disallow
5870 * 'md' ioctls, and switch to rw mode if started auto-readonly.
5872 if (_IOC_TYPE(cmd) == MD_MAJOR && mddev->ro && mddev->pers) {
5873 if (mddev->ro == 2) {
5875 sysfs_notify_dirent_safe(mddev->sysfs_state);
5876 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5877 md_wakeup_thread(mddev->thread);
5888 mdu_disk_info_t info;
5889 if (copy_from_user(&info, argp, sizeof(info)))
5892 err = add_new_disk(mddev, &info);
5896 case HOT_REMOVE_DISK:
5897 err = hot_remove_disk(mddev, new_decode_dev(arg));
5901 err = hot_add_disk(mddev, new_decode_dev(arg));
5904 case SET_DISK_FAULTY:
5905 err = set_disk_faulty(mddev, new_decode_dev(arg));
5909 err = do_md_run(mddev);
5912 case SET_BITMAP_FILE:
5913 err = set_bitmap_file(mddev, (int)arg);
5923 if (mddev->hold_active == UNTIL_IOCTL &&
5925 mddev->hold_active = 0;
5926 mddev_unlock(mddev);
5935 #ifdef CONFIG_COMPAT
5936 static int md_compat_ioctl(struct block_device *bdev, fmode_t mode,
5937 unsigned int cmd, unsigned long arg)
5940 case HOT_REMOVE_DISK:
5942 case SET_DISK_FAULTY:
5943 case SET_BITMAP_FILE:
5944 /* These take in integer arg, do not convert */
5947 arg = (unsigned long)compat_ptr(arg);
5951 return md_ioctl(bdev, mode, cmd, arg);
5953 #endif /* CONFIG_COMPAT */
5955 static int md_open(struct block_device *bdev, fmode_t mode)
5958 * Succeed if we can lock the mddev, which confirms that
5959 * it isn't being stopped right now.
5961 mddev_t *mddev = mddev_find(bdev->bd_dev);
5965 if (mddev->gendisk != bdev->bd_disk) {
5966 /* we are racing with mddev_put which is discarding this
5970 /* Wait until bdev->bd_disk is definitely gone */
5971 flush_scheduled_work();
5972 /* Then retry the open from the top */
5974 return -ERESTARTSYS;
5976 BUG_ON(mddev != bdev->bd_disk->private_data);
5978 if ((err = mutex_lock_interruptible(&mddev->open_mutex)))
5982 atomic_inc(&mddev->openers);
5983 mutex_unlock(&mddev->open_mutex);
5985 check_disk_size_change(mddev->gendisk, bdev);
5991 static int md_release(struct gendisk *disk, fmode_t mode)
5993 mddev_t *mddev = disk->private_data;
5997 atomic_dec(&mddev->openers);
6003 static const struct block_device_operations md_fops =
6005 .owner = THIS_MODULE,
6007 .release = md_release,
6009 #ifdef CONFIG_COMPAT
6010 .compat_ioctl = md_compat_ioctl,
6012 .getgeo = md_getgeo,
6015 static int md_thread(void * arg)
6017 mdk_thread_t *thread = arg;
6020 * md_thread is a 'system-thread', it's priority should be very
6021 * high. We avoid resource deadlocks individually in each
6022 * raid personality. (RAID5 does preallocation) We also use RR and
6023 * the very same RT priority as kswapd, thus we will never get
6024 * into a priority inversion deadlock.
6026 * we definitely have to have equal or higher priority than
6027 * bdflush, otherwise bdflush will deadlock if there are too
6028 * many dirty RAID5 blocks.
6031 allow_signal(SIGKILL);
6032 while (!kthread_should_stop()) {
6034 /* We need to wait INTERRUPTIBLE so that
6035 * we don't add to the load-average.
6036 * That means we need to be sure no signals are
6039 if (signal_pending(current))
6040 flush_signals(current);
6042 wait_event_interruptible_timeout
6044 test_bit(THREAD_WAKEUP, &thread->flags)
6045 || kthread_should_stop(),
6048 clear_bit(THREAD_WAKEUP, &thread->flags);
6050 thread->run(thread->mddev);
6056 void md_wakeup_thread(mdk_thread_t *thread)
6059 dprintk("md: waking up MD thread %s.\n", thread->tsk->comm);
6060 set_bit(THREAD_WAKEUP, &thread->flags);
6061 wake_up(&thread->wqueue);
6065 mdk_thread_t *md_register_thread(void (*run) (mddev_t *), mddev_t *mddev,
6068 mdk_thread_t *thread;
6070 thread = kzalloc(sizeof(mdk_thread_t), GFP_KERNEL);
6074 init_waitqueue_head(&thread->wqueue);
6077 thread->mddev = mddev;
6078 thread->timeout = MAX_SCHEDULE_TIMEOUT;
6079 thread->tsk = kthread_run(md_thread, thread,
6081 mdname(thread->mddev),
6082 name ?: mddev->pers->name);
6083 if (IS_ERR(thread->tsk)) {
6090 void md_unregister_thread(mdk_thread_t *thread)
6094 dprintk("interrupting MD-thread pid %d\n", task_pid_nr(thread->tsk));
6096 kthread_stop(thread->tsk);
6100 void md_error(mddev_t *mddev, mdk_rdev_t *rdev)
6107 if (!rdev || test_bit(Faulty, &rdev->flags))
6110 if (mddev->external)
6111 set_bit(Blocked, &rdev->flags);
6113 dprintk("md_error dev:%s, rdev:(%d:%d), (caller: %p,%p,%p,%p).\n",
6115 MAJOR(rdev->bdev->bd_dev), MINOR(rdev->bdev->bd_dev),
6116 __builtin_return_address(0),__builtin_return_address(1),
6117 __builtin_return_address(2),__builtin_return_address(3));
6121 if (!mddev->pers->error_handler)
6123 mddev->pers->error_handler(mddev,rdev);
6124 if (mddev->degraded)
6125 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
6126 sysfs_notify_dirent_safe(rdev->sysfs_state);
6127 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6128 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6129 md_wakeup_thread(mddev->thread);
6130 if (mddev->event_work.func)
6131 schedule_work(&mddev->event_work);
6132 md_new_event_inintr(mddev);
6135 /* seq_file implementation /proc/mdstat */
6137 static void status_unused(struct seq_file *seq)
6142 seq_printf(seq, "unused devices: ");
6144 list_for_each_entry(rdev, &pending_raid_disks, same_set) {
6145 char b[BDEVNAME_SIZE];
6147 seq_printf(seq, "%s ",
6148 bdevname(rdev->bdev,b));
6151 seq_printf(seq, "<none>");
6153 seq_printf(seq, "\n");
6157 static void status_resync(struct seq_file *seq, mddev_t * mddev)
6159 sector_t max_sectors, resync, res;
6160 unsigned long dt, db;
6163 unsigned int per_milli;
6165 resync = mddev->curr_resync - atomic_read(&mddev->recovery_active);
6167 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
6168 max_sectors = mddev->resync_max_sectors;
6170 max_sectors = mddev->dev_sectors;
6173 * Should not happen.
6179 /* Pick 'scale' such that (resync>>scale)*1000 will fit
6180 * in a sector_t, and (max_sectors>>scale) will fit in a
6181 * u32, as those are the requirements for sector_div.
6182 * Thus 'scale' must be at least 10
6185 if (sizeof(sector_t) > sizeof(unsigned long)) {
6186 while ( max_sectors/2 > (1ULL<<(scale+32)))
6189 res = (resync>>scale)*1000;
6190 sector_div(res, (u32)((max_sectors>>scale)+1));
6194 int i, x = per_milli/50, y = 20-x;
6195 seq_printf(seq, "[");
6196 for (i = 0; i < x; i++)
6197 seq_printf(seq, "=");
6198 seq_printf(seq, ">");
6199 for (i = 0; i < y; i++)
6200 seq_printf(seq, ".");
6201 seq_printf(seq, "] ");
6203 seq_printf(seq, " %s =%3u.%u%% (%llu/%llu)",
6204 (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)?
6206 (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)?
6208 (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
6209 "resync" : "recovery"))),
6210 per_milli/10, per_milli % 10,
6211 (unsigned long long) resync/2,
6212 (unsigned long long) max_sectors/2);
6215 * dt: time from mark until now
6216 * db: blocks written from mark until now
6217 * rt: remaining time
6219 * rt is a sector_t, so could be 32bit or 64bit.
6220 * So we divide before multiply in case it is 32bit and close
6222 * We scale the divisor (db) by 32 to avoid loosing precision
6223 * near the end of resync when the number of remaining sectors
6225 * We then divide rt by 32 after multiplying by db to compensate.
6226 * The '+1' avoids division by zero if db is very small.
6228 dt = ((jiffies - mddev->resync_mark) / HZ);
6230 db = (mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active))
6231 - mddev->resync_mark_cnt;
6233 rt = max_sectors - resync; /* number of remaining sectors */
6234 sector_div(rt, db/32+1);
6238 seq_printf(seq, " finish=%lu.%lumin", (unsigned long)rt / 60,
6239 ((unsigned long)rt % 60)/6);
6241 seq_printf(seq, " speed=%ldK/sec", db/2/dt);
6244 static void *md_seq_start(struct seq_file *seq, loff_t *pos)
6246 struct list_head *tmp;
6256 spin_lock(&all_mddevs_lock);
6257 list_for_each(tmp,&all_mddevs)
6259 mddev = list_entry(tmp, mddev_t, all_mddevs);
6261 spin_unlock(&all_mddevs_lock);
6264 spin_unlock(&all_mddevs_lock);
6266 return (void*)2;/* tail */
6270 static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
6272 struct list_head *tmp;
6273 mddev_t *next_mddev, *mddev = v;
6279 spin_lock(&all_mddevs_lock);
6281 tmp = all_mddevs.next;
6283 tmp = mddev->all_mddevs.next;
6284 if (tmp != &all_mddevs)
6285 next_mddev = mddev_get(list_entry(tmp,mddev_t,all_mddevs));
6287 next_mddev = (void*)2;
6290 spin_unlock(&all_mddevs_lock);
6298 static void md_seq_stop(struct seq_file *seq, void *v)
6302 if (mddev && v != (void*)1 && v != (void*)2)
6306 struct mdstat_info {
6310 static int md_seq_show(struct seq_file *seq, void *v)
6315 struct mdstat_info *mi = seq->private;
6316 struct bitmap *bitmap;
6318 if (v == (void*)1) {
6319 struct mdk_personality *pers;
6320 seq_printf(seq, "Personalities : ");
6321 spin_lock(&pers_lock);
6322 list_for_each_entry(pers, &pers_list, list)
6323 seq_printf(seq, "[%s] ", pers->name);
6325 spin_unlock(&pers_lock);
6326 seq_printf(seq, "\n");
6327 mi->event = atomic_read(&md_event_count);
6330 if (v == (void*)2) {
6335 if (mddev_lock(mddev) < 0)
6338 if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
6339 seq_printf(seq, "%s : %sactive", mdname(mddev),
6340 mddev->pers ? "" : "in");
6343 seq_printf(seq, " (read-only)");
6345 seq_printf(seq, " (auto-read-only)");
6346 seq_printf(seq, " %s", mddev->pers->name);
6350 list_for_each_entry(rdev, &mddev->disks, same_set) {
6351 char b[BDEVNAME_SIZE];
6352 seq_printf(seq, " %s[%d]",
6353 bdevname(rdev->bdev,b), rdev->desc_nr);
6354 if (test_bit(WriteMostly, &rdev->flags))
6355 seq_printf(seq, "(W)");
6356 if (test_bit(Faulty, &rdev->flags)) {
6357 seq_printf(seq, "(F)");
6359 } else if (rdev->raid_disk < 0)
6360 seq_printf(seq, "(S)"); /* spare */
6361 sectors += rdev->sectors;
6364 if (!list_empty(&mddev->disks)) {
6366 seq_printf(seq, "\n %llu blocks",
6367 (unsigned long long)
6368 mddev->array_sectors / 2);
6370 seq_printf(seq, "\n %llu blocks",
6371 (unsigned long long)sectors / 2);
6373 if (mddev->persistent) {
6374 if (mddev->major_version != 0 ||
6375 mddev->minor_version != 90) {
6376 seq_printf(seq," super %d.%d",
6377 mddev->major_version,
6378 mddev->minor_version);
6380 } else if (mddev->external)
6381 seq_printf(seq, " super external:%s",
6382 mddev->metadata_type);
6384 seq_printf(seq, " super non-persistent");
6387 mddev->pers->status(seq, mddev);
6388 seq_printf(seq, "\n ");
6389 if (mddev->pers->sync_request) {
6390 if (mddev->curr_resync > 2) {
6391 status_resync(seq, mddev);
6392 seq_printf(seq, "\n ");
6393 } else if (mddev->curr_resync == 1 || mddev->curr_resync == 2)
6394 seq_printf(seq, "\tresync=DELAYED\n ");
6395 else if (mddev->recovery_cp < MaxSector)
6396 seq_printf(seq, "\tresync=PENDING\n ");
6399 seq_printf(seq, "\n ");
6401 if ((bitmap = mddev->bitmap)) {
6402 unsigned long chunk_kb;
6403 unsigned long flags;
6404 spin_lock_irqsave(&bitmap->lock, flags);
6405 chunk_kb = mddev->bitmap_info.chunksize >> 10;
6406 seq_printf(seq, "bitmap: %lu/%lu pages [%luKB], "
6408 bitmap->pages - bitmap->missing_pages,
6410 (bitmap->pages - bitmap->missing_pages)
6411 << (PAGE_SHIFT - 10),
6412 chunk_kb ? chunk_kb : mddev->bitmap_info.chunksize,
6413 chunk_kb ? "KB" : "B");
6415 seq_printf(seq, ", file: ");
6416 seq_path(seq, &bitmap->file->f_path, " \t\n");
6419 seq_printf(seq, "\n");
6420 spin_unlock_irqrestore(&bitmap->lock, flags);
6423 seq_printf(seq, "\n");
6425 mddev_unlock(mddev);
6430 static const struct seq_operations md_seq_ops = {
6431 .start = md_seq_start,
6432 .next = md_seq_next,
6433 .stop = md_seq_stop,
6434 .show = md_seq_show,
6437 static int md_seq_open(struct inode *inode, struct file *file)
6440 struct mdstat_info *mi = kmalloc(sizeof(*mi), GFP_KERNEL);
6444 error = seq_open(file, &md_seq_ops);
6448 struct seq_file *p = file->private_data;
6450 mi->event = atomic_read(&md_event_count);
6455 static unsigned int mdstat_poll(struct file *filp, poll_table *wait)
6457 struct seq_file *m = filp->private_data;
6458 struct mdstat_info *mi = m->private;
6461 poll_wait(filp, &md_event_waiters, wait);
6463 /* always allow read */
6464 mask = POLLIN | POLLRDNORM;
6466 if (mi->event != atomic_read(&md_event_count))
6467 mask |= POLLERR | POLLPRI;
6471 static const struct file_operations md_seq_fops = {
6472 .owner = THIS_MODULE,
6473 .open = md_seq_open,
6475 .llseek = seq_lseek,
6476 .release = seq_release_private,
6477 .poll = mdstat_poll,
6480 int register_md_personality(struct mdk_personality *p)
6482 spin_lock(&pers_lock);
6483 list_add_tail(&p->list, &pers_list);
6484 printk(KERN_INFO "md: %s personality registered for level %d\n", p->name, p->level);
6485 spin_unlock(&pers_lock);
6489 int unregister_md_personality(struct mdk_personality *p)
6491 printk(KERN_INFO "md: %s personality unregistered\n", p->name);
6492 spin_lock(&pers_lock);
6493 list_del_init(&p->list);
6494 spin_unlock(&pers_lock);
6498 static int is_mddev_idle(mddev_t *mddev, int init)
6506 rdev_for_each_rcu(rdev, mddev) {
6507 struct gendisk *disk = rdev->bdev->bd_contains->bd_disk;
6508 curr_events = (int)part_stat_read(&disk->part0, sectors[0]) +
6509 (int)part_stat_read(&disk->part0, sectors[1]) -
6510 atomic_read(&disk->sync_io);
6511 /* sync IO will cause sync_io to increase before the disk_stats
6512 * as sync_io is counted when a request starts, and
6513 * disk_stats is counted when it completes.
6514 * So resync activity will cause curr_events to be smaller than
6515 * when there was no such activity.
6516 * non-sync IO will cause disk_stat to increase without
6517 * increasing sync_io so curr_events will (eventually)
6518 * be larger than it was before. Once it becomes
6519 * substantially larger, the test below will cause
6520 * the array to appear non-idle, and resync will slow
6522 * If there is a lot of outstanding resync activity when
6523 * we set last_event to curr_events, then all that activity
6524 * completing might cause the array to appear non-idle
6525 * and resync will be slowed down even though there might
6526 * not have been non-resync activity. This will only
6527 * happen once though. 'last_events' will soon reflect
6528 * the state where there is little or no outstanding
6529 * resync requests, and further resync activity will
6530 * always make curr_events less than last_events.
6533 if (init || curr_events - rdev->last_events > 64) {
6534 rdev->last_events = curr_events;
6542 void md_done_sync(mddev_t *mddev, int blocks, int ok)
6544 /* another "blocks" (512byte) blocks have been synced */
6545 atomic_sub(blocks, &mddev->recovery_active);
6546 wake_up(&mddev->recovery_wait);
6548 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6549 md_wakeup_thread(mddev->thread);
6550 // stop recovery, signal do_sync ....
6555 /* md_write_start(mddev, bi)
6556 * If we need to update some array metadata (e.g. 'active' flag
6557 * in superblock) before writing, schedule a superblock update
6558 * and wait for it to complete.
6560 void md_write_start(mddev_t *mddev, struct bio *bi)
6563 if (bio_data_dir(bi) != WRITE)
6566 BUG_ON(mddev->ro == 1);
6567 if (mddev->ro == 2) {
6568 /* need to switch to read/write */
6570 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6571 md_wakeup_thread(mddev->thread);
6572 md_wakeup_thread(mddev->sync_thread);
6575 atomic_inc(&mddev->writes_pending);
6576 if (mddev->safemode == 1)
6577 mddev->safemode = 0;
6578 if (mddev->in_sync) {
6579 spin_lock_irq(&mddev->write_lock);
6580 if (mddev->in_sync) {
6582 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
6583 md_wakeup_thread(mddev->thread);
6586 spin_unlock_irq(&mddev->write_lock);
6589 sysfs_notify_dirent_safe(mddev->sysfs_state);
6590 wait_event(mddev->sb_wait,
6591 !test_bit(MD_CHANGE_CLEAN, &mddev->flags) &&
6592 !test_bit(MD_CHANGE_PENDING, &mddev->flags));
6595 void md_write_end(mddev_t *mddev)
6597 if (atomic_dec_and_test(&mddev->writes_pending)) {
6598 if (mddev->safemode == 2)
6599 md_wakeup_thread(mddev->thread);
6600 else if (mddev->safemode_delay)
6601 mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay);
6605 /* md_allow_write(mddev)
6606 * Calling this ensures that the array is marked 'active' so that writes
6607 * may proceed without blocking. It is important to call this before
6608 * attempting a GFP_KERNEL allocation while holding the mddev lock.
6609 * Must be called with mddev_lock held.
6611 * In the ->external case MD_CHANGE_CLEAN can not be cleared until mddev->lock
6612 * is dropped, so return -EAGAIN after notifying userspace.
6614 int md_allow_write(mddev_t *mddev)
6620 if (!mddev->pers->sync_request)
6623 spin_lock_irq(&mddev->write_lock);
6624 if (mddev->in_sync) {
6626 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
6627 if (mddev->safemode_delay &&
6628 mddev->safemode == 0)
6629 mddev->safemode = 1;
6630 spin_unlock_irq(&mddev->write_lock);
6631 md_update_sb(mddev, 0);
6632 sysfs_notify_dirent_safe(mddev->sysfs_state);
6634 spin_unlock_irq(&mddev->write_lock);
6636 if (test_bit(MD_CHANGE_CLEAN, &mddev->flags))
6641 EXPORT_SYMBOL_GPL(md_allow_write);
6643 void md_unplug(mddev_t *mddev)
6646 blk_unplug(mddev->queue);
6648 mddev->plug->unplug_fn(mddev->plug);
6651 #define SYNC_MARKS 10
6652 #define SYNC_MARK_STEP (3*HZ)
6653 void md_do_sync(mddev_t *mddev)
6656 unsigned int currspeed = 0,
6658 sector_t max_sectors,j, io_sectors;
6659 unsigned long mark[SYNC_MARKS];
6660 sector_t mark_cnt[SYNC_MARKS];
6662 struct list_head *tmp;
6663 sector_t last_check;
6668 /* just incase thread restarts... */
6669 if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
6671 if (mddev->ro) /* never try to sync a read-only array */
6674 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
6675 if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
6676 desc = "data-check";
6677 else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
6678 desc = "requested-resync";
6681 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
6686 /* we overload curr_resync somewhat here.
6687 * 0 == not engaged in resync at all
6688 * 2 == checking that there is no conflict with another sync
6689 * 1 == like 2, but have yielded to allow conflicting resync to
6691 * other == active in resync - this many blocks
6693 * Before starting a resync we must have set curr_resync to
6694 * 2, and then checked that every "conflicting" array has curr_resync
6695 * less than ours. When we find one that is the same or higher
6696 * we wait on resync_wait. To avoid deadlock, we reduce curr_resync
6697 * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
6698 * This will mean we have to start checking from the beginning again.
6703 mddev->curr_resync = 2;
6706 if (kthread_should_stop())
6707 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6709 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
6711 for_each_mddev(mddev2, tmp) {
6712 if (mddev2 == mddev)
6714 if (!mddev->parallel_resync
6715 && mddev2->curr_resync
6716 && match_mddev_units(mddev, mddev2)) {
6718 if (mddev < mddev2 && mddev->curr_resync == 2) {
6719 /* arbitrarily yield */
6720 mddev->curr_resync = 1;
6721 wake_up(&resync_wait);
6723 if (mddev > mddev2 && mddev->curr_resync == 1)
6724 /* no need to wait here, we can wait the next
6725 * time 'round when curr_resync == 2
6728 /* We need to wait 'interruptible' so as not to
6729 * contribute to the load average, and not to
6730 * be caught by 'softlockup'
6732 prepare_to_wait(&resync_wait, &wq, TASK_INTERRUPTIBLE);
6733 if (!kthread_should_stop() &&
6734 mddev2->curr_resync >= mddev->curr_resync) {
6735 printk(KERN_INFO "md: delaying %s of %s"
6736 " until %s has finished (they"
6737 " share one or more physical units)\n",
6738 desc, mdname(mddev), mdname(mddev2));
6740 if (signal_pending(current))
6741 flush_signals(current);
6743 finish_wait(&resync_wait, &wq);
6746 finish_wait(&resync_wait, &wq);
6749 } while (mddev->curr_resync < 2);
6752 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
6753 /* resync follows the size requested by the personality,
6754 * which defaults to physical size, but can be virtual size
6756 max_sectors = mddev->resync_max_sectors;
6757 mddev->resync_mismatches = 0;
6758 /* we don't use the checkpoint if there's a bitmap */
6759 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
6760 j = mddev->resync_min;
6761 else if (!mddev->bitmap)
6762 j = mddev->recovery_cp;
6764 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
6765 max_sectors = mddev->dev_sectors;
6767 /* recovery follows the physical size of devices */
6768 max_sectors = mddev->dev_sectors;
6771 list_for_each_entry_rcu(rdev, &mddev->disks, same_set)
6772 if (rdev->raid_disk >= 0 &&
6773 !test_bit(Faulty, &rdev->flags) &&
6774 !test_bit(In_sync, &rdev->flags) &&
6775 rdev->recovery_offset < j)
6776 j = rdev->recovery_offset;
6780 printk(KERN_INFO "md: %s of RAID array %s\n", desc, mdname(mddev));
6781 printk(KERN_INFO "md: minimum _guaranteed_ speed:"
6782 " %d KB/sec/disk.\n", speed_min(mddev));
6783 printk(KERN_INFO "md: using maximum available idle IO bandwidth "
6784 "(but not more than %d KB/sec) for %s.\n",
6785 speed_max(mddev), desc);
6787 is_mddev_idle(mddev, 1); /* this initializes IO event counters */
6790 for (m = 0; m < SYNC_MARKS; m++) {
6792 mark_cnt[m] = io_sectors;
6795 mddev->resync_mark = mark[last_mark];
6796 mddev->resync_mark_cnt = mark_cnt[last_mark];
6799 * Tune reconstruction:
6801 window = 32*(PAGE_SIZE/512);
6802 printk(KERN_INFO "md: using %dk window, over a total of %llu blocks.\n",
6803 window/2,(unsigned long long) max_sectors/2);
6805 atomic_set(&mddev->recovery_active, 0);
6810 "md: resuming %s of %s from checkpoint.\n",
6811 desc, mdname(mddev));
6812 mddev->curr_resync = j;
6814 mddev->curr_resync_completed = mddev->curr_resync;
6816 while (j < max_sectors) {
6821 if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
6822 ((mddev->curr_resync > mddev->curr_resync_completed &&
6823 (mddev->curr_resync - mddev->curr_resync_completed)
6824 > (max_sectors >> 4)) ||
6825 (j - mddev->curr_resync_completed)*2
6826 >= mddev->resync_max - mddev->curr_resync_completed
6828 /* time to update curr_resync_completed */
6830 wait_event(mddev->recovery_wait,
6831 atomic_read(&mddev->recovery_active) == 0);
6832 mddev->curr_resync_completed =
6834 if (mddev->persistent)
6835 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
6836 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
6839 while (j >= mddev->resync_max && !kthread_should_stop()) {
6840 /* As this condition is controlled by user-space,
6841 * we can block indefinitely, so use '_interruptible'
6842 * to avoid triggering warnings.
6844 flush_signals(current); /* just in case */
6845 wait_event_interruptible(mddev->recovery_wait,
6846 mddev->resync_max > j
6847 || kthread_should_stop());
6850 if (kthread_should_stop())
6853 sectors = mddev->pers->sync_request(mddev, j, &skipped,
6854 currspeed < speed_min(mddev));
6856 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6860 if (!skipped) { /* actual IO requested */
6861 io_sectors += sectors;
6862 atomic_add(sectors, &mddev->recovery_active);
6866 if (j>1) mddev->curr_resync = j;
6867 mddev->curr_mark_cnt = io_sectors;
6868 if (last_check == 0)
6869 /* this is the earliers that rebuilt will be
6870 * visible in /proc/mdstat
6872 md_new_event(mddev);
6874 if (last_check + window > io_sectors || j == max_sectors)
6877 last_check = io_sectors;
6879 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
6883 if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) {
6885 int next = (last_mark+1) % SYNC_MARKS;
6887 mddev->resync_mark = mark[next];
6888 mddev->resync_mark_cnt = mark_cnt[next];
6889 mark[next] = jiffies;
6890 mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active);
6895 if (kthread_should_stop())
6900 * this loop exits only if either when we are slower than
6901 * the 'hard' speed limit, or the system was IO-idle for
6903 * the system might be non-idle CPU-wise, but we only care
6904 * about not overloading the IO subsystem. (things like an
6905 * e2fsck being done on the RAID array should execute fast)
6910 currspeed = ((unsigned long)(io_sectors-mddev->resync_mark_cnt))/2
6911 /((jiffies-mddev->resync_mark)/HZ +1) +1;
6913 if (currspeed > speed_min(mddev)) {
6914 if ((currspeed > speed_max(mddev)) ||
6915 !is_mddev_idle(mddev, 0)) {
6921 printk(KERN_INFO "md: %s: %s done.\n",mdname(mddev), desc);
6923 * this also signals 'finished resyncing' to md_stop
6928 wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
6930 /* tell personality that we are finished */
6931 mddev->pers->sync_request(mddev, max_sectors, &skipped, 1);
6933 if (!test_bit(MD_RECOVERY_CHECK, &mddev->recovery) &&
6934 mddev->curr_resync > 2) {
6935 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
6936 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
6937 if (mddev->curr_resync >= mddev->recovery_cp) {
6939 "md: checkpointing %s of %s.\n",
6940 desc, mdname(mddev));
6941 mddev->recovery_cp = mddev->curr_resync;
6944 mddev->recovery_cp = MaxSector;
6946 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery))
6947 mddev->curr_resync = MaxSector;
6949 list_for_each_entry_rcu(rdev, &mddev->disks, same_set)
6950 if (rdev->raid_disk >= 0 &&
6951 mddev->delta_disks >= 0 &&
6952 !test_bit(Faulty, &rdev->flags) &&
6953 !test_bit(In_sync, &rdev->flags) &&
6954 rdev->recovery_offset < mddev->curr_resync)
6955 rdev->recovery_offset = mddev->curr_resync;
6959 set_bit(MD_CHANGE_DEVS, &mddev->flags);
6962 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
6963 /* We completed so min/max setting can be forgotten if used. */
6964 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
6965 mddev->resync_min = 0;
6966 mddev->resync_max = MaxSector;
6967 } else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
6968 mddev->resync_min = mddev->curr_resync_completed;
6969 mddev->curr_resync = 0;
6970 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery))
6971 mddev->curr_resync_completed = 0;
6972 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
6973 wake_up(&resync_wait);
6974 set_bit(MD_RECOVERY_DONE, &mddev->recovery);
6975 md_wakeup_thread(mddev->thread);
6980 * got a signal, exit.
6983 "md: md_do_sync() got signal ... exiting\n");
6984 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6988 EXPORT_SYMBOL_GPL(md_do_sync);
6991 static int remove_and_add_spares(mddev_t *mddev)
6996 mddev->curr_resync_completed = 0;
6998 list_for_each_entry(rdev, &mddev->disks, same_set)
6999 if (rdev->raid_disk >= 0 &&
7000 !test_bit(Blocked, &rdev->flags) &&
7001 (test_bit(Faulty, &rdev->flags) ||
7002 ! test_bit(In_sync, &rdev->flags)) &&
7003 atomic_read(&rdev->nr_pending)==0) {
7004 if (mddev->pers->hot_remove_disk(
7005 mddev, rdev->raid_disk)==0) {
7007 sprintf(nm,"rd%d", rdev->raid_disk);
7008 sysfs_remove_link(&mddev->kobj, nm);
7009 rdev->raid_disk = -1;
7013 if (mddev->degraded && ! mddev->ro && !mddev->recovery_disabled) {
7014 list_for_each_entry(rdev, &mddev->disks, same_set) {
7015 if (rdev->raid_disk >= 0 &&
7016 !test_bit(In_sync, &rdev->flags) &&
7017 !test_bit(Blocked, &rdev->flags))
7019 if (rdev->raid_disk < 0
7020 && !test_bit(Faulty, &rdev->flags)) {
7021 rdev->recovery_offset = 0;
7023 hot_add_disk(mddev, rdev) == 0) {
7025 sprintf(nm, "rd%d", rdev->raid_disk);
7026 if (sysfs_create_link(&mddev->kobj,
7028 /* failure here is OK */;
7030 md_new_event(mddev);
7031 set_bit(MD_CHANGE_DEVS, &mddev->flags);
7040 * This routine is regularly called by all per-raid-array threads to
7041 * deal with generic issues like resync and super-block update.
7042 * Raid personalities that don't have a thread (linear/raid0) do not
7043 * need this as they never do any recovery or update the superblock.
7045 * It does not do any resync itself, but rather "forks" off other threads
7046 * to do that as needed.
7047 * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
7048 * "->recovery" and create a thread at ->sync_thread.
7049 * When the thread finishes it sets MD_RECOVERY_DONE
7050 * and wakeups up this thread which will reap the thread and finish up.
7051 * This thread also removes any faulty devices (with nr_pending == 0).
7053 * The overall approach is:
7054 * 1/ if the superblock needs updating, update it.
7055 * 2/ If a recovery thread is running, don't do anything else.
7056 * 3/ If recovery has finished, clean up, possibly marking spares active.
7057 * 4/ If there are any faulty devices, remove them.
7058 * 5/ If array is degraded, try to add spares devices
7059 * 6/ If array has spares or is not in-sync, start a resync thread.
7061 void md_check_recovery(mddev_t *mddev)
7067 bitmap_daemon_work(mddev);
7072 if (signal_pending(current)) {
7073 if (mddev->pers->sync_request && !mddev->external) {
7074 printk(KERN_INFO "md: %s in immediate safe mode\n",
7076 mddev->safemode = 2;
7078 flush_signals(current);
7081 if (mddev->ro && !test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
7084 (mddev->flags && !mddev->external) ||
7085 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
7086 test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
7087 (mddev->external == 0 && mddev->safemode == 1) ||
7088 (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending)
7089 && !mddev->in_sync && mddev->recovery_cp == MaxSector)
7093 if (mddev_trylock(mddev)) {
7097 /* Only thing we do on a ro array is remove
7100 remove_and_add_spares(mddev);
7101 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7105 if (!mddev->external) {
7107 spin_lock_irq(&mddev->write_lock);
7108 if (mddev->safemode &&
7109 !atomic_read(&mddev->writes_pending) &&
7111 mddev->recovery_cp == MaxSector) {
7114 if (mddev->persistent)
7115 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
7117 if (mddev->safemode == 1)
7118 mddev->safemode = 0;
7119 spin_unlock_irq(&mddev->write_lock);
7121 sysfs_notify_dirent_safe(mddev->sysfs_state);
7125 md_update_sb(mddev, 0);
7127 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
7128 !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
7129 /* resync/recovery still happening */
7130 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7133 if (mddev->sync_thread) {
7134 /* resync has finished, collect result */
7135 md_unregister_thread(mddev->sync_thread);
7136 mddev->sync_thread = NULL;
7137 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
7138 !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
7140 /* activate any spares */
7141 if (mddev->pers->spare_active(mddev))
7142 sysfs_notify(&mddev->kobj, NULL,
7145 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
7146 mddev->pers->finish_reshape)
7147 mddev->pers->finish_reshape(mddev);
7148 md_update_sb(mddev, 1);
7150 /* if array is no-longer degraded, then any saved_raid_disk
7151 * information must be scrapped
7153 if (!mddev->degraded)
7154 list_for_each_entry(rdev, &mddev->disks, same_set)
7155 rdev->saved_raid_disk = -1;
7157 mddev->recovery = 0;
7158 /* flag recovery needed just to double check */
7159 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7160 sysfs_notify_dirent_safe(mddev->sysfs_action);
7161 md_new_event(mddev);
7164 /* Set RUNNING before clearing NEEDED to avoid
7165 * any transients in the value of "sync_action".
7167 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
7168 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7169 /* Clear some bits that don't mean anything, but
7172 clear_bit(MD_RECOVERY_INTR, &mddev->recovery);
7173 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
7175 if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
7177 /* no recovery is running.
7178 * remove any failed drives, then
7179 * add spares if possible.
7180 * Spare are also removed and re-added, to allow
7181 * the personality to fail the re-add.
7184 if (mddev->reshape_position != MaxSector) {
7185 if (mddev->pers->check_reshape == NULL ||
7186 mddev->pers->check_reshape(mddev) != 0)
7187 /* Cannot proceed */
7189 set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
7190 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
7191 } else if ((spares = remove_and_add_spares(mddev))) {
7192 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
7193 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
7194 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
7195 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
7196 } else if (mddev->recovery_cp < MaxSector) {
7197 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
7198 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
7199 } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
7200 /* nothing to be done ... */
7203 if (mddev->pers->sync_request) {
7204 if (spares && mddev->bitmap && ! mddev->bitmap->file) {
7205 /* We are adding a device or devices to an array
7206 * which has the bitmap stored on all devices.
7207 * So make sure all bitmap pages get written
7209 bitmap_write_all(mddev->bitmap);
7211 mddev->sync_thread = md_register_thread(md_do_sync,
7214 if (!mddev->sync_thread) {
7215 printk(KERN_ERR "%s: could not start resync"
7218 /* leave the spares where they are, it shouldn't hurt */
7219 mddev->recovery = 0;
7221 md_wakeup_thread(mddev->sync_thread);
7222 sysfs_notify_dirent_safe(mddev->sysfs_action);
7223 md_new_event(mddev);
7226 if (!mddev->sync_thread) {
7227 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
7228 if (test_and_clear_bit(MD_RECOVERY_RECOVER,
7230 if (mddev->sysfs_action)
7231 sysfs_notify_dirent_safe(mddev->sysfs_action);
7233 mddev_unlock(mddev);
7237 void md_wait_for_blocked_rdev(mdk_rdev_t *rdev, mddev_t *mddev)
7239 sysfs_notify_dirent_safe(rdev->sysfs_state);
7240 wait_event_timeout(rdev->blocked_wait,
7241 !test_bit(Blocked, &rdev->flags),
7242 msecs_to_jiffies(5000));
7243 rdev_dec_pending(rdev, mddev);
7245 EXPORT_SYMBOL(md_wait_for_blocked_rdev);
7247 static int md_notify_reboot(struct notifier_block *this,
7248 unsigned long code, void *x)
7250 struct list_head *tmp;
7253 if ((code == SYS_DOWN) || (code == SYS_HALT) || (code == SYS_POWER_OFF)) {
7255 printk(KERN_INFO "md: stopping all md devices.\n");
7257 for_each_mddev(mddev, tmp)
7258 if (mddev_trylock(mddev)) {
7259 /* Force a switch to readonly even array
7260 * appears to still be in use. Hence
7263 md_set_readonly(mddev, 100);
7264 mddev_unlock(mddev);
7267 * certain more exotic SCSI devices are known to be
7268 * volatile wrt too early system reboots. While the
7269 * right place to handle this issue is the given
7270 * driver, we do want to have a safe RAID driver ...
7277 static struct notifier_block md_notifier = {
7278 .notifier_call = md_notify_reboot,
7280 .priority = INT_MAX, /* before any real devices */
7283 static void md_geninit(void)
7285 dprintk("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
7287 proc_create("mdstat", S_IRUGO, NULL, &md_seq_fops);
7290 static int __init md_init(void)
7292 if (register_blkdev(MD_MAJOR, "md"))
7294 if ((mdp_major=register_blkdev(0, "mdp"))<=0) {
7295 unregister_blkdev(MD_MAJOR, "md");
7298 blk_register_region(MKDEV(MD_MAJOR, 0), 1UL<<MINORBITS, THIS_MODULE,
7299 md_probe, NULL, NULL);
7300 blk_register_region(MKDEV(mdp_major, 0), 1UL<<MINORBITS, THIS_MODULE,
7301 md_probe, NULL, NULL);
7303 register_reboot_notifier(&md_notifier);
7304 raid_table_header = register_sysctl_table(raid_root_table);
7314 * Searches all registered partitions for autorun RAID arrays
7318 static LIST_HEAD(all_detected_devices);
7319 struct detected_devices_node {
7320 struct list_head list;
7324 void md_autodetect_dev(dev_t dev)
7326 struct detected_devices_node *node_detected_dev;
7328 node_detected_dev = kzalloc(sizeof(*node_detected_dev), GFP_KERNEL);
7329 if (node_detected_dev) {
7330 node_detected_dev->dev = dev;
7331 list_add_tail(&node_detected_dev->list, &all_detected_devices);
7333 printk(KERN_CRIT "md: md_autodetect_dev: kzalloc failed"
7334 ", skipping dev(%d,%d)\n", MAJOR(dev), MINOR(dev));
7339 static void autostart_arrays(int part)
7342 struct detected_devices_node *node_detected_dev;
7344 int i_scanned, i_passed;
7349 printk(KERN_INFO "md: Autodetecting RAID arrays.\n");
7351 while (!list_empty(&all_detected_devices) && i_scanned < INT_MAX) {
7353 node_detected_dev = list_entry(all_detected_devices.next,
7354 struct detected_devices_node, list);
7355 list_del(&node_detected_dev->list);
7356 dev = node_detected_dev->dev;
7357 kfree(node_detected_dev);
7358 rdev = md_import_device(dev,0, 90);
7362 if (test_bit(Faulty, &rdev->flags)) {
7366 set_bit(AutoDetected, &rdev->flags);
7367 list_add(&rdev->same_set, &pending_raid_disks);
7371 printk(KERN_INFO "md: Scanned %d and added %d devices.\n",
7372 i_scanned, i_passed);
7374 autorun_devices(part);
7377 #endif /* !MODULE */
7379 static __exit void md_exit(void)
7382 struct list_head *tmp;
7384 blk_unregister_region(MKDEV(MD_MAJOR,0), 1U << MINORBITS);
7385 blk_unregister_region(MKDEV(mdp_major,0), 1U << MINORBITS);
7387 unregister_blkdev(MD_MAJOR,"md");
7388 unregister_blkdev(mdp_major, "mdp");
7389 unregister_reboot_notifier(&md_notifier);
7390 unregister_sysctl_table(raid_table_header);
7391 remove_proc_entry("mdstat", NULL);
7392 for_each_mddev(mddev, tmp) {
7393 export_array(mddev);
7394 mddev->hold_active = 0;
7398 subsys_initcall(md_init);
7399 module_exit(md_exit)
7401 static int get_ro(char *buffer, struct kernel_param *kp)
7403 return sprintf(buffer, "%d", start_readonly);
7405 static int set_ro(const char *val, struct kernel_param *kp)
7408 int num = simple_strtoul(val, &e, 10);
7409 if (*val && (*e == '\0' || *e == '\n')) {
7410 start_readonly = num;
7416 module_param_call(start_ro, set_ro, get_ro, NULL, S_IRUSR|S_IWUSR);
7417 module_param(start_dirty_degraded, int, S_IRUGO|S_IWUSR);
7419 module_param_call(new_array, add_named_array, NULL, NULL, S_IWUSR);
7421 EXPORT_SYMBOL(register_md_personality);
7422 EXPORT_SYMBOL(unregister_md_personality);
7423 EXPORT_SYMBOL(md_error);
7424 EXPORT_SYMBOL(md_done_sync);
7425 EXPORT_SYMBOL(md_write_start);
7426 EXPORT_SYMBOL(md_write_end);
7427 EXPORT_SYMBOL(md_register_thread);
7428 EXPORT_SYMBOL(md_unregister_thread);
7429 EXPORT_SYMBOL(md_wakeup_thread);
7430 EXPORT_SYMBOL(md_check_recovery);
7431 MODULE_LICENSE("GPL");
7432 MODULE_DESCRIPTION("MD RAID framework");
7434 MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR);