2 md.c : Multiple Devices driver for Linux
3 Copyright (C) 1998, 1999, 2000 Ingo Molnar
5 completely rewritten, based on the MD driver code from Marc Zyngier
9 - RAID-1/RAID-5 extensions by Miguel de Icaza, Gadi Oxman, Ingo Molnar
10 - RAID-6 extensions by H. Peter Anvin <hpa@zytor.com>
11 - boot support for linear and striped mode by Harald Hoyer <HarryH@Royal.Net>
12 - kerneld support by Boris Tobotras <boris@xtalk.msk.su>
13 - kmod support by: Cyrus Durgin
14 - RAID0 bugfixes: Mark Anthony Lisher <markal@iname.com>
15 - Devfs support by Richard Gooch <rgooch@atnf.csiro.au>
17 - lots of fixes and improvements to the RAID1/RAID5 and generic
18 RAID code (such as request based resynchronization):
20 Neil Brown <neilb@cse.unsw.edu.au>.
22 - persistent bitmap code
23 Copyright (C) 2003-2004, Paul Clements, SteelEye Technology, Inc.
25 This program is free software; you can redistribute it and/or modify
26 it under the terms of the GNU General Public License as published by
27 the Free Software Foundation; either version 2, or (at your option)
30 You should have received a copy of the GNU General Public License
31 (for example /usr/src/linux/COPYING); if not, write to the Free
32 Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
35 #include <linux/kthread.h>
36 #include <linux/blkdev.h>
37 #include <linux/sysctl.h>
38 #include <linux/seq_file.h>
39 #include <linux/mutex.h>
40 #include <linux/buffer_head.h> /* for invalidate_bdev */
41 #include <linux/poll.h>
42 #include <linux/ctype.h>
43 #include <linux/string.h>
44 #include <linux/hdreg.h>
45 #include <linux/proc_fs.h>
46 #include <linux/random.h>
47 #include <linux/reboot.h>
48 #include <linux/file.h>
49 #include <linux/compat.h>
50 #include <linux/delay.h>
51 #include <linux/raid/md_p.h>
52 #include <linux/raid/md_u.h>
53 #include <linux/slab.h>
58 #define dprintk(x...) ((void)(DEBUG && printk(x)))
60 static DEFINE_MUTEX(md_mutex);
63 static void autostart_arrays(int part);
66 static LIST_HEAD(pers_list);
67 static DEFINE_SPINLOCK(pers_lock);
69 static void md_print_devices(void);
71 static DECLARE_WAIT_QUEUE_HEAD(resync_wait);
73 #define MD_BUG(x...) { printk("md: bug in file %s, line %d\n", __FILE__, __LINE__); md_print_devices(); }
76 * Default number of read corrections we'll attempt on an rdev
77 * before ejecting it from the array. We divide the read error
78 * count by 2 for every hour elapsed between read errors.
80 #define MD_DEFAULT_MAX_CORRECTED_READ_ERRORS 20
82 * Current RAID-1,4,5 parallel reconstruction 'guaranteed speed limit'
83 * is 1000 KB/sec, so the extra system load does not show up that much.
84 * Increase it if you want to have more _guaranteed_ speed. Note that
85 * the RAID driver will use the maximum available bandwidth if the IO
86 * subsystem is idle. There is also an 'absolute maximum' reconstruction
87 * speed limit - in case reconstruction slows down your system despite
90 * you can change it via /proc/sys/dev/raid/speed_limit_min and _max.
91 * or /sys/block/mdX/md/sync_speed_{min,max}
94 static int sysctl_speed_limit_min = 1000;
95 static int sysctl_speed_limit_max = 200000;
96 static inline int speed_min(mddev_t *mddev)
98 return mddev->sync_speed_min ?
99 mddev->sync_speed_min : sysctl_speed_limit_min;
102 static inline int speed_max(mddev_t *mddev)
104 return mddev->sync_speed_max ?
105 mddev->sync_speed_max : sysctl_speed_limit_max;
108 static struct ctl_table_header *raid_table_header;
110 static ctl_table raid_table[] = {
112 .procname = "speed_limit_min",
113 .data = &sysctl_speed_limit_min,
114 .maxlen = sizeof(int),
115 .mode = S_IRUGO|S_IWUSR,
116 .proc_handler = proc_dointvec,
119 .procname = "speed_limit_max",
120 .data = &sysctl_speed_limit_max,
121 .maxlen = sizeof(int),
122 .mode = S_IRUGO|S_IWUSR,
123 .proc_handler = proc_dointvec,
128 static ctl_table raid_dir_table[] = {
132 .mode = S_IRUGO|S_IXUGO,
138 static ctl_table raid_root_table[] = {
143 .child = raid_dir_table,
148 static const struct block_device_operations md_fops;
150 static int start_readonly;
153 * We have a system wide 'event count' that is incremented
154 * on any 'interesting' event, and readers of /proc/mdstat
155 * can use 'poll' or 'select' to find out when the event
159 * start array, stop array, error, add device, remove device,
160 * start build, activate spare
162 static DECLARE_WAIT_QUEUE_HEAD(md_event_waiters);
163 static atomic_t md_event_count;
164 void md_new_event(mddev_t *mddev)
166 atomic_inc(&md_event_count);
167 wake_up(&md_event_waiters);
169 EXPORT_SYMBOL_GPL(md_new_event);
171 /* Alternate version that can be called from interrupts
172 * when calling sysfs_notify isn't needed.
174 static void md_new_event_inintr(mddev_t *mddev)
176 atomic_inc(&md_event_count);
177 wake_up(&md_event_waiters);
181 * Enables to iterate over all existing md arrays
182 * all_mddevs_lock protects this list.
184 static LIST_HEAD(all_mddevs);
185 static DEFINE_SPINLOCK(all_mddevs_lock);
189 * iterates through all used mddevs in the system.
190 * We take care to grab the all_mddevs_lock whenever navigating
191 * the list, and to always hold a refcount when unlocked.
192 * Any code which breaks out of this loop while own
193 * a reference to the current mddev and must mddev_put it.
195 #define for_each_mddev(mddev,tmp) \
197 for (({ spin_lock(&all_mddevs_lock); \
198 tmp = all_mddevs.next; \
200 ({ if (tmp != &all_mddevs) \
201 mddev_get(list_entry(tmp, mddev_t, all_mddevs));\
202 spin_unlock(&all_mddevs_lock); \
203 if (mddev) mddev_put(mddev); \
204 mddev = list_entry(tmp, mddev_t, all_mddevs); \
205 tmp != &all_mddevs;}); \
206 ({ spin_lock(&all_mddevs_lock); \
211 /* Rather than calling directly into the personality make_request function,
212 * IO requests come here first so that we can check if the device is
213 * being suspended pending a reconfiguration.
214 * We hold a refcount over the call to ->make_request. By the time that
215 * call has finished, the bio has been linked into some internal structure
216 * and so is visible to ->quiesce(), so we don't need the refcount any more.
218 static int md_make_request(struct request_queue *q, struct bio *bio)
220 const int rw = bio_data_dir(bio);
221 mddev_t *mddev = q->queuedata;
225 if (mddev == NULL || mddev->pers == NULL) {
230 if (mddev->suspended || mddev->barrier) {
233 prepare_to_wait(&mddev->sb_wait, &__wait,
234 TASK_UNINTERRUPTIBLE);
235 if (!mddev->suspended && !mddev->barrier)
241 finish_wait(&mddev->sb_wait, &__wait);
243 atomic_inc(&mddev->active_io);
246 rv = mddev->pers->make_request(mddev, bio);
248 cpu = part_stat_lock();
249 part_stat_inc(cpu, &mddev->gendisk->part0, ios[rw]);
250 part_stat_add(cpu, &mddev->gendisk->part0, sectors[rw],
254 if (atomic_dec_and_test(&mddev->active_io) && mddev->suspended)
255 wake_up(&mddev->sb_wait);
260 /* mddev_suspend makes sure no new requests are submitted
261 * to the device, and that any requests that have been submitted
262 * are completely handled.
263 * Once ->stop is called and completes, the module will be completely
266 void mddev_suspend(mddev_t *mddev)
268 BUG_ON(mddev->suspended);
269 mddev->suspended = 1;
271 wait_event(mddev->sb_wait, atomic_read(&mddev->active_io) == 0);
272 mddev->pers->quiesce(mddev, 1);
274 EXPORT_SYMBOL_GPL(mddev_suspend);
276 void mddev_resume(mddev_t *mddev)
278 mddev->suspended = 0;
279 wake_up(&mddev->sb_wait);
280 mddev->pers->quiesce(mddev, 0);
282 EXPORT_SYMBOL_GPL(mddev_resume);
284 int mddev_congested(mddev_t *mddev, int bits)
288 return mddev->suspended;
290 EXPORT_SYMBOL(mddev_congested);
293 * Generic barrier handling for md
296 #define POST_REQUEST_BARRIER ((void*)1)
298 static void md_end_barrier(struct bio *bio, int err)
300 mdk_rdev_t *rdev = bio->bi_private;
301 mddev_t *mddev = rdev->mddev;
302 if (err == -EOPNOTSUPP && mddev->barrier != POST_REQUEST_BARRIER)
303 set_bit(BIO_EOPNOTSUPP, &mddev->barrier->bi_flags);
305 rdev_dec_pending(rdev, mddev);
307 if (atomic_dec_and_test(&mddev->flush_pending)) {
308 if (mddev->barrier == POST_REQUEST_BARRIER) {
309 /* This was a post-request barrier */
310 mddev->barrier = NULL;
311 wake_up(&mddev->sb_wait);
313 /* The pre-request barrier has finished */
314 schedule_work(&mddev->barrier_work);
319 static void submit_barriers(mddev_t *mddev)
324 list_for_each_entry_rcu(rdev, &mddev->disks, same_set)
325 if (rdev->raid_disk >= 0 &&
326 !test_bit(Faulty, &rdev->flags)) {
327 /* Take two references, one is dropped
328 * when request finishes, one after
329 * we reclaim rcu_read_lock
332 atomic_inc(&rdev->nr_pending);
333 atomic_inc(&rdev->nr_pending);
335 bi = bio_alloc(GFP_KERNEL, 0);
336 bi->bi_end_io = md_end_barrier;
337 bi->bi_private = rdev;
338 bi->bi_bdev = rdev->bdev;
339 atomic_inc(&mddev->flush_pending);
340 submit_bio(WRITE_BARRIER, bi);
342 rdev_dec_pending(rdev, mddev);
347 static void md_submit_barrier(struct work_struct *ws)
349 mddev_t *mddev = container_of(ws, mddev_t, barrier_work);
350 struct bio *bio = mddev->barrier;
352 atomic_set(&mddev->flush_pending, 1);
354 if (test_bit(BIO_EOPNOTSUPP, &bio->bi_flags))
355 bio_endio(bio, -EOPNOTSUPP);
356 else if (bio->bi_size == 0)
357 /* an empty barrier - all done */
360 bio->bi_rw &= ~REQ_HARDBARRIER;
361 if (mddev->pers->make_request(mddev, bio))
362 generic_make_request(bio);
363 mddev->barrier = POST_REQUEST_BARRIER;
364 submit_barriers(mddev);
366 if (atomic_dec_and_test(&mddev->flush_pending)) {
367 mddev->barrier = NULL;
368 wake_up(&mddev->sb_wait);
372 void md_barrier_request(mddev_t *mddev, struct bio *bio)
374 spin_lock_irq(&mddev->write_lock);
375 wait_event_lock_irq(mddev->sb_wait,
377 mddev->write_lock, /*nothing*/);
378 mddev->barrier = bio;
379 spin_unlock_irq(&mddev->write_lock);
381 atomic_set(&mddev->flush_pending, 1);
382 INIT_WORK(&mddev->barrier_work, md_submit_barrier);
384 submit_barriers(mddev);
386 if (atomic_dec_and_test(&mddev->flush_pending))
387 schedule_work(&mddev->barrier_work);
389 EXPORT_SYMBOL(md_barrier_request);
391 /* Support for plugging.
392 * This mirrors the plugging support in request_queue, but does not
393 * require having a whole queue
395 static void plugger_work(struct work_struct *work)
397 struct plug_handle *plug =
398 container_of(work, struct plug_handle, unplug_work);
399 plug->unplug_fn(plug);
401 static void plugger_timeout(unsigned long data)
403 struct plug_handle *plug = (void *)data;
404 kblockd_schedule_work(NULL, &plug->unplug_work);
406 void plugger_init(struct plug_handle *plug,
407 void (*unplug_fn)(struct plug_handle *))
409 plug->unplug_flag = 0;
410 plug->unplug_fn = unplug_fn;
411 init_timer(&plug->unplug_timer);
412 plug->unplug_timer.function = plugger_timeout;
413 plug->unplug_timer.data = (unsigned long)plug;
414 INIT_WORK(&plug->unplug_work, plugger_work);
416 EXPORT_SYMBOL_GPL(plugger_init);
418 void plugger_set_plug(struct plug_handle *plug)
420 if (!test_and_set_bit(PLUGGED_FLAG, &plug->unplug_flag))
421 mod_timer(&plug->unplug_timer, jiffies + msecs_to_jiffies(3)+1);
423 EXPORT_SYMBOL_GPL(plugger_set_plug);
425 int plugger_remove_plug(struct plug_handle *plug)
427 if (test_and_clear_bit(PLUGGED_FLAG, &plug->unplug_flag)) {
428 del_timer(&plug->unplug_timer);
433 EXPORT_SYMBOL_GPL(plugger_remove_plug);
436 static inline mddev_t *mddev_get(mddev_t *mddev)
438 atomic_inc(&mddev->active);
442 static void mddev_delayed_delete(struct work_struct *ws);
444 static void mddev_put(mddev_t *mddev)
446 if (!atomic_dec_and_lock(&mddev->active, &all_mddevs_lock))
448 if (!mddev->raid_disks && list_empty(&mddev->disks) &&
449 mddev->ctime == 0 && !mddev->hold_active) {
450 /* Array is not configured at all, and not held active,
452 list_del(&mddev->all_mddevs);
453 if (mddev->gendisk) {
454 /* we did a probe so need to clean up.
455 * Call schedule_work inside the spinlock
456 * so that flush_scheduled_work() after
457 * mddev_find will succeed in waiting for the
460 INIT_WORK(&mddev->del_work, mddev_delayed_delete);
461 schedule_work(&mddev->del_work);
465 spin_unlock(&all_mddevs_lock);
468 void mddev_init(mddev_t *mddev)
470 mutex_init(&mddev->open_mutex);
471 mutex_init(&mddev->reconfig_mutex);
472 mutex_init(&mddev->bitmap_info.mutex);
473 INIT_LIST_HEAD(&mddev->disks);
474 INIT_LIST_HEAD(&mddev->all_mddevs);
475 init_timer(&mddev->safemode_timer);
476 atomic_set(&mddev->active, 1);
477 atomic_set(&mddev->openers, 0);
478 atomic_set(&mddev->active_io, 0);
479 spin_lock_init(&mddev->write_lock);
480 atomic_set(&mddev->flush_pending, 0);
481 init_waitqueue_head(&mddev->sb_wait);
482 init_waitqueue_head(&mddev->recovery_wait);
483 mddev->reshape_position = MaxSector;
484 mddev->resync_min = 0;
485 mddev->resync_max = MaxSector;
486 mddev->level = LEVEL_NONE;
488 EXPORT_SYMBOL_GPL(mddev_init);
490 static mddev_t * mddev_find(dev_t unit)
492 mddev_t *mddev, *new = NULL;
495 spin_lock(&all_mddevs_lock);
498 list_for_each_entry(mddev, &all_mddevs, all_mddevs)
499 if (mddev->unit == unit) {
501 spin_unlock(&all_mddevs_lock);
507 list_add(&new->all_mddevs, &all_mddevs);
508 spin_unlock(&all_mddevs_lock);
509 new->hold_active = UNTIL_IOCTL;
513 /* find an unused unit number */
514 static int next_minor = 512;
515 int start = next_minor;
519 dev = MKDEV(MD_MAJOR, next_minor);
521 if (next_minor > MINORMASK)
523 if (next_minor == start) {
524 /* Oh dear, all in use. */
525 spin_unlock(&all_mddevs_lock);
531 list_for_each_entry(mddev, &all_mddevs, all_mddevs)
532 if (mddev->unit == dev) {
538 new->md_minor = MINOR(dev);
539 new->hold_active = UNTIL_STOP;
540 list_add(&new->all_mddevs, &all_mddevs);
541 spin_unlock(&all_mddevs_lock);
544 spin_unlock(&all_mddevs_lock);
546 new = kzalloc(sizeof(*new), GFP_KERNEL);
551 if (MAJOR(unit) == MD_MAJOR)
552 new->md_minor = MINOR(unit);
554 new->md_minor = MINOR(unit) >> MdpMinorShift;
561 static inline int mddev_lock(mddev_t * mddev)
563 return mutex_lock_interruptible(&mddev->reconfig_mutex);
566 static inline int mddev_is_locked(mddev_t *mddev)
568 return mutex_is_locked(&mddev->reconfig_mutex);
571 static inline int mddev_trylock(mddev_t * mddev)
573 return mutex_trylock(&mddev->reconfig_mutex);
576 static struct attribute_group md_redundancy_group;
578 static void mddev_unlock(mddev_t * mddev)
580 if (mddev->to_remove) {
581 /* These cannot be removed under reconfig_mutex as
582 * an access to the files will try to take reconfig_mutex
583 * while holding the file unremovable, which leads to
585 * So hold set sysfs_active while the remove in happeing,
586 * and anything else which might set ->to_remove or my
587 * otherwise change the sysfs namespace will fail with
588 * -EBUSY if sysfs_active is still set.
589 * We set sysfs_active under reconfig_mutex and elsewhere
590 * test it under the same mutex to ensure its correct value
593 struct attribute_group *to_remove = mddev->to_remove;
594 mddev->to_remove = NULL;
595 mddev->sysfs_active = 1;
596 mutex_unlock(&mddev->reconfig_mutex);
598 if (mddev->kobj.sd) {
599 if (to_remove != &md_redundancy_group)
600 sysfs_remove_group(&mddev->kobj, to_remove);
601 if (mddev->pers == NULL ||
602 mddev->pers->sync_request == NULL) {
603 sysfs_remove_group(&mddev->kobj, &md_redundancy_group);
604 if (mddev->sysfs_action)
605 sysfs_put(mddev->sysfs_action);
606 mddev->sysfs_action = NULL;
609 mddev->sysfs_active = 0;
611 mutex_unlock(&mddev->reconfig_mutex);
613 md_wakeup_thread(mddev->thread);
616 static mdk_rdev_t * find_rdev_nr(mddev_t *mddev, int nr)
620 list_for_each_entry(rdev, &mddev->disks, same_set)
621 if (rdev->desc_nr == nr)
627 static mdk_rdev_t * find_rdev(mddev_t * mddev, dev_t dev)
631 list_for_each_entry(rdev, &mddev->disks, same_set)
632 if (rdev->bdev->bd_dev == dev)
638 static struct mdk_personality *find_pers(int level, char *clevel)
640 struct mdk_personality *pers;
641 list_for_each_entry(pers, &pers_list, list) {
642 if (level != LEVEL_NONE && pers->level == level)
644 if (strcmp(pers->name, clevel)==0)
650 /* return the offset of the super block in 512byte sectors */
651 static inline sector_t calc_dev_sboffset(struct block_device *bdev)
653 sector_t num_sectors = bdev->bd_inode->i_size / 512;
654 return MD_NEW_SIZE_SECTORS(num_sectors);
657 static int alloc_disk_sb(mdk_rdev_t * rdev)
662 rdev->sb_page = alloc_page(GFP_KERNEL);
663 if (!rdev->sb_page) {
664 printk(KERN_ALERT "md: out of memory.\n");
671 static void free_disk_sb(mdk_rdev_t * rdev)
674 put_page(rdev->sb_page);
676 rdev->sb_page = NULL;
683 static void super_written(struct bio *bio, int error)
685 mdk_rdev_t *rdev = bio->bi_private;
686 mddev_t *mddev = rdev->mddev;
688 if (error || !test_bit(BIO_UPTODATE, &bio->bi_flags)) {
689 printk("md: super_written gets error=%d, uptodate=%d\n",
690 error, test_bit(BIO_UPTODATE, &bio->bi_flags));
691 WARN_ON(test_bit(BIO_UPTODATE, &bio->bi_flags));
692 md_error(mddev, rdev);
695 if (atomic_dec_and_test(&mddev->pending_writes))
696 wake_up(&mddev->sb_wait);
700 static void super_written_barrier(struct bio *bio, int error)
702 struct bio *bio2 = bio->bi_private;
703 mdk_rdev_t *rdev = bio2->bi_private;
704 mddev_t *mddev = rdev->mddev;
706 if (!test_bit(BIO_UPTODATE, &bio->bi_flags) &&
707 error == -EOPNOTSUPP) {
709 /* barriers don't appear to be supported :-( */
710 set_bit(BarriersNotsupp, &rdev->flags);
711 mddev->barriers_work = 0;
712 spin_lock_irqsave(&mddev->write_lock, flags);
713 bio2->bi_next = mddev->biolist;
714 mddev->biolist = bio2;
715 spin_unlock_irqrestore(&mddev->write_lock, flags);
716 wake_up(&mddev->sb_wait);
720 bio->bi_private = rdev;
721 super_written(bio, error);
725 void md_super_write(mddev_t *mddev, mdk_rdev_t *rdev,
726 sector_t sector, int size, struct page *page)
728 /* write first size bytes of page to sector of rdev
729 * Increment mddev->pending_writes before returning
730 * and decrement it on completion, waking up sb_wait
731 * if zero is reached.
732 * If an error occurred, call md_error
734 * As we might need to resubmit the request if REQ_HARDBARRIER
735 * causes ENOTSUPP, we allocate a spare bio...
737 struct bio *bio = bio_alloc(GFP_NOIO, 1);
738 int rw = REQ_WRITE | REQ_SYNC | REQ_UNPLUG;
740 bio->bi_bdev = rdev->bdev;
741 bio->bi_sector = sector;
742 bio_add_page(bio, page, size, 0);
743 bio->bi_private = rdev;
744 bio->bi_end_io = super_written;
747 atomic_inc(&mddev->pending_writes);
748 if (!test_bit(BarriersNotsupp, &rdev->flags)) {
750 rw |= REQ_HARDBARRIER;
751 rbio = bio_clone(bio, GFP_NOIO);
752 rbio->bi_private = bio;
753 rbio->bi_end_io = super_written_barrier;
754 submit_bio(rw, rbio);
759 void md_super_wait(mddev_t *mddev)
761 /* wait for all superblock writes that were scheduled to complete.
762 * if any had to be retried (due to BARRIER problems), retry them
766 prepare_to_wait(&mddev->sb_wait, &wq, TASK_UNINTERRUPTIBLE);
767 if (atomic_read(&mddev->pending_writes)==0)
769 while (mddev->biolist) {
771 spin_lock_irq(&mddev->write_lock);
772 bio = mddev->biolist;
773 mddev->biolist = bio->bi_next ;
775 spin_unlock_irq(&mddev->write_lock);
776 submit_bio(bio->bi_rw, bio);
780 finish_wait(&mddev->sb_wait, &wq);
783 static void bi_complete(struct bio *bio, int error)
785 complete((struct completion*)bio->bi_private);
788 int sync_page_io(struct block_device *bdev, sector_t sector, int size,
789 struct page *page, int rw)
791 struct bio *bio = bio_alloc(GFP_NOIO, 1);
792 struct completion event;
795 rw |= REQ_SYNC | REQ_UNPLUG;
798 bio->bi_sector = sector;
799 bio_add_page(bio, page, size, 0);
800 init_completion(&event);
801 bio->bi_private = &event;
802 bio->bi_end_io = bi_complete;
804 wait_for_completion(&event);
806 ret = test_bit(BIO_UPTODATE, &bio->bi_flags);
810 EXPORT_SYMBOL_GPL(sync_page_io);
812 static int read_disk_sb(mdk_rdev_t * rdev, int size)
814 char b[BDEVNAME_SIZE];
815 if (!rdev->sb_page) {
823 if (!sync_page_io(rdev->bdev, rdev->sb_start, size, rdev->sb_page, READ))
829 printk(KERN_WARNING "md: disabled device %s, could not read superblock.\n",
830 bdevname(rdev->bdev,b));
834 static int uuid_equal(mdp_super_t *sb1, mdp_super_t *sb2)
836 return sb1->set_uuid0 == sb2->set_uuid0 &&
837 sb1->set_uuid1 == sb2->set_uuid1 &&
838 sb1->set_uuid2 == sb2->set_uuid2 &&
839 sb1->set_uuid3 == sb2->set_uuid3;
842 static int sb_equal(mdp_super_t *sb1, mdp_super_t *sb2)
845 mdp_super_t *tmp1, *tmp2;
847 tmp1 = kmalloc(sizeof(*tmp1),GFP_KERNEL);
848 tmp2 = kmalloc(sizeof(*tmp2),GFP_KERNEL);
850 if (!tmp1 || !tmp2) {
852 printk(KERN_INFO "md.c sb_equal(): failed to allocate memory!\n");
860 * nr_disks is not constant
865 ret = (memcmp(tmp1, tmp2, MD_SB_GENERIC_CONSTANT_WORDS * 4) == 0);
873 static u32 md_csum_fold(u32 csum)
875 csum = (csum & 0xffff) + (csum >> 16);
876 return (csum & 0xffff) + (csum >> 16);
879 static unsigned int calc_sb_csum(mdp_super_t * sb)
882 u32 *sb32 = (u32*)sb;
884 unsigned int disk_csum, csum;
886 disk_csum = sb->sb_csum;
889 for (i = 0; i < MD_SB_BYTES/4 ; i++)
891 csum = (newcsum & 0xffffffff) + (newcsum>>32);
895 /* This used to use csum_partial, which was wrong for several
896 * reasons including that different results are returned on
897 * different architectures. It isn't critical that we get exactly
898 * the same return value as before (we always csum_fold before
899 * testing, and that removes any differences). However as we
900 * know that csum_partial always returned a 16bit value on
901 * alphas, do a fold to maximise conformity to previous behaviour.
903 sb->sb_csum = md_csum_fold(disk_csum);
905 sb->sb_csum = disk_csum;
912 * Handle superblock details.
913 * We want to be able to handle multiple superblock formats
914 * so we have a common interface to them all, and an array of
915 * different handlers.
916 * We rely on user-space to write the initial superblock, and support
917 * reading and updating of superblocks.
918 * Interface methods are:
919 * int load_super(mdk_rdev_t *dev, mdk_rdev_t *refdev, int minor_version)
920 * loads and validates a superblock on dev.
921 * if refdev != NULL, compare superblocks on both devices
923 * 0 - dev has a superblock that is compatible with refdev
924 * 1 - dev has a superblock that is compatible and newer than refdev
925 * so dev should be used as the refdev in future
926 * -EINVAL superblock incompatible or invalid
927 * -othererror e.g. -EIO
929 * int validate_super(mddev_t *mddev, mdk_rdev_t *dev)
930 * Verify that dev is acceptable into mddev.
931 * The first time, mddev->raid_disks will be 0, and data from
932 * dev should be merged in. Subsequent calls check that dev
933 * is new enough. Return 0 or -EINVAL
935 * void sync_super(mddev_t *mddev, mdk_rdev_t *dev)
936 * Update the superblock for rdev with data in mddev
937 * This does not write to disc.
943 struct module *owner;
944 int (*load_super)(mdk_rdev_t *rdev, mdk_rdev_t *refdev,
946 int (*validate_super)(mddev_t *mddev, mdk_rdev_t *rdev);
947 void (*sync_super)(mddev_t *mddev, mdk_rdev_t *rdev);
948 unsigned long long (*rdev_size_change)(mdk_rdev_t *rdev,
949 sector_t num_sectors);
953 * Check that the given mddev has no bitmap.
955 * This function is called from the run method of all personalities that do not
956 * support bitmaps. It prints an error message and returns non-zero if mddev
957 * has a bitmap. Otherwise, it returns 0.
960 int md_check_no_bitmap(mddev_t *mddev)
962 if (!mddev->bitmap_info.file && !mddev->bitmap_info.offset)
964 printk(KERN_ERR "%s: bitmaps are not supported for %s\n",
965 mdname(mddev), mddev->pers->name);
968 EXPORT_SYMBOL(md_check_no_bitmap);
971 * load_super for 0.90.0
973 static int super_90_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
975 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
980 * Calculate the position of the superblock (512byte sectors),
981 * it's at the end of the disk.
983 * It also happens to be a multiple of 4Kb.
985 rdev->sb_start = calc_dev_sboffset(rdev->bdev);
987 ret = read_disk_sb(rdev, MD_SB_BYTES);
992 bdevname(rdev->bdev, b);
993 sb = (mdp_super_t*)page_address(rdev->sb_page);
995 if (sb->md_magic != MD_SB_MAGIC) {
996 printk(KERN_ERR "md: invalid raid superblock magic on %s\n",
1001 if (sb->major_version != 0 ||
1002 sb->minor_version < 90 ||
1003 sb->minor_version > 91) {
1004 printk(KERN_WARNING "Bad version number %d.%d on %s\n",
1005 sb->major_version, sb->minor_version,
1010 if (sb->raid_disks <= 0)
1013 if (md_csum_fold(calc_sb_csum(sb)) != md_csum_fold(sb->sb_csum)) {
1014 printk(KERN_WARNING "md: invalid superblock checksum on %s\n",
1019 rdev->preferred_minor = sb->md_minor;
1020 rdev->data_offset = 0;
1021 rdev->sb_size = MD_SB_BYTES;
1023 if (sb->level == LEVEL_MULTIPATH)
1026 rdev->desc_nr = sb->this_disk.number;
1032 mdp_super_t *refsb = (mdp_super_t*)page_address(refdev->sb_page);
1033 if (!uuid_equal(refsb, sb)) {
1034 printk(KERN_WARNING "md: %s has different UUID to %s\n",
1035 b, bdevname(refdev->bdev,b2));
1038 if (!sb_equal(refsb, sb)) {
1039 printk(KERN_WARNING "md: %s has same UUID"
1040 " but different superblock to %s\n",
1041 b, bdevname(refdev->bdev, b2));
1045 ev2 = md_event(refsb);
1051 rdev->sectors = rdev->sb_start;
1053 if (rdev->sectors < sb->size * 2 && sb->level > 1)
1054 /* "this cannot possibly happen" ... */
1062 * validate_super for 0.90.0
1064 static int super_90_validate(mddev_t *mddev, mdk_rdev_t *rdev)
1067 mdp_super_t *sb = (mdp_super_t *)page_address(rdev->sb_page);
1068 __u64 ev1 = md_event(sb);
1070 rdev->raid_disk = -1;
1071 clear_bit(Faulty, &rdev->flags);
1072 clear_bit(In_sync, &rdev->flags);
1073 clear_bit(WriteMostly, &rdev->flags);
1074 clear_bit(BarriersNotsupp, &rdev->flags);
1076 if (mddev->raid_disks == 0) {
1077 mddev->major_version = 0;
1078 mddev->minor_version = sb->minor_version;
1079 mddev->patch_version = sb->patch_version;
1080 mddev->external = 0;
1081 mddev->chunk_sectors = sb->chunk_size >> 9;
1082 mddev->ctime = sb->ctime;
1083 mddev->utime = sb->utime;
1084 mddev->level = sb->level;
1085 mddev->clevel[0] = 0;
1086 mddev->layout = sb->layout;
1087 mddev->raid_disks = sb->raid_disks;
1088 mddev->dev_sectors = sb->size * 2;
1089 mddev->events = ev1;
1090 mddev->bitmap_info.offset = 0;
1091 mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9;
1093 if (mddev->minor_version >= 91) {
1094 mddev->reshape_position = sb->reshape_position;
1095 mddev->delta_disks = sb->delta_disks;
1096 mddev->new_level = sb->new_level;
1097 mddev->new_layout = sb->new_layout;
1098 mddev->new_chunk_sectors = sb->new_chunk >> 9;
1100 mddev->reshape_position = MaxSector;
1101 mddev->delta_disks = 0;
1102 mddev->new_level = mddev->level;
1103 mddev->new_layout = mddev->layout;
1104 mddev->new_chunk_sectors = mddev->chunk_sectors;
1107 if (sb->state & (1<<MD_SB_CLEAN))
1108 mddev->recovery_cp = MaxSector;
1110 if (sb->events_hi == sb->cp_events_hi &&
1111 sb->events_lo == sb->cp_events_lo) {
1112 mddev->recovery_cp = sb->recovery_cp;
1114 mddev->recovery_cp = 0;
1117 memcpy(mddev->uuid+0, &sb->set_uuid0, 4);
1118 memcpy(mddev->uuid+4, &sb->set_uuid1, 4);
1119 memcpy(mddev->uuid+8, &sb->set_uuid2, 4);
1120 memcpy(mddev->uuid+12,&sb->set_uuid3, 4);
1122 mddev->max_disks = MD_SB_DISKS;
1124 if (sb->state & (1<<MD_SB_BITMAP_PRESENT) &&
1125 mddev->bitmap_info.file == NULL)
1126 mddev->bitmap_info.offset =
1127 mddev->bitmap_info.default_offset;
1129 } else if (mddev->pers == NULL) {
1130 /* Insist on good event counter while assembling, except
1131 * for spares (which don't need an event count) */
1133 if (sb->disks[rdev->desc_nr].state & (
1134 (1<<MD_DISK_SYNC) | (1 << MD_DISK_ACTIVE)))
1135 if (ev1 < mddev->events)
1137 } else if (mddev->bitmap) {
1138 /* if adding to array with a bitmap, then we can accept an
1139 * older device ... but not too old.
1141 if (ev1 < mddev->bitmap->events_cleared)
1144 if (ev1 < mddev->events)
1145 /* just a hot-add of a new device, leave raid_disk at -1 */
1149 if (mddev->level != LEVEL_MULTIPATH) {
1150 desc = sb->disks + rdev->desc_nr;
1152 if (desc->state & (1<<MD_DISK_FAULTY))
1153 set_bit(Faulty, &rdev->flags);
1154 else if (desc->state & (1<<MD_DISK_SYNC) /* &&
1155 desc->raid_disk < mddev->raid_disks */) {
1156 set_bit(In_sync, &rdev->flags);
1157 rdev->raid_disk = desc->raid_disk;
1158 } else if (desc->state & (1<<MD_DISK_ACTIVE)) {
1159 /* active but not in sync implies recovery up to
1160 * reshape position. We don't know exactly where
1161 * that is, so set to zero for now */
1162 if (mddev->minor_version >= 91) {
1163 rdev->recovery_offset = 0;
1164 rdev->raid_disk = desc->raid_disk;
1167 if (desc->state & (1<<MD_DISK_WRITEMOSTLY))
1168 set_bit(WriteMostly, &rdev->flags);
1169 } else /* MULTIPATH are always insync */
1170 set_bit(In_sync, &rdev->flags);
1175 * sync_super for 0.90.0
1177 static void super_90_sync(mddev_t *mddev, mdk_rdev_t *rdev)
1181 int next_spare = mddev->raid_disks;
1184 /* make rdev->sb match mddev data..
1187 * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
1188 * 3/ any empty disks < next_spare become removed
1190 * disks[0] gets initialised to REMOVED because
1191 * we cannot be sure from other fields if it has
1192 * been initialised or not.
1195 int active=0, working=0,failed=0,spare=0,nr_disks=0;
1197 rdev->sb_size = MD_SB_BYTES;
1199 sb = (mdp_super_t*)page_address(rdev->sb_page);
1201 memset(sb, 0, sizeof(*sb));
1203 sb->md_magic = MD_SB_MAGIC;
1204 sb->major_version = mddev->major_version;
1205 sb->patch_version = mddev->patch_version;
1206 sb->gvalid_words = 0; /* ignored */
1207 memcpy(&sb->set_uuid0, mddev->uuid+0, 4);
1208 memcpy(&sb->set_uuid1, mddev->uuid+4, 4);
1209 memcpy(&sb->set_uuid2, mddev->uuid+8, 4);
1210 memcpy(&sb->set_uuid3, mddev->uuid+12,4);
1212 sb->ctime = mddev->ctime;
1213 sb->level = mddev->level;
1214 sb->size = mddev->dev_sectors / 2;
1215 sb->raid_disks = mddev->raid_disks;
1216 sb->md_minor = mddev->md_minor;
1217 sb->not_persistent = 0;
1218 sb->utime = mddev->utime;
1220 sb->events_hi = (mddev->events>>32);
1221 sb->events_lo = (u32)mddev->events;
1223 if (mddev->reshape_position == MaxSector)
1224 sb->minor_version = 90;
1226 sb->minor_version = 91;
1227 sb->reshape_position = mddev->reshape_position;
1228 sb->new_level = mddev->new_level;
1229 sb->delta_disks = mddev->delta_disks;
1230 sb->new_layout = mddev->new_layout;
1231 sb->new_chunk = mddev->new_chunk_sectors << 9;
1233 mddev->minor_version = sb->minor_version;
1236 sb->recovery_cp = mddev->recovery_cp;
1237 sb->cp_events_hi = (mddev->events>>32);
1238 sb->cp_events_lo = (u32)mddev->events;
1239 if (mddev->recovery_cp == MaxSector)
1240 sb->state = (1<< MD_SB_CLEAN);
1242 sb->recovery_cp = 0;
1244 sb->layout = mddev->layout;
1245 sb->chunk_size = mddev->chunk_sectors << 9;
1247 if (mddev->bitmap && mddev->bitmap_info.file == NULL)
1248 sb->state |= (1<<MD_SB_BITMAP_PRESENT);
1250 sb->disks[0].state = (1<<MD_DISK_REMOVED);
1251 list_for_each_entry(rdev2, &mddev->disks, same_set) {
1254 int is_active = test_bit(In_sync, &rdev2->flags);
1256 if (rdev2->raid_disk >= 0 &&
1257 sb->minor_version >= 91)
1258 /* we have nowhere to store the recovery_offset,
1259 * but if it is not below the reshape_position,
1260 * we can piggy-back on that.
1263 if (rdev2->raid_disk < 0 ||
1264 test_bit(Faulty, &rdev2->flags))
1267 desc_nr = rdev2->raid_disk;
1269 desc_nr = next_spare++;
1270 rdev2->desc_nr = desc_nr;
1271 d = &sb->disks[rdev2->desc_nr];
1273 d->number = rdev2->desc_nr;
1274 d->major = MAJOR(rdev2->bdev->bd_dev);
1275 d->minor = MINOR(rdev2->bdev->bd_dev);
1277 d->raid_disk = rdev2->raid_disk;
1279 d->raid_disk = rdev2->desc_nr; /* compatibility */
1280 if (test_bit(Faulty, &rdev2->flags))
1281 d->state = (1<<MD_DISK_FAULTY);
1282 else if (is_active) {
1283 d->state = (1<<MD_DISK_ACTIVE);
1284 if (test_bit(In_sync, &rdev2->flags))
1285 d->state |= (1<<MD_DISK_SYNC);
1293 if (test_bit(WriteMostly, &rdev2->flags))
1294 d->state |= (1<<MD_DISK_WRITEMOSTLY);
1296 /* now set the "removed" and "faulty" bits on any missing devices */
1297 for (i=0 ; i < mddev->raid_disks ; i++) {
1298 mdp_disk_t *d = &sb->disks[i];
1299 if (d->state == 0 && d->number == 0) {
1302 d->state = (1<<MD_DISK_REMOVED);
1303 d->state |= (1<<MD_DISK_FAULTY);
1307 sb->nr_disks = nr_disks;
1308 sb->active_disks = active;
1309 sb->working_disks = working;
1310 sb->failed_disks = failed;
1311 sb->spare_disks = spare;
1313 sb->this_disk = sb->disks[rdev->desc_nr];
1314 sb->sb_csum = calc_sb_csum(sb);
1318 * rdev_size_change for 0.90.0
1320 static unsigned long long
1321 super_90_rdev_size_change(mdk_rdev_t *rdev, sector_t num_sectors)
1323 if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
1324 return 0; /* component must fit device */
1325 if (rdev->mddev->bitmap_info.offset)
1326 return 0; /* can't move bitmap */
1327 rdev->sb_start = calc_dev_sboffset(rdev->bdev);
1328 if (!num_sectors || num_sectors > rdev->sb_start)
1329 num_sectors = rdev->sb_start;
1330 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
1332 md_super_wait(rdev->mddev);
1333 return num_sectors / 2; /* kB for sysfs */
1338 * version 1 superblock
1341 static __le32 calc_sb_1_csum(struct mdp_superblock_1 * sb)
1345 unsigned long long newcsum;
1346 int size = 256 + le32_to_cpu(sb->max_dev)*2;
1347 __le32 *isuper = (__le32*)sb;
1350 disk_csum = sb->sb_csum;
1353 for (i=0; size>=4; size -= 4 )
1354 newcsum += le32_to_cpu(*isuper++);
1357 newcsum += le16_to_cpu(*(__le16*) isuper);
1359 csum = (newcsum & 0xffffffff) + (newcsum >> 32);
1360 sb->sb_csum = disk_csum;
1361 return cpu_to_le32(csum);
1364 static int super_1_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
1366 struct mdp_superblock_1 *sb;
1369 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
1373 * Calculate the position of the superblock in 512byte sectors.
1374 * It is always aligned to a 4K boundary and
1375 * depeding on minor_version, it can be:
1376 * 0: At least 8K, but less than 12K, from end of device
1377 * 1: At start of device
1378 * 2: 4K from start of device.
1380 switch(minor_version) {
1382 sb_start = rdev->bdev->bd_inode->i_size >> 9;
1384 sb_start &= ~(sector_t)(4*2-1);
1395 rdev->sb_start = sb_start;
1397 /* superblock is rarely larger than 1K, but it can be larger,
1398 * and it is safe to read 4k, so we do that
1400 ret = read_disk_sb(rdev, 4096);
1401 if (ret) return ret;
1404 sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1406 if (sb->magic != cpu_to_le32(MD_SB_MAGIC) ||
1407 sb->major_version != cpu_to_le32(1) ||
1408 le32_to_cpu(sb->max_dev) > (4096-256)/2 ||
1409 le64_to_cpu(sb->super_offset) != rdev->sb_start ||
1410 (le32_to_cpu(sb->feature_map) & ~MD_FEATURE_ALL) != 0)
1413 if (calc_sb_1_csum(sb) != sb->sb_csum) {
1414 printk("md: invalid superblock checksum on %s\n",
1415 bdevname(rdev->bdev,b));
1418 if (le64_to_cpu(sb->data_size) < 10) {
1419 printk("md: data_size too small on %s\n",
1420 bdevname(rdev->bdev,b));
1424 rdev->preferred_minor = 0xffff;
1425 rdev->data_offset = le64_to_cpu(sb->data_offset);
1426 atomic_set(&rdev->corrected_errors, le32_to_cpu(sb->cnt_corrected_read));
1428 rdev->sb_size = le32_to_cpu(sb->max_dev) * 2 + 256;
1429 bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
1430 if (rdev->sb_size & bmask)
1431 rdev->sb_size = (rdev->sb_size | bmask) + 1;
1434 && rdev->data_offset < sb_start + (rdev->sb_size/512))
1437 if (sb->level == cpu_to_le32(LEVEL_MULTIPATH))
1440 rdev->desc_nr = le32_to_cpu(sb->dev_number);
1446 struct mdp_superblock_1 *refsb =
1447 (struct mdp_superblock_1*)page_address(refdev->sb_page);
1449 if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 ||
1450 sb->level != refsb->level ||
1451 sb->layout != refsb->layout ||
1452 sb->chunksize != refsb->chunksize) {
1453 printk(KERN_WARNING "md: %s has strangely different"
1454 " superblock to %s\n",
1455 bdevname(rdev->bdev,b),
1456 bdevname(refdev->bdev,b2));
1459 ev1 = le64_to_cpu(sb->events);
1460 ev2 = le64_to_cpu(refsb->events);
1468 rdev->sectors = (rdev->bdev->bd_inode->i_size >> 9) -
1469 le64_to_cpu(sb->data_offset);
1471 rdev->sectors = rdev->sb_start;
1472 if (rdev->sectors < le64_to_cpu(sb->data_size))
1474 rdev->sectors = le64_to_cpu(sb->data_size);
1475 if (le64_to_cpu(sb->size) > rdev->sectors)
1480 static int super_1_validate(mddev_t *mddev, mdk_rdev_t *rdev)
1482 struct mdp_superblock_1 *sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1483 __u64 ev1 = le64_to_cpu(sb->events);
1485 rdev->raid_disk = -1;
1486 clear_bit(Faulty, &rdev->flags);
1487 clear_bit(In_sync, &rdev->flags);
1488 clear_bit(WriteMostly, &rdev->flags);
1489 clear_bit(BarriersNotsupp, &rdev->flags);
1491 if (mddev->raid_disks == 0) {
1492 mddev->major_version = 1;
1493 mddev->patch_version = 0;
1494 mddev->external = 0;
1495 mddev->chunk_sectors = le32_to_cpu(sb->chunksize);
1496 mddev->ctime = le64_to_cpu(sb->ctime) & ((1ULL << 32)-1);
1497 mddev->utime = le64_to_cpu(sb->utime) & ((1ULL << 32)-1);
1498 mddev->level = le32_to_cpu(sb->level);
1499 mddev->clevel[0] = 0;
1500 mddev->layout = le32_to_cpu(sb->layout);
1501 mddev->raid_disks = le32_to_cpu(sb->raid_disks);
1502 mddev->dev_sectors = le64_to_cpu(sb->size);
1503 mddev->events = ev1;
1504 mddev->bitmap_info.offset = 0;
1505 mddev->bitmap_info.default_offset = 1024 >> 9;
1507 mddev->recovery_cp = le64_to_cpu(sb->resync_offset);
1508 memcpy(mddev->uuid, sb->set_uuid, 16);
1510 mddev->max_disks = (4096-256)/2;
1512 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) &&
1513 mddev->bitmap_info.file == NULL )
1514 mddev->bitmap_info.offset =
1515 (__s32)le32_to_cpu(sb->bitmap_offset);
1517 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) {
1518 mddev->reshape_position = le64_to_cpu(sb->reshape_position);
1519 mddev->delta_disks = le32_to_cpu(sb->delta_disks);
1520 mddev->new_level = le32_to_cpu(sb->new_level);
1521 mddev->new_layout = le32_to_cpu(sb->new_layout);
1522 mddev->new_chunk_sectors = le32_to_cpu(sb->new_chunk);
1524 mddev->reshape_position = MaxSector;
1525 mddev->delta_disks = 0;
1526 mddev->new_level = mddev->level;
1527 mddev->new_layout = mddev->layout;
1528 mddev->new_chunk_sectors = mddev->chunk_sectors;
1531 } else if (mddev->pers == NULL) {
1532 /* Insist of good event counter while assembling, except for
1533 * spares (which don't need an event count) */
1535 if (rdev->desc_nr >= 0 &&
1536 rdev->desc_nr < le32_to_cpu(sb->max_dev) &&
1537 le16_to_cpu(sb->dev_roles[rdev->desc_nr]) < 0xfffe)
1538 if (ev1 < mddev->events)
1540 } else if (mddev->bitmap) {
1541 /* If adding to array with a bitmap, then we can accept an
1542 * older device, but not too old.
1544 if (ev1 < mddev->bitmap->events_cleared)
1547 if (ev1 < mddev->events)
1548 /* just a hot-add of a new device, leave raid_disk at -1 */
1551 if (mddev->level != LEVEL_MULTIPATH) {
1553 if (rdev->desc_nr < 0 ||
1554 rdev->desc_nr >= le32_to_cpu(sb->max_dev)) {
1558 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
1560 case 0xffff: /* spare */
1562 case 0xfffe: /* faulty */
1563 set_bit(Faulty, &rdev->flags);
1566 if ((le32_to_cpu(sb->feature_map) &
1567 MD_FEATURE_RECOVERY_OFFSET))
1568 rdev->recovery_offset = le64_to_cpu(sb->recovery_offset);
1570 set_bit(In_sync, &rdev->flags);
1571 rdev->raid_disk = role;
1574 if (sb->devflags & WriteMostly1)
1575 set_bit(WriteMostly, &rdev->flags);
1576 } else /* MULTIPATH are always insync */
1577 set_bit(In_sync, &rdev->flags);
1582 static void super_1_sync(mddev_t *mddev, mdk_rdev_t *rdev)
1584 struct mdp_superblock_1 *sb;
1587 /* make rdev->sb match mddev and rdev data. */
1589 sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1591 sb->feature_map = 0;
1593 sb->recovery_offset = cpu_to_le64(0);
1594 memset(sb->pad1, 0, sizeof(sb->pad1));
1595 memset(sb->pad2, 0, sizeof(sb->pad2));
1596 memset(sb->pad3, 0, sizeof(sb->pad3));
1598 sb->utime = cpu_to_le64((__u64)mddev->utime);
1599 sb->events = cpu_to_le64(mddev->events);
1601 sb->resync_offset = cpu_to_le64(mddev->recovery_cp);
1603 sb->resync_offset = cpu_to_le64(0);
1605 sb->cnt_corrected_read = cpu_to_le32(atomic_read(&rdev->corrected_errors));
1607 sb->raid_disks = cpu_to_le32(mddev->raid_disks);
1608 sb->size = cpu_to_le64(mddev->dev_sectors);
1609 sb->chunksize = cpu_to_le32(mddev->chunk_sectors);
1610 sb->level = cpu_to_le32(mddev->level);
1611 sb->layout = cpu_to_le32(mddev->layout);
1613 if (mddev->bitmap && mddev->bitmap_info.file == NULL) {
1614 sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_info.offset);
1615 sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET);
1618 if (rdev->raid_disk >= 0 &&
1619 !test_bit(In_sync, &rdev->flags)) {
1621 cpu_to_le32(MD_FEATURE_RECOVERY_OFFSET);
1622 sb->recovery_offset =
1623 cpu_to_le64(rdev->recovery_offset);
1626 if (mddev->reshape_position != MaxSector) {
1627 sb->feature_map |= cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE);
1628 sb->reshape_position = cpu_to_le64(mddev->reshape_position);
1629 sb->new_layout = cpu_to_le32(mddev->new_layout);
1630 sb->delta_disks = cpu_to_le32(mddev->delta_disks);
1631 sb->new_level = cpu_to_le32(mddev->new_level);
1632 sb->new_chunk = cpu_to_le32(mddev->new_chunk_sectors);
1636 list_for_each_entry(rdev2, &mddev->disks, same_set)
1637 if (rdev2->desc_nr+1 > max_dev)
1638 max_dev = rdev2->desc_nr+1;
1640 if (max_dev > le32_to_cpu(sb->max_dev)) {
1642 sb->max_dev = cpu_to_le32(max_dev);
1643 rdev->sb_size = max_dev * 2 + 256;
1644 bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
1645 if (rdev->sb_size & bmask)
1646 rdev->sb_size = (rdev->sb_size | bmask) + 1;
1648 for (i=0; i<max_dev;i++)
1649 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1651 list_for_each_entry(rdev2, &mddev->disks, same_set) {
1653 if (test_bit(Faulty, &rdev2->flags))
1654 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1655 else if (test_bit(In_sync, &rdev2->flags))
1656 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1657 else if (rdev2->raid_disk >= 0)
1658 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1660 sb->dev_roles[i] = cpu_to_le16(0xffff);
1663 sb->sb_csum = calc_sb_1_csum(sb);
1666 static unsigned long long
1667 super_1_rdev_size_change(mdk_rdev_t *rdev, sector_t num_sectors)
1669 struct mdp_superblock_1 *sb;
1670 sector_t max_sectors;
1671 if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
1672 return 0; /* component must fit device */
1673 if (rdev->sb_start < rdev->data_offset) {
1674 /* minor versions 1 and 2; superblock before data */
1675 max_sectors = rdev->bdev->bd_inode->i_size >> 9;
1676 max_sectors -= rdev->data_offset;
1677 if (!num_sectors || num_sectors > max_sectors)
1678 num_sectors = max_sectors;
1679 } else if (rdev->mddev->bitmap_info.offset) {
1680 /* minor version 0 with bitmap we can't move */
1683 /* minor version 0; superblock after data */
1685 sb_start = (rdev->bdev->bd_inode->i_size >> 9) - 8*2;
1686 sb_start &= ~(sector_t)(4*2 - 1);
1687 max_sectors = rdev->sectors + sb_start - rdev->sb_start;
1688 if (!num_sectors || num_sectors > max_sectors)
1689 num_sectors = max_sectors;
1690 rdev->sb_start = sb_start;
1692 sb = (struct mdp_superblock_1 *) page_address(rdev->sb_page);
1693 sb->data_size = cpu_to_le64(num_sectors);
1694 sb->super_offset = rdev->sb_start;
1695 sb->sb_csum = calc_sb_1_csum(sb);
1696 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
1698 md_super_wait(rdev->mddev);
1699 return num_sectors / 2; /* kB for sysfs */
1702 static struct super_type super_types[] = {
1705 .owner = THIS_MODULE,
1706 .load_super = super_90_load,
1707 .validate_super = super_90_validate,
1708 .sync_super = super_90_sync,
1709 .rdev_size_change = super_90_rdev_size_change,
1713 .owner = THIS_MODULE,
1714 .load_super = super_1_load,
1715 .validate_super = super_1_validate,
1716 .sync_super = super_1_sync,
1717 .rdev_size_change = super_1_rdev_size_change,
1721 static int match_mddev_units(mddev_t *mddev1, mddev_t *mddev2)
1723 mdk_rdev_t *rdev, *rdev2;
1726 rdev_for_each_rcu(rdev, mddev1)
1727 rdev_for_each_rcu(rdev2, mddev2)
1728 if (rdev->bdev->bd_contains ==
1729 rdev2->bdev->bd_contains) {
1737 static LIST_HEAD(pending_raid_disks);
1740 * Try to register data integrity profile for an mddev
1742 * This is called when an array is started and after a disk has been kicked
1743 * from the array. It only succeeds if all working and active component devices
1744 * are integrity capable with matching profiles.
1746 int md_integrity_register(mddev_t *mddev)
1748 mdk_rdev_t *rdev, *reference = NULL;
1750 if (list_empty(&mddev->disks))
1751 return 0; /* nothing to do */
1752 if (blk_get_integrity(mddev->gendisk))
1753 return 0; /* already registered */
1754 list_for_each_entry(rdev, &mddev->disks, same_set) {
1755 /* skip spares and non-functional disks */
1756 if (test_bit(Faulty, &rdev->flags))
1758 if (rdev->raid_disk < 0)
1761 * If at least one rdev is not integrity capable, we can not
1762 * enable data integrity for the md device.
1764 if (!bdev_get_integrity(rdev->bdev))
1767 /* Use the first rdev as the reference */
1771 /* does this rdev's profile match the reference profile? */
1772 if (blk_integrity_compare(reference->bdev->bd_disk,
1773 rdev->bdev->bd_disk) < 0)
1777 * All component devices are integrity capable and have matching
1778 * profiles, register the common profile for the md device.
1780 if (blk_integrity_register(mddev->gendisk,
1781 bdev_get_integrity(reference->bdev)) != 0) {
1782 printk(KERN_ERR "md: failed to register integrity for %s\n",
1786 printk(KERN_NOTICE "md: data integrity on %s enabled\n",
1790 EXPORT_SYMBOL(md_integrity_register);
1792 /* Disable data integrity if non-capable/non-matching disk is being added */
1793 void md_integrity_add_rdev(mdk_rdev_t *rdev, mddev_t *mddev)
1795 struct blk_integrity *bi_rdev = bdev_get_integrity(rdev->bdev);
1796 struct blk_integrity *bi_mddev = blk_get_integrity(mddev->gendisk);
1798 if (!bi_mddev) /* nothing to do */
1800 if (rdev->raid_disk < 0) /* skip spares */
1802 if (bi_rdev && blk_integrity_compare(mddev->gendisk,
1803 rdev->bdev->bd_disk) >= 0)
1805 printk(KERN_NOTICE "disabling data integrity on %s\n", mdname(mddev));
1806 blk_integrity_unregister(mddev->gendisk);
1808 EXPORT_SYMBOL(md_integrity_add_rdev);
1810 static int bind_rdev_to_array(mdk_rdev_t * rdev, mddev_t * mddev)
1812 char b[BDEVNAME_SIZE];
1822 /* prevent duplicates */
1823 if (find_rdev(mddev, rdev->bdev->bd_dev))
1826 /* make sure rdev->sectors exceeds mddev->dev_sectors */
1827 if (rdev->sectors && (mddev->dev_sectors == 0 ||
1828 rdev->sectors < mddev->dev_sectors)) {
1830 /* Cannot change size, so fail
1831 * If mddev->level <= 0, then we don't care
1832 * about aligning sizes (e.g. linear)
1834 if (mddev->level > 0)
1837 mddev->dev_sectors = rdev->sectors;
1840 /* Verify rdev->desc_nr is unique.
1841 * If it is -1, assign a free number, else
1842 * check number is not in use
1844 if (rdev->desc_nr < 0) {
1846 if (mddev->pers) choice = mddev->raid_disks;
1847 while (find_rdev_nr(mddev, choice))
1849 rdev->desc_nr = choice;
1851 if (find_rdev_nr(mddev, rdev->desc_nr))
1854 if (mddev->max_disks && rdev->desc_nr >= mddev->max_disks) {
1855 printk(KERN_WARNING "md: %s: array is limited to %d devices\n",
1856 mdname(mddev), mddev->max_disks);
1859 bdevname(rdev->bdev,b);
1860 while ( (s=strchr(b, '/')) != NULL)
1863 rdev->mddev = mddev;
1864 printk(KERN_INFO "md: bind<%s>\n", b);
1866 if ((err = kobject_add(&rdev->kobj, &mddev->kobj, "dev-%s", b)))
1869 ko = &part_to_dev(rdev->bdev->bd_part)->kobj;
1870 if (sysfs_create_link(&rdev->kobj, ko, "block"))
1871 /* failure here is OK */;
1872 rdev->sysfs_state = sysfs_get_dirent_safe(rdev->kobj.sd, "state");
1874 list_add_rcu(&rdev->same_set, &mddev->disks);
1875 bd_claim_by_disk(rdev->bdev, rdev->bdev->bd_holder, mddev->gendisk);
1877 /* May as well allow recovery to be retried once */
1878 mddev->recovery_disabled = 0;
1883 printk(KERN_WARNING "md: failed to register dev-%s for %s\n",
1888 static void md_delayed_delete(struct work_struct *ws)
1890 mdk_rdev_t *rdev = container_of(ws, mdk_rdev_t, del_work);
1891 kobject_del(&rdev->kobj);
1892 kobject_put(&rdev->kobj);
1895 static void unbind_rdev_from_array(mdk_rdev_t * rdev)
1897 char b[BDEVNAME_SIZE];
1902 bd_release_from_disk(rdev->bdev, rdev->mddev->gendisk);
1903 list_del_rcu(&rdev->same_set);
1904 printk(KERN_INFO "md: unbind<%s>\n", bdevname(rdev->bdev,b));
1906 sysfs_remove_link(&rdev->kobj, "block");
1907 sysfs_put(rdev->sysfs_state);
1908 rdev->sysfs_state = NULL;
1909 /* We need to delay this, otherwise we can deadlock when
1910 * writing to 'remove' to "dev/state". We also need
1911 * to delay it due to rcu usage.
1914 INIT_WORK(&rdev->del_work, md_delayed_delete);
1915 kobject_get(&rdev->kobj);
1916 schedule_work(&rdev->del_work);
1920 * prevent the device from being mounted, repartitioned or
1921 * otherwise reused by a RAID array (or any other kernel
1922 * subsystem), by bd_claiming the device.
1924 static int lock_rdev(mdk_rdev_t *rdev, dev_t dev, int shared)
1927 struct block_device *bdev;
1928 char b[BDEVNAME_SIZE];
1930 bdev = open_by_devnum(dev, FMODE_READ|FMODE_WRITE);
1932 printk(KERN_ERR "md: could not open %s.\n",
1933 __bdevname(dev, b));
1934 return PTR_ERR(bdev);
1936 err = bd_claim(bdev, shared ? (mdk_rdev_t *)lock_rdev : rdev);
1938 printk(KERN_ERR "md: could not bd_claim %s.\n",
1940 blkdev_put(bdev, FMODE_READ|FMODE_WRITE);
1944 set_bit(AllReserved, &rdev->flags);
1949 static void unlock_rdev(mdk_rdev_t *rdev)
1951 struct block_device *bdev = rdev->bdev;
1956 blkdev_put(bdev, FMODE_READ|FMODE_WRITE);
1959 void md_autodetect_dev(dev_t dev);
1961 static void export_rdev(mdk_rdev_t * rdev)
1963 char b[BDEVNAME_SIZE];
1964 printk(KERN_INFO "md: export_rdev(%s)\n",
1965 bdevname(rdev->bdev,b));
1970 if (test_bit(AutoDetected, &rdev->flags))
1971 md_autodetect_dev(rdev->bdev->bd_dev);
1974 kobject_put(&rdev->kobj);
1977 static void kick_rdev_from_array(mdk_rdev_t * rdev)
1979 unbind_rdev_from_array(rdev);
1983 static void export_array(mddev_t *mddev)
1985 mdk_rdev_t *rdev, *tmp;
1987 rdev_for_each(rdev, tmp, mddev) {
1992 kick_rdev_from_array(rdev);
1994 if (!list_empty(&mddev->disks))
1996 mddev->raid_disks = 0;
1997 mddev->major_version = 0;
2000 static void print_desc(mdp_disk_t *desc)
2002 printk(" DISK<N:%d,(%d,%d),R:%d,S:%d>\n", desc->number,
2003 desc->major,desc->minor,desc->raid_disk,desc->state);
2006 static void print_sb_90(mdp_super_t *sb)
2011 "md: SB: (V:%d.%d.%d) ID:<%08x.%08x.%08x.%08x> CT:%08x\n",
2012 sb->major_version, sb->minor_version, sb->patch_version,
2013 sb->set_uuid0, sb->set_uuid1, sb->set_uuid2, sb->set_uuid3,
2015 printk(KERN_INFO "md: L%d S%08d ND:%d RD:%d md%d LO:%d CS:%d\n",
2016 sb->level, sb->size, sb->nr_disks, sb->raid_disks,
2017 sb->md_minor, sb->layout, sb->chunk_size);
2018 printk(KERN_INFO "md: UT:%08x ST:%d AD:%d WD:%d"
2019 " FD:%d SD:%d CSUM:%08x E:%08lx\n",
2020 sb->utime, sb->state, sb->active_disks, sb->working_disks,
2021 sb->failed_disks, sb->spare_disks,
2022 sb->sb_csum, (unsigned long)sb->events_lo);
2025 for (i = 0; i < MD_SB_DISKS; i++) {
2028 desc = sb->disks + i;
2029 if (desc->number || desc->major || desc->minor ||
2030 desc->raid_disk || (desc->state && (desc->state != 4))) {
2031 printk(" D %2d: ", i);
2035 printk(KERN_INFO "md: THIS: ");
2036 print_desc(&sb->this_disk);
2039 static void print_sb_1(struct mdp_superblock_1 *sb)
2043 uuid = sb->set_uuid;
2045 "md: SB: (V:%u) (F:0x%08x) Array-ID:<%pU>\n"
2046 "md: Name: \"%s\" CT:%llu\n",
2047 le32_to_cpu(sb->major_version),
2048 le32_to_cpu(sb->feature_map),
2051 (unsigned long long)le64_to_cpu(sb->ctime)
2052 & MD_SUPERBLOCK_1_TIME_SEC_MASK);
2054 uuid = sb->device_uuid;
2056 "md: L%u SZ%llu RD:%u LO:%u CS:%u DO:%llu DS:%llu SO:%llu"
2058 "md: Dev:%08x UUID: %pU\n"
2059 "md: (F:0x%08x) UT:%llu Events:%llu ResyncOffset:%llu CSUM:0x%08x\n"
2060 "md: (MaxDev:%u) \n",
2061 le32_to_cpu(sb->level),
2062 (unsigned long long)le64_to_cpu(sb->size),
2063 le32_to_cpu(sb->raid_disks),
2064 le32_to_cpu(sb->layout),
2065 le32_to_cpu(sb->chunksize),
2066 (unsigned long long)le64_to_cpu(sb->data_offset),
2067 (unsigned long long)le64_to_cpu(sb->data_size),
2068 (unsigned long long)le64_to_cpu(sb->super_offset),
2069 (unsigned long long)le64_to_cpu(sb->recovery_offset),
2070 le32_to_cpu(sb->dev_number),
2073 (unsigned long long)le64_to_cpu(sb->utime) & MD_SUPERBLOCK_1_TIME_SEC_MASK,
2074 (unsigned long long)le64_to_cpu(sb->events),
2075 (unsigned long long)le64_to_cpu(sb->resync_offset),
2076 le32_to_cpu(sb->sb_csum),
2077 le32_to_cpu(sb->max_dev)
2081 static void print_rdev(mdk_rdev_t *rdev, int major_version)
2083 char b[BDEVNAME_SIZE];
2084 printk(KERN_INFO "md: rdev %s, Sect:%08llu F:%d S:%d DN:%u\n",
2085 bdevname(rdev->bdev, b), (unsigned long long)rdev->sectors,
2086 test_bit(Faulty, &rdev->flags), test_bit(In_sync, &rdev->flags),
2088 if (rdev->sb_loaded) {
2089 printk(KERN_INFO "md: rdev superblock (MJ:%d):\n", major_version);
2090 switch (major_version) {
2092 print_sb_90((mdp_super_t*)page_address(rdev->sb_page));
2095 print_sb_1((struct mdp_superblock_1 *)page_address(rdev->sb_page));
2099 printk(KERN_INFO "md: no rdev superblock!\n");
2102 static void md_print_devices(void)
2104 struct list_head *tmp;
2107 char b[BDEVNAME_SIZE];
2110 printk("md: **********************************\n");
2111 printk("md: * <COMPLETE RAID STATE PRINTOUT> *\n");
2112 printk("md: **********************************\n");
2113 for_each_mddev(mddev, tmp) {
2116 bitmap_print_sb(mddev->bitmap);
2118 printk("%s: ", mdname(mddev));
2119 list_for_each_entry(rdev, &mddev->disks, same_set)
2120 printk("<%s>", bdevname(rdev->bdev,b));
2123 list_for_each_entry(rdev, &mddev->disks, same_set)
2124 print_rdev(rdev, mddev->major_version);
2126 printk("md: **********************************\n");
2131 static void sync_sbs(mddev_t * mddev, int nospares)
2133 /* Update each superblock (in-memory image), but
2134 * if we are allowed to, skip spares which already
2135 * have the right event counter, or have one earlier
2136 * (which would mean they aren't being marked as dirty
2137 * with the rest of the array)
2140 list_for_each_entry(rdev, &mddev->disks, same_set) {
2141 if (rdev->sb_events == mddev->events ||
2143 rdev->raid_disk < 0 &&
2144 rdev->sb_events+1 == mddev->events)) {
2145 /* Don't update this superblock */
2146 rdev->sb_loaded = 2;
2148 super_types[mddev->major_version].
2149 sync_super(mddev, rdev);
2150 rdev->sb_loaded = 1;
2155 static void md_update_sb(mddev_t * mddev, int force_change)
2162 /* First make sure individual recovery_offsets are correct */
2163 list_for_each_entry(rdev, &mddev->disks, same_set) {
2164 if (rdev->raid_disk >= 0 &&
2165 mddev->delta_disks >= 0 &&
2166 !test_bit(In_sync, &rdev->flags) &&
2167 mddev->curr_resync_completed > rdev->recovery_offset)
2168 rdev->recovery_offset = mddev->curr_resync_completed;
2171 if (!mddev->persistent) {
2172 clear_bit(MD_CHANGE_CLEAN, &mddev->flags);
2173 clear_bit(MD_CHANGE_DEVS, &mddev->flags);
2174 wake_up(&mddev->sb_wait);
2178 spin_lock_irq(&mddev->write_lock);
2180 mddev->utime = get_seconds();
2182 if (test_and_clear_bit(MD_CHANGE_DEVS, &mddev->flags))
2184 if (test_and_clear_bit(MD_CHANGE_CLEAN, &mddev->flags))
2185 /* just a clean<-> dirty transition, possibly leave spares alone,
2186 * though if events isn't the right even/odd, we will have to do
2192 if (mddev->degraded)
2193 /* If the array is degraded, then skipping spares is both
2194 * dangerous and fairly pointless.
2195 * Dangerous because a device that was removed from the array
2196 * might have a event_count that still looks up-to-date,
2197 * so it can be re-added without a resync.
2198 * Pointless because if there are any spares to skip,
2199 * then a recovery will happen and soon that array won't
2200 * be degraded any more and the spare can go back to sleep then.
2204 sync_req = mddev->in_sync;
2206 /* If this is just a dirty<->clean transition, and the array is clean
2207 * and 'events' is odd, we can roll back to the previous clean state */
2209 && (mddev->in_sync && mddev->recovery_cp == MaxSector)
2210 && mddev->can_decrease_events
2211 && mddev->events != 1) {
2213 mddev->can_decrease_events = 0;
2215 /* otherwise we have to go forward and ... */
2217 mddev->can_decrease_events = nospares;
2220 if (!mddev->events) {
2222 * oops, this 64-bit counter should never wrap.
2223 * Either we are in around ~1 trillion A.C., assuming
2224 * 1 reboot per second, or we have a bug:
2229 sync_sbs(mddev, nospares);
2230 spin_unlock_irq(&mddev->write_lock);
2233 "md: updating %s RAID superblock on device (in sync %d)\n",
2234 mdname(mddev),mddev->in_sync);
2236 bitmap_update_sb(mddev->bitmap);
2237 list_for_each_entry(rdev, &mddev->disks, same_set) {
2238 char b[BDEVNAME_SIZE];
2239 dprintk(KERN_INFO "md: ");
2240 if (rdev->sb_loaded != 1)
2241 continue; /* no noise on spare devices */
2242 if (test_bit(Faulty, &rdev->flags))
2243 dprintk("(skipping faulty ");
2245 dprintk("%s ", bdevname(rdev->bdev,b));
2246 if (!test_bit(Faulty, &rdev->flags)) {
2247 md_super_write(mddev,rdev,
2248 rdev->sb_start, rdev->sb_size,
2250 dprintk(KERN_INFO "(write) %s's sb offset: %llu\n",
2251 bdevname(rdev->bdev,b),
2252 (unsigned long long)rdev->sb_start);
2253 rdev->sb_events = mddev->events;
2257 if (mddev->level == LEVEL_MULTIPATH)
2258 /* only need to write one superblock... */
2261 md_super_wait(mddev);
2262 /* if there was a failure, MD_CHANGE_DEVS was set, and we re-write super */
2264 spin_lock_irq(&mddev->write_lock);
2265 if (mddev->in_sync != sync_req ||
2266 test_bit(MD_CHANGE_DEVS, &mddev->flags)) {
2267 /* have to write it out again */
2268 spin_unlock_irq(&mddev->write_lock);
2271 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
2272 spin_unlock_irq(&mddev->write_lock);
2273 wake_up(&mddev->sb_wait);
2274 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
2275 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
2279 /* words written to sysfs files may, or may not, be \n terminated.
2280 * We want to accept with case. For this we use cmd_match.
2282 static int cmd_match(const char *cmd, const char *str)
2284 /* See if cmd, written into a sysfs file, matches
2285 * str. They must either be the same, or cmd can
2286 * have a trailing newline
2288 while (*cmd && *str && *cmd == *str) {
2299 struct rdev_sysfs_entry {
2300 struct attribute attr;
2301 ssize_t (*show)(mdk_rdev_t *, char *);
2302 ssize_t (*store)(mdk_rdev_t *, const char *, size_t);
2306 state_show(mdk_rdev_t *rdev, char *page)
2311 if (test_bit(Faulty, &rdev->flags)) {
2312 len+= sprintf(page+len, "%sfaulty",sep);
2315 if (test_bit(In_sync, &rdev->flags)) {
2316 len += sprintf(page+len, "%sin_sync",sep);
2319 if (test_bit(WriteMostly, &rdev->flags)) {
2320 len += sprintf(page+len, "%swrite_mostly",sep);
2323 if (test_bit(Blocked, &rdev->flags)) {
2324 len += sprintf(page+len, "%sblocked", sep);
2327 if (!test_bit(Faulty, &rdev->flags) &&
2328 !test_bit(In_sync, &rdev->flags)) {
2329 len += sprintf(page+len, "%sspare", sep);
2332 return len+sprintf(page+len, "\n");
2336 state_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2339 * faulty - simulates and error
2340 * remove - disconnects the device
2341 * writemostly - sets write_mostly
2342 * -writemostly - clears write_mostly
2343 * blocked - sets the Blocked flag
2344 * -blocked - clears the Blocked flag
2345 * insync - sets Insync providing device isn't active
2348 if (cmd_match(buf, "faulty") && rdev->mddev->pers) {
2349 md_error(rdev->mddev, rdev);
2351 } else if (cmd_match(buf, "remove")) {
2352 if (rdev->raid_disk >= 0)
2355 mddev_t *mddev = rdev->mddev;
2356 kick_rdev_from_array(rdev);
2358 md_update_sb(mddev, 1);
2359 md_new_event(mddev);
2362 } else if (cmd_match(buf, "writemostly")) {
2363 set_bit(WriteMostly, &rdev->flags);
2365 } else if (cmd_match(buf, "-writemostly")) {
2366 clear_bit(WriteMostly, &rdev->flags);
2368 } else if (cmd_match(buf, "blocked")) {
2369 set_bit(Blocked, &rdev->flags);
2371 } else if (cmd_match(buf, "-blocked")) {
2372 clear_bit(Blocked, &rdev->flags);
2373 wake_up(&rdev->blocked_wait);
2374 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2375 md_wakeup_thread(rdev->mddev->thread);
2378 } else if (cmd_match(buf, "insync") && rdev->raid_disk == -1) {
2379 set_bit(In_sync, &rdev->flags);
2383 sysfs_notify_dirent_safe(rdev->sysfs_state);
2384 return err ? err : len;
2386 static struct rdev_sysfs_entry rdev_state =
2387 __ATTR(state, S_IRUGO|S_IWUSR, state_show, state_store);
2390 errors_show(mdk_rdev_t *rdev, char *page)
2392 return sprintf(page, "%d\n", atomic_read(&rdev->corrected_errors));
2396 errors_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2399 unsigned long n = simple_strtoul(buf, &e, 10);
2400 if (*buf && (*e == 0 || *e == '\n')) {
2401 atomic_set(&rdev->corrected_errors, n);
2406 static struct rdev_sysfs_entry rdev_errors =
2407 __ATTR(errors, S_IRUGO|S_IWUSR, errors_show, errors_store);
2410 slot_show(mdk_rdev_t *rdev, char *page)
2412 if (rdev->raid_disk < 0)
2413 return sprintf(page, "none\n");
2415 return sprintf(page, "%d\n", rdev->raid_disk);
2419 slot_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2424 int slot = simple_strtoul(buf, &e, 10);
2425 if (strncmp(buf, "none", 4)==0)
2427 else if (e==buf || (*e && *e!= '\n'))
2429 if (rdev->mddev->pers && slot == -1) {
2430 /* Setting 'slot' on an active array requires also
2431 * updating the 'rd%d' link, and communicating
2432 * with the personality with ->hot_*_disk.
2433 * For now we only support removing
2434 * failed/spare devices. This normally happens automatically,
2435 * but not when the metadata is externally managed.
2437 if (rdev->raid_disk == -1)
2439 /* personality does all needed checks */
2440 if (rdev->mddev->pers->hot_add_disk == NULL)
2442 err = rdev->mddev->pers->
2443 hot_remove_disk(rdev->mddev, rdev->raid_disk);
2446 sprintf(nm, "rd%d", rdev->raid_disk);
2447 sysfs_remove_link(&rdev->mddev->kobj, nm);
2448 rdev->raid_disk = -1;
2449 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2450 md_wakeup_thread(rdev->mddev->thread);
2451 } else if (rdev->mddev->pers) {
2453 /* Activating a spare .. or possibly reactivating
2454 * if we ever get bitmaps working here.
2457 if (rdev->raid_disk != -1)
2460 if (rdev->mddev->pers->hot_add_disk == NULL)
2463 list_for_each_entry(rdev2, &rdev->mddev->disks, same_set)
2464 if (rdev2->raid_disk == slot)
2467 rdev->raid_disk = slot;
2468 if (test_bit(In_sync, &rdev->flags))
2469 rdev->saved_raid_disk = slot;
2471 rdev->saved_raid_disk = -1;
2472 err = rdev->mddev->pers->
2473 hot_add_disk(rdev->mddev, rdev);
2475 rdev->raid_disk = -1;
2478 sysfs_notify_dirent_safe(rdev->sysfs_state);
2479 sprintf(nm, "rd%d", rdev->raid_disk);
2480 if (sysfs_create_link(&rdev->mddev->kobj, &rdev->kobj, nm))
2481 /* failure here is OK */;
2482 /* don't wakeup anyone, leave that to userspace. */
2484 if (slot >= rdev->mddev->raid_disks)
2486 rdev->raid_disk = slot;
2487 /* assume it is working */
2488 clear_bit(Faulty, &rdev->flags);
2489 clear_bit(WriteMostly, &rdev->flags);
2490 set_bit(In_sync, &rdev->flags);
2491 sysfs_notify_dirent_safe(rdev->sysfs_state);
2497 static struct rdev_sysfs_entry rdev_slot =
2498 __ATTR(slot, S_IRUGO|S_IWUSR, slot_show, slot_store);
2501 offset_show(mdk_rdev_t *rdev, char *page)
2503 return sprintf(page, "%llu\n", (unsigned long long)rdev->data_offset);
2507 offset_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2510 unsigned long long offset = simple_strtoull(buf, &e, 10);
2511 if (e==buf || (*e && *e != '\n'))
2513 if (rdev->mddev->pers && rdev->raid_disk >= 0)
2515 if (rdev->sectors && rdev->mddev->external)
2516 /* Must set offset before size, so overlap checks
2519 rdev->data_offset = offset;
2523 static struct rdev_sysfs_entry rdev_offset =
2524 __ATTR(offset, S_IRUGO|S_IWUSR, offset_show, offset_store);
2527 rdev_size_show(mdk_rdev_t *rdev, char *page)
2529 return sprintf(page, "%llu\n", (unsigned long long)rdev->sectors / 2);
2532 static int overlaps(sector_t s1, sector_t l1, sector_t s2, sector_t l2)
2534 /* check if two start/length pairs overlap */
2542 static int strict_blocks_to_sectors(const char *buf, sector_t *sectors)
2544 unsigned long long blocks;
2547 if (strict_strtoull(buf, 10, &blocks) < 0)
2550 if (blocks & 1ULL << (8 * sizeof(blocks) - 1))
2551 return -EINVAL; /* sector conversion overflow */
2554 if (new != blocks * 2)
2555 return -EINVAL; /* unsigned long long to sector_t overflow */
2562 rdev_size_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2564 mddev_t *my_mddev = rdev->mddev;
2565 sector_t oldsectors = rdev->sectors;
2568 if (strict_blocks_to_sectors(buf, §ors) < 0)
2570 if (my_mddev->pers && rdev->raid_disk >= 0) {
2571 if (my_mddev->persistent) {
2572 sectors = super_types[my_mddev->major_version].
2573 rdev_size_change(rdev, sectors);
2576 } else if (!sectors)
2577 sectors = (rdev->bdev->bd_inode->i_size >> 9) -
2580 if (sectors < my_mddev->dev_sectors)
2581 return -EINVAL; /* component must fit device */
2583 rdev->sectors = sectors;
2584 if (sectors > oldsectors && my_mddev->external) {
2585 /* need to check that all other rdevs with the same ->bdev
2586 * do not overlap. We need to unlock the mddev to avoid
2587 * a deadlock. We have already changed rdev->sectors, and if
2588 * we have to change it back, we will have the lock again.
2592 struct list_head *tmp;
2594 mddev_unlock(my_mddev);
2595 for_each_mddev(mddev, tmp) {
2599 list_for_each_entry(rdev2, &mddev->disks, same_set)
2600 if (test_bit(AllReserved, &rdev2->flags) ||
2601 (rdev->bdev == rdev2->bdev &&
2603 overlaps(rdev->data_offset, rdev->sectors,
2609 mddev_unlock(mddev);
2615 mddev_lock(my_mddev);
2617 /* Someone else could have slipped in a size
2618 * change here, but doing so is just silly.
2619 * We put oldsectors back because we *know* it is
2620 * safe, and trust userspace not to race with
2623 rdev->sectors = oldsectors;
2630 static struct rdev_sysfs_entry rdev_size =
2631 __ATTR(size, S_IRUGO|S_IWUSR, rdev_size_show, rdev_size_store);
2634 static ssize_t recovery_start_show(mdk_rdev_t *rdev, char *page)
2636 unsigned long long recovery_start = rdev->recovery_offset;
2638 if (test_bit(In_sync, &rdev->flags) ||
2639 recovery_start == MaxSector)
2640 return sprintf(page, "none\n");
2642 return sprintf(page, "%llu\n", recovery_start);
2645 static ssize_t recovery_start_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2647 unsigned long long recovery_start;
2649 if (cmd_match(buf, "none"))
2650 recovery_start = MaxSector;
2651 else if (strict_strtoull(buf, 10, &recovery_start))
2654 if (rdev->mddev->pers &&
2655 rdev->raid_disk >= 0)
2658 rdev->recovery_offset = recovery_start;
2659 if (recovery_start == MaxSector)
2660 set_bit(In_sync, &rdev->flags);
2662 clear_bit(In_sync, &rdev->flags);
2666 static struct rdev_sysfs_entry rdev_recovery_start =
2667 __ATTR(recovery_start, S_IRUGO|S_IWUSR, recovery_start_show, recovery_start_store);
2669 static struct attribute *rdev_default_attrs[] = {
2675 &rdev_recovery_start.attr,
2679 rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
2681 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
2682 mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
2683 mddev_t *mddev = rdev->mddev;
2689 rv = mddev ? mddev_lock(mddev) : -EBUSY;
2691 if (rdev->mddev == NULL)
2694 rv = entry->show(rdev, page);
2695 mddev_unlock(mddev);
2701 rdev_attr_store(struct kobject *kobj, struct attribute *attr,
2702 const char *page, size_t length)
2704 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
2705 mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
2707 mddev_t *mddev = rdev->mddev;
2711 if (!capable(CAP_SYS_ADMIN))
2713 rv = mddev ? mddev_lock(mddev): -EBUSY;
2715 if (rdev->mddev == NULL)
2718 rv = entry->store(rdev, page, length);
2719 mddev_unlock(mddev);
2724 static void rdev_free(struct kobject *ko)
2726 mdk_rdev_t *rdev = container_of(ko, mdk_rdev_t, kobj);
2729 static const struct sysfs_ops rdev_sysfs_ops = {
2730 .show = rdev_attr_show,
2731 .store = rdev_attr_store,
2733 static struct kobj_type rdev_ktype = {
2734 .release = rdev_free,
2735 .sysfs_ops = &rdev_sysfs_ops,
2736 .default_attrs = rdev_default_attrs,
2739 void md_rdev_init(mdk_rdev_t *rdev)
2742 rdev->saved_raid_disk = -1;
2743 rdev->raid_disk = -1;
2745 rdev->data_offset = 0;
2746 rdev->sb_events = 0;
2747 rdev->last_read_error.tv_sec = 0;
2748 rdev->last_read_error.tv_nsec = 0;
2749 atomic_set(&rdev->nr_pending, 0);
2750 atomic_set(&rdev->read_errors, 0);
2751 atomic_set(&rdev->corrected_errors, 0);
2753 INIT_LIST_HEAD(&rdev->same_set);
2754 init_waitqueue_head(&rdev->blocked_wait);
2756 EXPORT_SYMBOL_GPL(md_rdev_init);
2758 * Import a device. If 'super_format' >= 0, then sanity check the superblock
2760 * mark the device faulty if:
2762 * - the device is nonexistent (zero size)
2763 * - the device has no valid superblock
2765 * a faulty rdev _never_ has rdev->sb set.
2767 static mdk_rdev_t *md_import_device(dev_t newdev, int super_format, int super_minor)
2769 char b[BDEVNAME_SIZE];
2774 rdev = kzalloc(sizeof(*rdev), GFP_KERNEL);
2776 printk(KERN_ERR "md: could not alloc mem for new device!\n");
2777 return ERR_PTR(-ENOMEM);
2781 if ((err = alloc_disk_sb(rdev)))
2784 err = lock_rdev(rdev, newdev, super_format == -2);
2788 kobject_init(&rdev->kobj, &rdev_ktype);
2790 size = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
2793 "md: %s has zero or unknown size, marking faulty!\n",
2794 bdevname(rdev->bdev,b));
2799 if (super_format >= 0) {
2800 err = super_types[super_format].
2801 load_super(rdev, NULL, super_minor);
2802 if (err == -EINVAL) {
2804 "md: %s does not have a valid v%d.%d "
2805 "superblock, not importing!\n",
2806 bdevname(rdev->bdev,b),
2807 super_format, super_minor);
2812 "md: could not read %s's sb, not importing!\n",
2813 bdevname(rdev->bdev,b));
2821 if (rdev->sb_page) {
2827 return ERR_PTR(err);
2831 * Check a full RAID array for plausibility
2835 static void analyze_sbs(mddev_t * mddev)
2838 mdk_rdev_t *rdev, *freshest, *tmp;
2839 char b[BDEVNAME_SIZE];
2842 rdev_for_each(rdev, tmp, mddev)
2843 switch (super_types[mddev->major_version].
2844 load_super(rdev, freshest, mddev->minor_version)) {
2852 "md: fatal superblock inconsistency in %s"
2853 " -- removing from array\n",
2854 bdevname(rdev->bdev,b));
2855 kick_rdev_from_array(rdev);
2859 super_types[mddev->major_version].
2860 validate_super(mddev, freshest);
2863 rdev_for_each(rdev, tmp, mddev) {
2864 if (mddev->max_disks &&
2865 (rdev->desc_nr >= mddev->max_disks ||
2866 i > mddev->max_disks)) {
2868 "md: %s: %s: only %d devices permitted\n",
2869 mdname(mddev), bdevname(rdev->bdev, b),
2871 kick_rdev_from_array(rdev);
2874 if (rdev != freshest)
2875 if (super_types[mddev->major_version].
2876 validate_super(mddev, rdev)) {
2877 printk(KERN_WARNING "md: kicking non-fresh %s"
2879 bdevname(rdev->bdev,b));
2880 kick_rdev_from_array(rdev);
2883 if (mddev->level == LEVEL_MULTIPATH) {
2884 rdev->desc_nr = i++;
2885 rdev->raid_disk = rdev->desc_nr;
2886 set_bit(In_sync, &rdev->flags);
2887 } else if (rdev->raid_disk >= (mddev->raid_disks - min(0, mddev->delta_disks))) {
2888 rdev->raid_disk = -1;
2889 clear_bit(In_sync, &rdev->flags);
2894 /* Read a fixed-point number.
2895 * Numbers in sysfs attributes should be in "standard" units where
2896 * possible, so time should be in seconds.
2897 * However we internally use a a much smaller unit such as
2898 * milliseconds or jiffies.
2899 * This function takes a decimal number with a possible fractional
2900 * component, and produces an integer which is the result of
2901 * multiplying that number by 10^'scale'.
2902 * all without any floating-point arithmetic.
2904 int strict_strtoul_scaled(const char *cp, unsigned long *res, int scale)
2906 unsigned long result = 0;
2908 while (isdigit(*cp) || (*cp == '.' && decimals < 0)) {
2911 else if (decimals < scale) {
2914 result = result * 10 + value;
2926 while (decimals < scale) {
2935 static void md_safemode_timeout(unsigned long data);
2938 safe_delay_show(mddev_t *mddev, char *page)
2940 int msec = (mddev->safemode_delay*1000)/HZ;
2941 return sprintf(page, "%d.%03d\n", msec/1000, msec%1000);
2944 safe_delay_store(mddev_t *mddev, const char *cbuf, size_t len)
2948 if (strict_strtoul_scaled(cbuf, &msec, 3) < 0)
2951 mddev->safemode_delay = 0;
2953 unsigned long old_delay = mddev->safemode_delay;
2954 mddev->safemode_delay = (msec*HZ)/1000;
2955 if (mddev->safemode_delay == 0)
2956 mddev->safemode_delay = 1;
2957 if (mddev->safemode_delay < old_delay)
2958 md_safemode_timeout((unsigned long)mddev);
2962 static struct md_sysfs_entry md_safe_delay =
2963 __ATTR(safe_mode_delay, S_IRUGO|S_IWUSR,safe_delay_show, safe_delay_store);
2966 level_show(mddev_t *mddev, char *page)
2968 struct mdk_personality *p = mddev->pers;
2970 return sprintf(page, "%s\n", p->name);
2971 else if (mddev->clevel[0])
2972 return sprintf(page, "%s\n", mddev->clevel);
2973 else if (mddev->level != LEVEL_NONE)
2974 return sprintf(page, "%d\n", mddev->level);
2980 level_store(mddev_t *mddev, const char *buf, size_t len)
2984 struct mdk_personality *pers;
2989 if (mddev->pers == NULL) {
2992 if (len >= sizeof(mddev->clevel))
2994 strncpy(mddev->clevel, buf, len);
2995 if (mddev->clevel[len-1] == '\n')
2997 mddev->clevel[len] = 0;
2998 mddev->level = LEVEL_NONE;
3002 /* request to change the personality. Need to ensure:
3003 * - array is not engaged in resync/recovery/reshape
3004 * - old personality can be suspended
3005 * - new personality will access other array.
3008 if (mddev->sync_thread ||
3009 mddev->reshape_position != MaxSector ||
3010 mddev->sysfs_active)
3013 if (!mddev->pers->quiesce) {
3014 printk(KERN_WARNING "md: %s: %s does not support online personality change\n",
3015 mdname(mddev), mddev->pers->name);
3019 /* Now find the new personality */
3020 if (len == 0 || len >= sizeof(clevel))
3022 strncpy(clevel, buf, len);
3023 if (clevel[len-1] == '\n')
3026 if (strict_strtol(clevel, 10, &level))
3029 if (request_module("md-%s", clevel) != 0)
3030 request_module("md-level-%s", clevel);
3031 spin_lock(&pers_lock);
3032 pers = find_pers(level, clevel);
3033 if (!pers || !try_module_get(pers->owner)) {
3034 spin_unlock(&pers_lock);
3035 printk(KERN_WARNING "md: personality %s not loaded\n", clevel);
3038 spin_unlock(&pers_lock);
3040 if (pers == mddev->pers) {
3041 /* Nothing to do! */
3042 module_put(pers->owner);
3045 if (!pers->takeover) {
3046 module_put(pers->owner);
3047 printk(KERN_WARNING "md: %s: %s does not support personality takeover\n",
3048 mdname(mddev), clevel);
3052 list_for_each_entry(rdev, &mddev->disks, same_set)
3053 rdev->new_raid_disk = rdev->raid_disk;
3055 /* ->takeover must set new_* and/or delta_disks
3056 * if it succeeds, and may set them when it fails.
3058 priv = pers->takeover(mddev);
3060 mddev->new_level = mddev->level;
3061 mddev->new_layout = mddev->layout;
3062 mddev->new_chunk_sectors = mddev->chunk_sectors;
3063 mddev->raid_disks -= mddev->delta_disks;
3064 mddev->delta_disks = 0;
3065 module_put(pers->owner);
3066 printk(KERN_WARNING "md: %s: %s would not accept array\n",
3067 mdname(mddev), clevel);
3068 return PTR_ERR(priv);
3071 /* Looks like we have a winner */
3072 mddev_suspend(mddev);
3073 mddev->pers->stop(mddev);
3075 if (mddev->pers->sync_request == NULL &&
3076 pers->sync_request != NULL) {
3077 /* need to add the md_redundancy_group */
3078 if (sysfs_create_group(&mddev->kobj, &md_redundancy_group))
3080 "md: cannot register extra attributes for %s\n",
3082 mddev->sysfs_action = sysfs_get_dirent(mddev->kobj.sd, NULL, "sync_action");
3084 if (mddev->pers->sync_request != NULL &&
3085 pers->sync_request == NULL) {
3086 /* need to remove the md_redundancy_group */
3087 if (mddev->to_remove == NULL)
3088 mddev->to_remove = &md_redundancy_group;
3091 if (mddev->pers->sync_request == NULL &&
3093 /* We are converting from a no-redundancy array
3094 * to a redundancy array and metadata is managed
3095 * externally so we need to be sure that writes
3096 * won't block due to a need to transition
3098 * until external management is started.
3101 mddev->safemode_delay = 0;
3102 mddev->safemode = 0;
3105 list_for_each_entry(rdev, &mddev->disks, same_set) {
3107 if (rdev->raid_disk < 0)
3109 if (rdev->new_raid_disk > mddev->raid_disks)
3110 rdev->new_raid_disk = -1;
3111 if (rdev->new_raid_disk == rdev->raid_disk)
3113 sprintf(nm, "rd%d", rdev->raid_disk);
3114 sysfs_remove_link(&mddev->kobj, nm);
3116 list_for_each_entry(rdev, &mddev->disks, same_set) {
3117 if (rdev->raid_disk < 0)
3119 if (rdev->new_raid_disk == rdev->raid_disk)
3121 rdev->raid_disk = rdev->new_raid_disk;
3122 if (rdev->raid_disk < 0)
3123 clear_bit(In_sync, &rdev->flags);
3126 sprintf(nm, "rd%d", rdev->raid_disk);
3127 if(sysfs_create_link(&mddev->kobj, &rdev->kobj, nm))
3128 printk("md: cannot register %s for %s after level change\n",
3133 module_put(mddev->pers->owner);
3135 mddev->private = priv;
3136 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
3137 mddev->level = mddev->new_level;
3138 mddev->layout = mddev->new_layout;
3139 mddev->chunk_sectors = mddev->new_chunk_sectors;
3140 mddev->delta_disks = 0;
3141 if (mddev->pers->sync_request == NULL) {
3142 /* this is now an array without redundancy, so
3143 * it must always be in_sync
3146 del_timer_sync(&mddev->safemode_timer);
3149 mddev_resume(mddev);
3150 set_bit(MD_CHANGE_DEVS, &mddev->flags);
3151 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3152 md_wakeup_thread(mddev->thread);
3153 sysfs_notify(&mddev->kobj, NULL, "level");
3154 md_new_event(mddev);
3158 static struct md_sysfs_entry md_level =
3159 __ATTR(level, S_IRUGO|S_IWUSR, level_show, level_store);
3163 layout_show(mddev_t *mddev, char *page)
3165 /* just a number, not meaningful for all levels */
3166 if (mddev->reshape_position != MaxSector &&
3167 mddev->layout != mddev->new_layout)
3168 return sprintf(page, "%d (%d)\n",
3169 mddev->new_layout, mddev->layout);
3170 return sprintf(page, "%d\n", mddev->layout);
3174 layout_store(mddev_t *mddev, const char *buf, size_t len)
3177 unsigned long n = simple_strtoul(buf, &e, 10);
3179 if (!*buf || (*e && *e != '\n'))
3184 if (mddev->pers->check_reshape == NULL)
3186 mddev->new_layout = n;
3187 err = mddev->pers->check_reshape(mddev);
3189 mddev->new_layout = mddev->layout;
3193 mddev->new_layout = n;
3194 if (mddev->reshape_position == MaxSector)
3199 static struct md_sysfs_entry md_layout =
3200 __ATTR(layout, S_IRUGO|S_IWUSR, layout_show, layout_store);
3204 raid_disks_show(mddev_t *mddev, char *page)
3206 if (mddev->raid_disks == 0)
3208 if (mddev->reshape_position != MaxSector &&
3209 mddev->delta_disks != 0)
3210 return sprintf(page, "%d (%d)\n", mddev->raid_disks,
3211 mddev->raid_disks - mddev->delta_disks);
3212 return sprintf(page, "%d\n", mddev->raid_disks);
3215 static int update_raid_disks(mddev_t *mddev, int raid_disks);
3218 raid_disks_store(mddev_t *mddev, const char *buf, size_t len)
3222 unsigned long n = simple_strtoul(buf, &e, 10);
3224 if (!*buf || (*e && *e != '\n'))
3228 rv = update_raid_disks(mddev, n);
3229 else if (mddev->reshape_position != MaxSector) {
3230 int olddisks = mddev->raid_disks - mddev->delta_disks;
3231 mddev->delta_disks = n - olddisks;
3232 mddev->raid_disks = n;
3234 mddev->raid_disks = n;
3235 return rv ? rv : len;
3237 static struct md_sysfs_entry md_raid_disks =
3238 __ATTR(raid_disks, S_IRUGO|S_IWUSR, raid_disks_show, raid_disks_store);
3241 chunk_size_show(mddev_t *mddev, char *page)
3243 if (mddev->reshape_position != MaxSector &&
3244 mddev->chunk_sectors != mddev->new_chunk_sectors)
3245 return sprintf(page, "%d (%d)\n",
3246 mddev->new_chunk_sectors << 9,
3247 mddev->chunk_sectors << 9);
3248 return sprintf(page, "%d\n", mddev->chunk_sectors << 9);
3252 chunk_size_store(mddev_t *mddev, const char *buf, size_t len)
3255 unsigned long n = simple_strtoul(buf, &e, 10);
3257 if (!*buf || (*e && *e != '\n'))
3262 if (mddev->pers->check_reshape == NULL)
3264 mddev->new_chunk_sectors = n >> 9;
3265 err = mddev->pers->check_reshape(mddev);
3267 mddev->new_chunk_sectors = mddev->chunk_sectors;
3271 mddev->new_chunk_sectors = n >> 9;
3272 if (mddev->reshape_position == MaxSector)
3273 mddev->chunk_sectors = n >> 9;
3277 static struct md_sysfs_entry md_chunk_size =
3278 __ATTR(chunk_size, S_IRUGO|S_IWUSR, chunk_size_show, chunk_size_store);
3281 resync_start_show(mddev_t *mddev, char *page)
3283 if (mddev->recovery_cp == MaxSector)
3284 return sprintf(page, "none\n");
3285 return sprintf(page, "%llu\n", (unsigned long long)mddev->recovery_cp);
3289 resync_start_store(mddev_t *mddev, const char *buf, size_t len)
3292 unsigned long long n = simple_strtoull(buf, &e, 10);
3296 if (cmd_match(buf, "none"))
3298 else if (!*buf || (*e && *e != '\n'))
3301 mddev->recovery_cp = n;
3304 static struct md_sysfs_entry md_resync_start =
3305 __ATTR(resync_start, S_IRUGO|S_IWUSR, resync_start_show, resync_start_store);
3308 * The array state can be:
3311 * No devices, no size, no level
3312 * Equivalent to STOP_ARRAY ioctl
3314 * May have some settings, but array is not active
3315 * all IO results in error
3316 * When written, doesn't tear down array, but just stops it
3317 * suspended (not supported yet)
3318 * All IO requests will block. The array can be reconfigured.
3319 * Writing this, if accepted, will block until array is quiescent
3321 * no resync can happen. no superblocks get written.
3322 * write requests fail
3324 * like readonly, but behaves like 'clean' on a write request.
3326 * clean - no pending writes, but otherwise active.
3327 * When written to inactive array, starts without resync
3328 * If a write request arrives then
3329 * if metadata is known, mark 'dirty' and switch to 'active'.
3330 * if not known, block and switch to write-pending
3331 * If written to an active array that has pending writes, then fails.
3333 * fully active: IO and resync can be happening.
3334 * When written to inactive array, starts with resync
3337 * clean, but writes are blocked waiting for 'active' to be written.
3340 * like active, but no writes have been seen for a while (100msec).
3343 enum array_state { clear, inactive, suspended, readonly, read_auto, clean, active,
3344 write_pending, active_idle, bad_word};
3345 static char *array_states[] = {
3346 "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active",
3347 "write-pending", "active-idle", NULL };
3349 static int match_word(const char *word, char **list)
3352 for (n=0; list[n]; n++)
3353 if (cmd_match(word, list[n]))
3359 array_state_show(mddev_t *mddev, char *page)
3361 enum array_state st = inactive;
3374 else if (test_bit(MD_CHANGE_PENDING, &mddev->flags))
3376 else if (mddev->safemode)
3382 if (list_empty(&mddev->disks) &&
3383 mddev->raid_disks == 0 &&
3384 mddev->dev_sectors == 0)
3389 return sprintf(page, "%s\n", array_states[st]);
3392 static int do_md_stop(mddev_t * mddev, int ro, int is_open);
3393 static int md_set_readonly(mddev_t * mddev, int is_open);
3394 static int do_md_run(mddev_t * mddev);
3395 static int restart_array(mddev_t *mddev);
3398 array_state_store(mddev_t *mddev, const char *buf, size_t len)
3401 enum array_state st = match_word(buf, array_states);
3406 /* stopping an active array */
3407 if (atomic_read(&mddev->openers) > 0)
3409 err = do_md_stop(mddev, 0, 0);
3412 /* stopping an active array */
3414 if (atomic_read(&mddev->openers) > 0)
3416 err = do_md_stop(mddev, 2, 0);
3418 err = 0; /* already inactive */
3421 break; /* not supported yet */
3424 err = md_set_readonly(mddev, 0);
3427 set_disk_ro(mddev->gendisk, 1);
3428 err = do_md_run(mddev);
3434 err = md_set_readonly(mddev, 0);
3435 else if (mddev->ro == 1)
3436 err = restart_array(mddev);
3439 set_disk_ro(mddev->gendisk, 0);
3443 err = do_md_run(mddev);
3448 restart_array(mddev);
3449 spin_lock_irq(&mddev->write_lock);
3450 if (atomic_read(&mddev->writes_pending) == 0) {
3451 if (mddev->in_sync == 0) {
3453 if (mddev->safemode == 1)
3454 mddev->safemode = 0;
3455 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
3460 spin_unlock_irq(&mddev->write_lock);
3466 restart_array(mddev);
3467 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
3468 wake_up(&mddev->sb_wait);
3472 set_disk_ro(mddev->gendisk, 0);
3473 err = do_md_run(mddev);
3478 /* these cannot be set */
3484 sysfs_notify_dirent_safe(mddev->sysfs_state);
3488 static struct md_sysfs_entry md_array_state =
3489 __ATTR(array_state, S_IRUGO|S_IWUSR, array_state_show, array_state_store);
3492 max_corrected_read_errors_show(mddev_t *mddev, char *page) {
3493 return sprintf(page, "%d\n",
3494 atomic_read(&mddev->max_corr_read_errors));
3498 max_corrected_read_errors_store(mddev_t *mddev, const char *buf, size_t len)
3501 unsigned long n = simple_strtoul(buf, &e, 10);
3503 if (*buf && (*e == 0 || *e == '\n')) {
3504 atomic_set(&mddev->max_corr_read_errors, n);
3510 static struct md_sysfs_entry max_corr_read_errors =
3511 __ATTR(max_read_errors, S_IRUGO|S_IWUSR, max_corrected_read_errors_show,
3512 max_corrected_read_errors_store);
3515 null_show(mddev_t *mddev, char *page)
3521 new_dev_store(mddev_t *mddev, const char *buf, size_t len)
3523 /* buf must be %d:%d\n? giving major and minor numbers */
3524 /* The new device is added to the array.
3525 * If the array has a persistent superblock, we read the
3526 * superblock to initialise info and check validity.
3527 * Otherwise, only checking done is that in bind_rdev_to_array,
3528 * which mainly checks size.
3531 int major = simple_strtoul(buf, &e, 10);
3537 if (!*buf || *e != ':' || !e[1] || e[1] == '\n')
3539 minor = simple_strtoul(e+1, &e, 10);
3540 if (*e && *e != '\n')
3542 dev = MKDEV(major, minor);
3543 if (major != MAJOR(dev) ||
3544 minor != MINOR(dev))
3548 if (mddev->persistent) {
3549 rdev = md_import_device(dev, mddev->major_version,
3550 mddev->minor_version);
3551 if (!IS_ERR(rdev) && !list_empty(&mddev->disks)) {
3552 mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
3553 mdk_rdev_t, same_set);
3554 err = super_types[mddev->major_version]
3555 .load_super(rdev, rdev0, mddev->minor_version);
3559 } else if (mddev->external)
3560 rdev = md_import_device(dev, -2, -1);
3562 rdev = md_import_device(dev, -1, -1);
3565 return PTR_ERR(rdev);
3566 err = bind_rdev_to_array(rdev, mddev);
3570 return err ? err : len;
3573 static struct md_sysfs_entry md_new_device =
3574 __ATTR(new_dev, S_IWUSR, null_show, new_dev_store);
3577 bitmap_store(mddev_t *mddev, const char *buf, size_t len)
3580 unsigned long chunk, end_chunk;
3584 /* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */
3586 chunk = end_chunk = simple_strtoul(buf, &end, 0);
3587 if (buf == end) break;
3588 if (*end == '-') { /* range */
3590 end_chunk = simple_strtoul(buf, &end, 0);
3591 if (buf == end) break;
3593 if (*end && !isspace(*end)) break;
3594 bitmap_dirty_bits(mddev->bitmap, chunk, end_chunk);
3595 buf = skip_spaces(end);
3597 bitmap_unplug(mddev->bitmap); /* flush the bits to disk */
3602 static struct md_sysfs_entry md_bitmap =
3603 __ATTR(bitmap_set_bits, S_IWUSR, null_show, bitmap_store);
3606 size_show(mddev_t *mddev, char *page)
3608 return sprintf(page, "%llu\n",
3609 (unsigned long long)mddev->dev_sectors / 2);
3612 static int update_size(mddev_t *mddev, sector_t num_sectors);
3615 size_store(mddev_t *mddev, const char *buf, size_t len)
3617 /* If array is inactive, we can reduce the component size, but
3618 * not increase it (except from 0).
3619 * If array is active, we can try an on-line resize
3622 int err = strict_blocks_to_sectors(buf, §ors);
3627 err = update_size(mddev, sectors);
3628 md_update_sb(mddev, 1);
3630 if (mddev->dev_sectors == 0 ||
3631 mddev->dev_sectors > sectors)
3632 mddev->dev_sectors = sectors;
3636 return err ? err : len;
3639 static struct md_sysfs_entry md_size =
3640 __ATTR(component_size, S_IRUGO|S_IWUSR, size_show, size_store);
3645 * 'none' for arrays with no metadata (good luck...)
3646 * 'external' for arrays with externally managed metadata,
3647 * or N.M for internally known formats
3650 metadata_show(mddev_t *mddev, char *page)
3652 if (mddev->persistent)
3653 return sprintf(page, "%d.%d\n",
3654 mddev->major_version, mddev->minor_version);
3655 else if (mddev->external)
3656 return sprintf(page, "external:%s\n", mddev->metadata_type);
3658 return sprintf(page, "none\n");
3662 metadata_store(mddev_t *mddev, const char *buf, size_t len)
3666 /* Changing the details of 'external' metadata is
3667 * always permitted. Otherwise there must be
3668 * no devices attached to the array.
3670 if (mddev->external && strncmp(buf, "external:", 9) == 0)
3672 else if (!list_empty(&mddev->disks))
3675 if (cmd_match(buf, "none")) {
3676 mddev->persistent = 0;
3677 mddev->external = 0;
3678 mddev->major_version = 0;
3679 mddev->minor_version = 90;
3682 if (strncmp(buf, "external:", 9) == 0) {
3683 size_t namelen = len-9;
3684 if (namelen >= sizeof(mddev->metadata_type))
3685 namelen = sizeof(mddev->metadata_type)-1;
3686 strncpy(mddev->metadata_type, buf+9, namelen);
3687 mddev->metadata_type[namelen] = 0;
3688 if (namelen && mddev->metadata_type[namelen-1] == '\n')
3689 mddev->metadata_type[--namelen] = 0;
3690 mddev->persistent = 0;
3691 mddev->external = 1;
3692 mddev->major_version = 0;
3693 mddev->minor_version = 90;
3696 major = simple_strtoul(buf, &e, 10);
3697 if (e==buf || *e != '.')
3700 minor = simple_strtoul(buf, &e, 10);
3701 if (e==buf || (*e && *e != '\n') )
3703 if (major >= ARRAY_SIZE(super_types) || super_types[major].name == NULL)
3705 mddev->major_version = major;
3706 mddev->minor_version = minor;
3707 mddev->persistent = 1;
3708 mddev->external = 0;
3712 static struct md_sysfs_entry md_metadata =
3713 __ATTR(metadata_version, S_IRUGO|S_IWUSR, metadata_show, metadata_store);
3716 action_show(mddev_t *mddev, char *page)
3718 char *type = "idle";
3719 if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
3721 else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
3722 (!mddev->ro && test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))) {
3723 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
3725 else if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
3726 if (!test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
3728 else if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
3732 } else if (test_bit(MD_RECOVERY_RECOVER, &mddev->recovery))
3735 return sprintf(page, "%s\n", type);
3739 action_store(mddev_t *mddev, const char *page, size_t len)
3741 if (!mddev->pers || !mddev->pers->sync_request)
3744 if (cmd_match(page, "frozen"))
3745 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
3747 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
3749 if (cmd_match(page, "idle") || cmd_match(page, "frozen")) {
3750 if (mddev->sync_thread) {
3751 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
3752 md_unregister_thread(mddev->sync_thread);
3753 mddev->sync_thread = NULL;
3754 mddev->recovery = 0;
3756 } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
3757 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
3759 else if (cmd_match(page, "resync"))
3760 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3761 else if (cmd_match(page, "recover")) {
3762 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
3763 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3764 } else if (cmd_match(page, "reshape")) {
3766 if (mddev->pers->start_reshape == NULL)
3768 err = mddev->pers->start_reshape(mddev);
3771 sysfs_notify(&mddev->kobj, NULL, "degraded");
3773 if (cmd_match(page, "check"))
3774 set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
3775 else if (!cmd_match(page, "repair"))
3777 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
3778 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
3780 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3781 md_wakeup_thread(mddev->thread);
3782 sysfs_notify_dirent_safe(mddev->sysfs_action);
3787 mismatch_cnt_show(mddev_t *mddev, char *page)
3789 return sprintf(page, "%llu\n",
3790 (unsigned long long) mddev->resync_mismatches);
3793 static struct md_sysfs_entry md_scan_mode =
3794 __ATTR(sync_action, S_IRUGO|S_IWUSR, action_show, action_store);
3797 static struct md_sysfs_entry md_mismatches = __ATTR_RO(mismatch_cnt);
3800 sync_min_show(mddev_t *mddev, char *page)
3802 return sprintf(page, "%d (%s)\n", speed_min(mddev),
3803 mddev->sync_speed_min ? "local": "system");
3807 sync_min_store(mddev_t *mddev, const char *buf, size_t len)
3811 if (strncmp(buf, "system", 6)==0) {
3812 mddev->sync_speed_min = 0;
3815 min = simple_strtoul(buf, &e, 10);
3816 if (buf == e || (*e && *e != '\n') || min <= 0)
3818 mddev->sync_speed_min = min;
3822 static struct md_sysfs_entry md_sync_min =
3823 __ATTR(sync_speed_min, S_IRUGO|S_IWUSR, sync_min_show, sync_min_store);
3826 sync_max_show(mddev_t *mddev, char *page)
3828 return sprintf(page, "%d (%s)\n", speed_max(mddev),
3829 mddev->sync_speed_max ? "local": "system");
3833 sync_max_store(mddev_t *mddev, const char *buf, size_t len)
3837 if (strncmp(buf, "system", 6)==0) {
3838 mddev->sync_speed_max = 0;
3841 max = simple_strtoul(buf, &e, 10);
3842 if (buf == e || (*e && *e != '\n') || max <= 0)
3844 mddev->sync_speed_max = max;
3848 static struct md_sysfs_entry md_sync_max =
3849 __ATTR(sync_speed_max, S_IRUGO|S_IWUSR, sync_max_show, sync_max_store);
3852 degraded_show(mddev_t *mddev, char *page)
3854 return sprintf(page, "%d\n", mddev->degraded);
3856 static struct md_sysfs_entry md_degraded = __ATTR_RO(degraded);
3859 sync_force_parallel_show(mddev_t *mddev, char *page)
3861 return sprintf(page, "%d\n", mddev->parallel_resync);
3865 sync_force_parallel_store(mddev_t *mddev, const char *buf, size_t len)
3869 if (strict_strtol(buf, 10, &n))
3872 if (n != 0 && n != 1)
3875 mddev->parallel_resync = n;
3877 if (mddev->sync_thread)
3878 wake_up(&resync_wait);
3883 /* force parallel resync, even with shared block devices */
3884 static struct md_sysfs_entry md_sync_force_parallel =
3885 __ATTR(sync_force_parallel, S_IRUGO|S_IWUSR,
3886 sync_force_parallel_show, sync_force_parallel_store);
3889 sync_speed_show(mddev_t *mddev, char *page)
3891 unsigned long resync, dt, db;
3892 if (mddev->curr_resync == 0)
3893 return sprintf(page, "none\n");
3894 resync = mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active);
3895 dt = (jiffies - mddev->resync_mark) / HZ;
3897 db = resync - mddev->resync_mark_cnt;
3898 return sprintf(page, "%lu\n", db/dt/2); /* K/sec */
3901 static struct md_sysfs_entry md_sync_speed = __ATTR_RO(sync_speed);
3904 sync_completed_show(mddev_t *mddev, char *page)
3906 unsigned long max_sectors, resync;
3908 if (!test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
3909 return sprintf(page, "none\n");
3911 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
3912 max_sectors = mddev->resync_max_sectors;
3914 max_sectors = mddev->dev_sectors;
3916 resync = mddev->curr_resync_completed;
3917 return sprintf(page, "%lu / %lu\n", resync, max_sectors);
3920 static struct md_sysfs_entry md_sync_completed = __ATTR_RO(sync_completed);
3923 min_sync_show(mddev_t *mddev, char *page)
3925 return sprintf(page, "%llu\n",
3926 (unsigned long long)mddev->resync_min);
3929 min_sync_store(mddev_t *mddev, const char *buf, size_t len)
3931 unsigned long long min;
3932 if (strict_strtoull(buf, 10, &min))
3934 if (min > mddev->resync_max)
3936 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
3939 /* Must be a multiple of chunk_size */
3940 if (mddev->chunk_sectors) {
3941 sector_t temp = min;
3942 if (sector_div(temp, mddev->chunk_sectors))
3945 mddev->resync_min = min;
3950 static struct md_sysfs_entry md_min_sync =
3951 __ATTR(sync_min, S_IRUGO|S_IWUSR, min_sync_show, min_sync_store);
3954 max_sync_show(mddev_t *mddev, char *page)
3956 if (mddev->resync_max == MaxSector)
3957 return sprintf(page, "max\n");
3959 return sprintf(page, "%llu\n",
3960 (unsigned long long)mddev->resync_max);
3963 max_sync_store(mddev_t *mddev, const char *buf, size_t len)
3965 if (strncmp(buf, "max", 3) == 0)
3966 mddev->resync_max = MaxSector;
3968 unsigned long long max;
3969 if (strict_strtoull(buf, 10, &max))
3971 if (max < mddev->resync_min)
3973 if (max < mddev->resync_max &&
3975 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
3978 /* Must be a multiple of chunk_size */
3979 if (mddev->chunk_sectors) {
3980 sector_t temp = max;
3981 if (sector_div(temp, mddev->chunk_sectors))
3984 mddev->resync_max = max;
3986 wake_up(&mddev->recovery_wait);
3990 static struct md_sysfs_entry md_max_sync =
3991 __ATTR(sync_max, S_IRUGO|S_IWUSR, max_sync_show, max_sync_store);
3994 suspend_lo_show(mddev_t *mddev, char *page)
3996 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_lo);
4000 suspend_lo_store(mddev_t *mddev, const char *buf, size_t len)
4003 unsigned long long new = simple_strtoull(buf, &e, 10);
4005 if (mddev->pers == NULL ||
4006 mddev->pers->quiesce == NULL)
4008 if (buf == e || (*e && *e != '\n'))
4010 if (new >= mddev->suspend_hi ||
4011 (new > mddev->suspend_lo && new < mddev->suspend_hi)) {
4012 mddev->suspend_lo = new;
4013 mddev->pers->quiesce(mddev, 2);
4018 static struct md_sysfs_entry md_suspend_lo =
4019 __ATTR(suspend_lo, S_IRUGO|S_IWUSR, suspend_lo_show, suspend_lo_store);
4023 suspend_hi_show(mddev_t *mddev, char *page)
4025 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_hi);
4029 suspend_hi_store(mddev_t *mddev, const char *buf, size_t len)
4032 unsigned long long new = simple_strtoull(buf, &e, 10);
4034 if (mddev->pers == NULL ||
4035 mddev->pers->quiesce == NULL)
4037 if (buf == e || (*e && *e != '\n'))
4039 if ((new <= mddev->suspend_lo && mddev->suspend_lo >= mddev->suspend_hi) ||
4040 (new > mddev->suspend_lo && new > mddev->suspend_hi)) {
4041 mddev->suspend_hi = new;
4042 mddev->pers->quiesce(mddev, 1);
4043 mddev->pers->quiesce(mddev, 0);
4048 static struct md_sysfs_entry md_suspend_hi =
4049 __ATTR(suspend_hi, S_IRUGO|S_IWUSR, suspend_hi_show, suspend_hi_store);
4052 reshape_position_show(mddev_t *mddev, char *page)
4054 if (mddev->reshape_position != MaxSector)
4055 return sprintf(page, "%llu\n",
4056 (unsigned long long)mddev->reshape_position);
4057 strcpy(page, "none\n");
4062 reshape_position_store(mddev_t *mddev, const char *buf, size_t len)
4065 unsigned long long new = simple_strtoull(buf, &e, 10);
4068 if (buf == e || (*e && *e != '\n'))
4070 mddev->reshape_position = new;
4071 mddev->delta_disks = 0;
4072 mddev->new_level = mddev->level;
4073 mddev->new_layout = mddev->layout;
4074 mddev->new_chunk_sectors = mddev->chunk_sectors;
4078 static struct md_sysfs_entry md_reshape_position =
4079 __ATTR(reshape_position, S_IRUGO|S_IWUSR, reshape_position_show,
4080 reshape_position_store);
4083 array_size_show(mddev_t *mddev, char *page)
4085 if (mddev->external_size)
4086 return sprintf(page, "%llu\n",
4087 (unsigned long long)mddev->array_sectors/2);
4089 return sprintf(page, "default\n");
4093 array_size_store(mddev_t *mddev, const char *buf, size_t len)
4097 if (strncmp(buf, "default", 7) == 0) {
4099 sectors = mddev->pers->size(mddev, 0, 0);
4101 sectors = mddev->array_sectors;
4103 mddev->external_size = 0;
4105 if (strict_blocks_to_sectors(buf, §ors) < 0)
4107 if (mddev->pers && mddev->pers->size(mddev, 0, 0) < sectors)
4110 mddev->external_size = 1;
4113 mddev->array_sectors = sectors;
4114 set_capacity(mddev->gendisk, mddev->array_sectors);
4116 revalidate_disk(mddev->gendisk);
4121 static struct md_sysfs_entry md_array_size =
4122 __ATTR(array_size, S_IRUGO|S_IWUSR, array_size_show,
4125 static struct attribute *md_default_attrs[] = {
4128 &md_raid_disks.attr,
4129 &md_chunk_size.attr,
4131 &md_resync_start.attr,
4133 &md_new_device.attr,
4134 &md_safe_delay.attr,
4135 &md_array_state.attr,
4136 &md_reshape_position.attr,
4137 &md_array_size.attr,
4138 &max_corr_read_errors.attr,
4142 static struct attribute *md_redundancy_attrs[] = {
4144 &md_mismatches.attr,
4147 &md_sync_speed.attr,
4148 &md_sync_force_parallel.attr,
4149 &md_sync_completed.attr,
4152 &md_suspend_lo.attr,
4153 &md_suspend_hi.attr,
4158 static struct attribute_group md_redundancy_group = {
4160 .attrs = md_redundancy_attrs,
4165 md_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
4167 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
4168 mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
4173 rv = mddev_lock(mddev);
4175 rv = entry->show(mddev, page);
4176 mddev_unlock(mddev);
4182 md_attr_store(struct kobject *kobj, struct attribute *attr,
4183 const char *page, size_t length)
4185 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
4186 mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
4191 if (!capable(CAP_SYS_ADMIN))
4193 rv = mddev_lock(mddev);
4194 if (mddev->hold_active == UNTIL_IOCTL)
4195 mddev->hold_active = 0;
4197 rv = entry->store(mddev, page, length);
4198 mddev_unlock(mddev);
4203 static void md_free(struct kobject *ko)
4205 mddev_t *mddev = container_of(ko, mddev_t, kobj);
4207 if (mddev->sysfs_state)
4208 sysfs_put(mddev->sysfs_state);
4210 if (mddev->gendisk) {
4211 del_gendisk(mddev->gendisk);
4212 put_disk(mddev->gendisk);
4215 blk_cleanup_queue(mddev->queue);
4220 static const struct sysfs_ops md_sysfs_ops = {
4221 .show = md_attr_show,
4222 .store = md_attr_store,
4224 static struct kobj_type md_ktype = {
4226 .sysfs_ops = &md_sysfs_ops,
4227 .default_attrs = md_default_attrs,
4232 static void mddev_delayed_delete(struct work_struct *ws)
4234 mddev_t *mddev = container_of(ws, mddev_t, del_work);
4236 sysfs_remove_group(&mddev->kobj, &md_bitmap_group);
4237 kobject_del(&mddev->kobj);
4238 kobject_put(&mddev->kobj);
4241 static int md_alloc(dev_t dev, char *name)
4243 static DEFINE_MUTEX(disks_mutex);
4244 mddev_t *mddev = mddev_find(dev);
4245 struct gendisk *disk;
4254 partitioned = (MAJOR(mddev->unit) != MD_MAJOR);
4255 shift = partitioned ? MdpMinorShift : 0;
4256 unit = MINOR(mddev->unit) >> shift;
4258 /* wait for any previous instance if this device
4259 * to be completed removed (mddev_delayed_delete).
4261 flush_scheduled_work();
4263 mutex_lock(&disks_mutex);
4269 /* Need to ensure that 'name' is not a duplicate.
4272 spin_lock(&all_mddevs_lock);
4274 list_for_each_entry(mddev2, &all_mddevs, all_mddevs)
4275 if (mddev2->gendisk &&
4276 strcmp(mddev2->gendisk->disk_name, name) == 0) {
4277 spin_unlock(&all_mddevs_lock);
4280 spin_unlock(&all_mddevs_lock);
4284 mddev->queue = blk_alloc_queue(GFP_KERNEL);
4287 mddev->queue->queuedata = mddev;
4289 /* Can be unlocked because the queue is new: no concurrency */
4290 queue_flag_set_unlocked(QUEUE_FLAG_CLUSTER, mddev->queue);
4292 blk_queue_make_request(mddev->queue, md_make_request);
4294 disk = alloc_disk(1 << shift);
4296 blk_cleanup_queue(mddev->queue);
4297 mddev->queue = NULL;
4300 disk->major = MAJOR(mddev->unit);
4301 disk->first_minor = unit << shift;
4303 strcpy(disk->disk_name, name);
4304 else if (partitioned)
4305 sprintf(disk->disk_name, "md_d%d", unit);
4307 sprintf(disk->disk_name, "md%d", unit);
4308 disk->fops = &md_fops;
4309 disk->private_data = mddev;
4310 disk->queue = mddev->queue;
4311 /* Allow extended partitions. This makes the
4312 * 'mdp' device redundant, but we can't really
4315 disk->flags |= GENHD_FL_EXT_DEVT;
4317 mddev->gendisk = disk;
4318 error = kobject_init_and_add(&mddev->kobj, &md_ktype,
4319 &disk_to_dev(disk)->kobj, "%s", "md");
4321 /* This isn't possible, but as kobject_init_and_add is marked
4322 * __must_check, we must do something with the result
4324 printk(KERN_WARNING "md: cannot register %s/md - name in use\n",
4328 if (mddev->kobj.sd &&
4329 sysfs_create_group(&mddev->kobj, &md_bitmap_group))
4330 printk(KERN_DEBUG "pointless warning\n");
4332 mutex_unlock(&disks_mutex);
4333 if (!error && mddev->kobj.sd) {
4334 kobject_uevent(&mddev->kobj, KOBJ_ADD);
4335 mddev->sysfs_state = sysfs_get_dirent_safe(mddev->kobj.sd, "array_state");
4341 static struct kobject *md_probe(dev_t dev, int *part, void *data)
4343 md_alloc(dev, NULL);
4347 static int add_named_array(const char *val, struct kernel_param *kp)
4349 /* val must be "md_*" where * is not all digits.
4350 * We allocate an array with a large free minor number, and
4351 * set the name to val. val must not already be an active name.
4353 int len = strlen(val);
4354 char buf[DISK_NAME_LEN];
4356 while (len && val[len-1] == '\n')
4358 if (len >= DISK_NAME_LEN)
4360 strlcpy(buf, val, len+1);
4361 if (strncmp(buf, "md_", 3) != 0)
4363 return md_alloc(0, buf);
4366 static void md_safemode_timeout(unsigned long data)
4368 mddev_t *mddev = (mddev_t *) data;
4370 if (!atomic_read(&mddev->writes_pending)) {
4371 mddev->safemode = 1;
4372 if (mddev->external)
4373 sysfs_notify_dirent_safe(mddev->sysfs_state);
4375 md_wakeup_thread(mddev->thread);
4378 static int start_dirty_degraded;
4380 int md_run(mddev_t *mddev)
4384 struct mdk_personality *pers;
4386 if (list_empty(&mddev->disks))
4387 /* cannot run an array with no devices.. */
4392 /* Cannot run until previous stop completes properly */
4393 if (mddev->sysfs_active)
4397 * Analyze all RAID superblock(s)
4399 if (!mddev->raid_disks) {
4400 if (!mddev->persistent)
4405 if (mddev->level != LEVEL_NONE)
4406 request_module("md-level-%d", mddev->level);
4407 else if (mddev->clevel[0])
4408 request_module("md-%s", mddev->clevel);
4411 * Drop all container device buffers, from now on
4412 * the only valid external interface is through the md
4415 list_for_each_entry(rdev, &mddev->disks, same_set) {
4416 if (test_bit(Faulty, &rdev->flags))
4418 sync_blockdev(rdev->bdev);
4419 invalidate_bdev(rdev->bdev);
4421 /* perform some consistency tests on the device.
4422 * We don't want the data to overlap the metadata,
4423 * Internal Bitmap issues have been handled elsewhere.
4425 if (rdev->data_offset < rdev->sb_start) {
4426 if (mddev->dev_sectors &&
4427 rdev->data_offset + mddev->dev_sectors
4429 printk("md: %s: data overlaps metadata\n",
4434 if (rdev->sb_start + rdev->sb_size/512
4435 > rdev->data_offset) {
4436 printk("md: %s: metadata overlaps data\n",
4441 sysfs_notify_dirent_safe(rdev->sysfs_state);
4444 spin_lock(&pers_lock);
4445 pers = find_pers(mddev->level, mddev->clevel);
4446 if (!pers || !try_module_get(pers->owner)) {
4447 spin_unlock(&pers_lock);
4448 if (mddev->level != LEVEL_NONE)
4449 printk(KERN_WARNING "md: personality for level %d is not loaded!\n",
4452 printk(KERN_WARNING "md: personality for level %s is not loaded!\n",
4457 spin_unlock(&pers_lock);
4458 if (mddev->level != pers->level) {
4459 mddev->level = pers->level;
4460 mddev->new_level = pers->level;
4462 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
4464 if (mddev->reshape_position != MaxSector &&
4465 pers->start_reshape == NULL) {
4466 /* This personality cannot handle reshaping... */
4468 module_put(pers->owner);
4472 if (pers->sync_request) {
4473 /* Warn if this is a potentially silly
4476 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
4480 list_for_each_entry(rdev, &mddev->disks, same_set)
4481 list_for_each_entry(rdev2, &mddev->disks, same_set) {
4483 rdev->bdev->bd_contains ==
4484 rdev2->bdev->bd_contains) {
4486 "%s: WARNING: %s appears to be"
4487 " on the same physical disk as"
4490 bdevname(rdev->bdev,b),
4491 bdevname(rdev2->bdev,b2));
4498 "True protection against single-disk"
4499 " failure might be compromised.\n");
4502 mddev->recovery = 0;
4503 /* may be over-ridden by personality */
4504 mddev->resync_max_sectors = mddev->dev_sectors;
4506 mddev->barriers_work = 1;
4507 mddev->ok_start_degraded = start_dirty_degraded;
4509 if (start_readonly && mddev->ro == 0)
4510 mddev->ro = 2; /* read-only, but switch on first write */
4512 err = mddev->pers->run(mddev);
4514 printk(KERN_ERR "md: pers->run() failed ...\n");
4515 else if (mddev->pers->size(mddev, 0, 0) < mddev->array_sectors) {
4516 WARN_ONCE(!mddev->external_size, "%s: default size too small,"
4517 " but 'external_size' not in effect?\n", __func__);
4519 "md: invalid array_size %llu > default size %llu\n",
4520 (unsigned long long)mddev->array_sectors / 2,
4521 (unsigned long long)mddev->pers->size(mddev, 0, 0) / 2);
4523 mddev->pers->stop(mddev);
4525 if (err == 0 && mddev->pers->sync_request) {
4526 err = bitmap_create(mddev);
4528 printk(KERN_ERR "%s: failed to create bitmap (%d)\n",
4529 mdname(mddev), err);
4530 mddev->pers->stop(mddev);
4534 module_put(mddev->pers->owner);
4536 bitmap_destroy(mddev);
4539 if (mddev->pers->sync_request) {
4540 if (mddev->kobj.sd &&
4541 sysfs_create_group(&mddev->kobj, &md_redundancy_group))
4543 "md: cannot register extra attributes for %s\n",
4545 mddev->sysfs_action = sysfs_get_dirent_safe(mddev->kobj.sd, "sync_action");
4546 } else if (mddev->ro == 2) /* auto-readonly not meaningful */
4549 atomic_set(&mddev->writes_pending,0);
4550 atomic_set(&mddev->max_corr_read_errors,
4551 MD_DEFAULT_MAX_CORRECTED_READ_ERRORS);
4552 mddev->safemode = 0;
4553 mddev->safemode_timer.function = md_safemode_timeout;
4554 mddev->safemode_timer.data = (unsigned long) mddev;
4555 mddev->safemode_delay = (200 * HZ)/1000 +1; /* 200 msec delay */
4558 list_for_each_entry(rdev, &mddev->disks, same_set)
4559 if (rdev->raid_disk >= 0) {
4561 sprintf(nm, "rd%d", rdev->raid_disk);
4562 if (sysfs_create_link(&mddev->kobj, &rdev->kobj, nm))
4563 /* failure here is OK */;
4566 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4569 md_update_sb(mddev, 0);
4571 md_wakeup_thread(mddev->thread);
4572 md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
4574 md_new_event(mddev);
4575 sysfs_notify_dirent_safe(mddev->sysfs_state);
4576 sysfs_notify_dirent_safe(mddev->sysfs_action);
4577 sysfs_notify(&mddev->kobj, NULL, "degraded");
4580 EXPORT_SYMBOL_GPL(md_run);
4582 static int do_md_run(mddev_t *mddev)
4586 err = md_run(mddev);
4589 err = bitmap_load(mddev);
4591 bitmap_destroy(mddev);
4594 set_capacity(mddev->gendisk, mddev->array_sectors);
4595 revalidate_disk(mddev->gendisk);
4596 kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
4601 static int restart_array(mddev_t *mddev)
4603 struct gendisk *disk = mddev->gendisk;
4605 /* Complain if it has no devices */
4606 if (list_empty(&mddev->disks))
4612 mddev->safemode = 0;
4614 set_disk_ro(disk, 0);
4615 printk(KERN_INFO "md: %s switched to read-write mode.\n",
4617 /* Kick recovery or resync if necessary */
4618 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4619 md_wakeup_thread(mddev->thread);
4620 md_wakeup_thread(mddev->sync_thread);
4621 sysfs_notify_dirent_safe(mddev->sysfs_state);
4625 /* similar to deny_write_access, but accounts for our holding a reference
4626 * to the file ourselves */
4627 static int deny_bitmap_write_access(struct file * file)
4629 struct inode *inode = file->f_mapping->host;
4631 spin_lock(&inode->i_lock);
4632 if (atomic_read(&inode->i_writecount) > 1) {
4633 spin_unlock(&inode->i_lock);
4636 atomic_set(&inode->i_writecount, -1);
4637 spin_unlock(&inode->i_lock);
4642 void restore_bitmap_write_access(struct file *file)
4644 struct inode *inode = file->f_mapping->host;
4646 spin_lock(&inode->i_lock);
4647 atomic_set(&inode->i_writecount, 1);
4648 spin_unlock(&inode->i_lock);
4651 static void md_clean(mddev_t *mddev)
4653 mddev->array_sectors = 0;
4654 mddev->external_size = 0;
4655 mddev->dev_sectors = 0;
4656 mddev->raid_disks = 0;
4657 mddev->recovery_cp = 0;
4658 mddev->resync_min = 0;
4659 mddev->resync_max = MaxSector;
4660 mddev->reshape_position = MaxSector;
4661 mddev->external = 0;
4662 mddev->persistent = 0;
4663 mddev->level = LEVEL_NONE;
4664 mddev->clevel[0] = 0;
4667 mddev->metadata_type[0] = 0;
4668 mddev->chunk_sectors = 0;
4669 mddev->ctime = mddev->utime = 0;
4671 mddev->max_disks = 0;
4673 mddev->can_decrease_events = 0;
4674 mddev->delta_disks = 0;
4675 mddev->new_level = LEVEL_NONE;
4676 mddev->new_layout = 0;
4677 mddev->new_chunk_sectors = 0;
4678 mddev->curr_resync = 0;
4679 mddev->resync_mismatches = 0;
4680 mddev->suspend_lo = mddev->suspend_hi = 0;
4681 mddev->sync_speed_min = mddev->sync_speed_max = 0;
4682 mddev->recovery = 0;
4684 mddev->degraded = 0;
4685 mddev->barriers_work = 0;
4686 mddev->safemode = 0;
4687 mddev->bitmap_info.offset = 0;
4688 mddev->bitmap_info.default_offset = 0;
4689 mddev->bitmap_info.chunksize = 0;
4690 mddev->bitmap_info.daemon_sleep = 0;
4691 mddev->bitmap_info.max_write_behind = 0;
4695 void md_stop_writes(mddev_t *mddev)
4697 if (mddev->sync_thread) {
4698 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4699 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
4700 md_unregister_thread(mddev->sync_thread);
4701 mddev->sync_thread = NULL;
4704 del_timer_sync(&mddev->safemode_timer);
4706 bitmap_flush(mddev);
4707 md_super_wait(mddev);
4709 if (!mddev->in_sync || mddev->flags) {
4710 /* mark array as shutdown cleanly */
4712 md_update_sb(mddev, 1);
4715 EXPORT_SYMBOL_GPL(md_stop_writes);
4717 void md_stop(mddev_t *mddev)
4719 mddev->pers->stop(mddev);
4720 if (mddev->pers->sync_request && mddev->to_remove == NULL)
4721 mddev->to_remove = &md_redundancy_group;
4722 module_put(mddev->pers->owner);
4724 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4726 EXPORT_SYMBOL_GPL(md_stop);
4728 static int md_set_readonly(mddev_t *mddev, int is_open)
4731 mutex_lock(&mddev->open_mutex);
4732 if (atomic_read(&mddev->openers) > is_open) {
4733 printk("md: %s still in use.\n",mdname(mddev));
4738 md_stop_writes(mddev);
4744 set_disk_ro(mddev->gendisk, 1);
4745 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4746 sysfs_notify_dirent_safe(mddev->sysfs_state);
4750 mutex_unlock(&mddev->open_mutex);
4755 * 0 - completely stop and dis-assemble array
4756 * 2 - stop but do not disassemble array
4758 static int do_md_stop(mddev_t * mddev, int mode, int is_open)
4760 struct gendisk *disk = mddev->gendisk;
4763 mutex_lock(&mddev->open_mutex);
4764 if (atomic_read(&mddev->openers) > is_open ||
4765 mddev->sysfs_active) {
4766 printk("md: %s still in use.\n",mdname(mddev));
4767 mutex_unlock(&mddev->open_mutex);
4773 set_disk_ro(disk, 0);
4775 md_stop_writes(mddev);
4777 mddev->queue->merge_bvec_fn = NULL;
4778 mddev->queue->unplug_fn = NULL;
4779 mddev->queue->backing_dev_info.congested_fn = NULL;
4781 /* tell userspace to handle 'inactive' */
4782 sysfs_notify_dirent_safe(mddev->sysfs_state);
4784 list_for_each_entry(rdev, &mddev->disks, same_set)
4785 if (rdev->raid_disk >= 0) {
4787 sprintf(nm, "rd%d", rdev->raid_disk);
4788 sysfs_remove_link(&mddev->kobj, nm);
4791 set_capacity(disk, 0);
4792 mutex_unlock(&mddev->open_mutex);
4793 revalidate_disk(disk);
4798 mutex_unlock(&mddev->open_mutex);
4800 * Free resources if final stop
4803 printk(KERN_INFO "md: %s stopped.\n", mdname(mddev));
4805 bitmap_destroy(mddev);
4806 if (mddev->bitmap_info.file) {
4807 restore_bitmap_write_access(mddev->bitmap_info.file);
4808 fput(mddev->bitmap_info.file);
4809 mddev->bitmap_info.file = NULL;
4811 mddev->bitmap_info.offset = 0;
4813 export_array(mddev);
4816 kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
4817 if (mddev->hold_active == UNTIL_STOP)
4818 mddev->hold_active = 0;
4820 blk_integrity_unregister(disk);
4821 md_new_event(mddev);
4822 sysfs_notify_dirent_safe(mddev->sysfs_state);
4827 static void autorun_array(mddev_t *mddev)
4832 if (list_empty(&mddev->disks))
4835 printk(KERN_INFO "md: running: ");
4837 list_for_each_entry(rdev, &mddev->disks, same_set) {
4838 char b[BDEVNAME_SIZE];
4839 printk("<%s>", bdevname(rdev->bdev,b));
4843 err = do_md_run(mddev);
4845 printk(KERN_WARNING "md: do_md_run() returned %d\n", err);
4846 do_md_stop(mddev, 0, 0);
4851 * lets try to run arrays based on all disks that have arrived
4852 * until now. (those are in pending_raid_disks)
4854 * the method: pick the first pending disk, collect all disks with
4855 * the same UUID, remove all from the pending list and put them into
4856 * the 'same_array' list. Then order this list based on superblock
4857 * update time (freshest comes first), kick out 'old' disks and
4858 * compare superblocks. If everything's fine then run it.
4860 * If "unit" is allocated, then bump its reference count
4862 static void autorun_devices(int part)
4864 mdk_rdev_t *rdev0, *rdev, *tmp;
4866 char b[BDEVNAME_SIZE];
4868 printk(KERN_INFO "md: autorun ...\n");
4869 while (!list_empty(&pending_raid_disks)) {
4872 LIST_HEAD(candidates);
4873 rdev0 = list_entry(pending_raid_disks.next,
4874 mdk_rdev_t, same_set);
4876 printk(KERN_INFO "md: considering %s ...\n",
4877 bdevname(rdev0->bdev,b));
4878 INIT_LIST_HEAD(&candidates);
4879 rdev_for_each_list(rdev, tmp, &pending_raid_disks)
4880 if (super_90_load(rdev, rdev0, 0) >= 0) {
4881 printk(KERN_INFO "md: adding %s ...\n",
4882 bdevname(rdev->bdev,b));
4883 list_move(&rdev->same_set, &candidates);
4886 * now we have a set of devices, with all of them having
4887 * mostly sane superblocks. It's time to allocate the
4891 dev = MKDEV(mdp_major,
4892 rdev0->preferred_minor << MdpMinorShift);
4893 unit = MINOR(dev) >> MdpMinorShift;
4895 dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
4898 if (rdev0->preferred_minor != unit) {
4899 printk(KERN_INFO "md: unit number in %s is bad: %d\n",
4900 bdevname(rdev0->bdev, b), rdev0->preferred_minor);
4904 md_probe(dev, NULL, NULL);
4905 mddev = mddev_find(dev);
4906 if (!mddev || !mddev->gendisk) {
4910 "md: cannot allocate memory for md drive.\n");
4913 if (mddev_lock(mddev))
4914 printk(KERN_WARNING "md: %s locked, cannot run\n",
4916 else if (mddev->raid_disks || mddev->major_version
4917 || !list_empty(&mddev->disks)) {
4919 "md: %s already running, cannot run %s\n",
4920 mdname(mddev), bdevname(rdev0->bdev,b));
4921 mddev_unlock(mddev);
4923 printk(KERN_INFO "md: created %s\n", mdname(mddev));
4924 mddev->persistent = 1;
4925 rdev_for_each_list(rdev, tmp, &candidates) {
4926 list_del_init(&rdev->same_set);
4927 if (bind_rdev_to_array(rdev, mddev))
4930 autorun_array(mddev);
4931 mddev_unlock(mddev);
4933 /* on success, candidates will be empty, on error
4936 rdev_for_each_list(rdev, tmp, &candidates) {
4937 list_del_init(&rdev->same_set);
4942 printk(KERN_INFO "md: ... autorun DONE.\n");
4944 #endif /* !MODULE */
4946 static int get_version(void __user * arg)
4950 ver.major = MD_MAJOR_VERSION;
4951 ver.minor = MD_MINOR_VERSION;
4952 ver.patchlevel = MD_PATCHLEVEL_VERSION;
4954 if (copy_to_user(arg, &ver, sizeof(ver)))
4960 static int get_array_info(mddev_t * mddev, void __user * arg)
4962 mdu_array_info_t info;
4963 int nr,working,insync,failed,spare;
4966 nr=working=insync=failed=spare=0;
4967 list_for_each_entry(rdev, &mddev->disks, same_set) {
4969 if (test_bit(Faulty, &rdev->flags))
4973 if (test_bit(In_sync, &rdev->flags))
4980 info.major_version = mddev->major_version;
4981 info.minor_version = mddev->minor_version;
4982 info.patch_version = MD_PATCHLEVEL_VERSION;
4983 info.ctime = mddev->ctime;
4984 info.level = mddev->level;
4985 info.size = mddev->dev_sectors / 2;
4986 if (info.size != mddev->dev_sectors / 2) /* overflow */
4989 info.raid_disks = mddev->raid_disks;
4990 info.md_minor = mddev->md_minor;
4991 info.not_persistent= !mddev->persistent;
4993 info.utime = mddev->utime;
4996 info.state = (1<<MD_SB_CLEAN);
4997 if (mddev->bitmap && mddev->bitmap_info.offset)
4998 info.state = (1<<MD_SB_BITMAP_PRESENT);
4999 info.active_disks = insync;
5000 info.working_disks = working;
5001 info.failed_disks = failed;
5002 info.spare_disks = spare;
5004 info.layout = mddev->layout;
5005 info.chunk_size = mddev->chunk_sectors << 9;
5007 if (copy_to_user(arg, &info, sizeof(info)))
5013 static int get_bitmap_file(mddev_t * mddev, void __user * arg)
5015 mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */
5016 char *ptr, *buf = NULL;
5019 if (md_allow_write(mddev))
5020 file = kmalloc(sizeof(*file), GFP_NOIO);
5022 file = kmalloc(sizeof(*file), GFP_KERNEL);
5027 /* bitmap disabled, zero the first byte and copy out */
5028 if (!mddev->bitmap || !mddev->bitmap->file) {
5029 file->pathname[0] = '\0';
5033 buf = kmalloc(sizeof(file->pathname), GFP_KERNEL);
5037 ptr = d_path(&mddev->bitmap->file->f_path, buf, sizeof(file->pathname));
5041 strcpy(file->pathname, ptr);
5045 if (copy_to_user(arg, file, sizeof(*file)))
5053 static int get_disk_info(mddev_t * mddev, void __user * arg)
5055 mdu_disk_info_t info;
5058 if (copy_from_user(&info, arg, sizeof(info)))
5061 rdev = find_rdev_nr(mddev, info.number);
5063 info.major = MAJOR(rdev->bdev->bd_dev);
5064 info.minor = MINOR(rdev->bdev->bd_dev);
5065 info.raid_disk = rdev->raid_disk;
5067 if (test_bit(Faulty, &rdev->flags))
5068 info.state |= (1<<MD_DISK_FAULTY);
5069 else if (test_bit(In_sync, &rdev->flags)) {
5070 info.state |= (1<<MD_DISK_ACTIVE);
5071 info.state |= (1<<MD_DISK_SYNC);
5073 if (test_bit(WriteMostly, &rdev->flags))
5074 info.state |= (1<<MD_DISK_WRITEMOSTLY);
5076 info.major = info.minor = 0;
5077 info.raid_disk = -1;
5078 info.state = (1<<MD_DISK_REMOVED);
5081 if (copy_to_user(arg, &info, sizeof(info)))
5087 static int add_new_disk(mddev_t * mddev, mdu_disk_info_t *info)
5089 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
5091 dev_t dev = MKDEV(info->major,info->minor);
5093 if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
5096 if (!mddev->raid_disks) {
5098 /* expecting a device which has a superblock */
5099 rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
5102 "md: md_import_device returned %ld\n",
5104 return PTR_ERR(rdev);
5106 if (!list_empty(&mddev->disks)) {
5107 mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
5108 mdk_rdev_t, same_set);
5109 err = super_types[mddev->major_version]
5110 .load_super(rdev, rdev0, mddev->minor_version);
5113 "md: %s has different UUID to %s\n",
5114 bdevname(rdev->bdev,b),
5115 bdevname(rdev0->bdev,b2));
5120 err = bind_rdev_to_array(rdev, mddev);
5127 * add_new_disk can be used once the array is assembled
5128 * to add "hot spares". They must already have a superblock
5133 if (!mddev->pers->hot_add_disk) {
5135 "%s: personality does not support diskops!\n",
5139 if (mddev->persistent)
5140 rdev = md_import_device(dev, mddev->major_version,
5141 mddev->minor_version);
5143 rdev = md_import_device(dev, -1, -1);
5146 "md: md_import_device returned %ld\n",
5148 return PTR_ERR(rdev);
5150 /* set save_raid_disk if appropriate */
5151 if (!mddev->persistent) {
5152 if (info->state & (1<<MD_DISK_SYNC) &&
5153 info->raid_disk < mddev->raid_disks)
5154 rdev->raid_disk = info->raid_disk;
5156 rdev->raid_disk = -1;
5158 super_types[mddev->major_version].
5159 validate_super(mddev, rdev);
5160 rdev->saved_raid_disk = rdev->raid_disk;
5162 clear_bit(In_sync, &rdev->flags); /* just to be sure */
5163 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
5164 set_bit(WriteMostly, &rdev->flags);
5166 clear_bit(WriteMostly, &rdev->flags);
5168 rdev->raid_disk = -1;
5169 err = bind_rdev_to_array(rdev, mddev);
5170 if (!err && !mddev->pers->hot_remove_disk) {
5171 /* If there is hot_add_disk but no hot_remove_disk
5172 * then added disks for geometry changes,
5173 * and should be added immediately.
5175 super_types[mddev->major_version].
5176 validate_super(mddev, rdev);
5177 err = mddev->pers->hot_add_disk(mddev, rdev);
5179 unbind_rdev_from_array(rdev);
5184 sysfs_notify_dirent_safe(rdev->sysfs_state);
5186 md_update_sb(mddev, 1);
5187 if (mddev->degraded)
5188 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
5189 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5190 md_wakeup_thread(mddev->thread);
5194 /* otherwise, add_new_disk is only allowed
5195 * for major_version==0 superblocks
5197 if (mddev->major_version != 0) {
5198 printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n",
5203 if (!(info->state & (1<<MD_DISK_FAULTY))) {
5205 rdev = md_import_device(dev, -1, 0);
5208 "md: error, md_import_device() returned %ld\n",
5210 return PTR_ERR(rdev);
5212 rdev->desc_nr = info->number;
5213 if (info->raid_disk < mddev->raid_disks)
5214 rdev->raid_disk = info->raid_disk;
5216 rdev->raid_disk = -1;
5218 if (rdev->raid_disk < mddev->raid_disks)
5219 if (info->state & (1<<MD_DISK_SYNC))
5220 set_bit(In_sync, &rdev->flags);
5222 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
5223 set_bit(WriteMostly, &rdev->flags);
5225 if (!mddev->persistent) {
5226 printk(KERN_INFO "md: nonpersistent superblock ...\n");
5227 rdev->sb_start = rdev->bdev->bd_inode->i_size / 512;
5229 rdev->sb_start = calc_dev_sboffset(rdev->bdev);
5230 rdev->sectors = rdev->sb_start;
5232 err = bind_rdev_to_array(rdev, mddev);
5242 static int hot_remove_disk(mddev_t * mddev, dev_t dev)
5244 char b[BDEVNAME_SIZE];
5247 rdev = find_rdev(mddev, dev);
5251 if (rdev->raid_disk >= 0)
5254 kick_rdev_from_array(rdev);
5255 md_update_sb(mddev, 1);
5256 md_new_event(mddev);
5260 printk(KERN_WARNING "md: cannot remove active disk %s from %s ...\n",
5261 bdevname(rdev->bdev,b), mdname(mddev));
5265 static int hot_add_disk(mddev_t * mddev, dev_t dev)
5267 char b[BDEVNAME_SIZE];
5274 if (mddev->major_version != 0) {
5275 printk(KERN_WARNING "%s: HOT_ADD may only be used with"
5276 " version-0 superblocks.\n",
5280 if (!mddev->pers->hot_add_disk) {
5282 "%s: personality does not support diskops!\n",
5287 rdev = md_import_device(dev, -1, 0);
5290 "md: error, md_import_device() returned %ld\n",
5295 if (mddev->persistent)
5296 rdev->sb_start = calc_dev_sboffset(rdev->bdev);
5298 rdev->sb_start = rdev->bdev->bd_inode->i_size / 512;
5300 rdev->sectors = rdev->sb_start;
5302 if (test_bit(Faulty, &rdev->flags)) {
5304 "md: can not hot-add faulty %s disk to %s!\n",
5305 bdevname(rdev->bdev,b), mdname(mddev));
5309 clear_bit(In_sync, &rdev->flags);
5311 rdev->saved_raid_disk = -1;
5312 err = bind_rdev_to_array(rdev, mddev);
5317 * The rest should better be atomic, we can have disk failures
5318 * noticed in interrupt contexts ...
5321 rdev->raid_disk = -1;
5323 md_update_sb(mddev, 1);
5326 * Kick recovery, maybe this spare has to be added to the
5327 * array immediately.
5329 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5330 md_wakeup_thread(mddev->thread);
5331 md_new_event(mddev);
5339 static int set_bitmap_file(mddev_t *mddev, int fd)
5344 if (!mddev->pers->quiesce)
5346 if (mddev->recovery || mddev->sync_thread)
5348 /* we should be able to change the bitmap.. */
5354 return -EEXIST; /* cannot add when bitmap is present */
5355 mddev->bitmap_info.file = fget(fd);
5357 if (mddev->bitmap_info.file == NULL) {
5358 printk(KERN_ERR "%s: error: failed to get bitmap file\n",
5363 err = deny_bitmap_write_access(mddev->bitmap_info.file);
5365 printk(KERN_ERR "%s: error: bitmap file is already in use\n",
5367 fput(mddev->bitmap_info.file);
5368 mddev->bitmap_info.file = NULL;
5371 mddev->bitmap_info.offset = 0; /* file overrides offset */
5372 } else if (mddev->bitmap == NULL)
5373 return -ENOENT; /* cannot remove what isn't there */
5376 mddev->pers->quiesce(mddev, 1);
5378 err = bitmap_create(mddev);
5380 err = bitmap_load(mddev);
5382 if (fd < 0 || err) {
5383 bitmap_destroy(mddev);
5384 fd = -1; /* make sure to put the file */
5386 mddev->pers->quiesce(mddev, 0);
5389 if (mddev->bitmap_info.file) {
5390 restore_bitmap_write_access(mddev->bitmap_info.file);
5391 fput(mddev->bitmap_info.file);
5393 mddev->bitmap_info.file = NULL;
5400 * set_array_info is used two different ways
5401 * The original usage is when creating a new array.
5402 * In this usage, raid_disks is > 0 and it together with
5403 * level, size, not_persistent,layout,chunksize determine the
5404 * shape of the array.
5405 * This will always create an array with a type-0.90.0 superblock.
5406 * The newer usage is when assembling an array.
5407 * In this case raid_disks will be 0, and the major_version field is
5408 * use to determine which style super-blocks are to be found on the devices.
5409 * The minor and patch _version numbers are also kept incase the
5410 * super_block handler wishes to interpret them.
5412 static int set_array_info(mddev_t * mddev, mdu_array_info_t *info)
5415 if (info->raid_disks == 0) {
5416 /* just setting version number for superblock loading */
5417 if (info->major_version < 0 ||
5418 info->major_version >= ARRAY_SIZE(super_types) ||
5419 super_types[info->major_version].name == NULL) {
5420 /* maybe try to auto-load a module? */
5422 "md: superblock version %d not known\n",
5423 info->major_version);
5426 mddev->major_version = info->major_version;
5427 mddev->minor_version = info->minor_version;
5428 mddev->patch_version = info->patch_version;
5429 mddev->persistent = !info->not_persistent;
5430 /* ensure mddev_put doesn't delete this now that there
5431 * is some minimal configuration.
5433 mddev->ctime = get_seconds();
5436 mddev->major_version = MD_MAJOR_VERSION;
5437 mddev->minor_version = MD_MINOR_VERSION;
5438 mddev->patch_version = MD_PATCHLEVEL_VERSION;
5439 mddev->ctime = get_seconds();
5441 mddev->level = info->level;
5442 mddev->clevel[0] = 0;
5443 mddev->dev_sectors = 2 * (sector_t)info->size;
5444 mddev->raid_disks = info->raid_disks;
5445 /* don't set md_minor, it is determined by which /dev/md* was
5448 if (info->state & (1<<MD_SB_CLEAN))
5449 mddev->recovery_cp = MaxSector;
5451 mddev->recovery_cp = 0;
5452 mddev->persistent = ! info->not_persistent;
5453 mddev->external = 0;
5455 mddev->layout = info->layout;
5456 mddev->chunk_sectors = info->chunk_size >> 9;
5458 mddev->max_disks = MD_SB_DISKS;
5460 if (mddev->persistent)
5462 set_bit(MD_CHANGE_DEVS, &mddev->flags);
5464 mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9;
5465 mddev->bitmap_info.offset = 0;
5467 mddev->reshape_position = MaxSector;
5470 * Generate a 128 bit UUID
5472 get_random_bytes(mddev->uuid, 16);
5474 mddev->new_level = mddev->level;
5475 mddev->new_chunk_sectors = mddev->chunk_sectors;
5476 mddev->new_layout = mddev->layout;
5477 mddev->delta_disks = 0;
5482 void md_set_array_sectors(mddev_t *mddev, sector_t array_sectors)
5484 WARN(!mddev_is_locked(mddev), "%s: unlocked mddev!\n", __func__);
5486 if (mddev->external_size)
5489 mddev->array_sectors = array_sectors;
5491 EXPORT_SYMBOL(md_set_array_sectors);
5493 static int update_size(mddev_t *mddev, sector_t num_sectors)
5497 int fit = (num_sectors == 0);
5499 if (mddev->pers->resize == NULL)
5501 /* The "num_sectors" is the number of sectors of each device that
5502 * is used. This can only make sense for arrays with redundancy.
5503 * linear and raid0 always use whatever space is available. We can only
5504 * consider changing this number if no resync or reconstruction is
5505 * happening, and if the new size is acceptable. It must fit before the
5506 * sb_start or, if that is <data_offset, it must fit before the size
5507 * of each device. If num_sectors is zero, we find the largest size
5511 if (mddev->sync_thread)
5514 /* Sorry, cannot grow a bitmap yet, just remove it,
5518 list_for_each_entry(rdev, &mddev->disks, same_set) {
5519 sector_t avail = rdev->sectors;
5521 if (fit && (num_sectors == 0 || num_sectors > avail))
5522 num_sectors = avail;
5523 if (avail < num_sectors)
5526 rv = mddev->pers->resize(mddev, num_sectors);
5528 revalidate_disk(mddev->gendisk);
5532 static int update_raid_disks(mddev_t *mddev, int raid_disks)
5535 /* change the number of raid disks */
5536 if (mddev->pers->check_reshape == NULL)
5538 if (raid_disks <= 0 ||
5539 (mddev->max_disks && raid_disks >= mddev->max_disks))
5541 if (mddev->sync_thread || mddev->reshape_position != MaxSector)
5543 mddev->delta_disks = raid_disks - mddev->raid_disks;
5545 rv = mddev->pers->check_reshape(mddev);
5551 * update_array_info is used to change the configuration of an
5553 * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
5554 * fields in the info are checked against the array.
5555 * Any differences that cannot be handled will cause an error.
5556 * Normally, only one change can be managed at a time.
5558 static int update_array_info(mddev_t *mddev, mdu_array_info_t *info)
5564 /* calculate expected state,ignoring low bits */
5565 if (mddev->bitmap && mddev->bitmap_info.offset)
5566 state |= (1 << MD_SB_BITMAP_PRESENT);
5568 if (mddev->major_version != info->major_version ||
5569 mddev->minor_version != info->minor_version ||
5570 /* mddev->patch_version != info->patch_version || */
5571 mddev->ctime != info->ctime ||
5572 mddev->level != info->level ||
5573 /* mddev->layout != info->layout || */
5574 !mddev->persistent != info->not_persistent||
5575 mddev->chunk_sectors != info->chunk_size >> 9 ||
5576 /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
5577 ((state^info->state) & 0xfffffe00)
5580 /* Check there is only one change */
5581 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
5583 if (mddev->raid_disks != info->raid_disks)
5585 if (mddev->layout != info->layout)
5587 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT))
5594 if (mddev->layout != info->layout) {
5596 * we don't need to do anything at the md level, the
5597 * personality will take care of it all.
5599 if (mddev->pers->check_reshape == NULL)
5602 mddev->new_layout = info->layout;
5603 rv = mddev->pers->check_reshape(mddev);
5605 mddev->new_layout = mddev->layout;
5609 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
5610 rv = update_size(mddev, (sector_t)info->size * 2);
5612 if (mddev->raid_disks != info->raid_disks)
5613 rv = update_raid_disks(mddev, info->raid_disks);
5615 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) {
5616 if (mddev->pers->quiesce == NULL)
5618 if (mddev->recovery || mddev->sync_thread)
5620 if (info->state & (1<<MD_SB_BITMAP_PRESENT)) {
5621 /* add the bitmap */
5624 if (mddev->bitmap_info.default_offset == 0)
5626 mddev->bitmap_info.offset =
5627 mddev->bitmap_info.default_offset;
5628 mddev->pers->quiesce(mddev, 1);
5629 rv = bitmap_create(mddev);
5631 rv = bitmap_load(mddev);
5633 bitmap_destroy(mddev);
5634 mddev->pers->quiesce(mddev, 0);
5636 /* remove the bitmap */
5639 if (mddev->bitmap->file)
5641 mddev->pers->quiesce(mddev, 1);
5642 bitmap_destroy(mddev);
5643 mddev->pers->quiesce(mddev, 0);
5644 mddev->bitmap_info.offset = 0;
5647 md_update_sb(mddev, 1);
5651 static int set_disk_faulty(mddev_t *mddev, dev_t dev)
5655 if (mddev->pers == NULL)
5658 rdev = find_rdev(mddev, dev);
5662 md_error(mddev, rdev);
5667 * We have a problem here : there is no easy way to give a CHS
5668 * virtual geometry. We currently pretend that we have a 2 heads
5669 * 4 sectors (with a BIG number of cylinders...). This drives
5670 * dosfs just mad... ;-)
5672 static int md_getgeo(struct block_device *bdev, struct hd_geometry *geo)
5674 mddev_t *mddev = bdev->bd_disk->private_data;
5678 geo->cylinders = mddev->array_sectors / 8;
5682 static int md_ioctl(struct block_device *bdev, fmode_t mode,
5683 unsigned int cmd, unsigned long arg)
5686 void __user *argp = (void __user *)arg;
5687 mddev_t *mddev = NULL;
5690 if (!capable(CAP_SYS_ADMIN))
5694 * Commands dealing with the RAID driver but not any
5700 err = get_version(argp);
5703 case PRINT_RAID_DEBUG:
5711 autostart_arrays(arg);
5718 * Commands creating/starting a new array:
5721 mddev = bdev->bd_disk->private_data;
5728 err = mddev_lock(mddev);
5731 "md: ioctl lock interrupted, reason %d, cmd %d\n",
5738 case SET_ARRAY_INFO:
5740 mdu_array_info_t info;
5742 memset(&info, 0, sizeof(info));
5743 else if (copy_from_user(&info, argp, sizeof(info))) {
5748 err = update_array_info(mddev, &info);
5750 printk(KERN_WARNING "md: couldn't update"
5751 " array info. %d\n", err);
5756 if (!list_empty(&mddev->disks)) {
5758 "md: array %s already has disks!\n",
5763 if (mddev->raid_disks) {
5765 "md: array %s already initialised!\n",
5770 err = set_array_info(mddev, &info);
5772 printk(KERN_WARNING "md: couldn't set"
5773 " array info. %d\n", err);
5783 * Commands querying/configuring an existing array:
5785 /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
5786 * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */
5787 if ((!mddev->raid_disks && !mddev->external)
5788 && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY
5789 && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE
5790 && cmd != GET_BITMAP_FILE) {
5796 * Commands even a read-only array can execute:
5800 case GET_ARRAY_INFO:
5801 err = get_array_info(mddev, argp);
5804 case GET_BITMAP_FILE:
5805 err = get_bitmap_file(mddev, argp);
5809 err = get_disk_info(mddev, argp);
5812 case RESTART_ARRAY_RW:
5813 err = restart_array(mddev);
5817 err = do_md_stop(mddev, 0, 1);
5821 err = md_set_readonly(mddev, 1);
5825 if (get_user(ro, (int __user *)(arg))) {
5831 /* if the bdev is going readonly the value of mddev->ro
5832 * does not matter, no writes are coming
5837 /* are we are already prepared for writes? */
5841 /* transitioning to readauto need only happen for
5842 * arrays that call md_write_start
5845 err = restart_array(mddev);
5848 set_disk_ro(mddev->gendisk, 0);
5855 * The remaining ioctls are changing the state of the
5856 * superblock, so we do not allow them on read-only arrays.
5857 * However non-MD ioctls (e.g. get-size) will still come through
5858 * here and hit the 'default' below, so only disallow
5859 * 'md' ioctls, and switch to rw mode if started auto-readonly.
5861 if (_IOC_TYPE(cmd) == MD_MAJOR && mddev->ro && mddev->pers) {
5862 if (mddev->ro == 2) {
5864 sysfs_notify_dirent_safe(mddev->sysfs_state);
5865 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5866 md_wakeup_thread(mddev->thread);
5877 mdu_disk_info_t info;
5878 if (copy_from_user(&info, argp, sizeof(info)))
5881 err = add_new_disk(mddev, &info);
5885 case HOT_REMOVE_DISK:
5886 err = hot_remove_disk(mddev, new_decode_dev(arg));
5890 err = hot_add_disk(mddev, new_decode_dev(arg));
5893 case SET_DISK_FAULTY:
5894 err = set_disk_faulty(mddev, new_decode_dev(arg));
5898 err = do_md_run(mddev);
5901 case SET_BITMAP_FILE:
5902 err = set_bitmap_file(mddev, (int)arg);
5912 if (mddev->hold_active == UNTIL_IOCTL &&
5914 mddev->hold_active = 0;
5915 mddev_unlock(mddev);
5924 #ifdef CONFIG_COMPAT
5925 static int md_compat_ioctl(struct block_device *bdev, fmode_t mode,
5926 unsigned int cmd, unsigned long arg)
5929 case HOT_REMOVE_DISK:
5931 case SET_DISK_FAULTY:
5932 case SET_BITMAP_FILE:
5933 /* These take in integer arg, do not convert */
5936 arg = (unsigned long)compat_ptr(arg);
5940 return md_ioctl(bdev, mode, cmd, arg);
5942 #endif /* CONFIG_COMPAT */
5944 static int md_open(struct block_device *bdev, fmode_t mode)
5947 * Succeed if we can lock the mddev, which confirms that
5948 * it isn't being stopped right now.
5950 mddev_t *mddev = mddev_find(bdev->bd_dev);
5953 mutex_lock(&md_mutex);
5954 if (mddev->gendisk != bdev->bd_disk) {
5955 /* we are racing with mddev_put which is discarding this
5959 /* Wait until bdev->bd_disk is definitely gone */
5960 flush_scheduled_work();
5961 /* Then retry the open from the top */
5962 mutex_unlock(&md_mutex);
5963 return -ERESTARTSYS;
5965 BUG_ON(mddev != bdev->bd_disk->private_data);
5967 if ((err = mutex_lock_interruptible(&mddev->open_mutex)))
5971 atomic_inc(&mddev->openers);
5972 mutex_unlock(&mddev->open_mutex);
5974 check_disk_size_change(mddev->gendisk, bdev);
5976 mutex_unlock(&md_mutex);
5980 static int md_release(struct gendisk *disk, fmode_t mode)
5982 mddev_t *mddev = disk->private_data;
5985 mutex_lock(&md_mutex);
5986 atomic_dec(&mddev->openers);
5988 mutex_unlock(&md_mutex);
5992 static const struct block_device_operations md_fops =
5994 .owner = THIS_MODULE,
5996 .release = md_release,
5998 #ifdef CONFIG_COMPAT
5999 .compat_ioctl = md_compat_ioctl,
6001 .getgeo = md_getgeo,
6004 static int md_thread(void * arg)
6006 mdk_thread_t *thread = arg;
6009 * md_thread is a 'system-thread', it's priority should be very
6010 * high. We avoid resource deadlocks individually in each
6011 * raid personality. (RAID5 does preallocation) We also use RR and
6012 * the very same RT priority as kswapd, thus we will never get
6013 * into a priority inversion deadlock.
6015 * we definitely have to have equal or higher priority than
6016 * bdflush, otherwise bdflush will deadlock if there are too
6017 * many dirty RAID5 blocks.
6020 allow_signal(SIGKILL);
6021 while (!kthread_should_stop()) {
6023 /* We need to wait INTERRUPTIBLE so that
6024 * we don't add to the load-average.
6025 * That means we need to be sure no signals are
6028 if (signal_pending(current))
6029 flush_signals(current);
6031 wait_event_interruptible_timeout
6033 test_bit(THREAD_WAKEUP, &thread->flags)
6034 || kthread_should_stop(),
6037 clear_bit(THREAD_WAKEUP, &thread->flags);
6039 thread->run(thread->mddev);
6045 void md_wakeup_thread(mdk_thread_t *thread)
6048 dprintk("md: waking up MD thread %s.\n", thread->tsk->comm);
6049 set_bit(THREAD_WAKEUP, &thread->flags);
6050 wake_up(&thread->wqueue);
6054 mdk_thread_t *md_register_thread(void (*run) (mddev_t *), mddev_t *mddev,
6057 mdk_thread_t *thread;
6059 thread = kzalloc(sizeof(mdk_thread_t), GFP_KERNEL);
6063 init_waitqueue_head(&thread->wqueue);
6066 thread->mddev = mddev;
6067 thread->timeout = MAX_SCHEDULE_TIMEOUT;
6068 thread->tsk = kthread_run(md_thread, thread,
6070 mdname(thread->mddev),
6071 name ?: mddev->pers->name);
6072 if (IS_ERR(thread->tsk)) {
6079 void md_unregister_thread(mdk_thread_t *thread)
6083 dprintk("interrupting MD-thread pid %d\n", task_pid_nr(thread->tsk));
6085 kthread_stop(thread->tsk);
6089 void md_error(mddev_t *mddev, mdk_rdev_t *rdev)
6096 if (!rdev || test_bit(Faulty, &rdev->flags))
6099 if (mddev->external)
6100 set_bit(Blocked, &rdev->flags);
6102 dprintk("md_error dev:%s, rdev:(%d:%d), (caller: %p,%p,%p,%p).\n",
6104 MAJOR(rdev->bdev->bd_dev), MINOR(rdev->bdev->bd_dev),
6105 __builtin_return_address(0),__builtin_return_address(1),
6106 __builtin_return_address(2),__builtin_return_address(3));
6110 if (!mddev->pers->error_handler)
6112 mddev->pers->error_handler(mddev,rdev);
6113 if (mddev->degraded)
6114 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
6115 sysfs_notify_dirent_safe(rdev->sysfs_state);
6116 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6117 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6118 md_wakeup_thread(mddev->thread);
6119 if (mddev->event_work.func)
6120 schedule_work(&mddev->event_work);
6121 md_new_event_inintr(mddev);
6124 /* seq_file implementation /proc/mdstat */
6126 static void status_unused(struct seq_file *seq)
6131 seq_printf(seq, "unused devices: ");
6133 list_for_each_entry(rdev, &pending_raid_disks, same_set) {
6134 char b[BDEVNAME_SIZE];
6136 seq_printf(seq, "%s ",
6137 bdevname(rdev->bdev,b));
6140 seq_printf(seq, "<none>");
6142 seq_printf(seq, "\n");
6146 static void status_resync(struct seq_file *seq, mddev_t * mddev)
6148 sector_t max_sectors, resync, res;
6149 unsigned long dt, db;
6152 unsigned int per_milli;
6154 resync = mddev->curr_resync - atomic_read(&mddev->recovery_active);
6156 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
6157 max_sectors = mddev->resync_max_sectors;
6159 max_sectors = mddev->dev_sectors;
6162 * Should not happen.
6168 /* Pick 'scale' such that (resync>>scale)*1000 will fit
6169 * in a sector_t, and (max_sectors>>scale) will fit in a
6170 * u32, as those are the requirements for sector_div.
6171 * Thus 'scale' must be at least 10
6174 if (sizeof(sector_t) > sizeof(unsigned long)) {
6175 while ( max_sectors/2 > (1ULL<<(scale+32)))
6178 res = (resync>>scale)*1000;
6179 sector_div(res, (u32)((max_sectors>>scale)+1));
6183 int i, x = per_milli/50, y = 20-x;
6184 seq_printf(seq, "[");
6185 for (i = 0; i < x; i++)
6186 seq_printf(seq, "=");
6187 seq_printf(seq, ">");
6188 for (i = 0; i < y; i++)
6189 seq_printf(seq, ".");
6190 seq_printf(seq, "] ");
6192 seq_printf(seq, " %s =%3u.%u%% (%llu/%llu)",
6193 (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)?
6195 (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)?
6197 (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
6198 "resync" : "recovery"))),
6199 per_milli/10, per_milli % 10,
6200 (unsigned long long) resync/2,
6201 (unsigned long long) max_sectors/2);
6204 * dt: time from mark until now
6205 * db: blocks written from mark until now
6206 * rt: remaining time
6208 * rt is a sector_t, so could be 32bit or 64bit.
6209 * So we divide before multiply in case it is 32bit and close
6211 * We scale the divisor (db) by 32 to avoid loosing precision
6212 * near the end of resync when the number of remaining sectors
6214 * We then divide rt by 32 after multiplying by db to compensate.
6215 * The '+1' avoids division by zero if db is very small.
6217 dt = ((jiffies - mddev->resync_mark) / HZ);
6219 db = (mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active))
6220 - mddev->resync_mark_cnt;
6222 rt = max_sectors - resync; /* number of remaining sectors */
6223 sector_div(rt, db/32+1);
6227 seq_printf(seq, " finish=%lu.%lumin", (unsigned long)rt / 60,
6228 ((unsigned long)rt % 60)/6);
6230 seq_printf(seq, " speed=%ldK/sec", db/2/dt);
6233 static void *md_seq_start(struct seq_file *seq, loff_t *pos)
6235 struct list_head *tmp;
6245 spin_lock(&all_mddevs_lock);
6246 list_for_each(tmp,&all_mddevs)
6248 mddev = list_entry(tmp, mddev_t, all_mddevs);
6250 spin_unlock(&all_mddevs_lock);
6253 spin_unlock(&all_mddevs_lock);
6255 return (void*)2;/* tail */
6259 static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
6261 struct list_head *tmp;
6262 mddev_t *next_mddev, *mddev = v;
6268 spin_lock(&all_mddevs_lock);
6270 tmp = all_mddevs.next;
6272 tmp = mddev->all_mddevs.next;
6273 if (tmp != &all_mddevs)
6274 next_mddev = mddev_get(list_entry(tmp,mddev_t,all_mddevs));
6276 next_mddev = (void*)2;
6279 spin_unlock(&all_mddevs_lock);
6287 static void md_seq_stop(struct seq_file *seq, void *v)
6291 if (mddev && v != (void*)1 && v != (void*)2)
6295 struct mdstat_info {
6299 static int md_seq_show(struct seq_file *seq, void *v)
6304 struct mdstat_info *mi = seq->private;
6305 struct bitmap *bitmap;
6307 if (v == (void*)1) {
6308 struct mdk_personality *pers;
6309 seq_printf(seq, "Personalities : ");
6310 spin_lock(&pers_lock);
6311 list_for_each_entry(pers, &pers_list, list)
6312 seq_printf(seq, "[%s] ", pers->name);
6314 spin_unlock(&pers_lock);
6315 seq_printf(seq, "\n");
6316 mi->event = atomic_read(&md_event_count);
6319 if (v == (void*)2) {
6324 if (mddev_lock(mddev) < 0)
6327 if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
6328 seq_printf(seq, "%s : %sactive", mdname(mddev),
6329 mddev->pers ? "" : "in");
6332 seq_printf(seq, " (read-only)");
6334 seq_printf(seq, " (auto-read-only)");
6335 seq_printf(seq, " %s", mddev->pers->name);
6339 list_for_each_entry(rdev, &mddev->disks, same_set) {
6340 char b[BDEVNAME_SIZE];
6341 seq_printf(seq, " %s[%d]",
6342 bdevname(rdev->bdev,b), rdev->desc_nr);
6343 if (test_bit(WriteMostly, &rdev->flags))
6344 seq_printf(seq, "(W)");
6345 if (test_bit(Faulty, &rdev->flags)) {
6346 seq_printf(seq, "(F)");
6348 } else if (rdev->raid_disk < 0)
6349 seq_printf(seq, "(S)"); /* spare */
6350 sectors += rdev->sectors;
6353 if (!list_empty(&mddev->disks)) {
6355 seq_printf(seq, "\n %llu blocks",
6356 (unsigned long long)
6357 mddev->array_sectors / 2);
6359 seq_printf(seq, "\n %llu blocks",
6360 (unsigned long long)sectors / 2);
6362 if (mddev->persistent) {
6363 if (mddev->major_version != 0 ||
6364 mddev->minor_version != 90) {
6365 seq_printf(seq," super %d.%d",
6366 mddev->major_version,
6367 mddev->minor_version);
6369 } else if (mddev->external)
6370 seq_printf(seq, " super external:%s",
6371 mddev->metadata_type);
6373 seq_printf(seq, " super non-persistent");
6376 mddev->pers->status(seq, mddev);
6377 seq_printf(seq, "\n ");
6378 if (mddev->pers->sync_request) {
6379 if (mddev->curr_resync > 2) {
6380 status_resync(seq, mddev);
6381 seq_printf(seq, "\n ");
6382 } else if (mddev->curr_resync == 1 || mddev->curr_resync == 2)
6383 seq_printf(seq, "\tresync=DELAYED\n ");
6384 else if (mddev->recovery_cp < MaxSector)
6385 seq_printf(seq, "\tresync=PENDING\n ");
6388 seq_printf(seq, "\n ");
6390 if ((bitmap = mddev->bitmap)) {
6391 unsigned long chunk_kb;
6392 unsigned long flags;
6393 spin_lock_irqsave(&bitmap->lock, flags);
6394 chunk_kb = mddev->bitmap_info.chunksize >> 10;
6395 seq_printf(seq, "bitmap: %lu/%lu pages [%luKB], "
6397 bitmap->pages - bitmap->missing_pages,
6399 (bitmap->pages - bitmap->missing_pages)
6400 << (PAGE_SHIFT - 10),
6401 chunk_kb ? chunk_kb : mddev->bitmap_info.chunksize,
6402 chunk_kb ? "KB" : "B");
6404 seq_printf(seq, ", file: ");
6405 seq_path(seq, &bitmap->file->f_path, " \t\n");
6408 seq_printf(seq, "\n");
6409 spin_unlock_irqrestore(&bitmap->lock, flags);
6412 seq_printf(seq, "\n");
6414 mddev_unlock(mddev);
6419 static const struct seq_operations md_seq_ops = {
6420 .start = md_seq_start,
6421 .next = md_seq_next,
6422 .stop = md_seq_stop,
6423 .show = md_seq_show,
6426 static int md_seq_open(struct inode *inode, struct file *file)
6429 struct mdstat_info *mi = kmalloc(sizeof(*mi), GFP_KERNEL);
6433 error = seq_open(file, &md_seq_ops);
6437 struct seq_file *p = file->private_data;
6439 mi->event = atomic_read(&md_event_count);
6444 static unsigned int mdstat_poll(struct file *filp, poll_table *wait)
6446 struct seq_file *m = filp->private_data;
6447 struct mdstat_info *mi = m->private;
6450 poll_wait(filp, &md_event_waiters, wait);
6452 /* always allow read */
6453 mask = POLLIN | POLLRDNORM;
6455 if (mi->event != atomic_read(&md_event_count))
6456 mask |= POLLERR | POLLPRI;
6460 static const struct file_operations md_seq_fops = {
6461 .owner = THIS_MODULE,
6462 .open = md_seq_open,
6464 .llseek = seq_lseek,
6465 .release = seq_release_private,
6466 .poll = mdstat_poll,
6469 int register_md_personality(struct mdk_personality *p)
6471 spin_lock(&pers_lock);
6472 list_add_tail(&p->list, &pers_list);
6473 printk(KERN_INFO "md: %s personality registered for level %d\n", p->name, p->level);
6474 spin_unlock(&pers_lock);
6478 int unregister_md_personality(struct mdk_personality *p)
6480 printk(KERN_INFO "md: %s personality unregistered\n", p->name);
6481 spin_lock(&pers_lock);
6482 list_del_init(&p->list);
6483 spin_unlock(&pers_lock);
6487 static int is_mddev_idle(mddev_t *mddev, int init)
6495 rdev_for_each_rcu(rdev, mddev) {
6496 struct gendisk *disk = rdev->bdev->bd_contains->bd_disk;
6497 curr_events = (int)part_stat_read(&disk->part0, sectors[0]) +
6498 (int)part_stat_read(&disk->part0, sectors[1]) -
6499 atomic_read(&disk->sync_io);
6500 /* sync IO will cause sync_io to increase before the disk_stats
6501 * as sync_io is counted when a request starts, and
6502 * disk_stats is counted when it completes.
6503 * So resync activity will cause curr_events to be smaller than
6504 * when there was no such activity.
6505 * non-sync IO will cause disk_stat to increase without
6506 * increasing sync_io so curr_events will (eventually)
6507 * be larger than it was before. Once it becomes
6508 * substantially larger, the test below will cause
6509 * the array to appear non-idle, and resync will slow
6511 * If there is a lot of outstanding resync activity when
6512 * we set last_event to curr_events, then all that activity
6513 * completing might cause the array to appear non-idle
6514 * and resync will be slowed down even though there might
6515 * not have been non-resync activity. This will only
6516 * happen once though. 'last_events' will soon reflect
6517 * the state where there is little or no outstanding
6518 * resync requests, and further resync activity will
6519 * always make curr_events less than last_events.
6522 if (init || curr_events - rdev->last_events > 64) {
6523 rdev->last_events = curr_events;
6531 void md_done_sync(mddev_t *mddev, int blocks, int ok)
6533 /* another "blocks" (512byte) blocks have been synced */
6534 atomic_sub(blocks, &mddev->recovery_active);
6535 wake_up(&mddev->recovery_wait);
6537 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6538 md_wakeup_thread(mddev->thread);
6539 // stop recovery, signal do_sync ....
6544 /* md_write_start(mddev, bi)
6545 * If we need to update some array metadata (e.g. 'active' flag
6546 * in superblock) before writing, schedule a superblock update
6547 * and wait for it to complete.
6549 void md_write_start(mddev_t *mddev, struct bio *bi)
6552 if (bio_data_dir(bi) != WRITE)
6555 BUG_ON(mddev->ro == 1);
6556 if (mddev->ro == 2) {
6557 /* need to switch to read/write */
6559 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6560 md_wakeup_thread(mddev->thread);
6561 md_wakeup_thread(mddev->sync_thread);
6564 atomic_inc(&mddev->writes_pending);
6565 if (mddev->safemode == 1)
6566 mddev->safemode = 0;
6567 if (mddev->in_sync) {
6568 spin_lock_irq(&mddev->write_lock);
6569 if (mddev->in_sync) {
6571 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
6572 set_bit(MD_CHANGE_PENDING, &mddev->flags);
6573 md_wakeup_thread(mddev->thread);
6576 spin_unlock_irq(&mddev->write_lock);
6579 sysfs_notify_dirent_safe(mddev->sysfs_state);
6580 wait_event(mddev->sb_wait,
6581 !test_bit(MD_CHANGE_PENDING, &mddev->flags));
6584 void md_write_end(mddev_t *mddev)
6586 if (atomic_dec_and_test(&mddev->writes_pending)) {
6587 if (mddev->safemode == 2)
6588 md_wakeup_thread(mddev->thread);
6589 else if (mddev->safemode_delay)
6590 mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay);
6594 /* md_allow_write(mddev)
6595 * Calling this ensures that the array is marked 'active' so that writes
6596 * may proceed without blocking. It is important to call this before
6597 * attempting a GFP_KERNEL allocation while holding the mddev lock.
6598 * Must be called with mddev_lock held.
6600 * In the ->external case MD_CHANGE_CLEAN can not be cleared until mddev->lock
6601 * is dropped, so return -EAGAIN after notifying userspace.
6603 int md_allow_write(mddev_t *mddev)
6609 if (!mddev->pers->sync_request)
6612 spin_lock_irq(&mddev->write_lock);
6613 if (mddev->in_sync) {
6615 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
6616 set_bit(MD_CHANGE_PENDING, &mddev->flags);
6617 if (mddev->safemode_delay &&
6618 mddev->safemode == 0)
6619 mddev->safemode = 1;
6620 spin_unlock_irq(&mddev->write_lock);
6621 md_update_sb(mddev, 0);
6622 sysfs_notify_dirent_safe(mddev->sysfs_state);
6624 spin_unlock_irq(&mddev->write_lock);
6626 if (test_bit(MD_CHANGE_PENDING, &mddev->flags))
6631 EXPORT_SYMBOL_GPL(md_allow_write);
6633 void md_unplug(mddev_t *mddev)
6636 blk_unplug(mddev->queue);
6638 mddev->plug->unplug_fn(mddev->plug);
6641 #define SYNC_MARKS 10
6642 #define SYNC_MARK_STEP (3*HZ)
6643 void md_do_sync(mddev_t *mddev)
6646 unsigned int currspeed = 0,
6648 sector_t max_sectors,j, io_sectors;
6649 unsigned long mark[SYNC_MARKS];
6650 sector_t mark_cnt[SYNC_MARKS];
6652 struct list_head *tmp;
6653 sector_t last_check;
6658 /* just incase thread restarts... */
6659 if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
6661 if (mddev->ro) /* never try to sync a read-only array */
6664 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
6665 if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
6666 desc = "data-check";
6667 else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
6668 desc = "requested-resync";
6671 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
6676 /* we overload curr_resync somewhat here.
6677 * 0 == not engaged in resync at all
6678 * 2 == checking that there is no conflict with another sync
6679 * 1 == like 2, but have yielded to allow conflicting resync to
6681 * other == active in resync - this many blocks
6683 * Before starting a resync we must have set curr_resync to
6684 * 2, and then checked that every "conflicting" array has curr_resync
6685 * less than ours. When we find one that is the same or higher
6686 * we wait on resync_wait. To avoid deadlock, we reduce curr_resync
6687 * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
6688 * This will mean we have to start checking from the beginning again.
6693 mddev->curr_resync = 2;
6696 if (kthread_should_stop())
6697 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6699 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
6701 for_each_mddev(mddev2, tmp) {
6702 if (mddev2 == mddev)
6704 if (!mddev->parallel_resync
6705 && mddev2->curr_resync
6706 && match_mddev_units(mddev, mddev2)) {
6708 if (mddev < mddev2 && mddev->curr_resync == 2) {
6709 /* arbitrarily yield */
6710 mddev->curr_resync = 1;
6711 wake_up(&resync_wait);
6713 if (mddev > mddev2 && mddev->curr_resync == 1)
6714 /* no need to wait here, we can wait the next
6715 * time 'round when curr_resync == 2
6718 /* We need to wait 'interruptible' so as not to
6719 * contribute to the load average, and not to
6720 * be caught by 'softlockup'
6722 prepare_to_wait(&resync_wait, &wq, TASK_INTERRUPTIBLE);
6723 if (!kthread_should_stop() &&
6724 mddev2->curr_resync >= mddev->curr_resync) {
6725 printk(KERN_INFO "md: delaying %s of %s"
6726 " until %s has finished (they"
6727 " share one or more physical units)\n",
6728 desc, mdname(mddev), mdname(mddev2));
6730 if (signal_pending(current))
6731 flush_signals(current);
6733 finish_wait(&resync_wait, &wq);
6736 finish_wait(&resync_wait, &wq);
6739 } while (mddev->curr_resync < 2);
6742 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
6743 /* resync follows the size requested by the personality,
6744 * which defaults to physical size, but can be virtual size
6746 max_sectors = mddev->resync_max_sectors;
6747 mddev->resync_mismatches = 0;
6748 /* we don't use the checkpoint if there's a bitmap */
6749 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
6750 j = mddev->resync_min;
6751 else if (!mddev->bitmap)
6752 j = mddev->recovery_cp;
6754 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
6755 max_sectors = mddev->dev_sectors;
6757 /* recovery follows the physical size of devices */
6758 max_sectors = mddev->dev_sectors;
6761 list_for_each_entry_rcu(rdev, &mddev->disks, same_set)
6762 if (rdev->raid_disk >= 0 &&
6763 !test_bit(Faulty, &rdev->flags) &&
6764 !test_bit(In_sync, &rdev->flags) &&
6765 rdev->recovery_offset < j)
6766 j = rdev->recovery_offset;
6770 printk(KERN_INFO "md: %s of RAID array %s\n", desc, mdname(mddev));
6771 printk(KERN_INFO "md: minimum _guaranteed_ speed:"
6772 " %d KB/sec/disk.\n", speed_min(mddev));
6773 printk(KERN_INFO "md: using maximum available idle IO bandwidth "
6774 "(but not more than %d KB/sec) for %s.\n",
6775 speed_max(mddev), desc);
6777 is_mddev_idle(mddev, 1); /* this initializes IO event counters */
6780 for (m = 0; m < SYNC_MARKS; m++) {
6782 mark_cnt[m] = io_sectors;
6785 mddev->resync_mark = mark[last_mark];
6786 mddev->resync_mark_cnt = mark_cnt[last_mark];
6789 * Tune reconstruction:
6791 window = 32*(PAGE_SIZE/512);
6792 printk(KERN_INFO "md: using %dk window, over a total of %llu blocks.\n",
6793 window/2,(unsigned long long) max_sectors/2);
6795 atomic_set(&mddev->recovery_active, 0);
6800 "md: resuming %s of %s from checkpoint.\n",
6801 desc, mdname(mddev));
6802 mddev->curr_resync = j;
6804 mddev->curr_resync_completed = mddev->curr_resync;
6806 while (j < max_sectors) {
6811 if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
6812 ((mddev->curr_resync > mddev->curr_resync_completed &&
6813 (mddev->curr_resync - mddev->curr_resync_completed)
6814 > (max_sectors >> 4)) ||
6815 (j - mddev->curr_resync_completed)*2
6816 >= mddev->resync_max - mddev->curr_resync_completed
6818 /* time to update curr_resync_completed */
6820 wait_event(mddev->recovery_wait,
6821 atomic_read(&mddev->recovery_active) == 0);
6822 mddev->curr_resync_completed =
6824 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
6825 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
6828 while (j >= mddev->resync_max && !kthread_should_stop()) {
6829 /* As this condition is controlled by user-space,
6830 * we can block indefinitely, so use '_interruptible'
6831 * to avoid triggering warnings.
6833 flush_signals(current); /* just in case */
6834 wait_event_interruptible(mddev->recovery_wait,
6835 mddev->resync_max > j
6836 || kthread_should_stop());
6839 if (kthread_should_stop())
6842 sectors = mddev->pers->sync_request(mddev, j, &skipped,
6843 currspeed < speed_min(mddev));
6845 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6849 if (!skipped) { /* actual IO requested */
6850 io_sectors += sectors;
6851 atomic_add(sectors, &mddev->recovery_active);
6855 if (j>1) mddev->curr_resync = j;
6856 mddev->curr_mark_cnt = io_sectors;
6857 if (last_check == 0)
6858 /* this is the earliers that rebuilt will be
6859 * visible in /proc/mdstat
6861 md_new_event(mddev);
6863 if (last_check + window > io_sectors || j == max_sectors)
6866 last_check = io_sectors;
6868 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
6872 if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) {
6874 int next = (last_mark+1) % SYNC_MARKS;
6876 mddev->resync_mark = mark[next];
6877 mddev->resync_mark_cnt = mark_cnt[next];
6878 mark[next] = jiffies;
6879 mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active);
6884 if (kthread_should_stop())
6889 * this loop exits only if either when we are slower than
6890 * the 'hard' speed limit, or the system was IO-idle for
6892 * the system might be non-idle CPU-wise, but we only care
6893 * about not overloading the IO subsystem. (things like an
6894 * e2fsck being done on the RAID array should execute fast)
6899 currspeed = ((unsigned long)(io_sectors-mddev->resync_mark_cnt))/2
6900 /((jiffies-mddev->resync_mark)/HZ +1) +1;
6902 if (currspeed > speed_min(mddev)) {
6903 if ((currspeed > speed_max(mddev)) ||
6904 !is_mddev_idle(mddev, 0)) {
6910 printk(KERN_INFO "md: %s: %s done.\n",mdname(mddev), desc);
6912 * this also signals 'finished resyncing' to md_stop
6917 wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
6919 /* tell personality that we are finished */
6920 mddev->pers->sync_request(mddev, max_sectors, &skipped, 1);
6922 if (!test_bit(MD_RECOVERY_CHECK, &mddev->recovery) &&
6923 mddev->curr_resync > 2) {
6924 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
6925 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
6926 if (mddev->curr_resync >= mddev->recovery_cp) {
6928 "md: checkpointing %s of %s.\n",
6929 desc, mdname(mddev));
6930 mddev->recovery_cp = mddev->curr_resync;
6933 mddev->recovery_cp = MaxSector;
6935 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery))
6936 mddev->curr_resync = MaxSector;
6938 list_for_each_entry_rcu(rdev, &mddev->disks, same_set)
6939 if (rdev->raid_disk >= 0 &&
6940 mddev->delta_disks >= 0 &&
6941 !test_bit(Faulty, &rdev->flags) &&
6942 !test_bit(In_sync, &rdev->flags) &&
6943 rdev->recovery_offset < mddev->curr_resync)
6944 rdev->recovery_offset = mddev->curr_resync;
6948 set_bit(MD_CHANGE_DEVS, &mddev->flags);
6951 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
6952 /* We completed so min/max setting can be forgotten if used. */
6953 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
6954 mddev->resync_min = 0;
6955 mddev->resync_max = MaxSector;
6956 } else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
6957 mddev->resync_min = mddev->curr_resync_completed;
6958 mddev->curr_resync = 0;
6959 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery))
6960 mddev->curr_resync_completed = 0;
6961 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
6962 wake_up(&resync_wait);
6963 set_bit(MD_RECOVERY_DONE, &mddev->recovery);
6964 md_wakeup_thread(mddev->thread);
6969 * got a signal, exit.
6972 "md: md_do_sync() got signal ... exiting\n");
6973 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6977 EXPORT_SYMBOL_GPL(md_do_sync);
6980 static int remove_and_add_spares(mddev_t *mddev)
6985 mddev->curr_resync_completed = 0;
6987 list_for_each_entry(rdev, &mddev->disks, same_set)
6988 if (rdev->raid_disk >= 0 &&
6989 !test_bit(Blocked, &rdev->flags) &&
6990 (test_bit(Faulty, &rdev->flags) ||
6991 ! test_bit(In_sync, &rdev->flags)) &&
6992 atomic_read(&rdev->nr_pending)==0) {
6993 if (mddev->pers->hot_remove_disk(
6994 mddev, rdev->raid_disk)==0) {
6996 sprintf(nm,"rd%d", rdev->raid_disk);
6997 sysfs_remove_link(&mddev->kobj, nm);
6998 rdev->raid_disk = -1;
7002 if (mddev->degraded && ! mddev->ro && !mddev->recovery_disabled) {
7003 list_for_each_entry(rdev, &mddev->disks, same_set) {
7004 if (rdev->raid_disk >= 0 &&
7005 !test_bit(In_sync, &rdev->flags) &&
7006 !test_bit(Blocked, &rdev->flags))
7008 if (rdev->raid_disk < 0
7009 && !test_bit(Faulty, &rdev->flags)) {
7010 rdev->recovery_offset = 0;
7012 hot_add_disk(mddev, rdev) == 0) {
7014 sprintf(nm, "rd%d", rdev->raid_disk);
7015 if (sysfs_create_link(&mddev->kobj,
7017 /* failure here is OK */;
7019 md_new_event(mddev);
7020 set_bit(MD_CHANGE_DEVS, &mddev->flags);
7029 * This routine is regularly called by all per-raid-array threads to
7030 * deal with generic issues like resync and super-block update.
7031 * Raid personalities that don't have a thread (linear/raid0) do not
7032 * need this as they never do any recovery or update the superblock.
7034 * It does not do any resync itself, but rather "forks" off other threads
7035 * to do that as needed.
7036 * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
7037 * "->recovery" and create a thread at ->sync_thread.
7038 * When the thread finishes it sets MD_RECOVERY_DONE
7039 * and wakeups up this thread which will reap the thread and finish up.
7040 * This thread also removes any faulty devices (with nr_pending == 0).
7042 * The overall approach is:
7043 * 1/ if the superblock needs updating, update it.
7044 * 2/ If a recovery thread is running, don't do anything else.
7045 * 3/ If recovery has finished, clean up, possibly marking spares active.
7046 * 4/ If there are any faulty devices, remove them.
7047 * 5/ If array is degraded, try to add spares devices
7048 * 6/ If array has spares or is not in-sync, start a resync thread.
7050 void md_check_recovery(mddev_t *mddev)
7056 bitmap_daemon_work(mddev);
7061 if (signal_pending(current)) {
7062 if (mddev->pers->sync_request && !mddev->external) {
7063 printk(KERN_INFO "md: %s in immediate safe mode\n",
7065 mddev->safemode = 2;
7067 flush_signals(current);
7070 if (mddev->ro && !test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
7073 (mddev->flags && !mddev->external) ||
7074 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
7075 test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
7076 (mddev->external == 0 && mddev->safemode == 1) ||
7077 (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending)
7078 && !mddev->in_sync && mddev->recovery_cp == MaxSector)
7082 if (mddev_trylock(mddev)) {
7086 /* Only thing we do on a ro array is remove
7089 remove_and_add_spares(mddev);
7090 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7094 if (!mddev->external) {
7096 spin_lock_irq(&mddev->write_lock);
7097 if (mddev->safemode &&
7098 !atomic_read(&mddev->writes_pending) &&
7100 mddev->recovery_cp == MaxSector) {
7103 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
7105 if (mddev->safemode == 1)
7106 mddev->safemode = 0;
7107 spin_unlock_irq(&mddev->write_lock);
7109 sysfs_notify_dirent_safe(mddev->sysfs_state);
7113 md_update_sb(mddev, 0);
7115 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
7116 !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
7117 /* resync/recovery still happening */
7118 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7121 if (mddev->sync_thread) {
7122 /* resync has finished, collect result */
7123 md_unregister_thread(mddev->sync_thread);
7124 mddev->sync_thread = NULL;
7125 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
7126 !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
7128 /* activate any spares */
7129 if (mddev->pers->spare_active(mddev))
7130 sysfs_notify(&mddev->kobj, NULL,
7133 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
7134 mddev->pers->finish_reshape)
7135 mddev->pers->finish_reshape(mddev);
7136 md_update_sb(mddev, 1);
7138 /* if array is no-longer degraded, then any saved_raid_disk
7139 * information must be scrapped
7141 if (!mddev->degraded)
7142 list_for_each_entry(rdev, &mddev->disks, same_set)
7143 rdev->saved_raid_disk = -1;
7145 mddev->recovery = 0;
7146 /* flag recovery needed just to double check */
7147 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7148 sysfs_notify_dirent_safe(mddev->sysfs_action);
7149 md_new_event(mddev);
7152 /* Set RUNNING before clearing NEEDED to avoid
7153 * any transients in the value of "sync_action".
7155 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
7156 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7157 /* Clear some bits that don't mean anything, but
7160 clear_bit(MD_RECOVERY_INTR, &mddev->recovery);
7161 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
7163 if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
7165 /* no recovery is running.
7166 * remove any failed drives, then
7167 * add spares if possible.
7168 * Spare are also removed and re-added, to allow
7169 * the personality to fail the re-add.
7172 if (mddev->reshape_position != MaxSector) {
7173 if (mddev->pers->check_reshape == NULL ||
7174 mddev->pers->check_reshape(mddev) != 0)
7175 /* Cannot proceed */
7177 set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
7178 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
7179 } else if ((spares = remove_and_add_spares(mddev))) {
7180 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
7181 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
7182 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
7183 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
7184 } else if (mddev->recovery_cp < MaxSector) {
7185 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
7186 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
7187 } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
7188 /* nothing to be done ... */
7191 if (mddev->pers->sync_request) {
7192 if (spares && mddev->bitmap && ! mddev->bitmap->file) {
7193 /* We are adding a device or devices to an array
7194 * which has the bitmap stored on all devices.
7195 * So make sure all bitmap pages get written
7197 bitmap_write_all(mddev->bitmap);
7199 mddev->sync_thread = md_register_thread(md_do_sync,
7202 if (!mddev->sync_thread) {
7203 printk(KERN_ERR "%s: could not start resync"
7206 /* leave the spares where they are, it shouldn't hurt */
7207 mddev->recovery = 0;
7209 md_wakeup_thread(mddev->sync_thread);
7210 sysfs_notify_dirent_safe(mddev->sysfs_action);
7211 md_new_event(mddev);
7214 if (!mddev->sync_thread) {
7215 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
7216 if (test_and_clear_bit(MD_RECOVERY_RECOVER,
7218 if (mddev->sysfs_action)
7219 sysfs_notify_dirent_safe(mddev->sysfs_action);
7221 mddev_unlock(mddev);
7225 void md_wait_for_blocked_rdev(mdk_rdev_t *rdev, mddev_t *mddev)
7227 sysfs_notify_dirent_safe(rdev->sysfs_state);
7228 wait_event_timeout(rdev->blocked_wait,
7229 !test_bit(Blocked, &rdev->flags),
7230 msecs_to_jiffies(5000));
7231 rdev_dec_pending(rdev, mddev);
7233 EXPORT_SYMBOL(md_wait_for_blocked_rdev);
7235 static int md_notify_reboot(struct notifier_block *this,
7236 unsigned long code, void *x)
7238 struct list_head *tmp;
7241 if ((code == SYS_DOWN) || (code == SYS_HALT) || (code == SYS_POWER_OFF)) {
7243 printk(KERN_INFO "md: stopping all md devices.\n");
7245 for_each_mddev(mddev, tmp)
7246 if (mddev_trylock(mddev)) {
7247 /* Force a switch to readonly even array
7248 * appears to still be in use. Hence
7251 md_set_readonly(mddev, 100);
7252 mddev_unlock(mddev);
7255 * certain more exotic SCSI devices are known to be
7256 * volatile wrt too early system reboots. While the
7257 * right place to handle this issue is the given
7258 * driver, we do want to have a safe RAID driver ...
7265 static struct notifier_block md_notifier = {
7266 .notifier_call = md_notify_reboot,
7268 .priority = INT_MAX, /* before any real devices */
7271 static void md_geninit(void)
7273 dprintk("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
7275 proc_create("mdstat", S_IRUGO, NULL, &md_seq_fops);
7278 static int __init md_init(void)
7280 if (register_blkdev(MD_MAJOR, "md"))
7282 if ((mdp_major=register_blkdev(0, "mdp"))<=0) {
7283 unregister_blkdev(MD_MAJOR, "md");
7286 blk_register_region(MKDEV(MD_MAJOR, 0), 1UL<<MINORBITS, THIS_MODULE,
7287 md_probe, NULL, NULL);
7288 blk_register_region(MKDEV(mdp_major, 0), 1UL<<MINORBITS, THIS_MODULE,
7289 md_probe, NULL, NULL);
7291 register_reboot_notifier(&md_notifier);
7292 raid_table_header = register_sysctl_table(raid_root_table);
7302 * Searches all registered partitions for autorun RAID arrays
7306 static LIST_HEAD(all_detected_devices);
7307 struct detected_devices_node {
7308 struct list_head list;
7312 void md_autodetect_dev(dev_t dev)
7314 struct detected_devices_node *node_detected_dev;
7316 node_detected_dev = kzalloc(sizeof(*node_detected_dev), GFP_KERNEL);
7317 if (node_detected_dev) {
7318 node_detected_dev->dev = dev;
7319 list_add_tail(&node_detected_dev->list, &all_detected_devices);
7321 printk(KERN_CRIT "md: md_autodetect_dev: kzalloc failed"
7322 ", skipping dev(%d,%d)\n", MAJOR(dev), MINOR(dev));
7327 static void autostart_arrays(int part)
7330 struct detected_devices_node *node_detected_dev;
7332 int i_scanned, i_passed;
7337 printk(KERN_INFO "md: Autodetecting RAID arrays.\n");
7339 while (!list_empty(&all_detected_devices) && i_scanned < INT_MAX) {
7341 node_detected_dev = list_entry(all_detected_devices.next,
7342 struct detected_devices_node, list);
7343 list_del(&node_detected_dev->list);
7344 dev = node_detected_dev->dev;
7345 kfree(node_detected_dev);
7346 rdev = md_import_device(dev,0, 90);
7350 if (test_bit(Faulty, &rdev->flags)) {
7354 set_bit(AutoDetected, &rdev->flags);
7355 list_add(&rdev->same_set, &pending_raid_disks);
7359 printk(KERN_INFO "md: Scanned %d and added %d devices.\n",
7360 i_scanned, i_passed);
7362 autorun_devices(part);
7365 #endif /* !MODULE */
7367 static __exit void md_exit(void)
7370 struct list_head *tmp;
7372 blk_unregister_region(MKDEV(MD_MAJOR,0), 1U << MINORBITS);
7373 blk_unregister_region(MKDEV(mdp_major,0), 1U << MINORBITS);
7375 unregister_blkdev(MD_MAJOR,"md");
7376 unregister_blkdev(mdp_major, "mdp");
7377 unregister_reboot_notifier(&md_notifier);
7378 unregister_sysctl_table(raid_table_header);
7379 remove_proc_entry("mdstat", NULL);
7380 for_each_mddev(mddev, tmp) {
7381 export_array(mddev);
7382 mddev->hold_active = 0;
7386 subsys_initcall(md_init);
7387 module_exit(md_exit)
7389 static int get_ro(char *buffer, struct kernel_param *kp)
7391 return sprintf(buffer, "%d", start_readonly);
7393 static int set_ro(const char *val, struct kernel_param *kp)
7396 int num = simple_strtoul(val, &e, 10);
7397 if (*val && (*e == '\0' || *e == '\n')) {
7398 start_readonly = num;
7404 module_param_call(start_ro, set_ro, get_ro, NULL, S_IRUSR|S_IWUSR);
7405 module_param(start_dirty_degraded, int, S_IRUGO|S_IWUSR);
7407 module_param_call(new_array, add_named_array, NULL, NULL, S_IWUSR);
7409 EXPORT_SYMBOL(register_md_personality);
7410 EXPORT_SYMBOL(unregister_md_personality);
7411 EXPORT_SYMBOL(md_error);
7412 EXPORT_SYMBOL(md_done_sync);
7413 EXPORT_SYMBOL(md_write_start);
7414 EXPORT_SYMBOL(md_write_end);
7415 EXPORT_SYMBOL(md_register_thread);
7416 EXPORT_SYMBOL(md_unregister_thread);
7417 EXPORT_SYMBOL(md_wakeup_thread);
7418 EXPORT_SYMBOL(md_check_recovery);
7419 MODULE_LICENSE("GPL");
7420 MODULE_DESCRIPTION("MD RAID framework");
7422 MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR);