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)
2139 list_for_each_entry(rdev, &mddev->disks, same_set) {
2140 if (rdev->sb_events == mddev->events ||
2142 rdev->raid_disk < 0 &&
2143 rdev->sb_events+1 == mddev->events)) {
2144 /* Don't update this superblock */
2145 rdev->sb_loaded = 2;
2147 super_types[mddev->major_version].
2148 sync_super(mddev, rdev);
2149 rdev->sb_loaded = 1;
2154 static void md_update_sb(mddev_t * mddev, int force_change)
2161 /* First make sure individual recovery_offsets are correct */
2162 list_for_each_entry(rdev, &mddev->disks, same_set) {
2163 if (rdev->raid_disk >= 0 &&
2164 mddev->delta_disks >= 0 &&
2165 !test_bit(In_sync, &rdev->flags) &&
2166 mddev->curr_resync_completed > rdev->recovery_offset)
2167 rdev->recovery_offset = mddev->curr_resync_completed;
2170 if (!mddev->persistent) {
2171 clear_bit(MD_CHANGE_CLEAN, &mddev->flags);
2172 clear_bit(MD_CHANGE_DEVS, &mddev->flags);
2173 wake_up(&mddev->sb_wait);
2177 spin_lock_irq(&mddev->write_lock);
2179 mddev->utime = get_seconds();
2181 if (test_and_clear_bit(MD_CHANGE_DEVS, &mddev->flags))
2183 if (test_and_clear_bit(MD_CHANGE_CLEAN, &mddev->flags))
2184 /* just a clean<-> dirty transition, possibly leave spares alone,
2185 * though if events isn't the right even/odd, we will have to do
2191 if (mddev->degraded)
2192 /* If the array is degraded, then skipping spares is both
2193 * dangerous and fairly pointless.
2194 * Dangerous because a device that was removed from the array
2195 * might have a event_count that still looks up-to-date,
2196 * so it can be re-added without a resync.
2197 * Pointless because if there are any spares to skip,
2198 * then a recovery will happen and soon that array won't
2199 * be degraded any more and the spare can go back to sleep then.
2203 sync_req = mddev->in_sync;
2205 /* If this is just a dirty<->clean transition, and the array is clean
2206 * and 'events' is odd, we can roll back to the previous clean state */
2208 && (mddev->in_sync && mddev->recovery_cp == MaxSector)
2209 && mddev->can_decrease_events
2210 && mddev->events != 1) {
2212 mddev->can_decrease_events = 0;
2214 /* otherwise we have to go forward and ... */
2216 mddev->can_decrease_events = nospares;
2219 if (!mddev->events) {
2221 * oops, this 64-bit counter should never wrap.
2222 * Either we are in around ~1 trillion A.C., assuming
2223 * 1 reboot per second, or we have a bug:
2228 sync_sbs(mddev, nospares);
2229 spin_unlock_irq(&mddev->write_lock);
2232 "md: updating %s RAID superblock on device (in sync %d)\n",
2233 mdname(mddev),mddev->in_sync);
2235 bitmap_update_sb(mddev->bitmap);
2236 list_for_each_entry(rdev, &mddev->disks, same_set) {
2237 char b[BDEVNAME_SIZE];
2238 dprintk(KERN_INFO "md: ");
2239 if (rdev->sb_loaded != 1)
2240 continue; /* no noise on spare devices */
2241 if (test_bit(Faulty, &rdev->flags))
2242 dprintk("(skipping faulty ");
2244 dprintk("%s ", bdevname(rdev->bdev,b));
2245 if (!test_bit(Faulty, &rdev->flags)) {
2246 md_super_write(mddev,rdev,
2247 rdev->sb_start, rdev->sb_size,
2249 dprintk(KERN_INFO "(write) %s's sb offset: %llu\n",
2250 bdevname(rdev->bdev,b),
2251 (unsigned long long)rdev->sb_start);
2252 rdev->sb_events = mddev->events;
2256 if (mddev->level == LEVEL_MULTIPATH)
2257 /* only need to write one superblock... */
2260 md_super_wait(mddev);
2261 /* if there was a failure, MD_CHANGE_DEVS was set, and we re-write super */
2263 spin_lock_irq(&mddev->write_lock);
2264 if (mddev->in_sync != sync_req ||
2265 test_bit(MD_CHANGE_DEVS, &mddev->flags)) {
2266 /* have to write it out again */
2267 spin_unlock_irq(&mddev->write_lock);
2270 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
2271 spin_unlock_irq(&mddev->write_lock);
2272 wake_up(&mddev->sb_wait);
2273 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
2274 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
2278 /* words written to sysfs files may, or may not, be \n terminated.
2279 * We want to accept with case. For this we use cmd_match.
2281 static int cmd_match(const char *cmd, const char *str)
2283 /* See if cmd, written into a sysfs file, matches
2284 * str. They must either be the same, or cmd can
2285 * have a trailing newline
2287 while (*cmd && *str && *cmd == *str) {
2298 struct rdev_sysfs_entry {
2299 struct attribute attr;
2300 ssize_t (*show)(mdk_rdev_t *, char *);
2301 ssize_t (*store)(mdk_rdev_t *, const char *, size_t);
2305 state_show(mdk_rdev_t *rdev, char *page)
2310 if (test_bit(Faulty, &rdev->flags)) {
2311 len+= sprintf(page+len, "%sfaulty",sep);
2314 if (test_bit(In_sync, &rdev->flags)) {
2315 len += sprintf(page+len, "%sin_sync",sep);
2318 if (test_bit(WriteMostly, &rdev->flags)) {
2319 len += sprintf(page+len, "%swrite_mostly",sep);
2322 if (test_bit(Blocked, &rdev->flags)) {
2323 len += sprintf(page+len, "%sblocked", sep);
2326 if (!test_bit(Faulty, &rdev->flags) &&
2327 !test_bit(In_sync, &rdev->flags)) {
2328 len += sprintf(page+len, "%sspare", sep);
2331 return len+sprintf(page+len, "\n");
2335 state_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2338 * faulty - simulates and error
2339 * remove - disconnects the device
2340 * writemostly - sets write_mostly
2341 * -writemostly - clears write_mostly
2342 * blocked - sets the Blocked flag
2343 * -blocked - clears the Blocked flag
2344 * insync - sets Insync providing device isn't active
2347 if (cmd_match(buf, "faulty") && rdev->mddev->pers) {
2348 md_error(rdev->mddev, rdev);
2350 } else if (cmd_match(buf, "remove")) {
2351 if (rdev->raid_disk >= 0)
2354 mddev_t *mddev = rdev->mddev;
2355 kick_rdev_from_array(rdev);
2357 md_update_sb(mddev, 1);
2358 md_new_event(mddev);
2361 } else if (cmd_match(buf, "writemostly")) {
2362 set_bit(WriteMostly, &rdev->flags);
2364 } else if (cmd_match(buf, "-writemostly")) {
2365 clear_bit(WriteMostly, &rdev->flags);
2367 } else if (cmd_match(buf, "blocked")) {
2368 set_bit(Blocked, &rdev->flags);
2370 } else if (cmd_match(buf, "-blocked")) {
2371 clear_bit(Blocked, &rdev->flags);
2372 wake_up(&rdev->blocked_wait);
2373 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2374 md_wakeup_thread(rdev->mddev->thread);
2377 } else if (cmd_match(buf, "insync") && rdev->raid_disk == -1) {
2378 set_bit(In_sync, &rdev->flags);
2382 sysfs_notify_dirent_safe(rdev->sysfs_state);
2383 return err ? err : len;
2385 static struct rdev_sysfs_entry rdev_state =
2386 __ATTR(state, S_IRUGO|S_IWUSR, state_show, state_store);
2389 errors_show(mdk_rdev_t *rdev, char *page)
2391 return sprintf(page, "%d\n", atomic_read(&rdev->corrected_errors));
2395 errors_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2398 unsigned long n = simple_strtoul(buf, &e, 10);
2399 if (*buf && (*e == 0 || *e == '\n')) {
2400 atomic_set(&rdev->corrected_errors, n);
2405 static struct rdev_sysfs_entry rdev_errors =
2406 __ATTR(errors, S_IRUGO|S_IWUSR, errors_show, errors_store);
2409 slot_show(mdk_rdev_t *rdev, char *page)
2411 if (rdev->raid_disk < 0)
2412 return sprintf(page, "none\n");
2414 return sprintf(page, "%d\n", rdev->raid_disk);
2418 slot_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2423 int slot = simple_strtoul(buf, &e, 10);
2424 if (strncmp(buf, "none", 4)==0)
2426 else if (e==buf || (*e && *e!= '\n'))
2428 if (rdev->mddev->pers && slot == -1) {
2429 /* Setting 'slot' on an active array requires also
2430 * updating the 'rd%d' link, and communicating
2431 * with the personality with ->hot_*_disk.
2432 * For now we only support removing
2433 * failed/spare devices. This normally happens automatically,
2434 * but not when the metadata is externally managed.
2436 if (rdev->raid_disk == -1)
2438 /* personality does all needed checks */
2439 if (rdev->mddev->pers->hot_add_disk == NULL)
2441 err = rdev->mddev->pers->
2442 hot_remove_disk(rdev->mddev, rdev->raid_disk);
2445 sprintf(nm, "rd%d", rdev->raid_disk);
2446 sysfs_remove_link(&rdev->mddev->kobj, nm);
2447 rdev->raid_disk = -1;
2448 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2449 md_wakeup_thread(rdev->mddev->thread);
2450 } else if (rdev->mddev->pers) {
2452 /* Activating a spare .. or possibly reactivating
2453 * if we ever get bitmaps working here.
2456 if (rdev->raid_disk != -1)
2459 if (rdev->mddev->pers->hot_add_disk == NULL)
2462 list_for_each_entry(rdev2, &rdev->mddev->disks, same_set)
2463 if (rdev2->raid_disk == slot)
2466 rdev->raid_disk = slot;
2467 if (test_bit(In_sync, &rdev->flags))
2468 rdev->saved_raid_disk = slot;
2470 rdev->saved_raid_disk = -1;
2471 err = rdev->mddev->pers->
2472 hot_add_disk(rdev->mddev, rdev);
2474 rdev->raid_disk = -1;
2477 sysfs_notify_dirent_safe(rdev->sysfs_state);
2478 sprintf(nm, "rd%d", rdev->raid_disk);
2479 if (sysfs_create_link(&rdev->mddev->kobj, &rdev->kobj, nm))
2480 /* failure here is OK */;
2481 /* don't wakeup anyone, leave that to userspace. */
2483 if (slot >= rdev->mddev->raid_disks)
2485 rdev->raid_disk = slot;
2486 /* assume it is working */
2487 clear_bit(Faulty, &rdev->flags);
2488 clear_bit(WriteMostly, &rdev->flags);
2489 set_bit(In_sync, &rdev->flags);
2490 sysfs_notify_dirent_safe(rdev->sysfs_state);
2496 static struct rdev_sysfs_entry rdev_slot =
2497 __ATTR(slot, S_IRUGO|S_IWUSR, slot_show, slot_store);
2500 offset_show(mdk_rdev_t *rdev, char *page)
2502 return sprintf(page, "%llu\n", (unsigned long long)rdev->data_offset);
2506 offset_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2509 unsigned long long offset = simple_strtoull(buf, &e, 10);
2510 if (e==buf || (*e && *e != '\n'))
2512 if (rdev->mddev->pers && rdev->raid_disk >= 0)
2514 if (rdev->sectors && rdev->mddev->external)
2515 /* Must set offset before size, so overlap checks
2518 rdev->data_offset = offset;
2522 static struct rdev_sysfs_entry rdev_offset =
2523 __ATTR(offset, S_IRUGO|S_IWUSR, offset_show, offset_store);
2526 rdev_size_show(mdk_rdev_t *rdev, char *page)
2528 return sprintf(page, "%llu\n", (unsigned long long)rdev->sectors / 2);
2531 static int overlaps(sector_t s1, sector_t l1, sector_t s2, sector_t l2)
2533 /* check if two start/length pairs overlap */
2541 static int strict_blocks_to_sectors(const char *buf, sector_t *sectors)
2543 unsigned long long blocks;
2546 if (strict_strtoull(buf, 10, &blocks) < 0)
2549 if (blocks & 1ULL << (8 * sizeof(blocks) - 1))
2550 return -EINVAL; /* sector conversion overflow */
2553 if (new != blocks * 2)
2554 return -EINVAL; /* unsigned long long to sector_t overflow */
2561 rdev_size_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2563 mddev_t *my_mddev = rdev->mddev;
2564 sector_t oldsectors = rdev->sectors;
2567 if (strict_blocks_to_sectors(buf, §ors) < 0)
2569 if (my_mddev->pers && rdev->raid_disk >= 0) {
2570 if (my_mddev->persistent) {
2571 sectors = super_types[my_mddev->major_version].
2572 rdev_size_change(rdev, sectors);
2575 } else if (!sectors)
2576 sectors = (rdev->bdev->bd_inode->i_size >> 9) -
2579 if (sectors < my_mddev->dev_sectors)
2580 return -EINVAL; /* component must fit device */
2582 rdev->sectors = sectors;
2583 if (sectors > oldsectors && my_mddev->external) {
2584 /* need to check that all other rdevs with the same ->bdev
2585 * do not overlap. We need to unlock the mddev to avoid
2586 * a deadlock. We have already changed rdev->sectors, and if
2587 * we have to change it back, we will have the lock again.
2591 struct list_head *tmp;
2593 mddev_unlock(my_mddev);
2594 for_each_mddev(mddev, tmp) {
2598 list_for_each_entry(rdev2, &mddev->disks, same_set)
2599 if (test_bit(AllReserved, &rdev2->flags) ||
2600 (rdev->bdev == rdev2->bdev &&
2602 overlaps(rdev->data_offset, rdev->sectors,
2608 mddev_unlock(mddev);
2614 mddev_lock(my_mddev);
2616 /* Someone else could have slipped in a size
2617 * change here, but doing so is just silly.
2618 * We put oldsectors back because we *know* it is
2619 * safe, and trust userspace not to race with
2622 rdev->sectors = oldsectors;
2629 static struct rdev_sysfs_entry rdev_size =
2630 __ATTR(size, S_IRUGO|S_IWUSR, rdev_size_show, rdev_size_store);
2633 static ssize_t recovery_start_show(mdk_rdev_t *rdev, char *page)
2635 unsigned long long recovery_start = rdev->recovery_offset;
2637 if (test_bit(In_sync, &rdev->flags) ||
2638 recovery_start == MaxSector)
2639 return sprintf(page, "none\n");
2641 return sprintf(page, "%llu\n", recovery_start);
2644 static ssize_t recovery_start_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2646 unsigned long long recovery_start;
2648 if (cmd_match(buf, "none"))
2649 recovery_start = MaxSector;
2650 else if (strict_strtoull(buf, 10, &recovery_start))
2653 if (rdev->mddev->pers &&
2654 rdev->raid_disk >= 0)
2657 rdev->recovery_offset = recovery_start;
2658 if (recovery_start == MaxSector)
2659 set_bit(In_sync, &rdev->flags);
2661 clear_bit(In_sync, &rdev->flags);
2665 static struct rdev_sysfs_entry rdev_recovery_start =
2666 __ATTR(recovery_start, S_IRUGO|S_IWUSR, recovery_start_show, recovery_start_store);
2668 static struct attribute *rdev_default_attrs[] = {
2674 &rdev_recovery_start.attr,
2678 rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
2680 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
2681 mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
2682 mddev_t *mddev = rdev->mddev;
2688 rv = mddev ? mddev_lock(mddev) : -EBUSY;
2690 if (rdev->mddev == NULL)
2693 rv = entry->show(rdev, page);
2694 mddev_unlock(mddev);
2700 rdev_attr_store(struct kobject *kobj, struct attribute *attr,
2701 const char *page, size_t length)
2703 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
2704 mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
2706 mddev_t *mddev = rdev->mddev;
2710 if (!capable(CAP_SYS_ADMIN))
2712 rv = mddev ? mddev_lock(mddev): -EBUSY;
2714 if (rdev->mddev == NULL)
2717 rv = entry->store(rdev, page, length);
2718 mddev_unlock(mddev);
2723 static void rdev_free(struct kobject *ko)
2725 mdk_rdev_t *rdev = container_of(ko, mdk_rdev_t, kobj);
2728 static const struct sysfs_ops rdev_sysfs_ops = {
2729 .show = rdev_attr_show,
2730 .store = rdev_attr_store,
2732 static struct kobj_type rdev_ktype = {
2733 .release = rdev_free,
2734 .sysfs_ops = &rdev_sysfs_ops,
2735 .default_attrs = rdev_default_attrs,
2738 void md_rdev_init(mdk_rdev_t *rdev)
2741 rdev->saved_raid_disk = -1;
2742 rdev->raid_disk = -1;
2744 rdev->data_offset = 0;
2745 rdev->sb_events = 0;
2746 rdev->last_read_error.tv_sec = 0;
2747 rdev->last_read_error.tv_nsec = 0;
2748 atomic_set(&rdev->nr_pending, 0);
2749 atomic_set(&rdev->read_errors, 0);
2750 atomic_set(&rdev->corrected_errors, 0);
2752 INIT_LIST_HEAD(&rdev->same_set);
2753 init_waitqueue_head(&rdev->blocked_wait);
2755 EXPORT_SYMBOL_GPL(md_rdev_init);
2757 * Import a device. If 'super_format' >= 0, then sanity check the superblock
2759 * mark the device faulty if:
2761 * - the device is nonexistent (zero size)
2762 * - the device has no valid superblock
2764 * a faulty rdev _never_ has rdev->sb set.
2766 static mdk_rdev_t *md_import_device(dev_t newdev, int super_format, int super_minor)
2768 char b[BDEVNAME_SIZE];
2773 rdev = kzalloc(sizeof(*rdev), GFP_KERNEL);
2775 printk(KERN_ERR "md: could not alloc mem for new device!\n");
2776 return ERR_PTR(-ENOMEM);
2780 if ((err = alloc_disk_sb(rdev)))
2783 err = lock_rdev(rdev, newdev, super_format == -2);
2787 kobject_init(&rdev->kobj, &rdev_ktype);
2789 size = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
2792 "md: %s has zero or unknown size, marking faulty!\n",
2793 bdevname(rdev->bdev,b));
2798 if (super_format >= 0) {
2799 err = super_types[super_format].
2800 load_super(rdev, NULL, super_minor);
2801 if (err == -EINVAL) {
2803 "md: %s does not have a valid v%d.%d "
2804 "superblock, not importing!\n",
2805 bdevname(rdev->bdev,b),
2806 super_format, super_minor);
2811 "md: could not read %s's sb, not importing!\n",
2812 bdevname(rdev->bdev,b));
2820 if (rdev->sb_page) {
2826 return ERR_PTR(err);
2830 * Check a full RAID array for plausibility
2834 static void analyze_sbs(mddev_t * mddev)
2837 mdk_rdev_t *rdev, *freshest, *tmp;
2838 char b[BDEVNAME_SIZE];
2841 rdev_for_each(rdev, tmp, mddev)
2842 switch (super_types[mddev->major_version].
2843 load_super(rdev, freshest, mddev->minor_version)) {
2851 "md: fatal superblock inconsistency in %s"
2852 " -- removing from array\n",
2853 bdevname(rdev->bdev,b));
2854 kick_rdev_from_array(rdev);
2858 super_types[mddev->major_version].
2859 validate_super(mddev, freshest);
2862 rdev_for_each(rdev, tmp, mddev) {
2863 if (mddev->max_disks &&
2864 (rdev->desc_nr >= mddev->max_disks ||
2865 i > mddev->max_disks)) {
2867 "md: %s: %s: only %d devices permitted\n",
2868 mdname(mddev), bdevname(rdev->bdev, b),
2870 kick_rdev_from_array(rdev);
2873 if (rdev != freshest)
2874 if (super_types[mddev->major_version].
2875 validate_super(mddev, rdev)) {
2876 printk(KERN_WARNING "md: kicking non-fresh %s"
2878 bdevname(rdev->bdev,b));
2879 kick_rdev_from_array(rdev);
2882 if (mddev->level == LEVEL_MULTIPATH) {
2883 rdev->desc_nr = i++;
2884 rdev->raid_disk = rdev->desc_nr;
2885 set_bit(In_sync, &rdev->flags);
2886 } else if (rdev->raid_disk >= (mddev->raid_disks - min(0, mddev->delta_disks))) {
2887 rdev->raid_disk = -1;
2888 clear_bit(In_sync, &rdev->flags);
2893 /* Read a fixed-point number.
2894 * Numbers in sysfs attributes should be in "standard" units where
2895 * possible, so time should be in seconds.
2896 * However we internally use a a much smaller unit such as
2897 * milliseconds or jiffies.
2898 * This function takes a decimal number with a possible fractional
2899 * component, and produces an integer which is the result of
2900 * multiplying that number by 10^'scale'.
2901 * all without any floating-point arithmetic.
2903 int strict_strtoul_scaled(const char *cp, unsigned long *res, int scale)
2905 unsigned long result = 0;
2907 while (isdigit(*cp) || (*cp == '.' && decimals < 0)) {
2910 else if (decimals < scale) {
2913 result = result * 10 + value;
2925 while (decimals < scale) {
2934 static void md_safemode_timeout(unsigned long data);
2937 safe_delay_show(mddev_t *mddev, char *page)
2939 int msec = (mddev->safemode_delay*1000)/HZ;
2940 return sprintf(page, "%d.%03d\n", msec/1000, msec%1000);
2943 safe_delay_store(mddev_t *mddev, const char *cbuf, size_t len)
2947 if (strict_strtoul_scaled(cbuf, &msec, 3) < 0)
2950 mddev->safemode_delay = 0;
2952 unsigned long old_delay = mddev->safemode_delay;
2953 mddev->safemode_delay = (msec*HZ)/1000;
2954 if (mddev->safemode_delay == 0)
2955 mddev->safemode_delay = 1;
2956 if (mddev->safemode_delay < old_delay)
2957 md_safemode_timeout((unsigned long)mddev);
2961 static struct md_sysfs_entry md_safe_delay =
2962 __ATTR(safe_mode_delay, S_IRUGO|S_IWUSR,safe_delay_show, safe_delay_store);
2965 level_show(mddev_t *mddev, char *page)
2967 struct mdk_personality *p = mddev->pers;
2969 return sprintf(page, "%s\n", p->name);
2970 else if (mddev->clevel[0])
2971 return sprintf(page, "%s\n", mddev->clevel);
2972 else if (mddev->level != LEVEL_NONE)
2973 return sprintf(page, "%d\n", mddev->level);
2979 level_store(mddev_t *mddev, const char *buf, size_t len)
2983 struct mdk_personality *pers;
2988 if (mddev->pers == NULL) {
2991 if (len >= sizeof(mddev->clevel))
2993 strncpy(mddev->clevel, buf, len);
2994 if (mddev->clevel[len-1] == '\n')
2996 mddev->clevel[len] = 0;
2997 mddev->level = LEVEL_NONE;
3001 /* request to change the personality. Need to ensure:
3002 * - array is not engaged in resync/recovery/reshape
3003 * - old personality can be suspended
3004 * - new personality will access other array.
3007 if (mddev->sync_thread ||
3008 mddev->reshape_position != MaxSector ||
3009 mddev->sysfs_active)
3012 if (!mddev->pers->quiesce) {
3013 printk(KERN_WARNING "md: %s: %s does not support online personality change\n",
3014 mdname(mddev), mddev->pers->name);
3018 /* Now find the new personality */
3019 if (len == 0 || len >= sizeof(clevel))
3021 strncpy(clevel, buf, len);
3022 if (clevel[len-1] == '\n')
3025 if (strict_strtol(clevel, 10, &level))
3028 if (request_module("md-%s", clevel) != 0)
3029 request_module("md-level-%s", clevel);
3030 spin_lock(&pers_lock);
3031 pers = find_pers(level, clevel);
3032 if (!pers || !try_module_get(pers->owner)) {
3033 spin_unlock(&pers_lock);
3034 printk(KERN_WARNING "md: personality %s not loaded\n", clevel);
3037 spin_unlock(&pers_lock);
3039 if (pers == mddev->pers) {
3040 /* Nothing to do! */
3041 module_put(pers->owner);
3044 if (!pers->takeover) {
3045 module_put(pers->owner);
3046 printk(KERN_WARNING "md: %s: %s does not support personality takeover\n",
3047 mdname(mddev), clevel);
3051 list_for_each_entry(rdev, &mddev->disks, same_set)
3052 rdev->new_raid_disk = rdev->raid_disk;
3054 /* ->takeover must set new_* and/or delta_disks
3055 * if it succeeds, and may set them when it fails.
3057 priv = pers->takeover(mddev);
3059 mddev->new_level = mddev->level;
3060 mddev->new_layout = mddev->layout;
3061 mddev->new_chunk_sectors = mddev->chunk_sectors;
3062 mddev->raid_disks -= mddev->delta_disks;
3063 mddev->delta_disks = 0;
3064 module_put(pers->owner);
3065 printk(KERN_WARNING "md: %s: %s would not accept array\n",
3066 mdname(mddev), clevel);
3067 return PTR_ERR(priv);
3070 /* Looks like we have a winner */
3071 mddev_suspend(mddev);
3072 mddev->pers->stop(mddev);
3074 if (mddev->pers->sync_request == NULL &&
3075 pers->sync_request != NULL) {
3076 /* need to add the md_redundancy_group */
3077 if (sysfs_create_group(&mddev->kobj, &md_redundancy_group))
3079 "md: cannot register extra attributes for %s\n",
3081 mddev->sysfs_action = sysfs_get_dirent(mddev->kobj.sd, NULL, "sync_action");
3083 if (mddev->pers->sync_request != NULL &&
3084 pers->sync_request == NULL) {
3085 /* need to remove the md_redundancy_group */
3086 if (mddev->to_remove == NULL)
3087 mddev->to_remove = &md_redundancy_group;
3090 if (mddev->pers->sync_request == NULL &&
3092 /* We are converting from a no-redundancy array
3093 * to a redundancy array and metadata is managed
3094 * externally so we need to be sure that writes
3095 * won't block due to a need to transition
3097 * until external management is started.
3100 mddev->safemode_delay = 0;
3101 mddev->safemode = 0;
3104 list_for_each_entry(rdev, &mddev->disks, same_set) {
3106 if (rdev->raid_disk < 0)
3108 if (rdev->new_raid_disk > mddev->raid_disks)
3109 rdev->new_raid_disk = -1;
3110 if (rdev->new_raid_disk == rdev->raid_disk)
3112 sprintf(nm, "rd%d", rdev->raid_disk);
3113 sysfs_remove_link(&mddev->kobj, nm);
3115 list_for_each_entry(rdev, &mddev->disks, same_set) {
3116 if (rdev->raid_disk < 0)
3118 if (rdev->new_raid_disk == rdev->raid_disk)
3120 rdev->raid_disk = rdev->new_raid_disk;
3121 if (rdev->raid_disk < 0)
3122 clear_bit(In_sync, &rdev->flags);
3125 sprintf(nm, "rd%d", rdev->raid_disk);
3126 if(sysfs_create_link(&mddev->kobj, &rdev->kobj, nm))
3127 printk("md: cannot register %s for %s after level change\n",
3132 module_put(mddev->pers->owner);
3134 mddev->private = priv;
3135 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
3136 mddev->level = mddev->new_level;
3137 mddev->layout = mddev->new_layout;
3138 mddev->chunk_sectors = mddev->new_chunk_sectors;
3139 mddev->delta_disks = 0;
3140 if (mddev->pers->sync_request == NULL) {
3141 /* this is now an array without redundancy, so
3142 * it must always be in_sync
3145 del_timer_sync(&mddev->safemode_timer);
3148 mddev_resume(mddev);
3149 set_bit(MD_CHANGE_DEVS, &mddev->flags);
3150 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3151 md_wakeup_thread(mddev->thread);
3152 sysfs_notify(&mddev->kobj, NULL, "level");
3153 md_new_event(mddev);
3157 static struct md_sysfs_entry md_level =
3158 __ATTR(level, S_IRUGO|S_IWUSR, level_show, level_store);
3162 layout_show(mddev_t *mddev, char *page)
3164 /* just a number, not meaningful for all levels */
3165 if (mddev->reshape_position != MaxSector &&
3166 mddev->layout != mddev->new_layout)
3167 return sprintf(page, "%d (%d)\n",
3168 mddev->new_layout, mddev->layout);
3169 return sprintf(page, "%d\n", mddev->layout);
3173 layout_store(mddev_t *mddev, const char *buf, size_t len)
3176 unsigned long n = simple_strtoul(buf, &e, 10);
3178 if (!*buf || (*e && *e != '\n'))
3183 if (mddev->pers->check_reshape == NULL)
3185 mddev->new_layout = n;
3186 err = mddev->pers->check_reshape(mddev);
3188 mddev->new_layout = mddev->layout;
3192 mddev->new_layout = n;
3193 if (mddev->reshape_position == MaxSector)
3198 static struct md_sysfs_entry md_layout =
3199 __ATTR(layout, S_IRUGO|S_IWUSR, layout_show, layout_store);
3203 raid_disks_show(mddev_t *mddev, char *page)
3205 if (mddev->raid_disks == 0)
3207 if (mddev->reshape_position != MaxSector &&
3208 mddev->delta_disks != 0)
3209 return sprintf(page, "%d (%d)\n", mddev->raid_disks,
3210 mddev->raid_disks - mddev->delta_disks);
3211 return sprintf(page, "%d\n", mddev->raid_disks);
3214 static int update_raid_disks(mddev_t *mddev, int raid_disks);
3217 raid_disks_store(mddev_t *mddev, const char *buf, size_t len)
3221 unsigned long n = simple_strtoul(buf, &e, 10);
3223 if (!*buf || (*e && *e != '\n'))
3227 rv = update_raid_disks(mddev, n);
3228 else if (mddev->reshape_position != MaxSector) {
3229 int olddisks = mddev->raid_disks - mddev->delta_disks;
3230 mddev->delta_disks = n - olddisks;
3231 mddev->raid_disks = n;
3233 mddev->raid_disks = n;
3234 return rv ? rv : len;
3236 static struct md_sysfs_entry md_raid_disks =
3237 __ATTR(raid_disks, S_IRUGO|S_IWUSR, raid_disks_show, raid_disks_store);
3240 chunk_size_show(mddev_t *mddev, char *page)
3242 if (mddev->reshape_position != MaxSector &&
3243 mddev->chunk_sectors != mddev->new_chunk_sectors)
3244 return sprintf(page, "%d (%d)\n",
3245 mddev->new_chunk_sectors << 9,
3246 mddev->chunk_sectors << 9);
3247 return sprintf(page, "%d\n", mddev->chunk_sectors << 9);
3251 chunk_size_store(mddev_t *mddev, const char *buf, size_t len)
3254 unsigned long n = simple_strtoul(buf, &e, 10);
3256 if (!*buf || (*e && *e != '\n'))
3261 if (mddev->pers->check_reshape == NULL)
3263 mddev->new_chunk_sectors = n >> 9;
3264 err = mddev->pers->check_reshape(mddev);
3266 mddev->new_chunk_sectors = mddev->chunk_sectors;
3270 mddev->new_chunk_sectors = n >> 9;
3271 if (mddev->reshape_position == MaxSector)
3272 mddev->chunk_sectors = n >> 9;
3276 static struct md_sysfs_entry md_chunk_size =
3277 __ATTR(chunk_size, S_IRUGO|S_IWUSR, chunk_size_show, chunk_size_store);
3280 resync_start_show(mddev_t *mddev, char *page)
3282 if (mddev->recovery_cp == MaxSector)
3283 return sprintf(page, "none\n");
3284 return sprintf(page, "%llu\n", (unsigned long long)mddev->recovery_cp);
3288 resync_start_store(mddev_t *mddev, const char *buf, size_t len)
3291 unsigned long long n = simple_strtoull(buf, &e, 10);
3295 if (cmd_match(buf, "none"))
3297 else if (!*buf || (*e && *e != '\n'))
3300 mddev->recovery_cp = n;
3303 static struct md_sysfs_entry md_resync_start =
3304 __ATTR(resync_start, S_IRUGO|S_IWUSR, resync_start_show, resync_start_store);
3307 * The array state can be:
3310 * No devices, no size, no level
3311 * Equivalent to STOP_ARRAY ioctl
3313 * May have some settings, but array is not active
3314 * all IO results in error
3315 * When written, doesn't tear down array, but just stops it
3316 * suspended (not supported yet)
3317 * All IO requests will block. The array can be reconfigured.
3318 * Writing this, if accepted, will block until array is quiescent
3320 * no resync can happen. no superblocks get written.
3321 * write requests fail
3323 * like readonly, but behaves like 'clean' on a write request.
3325 * clean - no pending writes, but otherwise active.
3326 * When written to inactive array, starts without resync
3327 * If a write request arrives then
3328 * if metadata is known, mark 'dirty' and switch to 'active'.
3329 * if not known, block and switch to write-pending
3330 * If written to an active array that has pending writes, then fails.
3332 * fully active: IO and resync can be happening.
3333 * When written to inactive array, starts with resync
3336 * clean, but writes are blocked waiting for 'active' to be written.
3339 * like active, but no writes have been seen for a while (100msec).
3342 enum array_state { clear, inactive, suspended, readonly, read_auto, clean, active,
3343 write_pending, active_idle, bad_word};
3344 static char *array_states[] = {
3345 "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active",
3346 "write-pending", "active-idle", NULL };
3348 static int match_word(const char *word, char **list)
3351 for (n=0; list[n]; n++)
3352 if (cmd_match(word, list[n]))
3358 array_state_show(mddev_t *mddev, char *page)
3360 enum array_state st = inactive;
3373 else if (test_bit(MD_CHANGE_PENDING, &mddev->flags))
3375 else if (mddev->safemode)
3381 if (list_empty(&mddev->disks) &&
3382 mddev->raid_disks == 0 &&
3383 mddev->dev_sectors == 0)
3388 return sprintf(page, "%s\n", array_states[st]);
3391 static int do_md_stop(mddev_t * mddev, int ro, int is_open);
3392 static int md_set_readonly(mddev_t * mddev, int is_open);
3393 static int do_md_run(mddev_t * mddev);
3394 static int restart_array(mddev_t *mddev);
3397 array_state_store(mddev_t *mddev, const char *buf, size_t len)
3400 enum array_state st = match_word(buf, array_states);
3405 /* stopping an active array */
3406 if (atomic_read(&mddev->openers) > 0)
3408 err = do_md_stop(mddev, 0, 0);
3411 /* stopping an active array */
3413 if (atomic_read(&mddev->openers) > 0)
3415 err = do_md_stop(mddev, 2, 0);
3417 err = 0; /* already inactive */
3420 break; /* not supported yet */
3423 err = md_set_readonly(mddev, 0);
3426 set_disk_ro(mddev->gendisk, 1);
3427 err = do_md_run(mddev);
3433 err = md_set_readonly(mddev, 0);
3434 else if (mddev->ro == 1)
3435 err = restart_array(mddev);
3438 set_disk_ro(mddev->gendisk, 0);
3442 err = do_md_run(mddev);
3447 restart_array(mddev);
3448 spin_lock_irq(&mddev->write_lock);
3449 if (atomic_read(&mddev->writes_pending) == 0) {
3450 if (mddev->in_sync == 0) {
3452 if (mddev->safemode == 1)
3453 mddev->safemode = 0;
3454 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
3459 spin_unlock_irq(&mddev->write_lock);
3465 restart_array(mddev);
3466 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
3467 wake_up(&mddev->sb_wait);
3471 set_disk_ro(mddev->gendisk, 0);
3472 err = do_md_run(mddev);
3477 /* these cannot be set */
3483 sysfs_notify_dirent_safe(mddev->sysfs_state);
3487 static struct md_sysfs_entry md_array_state =
3488 __ATTR(array_state, S_IRUGO|S_IWUSR, array_state_show, array_state_store);
3491 max_corrected_read_errors_show(mddev_t *mddev, char *page) {
3492 return sprintf(page, "%d\n",
3493 atomic_read(&mddev->max_corr_read_errors));
3497 max_corrected_read_errors_store(mddev_t *mddev, const char *buf, size_t len)
3500 unsigned long n = simple_strtoul(buf, &e, 10);
3502 if (*buf && (*e == 0 || *e == '\n')) {
3503 atomic_set(&mddev->max_corr_read_errors, n);
3509 static struct md_sysfs_entry max_corr_read_errors =
3510 __ATTR(max_read_errors, S_IRUGO|S_IWUSR, max_corrected_read_errors_show,
3511 max_corrected_read_errors_store);
3514 null_show(mddev_t *mddev, char *page)
3520 new_dev_store(mddev_t *mddev, const char *buf, size_t len)
3522 /* buf must be %d:%d\n? giving major and minor numbers */
3523 /* The new device is added to the array.
3524 * If the array has a persistent superblock, we read the
3525 * superblock to initialise info and check validity.
3526 * Otherwise, only checking done is that in bind_rdev_to_array,
3527 * which mainly checks size.
3530 int major = simple_strtoul(buf, &e, 10);
3536 if (!*buf || *e != ':' || !e[1] || e[1] == '\n')
3538 minor = simple_strtoul(e+1, &e, 10);
3539 if (*e && *e != '\n')
3541 dev = MKDEV(major, minor);
3542 if (major != MAJOR(dev) ||
3543 minor != MINOR(dev))
3547 if (mddev->persistent) {
3548 rdev = md_import_device(dev, mddev->major_version,
3549 mddev->minor_version);
3550 if (!IS_ERR(rdev) && !list_empty(&mddev->disks)) {
3551 mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
3552 mdk_rdev_t, same_set);
3553 err = super_types[mddev->major_version]
3554 .load_super(rdev, rdev0, mddev->minor_version);
3558 } else if (mddev->external)
3559 rdev = md_import_device(dev, -2, -1);
3561 rdev = md_import_device(dev, -1, -1);
3564 return PTR_ERR(rdev);
3565 err = bind_rdev_to_array(rdev, mddev);
3569 return err ? err : len;
3572 static struct md_sysfs_entry md_new_device =
3573 __ATTR(new_dev, S_IWUSR, null_show, new_dev_store);
3576 bitmap_store(mddev_t *mddev, const char *buf, size_t len)
3579 unsigned long chunk, end_chunk;
3583 /* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */
3585 chunk = end_chunk = simple_strtoul(buf, &end, 0);
3586 if (buf == end) break;
3587 if (*end == '-') { /* range */
3589 end_chunk = simple_strtoul(buf, &end, 0);
3590 if (buf == end) break;
3592 if (*end && !isspace(*end)) break;
3593 bitmap_dirty_bits(mddev->bitmap, chunk, end_chunk);
3594 buf = skip_spaces(end);
3596 bitmap_unplug(mddev->bitmap); /* flush the bits to disk */
3601 static struct md_sysfs_entry md_bitmap =
3602 __ATTR(bitmap_set_bits, S_IWUSR, null_show, bitmap_store);
3605 size_show(mddev_t *mddev, char *page)
3607 return sprintf(page, "%llu\n",
3608 (unsigned long long)mddev->dev_sectors / 2);
3611 static int update_size(mddev_t *mddev, sector_t num_sectors);
3614 size_store(mddev_t *mddev, const char *buf, size_t len)
3616 /* If array is inactive, we can reduce the component size, but
3617 * not increase it (except from 0).
3618 * If array is active, we can try an on-line resize
3621 int err = strict_blocks_to_sectors(buf, §ors);
3626 err = update_size(mddev, sectors);
3627 md_update_sb(mddev, 1);
3629 if (mddev->dev_sectors == 0 ||
3630 mddev->dev_sectors > sectors)
3631 mddev->dev_sectors = sectors;
3635 return err ? err : len;
3638 static struct md_sysfs_entry md_size =
3639 __ATTR(component_size, S_IRUGO|S_IWUSR, size_show, size_store);
3644 * 'none' for arrays with no metadata (good luck...)
3645 * 'external' for arrays with externally managed metadata,
3646 * or N.M for internally known formats
3649 metadata_show(mddev_t *mddev, char *page)
3651 if (mddev->persistent)
3652 return sprintf(page, "%d.%d\n",
3653 mddev->major_version, mddev->minor_version);
3654 else if (mddev->external)
3655 return sprintf(page, "external:%s\n", mddev->metadata_type);
3657 return sprintf(page, "none\n");
3661 metadata_store(mddev_t *mddev, const char *buf, size_t len)
3665 /* Changing the details of 'external' metadata is
3666 * always permitted. Otherwise there must be
3667 * no devices attached to the array.
3669 if (mddev->external && strncmp(buf, "external:", 9) == 0)
3671 else if (!list_empty(&mddev->disks))
3674 if (cmd_match(buf, "none")) {
3675 mddev->persistent = 0;
3676 mddev->external = 0;
3677 mddev->major_version = 0;
3678 mddev->minor_version = 90;
3681 if (strncmp(buf, "external:", 9) == 0) {
3682 size_t namelen = len-9;
3683 if (namelen >= sizeof(mddev->metadata_type))
3684 namelen = sizeof(mddev->metadata_type)-1;
3685 strncpy(mddev->metadata_type, buf+9, namelen);
3686 mddev->metadata_type[namelen] = 0;
3687 if (namelen && mddev->metadata_type[namelen-1] == '\n')
3688 mddev->metadata_type[--namelen] = 0;
3689 mddev->persistent = 0;
3690 mddev->external = 1;
3691 mddev->major_version = 0;
3692 mddev->minor_version = 90;
3695 major = simple_strtoul(buf, &e, 10);
3696 if (e==buf || *e != '.')
3699 minor = simple_strtoul(buf, &e, 10);
3700 if (e==buf || (*e && *e != '\n') )
3702 if (major >= ARRAY_SIZE(super_types) || super_types[major].name == NULL)
3704 mddev->major_version = major;
3705 mddev->minor_version = minor;
3706 mddev->persistent = 1;
3707 mddev->external = 0;
3711 static struct md_sysfs_entry md_metadata =
3712 __ATTR(metadata_version, S_IRUGO|S_IWUSR, metadata_show, metadata_store);
3715 action_show(mddev_t *mddev, char *page)
3717 char *type = "idle";
3718 if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
3720 else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
3721 (!mddev->ro && test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))) {
3722 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
3724 else if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
3725 if (!test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
3727 else if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
3731 } else if (test_bit(MD_RECOVERY_RECOVER, &mddev->recovery))
3734 return sprintf(page, "%s\n", type);
3738 action_store(mddev_t *mddev, const char *page, size_t len)
3740 if (!mddev->pers || !mddev->pers->sync_request)
3743 if (cmd_match(page, "frozen"))
3744 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
3746 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
3748 if (cmd_match(page, "idle") || cmd_match(page, "frozen")) {
3749 if (mddev->sync_thread) {
3750 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
3751 md_unregister_thread(mddev->sync_thread);
3752 mddev->sync_thread = NULL;
3753 mddev->recovery = 0;
3755 } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
3756 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
3758 else if (cmd_match(page, "resync"))
3759 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3760 else if (cmd_match(page, "recover")) {
3761 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
3762 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3763 } else if (cmd_match(page, "reshape")) {
3765 if (mddev->pers->start_reshape == NULL)
3767 err = mddev->pers->start_reshape(mddev);
3770 sysfs_notify(&mddev->kobj, NULL, "degraded");
3772 if (cmd_match(page, "check"))
3773 set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
3774 else if (!cmd_match(page, "repair"))
3776 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
3777 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
3779 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3780 md_wakeup_thread(mddev->thread);
3781 sysfs_notify_dirent_safe(mddev->sysfs_action);
3786 mismatch_cnt_show(mddev_t *mddev, char *page)
3788 return sprintf(page, "%llu\n",
3789 (unsigned long long) mddev->resync_mismatches);
3792 static struct md_sysfs_entry md_scan_mode =
3793 __ATTR(sync_action, S_IRUGO|S_IWUSR, action_show, action_store);
3796 static struct md_sysfs_entry md_mismatches = __ATTR_RO(mismatch_cnt);
3799 sync_min_show(mddev_t *mddev, char *page)
3801 return sprintf(page, "%d (%s)\n", speed_min(mddev),
3802 mddev->sync_speed_min ? "local": "system");
3806 sync_min_store(mddev_t *mddev, const char *buf, size_t len)
3810 if (strncmp(buf, "system", 6)==0) {
3811 mddev->sync_speed_min = 0;
3814 min = simple_strtoul(buf, &e, 10);
3815 if (buf == e || (*e && *e != '\n') || min <= 0)
3817 mddev->sync_speed_min = min;
3821 static struct md_sysfs_entry md_sync_min =
3822 __ATTR(sync_speed_min, S_IRUGO|S_IWUSR, sync_min_show, sync_min_store);
3825 sync_max_show(mddev_t *mddev, char *page)
3827 return sprintf(page, "%d (%s)\n", speed_max(mddev),
3828 mddev->sync_speed_max ? "local": "system");
3832 sync_max_store(mddev_t *mddev, const char *buf, size_t len)
3836 if (strncmp(buf, "system", 6)==0) {
3837 mddev->sync_speed_max = 0;
3840 max = simple_strtoul(buf, &e, 10);
3841 if (buf == e || (*e && *e != '\n') || max <= 0)
3843 mddev->sync_speed_max = max;
3847 static struct md_sysfs_entry md_sync_max =
3848 __ATTR(sync_speed_max, S_IRUGO|S_IWUSR, sync_max_show, sync_max_store);
3851 degraded_show(mddev_t *mddev, char *page)
3853 return sprintf(page, "%d\n", mddev->degraded);
3855 static struct md_sysfs_entry md_degraded = __ATTR_RO(degraded);
3858 sync_force_parallel_show(mddev_t *mddev, char *page)
3860 return sprintf(page, "%d\n", mddev->parallel_resync);
3864 sync_force_parallel_store(mddev_t *mddev, const char *buf, size_t len)
3868 if (strict_strtol(buf, 10, &n))
3871 if (n != 0 && n != 1)
3874 mddev->parallel_resync = n;
3876 if (mddev->sync_thread)
3877 wake_up(&resync_wait);
3882 /* force parallel resync, even with shared block devices */
3883 static struct md_sysfs_entry md_sync_force_parallel =
3884 __ATTR(sync_force_parallel, S_IRUGO|S_IWUSR,
3885 sync_force_parallel_show, sync_force_parallel_store);
3888 sync_speed_show(mddev_t *mddev, char *page)
3890 unsigned long resync, dt, db;
3891 if (mddev->curr_resync == 0)
3892 return sprintf(page, "none\n");
3893 resync = mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active);
3894 dt = (jiffies - mddev->resync_mark) / HZ;
3896 db = resync - mddev->resync_mark_cnt;
3897 return sprintf(page, "%lu\n", db/dt/2); /* K/sec */
3900 static struct md_sysfs_entry md_sync_speed = __ATTR_RO(sync_speed);
3903 sync_completed_show(mddev_t *mddev, char *page)
3905 unsigned long max_sectors, resync;
3907 if (!test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
3908 return sprintf(page, "none\n");
3910 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
3911 max_sectors = mddev->resync_max_sectors;
3913 max_sectors = mddev->dev_sectors;
3915 resync = mddev->curr_resync_completed;
3916 return sprintf(page, "%lu / %lu\n", resync, max_sectors);
3919 static struct md_sysfs_entry md_sync_completed = __ATTR_RO(sync_completed);
3922 min_sync_show(mddev_t *mddev, char *page)
3924 return sprintf(page, "%llu\n",
3925 (unsigned long long)mddev->resync_min);
3928 min_sync_store(mddev_t *mddev, const char *buf, size_t len)
3930 unsigned long long min;
3931 if (strict_strtoull(buf, 10, &min))
3933 if (min > mddev->resync_max)
3935 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
3938 /* Must be a multiple of chunk_size */
3939 if (mddev->chunk_sectors) {
3940 sector_t temp = min;
3941 if (sector_div(temp, mddev->chunk_sectors))
3944 mddev->resync_min = min;
3949 static struct md_sysfs_entry md_min_sync =
3950 __ATTR(sync_min, S_IRUGO|S_IWUSR, min_sync_show, min_sync_store);
3953 max_sync_show(mddev_t *mddev, char *page)
3955 if (mddev->resync_max == MaxSector)
3956 return sprintf(page, "max\n");
3958 return sprintf(page, "%llu\n",
3959 (unsigned long long)mddev->resync_max);
3962 max_sync_store(mddev_t *mddev, const char *buf, size_t len)
3964 if (strncmp(buf, "max", 3) == 0)
3965 mddev->resync_max = MaxSector;
3967 unsigned long long max;
3968 if (strict_strtoull(buf, 10, &max))
3970 if (max < mddev->resync_min)
3972 if (max < mddev->resync_max &&
3974 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
3977 /* Must be a multiple of chunk_size */
3978 if (mddev->chunk_sectors) {
3979 sector_t temp = max;
3980 if (sector_div(temp, mddev->chunk_sectors))
3983 mddev->resync_max = max;
3985 wake_up(&mddev->recovery_wait);
3989 static struct md_sysfs_entry md_max_sync =
3990 __ATTR(sync_max, S_IRUGO|S_IWUSR, max_sync_show, max_sync_store);
3993 suspend_lo_show(mddev_t *mddev, char *page)
3995 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_lo);
3999 suspend_lo_store(mddev_t *mddev, const char *buf, size_t len)
4002 unsigned long long new = simple_strtoull(buf, &e, 10);
4004 if (mddev->pers == NULL ||
4005 mddev->pers->quiesce == NULL)
4007 if (buf == e || (*e && *e != '\n'))
4009 if (new >= mddev->suspend_hi ||
4010 (new > mddev->suspend_lo && new < mddev->suspend_hi)) {
4011 mddev->suspend_lo = new;
4012 mddev->pers->quiesce(mddev, 2);
4017 static struct md_sysfs_entry md_suspend_lo =
4018 __ATTR(suspend_lo, S_IRUGO|S_IWUSR, suspend_lo_show, suspend_lo_store);
4022 suspend_hi_show(mddev_t *mddev, char *page)
4024 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_hi);
4028 suspend_hi_store(mddev_t *mddev, const char *buf, size_t len)
4031 unsigned long long new = simple_strtoull(buf, &e, 10);
4033 if (mddev->pers == NULL ||
4034 mddev->pers->quiesce == NULL)
4036 if (buf == e || (*e && *e != '\n'))
4038 if ((new <= mddev->suspend_lo && mddev->suspend_lo >= mddev->suspend_hi) ||
4039 (new > mddev->suspend_lo && new > mddev->suspend_hi)) {
4040 mddev->suspend_hi = new;
4041 mddev->pers->quiesce(mddev, 1);
4042 mddev->pers->quiesce(mddev, 0);
4047 static struct md_sysfs_entry md_suspend_hi =
4048 __ATTR(suspend_hi, S_IRUGO|S_IWUSR, suspend_hi_show, suspend_hi_store);
4051 reshape_position_show(mddev_t *mddev, char *page)
4053 if (mddev->reshape_position != MaxSector)
4054 return sprintf(page, "%llu\n",
4055 (unsigned long long)mddev->reshape_position);
4056 strcpy(page, "none\n");
4061 reshape_position_store(mddev_t *mddev, const char *buf, size_t len)
4064 unsigned long long new = simple_strtoull(buf, &e, 10);
4067 if (buf == e || (*e && *e != '\n'))
4069 mddev->reshape_position = new;
4070 mddev->delta_disks = 0;
4071 mddev->new_level = mddev->level;
4072 mddev->new_layout = mddev->layout;
4073 mddev->new_chunk_sectors = mddev->chunk_sectors;
4077 static struct md_sysfs_entry md_reshape_position =
4078 __ATTR(reshape_position, S_IRUGO|S_IWUSR, reshape_position_show,
4079 reshape_position_store);
4082 array_size_show(mddev_t *mddev, char *page)
4084 if (mddev->external_size)
4085 return sprintf(page, "%llu\n",
4086 (unsigned long long)mddev->array_sectors/2);
4088 return sprintf(page, "default\n");
4092 array_size_store(mddev_t *mddev, const char *buf, size_t len)
4096 if (strncmp(buf, "default", 7) == 0) {
4098 sectors = mddev->pers->size(mddev, 0, 0);
4100 sectors = mddev->array_sectors;
4102 mddev->external_size = 0;
4104 if (strict_blocks_to_sectors(buf, §ors) < 0)
4106 if (mddev->pers && mddev->pers->size(mddev, 0, 0) < sectors)
4109 mddev->external_size = 1;
4112 mddev->array_sectors = sectors;
4113 set_capacity(mddev->gendisk, mddev->array_sectors);
4115 revalidate_disk(mddev->gendisk);
4120 static struct md_sysfs_entry md_array_size =
4121 __ATTR(array_size, S_IRUGO|S_IWUSR, array_size_show,
4124 static struct attribute *md_default_attrs[] = {
4127 &md_raid_disks.attr,
4128 &md_chunk_size.attr,
4130 &md_resync_start.attr,
4132 &md_new_device.attr,
4133 &md_safe_delay.attr,
4134 &md_array_state.attr,
4135 &md_reshape_position.attr,
4136 &md_array_size.attr,
4137 &max_corr_read_errors.attr,
4141 static struct attribute *md_redundancy_attrs[] = {
4143 &md_mismatches.attr,
4146 &md_sync_speed.attr,
4147 &md_sync_force_parallel.attr,
4148 &md_sync_completed.attr,
4151 &md_suspend_lo.attr,
4152 &md_suspend_hi.attr,
4157 static struct attribute_group md_redundancy_group = {
4159 .attrs = md_redundancy_attrs,
4164 md_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
4166 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
4167 mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
4172 rv = mddev_lock(mddev);
4174 rv = entry->show(mddev, page);
4175 mddev_unlock(mddev);
4181 md_attr_store(struct kobject *kobj, struct attribute *attr,
4182 const char *page, size_t length)
4184 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
4185 mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
4190 if (!capable(CAP_SYS_ADMIN))
4192 rv = mddev_lock(mddev);
4193 if (mddev->hold_active == UNTIL_IOCTL)
4194 mddev->hold_active = 0;
4196 rv = entry->store(mddev, page, length);
4197 mddev_unlock(mddev);
4202 static void md_free(struct kobject *ko)
4204 mddev_t *mddev = container_of(ko, mddev_t, kobj);
4206 if (mddev->sysfs_state)
4207 sysfs_put(mddev->sysfs_state);
4209 if (mddev->gendisk) {
4210 del_gendisk(mddev->gendisk);
4211 put_disk(mddev->gendisk);
4214 blk_cleanup_queue(mddev->queue);
4219 static const struct sysfs_ops md_sysfs_ops = {
4220 .show = md_attr_show,
4221 .store = md_attr_store,
4223 static struct kobj_type md_ktype = {
4225 .sysfs_ops = &md_sysfs_ops,
4226 .default_attrs = md_default_attrs,
4231 static void mddev_delayed_delete(struct work_struct *ws)
4233 mddev_t *mddev = container_of(ws, mddev_t, del_work);
4235 sysfs_remove_group(&mddev->kobj, &md_bitmap_group);
4236 kobject_del(&mddev->kobj);
4237 kobject_put(&mddev->kobj);
4240 static int md_alloc(dev_t dev, char *name)
4242 static DEFINE_MUTEX(disks_mutex);
4243 mddev_t *mddev = mddev_find(dev);
4244 struct gendisk *disk;
4253 partitioned = (MAJOR(mddev->unit) != MD_MAJOR);
4254 shift = partitioned ? MdpMinorShift : 0;
4255 unit = MINOR(mddev->unit) >> shift;
4257 /* wait for any previous instance if this device
4258 * to be completed removed (mddev_delayed_delete).
4260 flush_scheduled_work();
4262 mutex_lock(&disks_mutex);
4268 /* Need to ensure that 'name' is not a duplicate.
4271 spin_lock(&all_mddevs_lock);
4273 list_for_each_entry(mddev2, &all_mddevs, all_mddevs)
4274 if (mddev2->gendisk &&
4275 strcmp(mddev2->gendisk->disk_name, name) == 0) {
4276 spin_unlock(&all_mddevs_lock);
4279 spin_unlock(&all_mddevs_lock);
4283 mddev->queue = blk_alloc_queue(GFP_KERNEL);
4286 mddev->queue->queuedata = mddev;
4288 /* Can be unlocked because the queue is new: no concurrency */
4289 queue_flag_set_unlocked(QUEUE_FLAG_CLUSTER, mddev->queue);
4291 blk_queue_make_request(mddev->queue, md_make_request);
4293 disk = alloc_disk(1 << shift);
4295 blk_cleanup_queue(mddev->queue);
4296 mddev->queue = NULL;
4299 disk->major = MAJOR(mddev->unit);
4300 disk->first_minor = unit << shift;
4302 strcpy(disk->disk_name, name);
4303 else if (partitioned)
4304 sprintf(disk->disk_name, "md_d%d", unit);
4306 sprintf(disk->disk_name, "md%d", unit);
4307 disk->fops = &md_fops;
4308 disk->private_data = mddev;
4309 disk->queue = mddev->queue;
4310 /* Allow extended partitions. This makes the
4311 * 'mdp' device redundant, but we can't really
4314 disk->flags |= GENHD_FL_EXT_DEVT;
4316 mddev->gendisk = disk;
4317 error = kobject_init_and_add(&mddev->kobj, &md_ktype,
4318 &disk_to_dev(disk)->kobj, "%s", "md");
4320 /* This isn't possible, but as kobject_init_and_add is marked
4321 * __must_check, we must do something with the result
4323 printk(KERN_WARNING "md: cannot register %s/md - name in use\n",
4327 if (mddev->kobj.sd &&
4328 sysfs_create_group(&mddev->kobj, &md_bitmap_group))
4329 printk(KERN_DEBUG "pointless warning\n");
4331 mutex_unlock(&disks_mutex);
4332 if (!error && mddev->kobj.sd) {
4333 kobject_uevent(&mddev->kobj, KOBJ_ADD);
4334 mddev->sysfs_state = sysfs_get_dirent_safe(mddev->kobj.sd, "array_state");
4340 static struct kobject *md_probe(dev_t dev, int *part, void *data)
4342 md_alloc(dev, NULL);
4346 static int add_named_array(const char *val, struct kernel_param *kp)
4348 /* val must be "md_*" where * is not all digits.
4349 * We allocate an array with a large free minor number, and
4350 * set the name to val. val must not already be an active name.
4352 int len = strlen(val);
4353 char buf[DISK_NAME_LEN];
4355 while (len && val[len-1] == '\n')
4357 if (len >= DISK_NAME_LEN)
4359 strlcpy(buf, val, len+1);
4360 if (strncmp(buf, "md_", 3) != 0)
4362 return md_alloc(0, buf);
4365 static void md_safemode_timeout(unsigned long data)
4367 mddev_t *mddev = (mddev_t *) data;
4369 if (!atomic_read(&mddev->writes_pending)) {
4370 mddev->safemode = 1;
4371 if (mddev->external)
4372 sysfs_notify_dirent_safe(mddev->sysfs_state);
4374 md_wakeup_thread(mddev->thread);
4377 static int start_dirty_degraded;
4379 int md_run(mddev_t *mddev)
4383 struct mdk_personality *pers;
4385 if (list_empty(&mddev->disks))
4386 /* cannot run an array with no devices.. */
4391 /* Cannot run until previous stop completes properly */
4392 if (mddev->sysfs_active)
4396 * Analyze all RAID superblock(s)
4398 if (!mddev->raid_disks) {
4399 if (!mddev->persistent)
4404 if (mddev->level != LEVEL_NONE)
4405 request_module("md-level-%d", mddev->level);
4406 else if (mddev->clevel[0])
4407 request_module("md-%s", mddev->clevel);
4410 * Drop all container device buffers, from now on
4411 * the only valid external interface is through the md
4414 list_for_each_entry(rdev, &mddev->disks, same_set) {
4415 if (test_bit(Faulty, &rdev->flags))
4417 sync_blockdev(rdev->bdev);
4418 invalidate_bdev(rdev->bdev);
4420 /* perform some consistency tests on the device.
4421 * We don't want the data to overlap the metadata,
4422 * Internal Bitmap issues have been handled elsewhere.
4424 if (rdev->data_offset < rdev->sb_start) {
4425 if (mddev->dev_sectors &&
4426 rdev->data_offset + mddev->dev_sectors
4428 printk("md: %s: data overlaps metadata\n",
4433 if (rdev->sb_start + rdev->sb_size/512
4434 > rdev->data_offset) {
4435 printk("md: %s: metadata overlaps data\n",
4440 sysfs_notify_dirent_safe(rdev->sysfs_state);
4443 spin_lock(&pers_lock);
4444 pers = find_pers(mddev->level, mddev->clevel);
4445 if (!pers || !try_module_get(pers->owner)) {
4446 spin_unlock(&pers_lock);
4447 if (mddev->level != LEVEL_NONE)
4448 printk(KERN_WARNING "md: personality for level %d is not loaded!\n",
4451 printk(KERN_WARNING "md: personality for level %s is not loaded!\n",
4456 spin_unlock(&pers_lock);
4457 if (mddev->level != pers->level) {
4458 mddev->level = pers->level;
4459 mddev->new_level = pers->level;
4461 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
4463 if (mddev->reshape_position != MaxSector &&
4464 pers->start_reshape == NULL) {
4465 /* This personality cannot handle reshaping... */
4467 module_put(pers->owner);
4471 if (pers->sync_request) {
4472 /* Warn if this is a potentially silly
4475 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
4479 list_for_each_entry(rdev, &mddev->disks, same_set)
4480 list_for_each_entry(rdev2, &mddev->disks, same_set) {
4482 rdev->bdev->bd_contains ==
4483 rdev2->bdev->bd_contains) {
4485 "%s: WARNING: %s appears to be"
4486 " on the same physical disk as"
4489 bdevname(rdev->bdev,b),
4490 bdevname(rdev2->bdev,b2));
4497 "True protection against single-disk"
4498 " failure might be compromised.\n");
4501 mddev->recovery = 0;
4502 /* may be over-ridden by personality */
4503 mddev->resync_max_sectors = mddev->dev_sectors;
4505 mddev->barriers_work = 1;
4506 mddev->ok_start_degraded = start_dirty_degraded;
4508 if (start_readonly && mddev->ro == 0)
4509 mddev->ro = 2; /* read-only, but switch on first write */
4511 err = mddev->pers->run(mddev);
4513 printk(KERN_ERR "md: pers->run() failed ...\n");
4514 else if (mddev->pers->size(mddev, 0, 0) < mddev->array_sectors) {
4515 WARN_ONCE(!mddev->external_size, "%s: default size too small,"
4516 " but 'external_size' not in effect?\n", __func__);
4518 "md: invalid array_size %llu > default size %llu\n",
4519 (unsigned long long)mddev->array_sectors / 2,
4520 (unsigned long long)mddev->pers->size(mddev, 0, 0) / 2);
4522 mddev->pers->stop(mddev);
4524 if (err == 0 && mddev->pers->sync_request) {
4525 err = bitmap_create(mddev);
4527 printk(KERN_ERR "%s: failed to create bitmap (%d)\n",
4528 mdname(mddev), err);
4529 mddev->pers->stop(mddev);
4533 module_put(mddev->pers->owner);
4535 bitmap_destroy(mddev);
4538 if (mddev->pers->sync_request) {
4539 if (mddev->kobj.sd &&
4540 sysfs_create_group(&mddev->kobj, &md_redundancy_group))
4542 "md: cannot register extra attributes for %s\n",
4544 mddev->sysfs_action = sysfs_get_dirent_safe(mddev->kobj.sd, "sync_action");
4545 } else if (mddev->ro == 2) /* auto-readonly not meaningful */
4548 atomic_set(&mddev->writes_pending,0);
4549 atomic_set(&mddev->max_corr_read_errors,
4550 MD_DEFAULT_MAX_CORRECTED_READ_ERRORS);
4551 mddev->safemode = 0;
4552 mddev->safemode_timer.function = md_safemode_timeout;
4553 mddev->safemode_timer.data = (unsigned long) mddev;
4554 mddev->safemode_delay = (200 * HZ)/1000 +1; /* 200 msec delay */
4557 list_for_each_entry(rdev, &mddev->disks, same_set)
4558 if (rdev->raid_disk >= 0) {
4560 sprintf(nm, "rd%d", rdev->raid_disk);
4561 if (sysfs_create_link(&mddev->kobj, &rdev->kobj, nm))
4562 /* failure here is OK */;
4565 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4568 md_update_sb(mddev, 0);
4570 md_wakeup_thread(mddev->thread);
4571 md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
4573 md_new_event(mddev);
4574 sysfs_notify_dirent_safe(mddev->sysfs_state);
4575 sysfs_notify_dirent_safe(mddev->sysfs_action);
4576 sysfs_notify(&mddev->kobj, NULL, "degraded");
4579 EXPORT_SYMBOL_GPL(md_run);
4581 static int do_md_run(mddev_t *mddev)
4585 err = md_run(mddev);
4588 err = bitmap_load(mddev);
4590 bitmap_destroy(mddev);
4593 set_capacity(mddev->gendisk, mddev->array_sectors);
4594 revalidate_disk(mddev->gendisk);
4595 kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
4600 static int restart_array(mddev_t *mddev)
4602 struct gendisk *disk = mddev->gendisk;
4604 /* Complain if it has no devices */
4605 if (list_empty(&mddev->disks))
4611 mddev->safemode = 0;
4613 set_disk_ro(disk, 0);
4614 printk(KERN_INFO "md: %s switched to read-write mode.\n",
4616 /* Kick recovery or resync if necessary */
4617 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4618 md_wakeup_thread(mddev->thread);
4619 md_wakeup_thread(mddev->sync_thread);
4620 sysfs_notify_dirent_safe(mddev->sysfs_state);
4624 /* similar to deny_write_access, but accounts for our holding a reference
4625 * to the file ourselves */
4626 static int deny_bitmap_write_access(struct file * file)
4628 struct inode *inode = file->f_mapping->host;
4630 spin_lock(&inode->i_lock);
4631 if (atomic_read(&inode->i_writecount) > 1) {
4632 spin_unlock(&inode->i_lock);
4635 atomic_set(&inode->i_writecount, -1);
4636 spin_unlock(&inode->i_lock);
4641 void restore_bitmap_write_access(struct file *file)
4643 struct inode *inode = file->f_mapping->host;
4645 spin_lock(&inode->i_lock);
4646 atomic_set(&inode->i_writecount, 1);
4647 spin_unlock(&inode->i_lock);
4650 static void md_clean(mddev_t *mddev)
4652 mddev->array_sectors = 0;
4653 mddev->external_size = 0;
4654 mddev->dev_sectors = 0;
4655 mddev->raid_disks = 0;
4656 mddev->recovery_cp = 0;
4657 mddev->resync_min = 0;
4658 mddev->resync_max = MaxSector;
4659 mddev->reshape_position = MaxSector;
4660 mddev->external = 0;
4661 mddev->persistent = 0;
4662 mddev->level = LEVEL_NONE;
4663 mddev->clevel[0] = 0;
4666 mddev->metadata_type[0] = 0;
4667 mddev->chunk_sectors = 0;
4668 mddev->ctime = mddev->utime = 0;
4670 mddev->max_disks = 0;
4672 mddev->can_decrease_events = 0;
4673 mddev->delta_disks = 0;
4674 mddev->new_level = LEVEL_NONE;
4675 mddev->new_layout = 0;
4676 mddev->new_chunk_sectors = 0;
4677 mddev->curr_resync = 0;
4678 mddev->resync_mismatches = 0;
4679 mddev->suspend_lo = mddev->suspend_hi = 0;
4680 mddev->sync_speed_min = mddev->sync_speed_max = 0;
4681 mddev->recovery = 0;
4683 mddev->degraded = 0;
4684 mddev->barriers_work = 0;
4685 mddev->safemode = 0;
4686 mddev->bitmap_info.offset = 0;
4687 mddev->bitmap_info.default_offset = 0;
4688 mddev->bitmap_info.chunksize = 0;
4689 mddev->bitmap_info.daemon_sleep = 0;
4690 mddev->bitmap_info.max_write_behind = 0;
4694 void md_stop_writes(mddev_t *mddev)
4696 if (mddev->sync_thread) {
4697 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4698 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
4699 md_unregister_thread(mddev->sync_thread);
4700 mddev->sync_thread = NULL;
4703 del_timer_sync(&mddev->safemode_timer);
4705 bitmap_flush(mddev);
4706 md_super_wait(mddev);
4708 if (!mddev->in_sync || mddev->flags) {
4709 /* mark array as shutdown cleanly */
4711 md_update_sb(mddev, 1);
4714 EXPORT_SYMBOL_GPL(md_stop_writes);
4716 void md_stop(mddev_t *mddev)
4718 mddev->pers->stop(mddev);
4719 if (mddev->pers->sync_request && mddev->to_remove == NULL)
4720 mddev->to_remove = &md_redundancy_group;
4721 module_put(mddev->pers->owner);
4723 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4725 EXPORT_SYMBOL_GPL(md_stop);
4727 static int md_set_readonly(mddev_t *mddev, int is_open)
4730 mutex_lock(&mddev->open_mutex);
4731 if (atomic_read(&mddev->openers) > is_open) {
4732 printk("md: %s still in use.\n",mdname(mddev));
4737 md_stop_writes(mddev);
4743 set_disk_ro(mddev->gendisk, 1);
4744 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4745 sysfs_notify_dirent_safe(mddev->sysfs_state);
4749 mutex_unlock(&mddev->open_mutex);
4754 * 0 - completely stop and dis-assemble array
4755 * 2 - stop but do not disassemble array
4757 static int do_md_stop(mddev_t * mddev, int mode, int is_open)
4759 struct gendisk *disk = mddev->gendisk;
4762 mutex_lock(&mddev->open_mutex);
4763 if (atomic_read(&mddev->openers) > is_open ||
4764 mddev->sysfs_active) {
4765 printk("md: %s still in use.\n",mdname(mddev));
4766 mutex_unlock(&mddev->open_mutex);
4772 set_disk_ro(disk, 0);
4774 md_stop_writes(mddev);
4776 mddev->queue->merge_bvec_fn = NULL;
4777 mddev->queue->unplug_fn = NULL;
4778 mddev->queue->backing_dev_info.congested_fn = NULL;
4780 /* tell userspace to handle 'inactive' */
4781 sysfs_notify_dirent_safe(mddev->sysfs_state);
4783 list_for_each_entry(rdev, &mddev->disks, same_set)
4784 if (rdev->raid_disk >= 0) {
4786 sprintf(nm, "rd%d", rdev->raid_disk);
4787 sysfs_remove_link(&mddev->kobj, nm);
4790 set_capacity(disk, 0);
4791 mutex_unlock(&mddev->open_mutex);
4792 revalidate_disk(disk);
4797 mutex_unlock(&mddev->open_mutex);
4799 * Free resources if final stop
4802 printk(KERN_INFO "md: %s stopped.\n", mdname(mddev));
4804 bitmap_destroy(mddev);
4805 if (mddev->bitmap_info.file) {
4806 restore_bitmap_write_access(mddev->bitmap_info.file);
4807 fput(mddev->bitmap_info.file);
4808 mddev->bitmap_info.file = NULL;
4810 mddev->bitmap_info.offset = 0;
4812 export_array(mddev);
4815 kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
4816 if (mddev->hold_active == UNTIL_STOP)
4817 mddev->hold_active = 0;
4819 blk_integrity_unregister(disk);
4820 md_new_event(mddev);
4821 sysfs_notify_dirent_safe(mddev->sysfs_state);
4826 static void autorun_array(mddev_t *mddev)
4831 if (list_empty(&mddev->disks))
4834 printk(KERN_INFO "md: running: ");
4836 list_for_each_entry(rdev, &mddev->disks, same_set) {
4837 char b[BDEVNAME_SIZE];
4838 printk("<%s>", bdevname(rdev->bdev,b));
4842 err = do_md_run(mddev);
4844 printk(KERN_WARNING "md: do_md_run() returned %d\n", err);
4845 do_md_stop(mddev, 0, 0);
4850 * lets try to run arrays based on all disks that have arrived
4851 * until now. (those are in pending_raid_disks)
4853 * the method: pick the first pending disk, collect all disks with
4854 * the same UUID, remove all from the pending list and put them into
4855 * the 'same_array' list. Then order this list based on superblock
4856 * update time (freshest comes first), kick out 'old' disks and
4857 * compare superblocks. If everything's fine then run it.
4859 * If "unit" is allocated, then bump its reference count
4861 static void autorun_devices(int part)
4863 mdk_rdev_t *rdev0, *rdev, *tmp;
4865 char b[BDEVNAME_SIZE];
4867 printk(KERN_INFO "md: autorun ...\n");
4868 while (!list_empty(&pending_raid_disks)) {
4871 LIST_HEAD(candidates);
4872 rdev0 = list_entry(pending_raid_disks.next,
4873 mdk_rdev_t, same_set);
4875 printk(KERN_INFO "md: considering %s ...\n",
4876 bdevname(rdev0->bdev,b));
4877 INIT_LIST_HEAD(&candidates);
4878 rdev_for_each_list(rdev, tmp, &pending_raid_disks)
4879 if (super_90_load(rdev, rdev0, 0) >= 0) {
4880 printk(KERN_INFO "md: adding %s ...\n",
4881 bdevname(rdev->bdev,b));
4882 list_move(&rdev->same_set, &candidates);
4885 * now we have a set of devices, with all of them having
4886 * mostly sane superblocks. It's time to allocate the
4890 dev = MKDEV(mdp_major,
4891 rdev0->preferred_minor << MdpMinorShift);
4892 unit = MINOR(dev) >> MdpMinorShift;
4894 dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
4897 if (rdev0->preferred_minor != unit) {
4898 printk(KERN_INFO "md: unit number in %s is bad: %d\n",
4899 bdevname(rdev0->bdev, b), rdev0->preferred_minor);
4903 md_probe(dev, NULL, NULL);
4904 mddev = mddev_find(dev);
4905 if (!mddev || !mddev->gendisk) {
4909 "md: cannot allocate memory for md drive.\n");
4912 if (mddev_lock(mddev))
4913 printk(KERN_WARNING "md: %s locked, cannot run\n",
4915 else if (mddev->raid_disks || mddev->major_version
4916 || !list_empty(&mddev->disks)) {
4918 "md: %s already running, cannot run %s\n",
4919 mdname(mddev), bdevname(rdev0->bdev,b));
4920 mddev_unlock(mddev);
4922 printk(KERN_INFO "md: created %s\n", mdname(mddev));
4923 mddev->persistent = 1;
4924 rdev_for_each_list(rdev, tmp, &candidates) {
4925 list_del_init(&rdev->same_set);
4926 if (bind_rdev_to_array(rdev, mddev))
4929 autorun_array(mddev);
4930 mddev_unlock(mddev);
4932 /* on success, candidates will be empty, on error
4935 rdev_for_each_list(rdev, tmp, &candidates) {
4936 list_del_init(&rdev->same_set);
4941 printk(KERN_INFO "md: ... autorun DONE.\n");
4943 #endif /* !MODULE */
4945 static int get_version(void __user * arg)
4949 ver.major = MD_MAJOR_VERSION;
4950 ver.minor = MD_MINOR_VERSION;
4951 ver.patchlevel = MD_PATCHLEVEL_VERSION;
4953 if (copy_to_user(arg, &ver, sizeof(ver)))
4959 static int get_array_info(mddev_t * mddev, void __user * arg)
4961 mdu_array_info_t info;
4962 int nr,working,insync,failed,spare;
4965 nr=working=insync=failed=spare=0;
4966 list_for_each_entry(rdev, &mddev->disks, same_set) {
4968 if (test_bit(Faulty, &rdev->flags))
4972 if (test_bit(In_sync, &rdev->flags))
4979 info.major_version = mddev->major_version;
4980 info.minor_version = mddev->minor_version;
4981 info.patch_version = MD_PATCHLEVEL_VERSION;
4982 info.ctime = mddev->ctime;
4983 info.level = mddev->level;
4984 info.size = mddev->dev_sectors / 2;
4985 if (info.size != mddev->dev_sectors / 2) /* overflow */
4988 info.raid_disks = mddev->raid_disks;
4989 info.md_minor = mddev->md_minor;
4990 info.not_persistent= !mddev->persistent;
4992 info.utime = mddev->utime;
4995 info.state = (1<<MD_SB_CLEAN);
4996 if (mddev->bitmap && mddev->bitmap_info.offset)
4997 info.state = (1<<MD_SB_BITMAP_PRESENT);
4998 info.active_disks = insync;
4999 info.working_disks = working;
5000 info.failed_disks = failed;
5001 info.spare_disks = spare;
5003 info.layout = mddev->layout;
5004 info.chunk_size = mddev->chunk_sectors << 9;
5006 if (copy_to_user(arg, &info, sizeof(info)))
5012 static int get_bitmap_file(mddev_t * mddev, void __user * arg)
5014 mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */
5015 char *ptr, *buf = NULL;
5018 if (md_allow_write(mddev))
5019 file = kmalloc(sizeof(*file), GFP_NOIO);
5021 file = kmalloc(sizeof(*file), GFP_KERNEL);
5026 /* bitmap disabled, zero the first byte and copy out */
5027 if (!mddev->bitmap || !mddev->bitmap->file) {
5028 file->pathname[0] = '\0';
5032 buf = kmalloc(sizeof(file->pathname), GFP_KERNEL);
5036 ptr = d_path(&mddev->bitmap->file->f_path, buf, sizeof(file->pathname));
5040 strcpy(file->pathname, ptr);
5044 if (copy_to_user(arg, file, sizeof(*file)))
5052 static int get_disk_info(mddev_t * mddev, void __user * arg)
5054 mdu_disk_info_t info;
5057 if (copy_from_user(&info, arg, sizeof(info)))
5060 rdev = find_rdev_nr(mddev, info.number);
5062 info.major = MAJOR(rdev->bdev->bd_dev);
5063 info.minor = MINOR(rdev->bdev->bd_dev);
5064 info.raid_disk = rdev->raid_disk;
5066 if (test_bit(Faulty, &rdev->flags))
5067 info.state |= (1<<MD_DISK_FAULTY);
5068 else if (test_bit(In_sync, &rdev->flags)) {
5069 info.state |= (1<<MD_DISK_ACTIVE);
5070 info.state |= (1<<MD_DISK_SYNC);
5072 if (test_bit(WriteMostly, &rdev->flags))
5073 info.state |= (1<<MD_DISK_WRITEMOSTLY);
5075 info.major = info.minor = 0;
5076 info.raid_disk = -1;
5077 info.state = (1<<MD_DISK_REMOVED);
5080 if (copy_to_user(arg, &info, sizeof(info)))
5086 static int add_new_disk(mddev_t * mddev, mdu_disk_info_t *info)
5088 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
5090 dev_t dev = MKDEV(info->major,info->minor);
5092 if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
5095 if (!mddev->raid_disks) {
5097 /* expecting a device which has a superblock */
5098 rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
5101 "md: md_import_device returned %ld\n",
5103 return PTR_ERR(rdev);
5105 if (!list_empty(&mddev->disks)) {
5106 mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
5107 mdk_rdev_t, same_set);
5108 err = super_types[mddev->major_version]
5109 .load_super(rdev, rdev0, mddev->minor_version);
5112 "md: %s has different UUID to %s\n",
5113 bdevname(rdev->bdev,b),
5114 bdevname(rdev0->bdev,b2));
5119 err = bind_rdev_to_array(rdev, mddev);
5126 * add_new_disk can be used once the array is assembled
5127 * to add "hot spares". They must already have a superblock
5132 if (!mddev->pers->hot_add_disk) {
5134 "%s: personality does not support diskops!\n",
5138 if (mddev->persistent)
5139 rdev = md_import_device(dev, mddev->major_version,
5140 mddev->minor_version);
5142 rdev = md_import_device(dev, -1, -1);
5145 "md: md_import_device returned %ld\n",
5147 return PTR_ERR(rdev);
5149 /* set save_raid_disk if appropriate */
5150 if (!mddev->persistent) {
5151 if (info->state & (1<<MD_DISK_SYNC) &&
5152 info->raid_disk < mddev->raid_disks)
5153 rdev->raid_disk = info->raid_disk;
5155 rdev->raid_disk = -1;
5157 super_types[mddev->major_version].
5158 validate_super(mddev, rdev);
5159 rdev->saved_raid_disk = rdev->raid_disk;
5161 clear_bit(In_sync, &rdev->flags); /* just to be sure */
5162 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
5163 set_bit(WriteMostly, &rdev->flags);
5165 clear_bit(WriteMostly, &rdev->flags);
5167 rdev->raid_disk = -1;
5168 err = bind_rdev_to_array(rdev, mddev);
5169 if (!err && !mddev->pers->hot_remove_disk) {
5170 /* If there is hot_add_disk but no hot_remove_disk
5171 * then added disks for geometry changes,
5172 * and should be added immediately.
5174 super_types[mddev->major_version].
5175 validate_super(mddev, rdev);
5176 err = mddev->pers->hot_add_disk(mddev, rdev);
5178 unbind_rdev_from_array(rdev);
5183 sysfs_notify_dirent_safe(rdev->sysfs_state);
5185 md_update_sb(mddev, 1);
5186 if (mddev->degraded)
5187 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
5188 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5189 md_wakeup_thread(mddev->thread);
5193 /* otherwise, add_new_disk is only allowed
5194 * for major_version==0 superblocks
5196 if (mddev->major_version != 0) {
5197 printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n",
5202 if (!(info->state & (1<<MD_DISK_FAULTY))) {
5204 rdev = md_import_device(dev, -1, 0);
5207 "md: error, md_import_device() returned %ld\n",
5209 return PTR_ERR(rdev);
5211 rdev->desc_nr = info->number;
5212 if (info->raid_disk < mddev->raid_disks)
5213 rdev->raid_disk = info->raid_disk;
5215 rdev->raid_disk = -1;
5217 if (rdev->raid_disk < mddev->raid_disks)
5218 if (info->state & (1<<MD_DISK_SYNC))
5219 set_bit(In_sync, &rdev->flags);
5221 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
5222 set_bit(WriteMostly, &rdev->flags);
5224 if (!mddev->persistent) {
5225 printk(KERN_INFO "md: nonpersistent superblock ...\n");
5226 rdev->sb_start = rdev->bdev->bd_inode->i_size / 512;
5228 rdev->sb_start = calc_dev_sboffset(rdev->bdev);
5229 rdev->sectors = rdev->sb_start;
5231 err = bind_rdev_to_array(rdev, mddev);
5241 static int hot_remove_disk(mddev_t * mddev, dev_t dev)
5243 char b[BDEVNAME_SIZE];
5246 rdev = find_rdev(mddev, dev);
5250 if (rdev->raid_disk >= 0)
5253 kick_rdev_from_array(rdev);
5254 md_update_sb(mddev, 1);
5255 md_new_event(mddev);
5259 printk(KERN_WARNING "md: cannot remove active disk %s from %s ...\n",
5260 bdevname(rdev->bdev,b), mdname(mddev));
5264 static int hot_add_disk(mddev_t * mddev, dev_t dev)
5266 char b[BDEVNAME_SIZE];
5273 if (mddev->major_version != 0) {
5274 printk(KERN_WARNING "%s: HOT_ADD may only be used with"
5275 " version-0 superblocks.\n",
5279 if (!mddev->pers->hot_add_disk) {
5281 "%s: personality does not support diskops!\n",
5286 rdev = md_import_device(dev, -1, 0);
5289 "md: error, md_import_device() returned %ld\n",
5294 if (mddev->persistent)
5295 rdev->sb_start = calc_dev_sboffset(rdev->bdev);
5297 rdev->sb_start = rdev->bdev->bd_inode->i_size / 512;
5299 rdev->sectors = rdev->sb_start;
5301 if (test_bit(Faulty, &rdev->flags)) {
5303 "md: can not hot-add faulty %s disk to %s!\n",
5304 bdevname(rdev->bdev,b), mdname(mddev));
5308 clear_bit(In_sync, &rdev->flags);
5310 rdev->saved_raid_disk = -1;
5311 err = bind_rdev_to_array(rdev, mddev);
5316 * The rest should better be atomic, we can have disk failures
5317 * noticed in interrupt contexts ...
5320 rdev->raid_disk = -1;
5322 md_update_sb(mddev, 1);
5325 * Kick recovery, maybe this spare has to be added to the
5326 * array immediately.
5328 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5329 md_wakeup_thread(mddev->thread);
5330 md_new_event(mddev);
5338 static int set_bitmap_file(mddev_t *mddev, int fd)
5343 if (!mddev->pers->quiesce)
5345 if (mddev->recovery || mddev->sync_thread)
5347 /* we should be able to change the bitmap.. */
5353 return -EEXIST; /* cannot add when bitmap is present */
5354 mddev->bitmap_info.file = fget(fd);
5356 if (mddev->bitmap_info.file == NULL) {
5357 printk(KERN_ERR "%s: error: failed to get bitmap file\n",
5362 err = deny_bitmap_write_access(mddev->bitmap_info.file);
5364 printk(KERN_ERR "%s: error: bitmap file is already in use\n",
5366 fput(mddev->bitmap_info.file);
5367 mddev->bitmap_info.file = NULL;
5370 mddev->bitmap_info.offset = 0; /* file overrides offset */
5371 } else if (mddev->bitmap == NULL)
5372 return -ENOENT; /* cannot remove what isn't there */
5375 mddev->pers->quiesce(mddev, 1);
5377 err = bitmap_create(mddev);
5379 err = bitmap_load(mddev);
5381 if (fd < 0 || err) {
5382 bitmap_destroy(mddev);
5383 fd = -1; /* make sure to put the file */
5385 mddev->pers->quiesce(mddev, 0);
5388 if (mddev->bitmap_info.file) {
5389 restore_bitmap_write_access(mddev->bitmap_info.file);
5390 fput(mddev->bitmap_info.file);
5392 mddev->bitmap_info.file = NULL;
5399 * set_array_info is used two different ways
5400 * The original usage is when creating a new array.
5401 * In this usage, raid_disks is > 0 and it together with
5402 * level, size, not_persistent,layout,chunksize determine the
5403 * shape of the array.
5404 * This will always create an array with a type-0.90.0 superblock.
5405 * The newer usage is when assembling an array.
5406 * In this case raid_disks will be 0, and the major_version field is
5407 * use to determine which style super-blocks are to be found on the devices.
5408 * The minor and patch _version numbers are also kept incase the
5409 * super_block handler wishes to interpret them.
5411 static int set_array_info(mddev_t * mddev, mdu_array_info_t *info)
5414 if (info->raid_disks == 0) {
5415 /* just setting version number for superblock loading */
5416 if (info->major_version < 0 ||
5417 info->major_version >= ARRAY_SIZE(super_types) ||
5418 super_types[info->major_version].name == NULL) {
5419 /* maybe try to auto-load a module? */
5421 "md: superblock version %d not known\n",
5422 info->major_version);
5425 mddev->major_version = info->major_version;
5426 mddev->minor_version = info->minor_version;
5427 mddev->patch_version = info->patch_version;
5428 mddev->persistent = !info->not_persistent;
5429 /* ensure mddev_put doesn't delete this now that there
5430 * is some minimal configuration.
5432 mddev->ctime = get_seconds();
5435 mddev->major_version = MD_MAJOR_VERSION;
5436 mddev->minor_version = MD_MINOR_VERSION;
5437 mddev->patch_version = MD_PATCHLEVEL_VERSION;
5438 mddev->ctime = get_seconds();
5440 mddev->level = info->level;
5441 mddev->clevel[0] = 0;
5442 mddev->dev_sectors = 2 * (sector_t)info->size;
5443 mddev->raid_disks = info->raid_disks;
5444 /* don't set md_minor, it is determined by which /dev/md* was
5447 if (info->state & (1<<MD_SB_CLEAN))
5448 mddev->recovery_cp = MaxSector;
5450 mddev->recovery_cp = 0;
5451 mddev->persistent = ! info->not_persistent;
5452 mddev->external = 0;
5454 mddev->layout = info->layout;
5455 mddev->chunk_sectors = info->chunk_size >> 9;
5457 mddev->max_disks = MD_SB_DISKS;
5459 if (mddev->persistent)
5461 set_bit(MD_CHANGE_DEVS, &mddev->flags);
5463 mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9;
5464 mddev->bitmap_info.offset = 0;
5466 mddev->reshape_position = MaxSector;
5469 * Generate a 128 bit UUID
5471 get_random_bytes(mddev->uuid, 16);
5473 mddev->new_level = mddev->level;
5474 mddev->new_chunk_sectors = mddev->chunk_sectors;
5475 mddev->new_layout = mddev->layout;
5476 mddev->delta_disks = 0;
5481 void md_set_array_sectors(mddev_t *mddev, sector_t array_sectors)
5483 WARN(!mddev_is_locked(mddev), "%s: unlocked mddev!\n", __func__);
5485 if (mddev->external_size)
5488 mddev->array_sectors = array_sectors;
5490 EXPORT_SYMBOL(md_set_array_sectors);
5492 static int update_size(mddev_t *mddev, sector_t num_sectors)
5496 int fit = (num_sectors == 0);
5498 if (mddev->pers->resize == NULL)
5500 /* The "num_sectors" is the number of sectors of each device that
5501 * is used. This can only make sense for arrays with redundancy.
5502 * linear and raid0 always use whatever space is available. We can only
5503 * consider changing this number if no resync or reconstruction is
5504 * happening, and if the new size is acceptable. It must fit before the
5505 * sb_start or, if that is <data_offset, it must fit before the size
5506 * of each device. If num_sectors is zero, we find the largest size
5510 if (mddev->sync_thread)
5513 /* Sorry, cannot grow a bitmap yet, just remove it,
5517 list_for_each_entry(rdev, &mddev->disks, same_set) {
5518 sector_t avail = rdev->sectors;
5520 if (fit && (num_sectors == 0 || num_sectors > avail))
5521 num_sectors = avail;
5522 if (avail < num_sectors)
5525 rv = mddev->pers->resize(mddev, num_sectors);
5527 revalidate_disk(mddev->gendisk);
5531 static int update_raid_disks(mddev_t *mddev, int raid_disks)
5534 /* change the number of raid disks */
5535 if (mddev->pers->check_reshape == NULL)
5537 if (raid_disks <= 0 ||
5538 (mddev->max_disks && raid_disks >= mddev->max_disks))
5540 if (mddev->sync_thread || mddev->reshape_position != MaxSector)
5542 mddev->delta_disks = raid_disks - mddev->raid_disks;
5544 rv = mddev->pers->check_reshape(mddev);
5550 * update_array_info is used to change the configuration of an
5552 * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
5553 * fields in the info are checked against the array.
5554 * Any differences that cannot be handled will cause an error.
5555 * Normally, only one change can be managed at a time.
5557 static int update_array_info(mddev_t *mddev, mdu_array_info_t *info)
5563 /* calculate expected state,ignoring low bits */
5564 if (mddev->bitmap && mddev->bitmap_info.offset)
5565 state |= (1 << MD_SB_BITMAP_PRESENT);
5567 if (mddev->major_version != info->major_version ||
5568 mddev->minor_version != info->minor_version ||
5569 /* mddev->patch_version != info->patch_version || */
5570 mddev->ctime != info->ctime ||
5571 mddev->level != info->level ||
5572 /* mddev->layout != info->layout || */
5573 !mddev->persistent != info->not_persistent||
5574 mddev->chunk_sectors != info->chunk_size >> 9 ||
5575 /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
5576 ((state^info->state) & 0xfffffe00)
5579 /* Check there is only one change */
5580 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
5582 if (mddev->raid_disks != info->raid_disks)
5584 if (mddev->layout != info->layout)
5586 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT))
5593 if (mddev->layout != info->layout) {
5595 * we don't need to do anything at the md level, the
5596 * personality will take care of it all.
5598 if (mddev->pers->check_reshape == NULL)
5601 mddev->new_layout = info->layout;
5602 rv = mddev->pers->check_reshape(mddev);
5604 mddev->new_layout = mddev->layout;
5608 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
5609 rv = update_size(mddev, (sector_t)info->size * 2);
5611 if (mddev->raid_disks != info->raid_disks)
5612 rv = update_raid_disks(mddev, info->raid_disks);
5614 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) {
5615 if (mddev->pers->quiesce == NULL)
5617 if (mddev->recovery || mddev->sync_thread)
5619 if (info->state & (1<<MD_SB_BITMAP_PRESENT)) {
5620 /* add the bitmap */
5623 if (mddev->bitmap_info.default_offset == 0)
5625 mddev->bitmap_info.offset =
5626 mddev->bitmap_info.default_offset;
5627 mddev->pers->quiesce(mddev, 1);
5628 rv = bitmap_create(mddev);
5630 rv = bitmap_load(mddev);
5632 bitmap_destroy(mddev);
5633 mddev->pers->quiesce(mddev, 0);
5635 /* remove the bitmap */
5638 if (mddev->bitmap->file)
5640 mddev->pers->quiesce(mddev, 1);
5641 bitmap_destroy(mddev);
5642 mddev->pers->quiesce(mddev, 0);
5643 mddev->bitmap_info.offset = 0;
5646 md_update_sb(mddev, 1);
5650 static int set_disk_faulty(mddev_t *mddev, dev_t dev)
5654 if (mddev->pers == NULL)
5657 rdev = find_rdev(mddev, dev);
5661 md_error(mddev, rdev);
5666 * We have a problem here : there is no easy way to give a CHS
5667 * virtual geometry. We currently pretend that we have a 2 heads
5668 * 4 sectors (with a BIG number of cylinders...). This drives
5669 * dosfs just mad... ;-)
5671 static int md_getgeo(struct block_device *bdev, struct hd_geometry *geo)
5673 mddev_t *mddev = bdev->bd_disk->private_data;
5677 geo->cylinders = mddev->array_sectors / 8;
5681 static int md_ioctl(struct block_device *bdev, fmode_t mode,
5682 unsigned int cmd, unsigned long arg)
5685 void __user *argp = (void __user *)arg;
5686 mddev_t *mddev = NULL;
5689 if (!capable(CAP_SYS_ADMIN))
5693 * Commands dealing with the RAID driver but not any
5699 err = get_version(argp);
5702 case PRINT_RAID_DEBUG:
5710 autostart_arrays(arg);
5717 * Commands creating/starting a new array:
5720 mddev = bdev->bd_disk->private_data;
5727 err = mddev_lock(mddev);
5730 "md: ioctl lock interrupted, reason %d, cmd %d\n",
5737 case SET_ARRAY_INFO:
5739 mdu_array_info_t info;
5741 memset(&info, 0, sizeof(info));
5742 else if (copy_from_user(&info, argp, sizeof(info))) {
5747 err = update_array_info(mddev, &info);
5749 printk(KERN_WARNING "md: couldn't update"
5750 " array info. %d\n", err);
5755 if (!list_empty(&mddev->disks)) {
5757 "md: array %s already has disks!\n",
5762 if (mddev->raid_disks) {
5764 "md: array %s already initialised!\n",
5769 err = set_array_info(mddev, &info);
5771 printk(KERN_WARNING "md: couldn't set"
5772 " array info. %d\n", err);
5782 * Commands querying/configuring an existing array:
5784 /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
5785 * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */
5786 if ((!mddev->raid_disks && !mddev->external)
5787 && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY
5788 && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE
5789 && cmd != GET_BITMAP_FILE) {
5795 * Commands even a read-only array can execute:
5799 case GET_ARRAY_INFO:
5800 err = get_array_info(mddev, argp);
5803 case GET_BITMAP_FILE:
5804 err = get_bitmap_file(mddev, argp);
5808 err = get_disk_info(mddev, argp);
5811 case RESTART_ARRAY_RW:
5812 err = restart_array(mddev);
5816 err = do_md_stop(mddev, 0, 1);
5820 err = md_set_readonly(mddev, 1);
5824 if (get_user(ro, (int __user *)(arg))) {
5830 /* if the bdev is going readonly the value of mddev->ro
5831 * does not matter, no writes are coming
5836 /* are we are already prepared for writes? */
5840 /* transitioning to readauto need only happen for
5841 * arrays that call md_write_start
5844 err = restart_array(mddev);
5847 set_disk_ro(mddev->gendisk, 0);
5854 * The remaining ioctls are changing the state of the
5855 * superblock, so we do not allow them on read-only arrays.
5856 * However non-MD ioctls (e.g. get-size) will still come through
5857 * here and hit the 'default' below, so only disallow
5858 * 'md' ioctls, and switch to rw mode if started auto-readonly.
5860 if (_IOC_TYPE(cmd) == MD_MAJOR && mddev->ro && mddev->pers) {
5861 if (mddev->ro == 2) {
5863 sysfs_notify_dirent_safe(mddev->sysfs_state);
5864 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5865 md_wakeup_thread(mddev->thread);
5876 mdu_disk_info_t info;
5877 if (copy_from_user(&info, argp, sizeof(info)))
5880 err = add_new_disk(mddev, &info);
5884 case HOT_REMOVE_DISK:
5885 err = hot_remove_disk(mddev, new_decode_dev(arg));
5889 err = hot_add_disk(mddev, new_decode_dev(arg));
5892 case SET_DISK_FAULTY:
5893 err = set_disk_faulty(mddev, new_decode_dev(arg));
5897 err = do_md_run(mddev);
5900 case SET_BITMAP_FILE:
5901 err = set_bitmap_file(mddev, (int)arg);
5911 if (mddev->hold_active == UNTIL_IOCTL &&
5913 mddev->hold_active = 0;
5914 mddev_unlock(mddev);
5923 #ifdef CONFIG_COMPAT
5924 static int md_compat_ioctl(struct block_device *bdev, fmode_t mode,
5925 unsigned int cmd, unsigned long arg)
5928 case HOT_REMOVE_DISK:
5930 case SET_DISK_FAULTY:
5931 case SET_BITMAP_FILE:
5932 /* These take in integer arg, do not convert */
5935 arg = (unsigned long)compat_ptr(arg);
5939 return md_ioctl(bdev, mode, cmd, arg);
5941 #endif /* CONFIG_COMPAT */
5943 static int md_open(struct block_device *bdev, fmode_t mode)
5946 * Succeed if we can lock the mddev, which confirms that
5947 * it isn't being stopped right now.
5949 mddev_t *mddev = mddev_find(bdev->bd_dev);
5953 if (mddev->gendisk != bdev->bd_disk) {
5954 /* we are racing with mddev_put which is discarding this
5958 /* Wait until bdev->bd_disk is definitely gone */
5959 flush_scheduled_work();
5960 /* Then retry the open from the top */
5962 return -ERESTARTSYS;
5964 BUG_ON(mddev != bdev->bd_disk->private_data);
5966 if ((err = mutex_lock_interruptible(&mddev->open_mutex)))
5970 atomic_inc(&mddev->openers);
5971 mutex_unlock(&mddev->open_mutex);
5973 check_disk_size_change(mddev->gendisk, bdev);
5979 static int md_release(struct gendisk *disk, fmode_t mode)
5981 mddev_t *mddev = disk->private_data;
5985 atomic_dec(&mddev->openers);
5991 static const struct block_device_operations md_fops =
5993 .owner = THIS_MODULE,
5995 .release = md_release,
5997 #ifdef CONFIG_COMPAT
5998 .compat_ioctl = md_compat_ioctl,
6000 .getgeo = md_getgeo,
6003 static int md_thread(void * arg)
6005 mdk_thread_t *thread = arg;
6008 * md_thread is a 'system-thread', it's priority should be very
6009 * high. We avoid resource deadlocks individually in each
6010 * raid personality. (RAID5 does preallocation) We also use RR and
6011 * the very same RT priority as kswapd, thus we will never get
6012 * into a priority inversion deadlock.
6014 * we definitely have to have equal or higher priority than
6015 * bdflush, otherwise bdflush will deadlock if there are too
6016 * many dirty RAID5 blocks.
6019 allow_signal(SIGKILL);
6020 while (!kthread_should_stop()) {
6022 /* We need to wait INTERRUPTIBLE so that
6023 * we don't add to the load-average.
6024 * That means we need to be sure no signals are
6027 if (signal_pending(current))
6028 flush_signals(current);
6030 wait_event_interruptible_timeout
6032 test_bit(THREAD_WAKEUP, &thread->flags)
6033 || kthread_should_stop(),
6036 clear_bit(THREAD_WAKEUP, &thread->flags);
6038 thread->run(thread->mddev);
6044 void md_wakeup_thread(mdk_thread_t *thread)
6047 dprintk("md: waking up MD thread %s.\n", thread->tsk->comm);
6048 set_bit(THREAD_WAKEUP, &thread->flags);
6049 wake_up(&thread->wqueue);
6053 mdk_thread_t *md_register_thread(void (*run) (mddev_t *), mddev_t *mddev,
6056 mdk_thread_t *thread;
6058 thread = kzalloc(sizeof(mdk_thread_t), GFP_KERNEL);
6062 init_waitqueue_head(&thread->wqueue);
6065 thread->mddev = mddev;
6066 thread->timeout = MAX_SCHEDULE_TIMEOUT;
6067 thread->tsk = kthread_run(md_thread, thread,
6069 mdname(thread->mddev),
6070 name ?: mddev->pers->name);
6071 if (IS_ERR(thread->tsk)) {
6078 void md_unregister_thread(mdk_thread_t *thread)
6082 dprintk("interrupting MD-thread pid %d\n", task_pid_nr(thread->tsk));
6084 kthread_stop(thread->tsk);
6088 void md_error(mddev_t *mddev, mdk_rdev_t *rdev)
6095 if (!rdev || test_bit(Faulty, &rdev->flags))
6098 if (mddev->external)
6099 set_bit(Blocked, &rdev->flags);
6101 dprintk("md_error dev:%s, rdev:(%d:%d), (caller: %p,%p,%p,%p).\n",
6103 MAJOR(rdev->bdev->bd_dev), MINOR(rdev->bdev->bd_dev),
6104 __builtin_return_address(0),__builtin_return_address(1),
6105 __builtin_return_address(2),__builtin_return_address(3));
6109 if (!mddev->pers->error_handler)
6111 mddev->pers->error_handler(mddev,rdev);
6112 if (mddev->degraded)
6113 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
6114 sysfs_notify_dirent_safe(rdev->sysfs_state);
6115 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6116 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6117 md_wakeup_thread(mddev->thread);
6118 if (mddev->event_work.func)
6119 schedule_work(&mddev->event_work);
6120 md_new_event_inintr(mddev);
6123 /* seq_file implementation /proc/mdstat */
6125 static void status_unused(struct seq_file *seq)
6130 seq_printf(seq, "unused devices: ");
6132 list_for_each_entry(rdev, &pending_raid_disks, same_set) {
6133 char b[BDEVNAME_SIZE];
6135 seq_printf(seq, "%s ",
6136 bdevname(rdev->bdev,b));
6139 seq_printf(seq, "<none>");
6141 seq_printf(seq, "\n");
6145 static void status_resync(struct seq_file *seq, mddev_t * mddev)
6147 sector_t max_sectors, resync, res;
6148 unsigned long dt, db;
6151 unsigned int per_milli;
6153 resync = mddev->curr_resync - atomic_read(&mddev->recovery_active);
6155 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
6156 max_sectors = mddev->resync_max_sectors;
6158 max_sectors = mddev->dev_sectors;
6161 * Should not happen.
6167 /* Pick 'scale' such that (resync>>scale)*1000 will fit
6168 * in a sector_t, and (max_sectors>>scale) will fit in a
6169 * u32, as those are the requirements for sector_div.
6170 * Thus 'scale' must be at least 10
6173 if (sizeof(sector_t) > sizeof(unsigned long)) {
6174 while ( max_sectors/2 > (1ULL<<(scale+32)))
6177 res = (resync>>scale)*1000;
6178 sector_div(res, (u32)((max_sectors>>scale)+1));
6182 int i, x = per_milli/50, y = 20-x;
6183 seq_printf(seq, "[");
6184 for (i = 0; i < x; i++)
6185 seq_printf(seq, "=");
6186 seq_printf(seq, ">");
6187 for (i = 0; i < y; i++)
6188 seq_printf(seq, ".");
6189 seq_printf(seq, "] ");
6191 seq_printf(seq, " %s =%3u.%u%% (%llu/%llu)",
6192 (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)?
6194 (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)?
6196 (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
6197 "resync" : "recovery"))),
6198 per_milli/10, per_milli % 10,
6199 (unsigned long long) resync/2,
6200 (unsigned long long) max_sectors/2);
6203 * dt: time from mark until now
6204 * db: blocks written from mark until now
6205 * rt: remaining time
6207 * rt is a sector_t, so could be 32bit or 64bit.
6208 * So we divide before multiply in case it is 32bit and close
6210 * We scale the divisor (db) by 32 to avoid loosing precision
6211 * near the end of resync when the number of remaining sectors
6213 * We then divide rt by 32 after multiplying by db to compensate.
6214 * The '+1' avoids division by zero if db is very small.
6216 dt = ((jiffies - mddev->resync_mark) / HZ);
6218 db = (mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active))
6219 - mddev->resync_mark_cnt;
6221 rt = max_sectors - resync; /* number of remaining sectors */
6222 sector_div(rt, db/32+1);
6226 seq_printf(seq, " finish=%lu.%lumin", (unsigned long)rt / 60,
6227 ((unsigned long)rt % 60)/6);
6229 seq_printf(seq, " speed=%ldK/sec", db/2/dt);
6232 static void *md_seq_start(struct seq_file *seq, loff_t *pos)
6234 struct list_head *tmp;
6244 spin_lock(&all_mddevs_lock);
6245 list_for_each(tmp,&all_mddevs)
6247 mddev = list_entry(tmp, mddev_t, all_mddevs);
6249 spin_unlock(&all_mddevs_lock);
6252 spin_unlock(&all_mddevs_lock);
6254 return (void*)2;/* tail */
6258 static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
6260 struct list_head *tmp;
6261 mddev_t *next_mddev, *mddev = v;
6267 spin_lock(&all_mddevs_lock);
6269 tmp = all_mddevs.next;
6271 tmp = mddev->all_mddevs.next;
6272 if (tmp != &all_mddevs)
6273 next_mddev = mddev_get(list_entry(tmp,mddev_t,all_mddevs));
6275 next_mddev = (void*)2;
6278 spin_unlock(&all_mddevs_lock);
6286 static void md_seq_stop(struct seq_file *seq, void *v)
6290 if (mddev && v != (void*)1 && v != (void*)2)
6294 struct mdstat_info {
6298 static int md_seq_show(struct seq_file *seq, void *v)
6303 struct mdstat_info *mi = seq->private;
6304 struct bitmap *bitmap;
6306 if (v == (void*)1) {
6307 struct mdk_personality *pers;
6308 seq_printf(seq, "Personalities : ");
6309 spin_lock(&pers_lock);
6310 list_for_each_entry(pers, &pers_list, list)
6311 seq_printf(seq, "[%s] ", pers->name);
6313 spin_unlock(&pers_lock);
6314 seq_printf(seq, "\n");
6315 mi->event = atomic_read(&md_event_count);
6318 if (v == (void*)2) {
6323 if (mddev_lock(mddev) < 0)
6326 if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
6327 seq_printf(seq, "%s : %sactive", mdname(mddev),
6328 mddev->pers ? "" : "in");
6331 seq_printf(seq, " (read-only)");
6333 seq_printf(seq, " (auto-read-only)");
6334 seq_printf(seq, " %s", mddev->pers->name);
6338 list_for_each_entry(rdev, &mddev->disks, same_set) {
6339 char b[BDEVNAME_SIZE];
6340 seq_printf(seq, " %s[%d]",
6341 bdevname(rdev->bdev,b), rdev->desc_nr);
6342 if (test_bit(WriteMostly, &rdev->flags))
6343 seq_printf(seq, "(W)");
6344 if (test_bit(Faulty, &rdev->flags)) {
6345 seq_printf(seq, "(F)");
6347 } else if (rdev->raid_disk < 0)
6348 seq_printf(seq, "(S)"); /* spare */
6349 sectors += rdev->sectors;
6352 if (!list_empty(&mddev->disks)) {
6354 seq_printf(seq, "\n %llu blocks",
6355 (unsigned long long)
6356 mddev->array_sectors / 2);
6358 seq_printf(seq, "\n %llu blocks",
6359 (unsigned long long)sectors / 2);
6361 if (mddev->persistent) {
6362 if (mddev->major_version != 0 ||
6363 mddev->minor_version != 90) {
6364 seq_printf(seq," super %d.%d",
6365 mddev->major_version,
6366 mddev->minor_version);
6368 } else if (mddev->external)
6369 seq_printf(seq, " super external:%s",
6370 mddev->metadata_type);
6372 seq_printf(seq, " super non-persistent");
6375 mddev->pers->status(seq, mddev);
6376 seq_printf(seq, "\n ");
6377 if (mddev->pers->sync_request) {
6378 if (mddev->curr_resync > 2) {
6379 status_resync(seq, mddev);
6380 seq_printf(seq, "\n ");
6381 } else if (mddev->curr_resync == 1 || mddev->curr_resync == 2)
6382 seq_printf(seq, "\tresync=DELAYED\n ");
6383 else if (mddev->recovery_cp < MaxSector)
6384 seq_printf(seq, "\tresync=PENDING\n ");
6387 seq_printf(seq, "\n ");
6389 if ((bitmap = mddev->bitmap)) {
6390 unsigned long chunk_kb;
6391 unsigned long flags;
6392 spin_lock_irqsave(&bitmap->lock, flags);
6393 chunk_kb = mddev->bitmap_info.chunksize >> 10;
6394 seq_printf(seq, "bitmap: %lu/%lu pages [%luKB], "
6396 bitmap->pages - bitmap->missing_pages,
6398 (bitmap->pages - bitmap->missing_pages)
6399 << (PAGE_SHIFT - 10),
6400 chunk_kb ? chunk_kb : mddev->bitmap_info.chunksize,
6401 chunk_kb ? "KB" : "B");
6403 seq_printf(seq, ", file: ");
6404 seq_path(seq, &bitmap->file->f_path, " \t\n");
6407 seq_printf(seq, "\n");
6408 spin_unlock_irqrestore(&bitmap->lock, flags);
6411 seq_printf(seq, "\n");
6413 mddev_unlock(mddev);
6418 static const struct seq_operations md_seq_ops = {
6419 .start = md_seq_start,
6420 .next = md_seq_next,
6421 .stop = md_seq_stop,
6422 .show = md_seq_show,
6425 static int md_seq_open(struct inode *inode, struct file *file)
6428 struct mdstat_info *mi = kmalloc(sizeof(*mi), GFP_KERNEL);
6432 error = seq_open(file, &md_seq_ops);
6436 struct seq_file *p = file->private_data;
6438 mi->event = atomic_read(&md_event_count);
6443 static unsigned int mdstat_poll(struct file *filp, poll_table *wait)
6445 struct seq_file *m = filp->private_data;
6446 struct mdstat_info *mi = m->private;
6449 poll_wait(filp, &md_event_waiters, wait);
6451 /* always allow read */
6452 mask = POLLIN | POLLRDNORM;
6454 if (mi->event != atomic_read(&md_event_count))
6455 mask |= POLLERR | POLLPRI;
6459 static const struct file_operations md_seq_fops = {
6460 .owner = THIS_MODULE,
6461 .open = md_seq_open,
6463 .llseek = seq_lseek,
6464 .release = seq_release_private,
6465 .poll = mdstat_poll,
6468 int register_md_personality(struct mdk_personality *p)
6470 spin_lock(&pers_lock);
6471 list_add_tail(&p->list, &pers_list);
6472 printk(KERN_INFO "md: %s personality registered for level %d\n", p->name, p->level);
6473 spin_unlock(&pers_lock);
6477 int unregister_md_personality(struct mdk_personality *p)
6479 printk(KERN_INFO "md: %s personality unregistered\n", p->name);
6480 spin_lock(&pers_lock);
6481 list_del_init(&p->list);
6482 spin_unlock(&pers_lock);
6486 static int is_mddev_idle(mddev_t *mddev, int init)
6494 rdev_for_each_rcu(rdev, mddev) {
6495 struct gendisk *disk = rdev->bdev->bd_contains->bd_disk;
6496 curr_events = (int)part_stat_read(&disk->part0, sectors[0]) +
6497 (int)part_stat_read(&disk->part0, sectors[1]) -
6498 atomic_read(&disk->sync_io);
6499 /* sync IO will cause sync_io to increase before the disk_stats
6500 * as sync_io is counted when a request starts, and
6501 * disk_stats is counted when it completes.
6502 * So resync activity will cause curr_events to be smaller than
6503 * when there was no such activity.
6504 * non-sync IO will cause disk_stat to increase without
6505 * increasing sync_io so curr_events will (eventually)
6506 * be larger than it was before. Once it becomes
6507 * substantially larger, the test below will cause
6508 * the array to appear non-idle, and resync will slow
6510 * If there is a lot of outstanding resync activity when
6511 * we set last_event to curr_events, then all that activity
6512 * completing might cause the array to appear non-idle
6513 * and resync will be slowed down even though there might
6514 * not have been non-resync activity. This will only
6515 * happen once though. 'last_events' will soon reflect
6516 * the state where there is little or no outstanding
6517 * resync requests, and further resync activity will
6518 * always make curr_events less than last_events.
6521 if (init || curr_events - rdev->last_events > 64) {
6522 rdev->last_events = curr_events;
6530 void md_done_sync(mddev_t *mddev, int blocks, int ok)
6532 /* another "blocks" (512byte) blocks have been synced */
6533 atomic_sub(blocks, &mddev->recovery_active);
6534 wake_up(&mddev->recovery_wait);
6536 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6537 md_wakeup_thread(mddev->thread);
6538 // stop recovery, signal do_sync ....
6543 /* md_write_start(mddev, bi)
6544 * If we need to update some array metadata (e.g. 'active' flag
6545 * in superblock) before writing, schedule a superblock update
6546 * and wait for it to complete.
6548 void md_write_start(mddev_t *mddev, struct bio *bi)
6551 if (bio_data_dir(bi) != WRITE)
6554 BUG_ON(mddev->ro == 1);
6555 if (mddev->ro == 2) {
6556 /* need to switch to read/write */
6558 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6559 md_wakeup_thread(mddev->thread);
6560 md_wakeup_thread(mddev->sync_thread);
6563 atomic_inc(&mddev->writes_pending);
6564 if (mddev->safemode == 1)
6565 mddev->safemode = 0;
6566 if (mddev->in_sync) {
6567 spin_lock_irq(&mddev->write_lock);
6568 if (mddev->in_sync) {
6570 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
6571 set_bit(MD_CHANGE_PENDING, &mddev->flags);
6572 md_wakeup_thread(mddev->thread);
6575 spin_unlock_irq(&mddev->write_lock);
6578 sysfs_notify_dirent_safe(mddev->sysfs_state);
6579 wait_event(mddev->sb_wait,
6580 !test_bit(MD_CHANGE_PENDING, &mddev->flags));
6583 void md_write_end(mddev_t *mddev)
6585 if (atomic_dec_and_test(&mddev->writes_pending)) {
6586 if (mddev->safemode == 2)
6587 md_wakeup_thread(mddev->thread);
6588 else if (mddev->safemode_delay)
6589 mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay);
6593 /* md_allow_write(mddev)
6594 * Calling this ensures that the array is marked 'active' so that writes
6595 * may proceed without blocking. It is important to call this before
6596 * attempting a GFP_KERNEL allocation while holding the mddev lock.
6597 * Must be called with mddev_lock held.
6599 * In the ->external case MD_CHANGE_CLEAN can not be cleared until mddev->lock
6600 * is dropped, so return -EAGAIN after notifying userspace.
6602 int md_allow_write(mddev_t *mddev)
6608 if (!mddev->pers->sync_request)
6611 spin_lock_irq(&mddev->write_lock);
6612 if (mddev->in_sync) {
6614 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
6615 set_bit(MD_CHANGE_PENDING, &mddev->flags);
6616 if (mddev->safemode_delay &&
6617 mddev->safemode == 0)
6618 mddev->safemode = 1;
6619 spin_unlock_irq(&mddev->write_lock);
6620 md_update_sb(mddev, 0);
6621 sysfs_notify_dirent_safe(mddev->sysfs_state);
6623 spin_unlock_irq(&mddev->write_lock);
6625 if (test_bit(MD_CHANGE_PENDING, &mddev->flags))
6630 EXPORT_SYMBOL_GPL(md_allow_write);
6632 void md_unplug(mddev_t *mddev)
6635 blk_unplug(mddev->queue);
6637 mddev->plug->unplug_fn(mddev->plug);
6640 #define SYNC_MARKS 10
6641 #define SYNC_MARK_STEP (3*HZ)
6642 void md_do_sync(mddev_t *mddev)
6645 unsigned int currspeed = 0,
6647 sector_t max_sectors,j, io_sectors;
6648 unsigned long mark[SYNC_MARKS];
6649 sector_t mark_cnt[SYNC_MARKS];
6651 struct list_head *tmp;
6652 sector_t last_check;
6657 /* just incase thread restarts... */
6658 if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
6660 if (mddev->ro) /* never try to sync a read-only array */
6663 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
6664 if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
6665 desc = "data-check";
6666 else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
6667 desc = "requested-resync";
6670 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
6675 /* we overload curr_resync somewhat here.
6676 * 0 == not engaged in resync at all
6677 * 2 == checking that there is no conflict with another sync
6678 * 1 == like 2, but have yielded to allow conflicting resync to
6680 * other == active in resync - this many blocks
6682 * Before starting a resync we must have set curr_resync to
6683 * 2, and then checked that every "conflicting" array has curr_resync
6684 * less than ours. When we find one that is the same or higher
6685 * we wait on resync_wait. To avoid deadlock, we reduce curr_resync
6686 * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
6687 * This will mean we have to start checking from the beginning again.
6692 mddev->curr_resync = 2;
6695 if (kthread_should_stop())
6696 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6698 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
6700 for_each_mddev(mddev2, tmp) {
6701 if (mddev2 == mddev)
6703 if (!mddev->parallel_resync
6704 && mddev2->curr_resync
6705 && match_mddev_units(mddev, mddev2)) {
6707 if (mddev < mddev2 && mddev->curr_resync == 2) {
6708 /* arbitrarily yield */
6709 mddev->curr_resync = 1;
6710 wake_up(&resync_wait);
6712 if (mddev > mddev2 && mddev->curr_resync == 1)
6713 /* no need to wait here, we can wait the next
6714 * time 'round when curr_resync == 2
6717 /* We need to wait 'interruptible' so as not to
6718 * contribute to the load average, and not to
6719 * be caught by 'softlockup'
6721 prepare_to_wait(&resync_wait, &wq, TASK_INTERRUPTIBLE);
6722 if (!kthread_should_stop() &&
6723 mddev2->curr_resync >= mddev->curr_resync) {
6724 printk(KERN_INFO "md: delaying %s of %s"
6725 " until %s has finished (they"
6726 " share one or more physical units)\n",
6727 desc, mdname(mddev), mdname(mddev2));
6729 if (signal_pending(current))
6730 flush_signals(current);
6732 finish_wait(&resync_wait, &wq);
6735 finish_wait(&resync_wait, &wq);
6738 } while (mddev->curr_resync < 2);
6741 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
6742 /* resync follows the size requested by the personality,
6743 * which defaults to physical size, but can be virtual size
6745 max_sectors = mddev->resync_max_sectors;
6746 mddev->resync_mismatches = 0;
6747 /* we don't use the checkpoint if there's a bitmap */
6748 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
6749 j = mddev->resync_min;
6750 else if (!mddev->bitmap)
6751 j = mddev->recovery_cp;
6753 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
6754 max_sectors = mddev->dev_sectors;
6756 /* recovery follows the physical size of devices */
6757 max_sectors = mddev->dev_sectors;
6760 list_for_each_entry_rcu(rdev, &mddev->disks, same_set)
6761 if (rdev->raid_disk >= 0 &&
6762 !test_bit(Faulty, &rdev->flags) &&
6763 !test_bit(In_sync, &rdev->flags) &&
6764 rdev->recovery_offset < j)
6765 j = rdev->recovery_offset;
6769 printk(KERN_INFO "md: %s of RAID array %s\n", desc, mdname(mddev));
6770 printk(KERN_INFO "md: minimum _guaranteed_ speed:"
6771 " %d KB/sec/disk.\n", speed_min(mddev));
6772 printk(KERN_INFO "md: using maximum available idle IO bandwidth "
6773 "(but not more than %d KB/sec) for %s.\n",
6774 speed_max(mddev), desc);
6776 is_mddev_idle(mddev, 1); /* this initializes IO event counters */
6779 for (m = 0; m < SYNC_MARKS; m++) {
6781 mark_cnt[m] = io_sectors;
6784 mddev->resync_mark = mark[last_mark];
6785 mddev->resync_mark_cnt = mark_cnt[last_mark];
6788 * Tune reconstruction:
6790 window = 32*(PAGE_SIZE/512);
6791 printk(KERN_INFO "md: using %dk window, over a total of %llu blocks.\n",
6792 window/2,(unsigned long long) max_sectors/2);
6794 atomic_set(&mddev->recovery_active, 0);
6799 "md: resuming %s of %s from checkpoint.\n",
6800 desc, mdname(mddev));
6801 mddev->curr_resync = j;
6803 mddev->curr_resync_completed = mddev->curr_resync;
6805 while (j < max_sectors) {
6810 if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
6811 ((mddev->curr_resync > mddev->curr_resync_completed &&
6812 (mddev->curr_resync - mddev->curr_resync_completed)
6813 > (max_sectors >> 4)) ||
6814 (j - mddev->curr_resync_completed)*2
6815 >= mddev->resync_max - mddev->curr_resync_completed
6817 /* time to update curr_resync_completed */
6819 wait_event(mddev->recovery_wait,
6820 atomic_read(&mddev->recovery_active) == 0);
6821 mddev->curr_resync_completed =
6823 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
6824 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
6827 while (j >= mddev->resync_max && !kthread_should_stop()) {
6828 /* As this condition is controlled by user-space,
6829 * we can block indefinitely, so use '_interruptible'
6830 * to avoid triggering warnings.
6832 flush_signals(current); /* just in case */
6833 wait_event_interruptible(mddev->recovery_wait,
6834 mddev->resync_max > j
6835 || kthread_should_stop());
6838 if (kthread_should_stop())
6841 sectors = mddev->pers->sync_request(mddev, j, &skipped,
6842 currspeed < speed_min(mddev));
6844 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6848 if (!skipped) { /* actual IO requested */
6849 io_sectors += sectors;
6850 atomic_add(sectors, &mddev->recovery_active);
6854 if (j>1) mddev->curr_resync = j;
6855 mddev->curr_mark_cnt = io_sectors;
6856 if (last_check == 0)
6857 /* this is the earliers that rebuilt will be
6858 * visible in /proc/mdstat
6860 md_new_event(mddev);
6862 if (last_check + window > io_sectors || j == max_sectors)
6865 last_check = io_sectors;
6867 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
6871 if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) {
6873 int next = (last_mark+1) % SYNC_MARKS;
6875 mddev->resync_mark = mark[next];
6876 mddev->resync_mark_cnt = mark_cnt[next];
6877 mark[next] = jiffies;
6878 mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active);
6883 if (kthread_should_stop())
6888 * this loop exits only if either when we are slower than
6889 * the 'hard' speed limit, or the system was IO-idle for
6891 * the system might be non-idle CPU-wise, but we only care
6892 * about not overloading the IO subsystem. (things like an
6893 * e2fsck being done on the RAID array should execute fast)
6898 currspeed = ((unsigned long)(io_sectors-mddev->resync_mark_cnt))/2
6899 /((jiffies-mddev->resync_mark)/HZ +1) +1;
6901 if (currspeed > speed_min(mddev)) {
6902 if ((currspeed > speed_max(mddev)) ||
6903 !is_mddev_idle(mddev, 0)) {
6909 printk(KERN_INFO "md: %s: %s done.\n",mdname(mddev), desc);
6911 * this also signals 'finished resyncing' to md_stop
6916 wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
6918 /* tell personality that we are finished */
6919 mddev->pers->sync_request(mddev, max_sectors, &skipped, 1);
6921 if (!test_bit(MD_RECOVERY_CHECK, &mddev->recovery) &&
6922 mddev->curr_resync > 2) {
6923 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
6924 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
6925 if (mddev->curr_resync >= mddev->recovery_cp) {
6927 "md: checkpointing %s of %s.\n",
6928 desc, mdname(mddev));
6929 mddev->recovery_cp = mddev->curr_resync;
6932 mddev->recovery_cp = MaxSector;
6934 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery))
6935 mddev->curr_resync = MaxSector;
6937 list_for_each_entry_rcu(rdev, &mddev->disks, same_set)
6938 if (rdev->raid_disk >= 0 &&
6939 mddev->delta_disks >= 0 &&
6940 !test_bit(Faulty, &rdev->flags) &&
6941 !test_bit(In_sync, &rdev->flags) &&
6942 rdev->recovery_offset < mddev->curr_resync)
6943 rdev->recovery_offset = mddev->curr_resync;
6947 set_bit(MD_CHANGE_DEVS, &mddev->flags);
6950 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
6951 /* We completed so min/max setting can be forgotten if used. */
6952 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
6953 mddev->resync_min = 0;
6954 mddev->resync_max = MaxSector;
6955 } else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
6956 mddev->resync_min = mddev->curr_resync_completed;
6957 mddev->curr_resync = 0;
6958 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery))
6959 mddev->curr_resync_completed = 0;
6960 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
6961 wake_up(&resync_wait);
6962 set_bit(MD_RECOVERY_DONE, &mddev->recovery);
6963 md_wakeup_thread(mddev->thread);
6968 * got a signal, exit.
6971 "md: md_do_sync() got signal ... exiting\n");
6972 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6976 EXPORT_SYMBOL_GPL(md_do_sync);
6979 static int remove_and_add_spares(mddev_t *mddev)
6984 mddev->curr_resync_completed = 0;
6986 list_for_each_entry(rdev, &mddev->disks, same_set)
6987 if (rdev->raid_disk >= 0 &&
6988 !test_bit(Blocked, &rdev->flags) &&
6989 (test_bit(Faulty, &rdev->flags) ||
6990 ! test_bit(In_sync, &rdev->flags)) &&
6991 atomic_read(&rdev->nr_pending)==0) {
6992 if (mddev->pers->hot_remove_disk(
6993 mddev, rdev->raid_disk)==0) {
6995 sprintf(nm,"rd%d", rdev->raid_disk);
6996 sysfs_remove_link(&mddev->kobj, nm);
6997 rdev->raid_disk = -1;
7001 if (mddev->degraded && ! mddev->ro && !mddev->recovery_disabled) {
7002 list_for_each_entry(rdev, &mddev->disks, same_set) {
7003 if (rdev->raid_disk >= 0 &&
7004 !test_bit(In_sync, &rdev->flags) &&
7005 !test_bit(Blocked, &rdev->flags))
7007 if (rdev->raid_disk < 0
7008 && !test_bit(Faulty, &rdev->flags)) {
7009 rdev->recovery_offset = 0;
7011 hot_add_disk(mddev, rdev) == 0) {
7013 sprintf(nm, "rd%d", rdev->raid_disk);
7014 if (sysfs_create_link(&mddev->kobj,
7016 /* failure here is OK */;
7018 md_new_event(mddev);
7019 set_bit(MD_CHANGE_DEVS, &mddev->flags);
7028 * This routine is regularly called by all per-raid-array threads to
7029 * deal with generic issues like resync and super-block update.
7030 * Raid personalities that don't have a thread (linear/raid0) do not
7031 * need this as they never do any recovery or update the superblock.
7033 * It does not do any resync itself, but rather "forks" off other threads
7034 * to do that as needed.
7035 * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
7036 * "->recovery" and create a thread at ->sync_thread.
7037 * When the thread finishes it sets MD_RECOVERY_DONE
7038 * and wakeups up this thread which will reap the thread and finish up.
7039 * This thread also removes any faulty devices (with nr_pending == 0).
7041 * The overall approach is:
7042 * 1/ if the superblock needs updating, update it.
7043 * 2/ If a recovery thread is running, don't do anything else.
7044 * 3/ If recovery has finished, clean up, possibly marking spares active.
7045 * 4/ If there are any faulty devices, remove them.
7046 * 5/ If array is degraded, try to add spares devices
7047 * 6/ If array has spares or is not in-sync, start a resync thread.
7049 void md_check_recovery(mddev_t *mddev)
7055 bitmap_daemon_work(mddev);
7060 if (signal_pending(current)) {
7061 if (mddev->pers->sync_request && !mddev->external) {
7062 printk(KERN_INFO "md: %s in immediate safe mode\n",
7064 mddev->safemode = 2;
7066 flush_signals(current);
7069 if (mddev->ro && !test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
7072 (mddev->flags && !mddev->external) ||
7073 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
7074 test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
7075 (mddev->external == 0 && mddev->safemode == 1) ||
7076 (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending)
7077 && !mddev->in_sync && mddev->recovery_cp == MaxSector)
7081 if (mddev_trylock(mddev)) {
7085 /* Only thing we do on a ro array is remove
7088 remove_and_add_spares(mddev);
7089 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7093 if (!mddev->external) {
7095 spin_lock_irq(&mddev->write_lock);
7096 if (mddev->safemode &&
7097 !atomic_read(&mddev->writes_pending) &&
7099 mddev->recovery_cp == MaxSector) {
7102 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
7104 if (mddev->safemode == 1)
7105 mddev->safemode = 0;
7106 spin_unlock_irq(&mddev->write_lock);
7108 sysfs_notify_dirent_safe(mddev->sysfs_state);
7112 md_update_sb(mddev, 0);
7114 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
7115 !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
7116 /* resync/recovery still happening */
7117 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7120 if (mddev->sync_thread) {
7121 /* resync has finished, collect result */
7122 md_unregister_thread(mddev->sync_thread);
7123 mddev->sync_thread = NULL;
7124 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
7125 !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
7127 /* activate any spares */
7128 if (mddev->pers->spare_active(mddev))
7129 sysfs_notify(&mddev->kobj, NULL,
7132 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
7133 mddev->pers->finish_reshape)
7134 mddev->pers->finish_reshape(mddev);
7135 md_update_sb(mddev, 1);
7137 /* if array is no-longer degraded, then any saved_raid_disk
7138 * information must be scrapped
7140 if (!mddev->degraded)
7141 list_for_each_entry(rdev, &mddev->disks, same_set)
7142 rdev->saved_raid_disk = -1;
7144 mddev->recovery = 0;
7145 /* flag recovery needed just to double check */
7146 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7147 sysfs_notify_dirent_safe(mddev->sysfs_action);
7148 md_new_event(mddev);
7151 /* Set RUNNING before clearing NEEDED to avoid
7152 * any transients in the value of "sync_action".
7154 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
7155 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7156 /* Clear some bits that don't mean anything, but
7159 clear_bit(MD_RECOVERY_INTR, &mddev->recovery);
7160 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
7162 if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
7164 /* no recovery is running.
7165 * remove any failed drives, then
7166 * add spares if possible.
7167 * Spare are also removed and re-added, to allow
7168 * the personality to fail the re-add.
7171 if (mddev->reshape_position != MaxSector) {
7172 if (mddev->pers->check_reshape == NULL ||
7173 mddev->pers->check_reshape(mddev) != 0)
7174 /* Cannot proceed */
7176 set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
7177 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
7178 } else if ((spares = remove_and_add_spares(mddev))) {
7179 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
7180 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
7181 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
7182 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
7183 } else if (mddev->recovery_cp < MaxSector) {
7184 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
7185 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
7186 } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
7187 /* nothing to be done ... */
7190 if (mddev->pers->sync_request) {
7191 if (spares && mddev->bitmap && ! mddev->bitmap->file) {
7192 /* We are adding a device or devices to an array
7193 * which has the bitmap stored on all devices.
7194 * So make sure all bitmap pages get written
7196 bitmap_write_all(mddev->bitmap);
7198 mddev->sync_thread = md_register_thread(md_do_sync,
7201 if (!mddev->sync_thread) {
7202 printk(KERN_ERR "%s: could not start resync"
7205 /* leave the spares where they are, it shouldn't hurt */
7206 mddev->recovery = 0;
7208 md_wakeup_thread(mddev->sync_thread);
7209 sysfs_notify_dirent_safe(mddev->sysfs_action);
7210 md_new_event(mddev);
7213 if (!mddev->sync_thread) {
7214 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
7215 if (test_and_clear_bit(MD_RECOVERY_RECOVER,
7217 if (mddev->sysfs_action)
7218 sysfs_notify_dirent_safe(mddev->sysfs_action);
7220 mddev_unlock(mddev);
7224 void md_wait_for_blocked_rdev(mdk_rdev_t *rdev, mddev_t *mddev)
7226 sysfs_notify_dirent_safe(rdev->sysfs_state);
7227 wait_event_timeout(rdev->blocked_wait,
7228 !test_bit(Blocked, &rdev->flags),
7229 msecs_to_jiffies(5000));
7230 rdev_dec_pending(rdev, mddev);
7232 EXPORT_SYMBOL(md_wait_for_blocked_rdev);
7234 static int md_notify_reboot(struct notifier_block *this,
7235 unsigned long code, void *x)
7237 struct list_head *tmp;
7240 if ((code == SYS_DOWN) || (code == SYS_HALT) || (code == SYS_POWER_OFF)) {
7242 printk(KERN_INFO "md: stopping all md devices.\n");
7244 for_each_mddev(mddev, tmp)
7245 if (mddev_trylock(mddev)) {
7246 /* Force a switch to readonly even array
7247 * appears to still be in use. Hence
7250 md_set_readonly(mddev, 100);
7251 mddev_unlock(mddev);
7254 * certain more exotic SCSI devices are known to be
7255 * volatile wrt too early system reboots. While the
7256 * right place to handle this issue is the given
7257 * driver, we do want to have a safe RAID driver ...
7264 static struct notifier_block md_notifier = {
7265 .notifier_call = md_notify_reboot,
7267 .priority = INT_MAX, /* before any real devices */
7270 static void md_geninit(void)
7272 dprintk("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
7274 proc_create("mdstat", S_IRUGO, NULL, &md_seq_fops);
7277 static int __init md_init(void)
7279 if (register_blkdev(MD_MAJOR, "md"))
7281 if ((mdp_major=register_blkdev(0, "mdp"))<=0) {
7282 unregister_blkdev(MD_MAJOR, "md");
7285 blk_register_region(MKDEV(MD_MAJOR, 0), 1UL<<MINORBITS, THIS_MODULE,
7286 md_probe, NULL, NULL);
7287 blk_register_region(MKDEV(mdp_major, 0), 1UL<<MINORBITS, THIS_MODULE,
7288 md_probe, NULL, NULL);
7290 register_reboot_notifier(&md_notifier);
7291 raid_table_header = register_sysctl_table(raid_root_table);
7301 * Searches all registered partitions for autorun RAID arrays
7305 static LIST_HEAD(all_detected_devices);
7306 struct detected_devices_node {
7307 struct list_head list;
7311 void md_autodetect_dev(dev_t dev)
7313 struct detected_devices_node *node_detected_dev;
7315 node_detected_dev = kzalloc(sizeof(*node_detected_dev), GFP_KERNEL);
7316 if (node_detected_dev) {
7317 node_detected_dev->dev = dev;
7318 list_add_tail(&node_detected_dev->list, &all_detected_devices);
7320 printk(KERN_CRIT "md: md_autodetect_dev: kzalloc failed"
7321 ", skipping dev(%d,%d)\n", MAJOR(dev), MINOR(dev));
7326 static void autostart_arrays(int part)
7329 struct detected_devices_node *node_detected_dev;
7331 int i_scanned, i_passed;
7336 printk(KERN_INFO "md: Autodetecting RAID arrays.\n");
7338 while (!list_empty(&all_detected_devices) && i_scanned < INT_MAX) {
7340 node_detected_dev = list_entry(all_detected_devices.next,
7341 struct detected_devices_node, list);
7342 list_del(&node_detected_dev->list);
7343 dev = node_detected_dev->dev;
7344 kfree(node_detected_dev);
7345 rdev = md_import_device(dev,0, 90);
7349 if (test_bit(Faulty, &rdev->flags)) {
7353 set_bit(AutoDetected, &rdev->flags);
7354 list_add(&rdev->same_set, &pending_raid_disks);
7358 printk(KERN_INFO "md: Scanned %d and added %d devices.\n",
7359 i_scanned, i_passed);
7361 autorun_devices(part);
7364 #endif /* !MODULE */
7366 static __exit void md_exit(void)
7369 struct list_head *tmp;
7371 blk_unregister_region(MKDEV(MD_MAJOR,0), 1U << MINORBITS);
7372 blk_unregister_region(MKDEV(mdp_major,0), 1U << MINORBITS);
7374 unregister_blkdev(MD_MAJOR,"md");
7375 unregister_blkdev(mdp_major, "mdp");
7376 unregister_reboot_notifier(&md_notifier);
7377 unregister_sysctl_table(raid_table_header);
7378 remove_proc_entry("mdstat", NULL);
7379 for_each_mddev(mddev, tmp) {
7380 export_array(mddev);
7381 mddev->hold_active = 0;
7385 subsys_initcall(md_init);
7386 module_exit(md_exit)
7388 static int get_ro(char *buffer, struct kernel_param *kp)
7390 return sprintf(buffer, "%d", start_readonly);
7392 static int set_ro(const char *val, struct kernel_param *kp)
7395 int num = simple_strtoul(val, &e, 10);
7396 if (*val && (*e == '\0' || *e == '\n')) {
7397 start_readonly = num;
7403 module_param_call(start_ro, set_ro, get_ro, NULL, S_IRUSR|S_IWUSR);
7404 module_param(start_dirty_degraded, int, S_IRUGO|S_IWUSR);
7406 module_param_call(new_array, add_named_array, NULL, NULL, S_IWUSR);
7408 EXPORT_SYMBOL(register_md_personality);
7409 EXPORT_SYMBOL(unregister_md_personality);
7410 EXPORT_SYMBOL(md_error);
7411 EXPORT_SYMBOL(md_done_sync);
7412 EXPORT_SYMBOL(md_write_start);
7413 EXPORT_SYMBOL(md_write_end);
7414 EXPORT_SYMBOL(md_register_thread);
7415 EXPORT_SYMBOL(md_unregister_thread);
7416 EXPORT_SYMBOL(md_wakeup_thread);
7417 EXPORT_SYMBOL(md_check_recovery);
7418 MODULE_LICENSE("GPL");
7419 MODULE_DESCRIPTION("MD RAID framework");
7421 MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR);