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 max_dev = le32_to_cpu(sb->max_dev);
1650 for (i=0; i<max_dev;i++)
1651 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1653 list_for_each_entry(rdev2, &mddev->disks, same_set) {
1655 if (test_bit(Faulty, &rdev2->flags))
1656 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1657 else if (test_bit(In_sync, &rdev2->flags))
1658 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1659 else if (rdev2->raid_disk >= 0)
1660 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1662 sb->dev_roles[i] = cpu_to_le16(0xffff);
1665 sb->sb_csum = calc_sb_1_csum(sb);
1668 static unsigned long long
1669 super_1_rdev_size_change(mdk_rdev_t *rdev, sector_t num_sectors)
1671 struct mdp_superblock_1 *sb;
1672 sector_t max_sectors;
1673 if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
1674 return 0; /* component must fit device */
1675 if (rdev->sb_start < rdev->data_offset) {
1676 /* minor versions 1 and 2; superblock before data */
1677 max_sectors = rdev->bdev->bd_inode->i_size >> 9;
1678 max_sectors -= rdev->data_offset;
1679 if (!num_sectors || num_sectors > max_sectors)
1680 num_sectors = max_sectors;
1681 } else if (rdev->mddev->bitmap_info.offset) {
1682 /* minor version 0 with bitmap we can't move */
1685 /* minor version 0; superblock after data */
1687 sb_start = (rdev->bdev->bd_inode->i_size >> 9) - 8*2;
1688 sb_start &= ~(sector_t)(4*2 - 1);
1689 max_sectors = rdev->sectors + sb_start - rdev->sb_start;
1690 if (!num_sectors || num_sectors > max_sectors)
1691 num_sectors = max_sectors;
1692 rdev->sb_start = sb_start;
1694 sb = (struct mdp_superblock_1 *) page_address(rdev->sb_page);
1695 sb->data_size = cpu_to_le64(num_sectors);
1696 sb->super_offset = rdev->sb_start;
1697 sb->sb_csum = calc_sb_1_csum(sb);
1698 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
1700 md_super_wait(rdev->mddev);
1701 return num_sectors / 2; /* kB for sysfs */
1704 static struct super_type super_types[] = {
1707 .owner = THIS_MODULE,
1708 .load_super = super_90_load,
1709 .validate_super = super_90_validate,
1710 .sync_super = super_90_sync,
1711 .rdev_size_change = super_90_rdev_size_change,
1715 .owner = THIS_MODULE,
1716 .load_super = super_1_load,
1717 .validate_super = super_1_validate,
1718 .sync_super = super_1_sync,
1719 .rdev_size_change = super_1_rdev_size_change,
1723 static int match_mddev_units(mddev_t *mddev1, mddev_t *mddev2)
1725 mdk_rdev_t *rdev, *rdev2;
1728 rdev_for_each_rcu(rdev, mddev1)
1729 rdev_for_each_rcu(rdev2, mddev2)
1730 if (rdev->bdev->bd_contains ==
1731 rdev2->bdev->bd_contains) {
1739 static LIST_HEAD(pending_raid_disks);
1742 * Try to register data integrity profile for an mddev
1744 * This is called when an array is started and after a disk has been kicked
1745 * from the array. It only succeeds if all working and active component devices
1746 * are integrity capable with matching profiles.
1748 int md_integrity_register(mddev_t *mddev)
1750 mdk_rdev_t *rdev, *reference = NULL;
1752 if (list_empty(&mddev->disks))
1753 return 0; /* nothing to do */
1754 if (blk_get_integrity(mddev->gendisk))
1755 return 0; /* already registered */
1756 list_for_each_entry(rdev, &mddev->disks, same_set) {
1757 /* skip spares and non-functional disks */
1758 if (test_bit(Faulty, &rdev->flags))
1760 if (rdev->raid_disk < 0)
1763 * If at least one rdev is not integrity capable, we can not
1764 * enable data integrity for the md device.
1766 if (!bdev_get_integrity(rdev->bdev))
1769 /* Use the first rdev as the reference */
1773 /* does this rdev's profile match the reference profile? */
1774 if (blk_integrity_compare(reference->bdev->bd_disk,
1775 rdev->bdev->bd_disk) < 0)
1779 * All component devices are integrity capable and have matching
1780 * profiles, register the common profile for the md device.
1782 if (blk_integrity_register(mddev->gendisk,
1783 bdev_get_integrity(reference->bdev)) != 0) {
1784 printk(KERN_ERR "md: failed to register integrity for %s\n",
1788 printk(KERN_NOTICE "md: data integrity on %s enabled\n",
1792 EXPORT_SYMBOL(md_integrity_register);
1794 /* Disable data integrity if non-capable/non-matching disk is being added */
1795 void md_integrity_add_rdev(mdk_rdev_t *rdev, mddev_t *mddev)
1797 struct blk_integrity *bi_rdev = bdev_get_integrity(rdev->bdev);
1798 struct blk_integrity *bi_mddev = blk_get_integrity(mddev->gendisk);
1800 if (!bi_mddev) /* nothing to do */
1802 if (rdev->raid_disk < 0) /* skip spares */
1804 if (bi_rdev && blk_integrity_compare(mddev->gendisk,
1805 rdev->bdev->bd_disk) >= 0)
1807 printk(KERN_NOTICE "disabling data integrity on %s\n", mdname(mddev));
1808 blk_integrity_unregister(mddev->gendisk);
1810 EXPORT_SYMBOL(md_integrity_add_rdev);
1812 static int bind_rdev_to_array(mdk_rdev_t * rdev, mddev_t * mddev)
1814 char b[BDEVNAME_SIZE];
1824 /* prevent duplicates */
1825 if (find_rdev(mddev, rdev->bdev->bd_dev))
1828 /* make sure rdev->sectors exceeds mddev->dev_sectors */
1829 if (rdev->sectors && (mddev->dev_sectors == 0 ||
1830 rdev->sectors < mddev->dev_sectors)) {
1832 /* Cannot change size, so fail
1833 * If mddev->level <= 0, then we don't care
1834 * about aligning sizes (e.g. linear)
1836 if (mddev->level > 0)
1839 mddev->dev_sectors = rdev->sectors;
1842 /* Verify rdev->desc_nr is unique.
1843 * If it is -1, assign a free number, else
1844 * check number is not in use
1846 if (rdev->desc_nr < 0) {
1848 if (mddev->pers) choice = mddev->raid_disks;
1849 while (find_rdev_nr(mddev, choice))
1851 rdev->desc_nr = choice;
1853 if (find_rdev_nr(mddev, rdev->desc_nr))
1856 if (mddev->max_disks && rdev->desc_nr >= mddev->max_disks) {
1857 printk(KERN_WARNING "md: %s: array is limited to %d devices\n",
1858 mdname(mddev), mddev->max_disks);
1861 bdevname(rdev->bdev,b);
1862 while ( (s=strchr(b, '/')) != NULL)
1865 rdev->mddev = mddev;
1866 printk(KERN_INFO "md: bind<%s>\n", b);
1868 if ((err = kobject_add(&rdev->kobj, &mddev->kobj, "dev-%s", b)))
1871 ko = &part_to_dev(rdev->bdev->bd_part)->kobj;
1872 if (sysfs_create_link(&rdev->kobj, ko, "block"))
1873 /* failure here is OK */;
1874 rdev->sysfs_state = sysfs_get_dirent_safe(rdev->kobj.sd, "state");
1876 list_add_rcu(&rdev->same_set, &mddev->disks);
1877 bd_claim_by_disk(rdev->bdev, rdev->bdev->bd_holder, mddev->gendisk);
1879 /* May as well allow recovery to be retried once */
1880 mddev->recovery_disabled = 0;
1885 printk(KERN_WARNING "md: failed to register dev-%s for %s\n",
1890 static void md_delayed_delete(struct work_struct *ws)
1892 mdk_rdev_t *rdev = container_of(ws, mdk_rdev_t, del_work);
1893 kobject_del(&rdev->kobj);
1894 kobject_put(&rdev->kobj);
1897 static void unbind_rdev_from_array(mdk_rdev_t * rdev)
1899 char b[BDEVNAME_SIZE];
1904 bd_release_from_disk(rdev->bdev, rdev->mddev->gendisk);
1905 list_del_rcu(&rdev->same_set);
1906 printk(KERN_INFO "md: unbind<%s>\n", bdevname(rdev->bdev,b));
1908 sysfs_remove_link(&rdev->kobj, "block");
1909 sysfs_put(rdev->sysfs_state);
1910 rdev->sysfs_state = NULL;
1911 /* We need to delay this, otherwise we can deadlock when
1912 * writing to 'remove' to "dev/state". We also need
1913 * to delay it due to rcu usage.
1916 INIT_WORK(&rdev->del_work, md_delayed_delete);
1917 kobject_get(&rdev->kobj);
1918 schedule_work(&rdev->del_work);
1922 * prevent the device from being mounted, repartitioned or
1923 * otherwise reused by a RAID array (or any other kernel
1924 * subsystem), by bd_claiming the device.
1926 static int lock_rdev(mdk_rdev_t *rdev, dev_t dev, int shared)
1929 struct block_device *bdev;
1930 char b[BDEVNAME_SIZE];
1932 bdev = open_by_devnum(dev, FMODE_READ|FMODE_WRITE);
1934 printk(KERN_ERR "md: could not open %s.\n",
1935 __bdevname(dev, b));
1936 return PTR_ERR(bdev);
1938 err = bd_claim(bdev, shared ? (mdk_rdev_t *)lock_rdev : rdev);
1940 printk(KERN_ERR "md: could not bd_claim %s.\n",
1942 blkdev_put(bdev, FMODE_READ|FMODE_WRITE);
1946 set_bit(AllReserved, &rdev->flags);
1951 static void unlock_rdev(mdk_rdev_t *rdev)
1953 struct block_device *bdev = rdev->bdev;
1958 blkdev_put(bdev, FMODE_READ|FMODE_WRITE);
1961 void md_autodetect_dev(dev_t dev);
1963 static void export_rdev(mdk_rdev_t * rdev)
1965 char b[BDEVNAME_SIZE];
1966 printk(KERN_INFO "md: export_rdev(%s)\n",
1967 bdevname(rdev->bdev,b));
1972 if (test_bit(AutoDetected, &rdev->flags))
1973 md_autodetect_dev(rdev->bdev->bd_dev);
1976 kobject_put(&rdev->kobj);
1979 static void kick_rdev_from_array(mdk_rdev_t * rdev)
1981 unbind_rdev_from_array(rdev);
1985 static void export_array(mddev_t *mddev)
1987 mdk_rdev_t *rdev, *tmp;
1989 rdev_for_each(rdev, tmp, mddev) {
1994 kick_rdev_from_array(rdev);
1996 if (!list_empty(&mddev->disks))
1998 mddev->raid_disks = 0;
1999 mddev->major_version = 0;
2002 static void print_desc(mdp_disk_t *desc)
2004 printk(" DISK<N:%d,(%d,%d),R:%d,S:%d>\n", desc->number,
2005 desc->major,desc->minor,desc->raid_disk,desc->state);
2008 static void print_sb_90(mdp_super_t *sb)
2013 "md: SB: (V:%d.%d.%d) ID:<%08x.%08x.%08x.%08x> CT:%08x\n",
2014 sb->major_version, sb->minor_version, sb->patch_version,
2015 sb->set_uuid0, sb->set_uuid1, sb->set_uuid2, sb->set_uuid3,
2017 printk(KERN_INFO "md: L%d S%08d ND:%d RD:%d md%d LO:%d CS:%d\n",
2018 sb->level, sb->size, sb->nr_disks, sb->raid_disks,
2019 sb->md_minor, sb->layout, sb->chunk_size);
2020 printk(KERN_INFO "md: UT:%08x ST:%d AD:%d WD:%d"
2021 " FD:%d SD:%d CSUM:%08x E:%08lx\n",
2022 sb->utime, sb->state, sb->active_disks, sb->working_disks,
2023 sb->failed_disks, sb->spare_disks,
2024 sb->sb_csum, (unsigned long)sb->events_lo);
2027 for (i = 0; i < MD_SB_DISKS; i++) {
2030 desc = sb->disks + i;
2031 if (desc->number || desc->major || desc->minor ||
2032 desc->raid_disk || (desc->state && (desc->state != 4))) {
2033 printk(" D %2d: ", i);
2037 printk(KERN_INFO "md: THIS: ");
2038 print_desc(&sb->this_disk);
2041 static void print_sb_1(struct mdp_superblock_1 *sb)
2045 uuid = sb->set_uuid;
2047 "md: SB: (V:%u) (F:0x%08x) Array-ID:<%pU>\n"
2048 "md: Name: \"%s\" CT:%llu\n",
2049 le32_to_cpu(sb->major_version),
2050 le32_to_cpu(sb->feature_map),
2053 (unsigned long long)le64_to_cpu(sb->ctime)
2054 & MD_SUPERBLOCK_1_TIME_SEC_MASK);
2056 uuid = sb->device_uuid;
2058 "md: L%u SZ%llu RD:%u LO:%u CS:%u DO:%llu DS:%llu SO:%llu"
2060 "md: Dev:%08x UUID: %pU\n"
2061 "md: (F:0x%08x) UT:%llu Events:%llu ResyncOffset:%llu CSUM:0x%08x\n"
2062 "md: (MaxDev:%u) \n",
2063 le32_to_cpu(sb->level),
2064 (unsigned long long)le64_to_cpu(sb->size),
2065 le32_to_cpu(sb->raid_disks),
2066 le32_to_cpu(sb->layout),
2067 le32_to_cpu(sb->chunksize),
2068 (unsigned long long)le64_to_cpu(sb->data_offset),
2069 (unsigned long long)le64_to_cpu(sb->data_size),
2070 (unsigned long long)le64_to_cpu(sb->super_offset),
2071 (unsigned long long)le64_to_cpu(sb->recovery_offset),
2072 le32_to_cpu(sb->dev_number),
2075 (unsigned long long)le64_to_cpu(sb->utime) & MD_SUPERBLOCK_1_TIME_SEC_MASK,
2076 (unsigned long long)le64_to_cpu(sb->events),
2077 (unsigned long long)le64_to_cpu(sb->resync_offset),
2078 le32_to_cpu(sb->sb_csum),
2079 le32_to_cpu(sb->max_dev)
2083 static void print_rdev(mdk_rdev_t *rdev, int major_version)
2085 char b[BDEVNAME_SIZE];
2086 printk(KERN_INFO "md: rdev %s, Sect:%08llu F:%d S:%d DN:%u\n",
2087 bdevname(rdev->bdev, b), (unsigned long long)rdev->sectors,
2088 test_bit(Faulty, &rdev->flags), test_bit(In_sync, &rdev->flags),
2090 if (rdev->sb_loaded) {
2091 printk(KERN_INFO "md: rdev superblock (MJ:%d):\n", major_version);
2092 switch (major_version) {
2094 print_sb_90((mdp_super_t*)page_address(rdev->sb_page));
2097 print_sb_1((struct mdp_superblock_1 *)page_address(rdev->sb_page));
2101 printk(KERN_INFO "md: no rdev superblock!\n");
2104 static void md_print_devices(void)
2106 struct list_head *tmp;
2109 char b[BDEVNAME_SIZE];
2112 printk("md: **********************************\n");
2113 printk("md: * <COMPLETE RAID STATE PRINTOUT> *\n");
2114 printk("md: **********************************\n");
2115 for_each_mddev(mddev, tmp) {
2118 bitmap_print_sb(mddev->bitmap);
2120 printk("%s: ", mdname(mddev));
2121 list_for_each_entry(rdev, &mddev->disks, same_set)
2122 printk("<%s>", bdevname(rdev->bdev,b));
2125 list_for_each_entry(rdev, &mddev->disks, same_set)
2126 print_rdev(rdev, mddev->major_version);
2128 printk("md: **********************************\n");
2133 static void sync_sbs(mddev_t * mddev, int nospares)
2135 /* Update each superblock (in-memory image), but
2136 * if we are allowed to, skip spares which already
2137 * have the right event counter, or have one earlier
2138 * (which would mean they aren't being marked as dirty
2139 * with the rest of the array)
2142 list_for_each_entry(rdev, &mddev->disks, same_set) {
2143 if (rdev->sb_events == mddev->events ||
2145 rdev->raid_disk < 0 &&
2146 rdev->sb_events+1 == mddev->events)) {
2147 /* Don't update this superblock */
2148 rdev->sb_loaded = 2;
2150 super_types[mddev->major_version].
2151 sync_super(mddev, rdev);
2152 rdev->sb_loaded = 1;
2157 static void md_update_sb(mddev_t * mddev, int force_change)
2164 /* First make sure individual recovery_offsets are correct */
2165 list_for_each_entry(rdev, &mddev->disks, same_set) {
2166 if (rdev->raid_disk >= 0 &&
2167 mddev->delta_disks >= 0 &&
2168 !test_bit(In_sync, &rdev->flags) &&
2169 mddev->curr_resync_completed > rdev->recovery_offset)
2170 rdev->recovery_offset = mddev->curr_resync_completed;
2173 if (!mddev->persistent) {
2174 clear_bit(MD_CHANGE_CLEAN, &mddev->flags);
2175 clear_bit(MD_CHANGE_DEVS, &mddev->flags);
2176 wake_up(&mddev->sb_wait);
2180 spin_lock_irq(&mddev->write_lock);
2182 mddev->utime = get_seconds();
2184 if (test_and_clear_bit(MD_CHANGE_DEVS, &mddev->flags))
2186 if (test_and_clear_bit(MD_CHANGE_CLEAN, &mddev->flags))
2187 /* just a clean<-> dirty transition, possibly leave spares alone,
2188 * though if events isn't the right even/odd, we will have to do
2194 if (mddev->degraded)
2195 /* If the array is degraded, then skipping spares is both
2196 * dangerous and fairly pointless.
2197 * Dangerous because a device that was removed from the array
2198 * might have a event_count that still looks up-to-date,
2199 * so it can be re-added without a resync.
2200 * Pointless because if there are any spares to skip,
2201 * then a recovery will happen and soon that array won't
2202 * be degraded any more and the spare can go back to sleep then.
2206 sync_req = mddev->in_sync;
2208 /* If this is just a dirty<->clean transition, and the array is clean
2209 * and 'events' is odd, we can roll back to the previous clean state */
2211 && (mddev->in_sync && mddev->recovery_cp == MaxSector)
2212 && mddev->can_decrease_events
2213 && mddev->events != 1) {
2215 mddev->can_decrease_events = 0;
2217 /* otherwise we have to go forward and ... */
2219 mddev->can_decrease_events = nospares;
2222 if (!mddev->events) {
2224 * oops, this 64-bit counter should never wrap.
2225 * Either we are in around ~1 trillion A.C., assuming
2226 * 1 reboot per second, or we have a bug:
2231 sync_sbs(mddev, nospares);
2232 spin_unlock_irq(&mddev->write_lock);
2235 "md: updating %s RAID superblock on device (in sync %d)\n",
2236 mdname(mddev),mddev->in_sync);
2238 bitmap_update_sb(mddev->bitmap);
2239 list_for_each_entry(rdev, &mddev->disks, same_set) {
2240 char b[BDEVNAME_SIZE];
2241 dprintk(KERN_INFO "md: ");
2242 if (rdev->sb_loaded != 1)
2243 continue; /* no noise on spare devices */
2244 if (test_bit(Faulty, &rdev->flags))
2245 dprintk("(skipping faulty ");
2247 dprintk("%s ", bdevname(rdev->bdev,b));
2248 if (!test_bit(Faulty, &rdev->flags)) {
2249 md_super_write(mddev,rdev,
2250 rdev->sb_start, rdev->sb_size,
2252 dprintk(KERN_INFO "(write) %s's sb offset: %llu\n",
2253 bdevname(rdev->bdev,b),
2254 (unsigned long long)rdev->sb_start);
2255 rdev->sb_events = mddev->events;
2259 if (mddev->level == LEVEL_MULTIPATH)
2260 /* only need to write one superblock... */
2263 md_super_wait(mddev);
2264 /* if there was a failure, MD_CHANGE_DEVS was set, and we re-write super */
2266 spin_lock_irq(&mddev->write_lock);
2267 if (mddev->in_sync != sync_req ||
2268 test_bit(MD_CHANGE_DEVS, &mddev->flags)) {
2269 /* have to write it out again */
2270 spin_unlock_irq(&mddev->write_lock);
2273 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
2274 spin_unlock_irq(&mddev->write_lock);
2275 wake_up(&mddev->sb_wait);
2276 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
2277 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
2281 /* words written to sysfs files may, or may not, be \n terminated.
2282 * We want to accept with case. For this we use cmd_match.
2284 static int cmd_match(const char *cmd, const char *str)
2286 /* See if cmd, written into a sysfs file, matches
2287 * str. They must either be the same, or cmd can
2288 * have a trailing newline
2290 while (*cmd && *str && *cmd == *str) {
2301 struct rdev_sysfs_entry {
2302 struct attribute attr;
2303 ssize_t (*show)(mdk_rdev_t *, char *);
2304 ssize_t (*store)(mdk_rdev_t *, const char *, size_t);
2308 state_show(mdk_rdev_t *rdev, char *page)
2313 if (test_bit(Faulty, &rdev->flags)) {
2314 len+= sprintf(page+len, "%sfaulty",sep);
2317 if (test_bit(In_sync, &rdev->flags)) {
2318 len += sprintf(page+len, "%sin_sync",sep);
2321 if (test_bit(WriteMostly, &rdev->flags)) {
2322 len += sprintf(page+len, "%swrite_mostly",sep);
2325 if (test_bit(Blocked, &rdev->flags)) {
2326 len += sprintf(page+len, "%sblocked", sep);
2329 if (!test_bit(Faulty, &rdev->flags) &&
2330 !test_bit(In_sync, &rdev->flags)) {
2331 len += sprintf(page+len, "%sspare", sep);
2334 return len+sprintf(page+len, "\n");
2338 state_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2341 * faulty - simulates and error
2342 * remove - disconnects the device
2343 * writemostly - sets write_mostly
2344 * -writemostly - clears write_mostly
2345 * blocked - sets the Blocked flag
2346 * -blocked - clears the Blocked flag
2347 * insync - sets Insync providing device isn't active
2350 if (cmd_match(buf, "faulty") && rdev->mddev->pers) {
2351 md_error(rdev->mddev, rdev);
2353 } else if (cmd_match(buf, "remove")) {
2354 if (rdev->raid_disk >= 0)
2357 mddev_t *mddev = rdev->mddev;
2358 kick_rdev_from_array(rdev);
2360 md_update_sb(mddev, 1);
2361 md_new_event(mddev);
2364 } else if (cmd_match(buf, "writemostly")) {
2365 set_bit(WriteMostly, &rdev->flags);
2367 } else if (cmd_match(buf, "-writemostly")) {
2368 clear_bit(WriteMostly, &rdev->flags);
2370 } else if (cmd_match(buf, "blocked")) {
2371 set_bit(Blocked, &rdev->flags);
2373 } else if (cmd_match(buf, "-blocked")) {
2374 clear_bit(Blocked, &rdev->flags);
2375 wake_up(&rdev->blocked_wait);
2376 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2377 md_wakeup_thread(rdev->mddev->thread);
2380 } else if (cmd_match(buf, "insync") && rdev->raid_disk == -1) {
2381 set_bit(In_sync, &rdev->flags);
2385 sysfs_notify_dirent_safe(rdev->sysfs_state);
2386 return err ? err : len;
2388 static struct rdev_sysfs_entry rdev_state =
2389 __ATTR(state, S_IRUGO|S_IWUSR, state_show, state_store);
2392 errors_show(mdk_rdev_t *rdev, char *page)
2394 return sprintf(page, "%d\n", atomic_read(&rdev->corrected_errors));
2398 errors_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2401 unsigned long n = simple_strtoul(buf, &e, 10);
2402 if (*buf && (*e == 0 || *e == '\n')) {
2403 atomic_set(&rdev->corrected_errors, n);
2408 static struct rdev_sysfs_entry rdev_errors =
2409 __ATTR(errors, S_IRUGO|S_IWUSR, errors_show, errors_store);
2412 slot_show(mdk_rdev_t *rdev, char *page)
2414 if (rdev->raid_disk < 0)
2415 return sprintf(page, "none\n");
2417 return sprintf(page, "%d\n", rdev->raid_disk);
2421 slot_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2426 int slot = simple_strtoul(buf, &e, 10);
2427 if (strncmp(buf, "none", 4)==0)
2429 else if (e==buf || (*e && *e!= '\n'))
2431 if (rdev->mddev->pers && slot == -1) {
2432 /* Setting 'slot' on an active array requires also
2433 * updating the 'rd%d' link, and communicating
2434 * with the personality with ->hot_*_disk.
2435 * For now we only support removing
2436 * failed/spare devices. This normally happens automatically,
2437 * but not when the metadata is externally managed.
2439 if (rdev->raid_disk == -1)
2441 /* personality does all needed checks */
2442 if (rdev->mddev->pers->hot_add_disk == NULL)
2444 err = rdev->mddev->pers->
2445 hot_remove_disk(rdev->mddev, rdev->raid_disk);
2448 sprintf(nm, "rd%d", rdev->raid_disk);
2449 sysfs_remove_link(&rdev->mddev->kobj, nm);
2450 rdev->raid_disk = -1;
2451 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2452 md_wakeup_thread(rdev->mddev->thread);
2453 } else if (rdev->mddev->pers) {
2455 /* Activating a spare .. or possibly reactivating
2456 * if we ever get bitmaps working here.
2459 if (rdev->raid_disk != -1)
2462 if (rdev->mddev->pers->hot_add_disk == NULL)
2465 list_for_each_entry(rdev2, &rdev->mddev->disks, same_set)
2466 if (rdev2->raid_disk == slot)
2469 rdev->raid_disk = slot;
2470 if (test_bit(In_sync, &rdev->flags))
2471 rdev->saved_raid_disk = slot;
2473 rdev->saved_raid_disk = -1;
2474 err = rdev->mddev->pers->
2475 hot_add_disk(rdev->mddev, rdev);
2477 rdev->raid_disk = -1;
2480 sysfs_notify_dirent_safe(rdev->sysfs_state);
2481 sprintf(nm, "rd%d", rdev->raid_disk);
2482 if (sysfs_create_link(&rdev->mddev->kobj, &rdev->kobj, nm))
2483 /* failure here is OK */;
2484 /* don't wakeup anyone, leave that to userspace. */
2486 if (slot >= rdev->mddev->raid_disks)
2488 rdev->raid_disk = slot;
2489 /* assume it is working */
2490 clear_bit(Faulty, &rdev->flags);
2491 clear_bit(WriteMostly, &rdev->flags);
2492 set_bit(In_sync, &rdev->flags);
2493 sysfs_notify_dirent_safe(rdev->sysfs_state);
2499 static struct rdev_sysfs_entry rdev_slot =
2500 __ATTR(slot, S_IRUGO|S_IWUSR, slot_show, slot_store);
2503 offset_show(mdk_rdev_t *rdev, char *page)
2505 return sprintf(page, "%llu\n", (unsigned long long)rdev->data_offset);
2509 offset_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2512 unsigned long long offset = simple_strtoull(buf, &e, 10);
2513 if (e==buf || (*e && *e != '\n'))
2515 if (rdev->mddev->pers && rdev->raid_disk >= 0)
2517 if (rdev->sectors && rdev->mddev->external)
2518 /* Must set offset before size, so overlap checks
2521 rdev->data_offset = offset;
2525 static struct rdev_sysfs_entry rdev_offset =
2526 __ATTR(offset, S_IRUGO|S_IWUSR, offset_show, offset_store);
2529 rdev_size_show(mdk_rdev_t *rdev, char *page)
2531 return sprintf(page, "%llu\n", (unsigned long long)rdev->sectors / 2);
2534 static int overlaps(sector_t s1, sector_t l1, sector_t s2, sector_t l2)
2536 /* check if two start/length pairs overlap */
2544 static int strict_blocks_to_sectors(const char *buf, sector_t *sectors)
2546 unsigned long long blocks;
2549 if (strict_strtoull(buf, 10, &blocks) < 0)
2552 if (blocks & 1ULL << (8 * sizeof(blocks) - 1))
2553 return -EINVAL; /* sector conversion overflow */
2556 if (new != blocks * 2)
2557 return -EINVAL; /* unsigned long long to sector_t overflow */
2564 rdev_size_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2566 mddev_t *my_mddev = rdev->mddev;
2567 sector_t oldsectors = rdev->sectors;
2570 if (strict_blocks_to_sectors(buf, §ors) < 0)
2572 if (my_mddev->pers && rdev->raid_disk >= 0) {
2573 if (my_mddev->persistent) {
2574 sectors = super_types[my_mddev->major_version].
2575 rdev_size_change(rdev, sectors);
2578 } else if (!sectors)
2579 sectors = (rdev->bdev->bd_inode->i_size >> 9) -
2582 if (sectors < my_mddev->dev_sectors)
2583 return -EINVAL; /* component must fit device */
2585 rdev->sectors = sectors;
2586 if (sectors > oldsectors && my_mddev->external) {
2587 /* need to check that all other rdevs with the same ->bdev
2588 * do not overlap. We need to unlock the mddev to avoid
2589 * a deadlock. We have already changed rdev->sectors, and if
2590 * we have to change it back, we will have the lock again.
2594 struct list_head *tmp;
2596 mddev_unlock(my_mddev);
2597 for_each_mddev(mddev, tmp) {
2601 list_for_each_entry(rdev2, &mddev->disks, same_set)
2602 if (test_bit(AllReserved, &rdev2->flags) ||
2603 (rdev->bdev == rdev2->bdev &&
2605 overlaps(rdev->data_offset, rdev->sectors,
2611 mddev_unlock(mddev);
2617 mddev_lock(my_mddev);
2619 /* Someone else could have slipped in a size
2620 * change here, but doing so is just silly.
2621 * We put oldsectors back because we *know* it is
2622 * safe, and trust userspace not to race with
2625 rdev->sectors = oldsectors;
2632 static struct rdev_sysfs_entry rdev_size =
2633 __ATTR(size, S_IRUGO|S_IWUSR, rdev_size_show, rdev_size_store);
2636 static ssize_t recovery_start_show(mdk_rdev_t *rdev, char *page)
2638 unsigned long long recovery_start = rdev->recovery_offset;
2640 if (test_bit(In_sync, &rdev->flags) ||
2641 recovery_start == MaxSector)
2642 return sprintf(page, "none\n");
2644 return sprintf(page, "%llu\n", recovery_start);
2647 static ssize_t recovery_start_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2649 unsigned long long recovery_start;
2651 if (cmd_match(buf, "none"))
2652 recovery_start = MaxSector;
2653 else if (strict_strtoull(buf, 10, &recovery_start))
2656 if (rdev->mddev->pers &&
2657 rdev->raid_disk >= 0)
2660 rdev->recovery_offset = recovery_start;
2661 if (recovery_start == MaxSector)
2662 set_bit(In_sync, &rdev->flags);
2664 clear_bit(In_sync, &rdev->flags);
2668 static struct rdev_sysfs_entry rdev_recovery_start =
2669 __ATTR(recovery_start, S_IRUGO|S_IWUSR, recovery_start_show, recovery_start_store);
2671 static struct attribute *rdev_default_attrs[] = {
2677 &rdev_recovery_start.attr,
2681 rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
2683 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
2684 mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
2685 mddev_t *mddev = rdev->mddev;
2691 rv = mddev ? mddev_lock(mddev) : -EBUSY;
2693 if (rdev->mddev == NULL)
2696 rv = entry->show(rdev, page);
2697 mddev_unlock(mddev);
2703 rdev_attr_store(struct kobject *kobj, struct attribute *attr,
2704 const char *page, size_t length)
2706 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
2707 mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
2709 mddev_t *mddev = rdev->mddev;
2713 if (!capable(CAP_SYS_ADMIN))
2715 rv = mddev ? mddev_lock(mddev): -EBUSY;
2717 if (rdev->mddev == NULL)
2720 rv = entry->store(rdev, page, length);
2721 mddev_unlock(mddev);
2726 static void rdev_free(struct kobject *ko)
2728 mdk_rdev_t *rdev = container_of(ko, mdk_rdev_t, kobj);
2731 static const struct sysfs_ops rdev_sysfs_ops = {
2732 .show = rdev_attr_show,
2733 .store = rdev_attr_store,
2735 static struct kobj_type rdev_ktype = {
2736 .release = rdev_free,
2737 .sysfs_ops = &rdev_sysfs_ops,
2738 .default_attrs = rdev_default_attrs,
2741 void md_rdev_init(mdk_rdev_t *rdev)
2744 rdev->saved_raid_disk = -1;
2745 rdev->raid_disk = -1;
2747 rdev->data_offset = 0;
2748 rdev->sb_events = 0;
2749 rdev->last_read_error.tv_sec = 0;
2750 rdev->last_read_error.tv_nsec = 0;
2751 atomic_set(&rdev->nr_pending, 0);
2752 atomic_set(&rdev->read_errors, 0);
2753 atomic_set(&rdev->corrected_errors, 0);
2755 INIT_LIST_HEAD(&rdev->same_set);
2756 init_waitqueue_head(&rdev->blocked_wait);
2758 EXPORT_SYMBOL_GPL(md_rdev_init);
2760 * Import a device. If 'super_format' >= 0, then sanity check the superblock
2762 * mark the device faulty if:
2764 * - the device is nonexistent (zero size)
2765 * - the device has no valid superblock
2767 * a faulty rdev _never_ has rdev->sb set.
2769 static mdk_rdev_t *md_import_device(dev_t newdev, int super_format, int super_minor)
2771 char b[BDEVNAME_SIZE];
2776 rdev = kzalloc(sizeof(*rdev), GFP_KERNEL);
2778 printk(KERN_ERR "md: could not alloc mem for new device!\n");
2779 return ERR_PTR(-ENOMEM);
2783 if ((err = alloc_disk_sb(rdev)))
2786 err = lock_rdev(rdev, newdev, super_format == -2);
2790 kobject_init(&rdev->kobj, &rdev_ktype);
2792 size = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
2795 "md: %s has zero or unknown size, marking faulty!\n",
2796 bdevname(rdev->bdev,b));
2801 if (super_format >= 0) {
2802 err = super_types[super_format].
2803 load_super(rdev, NULL, super_minor);
2804 if (err == -EINVAL) {
2806 "md: %s does not have a valid v%d.%d "
2807 "superblock, not importing!\n",
2808 bdevname(rdev->bdev,b),
2809 super_format, super_minor);
2814 "md: could not read %s's sb, not importing!\n",
2815 bdevname(rdev->bdev,b));
2823 if (rdev->sb_page) {
2829 return ERR_PTR(err);
2833 * Check a full RAID array for plausibility
2837 static void analyze_sbs(mddev_t * mddev)
2840 mdk_rdev_t *rdev, *freshest, *tmp;
2841 char b[BDEVNAME_SIZE];
2844 rdev_for_each(rdev, tmp, mddev)
2845 switch (super_types[mddev->major_version].
2846 load_super(rdev, freshest, mddev->minor_version)) {
2854 "md: fatal superblock inconsistency in %s"
2855 " -- removing from array\n",
2856 bdevname(rdev->bdev,b));
2857 kick_rdev_from_array(rdev);
2861 super_types[mddev->major_version].
2862 validate_super(mddev, freshest);
2865 rdev_for_each(rdev, tmp, mddev) {
2866 if (mddev->max_disks &&
2867 (rdev->desc_nr >= mddev->max_disks ||
2868 i > mddev->max_disks)) {
2870 "md: %s: %s: only %d devices permitted\n",
2871 mdname(mddev), bdevname(rdev->bdev, b),
2873 kick_rdev_from_array(rdev);
2876 if (rdev != freshest)
2877 if (super_types[mddev->major_version].
2878 validate_super(mddev, rdev)) {
2879 printk(KERN_WARNING "md: kicking non-fresh %s"
2881 bdevname(rdev->bdev,b));
2882 kick_rdev_from_array(rdev);
2885 if (mddev->level == LEVEL_MULTIPATH) {
2886 rdev->desc_nr = i++;
2887 rdev->raid_disk = rdev->desc_nr;
2888 set_bit(In_sync, &rdev->flags);
2889 } else if (rdev->raid_disk >= (mddev->raid_disks - min(0, mddev->delta_disks))) {
2890 rdev->raid_disk = -1;
2891 clear_bit(In_sync, &rdev->flags);
2896 /* Read a fixed-point number.
2897 * Numbers in sysfs attributes should be in "standard" units where
2898 * possible, so time should be in seconds.
2899 * However we internally use a a much smaller unit such as
2900 * milliseconds or jiffies.
2901 * This function takes a decimal number with a possible fractional
2902 * component, and produces an integer which is the result of
2903 * multiplying that number by 10^'scale'.
2904 * all without any floating-point arithmetic.
2906 int strict_strtoul_scaled(const char *cp, unsigned long *res, int scale)
2908 unsigned long result = 0;
2910 while (isdigit(*cp) || (*cp == '.' && decimals < 0)) {
2913 else if (decimals < scale) {
2916 result = result * 10 + value;
2928 while (decimals < scale) {
2937 static void md_safemode_timeout(unsigned long data);
2940 safe_delay_show(mddev_t *mddev, char *page)
2942 int msec = (mddev->safemode_delay*1000)/HZ;
2943 return sprintf(page, "%d.%03d\n", msec/1000, msec%1000);
2946 safe_delay_store(mddev_t *mddev, const char *cbuf, size_t len)
2950 if (strict_strtoul_scaled(cbuf, &msec, 3) < 0)
2953 mddev->safemode_delay = 0;
2955 unsigned long old_delay = mddev->safemode_delay;
2956 mddev->safemode_delay = (msec*HZ)/1000;
2957 if (mddev->safemode_delay == 0)
2958 mddev->safemode_delay = 1;
2959 if (mddev->safemode_delay < old_delay)
2960 md_safemode_timeout((unsigned long)mddev);
2964 static struct md_sysfs_entry md_safe_delay =
2965 __ATTR(safe_mode_delay, S_IRUGO|S_IWUSR,safe_delay_show, safe_delay_store);
2968 level_show(mddev_t *mddev, char *page)
2970 struct mdk_personality *p = mddev->pers;
2972 return sprintf(page, "%s\n", p->name);
2973 else if (mddev->clevel[0])
2974 return sprintf(page, "%s\n", mddev->clevel);
2975 else if (mddev->level != LEVEL_NONE)
2976 return sprintf(page, "%d\n", mddev->level);
2982 level_store(mddev_t *mddev, const char *buf, size_t len)
2986 struct mdk_personality *pers;
2991 if (mddev->pers == NULL) {
2994 if (len >= sizeof(mddev->clevel))
2996 strncpy(mddev->clevel, buf, len);
2997 if (mddev->clevel[len-1] == '\n')
2999 mddev->clevel[len] = 0;
3000 mddev->level = LEVEL_NONE;
3004 /* request to change the personality. Need to ensure:
3005 * - array is not engaged in resync/recovery/reshape
3006 * - old personality can be suspended
3007 * - new personality will access other array.
3010 if (mddev->sync_thread ||
3011 mddev->reshape_position != MaxSector ||
3012 mddev->sysfs_active)
3015 if (!mddev->pers->quiesce) {
3016 printk(KERN_WARNING "md: %s: %s does not support online personality change\n",
3017 mdname(mddev), mddev->pers->name);
3021 /* Now find the new personality */
3022 if (len == 0 || len >= sizeof(clevel))
3024 strncpy(clevel, buf, len);
3025 if (clevel[len-1] == '\n')
3028 if (strict_strtol(clevel, 10, &level))
3031 if (request_module("md-%s", clevel) != 0)
3032 request_module("md-level-%s", clevel);
3033 spin_lock(&pers_lock);
3034 pers = find_pers(level, clevel);
3035 if (!pers || !try_module_get(pers->owner)) {
3036 spin_unlock(&pers_lock);
3037 printk(KERN_WARNING "md: personality %s not loaded\n", clevel);
3040 spin_unlock(&pers_lock);
3042 if (pers == mddev->pers) {
3043 /* Nothing to do! */
3044 module_put(pers->owner);
3047 if (!pers->takeover) {
3048 module_put(pers->owner);
3049 printk(KERN_WARNING "md: %s: %s does not support personality takeover\n",
3050 mdname(mddev), clevel);
3054 list_for_each_entry(rdev, &mddev->disks, same_set)
3055 rdev->new_raid_disk = rdev->raid_disk;
3057 /* ->takeover must set new_* and/or delta_disks
3058 * if it succeeds, and may set them when it fails.
3060 priv = pers->takeover(mddev);
3062 mddev->new_level = mddev->level;
3063 mddev->new_layout = mddev->layout;
3064 mddev->new_chunk_sectors = mddev->chunk_sectors;
3065 mddev->raid_disks -= mddev->delta_disks;
3066 mddev->delta_disks = 0;
3067 module_put(pers->owner);
3068 printk(KERN_WARNING "md: %s: %s would not accept array\n",
3069 mdname(mddev), clevel);
3070 return PTR_ERR(priv);
3073 /* Looks like we have a winner */
3074 mddev_suspend(mddev);
3075 mddev->pers->stop(mddev);
3077 if (mddev->pers->sync_request == NULL &&
3078 pers->sync_request != NULL) {
3079 /* need to add the md_redundancy_group */
3080 if (sysfs_create_group(&mddev->kobj, &md_redundancy_group))
3082 "md: cannot register extra attributes for %s\n",
3084 mddev->sysfs_action = sysfs_get_dirent(mddev->kobj.sd, NULL, "sync_action");
3086 if (mddev->pers->sync_request != NULL &&
3087 pers->sync_request == NULL) {
3088 /* need to remove the md_redundancy_group */
3089 if (mddev->to_remove == NULL)
3090 mddev->to_remove = &md_redundancy_group;
3093 if (mddev->pers->sync_request == NULL &&
3095 /* We are converting from a no-redundancy array
3096 * to a redundancy array and metadata is managed
3097 * externally so we need to be sure that writes
3098 * won't block due to a need to transition
3100 * until external management is started.
3103 mddev->safemode_delay = 0;
3104 mddev->safemode = 0;
3107 list_for_each_entry(rdev, &mddev->disks, same_set) {
3109 if (rdev->raid_disk < 0)
3111 if (rdev->new_raid_disk > mddev->raid_disks)
3112 rdev->new_raid_disk = -1;
3113 if (rdev->new_raid_disk == rdev->raid_disk)
3115 sprintf(nm, "rd%d", rdev->raid_disk);
3116 sysfs_remove_link(&mddev->kobj, nm);
3118 list_for_each_entry(rdev, &mddev->disks, same_set) {
3119 if (rdev->raid_disk < 0)
3121 if (rdev->new_raid_disk == rdev->raid_disk)
3123 rdev->raid_disk = rdev->new_raid_disk;
3124 if (rdev->raid_disk < 0)
3125 clear_bit(In_sync, &rdev->flags);
3128 sprintf(nm, "rd%d", rdev->raid_disk);
3129 if(sysfs_create_link(&mddev->kobj, &rdev->kobj, nm))
3130 printk("md: cannot register %s for %s after level change\n",
3135 module_put(mddev->pers->owner);
3137 mddev->private = priv;
3138 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
3139 mddev->level = mddev->new_level;
3140 mddev->layout = mddev->new_layout;
3141 mddev->chunk_sectors = mddev->new_chunk_sectors;
3142 mddev->delta_disks = 0;
3143 if (mddev->pers->sync_request == NULL) {
3144 /* this is now an array without redundancy, so
3145 * it must always be in_sync
3148 del_timer_sync(&mddev->safemode_timer);
3151 mddev_resume(mddev);
3152 set_bit(MD_CHANGE_DEVS, &mddev->flags);
3153 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3154 md_wakeup_thread(mddev->thread);
3155 sysfs_notify(&mddev->kobj, NULL, "level");
3156 md_new_event(mddev);
3160 static struct md_sysfs_entry md_level =
3161 __ATTR(level, S_IRUGO|S_IWUSR, level_show, level_store);
3165 layout_show(mddev_t *mddev, char *page)
3167 /* just a number, not meaningful for all levels */
3168 if (mddev->reshape_position != MaxSector &&
3169 mddev->layout != mddev->new_layout)
3170 return sprintf(page, "%d (%d)\n",
3171 mddev->new_layout, mddev->layout);
3172 return sprintf(page, "%d\n", mddev->layout);
3176 layout_store(mddev_t *mddev, const char *buf, size_t len)
3179 unsigned long n = simple_strtoul(buf, &e, 10);
3181 if (!*buf || (*e && *e != '\n'))
3186 if (mddev->pers->check_reshape == NULL)
3188 mddev->new_layout = n;
3189 err = mddev->pers->check_reshape(mddev);
3191 mddev->new_layout = mddev->layout;
3195 mddev->new_layout = n;
3196 if (mddev->reshape_position == MaxSector)
3201 static struct md_sysfs_entry md_layout =
3202 __ATTR(layout, S_IRUGO|S_IWUSR, layout_show, layout_store);
3206 raid_disks_show(mddev_t *mddev, char *page)
3208 if (mddev->raid_disks == 0)
3210 if (mddev->reshape_position != MaxSector &&
3211 mddev->delta_disks != 0)
3212 return sprintf(page, "%d (%d)\n", mddev->raid_disks,
3213 mddev->raid_disks - mddev->delta_disks);
3214 return sprintf(page, "%d\n", mddev->raid_disks);
3217 static int update_raid_disks(mddev_t *mddev, int raid_disks);
3220 raid_disks_store(mddev_t *mddev, const char *buf, size_t len)
3224 unsigned long n = simple_strtoul(buf, &e, 10);
3226 if (!*buf || (*e && *e != '\n'))
3230 rv = update_raid_disks(mddev, n);
3231 else if (mddev->reshape_position != MaxSector) {
3232 int olddisks = mddev->raid_disks - mddev->delta_disks;
3233 mddev->delta_disks = n - olddisks;
3234 mddev->raid_disks = n;
3236 mddev->raid_disks = n;
3237 return rv ? rv : len;
3239 static struct md_sysfs_entry md_raid_disks =
3240 __ATTR(raid_disks, S_IRUGO|S_IWUSR, raid_disks_show, raid_disks_store);
3243 chunk_size_show(mddev_t *mddev, char *page)
3245 if (mddev->reshape_position != MaxSector &&
3246 mddev->chunk_sectors != mddev->new_chunk_sectors)
3247 return sprintf(page, "%d (%d)\n",
3248 mddev->new_chunk_sectors << 9,
3249 mddev->chunk_sectors << 9);
3250 return sprintf(page, "%d\n", mddev->chunk_sectors << 9);
3254 chunk_size_store(mddev_t *mddev, const char *buf, size_t len)
3257 unsigned long n = simple_strtoul(buf, &e, 10);
3259 if (!*buf || (*e && *e != '\n'))
3264 if (mddev->pers->check_reshape == NULL)
3266 mddev->new_chunk_sectors = n >> 9;
3267 err = mddev->pers->check_reshape(mddev);
3269 mddev->new_chunk_sectors = mddev->chunk_sectors;
3273 mddev->new_chunk_sectors = n >> 9;
3274 if (mddev->reshape_position == MaxSector)
3275 mddev->chunk_sectors = n >> 9;
3279 static struct md_sysfs_entry md_chunk_size =
3280 __ATTR(chunk_size, S_IRUGO|S_IWUSR, chunk_size_show, chunk_size_store);
3283 resync_start_show(mddev_t *mddev, char *page)
3285 if (mddev->recovery_cp == MaxSector)
3286 return sprintf(page, "none\n");
3287 return sprintf(page, "%llu\n", (unsigned long long)mddev->recovery_cp);
3291 resync_start_store(mddev_t *mddev, const char *buf, size_t len)
3294 unsigned long long n = simple_strtoull(buf, &e, 10);
3298 if (cmd_match(buf, "none"))
3300 else if (!*buf || (*e && *e != '\n'))
3303 mddev->recovery_cp = n;
3306 static struct md_sysfs_entry md_resync_start =
3307 __ATTR(resync_start, S_IRUGO|S_IWUSR, resync_start_show, resync_start_store);
3310 * The array state can be:
3313 * No devices, no size, no level
3314 * Equivalent to STOP_ARRAY ioctl
3316 * May have some settings, but array is not active
3317 * all IO results in error
3318 * When written, doesn't tear down array, but just stops it
3319 * suspended (not supported yet)
3320 * All IO requests will block. The array can be reconfigured.
3321 * Writing this, if accepted, will block until array is quiescent
3323 * no resync can happen. no superblocks get written.
3324 * write requests fail
3326 * like readonly, but behaves like 'clean' on a write request.
3328 * clean - no pending writes, but otherwise active.
3329 * When written to inactive array, starts without resync
3330 * If a write request arrives then
3331 * if metadata is known, mark 'dirty' and switch to 'active'.
3332 * if not known, block and switch to write-pending
3333 * If written to an active array that has pending writes, then fails.
3335 * fully active: IO and resync can be happening.
3336 * When written to inactive array, starts with resync
3339 * clean, but writes are blocked waiting for 'active' to be written.
3342 * like active, but no writes have been seen for a while (100msec).
3345 enum array_state { clear, inactive, suspended, readonly, read_auto, clean, active,
3346 write_pending, active_idle, bad_word};
3347 static char *array_states[] = {
3348 "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active",
3349 "write-pending", "active-idle", NULL };
3351 static int match_word(const char *word, char **list)
3354 for (n=0; list[n]; n++)
3355 if (cmd_match(word, list[n]))
3361 array_state_show(mddev_t *mddev, char *page)
3363 enum array_state st = inactive;
3376 else if (test_bit(MD_CHANGE_PENDING, &mddev->flags))
3378 else if (mddev->safemode)
3384 if (list_empty(&mddev->disks) &&
3385 mddev->raid_disks == 0 &&
3386 mddev->dev_sectors == 0)
3391 return sprintf(page, "%s\n", array_states[st]);
3394 static int do_md_stop(mddev_t * mddev, int ro, int is_open);
3395 static int md_set_readonly(mddev_t * mddev, int is_open);
3396 static int do_md_run(mddev_t * mddev);
3397 static int restart_array(mddev_t *mddev);
3400 array_state_store(mddev_t *mddev, const char *buf, size_t len)
3403 enum array_state st = match_word(buf, array_states);
3408 /* stopping an active array */
3409 if (atomic_read(&mddev->openers) > 0)
3411 err = do_md_stop(mddev, 0, 0);
3414 /* stopping an active array */
3416 if (atomic_read(&mddev->openers) > 0)
3418 err = do_md_stop(mddev, 2, 0);
3420 err = 0; /* already inactive */
3423 break; /* not supported yet */
3426 err = md_set_readonly(mddev, 0);
3429 set_disk_ro(mddev->gendisk, 1);
3430 err = do_md_run(mddev);
3436 err = md_set_readonly(mddev, 0);
3437 else if (mddev->ro == 1)
3438 err = restart_array(mddev);
3441 set_disk_ro(mddev->gendisk, 0);
3445 err = do_md_run(mddev);
3450 restart_array(mddev);
3451 spin_lock_irq(&mddev->write_lock);
3452 if (atomic_read(&mddev->writes_pending) == 0) {
3453 if (mddev->in_sync == 0) {
3455 if (mddev->safemode == 1)
3456 mddev->safemode = 0;
3457 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
3462 spin_unlock_irq(&mddev->write_lock);
3468 restart_array(mddev);
3469 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
3470 wake_up(&mddev->sb_wait);
3474 set_disk_ro(mddev->gendisk, 0);
3475 err = do_md_run(mddev);
3480 /* these cannot be set */
3486 sysfs_notify_dirent_safe(mddev->sysfs_state);
3490 static struct md_sysfs_entry md_array_state =
3491 __ATTR(array_state, S_IRUGO|S_IWUSR, array_state_show, array_state_store);
3494 max_corrected_read_errors_show(mddev_t *mddev, char *page) {
3495 return sprintf(page, "%d\n",
3496 atomic_read(&mddev->max_corr_read_errors));
3500 max_corrected_read_errors_store(mddev_t *mddev, const char *buf, size_t len)
3503 unsigned long n = simple_strtoul(buf, &e, 10);
3505 if (*buf && (*e == 0 || *e == '\n')) {
3506 atomic_set(&mddev->max_corr_read_errors, n);
3512 static struct md_sysfs_entry max_corr_read_errors =
3513 __ATTR(max_read_errors, S_IRUGO|S_IWUSR, max_corrected_read_errors_show,
3514 max_corrected_read_errors_store);
3517 null_show(mddev_t *mddev, char *page)
3523 new_dev_store(mddev_t *mddev, const char *buf, size_t len)
3525 /* buf must be %d:%d\n? giving major and minor numbers */
3526 /* The new device is added to the array.
3527 * If the array has a persistent superblock, we read the
3528 * superblock to initialise info and check validity.
3529 * Otherwise, only checking done is that in bind_rdev_to_array,
3530 * which mainly checks size.
3533 int major = simple_strtoul(buf, &e, 10);
3539 if (!*buf || *e != ':' || !e[1] || e[1] == '\n')
3541 minor = simple_strtoul(e+1, &e, 10);
3542 if (*e && *e != '\n')
3544 dev = MKDEV(major, minor);
3545 if (major != MAJOR(dev) ||
3546 minor != MINOR(dev))
3550 if (mddev->persistent) {
3551 rdev = md_import_device(dev, mddev->major_version,
3552 mddev->minor_version);
3553 if (!IS_ERR(rdev) && !list_empty(&mddev->disks)) {
3554 mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
3555 mdk_rdev_t, same_set);
3556 err = super_types[mddev->major_version]
3557 .load_super(rdev, rdev0, mddev->minor_version);
3561 } else if (mddev->external)
3562 rdev = md_import_device(dev, -2, -1);
3564 rdev = md_import_device(dev, -1, -1);
3567 return PTR_ERR(rdev);
3568 err = bind_rdev_to_array(rdev, mddev);
3572 return err ? err : len;
3575 static struct md_sysfs_entry md_new_device =
3576 __ATTR(new_dev, S_IWUSR, null_show, new_dev_store);
3579 bitmap_store(mddev_t *mddev, const char *buf, size_t len)
3582 unsigned long chunk, end_chunk;
3586 /* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */
3588 chunk = end_chunk = simple_strtoul(buf, &end, 0);
3589 if (buf == end) break;
3590 if (*end == '-') { /* range */
3592 end_chunk = simple_strtoul(buf, &end, 0);
3593 if (buf == end) break;
3595 if (*end && !isspace(*end)) break;
3596 bitmap_dirty_bits(mddev->bitmap, chunk, end_chunk);
3597 buf = skip_spaces(end);
3599 bitmap_unplug(mddev->bitmap); /* flush the bits to disk */
3604 static struct md_sysfs_entry md_bitmap =
3605 __ATTR(bitmap_set_bits, S_IWUSR, null_show, bitmap_store);
3608 size_show(mddev_t *mddev, char *page)
3610 return sprintf(page, "%llu\n",
3611 (unsigned long long)mddev->dev_sectors / 2);
3614 static int update_size(mddev_t *mddev, sector_t num_sectors);
3617 size_store(mddev_t *mddev, const char *buf, size_t len)
3619 /* If array is inactive, we can reduce the component size, but
3620 * not increase it (except from 0).
3621 * If array is active, we can try an on-line resize
3624 int err = strict_blocks_to_sectors(buf, §ors);
3629 err = update_size(mddev, sectors);
3630 md_update_sb(mddev, 1);
3632 if (mddev->dev_sectors == 0 ||
3633 mddev->dev_sectors > sectors)
3634 mddev->dev_sectors = sectors;
3638 return err ? err : len;
3641 static struct md_sysfs_entry md_size =
3642 __ATTR(component_size, S_IRUGO|S_IWUSR, size_show, size_store);
3647 * 'none' for arrays with no metadata (good luck...)
3648 * 'external' for arrays with externally managed metadata,
3649 * or N.M for internally known formats
3652 metadata_show(mddev_t *mddev, char *page)
3654 if (mddev->persistent)
3655 return sprintf(page, "%d.%d\n",
3656 mddev->major_version, mddev->minor_version);
3657 else if (mddev->external)
3658 return sprintf(page, "external:%s\n", mddev->metadata_type);
3660 return sprintf(page, "none\n");
3664 metadata_store(mddev_t *mddev, const char *buf, size_t len)
3668 /* Changing the details of 'external' metadata is
3669 * always permitted. Otherwise there must be
3670 * no devices attached to the array.
3672 if (mddev->external && strncmp(buf, "external:", 9) == 0)
3674 else if (!list_empty(&mddev->disks))
3677 if (cmd_match(buf, "none")) {
3678 mddev->persistent = 0;
3679 mddev->external = 0;
3680 mddev->major_version = 0;
3681 mddev->minor_version = 90;
3684 if (strncmp(buf, "external:", 9) == 0) {
3685 size_t namelen = len-9;
3686 if (namelen >= sizeof(mddev->metadata_type))
3687 namelen = sizeof(mddev->metadata_type)-1;
3688 strncpy(mddev->metadata_type, buf+9, namelen);
3689 mddev->metadata_type[namelen] = 0;
3690 if (namelen && mddev->metadata_type[namelen-1] == '\n')
3691 mddev->metadata_type[--namelen] = 0;
3692 mddev->persistent = 0;
3693 mddev->external = 1;
3694 mddev->major_version = 0;
3695 mddev->minor_version = 90;
3698 major = simple_strtoul(buf, &e, 10);
3699 if (e==buf || *e != '.')
3702 minor = simple_strtoul(buf, &e, 10);
3703 if (e==buf || (*e && *e != '\n') )
3705 if (major >= ARRAY_SIZE(super_types) || super_types[major].name == NULL)
3707 mddev->major_version = major;
3708 mddev->minor_version = minor;
3709 mddev->persistent = 1;
3710 mddev->external = 0;
3714 static struct md_sysfs_entry md_metadata =
3715 __ATTR(metadata_version, S_IRUGO|S_IWUSR, metadata_show, metadata_store);
3718 action_show(mddev_t *mddev, char *page)
3720 char *type = "idle";
3721 if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
3723 else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
3724 (!mddev->ro && test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))) {
3725 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
3727 else if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
3728 if (!test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
3730 else if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
3734 } else if (test_bit(MD_RECOVERY_RECOVER, &mddev->recovery))
3737 return sprintf(page, "%s\n", type);
3741 action_store(mddev_t *mddev, const char *page, size_t len)
3743 if (!mddev->pers || !mddev->pers->sync_request)
3746 if (cmd_match(page, "frozen"))
3747 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
3749 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
3751 if (cmd_match(page, "idle") || cmd_match(page, "frozen")) {
3752 if (mddev->sync_thread) {
3753 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
3754 md_unregister_thread(mddev->sync_thread);
3755 mddev->sync_thread = NULL;
3756 mddev->recovery = 0;
3758 } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
3759 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
3761 else if (cmd_match(page, "resync"))
3762 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3763 else if (cmd_match(page, "recover")) {
3764 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
3765 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3766 } else if (cmd_match(page, "reshape")) {
3768 if (mddev->pers->start_reshape == NULL)
3770 err = mddev->pers->start_reshape(mddev);
3773 sysfs_notify(&mddev->kobj, NULL, "degraded");
3775 if (cmd_match(page, "check"))
3776 set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
3777 else if (!cmd_match(page, "repair"))
3779 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
3780 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
3782 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3783 md_wakeup_thread(mddev->thread);
3784 sysfs_notify_dirent_safe(mddev->sysfs_action);
3789 mismatch_cnt_show(mddev_t *mddev, char *page)
3791 return sprintf(page, "%llu\n",
3792 (unsigned long long) mddev->resync_mismatches);
3795 static struct md_sysfs_entry md_scan_mode =
3796 __ATTR(sync_action, S_IRUGO|S_IWUSR, action_show, action_store);
3799 static struct md_sysfs_entry md_mismatches = __ATTR_RO(mismatch_cnt);
3802 sync_min_show(mddev_t *mddev, char *page)
3804 return sprintf(page, "%d (%s)\n", speed_min(mddev),
3805 mddev->sync_speed_min ? "local": "system");
3809 sync_min_store(mddev_t *mddev, const char *buf, size_t len)
3813 if (strncmp(buf, "system", 6)==0) {
3814 mddev->sync_speed_min = 0;
3817 min = simple_strtoul(buf, &e, 10);
3818 if (buf == e || (*e && *e != '\n') || min <= 0)
3820 mddev->sync_speed_min = min;
3824 static struct md_sysfs_entry md_sync_min =
3825 __ATTR(sync_speed_min, S_IRUGO|S_IWUSR, sync_min_show, sync_min_store);
3828 sync_max_show(mddev_t *mddev, char *page)
3830 return sprintf(page, "%d (%s)\n", speed_max(mddev),
3831 mddev->sync_speed_max ? "local": "system");
3835 sync_max_store(mddev_t *mddev, const char *buf, size_t len)
3839 if (strncmp(buf, "system", 6)==0) {
3840 mddev->sync_speed_max = 0;
3843 max = simple_strtoul(buf, &e, 10);
3844 if (buf == e || (*e && *e != '\n') || max <= 0)
3846 mddev->sync_speed_max = max;
3850 static struct md_sysfs_entry md_sync_max =
3851 __ATTR(sync_speed_max, S_IRUGO|S_IWUSR, sync_max_show, sync_max_store);
3854 degraded_show(mddev_t *mddev, char *page)
3856 return sprintf(page, "%d\n", mddev->degraded);
3858 static struct md_sysfs_entry md_degraded = __ATTR_RO(degraded);
3861 sync_force_parallel_show(mddev_t *mddev, char *page)
3863 return sprintf(page, "%d\n", mddev->parallel_resync);
3867 sync_force_parallel_store(mddev_t *mddev, const char *buf, size_t len)
3871 if (strict_strtol(buf, 10, &n))
3874 if (n != 0 && n != 1)
3877 mddev->parallel_resync = n;
3879 if (mddev->sync_thread)
3880 wake_up(&resync_wait);
3885 /* force parallel resync, even with shared block devices */
3886 static struct md_sysfs_entry md_sync_force_parallel =
3887 __ATTR(sync_force_parallel, S_IRUGO|S_IWUSR,
3888 sync_force_parallel_show, sync_force_parallel_store);
3891 sync_speed_show(mddev_t *mddev, char *page)
3893 unsigned long resync, dt, db;
3894 if (mddev->curr_resync == 0)
3895 return sprintf(page, "none\n");
3896 resync = mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active);
3897 dt = (jiffies - mddev->resync_mark) / HZ;
3899 db = resync - mddev->resync_mark_cnt;
3900 return sprintf(page, "%lu\n", db/dt/2); /* K/sec */
3903 static struct md_sysfs_entry md_sync_speed = __ATTR_RO(sync_speed);
3906 sync_completed_show(mddev_t *mddev, char *page)
3908 unsigned long max_sectors, resync;
3910 if (!test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
3911 return sprintf(page, "none\n");
3913 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
3914 max_sectors = mddev->resync_max_sectors;
3916 max_sectors = mddev->dev_sectors;
3918 resync = mddev->curr_resync_completed;
3919 return sprintf(page, "%lu / %lu\n", resync, max_sectors);
3922 static struct md_sysfs_entry md_sync_completed = __ATTR_RO(sync_completed);
3925 min_sync_show(mddev_t *mddev, char *page)
3927 return sprintf(page, "%llu\n",
3928 (unsigned long long)mddev->resync_min);
3931 min_sync_store(mddev_t *mddev, const char *buf, size_t len)
3933 unsigned long long min;
3934 if (strict_strtoull(buf, 10, &min))
3936 if (min > mddev->resync_max)
3938 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
3941 /* Must be a multiple of chunk_size */
3942 if (mddev->chunk_sectors) {
3943 sector_t temp = min;
3944 if (sector_div(temp, mddev->chunk_sectors))
3947 mddev->resync_min = min;
3952 static struct md_sysfs_entry md_min_sync =
3953 __ATTR(sync_min, S_IRUGO|S_IWUSR, min_sync_show, min_sync_store);
3956 max_sync_show(mddev_t *mddev, char *page)
3958 if (mddev->resync_max == MaxSector)
3959 return sprintf(page, "max\n");
3961 return sprintf(page, "%llu\n",
3962 (unsigned long long)mddev->resync_max);
3965 max_sync_store(mddev_t *mddev, const char *buf, size_t len)
3967 if (strncmp(buf, "max", 3) == 0)
3968 mddev->resync_max = MaxSector;
3970 unsigned long long max;
3971 if (strict_strtoull(buf, 10, &max))
3973 if (max < mddev->resync_min)
3975 if (max < mddev->resync_max &&
3977 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
3980 /* Must be a multiple of chunk_size */
3981 if (mddev->chunk_sectors) {
3982 sector_t temp = max;
3983 if (sector_div(temp, mddev->chunk_sectors))
3986 mddev->resync_max = max;
3988 wake_up(&mddev->recovery_wait);
3992 static struct md_sysfs_entry md_max_sync =
3993 __ATTR(sync_max, S_IRUGO|S_IWUSR, max_sync_show, max_sync_store);
3996 suspend_lo_show(mddev_t *mddev, char *page)
3998 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_lo);
4002 suspend_lo_store(mddev_t *mddev, const char *buf, size_t len)
4005 unsigned long long new = simple_strtoull(buf, &e, 10);
4007 if (mddev->pers == NULL ||
4008 mddev->pers->quiesce == NULL)
4010 if (buf == e || (*e && *e != '\n'))
4012 if (new >= mddev->suspend_hi ||
4013 (new > mddev->suspend_lo && new < mddev->suspend_hi)) {
4014 mddev->suspend_lo = new;
4015 mddev->pers->quiesce(mddev, 2);
4020 static struct md_sysfs_entry md_suspend_lo =
4021 __ATTR(suspend_lo, S_IRUGO|S_IWUSR, suspend_lo_show, suspend_lo_store);
4025 suspend_hi_show(mddev_t *mddev, char *page)
4027 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_hi);
4031 suspend_hi_store(mddev_t *mddev, const char *buf, size_t len)
4034 unsigned long long new = simple_strtoull(buf, &e, 10);
4036 if (mddev->pers == NULL ||
4037 mddev->pers->quiesce == NULL)
4039 if (buf == e || (*e && *e != '\n'))
4041 if ((new <= mddev->suspend_lo && mddev->suspend_lo >= mddev->suspend_hi) ||
4042 (new > mddev->suspend_lo && new > mddev->suspend_hi)) {
4043 mddev->suspend_hi = new;
4044 mddev->pers->quiesce(mddev, 1);
4045 mddev->pers->quiesce(mddev, 0);
4050 static struct md_sysfs_entry md_suspend_hi =
4051 __ATTR(suspend_hi, S_IRUGO|S_IWUSR, suspend_hi_show, suspend_hi_store);
4054 reshape_position_show(mddev_t *mddev, char *page)
4056 if (mddev->reshape_position != MaxSector)
4057 return sprintf(page, "%llu\n",
4058 (unsigned long long)mddev->reshape_position);
4059 strcpy(page, "none\n");
4064 reshape_position_store(mddev_t *mddev, const char *buf, size_t len)
4067 unsigned long long new = simple_strtoull(buf, &e, 10);
4070 if (buf == e || (*e && *e != '\n'))
4072 mddev->reshape_position = new;
4073 mddev->delta_disks = 0;
4074 mddev->new_level = mddev->level;
4075 mddev->new_layout = mddev->layout;
4076 mddev->new_chunk_sectors = mddev->chunk_sectors;
4080 static struct md_sysfs_entry md_reshape_position =
4081 __ATTR(reshape_position, S_IRUGO|S_IWUSR, reshape_position_show,
4082 reshape_position_store);
4085 array_size_show(mddev_t *mddev, char *page)
4087 if (mddev->external_size)
4088 return sprintf(page, "%llu\n",
4089 (unsigned long long)mddev->array_sectors/2);
4091 return sprintf(page, "default\n");
4095 array_size_store(mddev_t *mddev, const char *buf, size_t len)
4099 if (strncmp(buf, "default", 7) == 0) {
4101 sectors = mddev->pers->size(mddev, 0, 0);
4103 sectors = mddev->array_sectors;
4105 mddev->external_size = 0;
4107 if (strict_blocks_to_sectors(buf, §ors) < 0)
4109 if (mddev->pers && mddev->pers->size(mddev, 0, 0) < sectors)
4112 mddev->external_size = 1;
4115 mddev->array_sectors = sectors;
4116 set_capacity(mddev->gendisk, mddev->array_sectors);
4118 revalidate_disk(mddev->gendisk);
4123 static struct md_sysfs_entry md_array_size =
4124 __ATTR(array_size, S_IRUGO|S_IWUSR, array_size_show,
4127 static struct attribute *md_default_attrs[] = {
4130 &md_raid_disks.attr,
4131 &md_chunk_size.attr,
4133 &md_resync_start.attr,
4135 &md_new_device.attr,
4136 &md_safe_delay.attr,
4137 &md_array_state.attr,
4138 &md_reshape_position.attr,
4139 &md_array_size.attr,
4140 &max_corr_read_errors.attr,
4144 static struct attribute *md_redundancy_attrs[] = {
4146 &md_mismatches.attr,
4149 &md_sync_speed.attr,
4150 &md_sync_force_parallel.attr,
4151 &md_sync_completed.attr,
4154 &md_suspend_lo.attr,
4155 &md_suspend_hi.attr,
4160 static struct attribute_group md_redundancy_group = {
4162 .attrs = md_redundancy_attrs,
4167 md_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
4169 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
4170 mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
4175 rv = mddev_lock(mddev);
4177 rv = entry->show(mddev, page);
4178 mddev_unlock(mddev);
4184 md_attr_store(struct kobject *kobj, struct attribute *attr,
4185 const char *page, size_t length)
4187 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
4188 mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
4193 if (!capable(CAP_SYS_ADMIN))
4195 rv = mddev_lock(mddev);
4196 if (mddev->hold_active == UNTIL_IOCTL)
4197 mddev->hold_active = 0;
4199 rv = entry->store(mddev, page, length);
4200 mddev_unlock(mddev);
4205 static void md_free(struct kobject *ko)
4207 mddev_t *mddev = container_of(ko, mddev_t, kobj);
4209 if (mddev->sysfs_state)
4210 sysfs_put(mddev->sysfs_state);
4212 if (mddev->gendisk) {
4213 del_gendisk(mddev->gendisk);
4214 put_disk(mddev->gendisk);
4217 blk_cleanup_queue(mddev->queue);
4222 static const struct sysfs_ops md_sysfs_ops = {
4223 .show = md_attr_show,
4224 .store = md_attr_store,
4226 static struct kobj_type md_ktype = {
4228 .sysfs_ops = &md_sysfs_ops,
4229 .default_attrs = md_default_attrs,
4234 static void mddev_delayed_delete(struct work_struct *ws)
4236 mddev_t *mddev = container_of(ws, mddev_t, del_work);
4238 sysfs_remove_group(&mddev->kobj, &md_bitmap_group);
4239 kobject_del(&mddev->kobj);
4240 kobject_put(&mddev->kobj);
4243 static int md_alloc(dev_t dev, char *name)
4245 static DEFINE_MUTEX(disks_mutex);
4246 mddev_t *mddev = mddev_find(dev);
4247 struct gendisk *disk;
4256 partitioned = (MAJOR(mddev->unit) != MD_MAJOR);
4257 shift = partitioned ? MdpMinorShift : 0;
4258 unit = MINOR(mddev->unit) >> shift;
4260 /* wait for any previous instance if this device
4261 * to be completed removed (mddev_delayed_delete).
4263 flush_scheduled_work();
4265 mutex_lock(&disks_mutex);
4271 /* Need to ensure that 'name' is not a duplicate.
4274 spin_lock(&all_mddevs_lock);
4276 list_for_each_entry(mddev2, &all_mddevs, all_mddevs)
4277 if (mddev2->gendisk &&
4278 strcmp(mddev2->gendisk->disk_name, name) == 0) {
4279 spin_unlock(&all_mddevs_lock);
4282 spin_unlock(&all_mddevs_lock);
4286 mddev->queue = blk_alloc_queue(GFP_KERNEL);
4289 mddev->queue->queuedata = mddev;
4291 /* Can be unlocked because the queue is new: no concurrency */
4292 queue_flag_set_unlocked(QUEUE_FLAG_CLUSTER, mddev->queue);
4294 blk_queue_make_request(mddev->queue, md_make_request);
4296 disk = alloc_disk(1 << shift);
4298 blk_cleanup_queue(mddev->queue);
4299 mddev->queue = NULL;
4302 disk->major = MAJOR(mddev->unit);
4303 disk->first_minor = unit << shift;
4305 strcpy(disk->disk_name, name);
4306 else if (partitioned)
4307 sprintf(disk->disk_name, "md_d%d", unit);
4309 sprintf(disk->disk_name, "md%d", unit);
4310 disk->fops = &md_fops;
4311 disk->private_data = mddev;
4312 disk->queue = mddev->queue;
4313 /* Allow extended partitions. This makes the
4314 * 'mdp' device redundant, but we can't really
4317 disk->flags |= GENHD_FL_EXT_DEVT;
4319 mddev->gendisk = disk;
4320 error = kobject_init_and_add(&mddev->kobj, &md_ktype,
4321 &disk_to_dev(disk)->kobj, "%s", "md");
4323 /* This isn't possible, but as kobject_init_and_add is marked
4324 * __must_check, we must do something with the result
4326 printk(KERN_WARNING "md: cannot register %s/md - name in use\n",
4330 if (mddev->kobj.sd &&
4331 sysfs_create_group(&mddev->kobj, &md_bitmap_group))
4332 printk(KERN_DEBUG "pointless warning\n");
4334 mutex_unlock(&disks_mutex);
4335 if (!error && mddev->kobj.sd) {
4336 kobject_uevent(&mddev->kobj, KOBJ_ADD);
4337 mddev->sysfs_state = sysfs_get_dirent_safe(mddev->kobj.sd, "array_state");
4343 static struct kobject *md_probe(dev_t dev, int *part, void *data)
4345 md_alloc(dev, NULL);
4349 static int add_named_array(const char *val, struct kernel_param *kp)
4351 /* val must be "md_*" where * is not all digits.
4352 * We allocate an array with a large free minor number, and
4353 * set the name to val. val must not already be an active name.
4355 int len = strlen(val);
4356 char buf[DISK_NAME_LEN];
4358 while (len && val[len-1] == '\n')
4360 if (len >= DISK_NAME_LEN)
4362 strlcpy(buf, val, len+1);
4363 if (strncmp(buf, "md_", 3) != 0)
4365 return md_alloc(0, buf);
4368 static void md_safemode_timeout(unsigned long data)
4370 mddev_t *mddev = (mddev_t *) data;
4372 if (!atomic_read(&mddev->writes_pending)) {
4373 mddev->safemode = 1;
4374 if (mddev->external)
4375 sysfs_notify_dirent_safe(mddev->sysfs_state);
4377 md_wakeup_thread(mddev->thread);
4380 static int start_dirty_degraded;
4382 int md_run(mddev_t *mddev)
4386 struct mdk_personality *pers;
4388 if (list_empty(&mddev->disks))
4389 /* cannot run an array with no devices.. */
4394 /* Cannot run until previous stop completes properly */
4395 if (mddev->sysfs_active)
4399 * Analyze all RAID superblock(s)
4401 if (!mddev->raid_disks) {
4402 if (!mddev->persistent)
4407 if (mddev->level != LEVEL_NONE)
4408 request_module("md-level-%d", mddev->level);
4409 else if (mddev->clevel[0])
4410 request_module("md-%s", mddev->clevel);
4413 * Drop all container device buffers, from now on
4414 * the only valid external interface is through the md
4417 list_for_each_entry(rdev, &mddev->disks, same_set) {
4418 if (test_bit(Faulty, &rdev->flags))
4420 sync_blockdev(rdev->bdev);
4421 invalidate_bdev(rdev->bdev);
4423 /* perform some consistency tests on the device.
4424 * We don't want the data to overlap the metadata,
4425 * Internal Bitmap issues have been handled elsewhere.
4427 if (rdev->data_offset < rdev->sb_start) {
4428 if (mddev->dev_sectors &&
4429 rdev->data_offset + mddev->dev_sectors
4431 printk("md: %s: data overlaps metadata\n",
4436 if (rdev->sb_start + rdev->sb_size/512
4437 > rdev->data_offset) {
4438 printk("md: %s: metadata overlaps data\n",
4443 sysfs_notify_dirent_safe(rdev->sysfs_state);
4446 spin_lock(&pers_lock);
4447 pers = find_pers(mddev->level, mddev->clevel);
4448 if (!pers || !try_module_get(pers->owner)) {
4449 spin_unlock(&pers_lock);
4450 if (mddev->level != LEVEL_NONE)
4451 printk(KERN_WARNING "md: personality for level %d is not loaded!\n",
4454 printk(KERN_WARNING "md: personality for level %s is not loaded!\n",
4459 spin_unlock(&pers_lock);
4460 if (mddev->level != pers->level) {
4461 mddev->level = pers->level;
4462 mddev->new_level = pers->level;
4464 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
4466 if (mddev->reshape_position != MaxSector &&
4467 pers->start_reshape == NULL) {
4468 /* This personality cannot handle reshaping... */
4470 module_put(pers->owner);
4474 if (pers->sync_request) {
4475 /* Warn if this is a potentially silly
4478 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
4482 list_for_each_entry(rdev, &mddev->disks, same_set)
4483 list_for_each_entry(rdev2, &mddev->disks, same_set) {
4485 rdev->bdev->bd_contains ==
4486 rdev2->bdev->bd_contains) {
4488 "%s: WARNING: %s appears to be"
4489 " on the same physical disk as"
4492 bdevname(rdev->bdev,b),
4493 bdevname(rdev2->bdev,b2));
4500 "True protection against single-disk"
4501 " failure might be compromised.\n");
4504 mddev->recovery = 0;
4505 /* may be over-ridden by personality */
4506 mddev->resync_max_sectors = mddev->dev_sectors;
4508 mddev->barriers_work = 1;
4509 mddev->ok_start_degraded = start_dirty_degraded;
4511 if (start_readonly && mddev->ro == 0)
4512 mddev->ro = 2; /* read-only, but switch on first write */
4514 err = mddev->pers->run(mddev);
4516 printk(KERN_ERR "md: pers->run() failed ...\n");
4517 else if (mddev->pers->size(mddev, 0, 0) < mddev->array_sectors) {
4518 WARN_ONCE(!mddev->external_size, "%s: default size too small,"
4519 " but 'external_size' not in effect?\n", __func__);
4521 "md: invalid array_size %llu > default size %llu\n",
4522 (unsigned long long)mddev->array_sectors / 2,
4523 (unsigned long long)mddev->pers->size(mddev, 0, 0) / 2);
4525 mddev->pers->stop(mddev);
4527 if (err == 0 && mddev->pers->sync_request) {
4528 err = bitmap_create(mddev);
4530 printk(KERN_ERR "%s: failed to create bitmap (%d)\n",
4531 mdname(mddev), err);
4532 mddev->pers->stop(mddev);
4536 module_put(mddev->pers->owner);
4538 bitmap_destroy(mddev);
4541 if (mddev->pers->sync_request) {
4542 if (mddev->kobj.sd &&
4543 sysfs_create_group(&mddev->kobj, &md_redundancy_group))
4545 "md: cannot register extra attributes for %s\n",
4547 mddev->sysfs_action = sysfs_get_dirent_safe(mddev->kobj.sd, "sync_action");
4548 } else if (mddev->ro == 2) /* auto-readonly not meaningful */
4551 atomic_set(&mddev->writes_pending,0);
4552 atomic_set(&mddev->max_corr_read_errors,
4553 MD_DEFAULT_MAX_CORRECTED_READ_ERRORS);
4554 mddev->safemode = 0;
4555 mddev->safemode_timer.function = md_safemode_timeout;
4556 mddev->safemode_timer.data = (unsigned long) mddev;
4557 mddev->safemode_delay = (200 * HZ)/1000 +1; /* 200 msec delay */
4560 list_for_each_entry(rdev, &mddev->disks, same_set)
4561 if (rdev->raid_disk >= 0) {
4563 sprintf(nm, "rd%d", rdev->raid_disk);
4564 if (sysfs_create_link(&mddev->kobj, &rdev->kobj, nm))
4565 /* failure here is OK */;
4568 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4571 md_update_sb(mddev, 0);
4573 md_wakeup_thread(mddev->thread);
4574 md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
4576 md_new_event(mddev);
4577 sysfs_notify_dirent_safe(mddev->sysfs_state);
4578 sysfs_notify_dirent_safe(mddev->sysfs_action);
4579 sysfs_notify(&mddev->kobj, NULL, "degraded");
4582 EXPORT_SYMBOL_GPL(md_run);
4584 static int do_md_run(mddev_t *mddev)
4588 err = md_run(mddev);
4591 err = bitmap_load(mddev);
4593 bitmap_destroy(mddev);
4596 set_capacity(mddev->gendisk, mddev->array_sectors);
4597 revalidate_disk(mddev->gendisk);
4598 kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
4603 static int restart_array(mddev_t *mddev)
4605 struct gendisk *disk = mddev->gendisk;
4607 /* Complain if it has no devices */
4608 if (list_empty(&mddev->disks))
4614 mddev->safemode = 0;
4616 set_disk_ro(disk, 0);
4617 printk(KERN_INFO "md: %s switched to read-write mode.\n",
4619 /* Kick recovery or resync if necessary */
4620 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4621 md_wakeup_thread(mddev->thread);
4622 md_wakeup_thread(mddev->sync_thread);
4623 sysfs_notify_dirent_safe(mddev->sysfs_state);
4627 /* similar to deny_write_access, but accounts for our holding a reference
4628 * to the file ourselves */
4629 static int deny_bitmap_write_access(struct file * file)
4631 struct inode *inode = file->f_mapping->host;
4633 spin_lock(&inode->i_lock);
4634 if (atomic_read(&inode->i_writecount) > 1) {
4635 spin_unlock(&inode->i_lock);
4638 atomic_set(&inode->i_writecount, -1);
4639 spin_unlock(&inode->i_lock);
4644 void restore_bitmap_write_access(struct file *file)
4646 struct inode *inode = file->f_mapping->host;
4648 spin_lock(&inode->i_lock);
4649 atomic_set(&inode->i_writecount, 1);
4650 spin_unlock(&inode->i_lock);
4653 static void md_clean(mddev_t *mddev)
4655 mddev->array_sectors = 0;
4656 mddev->external_size = 0;
4657 mddev->dev_sectors = 0;
4658 mddev->raid_disks = 0;
4659 mddev->recovery_cp = 0;
4660 mddev->resync_min = 0;
4661 mddev->resync_max = MaxSector;
4662 mddev->reshape_position = MaxSector;
4663 mddev->external = 0;
4664 mddev->persistent = 0;
4665 mddev->level = LEVEL_NONE;
4666 mddev->clevel[0] = 0;
4669 mddev->metadata_type[0] = 0;
4670 mddev->chunk_sectors = 0;
4671 mddev->ctime = mddev->utime = 0;
4673 mddev->max_disks = 0;
4675 mddev->can_decrease_events = 0;
4676 mddev->delta_disks = 0;
4677 mddev->new_level = LEVEL_NONE;
4678 mddev->new_layout = 0;
4679 mddev->new_chunk_sectors = 0;
4680 mddev->curr_resync = 0;
4681 mddev->resync_mismatches = 0;
4682 mddev->suspend_lo = mddev->suspend_hi = 0;
4683 mddev->sync_speed_min = mddev->sync_speed_max = 0;
4684 mddev->recovery = 0;
4686 mddev->degraded = 0;
4687 mddev->barriers_work = 0;
4688 mddev->safemode = 0;
4689 mddev->bitmap_info.offset = 0;
4690 mddev->bitmap_info.default_offset = 0;
4691 mddev->bitmap_info.chunksize = 0;
4692 mddev->bitmap_info.daemon_sleep = 0;
4693 mddev->bitmap_info.max_write_behind = 0;
4697 void md_stop_writes(mddev_t *mddev)
4699 if (mddev->sync_thread) {
4700 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4701 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
4702 md_unregister_thread(mddev->sync_thread);
4703 mddev->sync_thread = NULL;
4706 del_timer_sync(&mddev->safemode_timer);
4708 bitmap_flush(mddev);
4709 md_super_wait(mddev);
4711 if (!mddev->in_sync || mddev->flags) {
4712 /* mark array as shutdown cleanly */
4714 md_update_sb(mddev, 1);
4717 EXPORT_SYMBOL_GPL(md_stop_writes);
4719 void md_stop(mddev_t *mddev)
4721 mddev->pers->stop(mddev);
4722 if (mddev->pers->sync_request && mddev->to_remove == NULL)
4723 mddev->to_remove = &md_redundancy_group;
4724 module_put(mddev->pers->owner);
4726 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4728 EXPORT_SYMBOL_GPL(md_stop);
4730 static int md_set_readonly(mddev_t *mddev, int is_open)
4733 mutex_lock(&mddev->open_mutex);
4734 if (atomic_read(&mddev->openers) > is_open) {
4735 printk("md: %s still in use.\n",mdname(mddev));
4740 md_stop_writes(mddev);
4746 set_disk_ro(mddev->gendisk, 1);
4747 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4748 sysfs_notify_dirent_safe(mddev->sysfs_state);
4752 mutex_unlock(&mddev->open_mutex);
4757 * 0 - completely stop and dis-assemble array
4758 * 2 - stop but do not disassemble array
4760 static int do_md_stop(mddev_t * mddev, int mode, int is_open)
4762 struct gendisk *disk = mddev->gendisk;
4765 mutex_lock(&mddev->open_mutex);
4766 if (atomic_read(&mddev->openers) > is_open ||
4767 mddev->sysfs_active) {
4768 printk("md: %s still in use.\n",mdname(mddev));
4769 mutex_unlock(&mddev->open_mutex);
4775 set_disk_ro(disk, 0);
4777 md_stop_writes(mddev);
4779 mddev->queue->merge_bvec_fn = NULL;
4780 mddev->queue->unplug_fn = NULL;
4781 mddev->queue->backing_dev_info.congested_fn = NULL;
4783 /* tell userspace to handle 'inactive' */
4784 sysfs_notify_dirent_safe(mddev->sysfs_state);
4786 list_for_each_entry(rdev, &mddev->disks, same_set)
4787 if (rdev->raid_disk >= 0) {
4789 sprintf(nm, "rd%d", rdev->raid_disk);
4790 sysfs_remove_link(&mddev->kobj, nm);
4793 set_capacity(disk, 0);
4794 mutex_unlock(&mddev->open_mutex);
4795 revalidate_disk(disk);
4800 mutex_unlock(&mddev->open_mutex);
4802 * Free resources if final stop
4805 printk(KERN_INFO "md: %s stopped.\n", mdname(mddev));
4807 bitmap_destroy(mddev);
4808 if (mddev->bitmap_info.file) {
4809 restore_bitmap_write_access(mddev->bitmap_info.file);
4810 fput(mddev->bitmap_info.file);
4811 mddev->bitmap_info.file = NULL;
4813 mddev->bitmap_info.offset = 0;
4815 export_array(mddev);
4818 kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
4819 if (mddev->hold_active == UNTIL_STOP)
4820 mddev->hold_active = 0;
4822 blk_integrity_unregister(disk);
4823 md_new_event(mddev);
4824 sysfs_notify_dirent_safe(mddev->sysfs_state);
4829 static void autorun_array(mddev_t *mddev)
4834 if (list_empty(&mddev->disks))
4837 printk(KERN_INFO "md: running: ");
4839 list_for_each_entry(rdev, &mddev->disks, same_set) {
4840 char b[BDEVNAME_SIZE];
4841 printk("<%s>", bdevname(rdev->bdev,b));
4845 err = do_md_run(mddev);
4847 printk(KERN_WARNING "md: do_md_run() returned %d\n", err);
4848 do_md_stop(mddev, 0, 0);
4853 * lets try to run arrays based on all disks that have arrived
4854 * until now. (those are in pending_raid_disks)
4856 * the method: pick the first pending disk, collect all disks with
4857 * the same UUID, remove all from the pending list and put them into
4858 * the 'same_array' list. Then order this list based on superblock
4859 * update time (freshest comes first), kick out 'old' disks and
4860 * compare superblocks. If everything's fine then run it.
4862 * If "unit" is allocated, then bump its reference count
4864 static void autorun_devices(int part)
4866 mdk_rdev_t *rdev0, *rdev, *tmp;
4868 char b[BDEVNAME_SIZE];
4870 printk(KERN_INFO "md: autorun ...\n");
4871 while (!list_empty(&pending_raid_disks)) {
4874 LIST_HEAD(candidates);
4875 rdev0 = list_entry(pending_raid_disks.next,
4876 mdk_rdev_t, same_set);
4878 printk(KERN_INFO "md: considering %s ...\n",
4879 bdevname(rdev0->bdev,b));
4880 INIT_LIST_HEAD(&candidates);
4881 rdev_for_each_list(rdev, tmp, &pending_raid_disks)
4882 if (super_90_load(rdev, rdev0, 0) >= 0) {
4883 printk(KERN_INFO "md: adding %s ...\n",
4884 bdevname(rdev->bdev,b));
4885 list_move(&rdev->same_set, &candidates);
4888 * now we have a set of devices, with all of them having
4889 * mostly sane superblocks. It's time to allocate the
4893 dev = MKDEV(mdp_major,
4894 rdev0->preferred_minor << MdpMinorShift);
4895 unit = MINOR(dev) >> MdpMinorShift;
4897 dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
4900 if (rdev0->preferred_minor != unit) {
4901 printk(KERN_INFO "md: unit number in %s is bad: %d\n",
4902 bdevname(rdev0->bdev, b), rdev0->preferred_minor);
4906 md_probe(dev, NULL, NULL);
4907 mddev = mddev_find(dev);
4908 if (!mddev || !mddev->gendisk) {
4912 "md: cannot allocate memory for md drive.\n");
4915 if (mddev_lock(mddev))
4916 printk(KERN_WARNING "md: %s locked, cannot run\n",
4918 else if (mddev->raid_disks || mddev->major_version
4919 || !list_empty(&mddev->disks)) {
4921 "md: %s already running, cannot run %s\n",
4922 mdname(mddev), bdevname(rdev0->bdev,b));
4923 mddev_unlock(mddev);
4925 printk(KERN_INFO "md: created %s\n", mdname(mddev));
4926 mddev->persistent = 1;
4927 rdev_for_each_list(rdev, tmp, &candidates) {
4928 list_del_init(&rdev->same_set);
4929 if (bind_rdev_to_array(rdev, mddev))
4932 autorun_array(mddev);
4933 mddev_unlock(mddev);
4935 /* on success, candidates will be empty, on error
4938 rdev_for_each_list(rdev, tmp, &candidates) {
4939 list_del_init(&rdev->same_set);
4944 printk(KERN_INFO "md: ... autorun DONE.\n");
4946 #endif /* !MODULE */
4948 static int get_version(void __user * arg)
4952 ver.major = MD_MAJOR_VERSION;
4953 ver.minor = MD_MINOR_VERSION;
4954 ver.patchlevel = MD_PATCHLEVEL_VERSION;
4956 if (copy_to_user(arg, &ver, sizeof(ver)))
4962 static int get_array_info(mddev_t * mddev, void __user * arg)
4964 mdu_array_info_t info;
4965 int nr,working,insync,failed,spare;
4968 nr=working=insync=failed=spare=0;
4969 list_for_each_entry(rdev, &mddev->disks, same_set) {
4971 if (test_bit(Faulty, &rdev->flags))
4975 if (test_bit(In_sync, &rdev->flags))
4982 info.major_version = mddev->major_version;
4983 info.minor_version = mddev->minor_version;
4984 info.patch_version = MD_PATCHLEVEL_VERSION;
4985 info.ctime = mddev->ctime;
4986 info.level = mddev->level;
4987 info.size = mddev->dev_sectors / 2;
4988 if (info.size != mddev->dev_sectors / 2) /* overflow */
4991 info.raid_disks = mddev->raid_disks;
4992 info.md_minor = mddev->md_minor;
4993 info.not_persistent= !mddev->persistent;
4995 info.utime = mddev->utime;
4998 info.state = (1<<MD_SB_CLEAN);
4999 if (mddev->bitmap && mddev->bitmap_info.offset)
5000 info.state = (1<<MD_SB_BITMAP_PRESENT);
5001 info.active_disks = insync;
5002 info.working_disks = working;
5003 info.failed_disks = failed;
5004 info.spare_disks = spare;
5006 info.layout = mddev->layout;
5007 info.chunk_size = mddev->chunk_sectors << 9;
5009 if (copy_to_user(arg, &info, sizeof(info)))
5015 static int get_bitmap_file(mddev_t * mddev, void __user * arg)
5017 mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */
5018 char *ptr, *buf = NULL;
5021 if (md_allow_write(mddev))
5022 file = kmalloc(sizeof(*file), GFP_NOIO);
5024 file = kmalloc(sizeof(*file), GFP_KERNEL);
5029 /* bitmap disabled, zero the first byte and copy out */
5030 if (!mddev->bitmap || !mddev->bitmap->file) {
5031 file->pathname[0] = '\0';
5035 buf = kmalloc(sizeof(file->pathname), GFP_KERNEL);
5039 ptr = d_path(&mddev->bitmap->file->f_path, buf, sizeof(file->pathname));
5043 strcpy(file->pathname, ptr);
5047 if (copy_to_user(arg, file, sizeof(*file)))
5055 static int get_disk_info(mddev_t * mddev, void __user * arg)
5057 mdu_disk_info_t info;
5060 if (copy_from_user(&info, arg, sizeof(info)))
5063 rdev = find_rdev_nr(mddev, info.number);
5065 info.major = MAJOR(rdev->bdev->bd_dev);
5066 info.minor = MINOR(rdev->bdev->bd_dev);
5067 info.raid_disk = rdev->raid_disk;
5069 if (test_bit(Faulty, &rdev->flags))
5070 info.state |= (1<<MD_DISK_FAULTY);
5071 else if (test_bit(In_sync, &rdev->flags)) {
5072 info.state |= (1<<MD_DISK_ACTIVE);
5073 info.state |= (1<<MD_DISK_SYNC);
5075 if (test_bit(WriteMostly, &rdev->flags))
5076 info.state |= (1<<MD_DISK_WRITEMOSTLY);
5078 info.major = info.minor = 0;
5079 info.raid_disk = -1;
5080 info.state = (1<<MD_DISK_REMOVED);
5083 if (copy_to_user(arg, &info, sizeof(info)))
5089 static int add_new_disk(mddev_t * mddev, mdu_disk_info_t *info)
5091 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
5093 dev_t dev = MKDEV(info->major,info->minor);
5095 if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
5098 if (!mddev->raid_disks) {
5100 /* expecting a device which has a superblock */
5101 rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
5104 "md: md_import_device returned %ld\n",
5106 return PTR_ERR(rdev);
5108 if (!list_empty(&mddev->disks)) {
5109 mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
5110 mdk_rdev_t, same_set);
5111 err = super_types[mddev->major_version]
5112 .load_super(rdev, rdev0, mddev->minor_version);
5115 "md: %s has different UUID to %s\n",
5116 bdevname(rdev->bdev,b),
5117 bdevname(rdev0->bdev,b2));
5122 err = bind_rdev_to_array(rdev, mddev);
5129 * add_new_disk can be used once the array is assembled
5130 * to add "hot spares". They must already have a superblock
5135 if (!mddev->pers->hot_add_disk) {
5137 "%s: personality does not support diskops!\n",
5141 if (mddev->persistent)
5142 rdev = md_import_device(dev, mddev->major_version,
5143 mddev->minor_version);
5145 rdev = md_import_device(dev, -1, -1);
5148 "md: md_import_device returned %ld\n",
5150 return PTR_ERR(rdev);
5152 /* set save_raid_disk if appropriate */
5153 if (!mddev->persistent) {
5154 if (info->state & (1<<MD_DISK_SYNC) &&
5155 info->raid_disk < mddev->raid_disks)
5156 rdev->raid_disk = info->raid_disk;
5158 rdev->raid_disk = -1;
5160 super_types[mddev->major_version].
5161 validate_super(mddev, rdev);
5162 rdev->saved_raid_disk = rdev->raid_disk;
5164 clear_bit(In_sync, &rdev->flags); /* just to be sure */
5165 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
5166 set_bit(WriteMostly, &rdev->flags);
5168 clear_bit(WriteMostly, &rdev->flags);
5170 rdev->raid_disk = -1;
5171 err = bind_rdev_to_array(rdev, mddev);
5172 if (!err && !mddev->pers->hot_remove_disk) {
5173 /* If there is hot_add_disk but no hot_remove_disk
5174 * then added disks for geometry changes,
5175 * and should be added immediately.
5177 super_types[mddev->major_version].
5178 validate_super(mddev, rdev);
5179 err = mddev->pers->hot_add_disk(mddev, rdev);
5181 unbind_rdev_from_array(rdev);
5186 sysfs_notify_dirent_safe(rdev->sysfs_state);
5188 md_update_sb(mddev, 1);
5189 if (mddev->degraded)
5190 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
5191 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5192 md_wakeup_thread(mddev->thread);
5196 /* otherwise, add_new_disk is only allowed
5197 * for major_version==0 superblocks
5199 if (mddev->major_version != 0) {
5200 printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n",
5205 if (!(info->state & (1<<MD_DISK_FAULTY))) {
5207 rdev = md_import_device(dev, -1, 0);
5210 "md: error, md_import_device() returned %ld\n",
5212 return PTR_ERR(rdev);
5214 rdev->desc_nr = info->number;
5215 if (info->raid_disk < mddev->raid_disks)
5216 rdev->raid_disk = info->raid_disk;
5218 rdev->raid_disk = -1;
5220 if (rdev->raid_disk < mddev->raid_disks)
5221 if (info->state & (1<<MD_DISK_SYNC))
5222 set_bit(In_sync, &rdev->flags);
5224 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
5225 set_bit(WriteMostly, &rdev->flags);
5227 if (!mddev->persistent) {
5228 printk(KERN_INFO "md: nonpersistent superblock ...\n");
5229 rdev->sb_start = rdev->bdev->bd_inode->i_size / 512;
5231 rdev->sb_start = calc_dev_sboffset(rdev->bdev);
5232 rdev->sectors = rdev->sb_start;
5234 err = bind_rdev_to_array(rdev, mddev);
5244 static int hot_remove_disk(mddev_t * mddev, dev_t dev)
5246 char b[BDEVNAME_SIZE];
5249 rdev = find_rdev(mddev, dev);
5253 if (rdev->raid_disk >= 0)
5256 kick_rdev_from_array(rdev);
5257 md_update_sb(mddev, 1);
5258 md_new_event(mddev);
5262 printk(KERN_WARNING "md: cannot remove active disk %s from %s ...\n",
5263 bdevname(rdev->bdev,b), mdname(mddev));
5267 static int hot_add_disk(mddev_t * mddev, dev_t dev)
5269 char b[BDEVNAME_SIZE];
5276 if (mddev->major_version != 0) {
5277 printk(KERN_WARNING "%s: HOT_ADD may only be used with"
5278 " version-0 superblocks.\n",
5282 if (!mddev->pers->hot_add_disk) {
5284 "%s: personality does not support diskops!\n",
5289 rdev = md_import_device(dev, -1, 0);
5292 "md: error, md_import_device() returned %ld\n",
5297 if (mddev->persistent)
5298 rdev->sb_start = calc_dev_sboffset(rdev->bdev);
5300 rdev->sb_start = rdev->bdev->bd_inode->i_size / 512;
5302 rdev->sectors = rdev->sb_start;
5304 if (test_bit(Faulty, &rdev->flags)) {
5306 "md: can not hot-add faulty %s disk to %s!\n",
5307 bdevname(rdev->bdev,b), mdname(mddev));
5311 clear_bit(In_sync, &rdev->flags);
5313 rdev->saved_raid_disk = -1;
5314 err = bind_rdev_to_array(rdev, mddev);
5319 * The rest should better be atomic, we can have disk failures
5320 * noticed in interrupt contexts ...
5323 rdev->raid_disk = -1;
5325 md_update_sb(mddev, 1);
5328 * Kick recovery, maybe this spare has to be added to the
5329 * array immediately.
5331 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5332 md_wakeup_thread(mddev->thread);
5333 md_new_event(mddev);
5341 static int set_bitmap_file(mddev_t *mddev, int fd)
5346 if (!mddev->pers->quiesce)
5348 if (mddev->recovery || mddev->sync_thread)
5350 /* we should be able to change the bitmap.. */
5356 return -EEXIST; /* cannot add when bitmap is present */
5357 mddev->bitmap_info.file = fget(fd);
5359 if (mddev->bitmap_info.file == NULL) {
5360 printk(KERN_ERR "%s: error: failed to get bitmap file\n",
5365 err = deny_bitmap_write_access(mddev->bitmap_info.file);
5367 printk(KERN_ERR "%s: error: bitmap file is already in use\n",
5369 fput(mddev->bitmap_info.file);
5370 mddev->bitmap_info.file = NULL;
5373 mddev->bitmap_info.offset = 0; /* file overrides offset */
5374 } else if (mddev->bitmap == NULL)
5375 return -ENOENT; /* cannot remove what isn't there */
5378 mddev->pers->quiesce(mddev, 1);
5380 err = bitmap_create(mddev);
5382 err = bitmap_load(mddev);
5384 if (fd < 0 || err) {
5385 bitmap_destroy(mddev);
5386 fd = -1; /* make sure to put the file */
5388 mddev->pers->quiesce(mddev, 0);
5391 if (mddev->bitmap_info.file) {
5392 restore_bitmap_write_access(mddev->bitmap_info.file);
5393 fput(mddev->bitmap_info.file);
5395 mddev->bitmap_info.file = NULL;
5402 * set_array_info is used two different ways
5403 * The original usage is when creating a new array.
5404 * In this usage, raid_disks is > 0 and it together with
5405 * level, size, not_persistent,layout,chunksize determine the
5406 * shape of the array.
5407 * This will always create an array with a type-0.90.0 superblock.
5408 * The newer usage is when assembling an array.
5409 * In this case raid_disks will be 0, and the major_version field is
5410 * use to determine which style super-blocks are to be found on the devices.
5411 * The minor and patch _version numbers are also kept incase the
5412 * super_block handler wishes to interpret them.
5414 static int set_array_info(mddev_t * mddev, mdu_array_info_t *info)
5417 if (info->raid_disks == 0) {
5418 /* just setting version number for superblock loading */
5419 if (info->major_version < 0 ||
5420 info->major_version >= ARRAY_SIZE(super_types) ||
5421 super_types[info->major_version].name == NULL) {
5422 /* maybe try to auto-load a module? */
5424 "md: superblock version %d not known\n",
5425 info->major_version);
5428 mddev->major_version = info->major_version;
5429 mddev->minor_version = info->minor_version;
5430 mddev->patch_version = info->patch_version;
5431 mddev->persistent = !info->not_persistent;
5432 /* ensure mddev_put doesn't delete this now that there
5433 * is some minimal configuration.
5435 mddev->ctime = get_seconds();
5438 mddev->major_version = MD_MAJOR_VERSION;
5439 mddev->minor_version = MD_MINOR_VERSION;
5440 mddev->patch_version = MD_PATCHLEVEL_VERSION;
5441 mddev->ctime = get_seconds();
5443 mddev->level = info->level;
5444 mddev->clevel[0] = 0;
5445 mddev->dev_sectors = 2 * (sector_t)info->size;
5446 mddev->raid_disks = info->raid_disks;
5447 /* don't set md_minor, it is determined by which /dev/md* was
5450 if (info->state & (1<<MD_SB_CLEAN))
5451 mddev->recovery_cp = MaxSector;
5453 mddev->recovery_cp = 0;
5454 mddev->persistent = ! info->not_persistent;
5455 mddev->external = 0;
5457 mddev->layout = info->layout;
5458 mddev->chunk_sectors = info->chunk_size >> 9;
5460 mddev->max_disks = MD_SB_DISKS;
5462 if (mddev->persistent)
5464 set_bit(MD_CHANGE_DEVS, &mddev->flags);
5466 mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9;
5467 mddev->bitmap_info.offset = 0;
5469 mddev->reshape_position = MaxSector;
5472 * Generate a 128 bit UUID
5474 get_random_bytes(mddev->uuid, 16);
5476 mddev->new_level = mddev->level;
5477 mddev->new_chunk_sectors = mddev->chunk_sectors;
5478 mddev->new_layout = mddev->layout;
5479 mddev->delta_disks = 0;
5484 void md_set_array_sectors(mddev_t *mddev, sector_t array_sectors)
5486 WARN(!mddev_is_locked(mddev), "%s: unlocked mddev!\n", __func__);
5488 if (mddev->external_size)
5491 mddev->array_sectors = array_sectors;
5493 EXPORT_SYMBOL(md_set_array_sectors);
5495 static int update_size(mddev_t *mddev, sector_t num_sectors)
5499 int fit = (num_sectors == 0);
5501 if (mddev->pers->resize == NULL)
5503 /* The "num_sectors" is the number of sectors of each device that
5504 * is used. This can only make sense for arrays with redundancy.
5505 * linear and raid0 always use whatever space is available. We can only
5506 * consider changing this number if no resync or reconstruction is
5507 * happening, and if the new size is acceptable. It must fit before the
5508 * sb_start or, if that is <data_offset, it must fit before the size
5509 * of each device. If num_sectors is zero, we find the largest size
5513 if (mddev->sync_thread)
5516 /* Sorry, cannot grow a bitmap yet, just remove it,
5520 list_for_each_entry(rdev, &mddev->disks, same_set) {
5521 sector_t avail = rdev->sectors;
5523 if (fit && (num_sectors == 0 || num_sectors > avail))
5524 num_sectors = avail;
5525 if (avail < num_sectors)
5528 rv = mddev->pers->resize(mddev, num_sectors);
5530 revalidate_disk(mddev->gendisk);
5534 static int update_raid_disks(mddev_t *mddev, int raid_disks)
5537 /* change the number of raid disks */
5538 if (mddev->pers->check_reshape == NULL)
5540 if (raid_disks <= 0 ||
5541 (mddev->max_disks && raid_disks >= mddev->max_disks))
5543 if (mddev->sync_thread || mddev->reshape_position != MaxSector)
5545 mddev->delta_disks = raid_disks - mddev->raid_disks;
5547 rv = mddev->pers->check_reshape(mddev);
5553 * update_array_info is used to change the configuration of an
5555 * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
5556 * fields in the info are checked against the array.
5557 * Any differences that cannot be handled will cause an error.
5558 * Normally, only one change can be managed at a time.
5560 static int update_array_info(mddev_t *mddev, mdu_array_info_t *info)
5566 /* calculate expected state,ignoring low bits */
5567 if (mddev->bitmap && mddev->bitmap_info.offset)
5568 state |= (1 << MD_SB_BITMAP_PRESENT);
5570 if (mddev->major_version != info->major_version ||
5571 mddev->minor_version != info->minor_version ||
5572 /* mddev->patch_version != info->patch_version || */
5573 mddev->ctime != info->ctime ||
5574 mddev->level != info->level ||
5575 /* mddev->layout != info->layout || */
5576 !mddev->persistent != info->not_persistent||
5577 mddev->chunk_sectors != info->chunk_size >> 9 ||
5578 /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
5579 ((state^info->state) & 0xfffffe00)
5582 /* Check there is only one change */
5583 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
5585 if (mddev->raid_disks != info->raid_disks)
5587 if (mddev->layout != info->layout)
5589 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT))
5596 if (mddev->layout != info->layout) {
5598 * we don't need to do anything at the md level, the
5599 * personality will take care of it all.
5601 if (mddev->pers->check_reshape == NULL)
5604 mddev->new_layout = info->layout;
5605 rv = mddev->pers->check_reshape(mddev);
5607 mddev->new_layout = mddev->layout;
5611 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
5612 rv = update_size(mddev, (sector_t)info->size * 2);
5614 if (mddev->raid_disks != info->raid_disks)
5615 rv = update_raid_disks(mddev, info->raid_disks);
5617 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) {
5618 if (mddev->pers->quiesce == NULL)
5620 if (mddev->recovery || mddev->sync_thread)
5622 if (info->state & (1<<MD_SB_BITMAP_PRESENT)) {
5623 /* add the bitmap */
5626 if (mddev->bitmap_info.default_offset == 0)
5628 mddev->bitmap_info.offset =
5629 mddev->bitmap_info.default_offset;
5630 mddev->pers->quiesce(mddev, 1);
5631 rv = bitmap_create(mddev);
5633 rv = bitmap_load(mddev);
5635 bitmap_destroy(mddev);
5636 mddev->pers->quiesce(mddev, 0);
5638 /* remove the bitmap */
5641 if (mddev->bitmap->file)
5643 mddev->pers->quiesce(mddev, 1);
5644 bitmap_destroy(mddev);
5645 mddev->pers->quiesce(mddev, 0);
5646 mddev->bitmap_info.offset = 0;
5649 md_update_sb(mddev, 1);
5653 static int set_disk_faulty(mddev_t *mddev, dev_t dev)
5657 if (mddev->pers == NULL)
5660 rdev = find_rdev(mddev, dev);
5664 md_error(mddev, rdev);
5669 * We have a problem here : there is no easy way to give a CHS
5670 * virtual geometry. We currently pretend that we have a 2 heads
5671 * 4 sectors (with a BIG number of cylinders...). This drives
5672 * dosfs just mad... ;-)
5674 static int md_getgeo(struct block_device *bdev, struct hd_geometry *geo)
5676 mddev_t *mddev = bdev->bd_disk->private_data;
5680 geo->cylinders = mddev->array_sectors / 8;
5684 static int md_ioctl(struct block_device *bdev, fmode_t mode,
5685 unsigned int cmd, unsigned long arg)
5688 void __user *argp = (void __user *)arg;
5689 mddev_t *mddev = NULL;
5692 if (!capable(CAP_SYS_ADMIN))
5696 * Commands dealing with the RAID driver but not any
5702 err = get_version(argp);
5705 case PRINT_RAID_DEBUG:
5713 autostart_arrays(arg);
5720 * Commands creating/starting a new array:
5723 mddev = bdev->bd_disk->private_data;
5730 err = mddev_lock(mddev);
5733 "md: ioctl lock interrupted, reason %d, cmd %d\n",
5740 case SET_ARRAY_INFO:
5742 mdu_array_info_t info;
5744 memset(&info, 0, sizeof(info));
5745 else if (copy_from_user(&info, argp, sizeof(info))) {
5750 err = update_array_info(mddev, &info);
5752 printk(KERN_WARNING "md: couldn't update"
5753 " array info. %d\n", err);
5758 if (!list_empty(&mddev->disks)) {
5760 "md: array %s already has disks!\n",
5765 if (mddev->raid_disks) {
5767 "md: array %s already initialised!\n",
5772 err = set_array_info(mddev, &info);
5774 printk(KERN_WARNING "md: couldn't set"
5775 " array info. %d\n", err);
5785 * Commands querying/configuring an existing array:
5787 /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
5788 * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */
5789 if ((!mddev->raid_disks && !mddev->external)
5790 && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY
5791 && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE
5792 && cmd != GET_BITMAP_FILE) {
5798 * Commands even a read-only array can execute:
5802 case GET_ARRAY_INFO:
5803 err = get_array_info(mddev, argp);
5806 case GET_BITMAP_FILE:
5807 err = get_bitmap_file(mddev, argp);
5811 err = get_disk_info(mddev, argp);
5814 case RESTART_ARRAY_RW:
5815 err = restart_array(mddev);
5819 err = do_md_stop(mddev, 0, 1);
5823 err = md_set_readonly(mddev, 1);
5827 if (get_user(ro, (int __user *)(arg))) {
5833 /* if the bdev is going readonly the value of mddev->ro
5834 * does not matter, no writes are coming
5839 /* are we are already prepared for writes? */
5843 /* transitioning to readauto need only happen for
5844 * arrays that call md_write_start
5847 err = restart_array(mddev);
5850 set_disk_ro(mddev->gendisk, 0);
5857 * The remaining ioctls are changing the state of the
5858 * superblock, so we do not allow them on read-only arrays.
5859 * However non-MD ioctls (e.g. get-size) will still come through
5860 * here and hit the 'default' below, so only disallow
5861 * 'md' ioctls, and switch to rw mode if started auto-readonly.
5863 if (_IOC_TYPE(cmd) == MD_MAJOR && mddev->ro && mddev->pers) {
5864 if (mddev->ro == 2) {
5866 sysfs_notify_dirent_safe(mddev->sysfs_state);
5867 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5868 md_wakeup_thread(mddev->thread);
5879 mdu_disk_info_t info;
5880 if (copy_from_user(&info, argp, sizeof(info)))
5883 err = add_new_disk(mddev, &info);
5887 case HOT_REMOVE_DISK:
5888 err = hot_remove_disk(mddev, new_decode_dev(arg));
5892 err = hot_add_disk(mddev, new_decode_dev(arg));
5895 case SET_DISK_FAULTY:
5896 err = set_disk_faulty(mddev, new_decode_dev(arg));
5900 err = do_md_run(mddev);
5903 case SET_BITMAP_FILE:
5904 err = set_bitmap_file(mddev, (int)arg);
5914 if (mddev->hold_active == UNTIL_IOCTL &&
5916 mddev->hold_active = 0;
5917 mddev_unlock(mddev);
5926 #ifdef CONFIG_COMPAT
5927 static int md_compat_ioctl(struct block_device *bdev, fmode_t mode,
5928 unsigned int cmd, unsigned long arg)
5931 case HOT_REMOVE_DISK:
5933 case SET_DISK_FAULTY:
5934 case SET_BITMAP_FILE:
5935 /* These take in integer arg, do not convert */
5938 arg = (unsigned long)compat_ptr(arg);
5942 return md_ioctl(bdev, mode, cmd, arg);
5944 #endif /* CONFIG_COMPAT */
5946 static int md_open(struct block_device *bdev, fmode_t mode)
5949 * Succeed if we can lock the mddev, which confirms that
5950 * it isn't being stopped right now.
5952 mddev_t *mddev = mddev_find(bdev->bd_dev);
5955 mutex_lock(&md_mutex);
5956 if (mddev->gendisk != bdev->bd_disk) {
5957 /* we are racing with mddev_put which is discarding this
5961 /* Wait until bdev->bd_disk is definitely gone */
5962 flush_scheduled_work();
5963 /* Then retry the open from the top */
5964 mutex_unlock(&md_mutex);
5965 return -ERESTARTSYS;
5967 BUG_ON(mddev != bdev->bd_disk->private_data);
5969 if ((err = mutex_lock_interruptible(&mddev->open_mutex)))
5973 atomic_inc(&mddev->openers);
5974 mutex_unlock(&mddev->open_mutex);
5976 check_disk_size_change(mddev->gendisk, bdev);
5978 mutex_unlock(&md_mutex);
5982 static int md_release(struct gendisk *disk, fmode_t mode)
5984 mddev_t *mddev = disk->private_data;
5987 mutex_lock(&md_mutex);
5988 atomic_dec(&mddev->openers);
5990 mutex_unlock(&md_mutex);
5994 static const struct block_device_operations md_fops =
5996 .owner = THIS_MODULE,
5998 .release = md_release,
6000 #ifdef CONFIG_COMPAT
6001 .compat_ioctl = md_compat_ioctl,
6003 .getgeo = md_getgeo,
6006 static int md_thread(void * arg)
6008 mdk_thread_t *thread = arg;
6011 * md_thread is a 'system-thread', it's priority should be very
6012 * high. We avoid resource deadlocks individually in each
6013 * raid personality. (RAID5 does preallocation) We also use RR and
6014 * the very same RT priority as kswapd, thus we will never get
6015 * into a priority inversion deadlock.
6017 * we definitely have to have equal or higher priority than
6018 * bdflush, otherwise bdflush will deadlock if there are too
6019 * many dirty RAID5 blocks.
6022 allow_signal(SIGKILL);
6023 while (!kthread_should_stop()) {
6025 /* We need to wait INTERRUPTIBLE so that
6026 * we don't add to the load-average.
6027 * That means we need to be sure no signals are
6030 if (signal_pending(current))
6031 flush_signals(current);
6033 wait_event_interruptible_timeout
6035 test_bit(THREAD_WAKEUP, &thread->flags)
6036 || kthread_should_stop(),
6039 clear_bit(THREAD_WAKEUP, &thread->flags);
6041 thread->run(thread->mddev);
6047 void md_wakeup_thread(mdk_thread_t *thread)
6050 dprintk("md: waking up MD thread %s.\n", thread->tsk->comm);
6051 set_bit(THREAD_WAKEUP, &thread->flags);
6052 wake_up(&thread->wqueue);
6056 mdk_thread_t *md_register_thread(void (*run) (mddev_t *), mddev_t *mddev,
6059 mdk_thread_t *thread;
6061 thread = kzalloc(sizeof(mdk_thread_t), GFP_KERNEL);
6065 init_waitqueue_head(&thread->wqueue);
6068 thread->mddev = mddev;
6069 thread->timeout = MAX_SCHEDULE_TIMEOUT;
6070 thread->tsk = kthread_run(md_thread, thread,
6072 mdname(thread->mddev),
6073 name ?: mddev->pers->name);
6074 if (IS_ERR(thread->tsk)) {
6081 void md_unregister_thread(mdk_thread_t *thread)
6085 dprintk("interrupting MD-thread pid %d\n", task_pid_nr(thread->tsk));
6087 kthread_stop(thread->tsk);
6091 void md_error(mddev_t *mddev, mdk_rdev_t *rdev)
6098 if (!rdev || test_bit(Faulty, &rdev->flags))
6101 if (mddev->external)
6102 set_bit(Blocked, &rdev->flags);
6104 dprintk("md_error dev:%s, rdev:(%d:%d), (caller: %p,%p,%p,%p).\n",
6106 MAJOR(rdev->bdev->bd_dev), MINOR(rdev->bdev->bd_dev),
6107 __builtin_return_address(0),__builtin_return_address(1),
6108 __builtin_return_address(2),__builtin_return_address(3));
6112 if (!mddev->pers->error_handler)
6114 mddev->pers->error_handler(mddev,rdev);
6115 if (mddev->degraded)
6116 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
6117 sysfs_notify_dirent_safe(rdev->sysfs_state);
6118 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6119 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6120 md_wakeup_thread(mddev->thread);
6121 if (mddev->event_work.func)
6122 schedule_work(&mddev->event_work);
6123 md_new_event_inintr(mddev);
6126 /* seq_file implementation /proc/mdstat */
6128 static void status_unused(struct seq_file *seq)
6133 seq_printf(seq, "unused devices: ");
6135 list_for_each_entry(rdev, &pending_raid_disks, same_set) {
6136 char b[BDEVNAME_SIZE];
6138 seq_printf(seq, "%s ",
6139 bdevname(rdev->bdev,b));
6142 seq_printf(seq, "<none>");
6144 seq_printf(seq, "\n");
6148 static void status_resync(struct seq_file *seq, mddev_t * mddev)
6150 sector_t max_sectors, resync, res;
6151 unsigned long dt, db;
6154 unsigned int per_milli;
6156 resync = mddev->curr_resync - atomic_read(&mddev->recovery_active);
6158 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
6159 max_sectors = mddev->resync_max_sectors;
6161 max_sectors = mddev->dev_sectors;
6164 * Should not happen.
6170 /* Pick 'scale' such that (resync>>scale)*1000 will fit
6171 * in a sector_t, and (max_sectors>>scale) will fit in a
6172 * u32, as those are the requirements for sector_div.
6173 * Thus 'scale' must be at least 10
6176 if (sizeof(sector_t) > sizeof(unsigned long)) {
6177 while ( max_sectors/2 > (1ULL<<(scale+32)))
6180 res = (resync>>scale)*1000;
6181 sector_div(res, (u32)((max_sectors>>scale)+1));
6185 int i, x = per_milli/50, y = 20-x;
6186 seq_printf(seq, "[");
6187 for (i = 0; i < x; i++)
6188 seq_printf(seq, "=");
6189 seq_printf(seq, ">");
6190 for (i = 0; i < y; i++)
6191 seq_printf(seq, ".");
6192 seq_printf(seq, "] ");
6194 seq_printf(seq, " %s =%3u.%u%% (%llu/%llu)",
6195 (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)?
6197 (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)?
6199 (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
6200 "resync" : "recovery"))),
6201 per_milli/10, per_milli % 10,
6202 (unsigned long long) resync/2,
6203 (unsigned long long) max_sectors/2);
6206 * dt: time from mark until now
6207 * db: blocks written from mark until now
6208 * rt: remaining time
6210 * rt is a sector_t, so could be 32bit or 64bit.
6211 * So we divide before multiply in case it is 32bit and close
6213 * We scale the divisor (db) by 32 to avoid loosing precision
6214 * near the end of resync when the number of remaining sectors
6216 * We then divide rt by 32 after multiplying by db to compensate.
6217 * The '+1' avoids division by zero if db is very small.
6219 dt = ((jiffies - mddev->resync_mark) / HZ);
6221 db = (mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active))
6222 - mddev->resync_mark_cnt;
6224 rt = max_sectors - resync; /* number of remaining sectors */
6225 sector_div(rt, db/32+1);
6229 seq_printf(seq, " finish=%lu.%lumin", (unsigned long)rt / 60,
6230 ((unsigned long)rt % 60)/6);
6232 seq_printf(seq, " speed=%ldK/sec", db/2/dt);
6235 static void *md_seq_start(struct seq_file *seq, loff_t *pos)
6237 struct list_head *tmp;
6247 spin_lock(&all_mddevs_lock);
6248 list_for_each(tmp,&all_mddevs)
6250 mddev = list_entry(tmp, mddev_t, all_mddevs);
6252 spin_unlock(&all_mddevs_lock);
6255 spin_unlock(&all_mddevs_lock);
6257 return (void*)2;/* tail */
6261 static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
6263 struct list_head *tmp;
6264 mddev_t *next_mddev, *mddev = v;
6270 spin_lock(&all_mddevs_lock);
6272 tmp = all_mddevs.next;
6274 tmp = mddev->all_mddevs.next;
6275 if (tmp != &all_mddevs)
6276 next_mddev = mddev_get(list_entry(tmp,mddev_t,all_mddevs));
6278 next_mddev = (void*)2;
6281 spin_unlock(&all_mddevs_lock);
6289 static void md_seq_stop(struct seq_file *seq, void *v)
6293 if (mddev && v != (void*)1 && v != (void*)2)
6297 struct mdstat_info {
6301 static int md_seq_show(struct seq_file *seq, void *v)
6306 struct mdstat_info *mi = seq->private;
6307 struct bitmap *bitmap;
6309 if (v == (void*)1) {
6310 struct mdk_personality *pers;
6311 seq_printf(seq, "Personalities : ");
6312 spin_lock(&pers_lock);
6313 list_for_each_entry(pers, &pers_list, list)
6314 seq_printf(seq, "[%s] ", pers->name);
6316 spin_unlock(&pers_lock);
6317 seq_printf(seq, "\n");
6318 mi->event = atomic_read(&md_event_count);
6321 if (v == (void*)2) {
6326 if (mddev_lock(mddev) < 0)
6329 if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
6330 seq_printf(seq, "%s : %sactive", mdname(mddev),
6331 mddev->pers ? "" : "in");
6334 seq_printf(seq, " (read-only)");
6336 seq_printf(seq, " (auto-read-only)");
6337 seq_printf(seq, " %s", mddev->pers->name);
6341 list_for_each_entry(rdev, &mddev->disks, same_set) {
6342 char b[BDEVNAME_SIZE];
6343 seq_printf(seq, " %s[%d]",
6344 bdevname(rdev->bdev,b), rdev->desc_nr);
6345 if (test_bit(WriteMostly, &rdev->flags))
6346 seq_printf(seq, "(W)");
6347 if (test_bit(Faulty, &rdev->flags)) {
6348 seq_printf(seq, "(F)");
6350 } else if (rdev->raid_disk < 0)
6351 seq_printf(seq, "(S)"); /* spare */
6352 sectors += rdev->sectors;
6355 if (!list_empty(&mddev->disks)) {
6357 seq_printf(seq, "\n %llu blocks",
6358 (unsigned long long)
6359 mddev->array_sectors / 2);
6361 seq_printf(seq, "\n %llu blocks",
6362 (unsigned long long)sectors / 2);
6364 if (mddev->persistent) {
6365 if (mddev->major_version != 0 ||
6366 mddev->minor_version != 90) {
6367 seq_printf(seq," super %d.%d",
6368 mddev->major_version,
6369 mddev->minor_version);
6371 } else if (mddev->external)
6372 seq_printf(seq, " super external:%s",
6373 mddev->metadata_type);
6375 seq_printf(seq, " super non-persistent");
6378 mddev->pers->status(seq, mddev);
6379 seq_printf(seq, "\n ");
6380 if (mddev->pers->sync_request) {
6381 if (mddev->curr_resync > 2) {
6382 status_resync(seq, mddev);
6383 seq_printf(seq, "\n ");
6384 } else if (mddev->curr_resync == 1 || mddev->curr_resync == 2)
6385 seq_printf(seq, "\tresync=DELAYED\n ");
6386 else if (mddev->recovery_cp < MaxSector)
6387 seq_printf(seq, "\tresync=PENDING\n ");
6390 seq_printf(seq, "\n ");
6392 if ((bitmap = mddev->bitmap)) {
6393 unsigned long chunk_kb;
6394 unsigned long flags;
6395 spin_lock_irqsave(&bitmap->lock, flags);
6396 chunk_kb = mddev->bitmap_info.chunksize >> 10;
6397 seq_printf(seq, "bitmap: %lu/%lu pages [%luKB], "
6399 bitmap->pages - bitmap->missing_pages,
6401 (bitmap->pages - bitmap->missing_pages)
6402 << (PAGE_SHIFT - 10),
6403 chunk_kb ? chunk_kb : mddev->bitmap_info.chunksize,
6404 chunk_kb ? "KB" : "B");
6406 seq_printf(seq, ", file: ");
6407 seq_path(seq, &bitmap->file->f_path, " \t\n");
6410 seq_printf(seq, "\n");
6411 spin_unlock_irqrestore(&bitmap->lock, flags);
6414 seq_printf(seq, "\n");
6416 mddev_unlock(mddev);
6421 static const struct seq_operations md_seq_ops = {
6422 .start = md_seq_start,
6423 .next = md_seq_next,
6424 .stop = md_seq_stop,
6425 .show = md_seq_show,
6428 static int md_seq_open(struct inode *inode, struct file *file)
6431 struct mdstat_info *mi = kmalloc(sizeof(*mi), GFP_KERNEL);
6435 error = seq_open(file, &md_seq_ops);
6439 struct seq_file *p = file->private_data;
6441 mi->event = atomic_read(&md_event_count);
6446 static unsigned int mdstat_poll(struct file *filp, poll_table *wait)
6448 struct seq_file *m = filp->private_data;
6449 struct mdstat_info *mi = m->private;
6452 poll_wait(filp, &md_event_waiters, wait);
6454 /* always allow read */
6455 mask = POLLIN | POLLRDNORM;
6457 if (mi->event != atomic_read(&md_event_count))
6458 mask |= POLLERR | POLLPRI;
6462 static const struct file_operations md_seq_fops = {
6463 .owner = THIS_MODULE,
6464 .open = md_seq_open,
6466 .llseek = seq_lseek,
6467 .release = seq_release_private,
6468 .poll = mdstat_poll,
6471 int register_md_personality(struct mdk_personality *p)
6473 spin_lock(&pers_lock);
6474 list_add_tail(&p->list, &pers_list);
6475 printk(KERN_INFO "md: %s personality registered for level %d\n", p->name, p->level);
6476 spin_unlock(&pers_lock);
6480 int unregister_md_personality(struct mdk_personality *p)
6482 printk(KERN_INFO "md: %s personality unregistered\n", p->name);
6483 spin_lock(&pers_lock);
6484 list_del_init(&p->list);
6485 spin_unlock(&pers_lock);
6489 static int is_mddev_idle(mddev_t *mddev, int init)
6497 rdev_for_each_rcu(rdev, mddev) {
6498 struct gendisk *disk = rdev->bdev->bd_contains->bd_disk;
6499 curr_events = (int)part_stat_read(&disk->part0, sectors[0]) +
6500 (int)part_stat_read(&disk->part0, sectors[1]) -
6501 atomic_read(&disk->sync_io);
6502 /* sync IO will cause sync_io to increase before the disk_stats
6503 * as sync_io is counted when a request starts, and
6504 * disk_stats is counted when it completes.
6505 * So resync activity will cause curr_events to be smaller than
6506 * when there was no such activity.
6507 * non-sync IO will cause disk_stat to increase without
6508 * increasing sync_io so curr_events will (eventually)
6509 * be larger than it was before. Once it becomes
6510 * substantially larger, the test below will cause
6511 * the array to appear non-idle, and resync will slow
6513 * If there is a lot of outstanding resync activity when
6514 * we set last_event to curr_events, then all that activity
6515 * completing might cause the array to appear non-idle
6516 * and resync will be slowed down even though there might
6517 * not have been non-resync activity. This will only
6518 * happen once though. 'last_events' will soon reflect
6519 * the state where there is little or no outstanding
6520 * resync requests, and further resync activity will
6521 * always make curr_events less than last_events.
6524 if (init || curr_events - rdev->last_events > 64) {
6525 rdev->last_events = curr_events;
6533 void md_done_sync(mddev_t *mddev, int blocks, int ok)
6535 /* another "blocks" (512byte) blocks have been synced */
6536 atomic_sub(blocks, &mddev->recovery_active);
6537 wake_up(&mddev->recovery_wait);
6539 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6540 md_wakeup_thread(mddev->thread);
6541 // stop recovery, signal do_sync ....
6546 /* md_write_start(mddev, bi)
6547 * If we need to update some array metadata (e.g. 'active' flag
6548 * in superblock) before writing, schedule a superblock update
6549 * and wait for it to complete.
6551 void md_write_start(mddev_t *mddev, struct bio *bi)
6554 if (bio_data_dir(bi) != WRITE)
6557 BUG_ON(mddev->ro == 1);
6558 if (mddev->ro == 2) {
6559 /* need to switch to read/write */
6561 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6562 md_wakeup_thread(mddev->thread);
6563 md_wakeup_thread(mddev->sync_thread);
6566 atomic_inc(&mddev->writes_pending);
6567 if (mddev->safemode == 1)
6568 mddev->safemode = 0;
6569 if (mddev->in_sync) {
6570 spin_lock_irq(&mddev->write_lock);
6571 if (mddev->in_sync) {
6573 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
6574 set_bit(MD_CHANGE_PENDING, &mddev->flags);
6575 md_wakeup_thread(mddev->thread);
6578 spin_unlock_irq(&mddev->write_lock);
6581 sysfs_notify_dirent_safe(mddev->sysfs_state);
6582 wait_event(mddev->sb_wait,
6583 !test_bit(MD_CHANGE_PENDING, &mddev->flags));
6586 void md_write_end(mddev_t *mddev)
6588 if (atomic_dec_and_test(&mddev->writes_pending)) {
6589 if (mddev->safemode == 2)
6590 md_wakeup_thread(mddev->thread);
6591 else if (mddev->safemode_delay)
6592 mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay);
6596 /* md_allow_write(mddev)
6597 * Calling this ensures that the array is marked 'active' so that writes
6598 * may proceed without blocking. It is important to call this before
6599 * attempting a GFP_KERNEL allocation while holding the mddev lock.
6600 * Must be called with mddev_lock held.
6602 * In the ->external case MD_CHANGE_CLEAN can not be cleared until mddev->lock
6603 * is dropped, so return -EAGAIN after notifying userspace.
6605 int md_allow_write(mddev_t *mddev)
6611 if (!mddev->pers->sync_request)
6614 spin_lock_irq(&mddev->write_lock);
6615 if (mddev->in_sync) {
6617 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
6618 set_bit(MD_CHANGE_PENDING, &mddev->flags);
6619 if (mddev->safemode_delay &&
6620 mddev->safemode == 0)
6621 mddev->safemode = 1;
6622 spin_unlock_irq(&mddev->write_lock);
6623 md_update_sb(mddev, 0);
6624 sysfs_notify_dirent_safe(mddev->sysfs_state);
6626 spin_unlock_irq(&mddev->write_lock);
6628 if (test_bit(MD_CHANGE_PENDING, &mddev->flags))
6633 EXPORT_SYMBOL_GPL(md_allow_write);
6635 void md_unplug(mddev_t *mddev)
6638 blk_unplug(mddev->queue);
6640 mddev->plug->unplug_fn(mddev->plug);
6643 #define SYNC_MARKS 10
6644 #define SYNC_MARK_STEP (3*HZ)
6645 void md_do_sync(mddev_t *mddev)
6648 unsigned int currspeed = 0,
6650 sector_t max_sectors,j, io_sectors;
6651 unsigned long mark[SYNC_MARKS];
6652 sector_t mark_cnt[SYNC_MARKS];
6654 struct list_head *tmp;
6655 sector_t last_check;
6660 /* just incase thread restarts... */
6661 if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
6663 if (mddev->ro) /* never try to sync a read-only array */
6666 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
6667 if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
6668 desc = "data-check";
6669 else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
6670 desc = "requested-resync";
6673 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
6678 /* we overload curr_resync somewhat here.
6679 * 0 == not engaged in resync at all
6680 * 2 == checking that there is no conflict with another sync
6681 * 1 == like 2, but have yielded to allow conflicting resync to
6683 * other == active in resync - this many blocks
6685 * Before starting a resync we must have set curr_resync to
6686 * 2, and then checked that every "conflicting" array has curr_resync
6687 * less than ours. When we find one that is the same or higher
6688 * we wait on resync_wait. To avoid deadlock, we reduce curr_resync
6689 * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
6690 * This will mean we have to start checking from the beginning again.
6695 mddev->curr_resync = 2;
6698 if (kthread_should_stop())
6699 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6701 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
6703 for_each_mddev(mddev2, tmp) {
6704 if (mddev2 == mddev)
6706 if (!mddev->parallel_resync
6707 && mddev2->curr_resync
6708 && match_mddev_units(mddev, mddev2)) {
6710 if (mddev < mddev2 && mddev->curr_resync == 2) {
6711 /* arbitrarily yield */
6712 mddev->curr_resync = 1;
6713 wake_up(&resync_wait);
6715 if (mddev > mddev2 && mddev->curr_resync == 1)
6716 /* no need to wait here, we can wait the next
6717 * time 'round when curr_resync == 2
6720 /* We need to wait 'interruptible' so as not to
6721 * contribute to the load average, and not to
6722 * be caught by 'softlockup'
6724 prepare_to_wait(&resync_wait, &wq, TASK_INTERRUPTIBLE);
6725 if (!kthread_should_stop() &&
6726 mddev2->curr_resync >= mddev->curr_resync) {
6727 printk(KERN_INFO "md: delaying %s of %s"
6728 " until %s has finished (they"
6729 " share one or more physical units)\n",
6730 desc, mdname(mddev), mdname(mddev2));
6732 if (signal_pending(current))
6733 flush_signals(current);
6735 finish_wait(&resync_wait, &wq);
6738 finish_wait(&resync_wait, &wq);
6741 } while (mddev->curr_resync < 2);
6744 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
6745 /* resync follows the size requested by the personality,
6746 * which defaults to physical size, but can be virtual size
6748 max_sectors = mddev->resync_max_sectors;
6749 mddev->resync_mismatches = 0;
6750 /* we don't use the checkpoint if there's a bitmap */
6751 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
6752 j = mddev->resync_min;
6753 else if (!mddev->bitmap)
6754 j = mddev->recovery_cp;
6756 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
6757 max_sectors = mddev->dev_sectors;
6759 /* recovery follows the physical size of devices */
6760 max_sectors = mddev->dev_sectors;
6763 list_for_each_entry_rcu(rdev, &mddev->disks, same_set)
6764 if (rdev->raid_disk >= 0 &&
6765 !test_bit(Faulty, &rdev->flags) &&
6766 !test_bit(In_sync, &rdev->flags) &&
6767 rdev->recovery_offset < j)
6768 j = rdev->recovery_offset;
6772 printk(KERN_INFO "md: %s of RAID array %s\n", desc, mdname(mddev));
6773 printk(KERN_INFO "md: minimum _guaranteed_ speed:"
6774 " %d KB/sec/disk.\n", speed_min(mddev));
6775 printk(KERN_INFO "md: using maximum available idle IO bandwidth "
6776 "(but not more than %d KB/sec) for %s.\n",
6777 speed_max(mddev), desc);
6779 is_mddev_idle(mddev, 1); /* this initializes IO event counters */
6782 for (m = 0; m < SYNC_MARKS; m++) {
6784 mark_cnt[m] = io_sectors;
6787 mddev->resync_mark = mark[last_mark];
6788 mddev->resync_mark_cnt = mark_cnt[last_mark];
6791 * Tune reconstruction:
6793 window = 32*(PAGE_SIZE/512);
6794 printk(KERN_INFO "md: using %dk window, over a total of %llu blocks.\n",
6795 window/2,(unsigned long long) max_sectors/2);
6797 atomic_set(&mddev->recovery_active, 0);
6802 "md: resuming %s of %s from checkpoint.\n",
6803 desc, mdname(mddev));
6804 mddev->curr_resync = j;
6806 mddev->curr_resync_completed = mddev->curr_resync;
6808 while (j < max_sectors) {
6813 if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
6814 ((mddev->curr_resync > mddev->curr_resync_completed &&
6815 (mddev->curr_resync - mddev->curr_resync_completed)
6816 > (max_sectors >> 4)) ||
6817 (j - mddev->curr_resync_completed)*2
6818 >= mddev->resync_max - mddev->curr_resync_completed
6820 /* time to update curr_resync_completed */
6822 wait_event(mddev->recovery_wait,
6823 atomic_read(&mddev->recovery_active) == 0);
6824 mddev->curr_resync_completed =
6826 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
6827 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
6830 while (j >= mddev->resync_max && !kthread_should_stop()) {
6831 /* As this condition is controlled by user-space,
6832 * we can block indefinitely, so use '_interruptible'
6833 * to avoid triggering warnings.
6835 flush_signals(current); /* just in case */
6836 wait_event_interruptible(mddev->recovery_wait,
6837 mddev->resync_max > j
6838 || kthread_should_stop());
6841 if (kthread_should_stop())
6844 sectors = mddev->pers->sync_request(mddev, j, &skipped,
6845 currspeed < speed_min(mddev));
6847 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6851 if (!skipped) { /* actual IO requested */
6852 io_sectors += sectors;
6853 atomic_add(sectors, &mddev->recovery_active);
6857 if (j>1) mddev->curr_resync = j;
6858 mddev->curr_mark_cnt = io_sectors;
6859 if (last_check == 0)
6860 /* this is the earliers that rebuilt will be
6861 * visible in /proc/mdstat
6863 md_new_event(mddev);
6865 if (last_check + window > io_sectors || j == max_sectors)
6868 last_check = io_sectors;
6870 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
6874 if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) {
6876 int next = (last_mark+1) % SYNC_MARKS;
6878 mddev->resync_mark = mark[next];
6879 mddev->resync_mark_cnt = mark_cnt[next];
6880 mark[next] = jiffies;
6881 mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active);
6886 if (kthread_should_stop())
6891 * this loop exits only if either when we are slower than
6892 * the 'hard' speed limit, or the system was IO-idle for
6894 * the system might be non-idle CPU-wise, but we only care
6895 * about not overloading the IO subsystem. (things like an
6896 * e2fsck being done on the RAID array should execute fast)
6901 currspeed = ((unsigned long)(io_sectors-mddev->resync_mark_cnt))/2
6902 /((jiffies-mddev->resync_mark)/HZ +1) +1;
6904 if (currspeed > speed_min(mddev)) {
6905 if ((currspeed > speed_max(mddev)) ||
6906 !is_mddev_idle(mddev, 0)) {
6912 printk(KERN_INFO "md: %s: %s done.\n",mdname(mddev), desc);
6914 * this also signals 'finished resyncing' to md_stop
6919 wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
6921 /* tell personality that we are finished */
6922 mddev->pers->sync_request(mddev, max_sectors, &skipped, 1);
6924 if (!test_bit(MD_RECOVERY_CHECK, &mddev->recovery) &&
6925 mddev->curr_resync > 2) {
6926 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
6927 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
6928 if (mddev->curr_resync >= mddev->recovery_cp) {
6930 "md: checkpointing %s of %s.\n",
6931 desc, mdname(mddev));
6932 mddev->recovery_cp = mddev->curr_resync;
6935 mddev->recovery_cp = MaxSector;
6937 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery))
6938 mddev->curr_resync = MaxSector;
6940 list_for_each_entry_rcu(rdev, &mddev->disks, same_set)
6941 if (rdev->raid_disk >= 0 &&
6942 mddev->delta_disks >= 0 &&
6943 !test_bit(Faulty, &rdev->flags) &&
6944 !test_bit(In_sync, &rdev->flags) &&
6945 rdev->recovery_offset < mddev->curr_resync)
6946 rdev->recovery_offset = mddev->curr_resync;
6950 set_bit(MD_CHANGE_DEVS, &mddev->flags);
6953 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
6954 /* We completed so min/max setting can be forgotten if used. */
6955 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
6956 mddev->resync_min = 0;
6957 mddev->resync_max = MaxSector;
6958 } else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
6959 mddev->resync_min = mddev->curr_resync_completed;
6960 mddev->curr_resync = 0;
6961 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery))
6962 mddev->curr_resync_completed = 0;
6963 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
6964 wake_up(&resync_wait);
6965 set_bit(MD_RECOVERY_DONE, &mddev->recovery);
6966 md_wakeup_thread(mddev->thread);
6971 * got a signal, exit.
6974 "md: md_do_sync() got signal ... exiting\n");
6975 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6979 EXPORT_SYMBOL_GPL(md_do_sync);
6982 static int remove_and_add_spares(mddev_t *mddev)
6987 mddev->curr_resync_completed = 0;
6989 list_for_each_entry(rdev, &mddev->disks, same_set)
6990 if (rdev->raid_disk >= 0 &&
6991 !test_bit(Blocked, &rdev->flags) &&
6992 (test_bit(Faulty, &rdev->flags) ||
6993 ! test_bit(In_sync, &rdev->flags)) &&
6994 atomic_read(&rdev->nr_pending)==0) {
6995 if (mddev->pers->hot_remove_disk(
6996 mddev, rdev->raid_disk)==0) {
6998 sprintf(nm,"rd%d", rdev->raid_disk);
6999 sysfs_remove_link(&mddev->kobj, nm);
7000 rdev->raid_disk = -1;
7004 if (mddev->degraded && ! mddev->ro && !mddev->recovery_disabled) {
7005 list_for_each_entry(rdev, &mddev->disks, same_set) {
7006 if (rdev->raid_disk >= 0 &&
7007 !test_bit(In_sync, &rdev->flags) &&
7008 !test_bit(Blocked, &rdev->flags))
7010 if (rdev->raid_disk < 0
7011 && !test_bit(Faulty, &rdev->flags)) {
7012 rdev->recovery_offset = 0;
7014 hot_add_disk(mddev, rdev) == 0) {
7016 sprintf(nm, "rd%d", rdev->raid_disk);
7017 if (sysfs_create_link(&mddev->kobj,
7019 /* failure here is OK */;
7021 md_new_event(mddev);
7022 set_bit(MD_CHANGE_DEVS, &mddev->flags);
7031 * This routine is regularly called by all per-raid-array threads to
7032 * deal with generic issues like resync and super-block update.
7033 * Raid personalities that don't have a thread (linear/raid0) do not
7034 * need this as they never do any recovery or update the superblock.
7036 * It does not do any resync itself, but rather "forks" off other threads
7037 * to do that as needed.
7038 * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
7039 * "->recovery" and create a thread at ->sync_thread.
7040 * When the thread finishes it sets MD_RECOVERY_DONE
7041 * and wakeups up this thread which will reap the thread and finish up.
7042 * This thread also removes any faulty devices (with nr_pending == 0).
7044 * The overall approach is:
7045 * 1/ if the superblock needs updating, update it.
7046 * 2/ If a recovery thread is running, don't do anything else.
7047 * 3/ If recovery has finished, clean up, possibly marking spares active.
7048 * 4/ If there are any faulty devices, remove them.
7049 * 5/ If array is degraded, try to add spares devices
7050 * 6/ If array has spares or is not in-sync, start a resync thread.
7052 void md_check_recovery(mddev_t *mddev)
7058 bitmap_daemon_work(mddev);
7063 if (signal_pending(current)) {
7064 if (mddev->pers->sync_request && !mddev->external) {
7065 printk(KERN_INFO "md: %s in immediate safe mode\n",
7067 mddev->safemode = 2;
7069 flush_signals(current);
7072 if (mddev->ro && !test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
7075 (mddev->flags & ~ (1<<MD_CHANGE_PENDING)) ||
7076 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
7077 test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
7078 (mddev->external == 0 && mddev->safemode == 1) ||
7079 (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending)
7080 && !mddev->in_sync && mddev->recovery_cp == MaxSector)
7084 if (mddev_trylock(mddev)) {
7088 /* Only thing we do on a ro array is remove
7091 remove_and_add_spares(mddev);
7092 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7096 if (!mddev->external) {
7098 spin_lock_irq(&mddev->write_lock);
7099 if (mddev->safemode &&
7100 !atomic_read(&mddev->writes_pending) &&
7102 mddev->recovery_cp == MaxSector) {
7105 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
7107 if (mddev->safemode == 1)
7108 mddev->safemode = 0;
7109 spin_unlock_irq(&mddev->write_lock);
7111 sysfs_notify_dirent_safe(mddev->sysfs_state);
7115 md_update_sb(mddev, 0);
7117 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
7118 !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
7119 /* resync/recovery still happening */
7120 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7123 if (mddev->sync_thread) {
7124 /* resync has finished, collect result */
7125 md_unregister_thread(mddev->sync_thread);
7126 mddev->sync_thread = NULL;
7127 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
7128 !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
7130 /* activate any spares */
7131 if (mddev->pers->spare_active(mddev))
7132 sysfs_notify(&mddev->kobj, NULL,
7135 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
7136 mddev->pers->finish_reshape)
7137 mddev->pers->finish_reshape(mddev);
7138 md_update_sb(mddev, 1);
7140 /* if array is no-longer degraded, then any saved_raid_disk
7141 * information must be scrapped
7143 if (!mddev->degraded)
7144 list_for_each_entry(rdev, &mddev->disks, same_set)
7145 rdev->saved_raid_disk = -1;
7147 mddev->recovery = 0;
7148 /* flag recovery needed just to double check */
7149 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7150 sysfs_notify_dirent_safe(mddev->sysfs_action);
7151 md_new_event(mddev);
7154 /* Set RUNNING before clearing NEEDED to avoid
7155 * any transients in the value of "sync_action".
7157 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
7158 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7159 /* Clear some bits that don't mean anything, but
7162 clear_bit(MD_RECOVERY_INTR, &mddev->recovery);
7163 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
7165 if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
7167 /* no recovery is running.
7168 * remove any failed drives, then
7169 * add spares if possible.
7170 * Spare are also removed and re-added, to allow
7171 * the personality to fail the re-add.
7174 if (mddev->reshape_position != MaxSector) {
7175 if (mddev->pers->check_reshape == NULL ||
7176 mddev->pers->check_reshape(mddev) != 0)
7177 /* Cannot proceed */
7179 set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
7180 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
7181 } else if ((spares = remove_and_add_spares(mddev))) {
7182 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
7183 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
7184 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
7185 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
7186 } else if (mddev->recovery_cp < MaxSector) {
7187 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
7188 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
7189 } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
7190 /* nothing to be done ... */
7193 if (mddev->pers->sync_request) {
7194 if (spares && mddev->bitmap && ! mddev->bitmap->file) {
7195 /* We are adding a device or devices to an array
7196 * which has the bitmap stored on all devices.
7197 * So make sure all bitmap pages get written
7199 bitmap_write_all(mddev->bitmap);
7201 mddev->sync_thread = md_register_thread(md_do_sync,
7204 if (!mddev->sync_thread) {
7205 printk(KERN_ERR "%s: could not start resync"
7208 /* leave the spares where they are, it shouldn't hurt */
7209 mddev->recovery = 0;
7211 md_wakeup_thread(mddev->sync_thread);
7212 sysfs_notify_dirent_safe(mddev->sysfs_action);
7213 md_new_event(mddev);
7216 if (!mddev->sync_thread) {
7217 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
7218 if (test_and_clear_bit(MD_RECOVERY_RECOVER,
7220 if (mddev->sysfs_action)
7221 sysfs_notify_dirent_safe(mddev->sysfs_action);
7223 mddev_unlock(mddev);
7227 void md_wait_for_blocked_rdev(mdk_rdev_t *rdev, mddev_t *mddev)
7229 sysfs_notify_dirent_safe(rdev->sysfs_state);
7230 wait_event_timeout(rdev->blocked_wait,
7231 !test_bit(Blocked, &rdev->flags),
7232 msecs_to_jiffies(5000));
7233 rdev_dec_pending(rdev, mddev);
7235 EXPORT_SYMBOL(md_wait_for_blocked_rdev);
7237 static int md_notify_reboot(struct notifier_block *this,
7238 unsigned long code, void *x)
7240 struct list_head *tmp;
7243 if ((code == SYS_DOWN) || (code == SYS_HALT) || (code == SYS_POWER_OFF)) {
7245 printk(KERN_INFO "md: stopping all md devices.\n");
7247 for_each_mddev(mddev, tmp)
7248 if (mddev_trylock(mddev)) {
7249 /* Force a switch to readonly even array
7250 * appears to still be in use. Hence
7253 md_set_readonly(mddev, 100);
7254 mddev_unlock(mddev);
7257 * certain more exotic SCSI devices are known to be
7258 * volatile wrt too early system reboots. While the
7259 * right place to handle this issue is the given
7260 * driver, we do want to have a safe RAID driver ...
7267 static struct notifier_block md_notifier = {
7268 .notifier_call = md_notify_reboot,
7270 .priority = INT_MAX, /* before any real devices */
7273 static void md_geninit(void)
7275 dprintk("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
7277 proc_create("mdstat", S_IRUGO, NULL, &md_seq_fops);
7280 static int __init md_init(void)
7282 if (register_blkdev(MD_MAJOR, "md"))
7284 if ((mdp_major=register_blkdev(0, "mdp"))<=0) {
7285 unregister_blkdev(MD_MAJOR, "md");
7288 blk_register_region(MKDEV(MD_MAJOR, 0), 1UL<<MINORBITS, THIS_MODULE,
7289 md_probe, NULL, NULL);
7290 blk_register_region(MKDEV(mdp_major, 0), 1UL<<MINORBITS, THIS_MODULE,
7291 md_probe, NULL, NULL);
7293 register_reboot_notifier(&md_notifier);
7294 raid_table_header = register_sysctl_table(raid_root_table);
7304 * Searches all registered partitions for autorun RAID arrays
7308 static LIST_HEAD(all_detected_devices);
7309 struct detected_devices_node {
7310 struct list_head list;
7314 void md_autodetect_dev(dev_t dev)
7316 struct detected_devices_node *node_detected_dev;
7318 node_detected_dev = kzalloc(sizeof(*node_detected_dev), GFP_KERNEL);
7319 if (node_detected_dev) {
7320 node_detected_dev->dev = dev;
7321 list_add_tail(&node_detected_dev->list, &all_detected_devices);
7323 printk(KERN_CRIT "md: md_autodetect_dev: kzalloc failed"
7324 ", skipping dev(%d,%d)\n", MAJOR(dev), MINOR(dev));
7329 static void autostart_arrays(int part)
7332 struct detected_devices_node *node_detected_dev;
7334 int i_scanned, i_passed;
7339 printk(KERN_INFO "md: Autodetecting RAID arrays.\n");
7341 while (!list_empty(&all_detected_devices) && i_scanned < INT_MAX) {
7343 node_detected_dev = list_entry(all_detected_devices.next,
7344 struct detected_devices_node, list);
7345 list_del(&node_detected_dev->list);
7346 dev = node_detected_dev->dev;
7347 kfree(node_detected_dev);
7348 rdev = md_import_device(dev,0, 90);
7352 if (test_bit(Faulty, &rdev->flags)) {
7356 set_bit(AutoDetected, &rdev->flags);
7357 list_add(&rdev->same_set, &pending_raid_disks);
7361 printk(KERN_INFO "md: Scanned %d and added %d devices.\n",
7362 i_scanned, i_passed);
7364 autorun_devices(part);
7367 #endif /* !MODULE */
7369 static __exit void md_exit(void)
7372 struct list_head *tmp;
7374 blk_unregister_region(MKDEV(MD_MAJOR,0), 1U << MINORBITS);
7375 blk_unregister_region(MKDEV(mdp_major,0), 1U << MINORBITS);
7377 unregister_blkdev(MD_MAJOR,"md");
7378 unregister_blkdev(mdp_major, "mdp");
7379 unregister_reboot_notifier(&md_notifier);
7380 unregister_sysctl_table(raid_table_header);
7381 remove_proc_entry("mdstat", NULL);
7382 for_each_mddev(mddev, tmp) {
7383 export_array(mddev);
7384 mddev->hold_active = 0;
7388 subsys_initcall(md_init);
7389 module_exit(md_exit)
7391 static int get_ro(char *buffer, struct kernel_param *kp)
7393 return sprintf(buffer, "%d", start_readonly);
7395 static int set_ro(const char *val, struct kernel_param *kp)
7398 int num = simple_strtoul(val, &e, 10);
7399 if (*val && (*e == '\0' || *e == '\n')) {
7400 start_readonly = num;
7406 module_param_call(start_ro, set_ro, get_ro, NULL, S_IRUSR|S_IWUSR);
7407 module_param(start_dirty_degraded, int, S_IRUGO|S_IWUSR);
7409 module_param_call(new_array, add_named_array, NULL, NULL, S_IWUSR);
7411 EXPORT_SYMBOL(register_md_personality);
7412 EXPORT_SYMBOL(unregister_md_personality);
7413 EXPORT_SYMBOL(md_error);
7414 EXPORT_SYMBOL(md_done_sync);
7415 EXPORT_SYMBOL(md_write_start);
7416 EXPORT_SYMBOL(md_write_end);
7417 EXPORT_SYMBOL(md_register_thread);
7418 EXPORT_SYMBOL(md_unregister_thread);
7419 EXPORT_SYMBOL(md_wakeup_thread);
7420 EXPORT_SYMBOL(md_check_recovery);
7421 MODULE_LICENSE("GPL");
7422 MODULE_DESCRIPTION("MD RAID framework");
7424 MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR);