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/buffer_head.h> /* for invalidate_bdev */
40 #include <linux/poll.h>
41 #include <linux/ctype.h>
42 #include <linux/hdreg.h>
43 #include <linux/proc_fs.h>
44 #include <linux/random.h>
45 #include <linux/reboot.h>
46 #include <linux/file.h>
47 #include <linux/compat.h>
48 #include <linux/delay.h>
49 #include <linux/raid/md_p.h>
50 #include <linux/raid/md_u.h>
55 #define dprintk(x...) ((void)(DEBUG && printk(x)))
59 static void autostart_arrays(int part);
62 static LIST_HEAD(pers_list);
63 static DEFINE_SPINLOCK(pers_lock);
65 static void md_print_devices(void);
67 static DECLARE_WAIT_QUEUE_HEAD(resync_wait);
69 #define MD_BUG(x...) { printk("md: bug in file %s, line %d\n", __FILE__, __LINE__); md_print_devices(); }
72 * Default number of read corrections we'll attempt on an rdev
73 * before ejecting it from the array. We divide the read error
74 * count by 2 for every hour elapsed between read errors.
76 #define MD_DEFAULT_MAX_CORRECTED_READ_ERRORS 20
78 * Current RAID-1,4,5 parallel reconstruction 'guaranteed speed limit'
79 * is 1000 KB/sec, so the extra system load does not show up that much.
80 * Increase it if you want to have more _guaranteed_ speed. Note that
81 * the RAID driver will use the maximum available bandwidth if the IO
82 * subsystem is idle. There is also an 'absolute maximum' reconstruction
83 * speed limit - in case reconstruction slows down your system despite
86 * you can change it via /proc/sys/dev/raid/speed_limit_min and _max.
87 * or /sys/block/mdX/md/sync_speed_{min,max}
90 static int sysctl_speed_limit_min = 1000;
91 static int sysctl_speed_limit_max = 200000;
92 static inline int speed_min(mddev_t *mddev)
94 return mddev->sync_speed_min ?
95 mddev->sync_speed_min : sysctl_speed_limit_min;
98 static inline int speed_max(mddev_t *mddev)
100 return mddev->sync_speed_max ?
101 mddev->sync_speed_max : sysctl_speed_limit_max;
104 static struct ctl_table_header *raid_table_header;
106 static ctl_table raid_table[] = {
108 .procname = "speed_limit_min",
109 .data = &sysctl_speed_limit_min,
110 .maxlen = sizeof(int),
111 .mode = S_IRUGO|S_IWUSR,
112 .proc_handler = proc_dointvec,
115 .procname = "speed_limit_max",
116 .data = &sysctl_speed_limit_max,
117 .maxlen = sizeof(int),
118 .mode = S_IRUGO|S_IWUSR,
119 .proc_handler = proc_dointvec,
124 static ctl_table raid_dir_table[] = {
128 .mode = S_IRUGO|S_IXUGO,
134 static ctl_table raid_root_table[] = {
139 .child = raid_dir_table,
144 static const struct block_device_operations md_fops;
146 static int start_readonly;
149 * We have a system wide 'event count' that is incremented
150 * on any 'interesting' event, and readers of /proc/mdstat
151 * can use 'poll' or 'select' to find out when the event
155 * start array, stop array, error, add device, remove device,
156 * start build, activate spare
158 static DECLARE_WAIT_QUEUE_HEAD(md_event_waiters);
159 static atomic_t md_event_count;
160 void md_new_event(mddev_t *mddev)
162 atomic_inc(&md_event_count);
163 wake_up(&md_event_waiters);
165 EXPORT_SYMBOL_GPL(md_new_event);
167 /* Alternate version that can be called from interrupts
168 * when calling sysfs_notify isn't needed.
170 static void md_new_event_inintr(mddev_t *mddev)
172 atomic_inc(&md_event_count);
173 wake_up(&md_event_waiters);
177 * Enables to iterate over all existing md arrays
178 * all_mddevs_lock protects this list.
180 static LIST_HEAD(all_mddevs);
181 static DEFINE_SPINLOCK(all_mddevs_lock);
185 * iterates through all used mddevs in the system.
186 * We take care to grab the all_mddevs_lock whenever navigating
187 * the list, and to always hold a refcount when unlocked.
188 * Any code which breaks out of this loop while own
189 * a reference to the current mddev and must mddev_put it.
191 #define for_each_mddev(mddev,tmp) \
193 for (({ spin_lock(&all_mddevs_lock); \
194 tmp = all_mddevs.next; \
196 ({ if (tmp != &all_mddevs) \
197 mddev_get(list_entry(tmp, mddev_t, all_mddevs));\
198 spin_unlock(&all_mddevs_lock); \
199 if (mddev) mddev_put(mddev); \
200 mddev = list_entry(tmp, mddev_t, all_mddevs); \
201 tmp != &all_mddevs;}); \
202 ({ spin_lock(&all_mddevs_lock); \
207 /* Rather than calling directly into the personality make_request function,
208 * IO requests come here first so that we can check if the device is
209 * being suspended pending a reconfiguration.
210 * We hold a refcount over the call to ->make_request. By the time that
211 * call has finished, the bio has been linked into some internal structure
212 * and so is visible to ->quiesce(), so we don't need the refcount any more.
214 static int md_make_request(struct request_queue *q, struct bio *bio)
216 mddev_t *mddev = q->queuedata;
218 if (mddev == NULL || mddev->pers == NULL) {
223 if (mddev->suspended || mddev->barrier) {
226 prepare_to_wait(&mddev->sb_wait, &__wait,
227 TASK_UNINTERRUPTIBLE);
228 if (!mddev->suspended && !mddev->barrier)
234 finish_wait(&mddev->sb_wait, &__wait);
236 atomic_inc(&mddev->active_io);
238 rv = mddev->pers->make_request(q, bio);
239 if (atomic_dec_and_test(&mddev->active_io) && mddev->suspended)
240 wake_up(&mddev->sb_wait);
245 static void mddev_suspend(mddev_t *mddev)
247 BUG_ON(mddev->suspended);
248 mddev->suspended = 1;
250 wait_event(mddev->sb_wait, atomic_read(&mddev->active_io) == 0);
251 mddev->pers->quiesce(mddev, 1);
252 md_unregister_thread(mddev->thread);
253 mddev->thread = NULL;
254 /* we now know that no code is executing in the personality module,
255 * except possibly the tail end of a ->bi_end_io function, but that
256 * is certain to complete before the module has a chance to get
261 static void mddev_resume(mddev_t *mddev)
263 mddev->suspended = 0;
264 wake_up(&mddev->sb_wait);
265 mddev->pers->quiesce(mddev, 0);
268 int mddev_congested(mddev_t *mddev, int bits)
272 return mddev->suspended;
274 EXPORT_SYMBOL(mddev_congested);
277 * Generic barrier handling for md
280 #define POST_REQUEST_BARRIER ((void*)1)
282 static void md_end_barrier(struct bio *bio, int err)
284 mdk_rdev_t *rdev = bio->bi_private;
285 mddev_t *mddev = rdev->mddev;
286 if (err == -EOPNOTSUPP && mddev->barrier != POST_REQUEST_BARRIER)
287 set_bit(BIO_EOPNOTSUPP, &mddev->barrier->bi_flags);
289 rdev_dec_pending(rdev, mddev);
291 if (atomic_dec_and_test(&mddev->flush_pending)) {
292 if (mddev->barrier == POST_REQUEST_BARRIER) {
293 /* This was a post-request barrier */
294 mddev->barrier = NULL;
295 wake_up(&mddev->sb_wait);
297 /* The pre-request barrier has finished */
298 schedule_work(&mddev->barrier_work);
303 static void submit_barriers(mddev_t *mddev)
308 list_for_each_entry_rcu(rdev, &mddev->disks, same_set)
309 if (rdev->raid_disk >= 0 &&
310 !test_bit(Faulty, &rdev->flags)) {
311 /* Take two references, one is dropped
312 * when request finishes, one after
313 * we reclaim rcu_read_lock
316 atomic_inc(&rdev->nr_pending);
317 atomic_inc(&rdev->nr_pending);
319 bi = bio_alloc(GFP_KERNEL, 0);
320 bi->bi_end_io = md_end_barrier;
321 bi->bi_private = rdev;
322 bi->bi_bdev = rdev->bdev;
323 atomic_inc(&mddev->flush_pending);
324 submit_bio(WRITE_BARRIER, bi);
326 rdev_dec_pending(rdev, mddev);
331 static void md_submit_barrier(struct work_struct *ws)
333 mddev_t *mddev = container_of(ws, mddev_t, barrier_work);
334 struct bio *bio = mddev->barrier;
336 atomic_set(&mddev->flush_pending, 1);
338 if (test_bit(BIO_EOPNOTSUPP, &bio->bi_flags))
339 bio_endio(bio, -EOPNOTSUPP);
340 else if (bio->bi_size == 0)
341 /* an empty barrier - all done */
344 bio->bi_rw &= ~(1<<BIO_RW_BARRIER);
345 if (mddev->pers->make_request(mddev->queue, bio))
346 generic_make_request(bio);
347 mddev->barrier = POST_REQUEST_BARRIER;
348 submit_barriers(mddev);
350 if (atomic_dec_and_test(&mddev->flush_pending)) {
351 mddev->barrier = NULL;
352 wake_up(&mddev->sb_wait);
356 void md_barrier_request(mddev_t *mddev, struct bio *bio)
358 spin_lock_irq(&mddev->write_lock);
359 wait_event_lock_irq(mddev->sb_wait,
361 mddev->write_lock, /*nothing*/);
362 mddev->barrier = bio;
363 spin_unlock_irq(&mddev->write_lock);
365 atomic_set(&mddev->flush_pending, 1);
366 INIT_WORK(&mddev->barrier_work, md_submit_barrier);
368 submit_barriers(mddev);
370 if (atomic_dec_and_test(&mddev->flush_pending))
371 schedule_work(&mddev->barrier_work);
373 EXPORT_SYMBOL(md_barrier_request);
375 static inline mddev_t *mddev_get(mddev_t *mddev)
377 atomic_inc(&mddev->active);
381 static void mddev_delayed_delete(struct work_struct *ws);
383 static void mddev_put(mddev_t *mddev)
385 if (!atomic_dec_and_lock(&mddev->active, &all_mddevs_lock))
387 if (!mddev->raid_disks && list_empty(&mddev->disks) &&
388 !mddev->hold_active) {
389 list_del(&mddev->all_mddevs);
390 if (mddev->gendisk) {
391 /* we did a probe so need to clean up.
392 * Call schedule_work inside the spinlock
393 * so that flush_scheduled_work() after
394 * mddev_find will succeed in waiting for the
397 INIT_WORK(&mddev->del_work, mddev_delayed_delete);
398 schedule_work(&mddev->del_work);
402 spin_unlock(&all_mddevs_lock);
405 static mddev_t * mddev_find(dev_t unit)
407 mddev_t *mddev, *new = NULL;
410 spin_lock(&all_mddevs_lock);
413 list_for_each_entry(mddev, &all_mddevs, all_mddevs)
414 if (mddev->unit == unit) {
416 spin_unlock(&all_mddevs_lock);
422 list_add(&new->all_mddevs, &all_mddevs);
423 spin_unlock(&all_mddevs_lock);
424 new->hold_active = UNTIL_IOCTL;
428 /* find an unused unit number */
429 static int next_minor = 512;
430 int start = next_minor;
434 dev = MKDEV(MD_MAJOR, next_minor);
436 if (next_minor > MINORMASK)
438 if (next_minor == start) {
439 /* Oh dear, all in use. */
440 spin_unlock(&all_mddevs_lock);
446 list_for_each_entry(mddev, &all_mddevs, all_mddevs)
447 if (mddev->unit == dev) {
453 new->md_minor = MINOR(dev);
454 new->hold_active = UNTIL_STOP;
455 list_add(&new->all_mddevs, &all_mddevs);
456 spin_unlock(&all_mddevs_lock);
459 spin_unlock(&all_mddevs_lock);
461 new = kzalloc(sizeof(*new), GFP_KERNEL);
466 if (MAJOR(unit) == MD_MAJOR)
467 new->md_minor = MINOR(unit);
469 new->md_minor = MINOR(unit) >> MdpMinorShift;
471 mutex_init(&new->open_mutex);
472 mutex_init(&new->reconfig_mutex);
473 mutex_init(&new->bitmap_info.mutex);
474 INIT_LIST_HEAD(&new->disks);
475 INIT_LIST_HEAD(&new->all_mddevs);
476 init_timer(&new->safemode_timer);
477 atomic_set(&new->active, 1);
478 atomic_set(&new->openers, 0);
479 atomic_set(&new->active_io, 0);
480 spin_lock_init(&new->write_lock);
481 atomic_set(&new->flush_pending, 0);
482 init_waitqueue_head(&new->sb_wait);
483 init_waitqueue_head(&new->recovery_wait);
484 new->reshape_position = MaxSector;
486 new->resync_max = MaxSector;
487 new->level = LEVEL_NONE;
492 static inline int mddev_lock(mddev_t * mddev)
494 return mutex_lock_interruptible(&mddev->reconfig_mutex);
497 static inline int mddev_is_locked(mddev_t *mddev)
499 return mutex_is_locked(&mddev->reconfig_mutex);
502 static inline int mddev_trylock(mddev_t * mddev)
504 return mutex_trylock(&mddev->reconfig_mutex);
507 static inline void mddev_unlock(mddev_t * mddev)
509 mutex_unlock(&mddev->reconfig_mutex);
511 md_wakeup_thread(mddev->thread);
514 static mdk_rdev_t * find_rdev_nr(mddev_t *mddev, int nr)
518 list_for_each_entry(rdev, &mddev->disks, same_set)
519 if (rdev->desc_nr == nr)
525 static mdk_rdev_t * find_rdev(mddev_t * mddev, dev_t dev)
529 list_for_each_entry(rdev, &mddev->disks, same_set)
530 if (rdev->bdev->bd_dev == dev)
536 static struct mdk_personality *find_pers(int level, char *clevel)
538 struct mdk_personality *pers;
539 list_for_each_entry(pers, &pers_list, list) {
540 if (level != LEVEL_NONE && pers->level == level)
542 if (strcmp(pers->name, clevel)==0)
548 /* return the offset of the super block in 512byte sectors */
549 static inline sector_t calc_dev_sboffset(struct block_device *bdev)
551 sector_t num_sectors = bdev->bd_inode->i_size / 512;
552 return MD_NEW_SIZE_SECTORS(num_sectors);
555 static int alloc_disk_sb(mdk_rdev_t * rdev)
560 rdev->sb_page = alloc_page(GFP_KERNEL);
561 if (!rdev->sb_page) {
562 printk(KERN_ALERT "md: out of memory.\n");
569 static void free_disk_sb(mdk_rdev_t * rdev)
572 put_page(rdev->sb_page);
574 rdev->sb_page = NULL;
581 static void super_written(struct bio *bio, int error)
583 mdk_rdev_t *rdev = bio->bi_private;
584 mddev_t *mddev = rdev->mddev;
586 if (error || !test_bit(BIO_UPTODATE, &bio->bi_flags)) {
587 printk("md: super_written gets error=%d, uptodate=%d\n",
588 error, test_bit(BIO_UPTODATE, &bio->bi_flags));
589 WARN_ON(test_bit(BIO_UPTODATE, &bio->bi_flags));
590 md_error(mddev, rdev);
593 if (atomic_dec_and_test(&mddev->pending_writes))
594 wake_up(&mddev->sb_wait);
598 static void super_written_barrier(struct bio *bio, int error)
600 struct bio *bio2 = bio->bi_private;
601 mdk_rdev_t *rdev = bio2->bi_private;
602 mddev_t *mddev = rdev->mddev;
604 if (!test_bit(BIO_UPTODATE, &bio->bi_flags) &&
605 error == -EOPNOTSUPP) {
607 /* barriers don't appear to be supported :-( */
608 set_bit(BarriersNotsupp, &rdev->flags);
609 mddev->barriers_work = 0;
610 spin_lock_irqsave(&mddev->write_lock, flags);
611 bio2->bi_next = mddev->biolist;
612 mddev->biolist = bio2;
613 spin_unlock_irqrestore(&mddev->write_lock, flags);
614 wake_up(&mddev->sb_wait);
618 bio->bi_private = rdev;
619 super_written(bio, error);
623 void md_super_write(mddev_t *mddev, mdk_rdev_t *rdev,
624 sector_t sector, int size, struct page *page)
626 /* write first size bytes of page to sector of rdev
627 * Increment mddev->pending_writes before returning
628 * and decrement it on completion, waking up sb_wait
629 * if zero is reached.
630 * If an error occurred, call md_error
632 * As we might need to resubmit the request if BIO_RW_BARRIER
633 * causes ENOTSUPP, we allocate a spare bio...
635 struct bio *bio = bio_alloc(GFP_NOIO, 1);
636 int rw = (1<<BIO_RW) | (1<<BIO_RW_SYNCIO) | (1<<BIO_RW_UNPLUG);
638 bio->bi_bdev = rdev->bdev;
639 bio->bi_sector = sector;
640 bio_add_page(bio, page, size, 0);
641 bio->bi_private = rdev;
642 bio->bi_end_io = super_written;
645 atomic_inc(&mddev->pending_writes);
646 if (!test_bit(BarriersNotsupp, &rdev->flags)) {
648 rw |= (1<<BIO_RW_BARRIER);
649 rbio = bio_clone(bio, GFP_NOIO);
650 rbio->bi_private = bio;
651 rbio->bi_end_io = super_written_barrier;
652 submit_bio(rw, rbio);
657 void md_super_wait(mddev_t *mddev)
659 /* wait for all superblock writes that were scheduled to complete.
660 * if any had to be retried (due to BARRIER problems), retry them
664 prepare_to_wait(&mddev->sb_wait, &wq, TASK_UNINTERRUPTIBLE);
665 if (atomic_read(&mddev->pending_writes)==0)
667 while (mddev->biolist) {
669 spin_lock_irq(&mddev->write_lock);
670 bio = mddev->biolist;
671 mddev->biolist = bio->bi_next ;
673 spin_unlock_irq(&mddev->write_lock);
674 submit_bio(bio->bi_rw, bio);
678 finish_wait(&mddev->sb_wait, &wq);
681 static void bi_complete(struct bio *bio, int error)
683 complete((struct completion*)bio->bi_private);
686 int sync_page_io(struct block_device *bdev, sector_t sector, int size,
687 struct page *page, int rw)
689 struct bio *bio = bio_alloc(GFP_NOIO, 1);
690 struct completion event;
693 rw |= (1 << BIO_RW_SYNCIO) | (1 << BIO_RW_UNPLUG);
696 bio->bi_sector = sector;
697 bio_add_page(bio, page, size, 0);
698 init_completion(&event);
699 bio->bi_private = &event;
700 bio->bi_end_io = bi_complete;
702 wait_for_completion(&event);
704 ret = test_bit(BIO_UPTODATE, &bio->bi_flags);
708 EXPORT_SYMBOL_GPL(sync_page_io);
710 static int read_disk_sb(mdk_rdev_t * rdev, int size)
712 char b[BDEVNAME_SIZE];
713 if (!rdev->sb_page) {
721 if (!sync_page_io(rdev->bdev, rdev->sb_start, size, rdev->sb_page, READ))
727 printk(KERN_WARNING "md: disabled device %s, could not read superblock.\n",
728 bdevname(rdev->bdev,b));
732 static int uuid_equal(mdp_super_t *sb1, mdp_super_t *sb2)
734 return sb1->set_uuid0 == sb2->set_uuid0 &&
735 sb1->set_uuid1 == sb2->set_uuid1 &&
736 sb1->set_uuid2 == sb2->set_uuid2 &&
737 sb1->set_uuid3 == sb2->set_uuid3;
740 static int sb_equal(mdp_super_t *sb1, mdp_super_t *sb2)
743 mdp_super_t *tmp1, *tmp2;
745 tmp1 = kmalloc(sizeof(*tmp1),GFP_KERNEL);
746 tmp2 = kmalloc(sizeof(*tmp2),GFP_KERNEL);
748 if (!tmp1 || !tmp2) {
750 printk(KERN_INFO "md.c sb_equal(): failed to allocate memory!\n");
758 * nr_disks is not constant
763 ret = (memcmp(tmp1, tmp2, MD_SB_GENERIC_CONSTANT_WORDS * 4) == 0);
771 static u32 md_csum_fold(u32 csum)
773 csum = (csum & 0xffff) + (csum >> 16);
774 return (csum & 0xffff) + (csum >> 16);
777 static unsigned int calc_sb_csum(mdp_super_t * sb)
780 u32 *sb32 = (u32*)sb;
782 unsigned int disk_csum, csum;
784 disk_csum = sb->sb_csum;
787 for (i = 0; i < MD_SB_BYTES/4 ; i++)
789 csum = (newcsum & 0xffffffff) + (newcsum>>32);
793 /* This used to use csum_partial, which was wrong for several
794 * reasons including that different results are returned on
795 * different architectures. It isn't critical that we get exactly
796 * the same return value as before (we always csum_fold before
797 * testing, and that removes any differences). However as we
798 * know that csum_partial always returned a 16bit value on
799 * alphas, do a fold to maximise conformity to previous behaviour.
801 sb->sb_csum = md_csum_fold(disk_csum);
803 sb->sb_csum = disk_csum;
810 * Handle superblock details.
811 * We want to be able to handle multiple superblock formats
812 * so we have a common interface to them all, and an array of
813 * different handlers.
814 * We rely on user-space to write the initial superblock, and support
815 * reading and updating of superblocks.
816 * Interface methods are:
817 * int load_super(mdk_rdev_t *dev, mdk_rdev_t *refdev, int minor_version)
818 * loads and validates a superblock on dev.
819 * if refdev != NULL, compare superblocks on both devices
821 * 0 - dev has a superblock that is compatible with refdev
822 * 1 - dev has a superblock that is compatible and newer than refdev
823 * so dev should be used as the refdev in future
824 * -EINVAL superblock incompatible or invalid
825 * -othererror e.g. -EIO
827 * int validate_super(mddev_t *mddev, mdk_rdev_t *dev)
828 * Verify that dev is acceptable into mddev.
829 * The first time, mddev->raid_disks will be 0, and data from
830 * dev should be merged in. Subsequent calls check that dev
831 * is new enough. Return 0 or -EINVAL
833 * void sync_super(mddev_t *mddev, mdk_rdev_t *dev)
834 * Update the superblock for rdev with data in mddev
835 * This does not write to disc.
841 struct module *owner;
842 int (*load_super)(mdk_rdev_t *rdev, mdk_rdev_t *refdev,
844 int (*validate_super)(mddev_t *mddev, mdk_rdev_t *rdev);
845 void (*sync_super)(mddev_t *mddev, mdk_rdev_t *rdev);
846 unsigned long long (*rdev_size_change)(mdk_rdev_t *rdev,
847 sector_t num_sectors);
851 * Check that the given mddev has no bitmap.
853 * This function is called from the run method of all personalities that do not
854 * support bitmaps. It prints an error message and returns non-zero if mddev
855 * has a bitmap. Otherwise, it returns 0.
858 int md_check_no_bitmap(mddev_t *mddev)
860 if (!mddev->bitmap_info.file && !mddev->bitmap_info.offset)
862 printk(KERN_ERR "%s: bitmaps are not supported for %s\n",
863 mdname(mddev), mddev->pers->name);
866 EXPORT_SYMBOL(md_check_no_bitmap);
869 * load_super for 0.90.0
871 static int super_90_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
873 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
878 * Calculate the position of the superblock (512byte sectors),
879 * it's at the end of the disk.
881 * It also happens to be a multiple of 4Kb.
883 rdev->sb_start = calc_dev_sboffset(rdev->bdev);
885 ret = read_disk_sb(rdev, MD_SB_BYTES);
890 bdevname(rdev->bdev, b);
891 sb = (mdp_super_t*)page_address(rdev->sb_page);
893 if (sb->md_magic != MD_SB_MAGIC) {
894 printk(KERN_ERR "md: invalid raid superblock magic on %s\n",
899 if (sb->major_version != 0 ||
900 sb->minor_version < 90 ||
901 sb->minor_version > 91) {
902 printk(KERN_WARNING "Bad version number %d.%d on %s\n",
903 sb->major_version, sb->minor_version,
908 if (sb->raid_disks <= 0)
911 if (md_csum_fold(calc_sb_csum(sb)) != md_csum_fold(sb->sb_csum)) {
912 printk(KERN_WARNING "md: invalid superblock checksum on %s\n",
917 rdev->preferred_minor = sb->md_minor;
918 rdev->data_offset = 0;
919 rdev->sb_size = MD_SB_BYTES;
921 if (sb->level == LEVEL_MULTIPATH)
924 rdev->desc_nr = sb->this_disk.number;
930 mdp_super_t *refsb = (mdp_super_t*)page_address(refdev->sb_page);
931 if (!uuid_equal(refsb, sb)) {
932 printk(KERN_WARNING "md: %s has different UUID to %s\n",
933 b, bdevname(refdev->bdev,b2));
936 if (!sb_equal(refsb, sb)) {
937 printk(KERN_WARNING "md: %s has same UUID"
938 " but different superblock to %s\n",
939 b, bdevname(refdev->bdev, b2));
943 ev2 = md_event(refsb);
949 rdev->sectors = rdev->sb_start;
951 if (rdev->sectors < sb->size * 2 && sb->level > 1)
952 /* "this cannot possibly happen" ... */
960 * validate_super for 0.90.0
962 static int super_90_validate(mddev_t *mddev, mdk_rdev_t *rdev)
965 mdp_super_t *sb = (mdp_super_t *)page_address(rdev->sb_page);
966 __u64 ev1 = md_event(sb);
968 rdev->raid_disk = -1;
969 clear_bit(Faulty, &rdev->flags);
970 clear_bit(In_sync, &rdev->flags);
971 clear_bit(WriteMostly, &rdev->flags);
972 clear_bit(BarriersNotsupp, &rdev->flags);
974 if (mddev->raid_disks == 0) {
975 mddev->major_version = 0;
976 mddev->minor_version = sb->minor_version;
977 mddev->patch_version = sb->patch_version;
979 mddev->chunk_sectors = sb->chunk_size >> 9;
980 mddev->ctime = sb->ctime;
981 mddev->utime = sb->utime;
982 mddev->level = sb->level;
983 mddev->clevel[0] = 0;
984 mddev->layout = sb->layout;
985 mddev->raid_disks = sb->raid_disks;
986 mddev->dev_sectors = sb->size * 2;
988 mddev->bitmap_info.offset = 0;
989 mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9;
991 if (mddev->minor_version >= 91) {
992 mddev->reshape_position = sb->reshape_position;
993 mddev->delta_disks = sb->delta_disks;
994 mddev->new_level = sb->new_level;
995 mddev->new_layout = sb->new_layout;
996 mddev->new_chunk_sectors = sb->new_chunk >> 9;
998 mddev->reshape_position = MaxSector;
999 mddev->delta_disks = 0;
1000 mddev->new_level = mddev->level;
1001 mddev->new_layout = mddev->layout;
1002 mddev->new_chunk_sectors = mddev->chunk_sectors;
1005 if (sb->state & (1<<MD_SB_CLEAN))
1006 mddev->recovery_cp = MaxSector;
1008 if (sb->events_hi == sb->cp_events_hi &&
1009 sb->events_lo == sb->cp_events_lo) {
1010 mddev->recovery_cp = sb->recovery_cp;
1012 mddev->recovery_cp = 0;
1015 memcpy(mddev->uuid+0, &sb->set_uuid0, 4);
1016 memcpy(mddev->uuid+4, &sb->set_uuid1, 4);
1017 memcpy(mddev->uuid+8, &sb->set_uuid2, 4);
1018 memcpy(mddev->uuid+12,&sb->set_uuid3, 4);
1020 mddev->max_disks = MD_SB_DISKS;
1022 if (sb->state & (1<<MD_SB_BITMAP_PRESENT) &&
1023 mddev->bitmap_info.file == NULL)
1024 mddev->bitmap_info.offset =
1025 mddev->bitmap_info.default_offset;
1027 } else if (mddev->pers == NULL) {
1028 /* Insist on good event counter while assembling */
1030 if (ev1 < mddev->events)
1032 } else if (mddev->bitmap) {
1033 /* if adding to array with a bitmap, then we can accept an
1034 * older device ... but not too old.
1036 if (ev1 < mddev->bitmap->events_cleared)
1039 if (ev1 < mddev->events)
1040 /* just a hot-add of a new device, leave raid_disk at -1 */
1044 if (mddev->level != LEVEL_MULTIPATH) {
1045 desc = sb->disks + rdev->desc_nr;
1047 if (desc->state & (1<<MD_DISK_FAULTY))
1048 set_bit(Faulty, &rdev->flags);
1049 else if (desc->state & (1<<MD_DISK_SYNC) /* &&
1050 desc->raid_disk < mddev->raid_disks */) {
1051 set_bit(In_sync, &rdev->flags);
1052 rdev->raid_disk = desc->raid_disk;
1053 } else if (desc->state & (1<<MD_DISK_ACTIVE)) {
1054 /* active but not in sync implies recovery up to
1055 * reshape position. We don't know exactly where
1056 * that is, so set to zero for now */
1057 if (mddev->minor_version >= 91) {
1058 rdev->recovery_offset = 0;
1059 rdev->raid_disk = desc->raid_disk;
1062 if (desc->state & (1<<MD_DISK_WRITEMOSTLY))
1063 set_bit(WriteMostly, &rdev->flags);
1064 } else /* MULTIPATH are always insync */
1065 set_bit(In_sync, &rdev->flags);
1070 * sync_super for 0.90.0
1072 static void super_90_sync(mddev_t *mddev, mdk_rdev_t *rdev)
1076 int next_spare = mddev->raid_disks;
1079 /* make rdev->sb match mddev data..
1082 * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
1083 * 3/ any empty disks < next_spare become removed
1085 * disks[0] gets initialised to REMOVED because
1086 * we cannot be sure from other fields if it has
1087 * been initialised or not.
1090 int active=0, working=0,failed=0,spare=0,nr_disks=0;
1092 rdev->sb_size = MD_SB_BYTES;
1094 sb = (mdp_super_t*)page_address(rdev->sb_page);
1096 memset(sb, 0, sizeof(*sb));
1098 sb->md_magic = MD_SB_MAGIC;
1099 sb->major_version = mddev->major_version;
1100 sb->patch_version = mddev->patch_version;
1101 sb->gvalid_words = 0; /* ignored */
1102 memcpy(&sb->set_uuid0, mddev->uuid+0, 4);
1103 memcpy(&sb->set_uuid1, mddev->uuid+4, 4);
1104 memcpy(&sb->set_uuid2, mddev->uuid+8, 4);
1105 memcpy(&sb->set_uuid3, mddev->uuid+12,4);
1107 sb->ctime = mddev->ctime;
1108 sb->level = mddev->level;
1109 sb->size = mddev->dev_sectors / 2;
1110 sb->raid_disks = mddev->raid_disks;
1111 sb->md_minor = mddev->md_minor;
1112 sb->not_persistent = 0;
1113 sb->utime = mddev->utime;
1115 sb->events_hi = (mddev->events>>32);
1116 sb->events_lo = (u32)mddev->events;
1118 if (mddev->reshape_position == MaxSector)
1119 sb->minor_version = 90;
1121 sb->minor_version = 91;
1122 sb->reshape_position = mddev->reshape_position;
1123 sb->new_level = mddev->new_level;
1124 sb->delta_disks = mddev->delta_disks;
1125 sb->new_layout = mddev->new_layout;
1126 sb->new_chunk = mddev->new_chunk_sectors << 9;
1128 mddev->minor_version = sb->minor_version;
1131 sb->recovery_cp = mddev->recovery_cp;
1132 sb->cp_events_hi = (mddev->events>>32);
1133 sb->cp_events_lo = (u32)mddev->events;
1134 if (mddev->recovery_cp == MaxSector)
1135 sb->state = (1<< MD_SB_CLEAN);
1137 sb->recovery_cp = 0;
1139 sb->layout = mddev->layout;
1140 sb->chunk_size = mddev->chunk_sectors << 9;
1142 if (mddev->bitmap && mddev->bitmap_info.file == NULL)
1143 sb->state |= (1<<MD_SB_BITMAP_PRESENT);
1145 sb->disks[0].state = (1<<MD_DISK_REMOVED);
1146 list_for_each_entry(rdev2, &mddev->disks, same_set) {
1149 int is_active = test_bit(In_sync, &rdev2->flags);
1151 if (rdev2->raid_disk >= 0 &&
1152 sb->minor_version >= 91)
1153 /* we have nowhere to store the recovery_offset,
1154 * but if it is not below the reshape_position,
1155 * we can piggy-back on that.
1158 if (rdev2->raid_disk < 0 ||
1159 test_bit(Faulty, &rdev2->flags))
1162 desc_nr = rdev2->raid_disk;
1164 desc_nr = next_spare++;
1165 rdev2->desc_nr = desc_nr;
1166 d = &sb->disks[rdev2->desc_nr];
1168 d->number = rdev2->desc_nr;
1169 d->major = MAJOR(rdev2->bdev->bd_dev);
1170 d->minor = MINOR(rdev2->bdev->bd_dev);
1172 d->raid_disk = rdev2->raid_disk;
1174 d->raid_disk = rdev2->desc_nr; /* compatibility */
1175 if (test_bit(Faulty, &rdev2->flags))
1176 d->state = (1<<MD_DISK_FAULTY);
1177 else if (is_active) {
1178 d->state = (1<<MD_DISK_ACTIVE);
1179 if (test_bit(In_sync, &rdev2->flags))
1180 d->state |= (1<<MD_DISK_SYNC);
1188 if (test_bit(WriteMostly, &rdev2->flags))
1189 d->state |= (1<<MD_DISK_WRITEMOSTLY);
1191 /* now set the "removed" and "faulty" bits on any missing devices */
1192 for (i=0 ; i < mddev->raid_disks ; i++) {
1193 mdp_disk_t *d = &sb->disks[i];
1194 if (d->state == 0 && d->number == 0) {
1197 d->state = (1<<MD_DISK_REMOVED);
1198 d->state |= (1<<MD_DISK_FAULTY);
1202 sb->nr_disks = nr_disks;
1203 sb->active_disks = active;
1204 sb->working_disks = working;
1205 sb->failed_disks = failed;
1206 sb->spare_disks = spare;
1208 sb->this_disk = sb->disks[rdev->desc_nr];
1209 sb->sb_csum = calc_sb_csum(sb);
1213 * rdev_size_change for 0.90.0
1215 static unsigned long long
1216 super_90_rdev_size_change(mdk_rdev_t *rdev, sector_t num_sectors)
1218 if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
1219 return 0; /* component must fit device */
1220 if (rdev->mddev->bitmap_info.offset)
1221 return 0; /* can't move bitmap */
1222 rdev->sb_start = calc_dev_sboffset(rdev->bdev);
1223 if (!num_sectors || num_sectors > rdev->sb_start)
1224 num_sectors = rdev->sb_start;
1225 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
1227 md_super_wait(rdev->mddev);
1228 return num_sectors / 2; /* kB for sysfs */
1233 * version 1 superblock
1236 static __le32 calc_sb_1_csum(struct mdp_superblock_1 * sb)
1240 unsigned long long newcsum;
1241 int size = 256 + le32_to_cpu(sb->max_dev)*2;
1242 __le32 *isuper = (__le32*)sb;
1245 disk_csum = sb->sb_csum;
1248 for (i=0; size>=4; size -= 4 )
1249 newcsum += le32_to_cpu(*isuper++);
1252 newcsum += le16_to_cpu(*(__le16*) isuper);
1254 csum = (newcsum & 0xffffffff) + (newcsum >> 32);
1255 sb->sb_csum = disk_csum;
1256 return cpu_to_le32(csum);
1259 static int super_1_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
1261 struct mdp_superblock_1 *sb;
1264 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
1268 * Calculate the position of the superblock in 512byte sectors.
1269 * It is always aligned to a 4K boundary and
1270 * depeding on minor_version, it can be:
1271 * 0: At least 8K, but less than 12K, from end of device
1272 * 1: At start of device
1273 * 2: 4K from start of device.
1275 switch(minor_version) {
1277 sb_start = rdev->bdev->bd_inode->i_size >> 9;
1279 sb_start &= ~(sector_t)(4*2-1);
1290 rdev->sb_start = sb_start;
1292 /* superblock is rarely larger than 1K, but it can be larger,
1293 * and it is safe to read 4k, so we do that
1295 ret = read_disk_sb(rdev, 4096);
1296 if (ret) return ret;
1299 sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1301 if (sb->magic != cpu_to_le32(MD_SB_MAGIC) ||
1302 sb->major_version != cpu_to_le32(1) ||
1303 le32_to_cpu(sb->max_dev) > (4096-256)/2 ||
1304 le64_to_cpu(sb->super_offset) != rdev->sb_start ||
1305 (le32_to_cpu(sb->feature_map) & ~MD_FEATURE_ALL) != 0)
1308 if (calc_sb_1_csum(sb) != sb->sb_csum) {
1309 printk("md: invalid superblock checksum on %s\n",
1310 bdevname(rdev->bdev,b));
1313 if (le64_to_cpu(sb->data_size) < 10) {
1314 printk("md: data_size too small on %s\n",
1315 bdevname(rdev->bdev,b));
1319 rdev->preferred_minor = 0xffff;
1320 rdev->data_offset = le64_to_cpu(sb->data_offset);
1321 atomic_set(&rdev->corrected_errors, le32_to_cpu(sb->cnt_corrected_read));
1323 rdev->sb_size = le32_to_cpu(sb->max_dev) * 2 + 256;
1324 bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
1325 if (rdev->sb_size & bmask)
1326 rdev->sb_size = (rdev->sb_size | bmask) + 1;
1329 && rdev->data_offset < sb_start + (rdev->sb_size/512))
1332 if (sb->level == cpu_to_le32(LEVEL_MULTIPATH))
1335 rdev->desc_nr = le32_to_cpu(sb->dev_number);
1341 struct mdp_superblock_1 *refsb =
1342 (struct mdp_superblock_1*)page_address(refdev->sb_page);
1344 if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 ||
1345 sb->level != refsb->level ||
1346 sb->layout != refsb->layout ||
1347 sb->chunksize != refsb->chunksize) {
1348 printk(KERN_WARNING "md: %s has strangely different"
1349 " superblock to %s\n",
1350 bdevname(rdev->bdev,b),
1351 bdevname(refdev->bdev,b2));
1354 ev1 = le64_to_cpu(sb->events);
1355 ev2 = le64_to_cpu(refsb->events);
1363 rdev->sectors = (rdev->bdev->bd_inode->i_size >> 9) -
1364 le64_to_cpu(sb->data_offset);
1366 rdev->sectors = rdev->sb_start;
1367 if (rdev->sectors < le64_to_cpu(sb->data_size))
1369 rdev->sectors = le64_to_cpu(sb->data_size);
1370 if (le64_to_cpu(sb->size) > rdev->sectors)
1375 static int super_1_validate(mddev_t *mddev, mdk_rdev_t *rdev)
1377 struct mdp_superblock_1 *sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1378 __u64 ev1 = le64_to_cpu(sb->events);
1380 rdev->raid_disk = -1;
1381 clear_bit(Faulty, &rdev->flags);
1382 clear_bit(In_sync, &rdev->flags);
1383 clear_bit(WriteMostly, &rdev->flags);
1384 clear_bit(BarriersNotsupp, &rdev->flags);
1386 if (mddev->raid_disks == 0) {
1387 mddev->major_version = 1;
1388 mddev->patch_version = 0;
1389 mddev->external = 0;
1390 mddev->chunk_sectors = le32_to_cpu(sb->chunksize);
1391 mddev->ctime = le64_to_cpu(sb->ctime) & ((1ULL << 32)-1);
1392 mddev->utime = le64_to_cpu(sb->utime) & ((1ULL << 32)-1);
1393 mddev->level = le32_to_cpu(sb->level);
1394 mddev->clevel[0] = 0;
1395 mddev->layout = le32_to_cpu(sb->layout);
1396 mddev->raid_disks = le32_to_cpu(sb->raid_disks);
1397 mddev->dev_sectors = le64_to_cpu(sb->size);
1398 mddev->events = ev1;
1399 mddev->bitmap_info.offset = 0;
1400 mddev->bitmap_info.default_offset = 1024 >> 9;
1402 mddev->recovery_cp = le64_to_cpu(sb->resync_offset);
1403 memcpy(mddev->uuid, sb->set_uuid, 16);
1405 mddev->max_disks = (4096-256)/2;
1407 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) &&
1408 mddev->bitmap_info.file == NULL )
1409 mddev->bitmap_info.offset =
1410 (__s32)le32_to_cpu(sb->bitmap_offset);
1412 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) {
1413 mddev->reshape_position = le64_to_cpu(sb->reshape_position);
1414 mddev->delta_disks = le32_to_cpu(sb->delta_disks);
1415 mddev->new_level = le32_to_cpu(sb->new_level);
1416 mddev->new_layout = le32_to_cpu(sb->new_layout);
1417 mddev->new_chunk_sectors = le32_to_cpu(sb->new_chunk);
1419 mddev->reshape_position = MaxSector;
1420 mddev->delta_disks = 0;
1421 mddev->new_level = mddev->level;
1422 mddev->new_layout = mddev->layout;
1423 mddev->new_chunk_sectors = mddev->chunk_sectors;
1426 } else if (mddev->pers == NULL) {
1427 /* Insist of good event counter while assembling */
1429 if (ev1 < mddev->events)
1431 } else if (mddev->bitmap) {
1432 /* If adding to array with a bitmap, then we can accept an
1433 * older device, but not too old.
1435 if (ev1 < mddev->bitmap->events_cleared)
1438 if (ev1 < mddev->events)
1439 /* just a hot-add of a new device, leave raid_disk at -1 */
1442 if (mddev->level != LEVEL_MULTIPATH) {
1444 if (rdev->desc_nr < 0 ||
1445 rdev->desc_nr >= le32_to_cpu(sb->max_dev)) {
1449 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
1451 case 0xffff: /* spare */
1453 case 0xfffe: /* faulty */
1454 set_bit(Faulty, &rdev->flags);
1457 if ((le32_to_cpu(sb->feature_map) &
1458 MD_FEATURE_RECOVERY_OFFSET))
1459 rdev->recovery_offset = le64_to_cpu(sb->recovery_offset);
1461 set_bit(In_sync, &rdev->flags);
1462 rdev->raid_disk = role;
1465 if (sb->devflags & WriteMostly1)
1466 set_bit(WriteMostly, &rdev->flags);
1467 } else /* MULTIPATH are always insync */
1468 set_bit(In_sync, &rdev->flags);
1473 static void super_1_sync(mddev_t *mddev, mdk_rdev_t *rdev)
1475 struct mdp_superblock_1 *sb;
1478 /* make rdev->sb match mddev and rdev data. */
1480 sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1482 sb->feature_map = 0;
1484 sb->recovery_offset = cpu_to_le64(0);
1485 memset(sb->pad1, 0, sizeof(sb->pad1));
1486 memset(sb->pad2, 0, sizeof(sb->pad2));
1487 memset(sb->pad3, 0, sizeof(sb->pad3));
1489 sb->utime = cpu_to_le64((__u64)mddev->utime);
1490 sb->events = cpu_to_le64(mddev->events);
1492 sb->resync_offset = cpu_to_le64(mddev->recovery_cp);
1494 sb->resync_offset = cpu_to_le64(0);
1496 sb->cnt_corrected_read = cpu_to_le32(atomic_read(&rdev->corrected_errors));
1498 sb->raid_disks = cpu_to_le32(mddev->raid_disks);
1499 sb->size = cpu_to_le64(mddev->dev_sectors);
1500 sb->chunksize = cpu_to_le32(mddev->chunk_sectors);
1501 sb->level = cpu_to_le32(mddev->level);
1502 sb->layout = cpu_to_le32(mddev->layout);
1504 if (mddev->bitmap && mddev->bitmap_info.file == NULL) {
1505 sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_info.offset);
1506 sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET);
1509 if (rdev->raid_disk >= 0 &&
1510 !test_bit(In_sync, &rdev->flags)) {
1512 cpu_to_le32(MD_FEATURE_RECOVERY_OFFSET);
1513 sb->recovery_offset =
1514 cpu_to_le64(rdev->recovery_offset);
1517 if (mddev->reshape_position != MaxSector) {
1518 sb->feature_map |= cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE);
1519 sb->reshape_position = cpu_to_le64(mddev->reshape_position);
1520 sb->new_layout = cpu_to_le32(mddev->new_layout);
1521 sb->delta_disks = cpu_to_le32(mddev->delta_disks);
1522 sb->new_level = cpu_to_le32(mddev->new_level);
1523 sb->new_chunk = cpu_to_le32(mddev->new_chunk_sectors);
1527 list_for_each_entry(rdev2, &mddev->disks, same_set)
1528 if (rdev2->desc_nr+1 > max_dev)
1529 max_dev = rdev2->desc_nr+1;
1531 if (max_dev > le32_to_cpu(sb->max_dev)) {
1533 sb->max_dev = cpu_to_le32(max_dev);
1534 rdev->sb_size = max_dev * 2 + 256;
1535 bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
1536 if (rdev->sb_size & bmask)
1537 rdev->sb_size = (rdev->sb_size | bmask) + 1;
1539 for (i=0; i<max_dev;i++)
1540 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1542 list_for_each_entry(rdev2, &mddev->disks, same_set) {
1544 if (test_bit(Faulty, &rdev2->flags))
1545 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1546 else if (test_bit(In_sync, &rdev2->flags))
1547 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1548 else if (rdev2->raid_disk >= 0)
1549 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1551 sb->dev_roles[i] = cpu_to_le16(0xffff);
1554 sb->sb_csum = calc_sb_1_csum(sb);
1557 static unsigned long long
1558 super_1_rdev_size_change(mdk_rdev_t *rdev, sector_t num_sectors)
1560 struct mdp_superblock_1 *sb;
1561 sector_t max_sectors;
1562 if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
1563 return 0; /* component must fit device */
1564 if (rdev->sb_start < rdev->data_offset) {
1565 /* minor versions 1 and 2; superblock before data */
1566 max_sectors = rdev->bdev->bd_inode->i_size >> 9;
1567 max_sectors -= rdev->data_offset;
1568 if (!num_sectors || num_sectors > max_sectors)
1569 num_sectors = max_sectors;
1570 } else if (rdev->mddev->bitmap_info.offset) {
1571 /* minor version 0 with bitmap we can't move */
1574 /* minor version 0; superblock after data */
1576 sb_start = (rdev->bdev->bd_inode->i_size >> 9) - 8*2;
1577 sb_start &= ~(sector_t)(4*2 - 1);
1578 max_sectors = rdev->sectors + sb_start - rdev->sb_start;
1579 if (!num_sectors || num_sectors > max_sectors)
1580 num_sectors = max_sectors;
1581 rdev->sb_start = sb_start;
1583 sb = (struct mdp_superblock_1 *) page_address(rdev->sb_page);
1584 sb->data_size = cpu_to_le64(num_sectors);
1585 sb->super_offset = rdev->sb_start;
1586 sb->sb_csum = calc_sb_1_csum(sb);
1587 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
1589 md_super_wait(rdev->mddev);
1590 return num_sectors / 2; /* kB for sysfs */
1593 static struct super_type super_types[] = {
1596 .owner = THIS_MODULE,
1597 .load_super = super_90_load,
1598 .validate_super = super_90_validate,
1599 .sync_super = super_90_sync,
1600 .rdev_size_change = super_90_rdev_size_change,
1604 .owner = THIS_MODULE,
1605 .load_super = super_1_load,
1606 .validate_super = super_1_validate,
1607 .sync_super = super_1_sync,
1608 .rdev_size_change = super_1_rdev_size_change,
1612 static int match_mddev_units(mddev_t *mddev1, mddev_t *mddev2)
1614 mdk_rdev_t *rdev, *rdev2;
1617 rdev_for_each_rcu(rdev, mddev1)
1618 rdev_for_each_rcu(rdev2, mddev2)
1619 if (rdev->bdev->bd_contains ==
1620 rdev2->bdev->bd_contains) {
1628 static LIST_HEAD(pending_raid_disks);
1631 * Try to register data integrity profile for an mddev
1633 * This is called when an array is started and after a disk has been kicked
1634 * from the array. It only succeeds if all working and active component devices
1635 * are integrity capable with matching profiles.
1637 int md_integrity_register(mddev_t *mddev)
1639 mdk_rdev_t *rdev, *reference = NULL;
1641 if (list_empty(&mddev->disks))
1642 return 0; /* nothing to do */
1643 if (blk_get_integrity(mddev->gendisk))
1644 return 0; /* already registered */
1645 list_for_each_entry(rdev, &mddev->disks, same_set) {
1646 /* skip spares and non-functional disks */
1647 if (test_bit(Faulty, &rdev->flags))
1649 if (rdev->raid_disk < 0)
1652 * If at least one rdev is not integrity capable, we can not
1653 * enable data integrity for the md device.
1655 if (!bdev_get_integrity(rdev->bdev))
1658 /* Use the first rdev as the reference */
1662 /* does this rdev's profile match the reference profile? */
1663 if (blk_integrity_compare(reference->bdev->bd_disk,
1664 rdev->bdev->bd_disk) < 0)
1668 * All component devices are integrity capable and have matching
1669 * profiles, register the common profile for the md device.
1671 if (blk_integrity_register(mddev->gendisk,
1672 bdev_get_integrity(reference->bdev)) != 0) {
1673 printk(KERN_ERR "md: failed to register integrity for %s\n",
1677 printk(KERN_NOTICE "md: data integrity on %s enabled\n",
1681 EXPORT_SYMBOL(md_integrity_register);
1683 /* Disable data integrity if non-capable/non-matching disk is being added */
1684 void md_integrity_add_rdev(mdk_rdev_t *rdev, mddev_t *mddev)
1686 struct blk_integrity *bi_rdev = bdev_get_integrity(rdev->bdev);
1687 struct blk_integrity *bi_mddev = blk_get_integrity(mddev->gendisk);
1689 if (!bi_mddev) /* nothing to do */
1691 if (rdev->raid_disk < 0) /* skip spares */
1693 if (bi_rdev && blk_integrity_compare(mddev->gendisk,
1694 rdev->bdev->bd_disk) >= 0)
1696 printk(KERN_NOTICE "disabling data integrity on %s\n", mdname(mddev));
1697 blk_integrity_unregister(mddev->gendisk);
1699 EXPORT_SYMBOL(md_integrity_add_rdev);
1701 static int bind_rdev_to_array(mdk_rdev_t * rdev, mddev_t * mddev)
1703 char b[BDEVNAME_SIZE];
1713 /* prevent duplicates */
1714 if (find_rdev(mddev, rdev->bdev->bd_dev))
1717 /* make sure rdev->sectors exceeds mddev->dev_sectors */
1718 if (rdev->sectors && (mddev->dev_sectors == 0 ||
1719 rdev->sectors < mddev->dev_sectors)) {
1721 /* Cannot change size, so fail
1722 * If mddev->level <= 0, then we don't care
1723 * about aligning sizes (e.g. linear)
1725 if (mddev->level > 0)
1728 mddev->dev_sectors = rdev->sectors;
1731 /* Verify rdev->desc_nr is unique.
1732 * If it is -1, assign a free number, else
1733 * check number is not in use
1735 if (rdev->desc_nr < 0) {
1737 if (mddev->pers) choice = mddev->raid_disks;
1738 while (find_rdev_nr(mddev, choice))
1740 rdev->desc_nr = choice;
1742 if (find_rdev_nr(mddev, rdev->desc_nr))
1745 if (mddev->max_disks && rdev->desc_nr >= mddev->max_disks) {
1746 printk(KERN_WARNING "md: %s: array is limited to %d devices\n",
1747 mdname(mddev), mddev->max_disks);
1750 bdevname(rdev->bdev,b);
1751 while ( (s=strchr(b, '/')) != NULL)
1754 rdev->mddev = mddev;
1755 printk(KERN_INFO "md: bind<%s>\n", b);
1757 if ((err = kobject_add(&rdev->kobj, &mddev->kobj, "dev-%s", b)))
1760 ko = &part_to_dev(rdev->bdev->bd_part)->kobj;
1761 if ((err = sysfs_create_link(&rdev->kobj, ko, "block"))) {
1762 kobject_del(&rdev->kobj);
1765 rdev->sysfs_state = sysfs_get_dirent(rdev->kobj.sd, "state");
1767 list_add_rcu(&rdev->same_set, &mddev->disks);
1768 bd_claim_by_disk(rdev->bdev, rdev->bdev->bd_holder, mddev->gendisk);
1770 /* May as well allow recovery to be retried once */
1771 mddev->recovery_disabled = 0;
1776 printk(KERN_WARNING "md: failed to register dev-%s for %s\n",
1781 static void md_delayed_delete(struct work_struct *ws)
1783 mdk_rdev_t *rdev = container_of(ws, mdk_rdev_t, del_work);
1784 kobject_del(&rdev->kobj);
1785 kobject_put(&rdev->kobj);
1788 static void unbind_rdev_from_array(mdk_rdev_t * rdev)
1790 char b[BDEVNAME_SIZE];
1795 bd_release_from_disk(rdev->bdev, rdev->mddev->gendisk);
1796 list_del_rcu(&rdev->same_set);
1797 printk(KERN_INFO "md: unbind<%s>\n", bdevname(rdev->bdev,b));
1799 sysfs_remove_link(&rdev->kobj, "block");
1800 sysfs_put(rdev->sysfs_state);
1801 rdev->sysfs_state = NULL;
1802 /* We need to delay this, otherwise we can deadlock when
1803 * writing to 'remove' to "dev/state". We also need
1804 * to delay it due to rcu usage.
1807 INIT_WORK(&rdev->del_work, md_delayed_delete);
1808 kobject_get(&rdev->kobj);
1809 schedule_work(&rdev->del_work);
1813 * prevent the device from being mounted, repartitioned or
1814 * otherwise reused by a RAID array (or any other kernel
1815 * subsystem), by bd_claiming the device.
1817 static int lock_rdev(mdk_rdev_t *rdev, dev_t dev, int shared)
1820 struct block_device *bdev;
1821 char b[BDEVNAME_SIZE];
1823 bdev = open_by_devnum(dev, FMODE_READ|FMODE_WRITE);
1825 printk(KERN_ERR "md: could not open %s.\n",
1826 __bdevname(dev, b));
1827 return PTR_ERR(bdev);
1829 err = bd_claim(bdev, shared ? (mdk_rdev_t *)lock_rdev : rdev);
1831 printk(KERN_ERR "md: could not bd_claim %s.\n",
1833 blkdev_put(bdev, FMODE_READ|FMODE_WRITE);
1837 set_bit(AllReserved, &rdev->flags);
1842 static void unlock_rdev(mdk_rdev_t *rdev)
1844 struct block_device *bdev = rdev->bdev;
1849 blkdev_put(bdev, FMODE_READ|FMODE_WRITE);
1852 void md_autodetect_dev(dev_t dev);
1854 static void export_rdev(mdk_rdev_t * rdev)
1856 char b[BDEVNAME_SIZE];
1857 printk(KERN_INFO "md: export_rdev(%s)\n",
1858 bdevname(rdev->bdev,b));
1863 if (test_bit(AutoDetected, &rdev->flags))
1864 md_autodetect_dev(rdev->bdev->bd_dev);
1867 kobject_put(&rdev->kobj);
1870 static void kick_rdev_from_array(mdk_rdev_t * rdev)
1872 unbind_rdev_from_array(rdev);
1876 static void export_array(mddev_t *mddev)
1878 mdk_rdev_t *rdev, *tmp;
1880 rdev_for_each(rdev, tmp, mddev) {
1885 kick_rdev_from_array(rdev);
1887 if (!list_empty(&mddev->disks))
1889 mddev->raid_disks = 0;
1890 mddev->major_version = 0;
1893 static void print_desc(mdp_disk_t *desc)
1895 printk(" DISK<N:%d,(%d,%d),R:%d,S:%d>\n", desc->number,
1896 desc->major,desc->minor,desc->raid_disk,desc->state);
1899 static void print_sb_90(mdp_super_t *sb)
1904 "md: SB: (V:%d.%d.%d) ID:<%08x.%08x.%08x.%08x> CT:%08x\n",
1905 sb->major_version, sb->minor_version, sb->patch_version,
1906 sb->set_uuid0, sb->set_uuid1, sb->set_uuid2, sb->set_uuid3,
1908 printk(KERN_INFO "md: L%d S%08d ND:%d RD:%d md%d LO:%d CS:%d\n",
1909 sb->level, sb->size, sb->nr_disks, sb->raid_disks,
1910 sb->md_minor, sb->layout, sb->chunk_size);
1911 printk(KERN_INFO "md: UT:%08x ST:%d AD:%d WD:%d"
1912 " FD:%d SD:%d CSUM:%08x E:%08lx\n",
1913 sb->utime, sb->state, sb->active_disks, sb->working_disks,
1914 sb->failed_disks, sb->spare_disks,
1915 sb->sb_csum, (unsigned long)sb->events_lo);
1918 for (i = 0; i < MD_SB_DISKS; i++) {
1921 desc = sb->disks + i;
1922 if (desc->number || desc->major || desc->minor ||
1923 desc->raid_disk || (desc->state && (desc->state != 4))) {
1924 printk(" D %2d: ", i);
1928 printk(KERN_INFO "md: THIS: ");
1929 print_desc(&sb->this_disk);
1932 static void print_sb_1(struct mdp_superblock_1 *sb)
1936 uuid = sb->set_uuid;
1938 "md: SB: (V:%u) (F:0x%08x) Array-ID:<%02x%02x%02x%02x"
1939 ":%02x%02x:%02x%02x:%02x%02x:%02x%02x%02x%02x%02x%02x>\n"
1940 "md: Name: \"%s\" CT:%llu\n",
1941 le32_to_cpu(sb->major_version),
1942 le32_to_cpu(sb->feature_map),
1943 uuid[0], uuid[1], uuid[2], uuid[3],
1944 uuid[4], uuid[5], uuid[6], uuid[7],
1945 uuid[8], uuid[9], uuid[10], uuid[11],
1946 uuid[12], uuid[13], uuid[14], uuid[15],
1948 (unsigned long long)le64_to_cpu(sb->ctime)
1949 & MD_SUPERBLOCK_1_TIME_SEC_MASK);
1951 uuid = sb->device_uuid;
1953 "md: L%u SZ%llu RD:%u LO:%u CS:%u DO:%llu DS:%llu SO:%llu"
1955 "md: Dev:%08x UUID: %02x%02x%02x%02x:%02x%02x:%02x%02x:%02x%02x"
1956 ":%02x%02x%02x%02x%02x%02x\n"
1957 "md: (F:0x%08x) UT:%llu Events:%llu ResyncOffset:%llu CSUM:0x%08x\n"
1958 "md: (MaxDev:%u) \n",
1959 le32_to_cpu(sb->level),
1960 (unsigned long long)le64_to_cpu(sb->size),
1961 le32_to_cpu(sb->raid_disks),
1962 le32_to_cpu(sb->layout),
1963 le32_to_cpu(sb->chunksize),
1964 (unsigned long long)le64_to_cpu(sb->data_offset),
1965 (unsigned long long)le64_to_cpu(sb->data_size),
1966 (unsigned long long)le64_to_cpu(sb->super_offset),
1967 (unsigned long long)le64_to_cpu(sb->recovery_offset),
1968 le32_to_cpu(sb->dev_number),
1969 uuid[0], uuid[1], uuid[2], uuid[3],
1970 uuid[4], uuid[5], uuid[6], uuid[7],
1971 uuid[8], uuid[9], uuid[10], uuid[11],
1972 uuid[12], uuid[13], uuid[14], uuid[15],
1974 (unsigned long long)le64_to_cpu(sb->utime) & MD_SUPERBLOCK_1_TIME_SEC_MASK,
1975 (unsigned long long)le64_to_cpu(sb->events),
1976 (unsigned long long)le64_to_cpu(sb->resync_offset),
1977 le32_to_cpu(sb->sb_csum),
1978 le32_to_cpu(sb->max_dev)
1982 static void print_rdev(mdk_rdev_t *rdev, int major_version)
1984 char b[BDEVNAME_SIZE];
1985 printk(KERN_INFO "md: rdev %s, Sect:%08llu F:%d S:%d DN:%u\n",
1986 bdevname(rdev->bdev, b), (unsigned long long)rdev->sectors,
1987 test_bit(Faulty, &rdev->flags), test_bit(In_sync, &rdev->flags),
1989 if (rdev->sb_loaded) {
1990 printk(KERN_INFO "md: rdev superblock (MJ:%d):\n", major_version);
1991 switch (major_version) {
1993 print_sb_90((mdp_super_t*)page_address(rdev->sb_page));
1996 print_sb_1((struct mdp_superblock_1 *)page_address(rdev->sb_page));
2000 printk(KERN_INFO "md: no rdev superblock!\n");
2003 static void md_print_devices(void)
2005 struct list_head *tmp;
2008 char b[BDEVNAME_SIZE];
2011 printk("md: **********************************\n");
2012 printk("md: * <COMPLETE RAID STATE PRINTOUT> *\n");
2013 printk("md: **********************************\n");
2014 for_each_mddev(mddev, tmp) {
2017 bitmap_print_sb(mddev->bitmap);
2019 printk("%s: ", mdname(mddev));
2020 list_for_each_entry(rdev, &mddev->disks, same_set)
2021 printk("<%s>", bdevname(rdev->bdev,b));
2024 list_for_each_entry(rdev, &mddev->disks, same_set)
2025 print_rdev(rdev, mddev->major_version);
2027 printk("md: **********************************\n");
2032 static void sync_sbs(mddev_t * mddev, int nospares)
2034 /* Update each superblock (in-memory image), but
2035 * if we are allowed to, skip spares which already
2036 * have the right event counter, or have one earlier
2037 * (which would mean they aren't being marked as dirty
2038 * with the rest of the array)
2042 /* First make sure individual recovery_offsets are correct */
2043 list_for_each_entry(rdev, &mddev->disks, same_set) {
2044 if (rdev->raid_disk >= 0 &&
2045 !test_bit(In_sync, &rdev->flags) &&
2046 mddev->curr_resync_completed > rdev->recovery_offset)
2047 rdev->recovery_offset = mddev->curr_resync_completed;
2050 list_for_each_entry(rdev, &mddev->disks, same_set) {
2051 if (rdev->sb_events == mddev->events ||
2053 rdev->raid_disk < 0 &&
2054 (rdev->sb_events&1)==0 &&
2055 rdev->sb_events+1 == mddev->events)) {
2056 /* Don't update this superblock */
2057 rdev->sb_loaded = 2;
2059 super_types[mddev->major_version].
2060 sync_super(mddev, rdev);
2061 rdev->sb_loaded = 1;
2066 static void md_update_sb(mddev_t * mddev, int force_change)
2072 mddev->utime = get_seconds();
2073 if (mddev->external)
2076 spin_lock_irq(&mddev->write_lock);
2078 set_bit(MD_CHANGE_PENDING, &mddev->flags);
2079 if (test_and_clear_bit(MD_CHANGE_DEVS, &mddev->flags))
2081 if (test_and_clear_bit(MD_CHANGE_CLEAN, &mddev->flags))
2082 /* just a clean<-> dirty transition, possibly leave spares alone,
2083 * though if events isn't the right even/odd, we will have to do
2089 if (mddev->degraded)
2090 /* If the array is degraded, then skipping spares is both
2091 * dangerous and fairly pointless.
2092 * Dangerous because a device that was removed from the array
2093 * might have a event_count that still looks up-to-date,
2094 * so it can be re-added without a resync.
2095 * Pointless because if there are any spares to skip,
2096 * then a recovery will happen and soon that array won't
2097 * be degraded any more and the spare can go back to sleep then.
2101 sync_req = mddev->in_sync;
2103 /* If this is just a dirty<->clean transition, and the array is clean
2104 * and 'events' is odd, we can roll back to the previous clean state */
2106 && (mddev->in_sync && mddev->recovery_cp == MaxSector)
2107 && (mddev->events & 1)
2108 && mddev->events != 1)
2111 /* otherwise we have to go forward and ... */
2113 if (!mddev->in_sync || mddev->recovery_cp != MaxSector) { /* not clean */
2114 /* .. if the array isn't clean, an 'even' event must also go
2116 if ((mddev->events&1)==0)
2119 /* otherwise an 'odd' event must go to spares */
2120 if ((mddev->events&1))
2125 if (!mddev->events) {
2127 * oops, this 64-bit counter should never wrap.
2128 * Either we are in around ~1 trillion A.C., assuming
2129 * 1 reboot per second, or we have a bug:
2136 * do not write anything to disk if using
2137 * nonpersistent superblocks
2139 if (!mddev->persistent) {
2140 if (!mddev->external)
2141 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
2143 spin_unlock_irq(&mddev->write_lock);
2144 wake_up(&mddev->sb_wait);
2147 sync_sbs(mddev, nospares);
2148 spin_unlock_irq(&mddev->write_lock);
2151 "md: updating %s RAID superblock on device (in sync %d)\n",
2152 mdname(mddev),mddev->in_sync);
2154 bitmap_update_sb(mddev->bitmap);
2155 list_for_each_entry(rdev, &mddev->disks, same_set) {
2156 char b[BDEVNAME_SIZE];
2157 dprintk(KERN_INFO "md: ");
2158 if (rdev->sb_loaded != 1)
2159 continue; /* no noise on spare devices */
2160 if (test_bit(Faulty, &rdev->flags))
2161 dprintk("(skipping faulty ");
2163 dprintk("%s ", bdevname(rdev->bdev,b));
2164 if (!test_bit(Faulty, &rdev->flags)) {
2165 md_super_write(mddev,rdev,
2166 rdev->sb_start, rdev->sb_size,
2168 dprintk(KERN_INFO "(write) %s's sb offset: %llu\n",
2169 bdevname(rdev->bdev,b),
2170 (unsigned long long)rdev->sb_start);
2171 rdev->sb_events = mddev->events;
2175 if (mddev->level == LEVEL_MULTIPATH)
2176 /* only need to write one superblock... */
2179 md_super_wait(mddev);
2180 /* if there was a failure, MD_CHANGE_DEVS was set, and we re-write super */
2182 spin_lock_irq(&mddev->write_lock);
2183 if (mddev->in_sync != sync_req ||
2184 test_bit(MD_CHANGE_DEVS, &mddev->flags)) {
2185 /* have to write it out again */
2186 spin_unlock_irq(&mddev->write_lock);
2189 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
2190 spin_unlock_irq(&mddev->write_lock);
2191 wake_up(&mddev->sb_wait);
2192 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
2193 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
2197 /* words written to sysfs files may, or may not, be \n terminated.
2198 * We want to accept with case. For this we use cmd_match.
2200 static int cmd_match(const char *cmd, const char *str)
2202 /* See if cmd, written into a sysfs file, matches
2203 * str. They must either be the same, or cmd can
2204 * have a trailing newline
2206 while (*cmd && *str && *cmd == *str) {
2217 struct rdev_sysfs_entry {
2218 struct attribute attr;
2219 ssize_t (*show)(mdk_rdev_t *, char *);
2220 ssize_t (*store)(mdk_rdev_t *, const char *, size_t);
2224 state_show(mdk_rdev_t *rdev, char *page)
2229 if (test_bit(Faulty, &rdev->flags)) {
2230 len+= sprintf(page+len, "%sfaulty",sep);
2233 if (test_bit(In_sync, &rdev->flags)) {
2234 len += sprintf(page+len, "%sin_sync",sep);
2237 if (test_bit(WriteMostly, &rdev->flags)) {
2238 len += sprintf(page+len, "%swrite_mostly",sep);
2241 if (test_bit(Blocked, &rdev->flags)) {
2242 len += sprintf(page+len, "%sblocked", sep);
2245 if (!test_bit(Faulty, &rdev->flags) &&
2246 !test_bit(In_sync, &rdev->flags)) {
2247 len += sprintf(page+len, "%sspare", sep);
2250 return len+sprintf(page+len, "\n");
2254 state_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2257 * faulty - simulates and error
2258 * remove - disconnects the device
2259 * writemostly - sets write_mostly
2260 * -writemostly - clears write_mostly
2261 * blocked - sets the Blocked flag
2262 * -blocked - clears the Blocked flag
2263 * insync - sets Insync providing device isn't active
2266 if (cmd_match(buf, "faulty") && rdev->mddev->pers) {
2267 md_error(rdev->mddev, rdev);
2269 } else if (cmd_match(buf, "remove")) {
2270 if (rdev->raid_disk >= 0)
2273 mddev_t *mddev = rdev->mddev;
2274 kick_rdev_from_array(rdev);
2276 md_update_sb(mddev, 1);
2277 md_new_event(mddev);
2280 } else if (cmd_match(buf, "writemostly")) {
2281 set_bit(WriteMostly, &rdev->flags);
2283 } else if (cmd_match(buf, "-writemostly")) {
2284 clear_bit(WriteMostly, &rdev->flags);
2286 } else if (cmd_match(buf, "blocked")) {
2287 set_bit(Blocked, &rdev->flags);
2289 } else if (cmd_match(buf, "-blocked")) {
2290 clear_bit(Blocked, &rdev->flags);
2291 wake_up(&rdev->blocked_wait);
2292 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2293 md_wakeup_thread(rdev->mddev->thread);
2296 } else if (cmd_match(buf, "insync") && rdev->raid_disk == -1) {
2297 set_bit(In_sync, &rdev->flags);
2300 if (!err && rdev->sysfs_state)
2301 sysfs_notify_dirent(rdev->sysfs_state);
2302 return err ? err : len;
2304 static struct rdev_sysfs_entry rdev_state =
2305 __ATTR(state, S_IRUGO|S_IWUSR, state_show, state_store);
2308 errors_show(mdk_rdev_t *rdev, char *page)
2310 return sprintf(page, "%d\n", atomic_read(&rdev->corrected_errors));
2314 errors_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2317 unsigned long n = simple_strtoul(buf, &e, 10);
2318 if (*buf && (*e == 0 || *e == '\n')) {
2319 atomic_set(&rdev->corrected_errors, n);
2324 static struct rdev_sysfs_entry rdev_errors =
2325 __ATTR(errors, S_IRUGO|S_IWUSR, errors_show, errors_store);
2328 slot_show(mdk_rdev_t *rdev, char *page)
2330 if (rdev->raid_disk < 0)
2331 return sprintf(page, "none\n");
2333 return sprintf(page, "%d\n", rdev->raid_disk);
2337 slot_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2342 int slot = simple_strtoul(buf, &e, 10);
2343 if (strncmp(buf, "none", 4)==0)
2345 else if (e==buf || (*e && *e!= '\n'))
2347 if (rdev->mddev->pers && slot == -1) {
2348 /* Setting 'slot' on an active array requires also
2349 * updating the 'rd%d' link, and communicating
2350 * with the personality with ->hot_*_disk.
2351 * For now we only support removing
2352 * failed/spare devices. This normally happens automatically,
2353 * but not when the metadata is externally managed.
2355 if (rdev->raid_disk == -1)
2357 /* personality does all needed checks */
2358 if (rdev->mddev->pers->hot_add_disk == NULL)
2360 err = rdev->mddev->pers->
2361 hot_remove_disk(rdev->mddev, rdev->raid_disk);
2364 sprintf(nm, "rd%d", rdev->raid_disk);
2365 sysfs_remove_link(&rdev->mddev->kobj, nm);
2366 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2367 md_wakeup_thread(rdev->mddev->thread);
2368 } else if (rdev->mddev->pers) {
2370 /* Activating a spare .. or possibly reactivating
2371 * if we ever get bitmaps working here.
2374 if (rdev->raid_disk != -1)
2377 if (rdev->mddev->pers->hot_add_disk == NULL)
2380 list_for_each_entry(rdev2, &rdev->mddev->disks, same_set)
2381 if (rdev2->raid_disk == slot)
2384 rdev->raid_disk = slot;
2385 if (test_bit(In_sync, &rdev->flags))
2386 rdev->saved_raid_disk = slot;
2388 rdev->saved_raid_disk = -1;
2389 err = rdev->mddev->pers->
2390 hot_add_disk(rdev->mddev, rdev);
2392 rdev->raid_disk = -1;
2395 sysfs_notify_dirent(rdev->sysfs_state);
2396 sprintf(nm, "rd%d", rdev->raid_disk);
2397 if (sysfs_create_link(&rdev->mddev->kobj, &rdev->kobj, nm))
2399 "md: cannot register "
2401 nm, mdname(rdev->mddev));
2403 /* don't wakeup anyone, leave that to userspace. */
2405 if (slot >= rdev->mddev->raid_disks)
2407 rdev->raid_disk = slot;
2408 /* assume it is working */
2409 clear_bit(Faulty, &rdev->flags);
2410 clear_bit(WriteMostly, &rdev->flags);
2411 set_bit(In_sync, &rdev->flags);
2412 sysfs_notify_dirent(rdev->sysfs_state);
2418 static struct rdev_sysfs_entry rdev_slot =
2419 __ATTR(slot, S_IRUGO|S_IWUSR, slot_show, slot_store);
2422 offset_show(mdk_rdev_t *rdev, char *page)
2424 return sprintf(page, "%llu\n", (unsigned long long)rdev->data_offset);
2428 offset_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2431 unsigned long long offset = simple_strtoull(buf, &e, 10);
2432 if (e==buf || (*e && *e != '\n'))
2434 if (rdev->mddev->pers && rdev->raid_disk >= 0)
2436 if (rdev->sectors && rdev->mddev->external)
2437 /* Must set offset before size, so overlap checks
2440 rdev->data_offset = offset;
2444 static struct rdev_sysfs_entry rdev_offset =
2445 __ATTR(offset, S_IRUGO|S_IWUSR, offset_show, offset_store);
2448 rdev_size_show(mdk_rdev_t *rdev, char *page)
2450 return sprintf(page, "%llu\n", (unsigned long long)rdev->sectors / 2);
2453 static int overlaps(sector_t s1, sector_t l1, sector_t s2, sector_t l2)
2455 /* check if two start/length pairs overlap */
2463 static int strict_blocks_to_sectors(const char *buf, sector_t *sectors)
2465 unsigned long long blocks;
2468 if (strict_strtoull(buf, 10, &blocks) < 0)
2471 if (blocks & 1ULL << (8 * sizeof(blocks) - 1))
2472 return -EINVAL; /* sector conversion overflow */
2475 if (new != blocks * 2)
2476 return -EINVAL; /* unsigned long long to sector_t overflow */
2483 rdev_size_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2485 mddev_t *my_mddev = rdev->mddev;
2486 sector_t oldsectors = rdev->sectors;
2489 if (strict_blocks_to_sectors(buf, §ors) < 0)
2491 if (my_mddev->pers && rdev->raid_disk >= 0) {
2492 if (my_mddev->persistent) {
2493 sectors = super_types[my_mddev->major_version].
2494 rdev_size_change(rdev, sectors);
2497 } else if (!sectors)
2498 sectors = (rdev->bdev->bd_inode->i_size >> 9) -
2501 if (sectors < my_mddev->dev_sectors)
2502 return -EINVAL; /* component must fit device */
2504 rdev->sectors = sectors;
2505 if (sectors > oldsectors && my_mddev->external) {
2506 /* need to check that all other rdevs with the same ->bdev
2507 * do not overlap. We need to unlock the mddev to avoid
2508 * a deadlock. We have already changed rdev->sectors, and if
2509 * we have to change it back, we will have the lock again.
2513 struct list_head *tmp;
2515 mddev_unlock(my_mddev);
2516 for_each_mddev(mddev, tmp) {
2520 list_for_each_entry(rdev2, &mddev->disks, same_set)
2521 if (test_bit(AllReserved, &rdev2->flags) ||
2522 (rdev->bdev == rdev2->bdev &&
2524 overlaps(rdev->data_offset, rdev->sectors,
2530 mddev_unlock(mddev);
2536 mddev_lock(my_mddev);
2538 /* Someone else could have slipped in a size
2539 * change here, but doing so is just silly.
2540 * We put oldsectors back because we *know* it is
2541 * safe, and trust userspace not to race with
2544 rdev->sectors = oldsectors;
2551 static struct rdev_sysfs_entry rdev_size =
2552 __ATTR(size, S_IRUGO|S_IWUSR, rdev_size_show, rdev_size_store);
2555 static ssize_t recovery_start_show(mdk_rdev_t *rdev, char *page)
2557 unsigned long long recovery_start = rdev->recovery_offset;
2559 if (test_bit(In_sync, &rdev->flags) ||
2560 recovery_start == MaxSector)
2561 return sprintf(page, "none\n");
2563 return sprintf(page, "%llu\n", recovery_start);
2566 static ssize_t recovery_start_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2568 unsigned long long recovery_start;
2570 if (cmd_match(buf, "none"))
2571 recovery_start = MaxSector;
2572 else if (strict_strtoull(buf, 10, &recovery_start))
2575 if (rdev->mddev->pers &&
2576 rdev->raid_disk >= 0)
2579 rdev->recovery_offset = recovery_start;
2580 if (recovery_start == MaxSector)
2581 set_bit(In_sync, &rdev->flags);
2583 clear_bit(In_sync, &rdev->flags);
2587 static struct rdev_sysfs_entry rdev_recovery_start =
2588 __ATTR(recovery_start, S_IRUGO|S_IWUSR, recovery_start_show, recovery_start_store);
2590 static struct attribute *rdev_default_attrs[] = {
2596 &rdev_recovery_start.attr,
2600 rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
2602 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
2603 mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
2604 mddev_t *mddev = rdev->mddev;
2610 rv = mddev ? mddev_lock(mddev) : -EBUSY;
2612 if (rdev->mddev == NULL)
2615 rv = entry->show(rdev, page);
2616 mddev_unlock(mddev);
2622 rdev_attr_store(struct kobject *kobj, struct attribute *attr,
2623 const char *page, size_t length)
2625 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
2626 mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
2628 mddev_t *mddev = rdev->mddev;
2632 if (!capable(CAP_SYS_ADMIN))
2634 rv = mddev ? mddev_lock(mddev): -EBUSY;
2636 if (rdev->mddev == NULL)
2639 rv = entry->store(rdev, page, length);
2640 mddev_unlock(mddev);
2645 static void rdev_free(struct kobject *ko)
2647 mdk_rdev_t *rdev = container_of(ko, mdk_rdev_t, kobj);
2650 static struct sysfs_ops rdev_sysfs_ops = {
2651 .show = rdev_attr_show,
2652 .store = rdev_attr_store,
2654 static struct kobj_type rdev_ktype = {
2655 .release = rdev_free,
2656 .sysfs_ops = &rdev_sysfs_ops,
2657 .default_attrs = rdev_default_attrs,
2661 * Import a device. If 'super_format' >= 0, then sanity check the superblock
2663 * mark the device faulty if:
2665 * - the device is nonexistent (zero size)
2666 * - the device has no valid superblock
2668 * a faulty rdev _never_ has rdev->sb set.
2670 static mdk_rdev_t *md_import_device(dev_t newdev, int super_format, int super_minor)
2672 char b[BDEVNAME_SIZE];
2677 rdev = kzalloc(sizeof(*rdev), GFP_KERNEL);
2679 printk(KERN_ERR "md: could not alloc mem for new device!\n");
2680 return ERR_PTR(-ENOMEM);
2683 if ((err = alloc_disk_sb(rdev)))
2686 err = lock_rdev(rdev, newdev, super_format == -2);
2690 kobject_init(&rdev->kobj, &rdev_ktype);
2693 rdev->saved_raid_disk = -1;
2694 rdev->raid_disk = -1;
2696 rdev->data_offset = 0;
2697 rdev->sb_events = 0;
2698 rdev->last_read_error.tv_sec = 0;
2699 rdev->last_read_error.tv_nsec = 0;
2700 atomic_set(&rdev->nr_pending, 0);
2701 atomic_set(&rdev->read_errors, 0);
2702 atomic_set(&rdev->corrected_errors, 0);
2704 size = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
2707 "md: %s has zero or unknown size, marking faulty!\n",
2708 bdevname(rdev->bdev,b));
2713 if (super_format >= 0) {
2714 err = super_types[super_format].
2715 load_super(rdev, NULL, super_minor);
2716 if (err == -EINVAL) {
2718 "md: %s does not have a valid v%d.%d "
2719 "superblock, not importing!\n",
2720 bdevname(rdev->bdev,b),
2721 super_format, super_minor);
2726 "md: could not read %s's sb, not importing!\n",
2727 bdevname(rdev->bdev,b));
2732 INIT_LIST_HEAD(&rdev->same_set);
2733 init_waitqueue_head(&rdev->blocked_wait);
2738 if (rdev->sb_page) {
2744 return ERR_PTR(err);
2748 * Check a full RAID array for plausibility
2752 static void analyze_sbs(mddev_t * mddev)
2755 mdk_rdev_t *rdev, *freshest, *tmp;
2756 char b[BDEVNAME_SIZE];
2759 rdev_for_each(rdev, tmp, mddev)
2760 switch (super_types[mddev->major_version].
2761 load_super(rdev, freshest, mddev->minor_version)) {
2769 "md: fatal superblock inconsistency in %s"
2770 " -- removing from array\n",
2771 bdevname(rdev->bdev,b));
2772 kick_rdev_from_array(rdev);
2776 super_types[mddev->major_version].
2777 validate_super(mddev, freshest);
2780 rdev_for_each(rdev, tmp, mddev) {
2781 if (rdev->desc_nr >= mddev->max_disks ||
2782 i > mddev->max_disks) {
2784 "md: %s: %s: only %d devices permitted\n",
2785 mdname(mddev), bdevname(rdev->bdev, b),
2787 kick_rdev_from_array(rdev);
2790 if (rdev != freshest)
2791 if (super_types[mddev->major_version].
2792 validate_super(mddev, rdev)) {
2793 printk(KERN_WARNING "md: kicking non-fresh %s"
2795 bdevname(rdev->bdev,b));
2796 kick_rdev_from_array(rdev);
2799 if (mddev->level == LEVEL_MULTIPATH) {
2800 rdev->desc_nr = i++;
2801 rdev->raid_disk = rdev->desc_nr;
2802 set_bit(In_sync, &rdev->flags);
2803 } else if (rdev->raid_disk >= (mddev->raid_disks - min(0, mddev->delta_disks))) {
2804 rdev->raid_disk = -1;
2805 clear_bit(In_sync, &rdev->flags);
2810 /* Read a fixed-point number.
2811 * Numbers in sysfs attributes should be in "standard" units where
2812 * possible, so time should be in seconds.
2813 * However we internally use a a much smaller unit such as
2814 * milliseconds or jiffies.
2815 * This function takes a decimal number with a possible fractional
2816 * component, and produces an integer which is the result of
2817 * multiplying that number by 10^'scale'.
2818 * all without any floating-point arithmetic.
2820 int strict_strtoul_scaled(const char *cp, unsigned long *res, int scale)
2822 unsigned long result = 0;
2824 while (isdigit(*cp) || (*cp == '.' && decimals < 0)) {
2827 else if (decimals < scale) {
2830 result = result * 10 + value;
2842 while (decimals < scale) {
2851 static void md_safemode_timeout(unsigned long data);
2854 safe_delay_show(mddev_t *mddev, char *page)
2856 int msec = (mddev->safemode_delay*1000)/HZ;
2857 return sprintf(page, "%d.%03d\n", msec/1000, msec%1000);
2860 safe_delay_store(mddev_t *mddev, const char *cbuf, size_t len)
2864 if (strict_strtoul_scaled(cbuf, &msec, 3) < 0)
2867 mddev->safemode_delay = 0;
2869 unsigned long old_delay = mddev->safemode_delay;
2870 mddev->safemode_delay = (msec*HZ)/1000;
2871 if (mddev->safemode_delay == 0)
2872 mddev->safemode_delay = 1;
2873 if (mddev->safemode_delay < old_delay)
2874 md_safemode_timeout((unsigned long)mddev);
2878 static struct md_sysfs_entry md_safe_delay =
2879 __ATTR(safe_mode_delay, S_IRUGO|S_IWUSR,safe_delay_show, safe_delay_store);
2882 level_show(mddev_t *mddev, char *page)
2884 struct mdk_personality *p = mddev->pers;
2886 return sprintf(page, "%s\n", p->name);
2887 else if (mddev->clevel[0])
2888 return sprintf(page, "%s\n", mddev->clevel);
2889 else if (mddev->level != LEVEL_NONE)
2890 return sprintf(page, "%d\n", mddev->level);
2896 level_store(mddev_t *mddev, const char *buf, size_t len)
2900 struct mdk_personality *pers;
2904 if (mddev->pers == NULL) {
2907 if (len >= sizeof(mddev->clevel))
2909 strncpy(mddev->clevel, buf, len);
2910 if (mddev->clevel[len-1] == '\n')
2912 mddev->clevel[len] = 0;
2913 mddev->level = LEVEL_NONE;
2917 /* request to change the personality. Need to ensure:
2918 * - array is not engaged in resync/recovery/reshape
2919 * - old personality can be suspended
2920 * - new personality will access other array.
2923 if (mddev->sync_thread || mddev->reshape_position != MaxSector)
2926 if (!mddev->pers->quiesce) {
2927 printk(KERN_WARNING "md: %s: %s does not support online personality change\n",
2928 mdname(mddev), mddev->pers->name);
2932 /* Now find the new personality */
2933 if (len == 0 || len >= sizeof(level))
2935 strncpy(level, buf, len);
2936 if (level[len-1] == '\n')
2940 request_module("md-%s", level);
2941 spin_lock(&pers_lock);
2942 pers = find_pers(LEVEL_NONE, level);
2943 if (!pers || !try_module_get(pers->owner)) {
2944 spin_unlock(&pers_lock);
2945 printk(KERN_WARNING "md: personality %s not loaded\n", level);
2948 spin_unlock(&pers_lock);
2950 if (pers == mddev->pers) {
2951 /* Nothing to do! */
2952 module_put(pers->owner);
2955 if (!pers->takeover) {
2956 module_put(pers->owner);
2957 printk(KERN_WARNING "md: %s: %s does not support personality takeover\n",
2958 mdname(mddev), level);
2962 /* ->takeover must set new_* and/or delta_disks
2963 * if it succeeds, and may set them when it fails.
2965 priv = pers->takeover(mddev);
2967 mddev->new_level = mddev->level;
2968 mddev->new_layout = mddev->layout;
2969 mddev->new_chunk_sectors = mddev->chunk_sectors;
2970 mddev->raid_disks -= mddev->delta_disks;
2971 mddev->delta_disks = 0;
2972 module_put(pers->owner);
2973 printk(KERN_WARNING "md: %s: %s would not accept array\n",
2974 mdname(mddev), level);
2975 return PTR_ERR(priv);
2978 /* Looks like we have a winner */
2979 mddev_suspend(mddev);
2980 mddev->pers->stop(mddev);
2981 module_put(mddev->pers->owner);
2982 /* Invalidate devices that are now superfluous */
2983 list_for_each_entry(rdev, &mddev->disks, same_set)
2984 if (rdev->raid_disk >= mddev->raid_disks) {
2985 rdev->raid_disk = -1;
2986 clear_bit(In_sync, &rdev->flags);
2989 mddev->private = priv;
2990 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
2991 mddev->level = mddev->new_level;
2992 mddev->layout = mddev->new_layout;
2993 mddev->chunk_sectors = mddev->new_chunk_sectors;
2994 mddev->delta_disks = 0;
2996 mddev_resume(mddev);
2997 set_bit(MD_CHANGE_DEVS, &mddev->flags);
2998 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2999 md_wakeup_thread(mddev->thread);
3003 static struct md_sysfs_entry md_level =
3004 __ATTR(level, S_IRUGO|S_IWUSR, level_show, level_store);
3008 layout_show(mddev_t *mddev, char *page)
3010 /* just a number, not meaningful for all levels */
3011 if (mddev->reshape_position != MaxSector &&
3012 mddev->layout != mddev->new_layout)
3013 return sprintf(page, "%d (%d)\n",
3014 mddev->new_layout, mddev->layout);
3015 return sprintf(page, "%d\n", mddev->layout);
3019 layout_store(mddev_t *mddev, const char *buf, size_t len)
3022 unsigned long n = simple_strtoul(buf, &e, 10);
3024 if (!*buf || (*e && *e != '\n'))
3029 if (mddev->pers->check_reshape == NULL)
3031 mddev->new_layout = n;
3032 err = mddev->pers->check_reshape(mddev);
3034 mddev->new_layout = mddev->layout;
3038 mddev->new_layout = n;
3039 if (mddev->reshape_position == MaxSector)
3044 static struct md_sysfs_entry md_layout =
3045 __ATTR(layout, S_IRUGO|S_IWUSR, layout_show, layout_store);
3049 raid_disks_show(mddev_t *mddev, char *page)
3051 if (mddev->raid_disks == 0)
3053 if (mddev->reshape_position != MaxSector &&
3054 mddev->delta_disks != 0)
3055 return sprintf(page, "%d (%d)\n", mddev->raid_disks,
3056 mddev->raid_disks - mddev->delta_disks);
3057 return sprintf(page, "%d\n", mddev->raid_disks);
3060 static int update_raid_disks(mddev_t *mddev, int raid_disks);
3063 raid_disks_store(mddev_t *mddev, const char *buf, size_t len)
3067 unsigned long n = simple_strtoul(buf, &e, 10);
3069 if (!*buf || (*e && *e != '\n'))
3073 rv = update_raid_disks(mddev, n);
3074 else if (mddev->reshape_position != MaxSector) {
3075 int olddisks = mddev->raid_disks - mddev->delta_disks;
3076 mddev->delta_disks = n - olddisks;
3077 mddev->raid_disks = n;
3079 mddev->raid_disks = n;
3080 return rv ? rv : len;
3082 static struct md_sysfs_entry md_raid_disks =
3083 __ATTR(raid_disks, S_IRUGO|S_IWUSR, raid_disks_show, raid_disks_store);
3086 chunk_size_show(mddev_t *mddev, char *page)
3088 if (mddev->reshape_position != MaxSector &&
3089 mddev->chunk_sectors != mddev->new_chunk_sectors)
3090 return sprintf(page, "%d (%d)\n",
3091 mddev->new_chunk_sectors << 9,
3092 mddev->chunk_sectors << 9);
3093 return sprintf(page, "%d\n", mddev->chunk_sectors << 9);
3097 chunk_size_store(mddev_t *mddev, const char *buf, size_t len)
3100 unsigned long n = simple_strtoul(buf, &e, 10);
3102 if (!*buf || (*e && *e != '\n'))
3107 if (mddev->pers->check_reshape == NULL)
3109 mddev->new_chunk_sectors = n >> 9;
3110 err = mddev->pers->check_reshape(mddev);
3112 mddev->new_chunk_sectors = mddev->chunk_sectors;
3116 mddev->new_chunk_sectors = n >> 9;
3117 if (mddev->reshape_position == MaxSector)
3118 mddev->chunk_sectors = n >> 9;
3122 static struct md_sysfs_entry md_chunk_size =
3123 __ATTR(chunk_size, S_IRUGO|S_IWUSR, chunk_size_show, chunk_size_store);
3126 resync_start_show(mddev_t *mddev, char *page)
3128 if (mddev->recovery_cp == MaxSector)
3129 return sprintf(page, "none\n");
3130 return sprintf(page, "%llu\n", (unsigned long long)mddev->recovery_cp);
3134 resync_start_store(mddev_t *mddev, const char *buf, size_t len)
3137 unsigned long long n = simple_strtoull(buf, &e, 10);
3141 if (cmd_match(buf, "none"))
3143 else if (!*buf || (*e && *e != '\n'))
3146 mddev->recovery_cp = n;
3149 static struct md_sysfs_entry md_resync_start =
3150 __ATTR(resync_start, S_IRUGO|S_IWUSR, resync_start_show, resync_start_store);
3153 * The array state can be:
3156 * No devices, no size, no level
3157 * Equivalent to STOP_ARRAY ioctl
3159 * May have some settings, but array is not active
3160 * all IO results in error
3161 * When written, doesn't tear down array, but just stops it
3162 * suspended (not supported yet)
3163 * All IO requests will block. The array can be reconfigured.
3164 * Writing this, if accepted, will block until array is quiescent
3166 * no resync can happen. no superblocks get written.
3167 * write requests fail
3169 * like readonly, but behaves like 'clean' on a write request.
3171 * clean - no pending writes, but otherwise active.
3172 * When written to inactive array, starts without resync
3173 * If a write request arrives then
3174 * if metadata is known, mark 'dirty' and switch to 'active'.
3175 * if not known, block and switch to write-pending
3176 * If written to an active array that has pending writes, then fails.
3178 * fully active: IO and resync can be happening.
3179 * When written to inactive array, starts with resync
3182 * clean, but writes are blocked waiting for 'active' to be written.
3185 * like active, but no writes have been seen for a while (100msec).
3188 enum array_state { clear, inactive, suspended, readonly, read_auto, clean, active,
3189 write_pending, active_idle, bad_word};
3190 static char *array_states[] = {
3191 "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active",
3192 "write-pending", "active-idle", NULL };
3194 static int match_word(const char *word, char **list)
3197 for (n=0; list[n]; n++)
3198 if (cmd_match(word, list[n]))
3204 array_state_show(mddev_t *mddev, char *page)
3206 enum array_state st = inactive;
3219 else if (test_bit(MD_CHANGE_CLEAN, &mddev->flags))
3221 else if (mddev->safemode)
3227 if (list_empty(&mddev->disks) &&
3228 mddev->raid_disks == 0 &&
3229 mddev->dev_sectors == 0)
3234 return sprintf(page, "%s\n", array_states[st]);
3237 static int do_md_stop(mddev_t * mddev, int ro, int is_open);
3238 static int do_md_run(mddev_t * mddev);
3239 static int restart_array(mddev_t *mddev);
3242 array_state_store(mddev_t *mddev, const char *buf, size_t len)
3245 enum array_state st = match_word(buf, array_states);
3250 /* stopping an active array */
3251 if (atomic_read(&mddev->openers) > 0)
3253 err = do_md_stop(mddev, 0, 0);
3256 /* stopping an active array */
3258 if (atomic_read(&mddev->openers) > 0)
3260 err = do_md_stop(mddev, 2, 0);
3262 err = 0; /* already inactive */
3265 break; /* not supported yet */
3268 err = do_md_stop(mddev, 1, 0);
3271 set_disk_ro(mddev->gendisk, 1);
3272 err = do_md_run(mddev);
3278 err = do_md_stop(mddev, 1, 0);
3279 else if (mddev->ro == 1)
3280 err = restart_array(mddev);
3283 set_disk_ro(mddev->gendisk, 0);
3287 err = do_md_run(mddev);
3292 restart_array(mddev);
3293 spin_lock_irq(&mddev->write_lock);
3294 if (atomic_read(&mddev->writes_pending) == 0) {
3295 if (mddev->in_sync == 0) {
3297 if (mddev->safemode == 1)
3298 mddev->safemode = 0;
3299 if (mddev->persistent)
3300 set_bit(MD_CHANGE_CLEAN,
3306 spin_unlock_irq(&mddev->write_lock);
3312 restart_array(mddev);
3313 if (mddev->external)
3314 clear_bit(MD_CHANGE_CLEAN, &mddev->flags);
3315 wake_up(&mddev->sb_wait);
3319 set_disk_ro(mddev->gendisk, 0);
3320 err = do_md_run(mddev);
3325 /* these cannot be set */
3331 sysfs_notify_dirent(mddev->sysfs_state);
3335 static struct md_sysfs_entry md_array_state =
3336 __ATTR(array_state, S_IRUGO|S_IWUSR, array_state_show, array_state_store);
3339 max_corrected_read_errors_show(mddev_t *mddev, char *page) {
3340 return sprintf(page, "%d\n",
3341 atomic_read(&mddev->max_corr_read_errors));
3345 max_corrected_read_errors_store(mddev_t *mddev, const char *buf, size_t len)
3348 unsigned long n = simple_strtoul(buf, &e, 10);
3350 if (*buf && (*e == 0 || *e == '\n')) {
3351 atomic_set(&mddev->max_corr_read_errors, n);
3357 static struct md_sysfs_entry max_corr_read_errors =
3358 __ATTR(max_read_errors, S_IRUGO|S_IWUSR, max_corrected_read_errors_show,
3359 max_corrected_read_errors_store);
3362 null_show(mddev_t *mddev, char *page)
3368 new_dev_store(mddev_t *mddev, const char *buf, size_t len)
3370 /* buf must be %d:%d\n? giving major and minor numbers */
3371 /* The new device is added to the array.
3372 * If the array has a persistent superblock, we read the
3373 * superblock to initialise info and check validity.
3374 * Otherwise, only checking done is that in bind_rdev_to_array,
3375 * which mainly checks size.
3378 int major = simple_strtoul(buf, &e, 10);
3384 if (!*buf || *e != ':' || !e[1] || e[1] == '\n')
3386 minor = simple_strtoul(e+1, &e, 10);
3387 if (*e && *e != '\n')
3389 dev = MKDEV(major, minor);
3390 if (major != MAJOR(dev) ||
3391 minor != MINOR(dev))
3395 if (mddev->persistent) {
3396 rdev = md_import_device(dev, mddev->major_version,
3397 mddev->minor_version);
3398 if (!IS_ERR(rdev) && !list_empty(&mddev->disks)) {
3399 mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
3400 mdk_rdev_t, same_set);
3401 err = super_types[mddev->major_version]
3402 .load_super(rdev, rdev0, mddev->minor_version);
3406 } else if (mddev->external)
3407 rdev = md_import_device(dev, -2, -1);
3409 rdev = md_import_device(dev, -1, -1);
3412 return PTR_ERR(rdev);
3413 err = bind_rdev_to_array(rdev, mddev);
3417 return err ? err : len;
3420 static struct md_sysfs_entry md_new_device =
3421 __ATTR(new_dev, S_IWUSR, null_show, new_dev_store);
3424 bitmap_store(mddev_t *mddev, const char *buf, size_t len)
3427 unsigned long chunk, end_chunk;
3431 /* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */
3433 chunk = end_chunk = simple_strtoul(buf, &end, 0);
3434 if (buf == end) break;
3435 if (*end == '-') { /* range */
3437 end_chunk = simple_strtoul(buf, &end, 0);
3438 if (buf == end) break;
3440 if (*end && !isspace(*end)) break;
3441 bitmap_dirty_bits(mddev->bitmap, chunk, end_chunk);
3443 while (isspace(*buf)) buf++;
3445 bitmap_unplug(mddev->bitmap); /* flush the bits to disk */
3450 static struct md_sysfs_entry md_bitmap =
3451 __ATTR(bitmap_set_bits, S_IWUSR, null_show, bitmap_store);
3454 size_show(mddev_t *mddev, char *page)
3456 return sprintf(page, "%llu\n",
3457 (unsigned long long)mddev->dev_sectors / 2);
3460 static int update_size(mddev_t *mddev, sector_t num_sectors);
3463 size_store(mddev_t *mddev, const char *buf, size_t len)
3465 /* If array is inactive, we can reduce the component size, but
3466 * not increase it (except from 0).
3467 * If array is active, we can try an on-line resize
3470 int err = strict_blocks_to_sectors(buf, §ors);
3475 err = update_size(mddev, sectors);
3476 md_update_sb(mddev, 1);
3478 if (mddev->dev_sectors == 0 ||
3479 mddev->dev_sectors > sectors)
3480 mddev->dev_sectors = sectors;
3484 return err ? err : len;
3487 static struct md_sysfs_entry md_size =
3488 __ATTR(component_size, S_IRUGO|S_IWUSR, size_show, size_store);
3493 * 'none' for arrays with no metadata (good luck...)
3494 * 'external' for arrays with externally managed metadata,
3495 * or N.M for internally known formats
3498 metadata_show(mddev_t *mddev, char *page)
3500 if (mddev->persistent)
3501 return sprintf(page, "%d.%d\n",
3502 mddev->major_version, mddev->minor_version);
3503 else if (mddev->external)
3504 return sprintf(page, "external:%s\n", mddev->metadata_type);
3506 return sprintf(page, "none\n");
3510 metadata_store(mddev_t *mddev, const char *buf, size_t len)
3514 /* Changing the details of 'external' metadata is
3515 * always permitted. Otherwise there must be
3516 * no devices attached to the array.
3518 if (mddev->external && strncmp(buf, "external:", 9) == 0)
3520 else if (!list_empty(&mddev->disks))
3523 if (cmd_match(buf, "none")) {
3524 mddev->persistent = 0;
3525 mddev->external = 0;
3526 mddev->major_version = 0;
3527 mddev->minor_version = 90;
3530 if (strncmp(buf, "external:", 9) == 0) {
3531 size_t namelen = len-9;
3532 if (namelen >= sizeof(mddev->metadata_type))
3533 namelen = sizeof(mddev->metadata_type)-1;
3534 strncpy(mddev->metadata_type, buf+9, namelen);
3535 mddev->metadata_type[namelen] = 0;
3536 if (namelen && mddev->metadata_type[namelen-1] == '\n')
3537 mddev->metadata_type[--namelen] = 0;
3538 mddev->persistent = 0;
3539 mddev->external = 1;
3540 mddev->major_version = 0;
3541 mddev->minor_version = 90;
3544 major = simple_strtoul(buf, &e, 10);
3545 if (e==buf || *e != '.')
3548 minor = simple_strtoul(buf, &e, 10);
3549 if (e==buf || (*e && *e != '\n') )
3551 if (major >= ARRAY_SIZE(super_types) || super_types[major].name == NULL)
3553 mddev->major_version = major;
3554 mddev->minor_version = minor;
3555 mddev->persistent = 1;
3556 mddev->external = 0;
3560 static struct md_sysfs_entry md_metadata =
3561 __ATTR(metadata_version, S_IRUGO|S_IWUSR, metadata_show, metadata_store);
3564 action_show(mddev_t *mddev, char *page)
3566 char *type = "idle";
3567 if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
3569 else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
3570 (!mddev->ro && test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))) {
3571 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
3573 else if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
3574 if (!test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
3576 else if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
3580 } else if (test_bit(MD_RECOVERY_RECOVER, &mddev->recovery))
3583 return sprintf(page, "%s\n", type);
3587 action_store(mddev_t *mddev, const char *page, size_t len)
3589 if (!mddev->pers || !mddev->pers->sync_request)
3592 if (cmd_match(page, "frozen"))
3593 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
3595 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
3597 if (cmd_match(page, "idle") || cmd_match(page, "frozen")) {
3598 if (mddev->sync_thread) {
3599 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
3600 md_unregister_thread(mddev->sync_thread);
3601 mddev->sync_thread = NULL;
3602 mddev->recovery = 0;
3604 } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
3605 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
3607 else if (cmd_match(page, "resync"))
3608 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3609 else if (cmd_match(page, "recover")) {
3610 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
3611 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3612 } else if (cmd_match(page, "reshape")) {
3614 if (mddev->pers->start_reshape == NULL)
3616 err = mddev->pers->start_reshape(mddev);
3619 sysfs_notify(&mddev->kobj, NULL, "degraded");
3621 if (cmd_match(page, "check"))
3622 set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
3623 else if (!cmd_match(page, "repair"))
3625 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
3626 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
3628 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3629 md_wakeup_thread(mddev->thread);
3630 sysfs_notify_dirent(mddev->sysfs_action);
3635 mismatch_cnt_show(mddev_t *mddev, char *page)
3637 return sprintf(page, "%llu\n",
3638 (unsigned long long) mddev->resync_mismatches);
3641 static struct md_sysfs_entry md_scan_mode =
3642 __ATTR(sync_action, S_IRUGO|S_IWUSR, action_show, action_store);
3645 static struct md_sysfs_entry md_mismatches = __ATTR_RO(mismatch_cnt);
3648 sync_min_show(mddev_t *mddev, char *page)
3650 return sprintf(page, "%d (%s)\n", speed_min(mddev),
3651 mddev->sync_speed_min ? "local": "system");
3655 sync_min_store(mddev_t *mddev, const char *buf, size_t len)
3659 if (strncmp(buf, "system", 6)==0) {
3660 mddev->sync_speed_min = 0;
3663 min = simple_strtoul(buf, &e, 10);
3664 if (buf == e || (*e && *e != '\n') || min <= 0)
3666 mddev->sync_speed_min = min;
3670 static struct md_sysfs_entry md_sync_min =
3671 __ATTR(sync_speed_min, S_IRUGO|S_IWUSR, sync_min_show, sync_min_store);
3674 sync_max_show(mddev_t *mddev, char *page)
3676 return sprintf(page, "%d (%s)\n", speed_max(mddev),
3677 mddev->sync_speed_max ? "local": "system");
3681 sync_max_store(mddev_t *mddev, const char *buf, size_t len)
3685 if (strncmp(buf, "system", 6)==0) {
3686 mddev->sync_speed_max = 0;
3689 max = simple_strtoul(buf, &e, 10);
3690 if (buf == e || (*e && *e != '\n') || max <= 0)
3692 mddev->sync_speed_max = max;
3696 static struct md_sysfs_entry md_sync_max =
3697 __ATTR(sync_speed_max, S_IRUGO|S_IWUSR, sync_max_show, sync_max_store);
3700 degraded_show(mddev_t *mddev, char *page)
3702 return sprintf(page, "%d\n", mddev->degraded);
3704 static struct md_sysfs_entry md_degraded = __ATTR_RO(degraded);
3707 sync_force_parallel_show(mddev_t *mddev, char *page)
3709 return sprintf(page, "%d\n", mddev->parallel_resync);
3713 sync_force_parallel_store(mddev_t *mddev, const char *buf, size_t len)
3717 if (strict_strtol(buf, 10, &n))
3720 if (n != 0 && n != 1)
3723 mddev->parallel_resync = n;
3725 if (mddev->sync_thread)
3726 wake_up(&resync_wait);
3731 /* force parallel resync, even with shared block devices */
3732 static struct md_sysfs_entry md_sync_force_parallel =
3733 __ATTR(sync_force_parallel, S_IRUGO|S_IWUSR,
3734 sync_force_parallel_show, sync_force_parallel_store);
3737 sync_speed_show(mddev_t *mddev, char *page)
3739 unsigned long resync, dt, db;
3740 if (mddev->curr_resync == 0)
3741 return sprintf(page, "none\n");
3742 resync = mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active);
3743 dt = (jiffies - mddev->resync_mark) / HZ;
3745 db = resync - mddev->resync_mark_cnt;
3746 return sprintf(page, "%lu\n", db/dt/2); /* K/sec */
3749 static struct md_sysfs_entry md_sync_speed = __ATTR_RO(sync_speed);
3752 sync_completed_show(mddev_t *mddev, char *page)
3754 unsigned long max_sectors, resync;
3756 if (!test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
3757 return sprintf(page, "none\n");
3759 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
3760 max_sectors = mddev->resync_max_sectors;
3762 max_sectors = mddev->dev_sectors;
3764 resync = mddev->curr_resync_completed;
3765 return sprintf(page, "%lu / %lu\n", resync, max_sectors);
3768 static struct md_sysfs_entry md_sync_completed = __ATTR_RO(sync_completed);
3771 min_sync_show(mddev_t *mddev, char *page)
3773 return sprintf(page, "%llu\n",
3774 (unsigned long long)mddev->resync_min);
3777 min_sync_store(mddev_t *mddev, const char *buf, size_t len)
3779 unsigned long long min;
3780 if (strict_strtoull(buf, 10, &min))
3782 if (min > mddev->resync_max)
3784 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
3787 /* Must be a multiple of chunk_size */
3788 if (mddev->chunk_sectors) {
3789 sector_t temp = min;
3790 if (sector_div(temp, mddev->chunk_sectors))
3793 mddev->resync_min = min;
3798 static struct md_sysfs_entry md_min_sync =
3799 __ATTR(sync_min, S_IRUGO|S_IWUSR, min_sync_show, min_sync_store);
3802 max_sync_show(mddev_t *mddev, char *page)
3804 if (mddev->resync_max == MaxSector)
3805 return sprintf(page, "max\n");
3807 return sprintf(page, "%llu\n",
3808 (unsigned long long)mddev->resync_max);
3811 max_sync_store(mddev_t *mddev, const char *buf, size_t len)
3813 if (strncmp(buf, "max", 3) == 0)
3814 mddev->resync_max = MaxSector;
3816 unsigned long long max;
3817 if (strict_strtoull(buf, 10, &max))
3819 if (max < mddev->resync_min)
3821 if (max < mddev->resync_max &&
3823 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
3826 /* Must be a multiple of chunk_size */
3827 if (mddev->chunk_sectors) {
3828 sector_t temp = max;
3829 if (sector_div(temp, mddev->chunk_sectors))
3832 mddev->resync_max = max;
3834 wake_up(&mddev->recovery_wait);
3838 static struct md_sysfs_entry md_max_sync =
3839 __ATTR(sync_max, S_IRUGO|S_IWUSR, max_sync_show, max_sync_store);
3842 suspend_lo_show(mddev_t *mddev, char *page)
3844 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_lo);
3848 suspend_lo_store(mddev_t *mddev, const char *buf, size_t len)
3851 unsigned long long new = simple_strtoull(buf, &e, 10);
3853 if (mddev->pers == NULL ||
3854 mddev->pers->quiesce == NULL)
3856 if (buf == e || (*e && *e != '\n'))
3858 if (new >= mddev->suspend_hi ||
3859 (new > mddev->suspend_lo && new < mddev->suspend_hi)) {
3860 mddev->suspend_lo = new;
3861 mddev->pers->quiesce(mddev, 2);
3866 static struct md_sysfs_entry md_suspend_lo =
3867 __ATTR(suspend_lo, S_IRUGO|S_IWUSR, suspend_lo_show, suspend_lo_store);
3871 suspend_hi_show(mddev_t *mddev, char *page)
3873 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_hi);
3877 suspend_hi_store(mddev_t *mddev, const char *buf, size_t len)
3880 unsigned long long new = simple_strtoull(buf, &e, 10);
3882 if (mddev->pers == NULL ||
3883 mddev->pers->quiesce == NULL)
3885 if (buf == e || (*e && *e != '\n'))
3887 if ((new <= mddev->suspend_lo && mddev->suspend_lo >= mddev->suspend_hi) ||
3888 (new > mddev->suspend_lo && new > mddev->suspend_hi)) {
3889 mddev->suspend_hi = new;
3890 mddev->pers->quiesce(mddev, 1);
3891 mddev->pers->quiesce(mddev, 0);
3896 static struct md_sysfs_entry md_suspend_hi =
3897 __ATTR(suspend_hi, S_IRUGO|S_IWUSR, suspend_hi_show, suspend_hi_store);
3900 reshape_position_show(mddev_t *mddev, char *page)
3902 if (mddev->reshape_position != MaxSector)
3903 return sprintf(page, "%llu\n",
3904 (unsigned long long)mddev->reshape_position);
3905 strcpy(page, "none\n");
3910 reshape_position_store(mddev_t *mddev, const char *buf, size_t len)
3913 unsigned long long new = simple_strtoull(buf, &e, 10);
3916 if (buf == e || (*e && *e != '\n'))
3918 mddev->reshape_position = new;
3919 mddev->delta_disks = 0;
3920 mddev->new_level = mddev->level;
3921 mddev->new_layout = mddev->layout;
3922 mddev->new_chunk_sectors = mddev->chunk_sectors;
3926 static struct md_sysfs_entry md_reshape_position =
3927 __ATTR(reshape_position, S_IRUGO|S_IWUSR, reshape_position_show,
3928 reshape_position_store);
3931 array_size_show(mddev_t *mddev, char *page)
3933 if (mddev->external_size)
3934 return sprintf(page, "%llu\n",
3935 (unsigned long long)mddev->array_sectors/2);
3937 return sprintf(page, "default\n");
3941 array_size_store(mddev_t *mddev, const char *buf, size_t len)
3945 if (strncmp(buf, "default", 7) == 0) {
3947 sectors = mddev->pers->size(mddev, 0, 0);
3949 sectors = mddev->array_sectors;
3951 mddev->external_size = 0;
3953 if (strict_blocks_to_sectors(buf, §ors) < 0)
3955 if (mddev->pers && mddev->pers->size(mddev, 0, 0) < sectors)
3958 mddev->external_size = 1;
3961 mddev->array_sectors = sectors;
3962 set_capacity(mddev->gendisk, mddev->array_sectors);
3964 revalidate_disk(mddev->gendisk);
3969 static struct md_sysfs_entry md_array_size =
3970 __ATTR(array_size, S_IRUGO|S_IWUSR, array_size_show,
3973 static struct attribute *md_default_attrs[] = {
3976 &md_raid_disks.attr,
3977 &md_chunk_size.attr,
3979 &md_resync_start.attr,
3981 &md_new_device.attr,
3982 &md_safe_delay.attr,
3983 &md_array_state.attr,
3984 &md_reshape_position.attr,
3985 &md_array_size.attr,
3986 &max_corr_read_errors.attr,
3990 static struct attribute *md_redundancy_attrs[] = {
3992 &md_mismatches.attr,
3995 &md_sync_speed.attr,
3996 &md_sync_force_parallel.attr,
3997 &md_sync_completed.attr,
4000 &md_suspend_lo.attr,
4001 &md_suspend_hi.attr,
4006 static struct attribute_group md_redundancy_group = {
4008 .attrs = md_redundancy_attrs,
4013 md_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
4015 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
4016 mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
4021 rv = mddev_lock(mddev);
4023 rv = entry->show(mddev, page);
4024 mddev_unlock(mddev);
4030 md_attr_store(struct kobject *kobj, struct attribute *attr,
4031 const char *page, size_t length)
4033 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
4034 mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
4039 if (!capable(CAP_SYS_ADMIN))
4041 rv = mddev_lock(mddev);
4042 if (mddev->hold_active == UNTIL_IOCTL)
4043 mddev->hold_active = 0;
4045 rv = entry->store(mddev, page, length);
4046 mddev_unlock(mddev);
4051 static void md_free(struct kobject *ko)
4053 mddev_t *mddev = container_of(ko, mddev_t, kobj);
4055 if (mddev->sysfs_state)
4056 sysfs_put(mddev->sysfs_state);
4058 if (mddev->gendisk) {
4059 del_gendisk(mddev->gendisk);
4060 put_disk(mddev->gendisk);
4063 blk_cleanup_queue(mddev->queue);
4068 static struct sysfs_ops md_sysfs_ops = {
4069 .show = md_attr_show,
4070 .store = md_attr_store,
4072 static struct kobj_type md_ktype = {
4074 .sysfs_ops = &md_sysfs_ops,
4075 .default_attrs = md_default_attrs,
4080 static void mddev_delayed_delete(struct work_struct *ws)
4082 mddev_t *mddev = container_of(ws, mddev_t, del_work);
4084 if (mddev->private == &md_redundancy_group) {
4085 sysfs_remove_group(&mddev->kobj, &md_redundancy_group);
4086 if (mddev->sysfs_action)
4087 sysfs_put(mddev->sysfs_action);
4088 mddev->sysfs_action = NULL;
4089 mddev->private = NULL;
4091 sysfs_remove_group(&mddev->kobj, &md_bitmap_group);
4092 kobject_del(&mddev->kobj);
4093 kobject_put(&mddev->kobj);
4096 static int md_alloc(dev_t dev, char *name)
4098 static DEFINE_MUTEX(disks_mutex);
4099 mddev_t *mddev = mddev_find(dev);
4100 struct gendisk *disk;
4109 partitioned = (MAJOR(mddev->unit) != MD_MAJOR);
4110 shift = partitioned ? MdpMinorShift : 0;
4111 unit = MINOR(mddev->unit) >> shift;
4113 /* wait for any previous instance if this device
4114 * to be completed removed (mddev_delayed_delete).
4116 flush_scheduled_work();
4118 mutex_lock(&disks_mutex);
4124 /* Need to ensure that 'name' is not a duplicate.
4127 spin_lock(&all_mddevs_lock);
4129 list_for_each_entry(mddev2, &all_mddevs, all_mddevs)
4130 if (mddev2->gendisk &&
4131 strcmp(mddev2->gendisk->disk_name, name) == 0) {
4132 spin_unlock(&all_mddevs_lock);
4135 spin_unlock(&all_mddevs_lock);
4139 mddev->queue = blk_alloc_queue(GFP_KERNEL);
4142 mddev->queue->queuedata = mddev;
4144 /* Can be unlocked because the queue is new: no concurrency */
4145 queue_flag_set_unlocked(QUEUE_FLAG_CLUSTER, mddev->queue);
4147 blk_queue_make_request(mddev->queue, md_make_request);
4149 disk = alloc_disk(1 << shift);
4151 blk_cleanup_queue(mddev->queue);
4152 mddev->queue = NULL;
4155 disk->major = MAJOR(mddev->unit);
4156 disk->first_minor = unit << shift;
4158 strcpy(disk->disk_name, name);
4159 else if (partitioned)
4160 sprintf(disk->disk_name, "md_d%d", unit);
4162 sprintf(disk->disk_name, "md%d", unit);
4163 disk->fops = &md_fops;
4164 disk->private_data = mddev;
4165 disk->queue = mddev->queue;
4166 /* Allow extended partitions. This makes the
4167 * 'mdp' device redundant, but we can't really
4170 disk->flags |= GENHD_FL_EXT_DEVT;
4172 mddev->gendisk = disk;
4173 error = kobject_init_and_add(&mddev->kobj, &md_ktype,
4174 &disk_to_dev(disk)->kobj, "%s", "md");
4176 /* This isn't possible, but as kobject_init_and_add is marked
4177 * __must_check, we must do something with the result
4179 printk(KERN_WARNING "md: cannot register %s/md - name in use\n",
4183 if (sysfs_create_group(&mddev->kobj, &md_bitmap_group))
4184 printk(KERN_DEBUG "pointless warning\n");
4186 mutex_unlock(&disks_mutex);
4188 kobject_uevent(&mddev->kobj, KOBJ_ADD);
4189 mddev->sysfs_state = sysfs_get_dirent(mddev->kobj.sd, "array_state");
4195 static struct kobject *md_probe(dev_t dev, int *part, void *data)
4197 md_alloc(dev, NULL);
4201 static int add_named_array(const char *val, struct kernel_param *kp)
4203 /* val must be "md_*" where * is not all digits.
4204 * We allocate an array with a large free minor number, and
4205 * set the name to val. val must not already be an active name.
4207 int len = strlen(val);
4208 char buf[DISK_NAME_LEN];
4210 while (len && val[len-1] == '\n')
4212 if (len >= DISK_NAME_LEN)
4214 strlcpy(buf, val, len+1);
4215 if (strncmp(buf, "md_", 3) != 0)
4217 return md_alloc(0, buf);
4220 static void md_safemode_timeout(unsigned long data)
4222 mddev_t *mddev = (mddev_t *) data;
4224 if (!atomic_read(&mddev->writes_pending)) {
4225 mddev->safemode = 1;
4226 if (mddev->external)
4227 sysfs_notify_dirent(mddev->sysfs_state);
4229 md_wakeup_thread(mddev->thread);
4232 static int start_dirty_degraded;
4234 static int do_md_run(mddev_t * mddev)
4238 struct gendisk *disk;
4239 struct mdk_personality *pers;
4241 if (list_empty(&mddev->disks))
4242 /* cannot run an array with no devices.. */
4249 * Analyze all RAID superblock(s)
4251 if (!mddev->raid_disks) {
4252 if (!mddev->persistent)
4257 if (mddev->level != LEVEL_NONE)
4258 request_module("md-level-%d", mddev->level);
4259 else if (mddev->clevel[0])
4260 request_module("md-%s", mddev->clevel);
4263 * Drop all container device buffers, from now on
4264 * the only valid external interface is through the md
4267 list_for_each_entry(rdev, &mddev->disks, same_set) {
4268 if (test_bit(Faulty, &rdev->flags))
4270 sync_blockdev(rdev->bdev);
4271 invalidate_bdev(rdev->bdev);
4273 /* perform some consistency tests on the device.
4274 * We don't want the data to overlap the metadata,
4275 * Internal Bitmap issues have been handled elsewhere.
4277 if (rdev->data_offset < rdev->sb_start) {
4278 if (mddev->dev_sectors &&
4279 rdev->data_offset + mddev->dev_sectors
4281 printk("md: %s: data overlaps metadata\n",
4286 if (rdev->sb_start + rdev->sb_size/512
4287 > rdev->data_offset) {
4288 printk("md: %s: metadata overlaps data\n",
4293 sysfs_notify_dirent(rdev->sysfs_state);
4296 md_probe(mddev->unit, NULL, NULL);
4297 disk = mddev->gendisk;
4301 spin_lock(&pers_lock);
4302 pers = find_pers(mddev->level, mddev->clevel);
4303 if (!pers || !try_module_get(pers->owner)) {
4304 spin_unlock(&pers_lock);
4305 if (mddev->level != LEVEL_NONE)
4306 printk(KERN_WARNING "md: personality for level %d is not loaded!\n",
4309 printk(KERN_WARNING "md: personality for level %s is not loaded!\n",
4314 spin_unlock(&pers_lock);
4315 if (mddev->level != pers->level) {
4316 mddev->level = pers->level;
4317 mddev->new_level = pers->level;
4319 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
4321 if (mddev->reshape_position != MaxSector &&
4322 pers->start_reshape == NULL) {
4323 /* This personality cannot handle reshaping... */
4325 module_put(pers->owner);
4329 if (pers->sync_request) {
4330 /* Warn if this is a potentially silly
4333 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
4337 list_for_each_entry(rdev, &mddev->disks, same_set)
4338 list_for_each_entry(rdev2, &mddev->disks, same_set) {
4340 rdev->bdev->bd_contains ==
4341 rdev2->bdev->bd_contains) {
4343 "%s: WARNING: %s appears to be"
4344 " on the same physical disk as"
4347 bdevname(rdev->bdev,b),
4348 bdevname(rdev2->bdev,b2));
4355 "True protection against single-disk"
4356 " failure might be compromised.\n");
4359 mddev->recovery = 0;
4360 /* may be over-ridden by personality */
4361 mddev->resync_max_sectors = mddev->dev_sectors;
4363 mddev->barriers_work = 1;
4364 mddev->ok_start_degraded = start_dirty_degraded;
4367 mddev->ro = 2; /* read-only, but switch on first write */
4369 err = mddev->pers->run(mddev);
4371 printk(KERN_ERR "md: pers->run() failed ...\n");
4372 else if (mddev->pers->size(mddev, 0, 0) < mddev->array_sectors) {
4373 WARN_ONCE(!mddev->external_size, "%s: default size too small,"
4374 " but 'external_size' not in effect?\n", __func__);
4376 "md: invalid array_size %llu > default size %llu\n",
4377 (unsigned long long)mddev->array_sectors / 2,
4378 (unsigned long long)mddev->pers->size(mddev, 0, 0) / 2);
4380 mddev->pers->stop(mddev);
4382 if (err == 0 && mddev->pers->sync_request) {
4383 err = bitmap_create(mddev);
4385 printk(KERN_ERR "%s: failed to create bitmap (%d)\n",
4386 mdname(mddev), err);
4387 mddev->pers->stop(mddev);
4391 module_put(mddev->pers->owner);
4393 bitmap_destroy(mddev);
4396 if (mddev->pers->sync_request) {
4397 if (sysfs_create_group(&mddev->kobj, &md_redundancy_group))
4399 "md: cannot register extra attributes for %s\n",
4401 mddev->sysfs_action = sysfs_get_dirent(mddev->kobj.sd, "sync_action");
4402 } else if (mddev->ro == 2) /* auto-readonly not meaningful */
4405 atomic_set(&mddev->writes_pending,0);
4406 atomic_set(&mddev->max_corr_read_errors,
4407 MD_DEFAULT_MAX_CORRECTED_READ_ERRORS);
4408 mddev->safemode = 0;
4409 mddev->safemode_timer.function = md_safemode_timeout;
4410 mddev->safemode_timer.data = (unsigned long) mddev;
4411 mddev->safemode_delay = (200 * HZ)/1000 +1; /* 200 msec delay */
4414 list_for_each_entry(rdev, &mddev->disks, same_set)
4415 if (rdev->raid_disk >= 0) {
4417 sprintf(nm, "rd%d", rdev->raid_disk);
4418 if (sysfs_create_link(&mddev->kobj, &rdev->kobj, nm))
4419 printk("md: cannot register %s for %s\n",
4423 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4426 md_update_sb(mddev, 0);
4428 set_capacity(disk, mddev->array_sectors);
4430 /* If there is a partially-recovered drive we need to
4431 * start recovery here. If we leave it to md_check_recovery,
4432 * it will remove the drives and not do the right thing
4434 if (mddev->degraded && !mddev->sync_thread) {
4436 list_for_each_entry(rdev, &mddev->disks, same_set)
4437 if (rdev->raid_disk >= 0 &&
4438 !test_bit(In_sync, &rdev->flags) &&
4439 !test_bit(Faulty, &rdev->flags))
4440 /* complete an interrupted recovery */
4442 if (spares && mddev->pers->sync_request) {
4443 mddev->recovery = 0;
4444 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
4445 mddev->sync_thread = md_register_thread(md_do_sync,
4448 if (!mddev->sync_thread) {
4449 printk(KERN_ERR "%s: could not start resync"
4452 /* leave the spares where they are, it shouldn't hurt */
4453 mddev->recovery = 0;
4457 md_wakeup_thread(mddev->thread);
4458 md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
4460 revalidate_disk(mddev->gendisk);
4462 md_new_event(mddev);
4463 sysfs_notify_dirent(mddev->sysfs_state);
4464 if (mddev->sysfs_action)
4465 sysfs_notify_dirent(mddev->sysfs_action);
4466 sysfs_notify(&mddev->kobj, NULL, "degraded");
4467 kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
4471 static int restart_array(mddev_t *mddev)
4473 struct gendisk *disk = mddev->gendisk;
4475 /* Complain if it has no devices */
4476 if (list_empty(&mddev->disks))
4482 mddev->safemode = 0;
4484 set_disk_ro(disk, 0);
4485 printk(KERN_INFO "md: %s switched to read-write mode.\n",
4487 /* Kick recovery or resync if necessary */
4488 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4489 md_wakeup_thread(mddev->thread);
4490 md_wakeup_thread(mddev->sync_thread);
4491 sysfs_notify_dirent(mddev->sysfs_state);
4495 /* similar to deny_write_access, but accounts for our holding a reference
4496 * to the file ourselves */
4497 static int deny_bitmap_write_access(struct file * file)
4499 struct inode *inode = file->f_mapping->host;
4501 spin_lock(&inode->i_lock);
4502 if (atomic_read(&inode->i_writecount) > 1) {
4503 spin_unlock(&inode->i_lock);
4506 atomic_set(&inode->i_writecount, -1);
4507 spin_unlock(&inode->i_lock);
4512 void restore_bitmap_write_access(struct file *file)
4514 struct inode *inode = file->f_mapping->host;
4516 spin_lock(&inode->i_lock);
4517 atomic_set(&inode->i_writecount, 1);
4518 spin_unlock(&inode->i_lock);
4522 * 0 - completely stop and dis-assemble array
4523 * 1 - switch to readonly
4524 * 2 - stop but do not disassemble array
4526 static int do_md_stop(mddev_t * mddev, int mode, int is_open)
4529 struct gendisk *disk = mddev->gendisk;
4532 mutex_lock(&mddev->open_mutex);
4533 if (atomic_read(&mddev->openers) > is_open) {
4534 printk("md: %s still in use.\n",mdname(mddev));
4536 } else if (mddev->pers) {
4538 if (mddev->sync_thread) {
4539 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4540 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
4541 md_unregister_thread(mddev->sync_thread);
4542 mddev->sync_thread = NULL;
4545 del_timer_sync(&mddev->safemode_timer);
4548 case 1: /* readonly */
4554 case 0: /* disassemble */
4556 bitmap_flush(mddev);
4557 md_super_wait(mddev);
4559 set_disk_ro(disk, 0);
4561 mddev->pers->stop(mddev);
4562 mddev->queue->merge_bvec_fn = NULL;
4563 mddev->queue->unplug_fn = NULL;
4564 mddev->queue->backing_dev_info.congested_fn = NULL;
4565 module_put(mddev->pers->owner);
4566 if (mddev->pers->sync_request)
4567 mddev->private = &md_redundancy_group;
4569 /* tell userspace to handle 'inactive' */
4570 sysfs_notify_dirent(mddev->sysfs_state);
4572 list_for_each_entry(rdev, &mddev->disks, same_set)
4573 if (rdev->raid_disk >= 0) {
4575 sprintf(nm, "rd%d", rdev->raid_disk);
4576 sysfs_remove_link(&mddev->kobj, nm);
4579 set_capacity(disk, 0);
4585 if (!mddev->in_sync || mddev->flags) {
4586 /* mark array as shutdown cleanly */
4588 md_update_sb(mddev, 1);
4591 set_disk_ro(disk, 1);
4592 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4596 mutex_unlock(&mddev->open_mutex);
4600 * Free resources if final stop
4604 printk(KERN_INFO "md: %s stopped.\n", mdname(mddev));
4606 bitmap_destroy(mddev);
4607 if (mddev->bitmap_info.file) {
4608 restore_bitmap_write_access(mddev->bitmap_info.file);
4609 fput(mddev->bitmap_info.file);
4610 mddev->bitmap_info.file = NULL;
4612 mddev->bitmap_info.offset = 0;
4614 /* make sure all md_delayed_delete calls have finished */
4615 flush_scheduled_work();
4617 export_array(mddev);
4619 mddev->array_sectors = 0;
4620 mddev->external_size = 0;
4621 mddev->dev_sectors = 0;
4622 mddev->raid_disks = 0;
4623 mddev->recovery_cp = 0;
4624 mddev->resync_min = 0;
4625 mddev->resync_max = MaxSector;
4626 mddev->reshape_position = MaxSector;
4627 mddev->external = 0;
4628 mddev->persistent = 0;
4629 mddev->level = LEVEL_NONE;
4630 mddev->clevel[0] = 0;
4633 mddev->metadata_type[0] = 0;
4634 mddev->chunk_sectors = 0;
4635 mddev->ctime = mddev->utime = 0;
4637 mddev->max_disks = 0;
4639 mddev->delta_disks = 0;
4640 mddev->new_level = LEVEL_NONE;
4641 mddev->new_layout = 0;
4642 mddev->new_chunk_sectors = 0;
4643 mddev->curr_resync = 0;
4644 mddev->resync_mismatches = 0;
4645 mddev->suspend_lo = mddev->suspend_hi = 0;
4646 mddev->sync_speed_min = mddev->sync_speed_max = 0;
4647 mddev->recovery = 0;
4650 mddev->degraded = 0;
4651 mddev->barriers_work = 0;
4652 mddev->safemode = 0;
4653 mddev->bitmap_info.offset = 0;
4654 mddev->bitmap_info.default_offset = 0;
4655 mddev->bitmap_info.chunksize = 0;
4656 mddev->bitmap_info.daemon_sleep = 0;
4657 mddev->bitmap_info.max_write_behind = 0;
4658 kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
4659 if (mddev->hold_active == UNTIL_STOP)
4660 mddev->hold_active = 0;
4662 } else if (mddev->pers)
4663 printk(KERN_INFO "md: %s switched to read-only mode.\n",
4666 blk_integrity_unregister(disk);
4667 md_new_event(mddev);
4668 sysfs_notify_dirent(mddev->sysfs_state);
4673 static void autorun_array(mddev_t *mddev)
4678 if (list_empty(&mddev->disks))
4681 printk(KERN_INFO "md: running: ");
4683 list_for_each_entry(rdev, &mddev->disks, same_set) {
4684 char b[BDEVNAME_SIZE];
4685 printk("<%s>", bdevname(rdev->bdev,b));
4689 err = do_md_run(mddev);
4691 printk(KERN_WARNING "md: do_md_run() returned %d\n", err);
4692 do_md_stop(mddev, 0, 0);
4697 * lets try to run arrays based on all disks that have arrived
4698 * until now. (those are in pending_raid_disks)
4700 * the method: pick the first pending disk, collect all disks with
4701 * the same UUID, remove all from the pending list and put them into
4702 * the 'same_array' list. Then order this list based on superblock
4703 * update time (freshest comes first), kick out 'old' disks and
4704 * compare superblocks. If everything's fine then run it.
4706 * If "unit" is allocated, then bump its reference count
4708 static void autorun_devices(int part)
4710 mdk_rdev_t *rdev0, *rdev, *tmp;
4712 char b[BDEVNAME_SIZE];
4714 printk(KERN_INFO "md: autorun ...\n");
4715 while (!list_empty(&pending_raid_disks)) {
4718 LIST_HEAD(candidates);
4719 rdev0 = list_entry(pending_raid_disks.next,
4720 mdk_rdev_t, same_set);
4722 printk(KERN_INFO "md: considering %s ...\n",
4723 bdevname(rdev0->bdev,b));
4724 INIT_LIST_HEAD(&candidates);
4725 rdev_for_each_list(rdev, tmp, &pending_raid_disks)
4726 if (super_90_load(rdev, rdev0, 0) >= 0) {
4727 printk(KERN_INFO "md: adding %s ...\n",
4728 bdevname(rdev->bdev,b));
4729 list_move(&rdev->same_set, &candidates);
4732 * now we have a set of devices, with all of them having
4733 * mostly sane superblocks. It's time to allocate the
4737 dev = MKDEV(mdp_major,
4738 rdev0->preferred_minor << MdpMinorShift);
4739 unit = MINOR(dev) >> MdpMinorShift;
4741 dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
4744 if (rdev0->preferred_minor != unit) {
4745 printk(KERN_INFO "md: unit number in %s is bad: %d\n",
4746 bdevname(rdev0->bdev, b), rdev0->preferred_minor);
4750 md_probe(dev, NULL, NULL);
4751 mddev = mddev_find(dev);
4752 if (!mddev || !mddev->gendisk) {
4756 "md: cannot allocate memory for md drive.\n");
4759 if (mddev_lock(mddev))
4760 printk(KERN_WARNING "md: %s locked, cannot run\n",
4762 else if (mddev->raid_disks || mddev->major_version
4763 || !list_empty(&mddev->disks)) {
4765 "md: %s already running, cannot run %s\n",
4766 mdname(mddev), bdevname(rdev0->bdev,b));
4767 mddev_unlock(mddev);
4769 printk(KERN_INFO "md: created %s\n", mdname(mddev));
4770 mddev->persistent = 1;
4771 rdev_for_each_list(rdev, tmp, &candidates) {
4772 list_del_init(&rdev->same_set);
4773 if (bind_rdev_to_array(rdev, mddev))
4776 autorun_array(mddev);
4777 mddev_unlock(mddev);
4779 /* on success, candidates will be empty, on error
4782 rdev_for_each_list(rdev, tmp, &candidates) {
4783 list_del_init(&rdev->same_set);
4788 printk(KERN_INFO "md: ... autorun DONE.\n");
4790 #endif /* !MODULE */
4792 static int get_version(void __user * arg)
4796 ver.major = MD_MAJOR_VERSION;
4797 ver.minor = MD_MINOR_VERSION;
4798 ver.patchlevel = MD_PATCHLEVEL_VERSION;
4800 if (copy_to_user(arg, &ver, sizeof(ver)))
4806 static int get_array_info(mddev_t * mddev, void __user * arg)
4808 mdu_array_info_t info;
4809 int nr,working,insync,failed,spare;
4812 nr=working=insync=failed=spare=0;
4813 list_for_each_entry(rdev, &mddev->disks, same_set) {
4815 if (test_bit(Faulty, &rdev->flags))
4819 if (test_bit(In_sync, &rdev->flags))
4826 info.major_version = mddev->major_version;
4827 info.minor_version = mddev->minor_version;
4828 info.patch_version = MD_PATCHLEVEL_VERSION;
4829 info.ctime = mddev->ctime;
4830 info.level = mddev->level;
4831 info.size = mddev->dev_sectors / 2;
4832 if (info.size != mddev->dev_sectors / 2) /* overflow */
4835 info.raid_disks = mddev->raid_disks;
4836 info.md_minor = mddev->md_minor;
4837 info.not_persistent= !mddev->persistent;
4839 info.utime = mddev->utime;
4842 info.state = (1<<MD_SB_CLEAN);
4843 if (mddev->bitmap && mddev->bitmap_info.offset)
4844 info.state = (1<<MD_SB_BITMAP_PRESENT);
4845 info.active_disks = insync;
4846 info.working_disks = working;
4847 info.failed_disks = failed;
4848 info.spare_disks = spare;
4850 info.layout = mddev->layout;
4851 info.chunk_size = mddev->chunk_sectors << 9;
4853 if (copy_to_user(arg, &info, sizeof(info)))
4859 static int get_bitmap_file(mddev_t * mddev, void __user * arg)
4861 mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */
4862 char *ptr, *buf = NULL;
4865 if (md_allow_write(mddev))
4866 file = kmalloc(sizeof(*file), GFP_NOIO);
4868 file = kmalloc(sizeof(*file), GFP_KERNEL);
4873 /* bitmap disabled, zero the first byte and copy out */
4874 if (!mddev->bitmap || !mddev->bitmap->file) {
4875 file->pathname[0] = '\0';
4879 buf = kmalloc(sizeof(file->pathname), GFP_KERNEL);
4883 ptr = d_path(&mddev->bitmap->file->f_path, buf, sizeof(file->pathname));
4887 strcpy(file->pathname, ptr);
4891 if (copy_to_user(arg, file, sizeof(*file)))
4899 static int get_disk_info(mddev_t * mddev, void __user * arg)
4901 mdu_disk_info_t info;
4904 if (copy_from_user(&info, arg, sizeof(info)))
4907 rdev = find_rdev_nr(mddev, info.number);
4909 info.major = MAJOR(rdev->bdev->bd_dev);
4910 info.minor = MINOR(rdev->bdev->bd_dev);
4911 info.raid_disk = rdev->raid_disk;
4913 if (test_bit(Faulty, &rdev->flags))
4914 info.state |= (1<<MD_DISK_FAULTY);
4915 else if (test_bit(In_sync, &rdev->flags)) {
4916 info.state |= (1<<MD_DISK_ACTIVE);
4917 info.state |= (1<<MD_DISK_SYNC);
4919 if (test_bit(WriteMostly, &rdev->flags))
4920 info.state |= (1<<MD_DISK_WRITEMOSTLY);
4922 info.major = info.minor = 0;
4923 info.raid_disk = -1;
4924 info.state = (1<<MD_DISK_REMOVED);
4927 if (copy_to_user(arg, &info, sizeof(info)))
4933 static int add_new_disk(mddev_t * mddev, mdu_disk_info_t *info)
4935 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
4937 dev_t dev = MKDEV(info->major,info->minor);
4939 if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
4942 if (!mddev->raid_disks) {
4944 /* expecting a device which has a superblock */
4945 rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
4948 "md: md_import_device returned %ld\n",
4950 return PTR_ERR(rdev);
4952 if (!list_empty(&mddev->disks)) {
4953 mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
4954 mdk_rdev_t, same_set);
4955 err = super_types[mddev->major_version]
4956 .load_super(rdev, rdev0, mddev->minor_version);
4959 "md: %s has different UUID to %s\n",
4960 bdevname(rdev->bdev,b),
4961 bdevname(rdev0->bdev,b2));
4966 err = bind_rdev_to_array(rdev, mddev);
4973 * add_new_disk can be used once the array is assembled
4974 * to add "hot spares". They must already have a superblock
4979 if (!mddev->pers->hot_add_disk) {
4981 "%s: personality does not support diskops!\n",
4985 if (mddev->persistent)
4986 rdev = md_import_device(dev, mddev->major_version,
4987 mddev->minor_version);
4989 rdev = md_import_device(dev, -1, -1);
4992 "md: md_import_device returned %ld\n",
4994 return PTR_ERR(rdev);
4996 /* set save_raid_disk if appropriate */
4997 if (!mddev->persistent) {
4998 if (info->state & (1<<MD_DISK_SYNC) &&
4999 info->raid_disk < mddev->raid_disks)
5000 rdev->raid_disk = info->raid_disk;
5002 rdev->raid_disk = -1;
5004 super_types[mddev->major_version].
5005 validate_super(mddev, rdev);
5006 rdev->saved_raid_disk = rdev->raid_disk;
5008 clear_bit(In_sync, &rdev->flags); /* just to be sure */
5009 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
5010 set_bit(WriteMostly, &rdev->flags);
5012 clear_bit(WriteMostly, &rdev->flags);
5014 rdev->raid_disk = -1;
5015 err = bind_rdev_to_array(rdev, mddev);
5016 if (!err && !mddev->pers->hot_remove_disk) {
5017 /* If there is hot_add_disk but no hot_remove_disk
5018 * then added disks for geometry changes,
5019 * and should be added immediately.
5021 super_types[mddev->major_version].
5022 validate_super(mddev, rdev);
5023 err = mddev->pers->hot_add_disk(mddev, rdev);
5025 unbind_rdev_from_array(rdev);
5030 sysfs_notify_dirent(rdev->sysfs_state);
5032 md_update_sb(mddev, 1);
5033 if (mddev->degraded)
5034 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
5035 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5036 md_wakeup_thread(mddev->thread);
5040 /* otherwise, add_new_disk is only allowed
5041 * for major_version==0 superblocks
5043 if (mddev->major_version != 0) {
5044 printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n",
5049 if (!(info->state & (1<<MD_DISK_FAULTY))) {
5051 rdev = md_import_device(dev, -1, 0);
5054 "md: error, md_import_device() returned %ld\n",
5056 return PTR_ERR(rdev);
5058 rdev->desc_nr = info->number;
5059 if (info->raid_disk < mddev->raid_disks)
5060 rdev->raid_disk = info->raid_disk;
5062 rdev->raid_disk = -1;
5064 if (rdev->raid_disk < mddev->raid_disks)
5065 if (info->state & (1<<MD_DISK_SYNC))
5066 set_bit(In_sync, &rdev->flags);
5068 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
5069 set_bit(WriteMostly, &rdev->flags);
5071 if (!mddev->persistent) {
5072 printk(KERN_INFO "md: nonpersistent superblock ...\n");
5073 rdev->sb_start = rdev->bdev->bd_inode->i_size / 512;
5075 rdev->sb_start = calc_dev_sboffset(rdev->bdev);
5076 rdev->sectors = rdev->sb_start;
5078 err = bind_rdev_to_array(rdev, mddev);
5088 static int hot_remove_disk(mddev_t * mddev, dev_t dev)
5090 char b[BDEVNAME_SIZE];
5093 rdev = find_rdev(mddev, dev);
5097 if (rdev->raid_disk >= 0)
5100 kick_rdev_from_array(rdev);
5101 md_update_sb(mddev, 1);
5102 md_new_event(mddev);
5106 printk(KERN_WARNING "md: cannot remove active disk %s from %s ...\n",
5107 bdevname(rdev->bdev,b), mdname(mddev));
5111 static int hot_add_disk(mddev_t * mddev, dev_t dev)
5113 char b[BDEVNAME_SIZE];
5120 if (mddev->major_version != 0) {
5121 printk(KERN_WARNING "%s: HOT_ADD may only be used with"
5122 " version-0 superblocks.\n",
5126 if (!mddev->pers->hot_add_disk) {
5128 "%s: personality does not support diskops!\n",
5133 rdev = md_import_device(dev, -1, 0);
5136 "md: error, md_import_device() returned %ld\n",
5141 if (mddev->persistent)
5142 rdev->sb_start = calc_dev_sboffset(rdev->bdev);
5144 rdev->sb_start = rdev->bdev->bd_inode->i_size / 512;
5146 rdev->sectors = rdev->sb_start;
5148 if (test_bit(Faulty, &rdev->flags)) {
5150 "md: can not hot-add faulty %s disk to %s!\n",
5151 bdevname(rdev->bdev,b), mdname(mddev));
5155 clear_bit(In_sync, &rdev->flags);
5157 rdev->saved_raid_disk = -1;
5158 err = bind_rdev_to_array(rdev, mddev);
5163 * The rest should better be atomic, we can have disk failures
5164 * noticed in interrupt contexts ...
5167 rdev->raid_disk = -1;
5169 md_update_sb(mddev, 1);
5172 * Kick recovery, maybe this spare has to be added to the
5173 * array immediately.
5175 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5176 md_wakeup_thread(mddev->thread);
5177 md_new_event(mddev);
5185 static int set_bitmap_file(mddev_t *mddev, int fd)
5190 if (!mddev->pers->quiesce)
5192 if (mddev->recovery || mddev->sync_thread)
5194 /* we should be able to change the bitmap.. */
5200 return -EEXIST; /* cannot add when bitmap is present */
5201 mddev->bitmap_info.file = fget(fd);
5203 if (mddev->bitmap_info.file == NULL) {
5204 printk(KERN_ERR "%s: error: failed to get bitmap file\n",
5209 err = deny_bitmap_write_access(mddev->bitmap_info.file);
5211 printk(KERN_ERR "%s: error: bitmap file is already in use\n",
5213 fput(mddev->bitmap_info.file);
5214 mddev->bitmap_info.file = NULL;
5217 mddev->bitmap_info.offset = 0; /* file overrides offset */
5218 } else if (mddev->bitmap == NULL)
5219 return -ENOENT; /* cannot remove what isn't there */
5222 mddev->pers->quiesce(mddev, 1);
5224 err = bitmap_create(mddev);
5225 if (fd < 0 || err) {
5226 bitmap_destroy(mddev);
5227 fd = -1; /* make sure to put the file */
5229 mddev->pers->quiesce(mddev, 0);
5232 if (mddev->bitmap_info.file) {
5233 restore_bitmap_write_access(mddev->bitmap_info.file);
5234 fput(mddev->bitmap_info.file);
5236 mddev->bitmap_info.file = NULL;
5243 * set_array_info is used two different ways
5244 * The original usage is when creating a new array.
5245 * In this usage, raid_disks is > 0 and it together with
5246 * level, size, not_persistent,layout,chunksize determine the
5247 * shape of the array.
5248 * This will always create an array with a type-0.90.0 superblock.
5249 * The newer usage is when assembling an array.
5250 * In this case raid_disks will be 0, and the major_version field is
5251 * use to determine which style super-blocks are to be found on the devices.
5252 * The minor and patch _version numbers are also kept incase the
5253 * super_block handler wishes to interpret them.
5255 static int set_array_info(mddev_t * mddev, mdu_array_info_t *info)
5258 if (info->raid_disks == 0) {
5259 /* just setting version number for superblock loading */
5260 if (info->major_version < 0 ||
5261 info->major_version >= ARRAY_SIZE(super_types) ||
5262 super_types[info->major_version].name == NULL) {
5263 /* maybe try to auto-load a module? */
5265 "md: superblock version %d not known\n",
5266 info->major_version);
5269 mddev->major_version = info->major_version;
5270 mddev->minor_version = info->minor_version;
5271 mddev->patch_version = info->patch_version;
5272 mddev->persistent = !info->not_persistent;
5275 mddev->major_version = MD_MAJOR_VERSION;
5276 mddev->minor_version = MD_MINOR_VERSION;
5277 mddev->patch_version = MD_PATCHLEVEL_VERSION;
5278 mddev->ctime = get_seconds();
5280 mddev->level = info->level;
5281 mddev->clevel[0] = 0;
5282 mddev->dev_sectors = 2 * (sector_t)info->size;
5283 mddev->raid_disks = info->raid_disks;
5284 /* don't set md_minor, it is determined by which /dev/md* was
5287 if (info->state & (1<<MD_SB_CLEAN))
5288 mddev->recovery_cp = MaxSector;
5290 mddev->recovery_cp = 0;
5291 mddev->persistent = ! info->not_persistent;
5292 mddev->external = 0;
5294 mddev->layout = info->layout;
5295 mddev->chunk_sectors = info->chunk_size >> 9;
5297 mddev->max_disks = MD_SB_DISKS;
5299 if (mddev->persistent)
5301 set_bit(MD_CHANGE_DEVS, &mddev->flags);
5303 mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9;
5304 mddev->bitmap_info.offset = 0;
5306 mddev->reshape_position = MaxSector;
5309 * Generate a 128 bit UUID
5311 get_random_bytes(mddev->uuid, 16);
5313 mddev->new_level = mddev->level;
5314 mddev->new_chunk_sectors = mddev->chunk_sectors;
5315 mddev->new_layout = mddev->layout;
5316 mddev->delta_disks = 0;
5321 void md_set_array_sectors(mddev_t *mddev, sector_t array_sectors)
5323 WARN(!mddev_is_locked(mddev), "%s: unlocked mddev!\n", __func__);
5325 if (mddev->external_size)
5328 mddev->array_sectors = array_sectors;
5330 EXPORT_SYMBOL(md_set_array_sectors);
5332 static int update_size(mddev_t *mddev, sector_t num_sectors)
5336 int fit = (num_sectors == 0);
5338 if (mddev->pers->resize == NULL)
5340 /* The "num_sectors" is the number of sectors of each device that
5341 * is used. This can only make sense for arrays with redundancy.
5342 * linear and raid0 always use whatever space is available. We can only
5343 * consider changing this number if no resync or reconstruction is
5344 * happening, and if the new size is acceptable. It must fit before the
5345 * sb_start or, if that is <data_offset, it must fit before the size
5346 * of each device. If num_sectors is zero, we find the largest size
5350 if (mddev->sync_thread)
5353 /* Sorry, cannot grow a bitmap yet, just remove it,
5357 list_for_each_entry(rdev, &mddev->disks, same_set) {
5358 sector_t avail = rdev->sectors;
5360 if (fit && (num_sectors == 0 || num_sectors > avail))
5361 num_sectors = avail;
5362 if (avail < num_sectors)
5365 rv = mddev->pers->resize(mddev, num_sectors);
5367 revalidate_disk(mddev->gendisk);
5371 static int update_raid_disks(mddev_t *mddev, int raid_disks)
5374 /* change the number of raid disks */
5375 if (mddev->pers->check_reshape == NULL)
5377 if (raid_disks <= 0 ||
5378 raid_disks >= mddev->max_disks)
5380 if (mddev->sync_thread || mddev->reshape_position != MaxSector)
5382 mddev->delta_disks = raid_disks - mddev->raid_disks;
5384 rv = mddev->pers->check_reshape(mddev);
5390 * update_array_info is used to change the configuration of an
5392 * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
5393 * fields in the info are checked against the array.
5394 * Any differences that cannot be handled will cause an error.
5395 * Normally, only one change can be managed at a time.
5397 static int update_array_info(mddev_t *mddev, mdu_array_info_t *info)
5403 /* calculate expected state,ignoring low bits */
5404 if (mddev->bitmap && mddev->bitmap_info.offset)
5405 state |= (1 << MD_SB_BITMAP_PRESENT);
5407 if (mddev->major_version != info->major_version ||
5408 mddev->minor_version != info->minor_version ||
5409 /* mddev->patch_version != info->patch_version || */
5410 mddev->ctime != info->ctime ||
5411 mddev->level != info->level ||
5412 /* mddev->layout != info->layout || */
5413 !mddev->persistent != info->not_persistent||
5414 mddev->chunk_sectors != info->chunk_size >> 9 ||
5415 /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
5416 ((state^info->state) & 0xfffffe00)
5419 /* Check there is only one change */
5420 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
5422 if (mddev->raid_disks != info->raid_disks)
5424 if (mddev->layout != info->layout)
5426 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT))
5433 if (mddev->layout != info->layout) {
5435 * we don't need to do anything at the md level, the
5436 * personality will take care of it all.
5438 if (mddev->pers->check_reshape == NULL)
5441 mddev->new_layout = info->layout;
5442 rv = mddev->pers->check_reshape(mddev);
5444 mddev->new_layout = mddev->layout;
5448 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
5449 rv = update_size(mddev, (sector_t)info->size * 2);
5451 if (mddev->raid_disks != info->raid_disks)
5452 rv = update_raid_disks(mddev, info->raid_disks);
5454 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) {
5455 if (mddev->pers->quiesce == NULL)
5457 if (mddev->recovery || mddev->sync_thread)
5459 if (info->state & (1<<MD_SB_BITMAP_PRESENT)) {
5460 /* add the bitmap */
5463 if (mddev->bitmap_info.default_offset == 0)
5465 mddev->bitmap_info.offset =
5466 mddev->bitmap_info.default_offset;
5467 mddev->pers->quiesce(mddev, 1);
5468 rv = bitmap_create(mddev);
5470 bitmap_destroy(mddev);
5471 mddev->pers->quiesce(mddev, 0);
5473 /* remove the bitmap */
5476 if (mddev->bitmap->file)
5478 mddev->pers->quiesce(mddev, 1);
5479 bitmap_destroy(mddev);
5480 mddev->pers->quiesce(mddev, 0);
5481 mddev->bitmap_info.offset = 0;
5484 md_update_sb(mddev, 1);
5488 static int set_disk_faulty(mddev_t *mddev, dev_t dev)
5492 if (mddev->pers == NULL)
5495 rdev = find_rdev(mddev, dev);
5499 md_error(mddev, rdev);
5504 * We have a problem here : there is no easy way to give a CHS
5505 * virtual geometry. We currently pretend that we have a 2 heads
5506 * 4 sectors (with a BIG number of cylinders...). This drives
5507 * dosfs just mad... ;-)
5509 static int md_getgeo(struct block_device *bdev, struct hd_geometry *geo)
5511 mddev_t *mddev = bdev->bd_disk->private_data;
5515 geo->cylinders = get_capacity(mddev->gendisk) / 8;
5519 static int md_ioctl(struct block_device *bdev, fmode_t mode,
5520 unsigned int cmd, unsigned long arg)
5523 void __user *argp = (void __user *)arg;
5524 mddev_t *mddev = NULL;
5526 if (!capable(CAP_SYS_ADMIN))
5530 * Commands dealing with the RAID driver but not any
5536 err = get_version(argp);
5539 case PRINT_RAID_DEBUG:
5547 autostart_arrays(arg);
5554 * Commands creating/starting a new array:
5557 mddev = bdev->bd_disk->private_data;
5564 err = mddev_lock(mddev);
5567 "md: ioctl lock interrupted, reason %d, cmd %d\n",
5574 case SET_ARRAY_INFO:
5576 mdu_array_info_t info;
5578 memset(&info, 0, sizeof(info));
5579 else if (copy_from_user(&info, argp, sizeof(info))) {
5584 err = update_array_info(mddev, &info);
5586 printk(KERN_WARNING "md: couldn't update"
5587 " array info. %d\n", err);
5592 if (!list_empty(&mddev->disks)) {
5594 "md: array %s already has disks!\n",
5599 if (mddev->raid_disks) {
5601 "md: array %s already initialised!\n",
5606 err = set_array_info(mddev, &info);
5608 printk(KERN_WARNING "md: couldn't set"
5609 " array info. %d\n", err);
5619 * Commands querying/configuring an existing array:
5621 /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
5622 * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */
5623 if ((!mddev->raid_disks && !mddev->external)
5624 && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY
5625 && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE
5626 && cmd != GET_BITMAP_FILE) {
5632 * Commands even a read-only array can execute:
5636 case GET_ARRAY_INFO:
5637 err = get_array_info(mddev, argp);
5640 case GET_BITMAP_FILE:
5641 err = get_bitmap_file(mddev, argp);
5645 err = get_disk_info(mddev, argp);
5648 case RESTART_ARRAY_RW:
5649 err = restart_array(mddev);
5653 err = do_md_stop(mddev, 0, 1);
5657 err = do_md_stop(mddev, 1, 1);
5663 * The remaining ioctls are changing the state of the
5664 * superblock, so we do not allow them on read-only arrays.
5665 * However non-MD ioctls (e.g. get-size) will still come through
5666 * here and hit the 'default' below, so only disallow
5667 * 'md' ioctls, and switch to rw mode if started auto-readonly.
5669 if (_IOC_TYPE(cmd) == MD_MAJOR && mddev->ro && mddev->pers) {
5670 if (mddev->ro == 2) {
5672 sysfs_notify_dirent(mddev->sysfs_state);
5673 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5674 md_wakeup_thread(mddev->thread);
5685 mdu_disk_info_t info;
5686 if (copy_from_user(&info, argp, sizeof(info)))
5689 err = add_new_disk(mddev, &info);
5693 case HOT_REMOVE_DISK:
5694 err = hot_remove_disk(mddev, new_decode_dev(arg));
5698 err = hot_add_disk(mddev, new_decode_dev(arg));
5701 case SET_DISK_FAULTY:
5702 err = set_disk_faulty(mddev, new_decode_dev(arg));
5706 err = do_md_run(mddev);
5709 case SET_BITMAP_FILE:
5710 err = set_bitmap_file(mddev, (int)arg);
5720 if (mddev->hold_active == UNTIL_IOCTL &&
5722 mddev->hold_active = 0;
5723 mddev_unlock(mddev);
5732 #ifdef CONFIG_COMPAT
5733 static int md_compat_ioctl(struct block_device *bdev, fmode_t mode,
5734 unsigned int cmd, unsigned long arg)
5737 case HOT_REMOVE_DISK:
5739 case SET_DISK_FAULTY:
5740 case SET_BITMAP_FILE:
5741 /* These take in integer arg, do not convert */
5744 arg = (unsigned long)compat_ptr(arg);
5748 return md_ioctl(bdev, mode, cmd, arg);
5750 #endif /* CONFIG_COMPAT */
5752 static int md_open(struct block_device *bdev, fmode_t mode)
5755 * Succeed if we can lock the mddev, which confirms that
5756 * it isn't being stopped right now.
5758 mddev_t *mddev = mddev_find(bdev->bd_dev);
5761 if (mddev->gendisk != bdev->bd_disk) {
5762 /* we are racing with mddev_put which is discarding this
5766 /* Wait until bdev->bd_disk is definitely gone */
5767 flush_scheduled_work();
5768 /* Then retry the open from the top */
5769 return -ERESTARTSYS;
5771 BUG_ON(mddev != bdev->bd_disk->private_data);
5773 if ((err = mutex_lock_interruptible(&mddev->open_mutex)))
5777 atomic_inc(&mddev->openers);
5778 mutex_unlock(&mddev->open_mutex);
5780 check_disk_change(bdev);
5785 static int md_release(struct gendisk *disk, fmode_t mode)
5787 mddev_t *mddev = disk->private_data;
5790 atomic_dec(&mddev->openers);
5796 static int md_media_changed(struct gendisk *disk)
5798 mddev_t *mddev = disk->private_data;
5800 return mddev->changed;
5803 static int md_revalidate(struct gendisk *disk)
5805 mddev_t *mddev = disk->private_data;
5810 static const struct block_device_operations md_fops =
5812 .owner = THIS_MODULE,
5814 .release = md_release,
5816 #ifdef CONFIG_COMPAT
5817 .compat_ioctl = md_compat_ioctl,
5819 .getgeo = md_getgeo,
5820 .media_changed = md_media_changed,
5821 .revalidate_disk= md_revalidate,
5824 static int md_thread(void * arg)
5826 mdk_thread_t *thread = arg;
5829 * md_thread is a 'system-thread', it's priority should be very
5830 * high. We avoid resource deadlocks individually in each
5831 * raid personality. (RAID5 does preallocation) We also use RR and
5832 * the very same RT priority as kswapd, thus we will never get
5833 * into a priority inversion deadlock.
5835 * we definitely have to have equal or higher priority than
5836 * bdflush, otherwise bdflush will deadlock if there are too
5837 * many dirty RAID5 blocks.
5840 allow_signal(SIGKILL);
5841 while (!kthread_should_stop()) {
5843 /* We need to wait INTERRUPTIBLE so that
5844 * we don't add to the load-average.
5845 * That means we need to be sure no signals are
5848 if (signal_pending(current))
5849 flush_signals(current);
5851 wait_event_interruptible_timeout
5853 test_bit(THREAD_WAKEUP, &thread->flags)
5854 || kthread_should_stop(),
5857 clear_bit(THREAD_WAKEUP, &thread->flags);
5859 thread->run(thread->mddev);
5865 void md_wakeup_thread(mdk_thread_t *thread)
5868 dprintk("md: waking up MD thread %s.\n", thread->tsk->comm);
5869 set_bit(THREAD_WAKEUP, &thread->flags);
5870 wake_up(&thread->wqueue);
5874 mdk_thread_t *md_register_thread(void (*run) (mddev_t *), mddev_t *mddev,
5877 mdk_thread_t *thread;
5879 thread = kzalloc(sizeof(mdk_thread_t), GFP_KERNEL);
5883 init_waitqueue_head(&thread->wqueue);
5886 thread->mddev = mddev;
5887 thread->timeout = MAX_SCHEDULE_TIMEOUT;
5888 thread->tsk = kthread_run(md_thread, thread,
5890 mdname(thread->mddev),
5891 name ?: mddev->pers->name);
5892 if (IS_ERR(thread->tsk)) {
5899 void md_unregister_thread(mdk_thread_t *thread)
5903 dprintk("interrupting MD-thread pid %d\n", task_pid_nr(thread->tsk));
5905 kthread_stop(thread->tsk);
5909 void md_error(mddev_t *mddev, mdk_rdev_t *rdev)
5916 if (!rdev || test_bit(Faulty, &rdev->flags))
5919 if (mddev->external)
5920 set_bit(Blocked, &rdev->flags);
5922 dprintk("md_error dev:%s, rdev:(%d:%d), (caller: %p,%p,%p,%p).\n",
5924 MAJOR(rdev->bdev->bd_dev), MINOR(rdev->bdev->bd_dev),
5925 __builtin_return_address(0),__builtin_return_address(1),
5926 __builtin_return_address(2),__builtin_return_address(3));
5930 if (!mddev->pers->error_handler)
5932 mddev->pers->error_handler(mddev,rdev);
5933 if (mddev->degraded)
5934 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
5935 set_bit(StateChanged, &rdev->flags);
5936 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5937 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5938 md_wakeup_thread(mddev->thread);
5939 md_new_event_inintr(mddev);
5942 /* seq_file implementation /proc/mdstat */
5944 static void status_unused(struct seq_file *seq)
5949 seq_printf(seq, "unused devices: ");
5951 list_for_each_entry(rdev, &pending_raid_disks, same_set) {
5952 char b[BDEVNAME_SIZE];
5954 seq_printf(seq, "%s ",
5955 bdevname(rdev->bdev,b));
5958 seq_printf(seq, "<none>");
5960 seq_printf(seq, "\n");
5964 static void status_resync(struct seq_file *seq, mddev_t * mddev)
5966 sector_t max_sectors, resync, res;
5967 unsigned long dt, db;
5970 unsigned int per_milli;
5972 resync = mddev->curr_resync - atomic_read(&mddev->recovery_active);
5974 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
5975 max_sectors = mddev->resync_max_sectors;
5977 max_sectors = mddev->dev_sectors;
5980 * Should not happen.
5986 /* Pick 'scale' such that (resync>>scale)*1000 will fit
5987 * in a sector_t, and (max_sectors>>scale) will fit in a
5988 * u32, as those are the requirements for sector_div.
5989 * Thus 'scale' must be at least 10
5992 if (sizeof(sector_t) > sizeof(unsigned long)) {
5993 while ( max_sectors/2 > (1ULL<<(scale+32)))
5996 res = (resync>>scale)*1000;
5997 sector_div(res, (u32)((max_sectors>>scale)+1));
6001 int i, x = per_milli/50, y = 20-x;
6002 seq_printf(seq, "[");
6003 for (i = 0; i < x; i++)
6004 seq_printf(seq, "=");
6005 seq_printf(seq, ">");
6006 for (i = 0; i < y; i++)
6007 seq_printf(seq, ".");
6008 seq_printf(seq, "] ");
6010 seq_printf(seq, " %s =%3u.%u%% (%llu/%llu)",
6011 (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)?
6013 (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)?
6015 (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
6016 "resync" : "recovery"))),
6017 per_milli/10, per_milli % 10,
6018 (unsigned long long) resync/2,
6019 (unsigned long long) max_sectors/2);
6022 * dt: time from mark until now
6023 * db: blocks written from mark until now
6024 * rt: remaining time
6026 * rt is a sector_t, so could be 32bit or 64bit.
6027 * So we divide before multiply in case it is 32bit and close
6029 * We scale the divisor (db) by 32 to avoid loosing precision
6030 * near the end of resync when the number of remaining sectors
6032 * We then divide rt by 32 after multiplying by db to compensate.
6033 * The '+1' avoids division by zero if db is very small.
6035 dt = ((jiffies - mddev->resync_mark) / HZ);
6037 db = (mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active))
6038 - mddev->resync_mark_cnt;
6040 rt = max_sectors - resync; /* number of remaining sectors */
6041 sector_div(rt, db/32+1);
6045 seq_printf(seq, " finish=%lu.%lumin", (unsigned long)rt / 60,
6046 ((unsigned long)rt % 60)/6);
6048 seq_printf(seq, " speed=%ldK/sec", db/2/dt);
6051 static void *md_seq_start(struct seq_file *seq, loff_t *pos)
6053 struct list_head *tmp;
6063 spin_lock(&all_mddevs_lock);
6064 list_for_each(tmp,&all_mddevs)
6066 mddev = list_entry(tmp, mddev_t, all_mddevs);
6068 spin_unlock(&all_mddevs_lock);
6071 spin_unlock(&all_mddevs_lock);
6073 return (void*)2;/* tail */
6077 static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
6079 struct list_head *tmp;
6080 mddev_t *next_mddev, *mddev = v;
6086 spin_lock(&all_mddevs_lock);
6088 tmp = all_mddevs.next;
6090 tmp = mddev->all_mddevs.next;
6091 if (tmp != &all_mddevs)
6092 next_mddev = mddev_get(list_entry(tmp,mddev_t,all_mddevs));
6094 next_mddev = (void*)2;
6097 spin_unlock(&all_mddevs_lock);
6105 static void md_seq_stop(struct seq_file *seq, void *v)
6109 if (mddev && v != (void*)1 && v != (void*)2)
6113 struct mdstat_info {
6117 static int md_seq_show(struct seq_file *seq, void *v)
6122 struct mdstat_info *mi = seq->private;
6123 struct bitmap *bitmap;
6125 if (v == (void*)1) {
6126 struct mdk_personality *pers;
6127 seq_printf(seq, "Personalities : ");
6128 spin_lock(&pers_lock);
6129 list_for_each_entry(pers, &pers_list, list)
6130 seq_printf(seq, "[%s] ", pers->name);
6132 spin_unlock(&pers_lock);
6133 seq_printf(seq, "\n");
6134 mi->event = atomic_read(&md_event_count);
6137 if (v == (void*)2) {
6142 if (mddev_lock(mddev) < 0)
6145 if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
6146 seq_printf(seq, "%s : %sactive", mdname(mddev),
6147 mddev->pers ? "" : "in");
6150 seq_printf(seq, " (read-only)");
6152 seq_printf(seq, " (auto-read-only)");
6153 seq_printf(seq, " %s", mddev->pers->name);
6157 list_for_each_entry(rdev, &mddev->disks, same_set) {
6158 char b[BDEVNAME_SIZE];
6159 seq_printf(seq, " %s[%d]",
6160 bdevname(rdev->bdev,b), rdev->desc_nr);
6161 if (test_bit(WriteMostly, &rdev->flags))
6162 seq_printf(seq, "(W)");
6163 if (test_bit(Faulty, &rdev->flags)) {
6164 seq_printf(seq, "(F)");
6166 } else if (rdev->raid_disk < 0)
6167 seq_printf(seq, "(S)"); /* spare */
6168 sectors += rdev->sectors;
6171 if (!list_empty(&mddev->disks)) {
6173 seq_printf(seq, "\n %llu blocks",
6174 (unsigned long long)
6175 mddev->array_sectors / 2);
6177 seq_printf(seq, "\n %llu blocks",
6178 (unsigned long long)sectors / 2);
6180 if (mddev->persistent) {
6181 if (mddev->major_version != 0 ||
6182 mddev->minor_version != 90) {
6183 seq_printf(seq," super %d.%d",
6184 mddev->major_version,
6185 mddev->minor_version);
6187 } else if (mddev->external)
6188 seq_printf(seq, " super external:%s",
6189 mddev->metadata_type);
6191 seq_printf(seq, " super non-persistent");
6194 mddev->pers->status(seq, mddev);
6195 seq_printf(seq, "\n ");
6196 if (mddev->pers->sync_request) {
6197 if (mddev->curr_resync > 2) {
6198 status_resync(seq, mddev);
6199 seq_printf(seq, "\n ");
6200 } else if (mddev->curr_resync == 1 || mddev->curr_resync == 2)
6201 seq_printf(seq, "\tresync=DELAYED\n ");
6202 else if (mddev->recovery_cp < MaxSector)
6203 seq_printf(seq, "\tresync=PENDING\n ");
6206 seq_printf(seq, "\n ");
6208 if ((bitmap = mddev->bitmap)) {
6209 unsigned long chunk_kb;
6210 unsigned long flags;
6211 spin_lock_irqsave(&bitmap->lock, flags);
6212 chunk_kb = mddev->bitmap_info.chunksize >> 10;
6213 seq_printf(seq, "bitmap: %lu/%lu pages [%luKB], "
6215 bitmap->pages - bitmap->missing_pages,
6217 (bitmap->pages - bitmap->missing_pages)
6218 << (PAGE_SHIFT - 10),
6219 chunk_kb ? chunk_kb : mddev->bitmap_info.chunksize,
6220 chunk_kb ? "KB" : "B");
6222 seq_printf(seq, ", file: ");
6223 seq_path(seq, &bitmap->file->f_path, " \t\n");
6226 seq_printf(seq, "\n");
6227 spin_unlock_irqrestore(&bitmap->lock, flags);
6230 seq_printf(seq, "\n");
6232 mddev_unlock(mddev);
6237 static const struct seq_operations md_seq_ops = {
6238 .start = md_seq_start,
6239 .next = md_seq_next,
6240 .stop = md_seq_stop,
6241 .show = md_seq_show,
6244 static int md_seq_open(struct inode *inode, struct file *file)
6247 struct mdstat_info *mi = kmalloc(sizeof(*mi), GFP_KERNEL);
6251 error = seq_open(file, &md_seq_ops);
6255 struct seq_file *p = file->private_data;
6257 mi->event = atomic_read(&md_event_count);
6262 static unsigned int mdstat_poll(struct file *filp, poll_table *wait)
6264 struct seq_file *m = filp->private_data;
6265 struct mdstat_info *mi = m->private;
6268 poll_wait(filp, &md_event_waiters, wait);
6270 /* always allow read */
6271 mask = POLLIN | POLLRDNORM;
6273 if (mi->event != atomic_read(&md_event_count))
6274 mask |= POLLERR | POLLPRI;
6278 static const struct file_operations md_seq_fops = {
6279 .owner = THIS_MODULE,
6280 .open = md_seq_open,
6282 .llseek = seq_lseek,
6283 .release = seq_release_private,
6284 .poll = mdstat_poll,
6287 int register_md_personality(struct mdk_personality *p)
6289 spin_lock(&pers_lock);
6290 list_add_tail(&p->list, &pers_list);
6291 printk(KERN_INFO "md: %s personality registered for level %d\n", p->name, p->level);
6292 spin_unlock(&pers_lock);
6296 int unregister_md_personality(struct mdk_personality *p)
6298 printk(KERN_INFO "md: %s personality unregistered\n", p->name);
6299 spin_lock(&pers_lock);
6300 list_del_init(&p->list);
6301 spin_unlock(&pers_lock);
6305 static int is_mddev_idle(mddev_t *mddev, int init)
6313 rdev_for_each_rcu(rdev, mddev) {
6314 struct gendisk *disk = rdev->bdev->bd_contains->bd_disk;
6315 curr_events = (int)part_stat_read(&disk->part0, sectors[0]) +
6316 (int)part_stat_read(&disk->part0, sectors[1]) -
6317 atomic_read(&disk->sync_io);
6318 /* sync IO will cause sync_io to increase before the disk_stats
6319 * as sync_io is counted when a request starts, and
6320 * disk_stats is counted when it completes.
6321 * So resync activity will cause curr_events to be smaller than
6322 * when there was no such activity.
6323 * non-sync IO will cause disk_stat to increase without
6324 * increasing sync_io so curr_events will (eventually)
6325 * be larger than it was before. Once it becomes
6326 * substantially larger, the test below will cause
6327 * the array to appear non-idle, and resync will slow
6329 * If there is a lot of outstanding resync activity when
6330 * we set last_event to curr_events, then all that activity
6331 * completing might cause the array to appear non-idle
6332 * and resync will be slowed down even though there might
6333 * not have been non-resync activity. This will only
6334 * happen once though. 'last_events' will soon reflect
6335 * the state where there is little or no outstanding
6336 * resync requests, and further resync activity will
6337 * always make curr_events less than last_events.
6340 if (init || curr_events - rdev->last_events > 64) {
6341 rdev->last_events = curr_events;
6349 void md_done_sync(mddev_t *mddev, int blocks, int ok)
6351 /* another "blocks" (512byte) blocks have been synced */
6352 atomic_sub(blocks, &mddev->recovery_active);
6353 wake_up(&mddev->recovery_wait);
6355 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6356 md_wakeup_thread(mddev->thread);
6357 // stop recovery, signal do_sync ....
6362 /* md_write_start(mddev, bi)
6363 * If we need to update some array metadata (e.g. 'active' flag
6364 * in superblock) before writing, schedule a superblock update
6365 * and wait for it to complete.
6367 void md_write_start(mddev_t *mddev, struct bio *bi)
6370 if (bio_data_dir(bi) != WRITE)
6373 BUG_ON(mddev->ro == 1);
6374 if (mddev->ro == 2) {
6375 /* need to switch to read/write */
6377 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6378 md_wakeup_thread(mddev->thread);
6379 md_wakeup_thread(mddev->sync_thread);
6382 atomic_inc(&mddev->writes_pending);
6383 if (mddev->safemode == 1)
6384 mddev->safemode = 0;
6385 if (mddev->in_sync) {
6386 spin_lock_irq(&mddev->write_lock);
6387 if (mddev->in_sync) {
6389 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
6390 md_wakeup_thread(mddev->thread);
6393 spin_unlock_irq(&mddev->write_lock);
6396 sysfs_notify_dirent(mddev->sysfs_state);
6397 wait_event(mddev->sb_wait,
6398 !test_bit(MD_CHANGE_CLEAN, &mddev->flags) &&
6399 !test_bit(MD_CHANGE_PENDING, &mddev->flags));
6402 void md_write_end(mddev_t *mddev)
6404 if (atomic_dec_and_test(&mddev->writes_pending)) {
6405 if (mddev->safemode == 2)
6406 md_wakeup_thread(mddev->thread);
6407 else if (mddev->safemode_delay)
6408 mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay);
6412 /* md_allow_write(mddev)
6413 * Calling this ensures that the array is marked 'active' so that writes
6414 * may proceed without blocking. It is important to call this before
6415 * attempting a GFP_KERNEL allocation while holding the mddev lock.
6416 * Must be called with mddev_lock held.
6418 * In the ->external case MD_CHANGE_CLEAN can not be cleared until mddev->lock
6419 * is dropped, so return -EAGAIN after notifying userspace.
6421 int md_allow_write(mddev_t *mddev)
6427 if (!mddev->pers->sync_request)
6430 spin_lock_irq(&mddev->write_lock);
6431 if (mddev->in_sync) {
6433 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
6434 if (mddev->safemode_delay &&
6435 mddev->safemode == 0)
6436 mddev->safemode = 1;
6437 spin_unlock_irq(&mddev->write_lock);
6438 md_update_sb(mddev, 0);
6439 sysfs_notify_dirent(mddev->sysfs_state);
6441 spin_unlock_irq(&mddev->write_lock);
6443 if (test_bit(MD_CHANGE_CLEAN, &mddev->flags))
6448 EXPORT_SYMBOL_GPL(md_allow_write);
6450 #define SYNC_MARKS 10
6451 #define SYNC_MARK_STEP (3*HZ)
6452 void md_do_sync(mddev_t *mddev)
6455 unsigned int currspeed = 0,
6457 sector_t max_sectors,j, io_sectors;
6458 unsigned long mark[SYNC_MARKS];
6459 sector_t mark_cnt[SYNC_MARKS];
6461 struct list_head *tmp;
6462 sector_t last_check;
6467 /* just incase thread restarts... */
6468 if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
6470 if (mddev->ro) /* never try to sync a read-only array */
6473 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
6474 if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
6475 desc = "data-check";
6476 else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
6477 desc = "requested-resync";
6480 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
6485 /* we overload curr_resync somewhat here.
6486 * 0 == not engaged in resync at all
6487 * 2 == checking that there is no conflict with another sync
6488 * 1 == like 2, but have yielded to allow conflicting resync to
6490 * other == active in resync - this many blocks
6492 * Before starting a resync we must have set curr_resync to
6493 * 2, and then checked that every "conflicting" array has curr_resync
6494 * less than ours. When we find one that is the same or higher
6495 * we wait on resync_wait. To avoid deadlock, we reduce curr_resync
6496 * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
6497 * This will mean we have to start checking from the beginning again.
6502 mddev->curr_resync = 2;
6505 if (kthread_should_stop()) {
6506 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6509 for_each_mddev(mddev2, tmp) {
6510 if (mddev2 == mddev)
6512 if (!mddev->parallel_resync
6513 && mddev2->curr_resync
6514 && match_mddev_units(mddev, mddev2)) {
6516 if (mddev < mddev2 && mddev->curr_resync == 2) {
6517 /* arbitrarily yield */
6518 mddev->curr_resync = 1;
6519 wake_up(&resync_wait);
6521 if (mddev > mddev2 && mddev->curr_resync == 1)
6522 /* no need to wait here, we can wait the next
6523 * time 'round when curr_resync == 2
6526 /* We need to wait 'interruptible' so as not to
6527 * contribute to the load average, and not to
6528 * be caught by 'softlockup'
6530 prepare_to_wait(&resync_wait, &wq, TASK_INTERRUPTIBLE);
6531 if (!kthread_should_stop() &&
6532 mddev2->curr_resync >= mddev->curr_resync) {
6533 printk(KERN_INFO "md: delaying %s of %s"
6534 " until %s has finished (they"
6535 " share one or more physical units)\n",
6536 desc, mdname(mddev), mdname(mddev2));
6538 if (signal_pending(current))
6539 flush_signals(current);
6541 finish_wait(&resync_wait, &wq);
6544 finish_wait(&resync_wait, &wq);
6547 } while (mddev->curr_resync < 2);
6550 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
6551 /* resync follows the size requested by the personality,
6552 * which defaults to physical size, but can be virtual size
6554 max_sectors = mddev->resync_max_sectors;
6555 mddev->resync_mismatches = 0;
6556 /* we don't use the checkpoint if there's a bitmap */
6557 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
6558 j = mddev->resync_min;
6559 else if (!mddev->bitmap)
6560 j = mddev->recovery_cp;
6562 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
6563 max_sectors = mddev->dev_sectors;
6565 /* recovery follows the physical size of devices */
6566 max_sectors = mddev->dev_sectors;
6569 list_for_each_entry_rcu(rdev, &mddev->disks, same_set)
6570 if (rdev->raid_disk >= 0 &&
6571 !test_bit(Faulty, &rdev->flags) &&
6572 !test_bit(In_sync, &rdev->flags) &&
6573 rdev->recovery_offset < j)
6574 j = rdev->recovery_offset;
6578 printk(KERN_INFO "md: %s of RAID array %s\n", desc, mdname(mddev));
6579 printk(KERN_INFO "md: minimum _guaranteed_ speed:"
6580 " %d KB/sec/disk.\n", speed_min(mddev));
6581 printk(KERN_INFO "md: using maximum available idle IO bandwidth "
6582 "(but not more than %d KB/sec) for %s.\n",
6583 speed_max(mddev), desc);
6585 is_mddev_idle(mddev, 1); /* this initializes IO event counters */
6588 for (m = 0; m < SYNC_MARKS; m++) {
6590 mark_cnt[m] = io_sectors;
6593 mddev->resync_mark = mark[last_mark];
6594 mddev->resync_mark_cnt = mark_cnt[last_mark];
6597 * Tune reconstruction:
6599 window = 32*(PAGE_SIZE/512);
6600 printk(KERN_INFO "md: using %dk window, over a total of %llu blocks.\n",
6601 window/2,(unsigned long long) max_sectors/2);
6603 atomic_set(&mddev->recovery_active, 0);
6608 "md: resuming %s of %s from checkpoint.\n",
6609 desc, mdname(mddev));
6610 mddev->curr_resync = j;
6612 mddev->curr_resync_completed = mddev->curr_resync;
6614 while (j < max_sectors) {
6619 if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
6620 ((mddev->curr_resync > mddev->curr_resync_completed &&
6621 (mddev->curr_resync - mddev->curr_resync_completed)
6622 > (max_sectors >> 4)) ||
6623 (j - mddev->curr_resync_completed)*2
6624 >= mddev->resync_max - mddev->curr_resync_completed
6626 /* time to update curr_resync_completed */
6627 blk_unplug(mddev->queue);
6628 wait_event(mddev->recovery_wait,
6629 atomic_read(&mddev->recovery_active) == 0);
6630 mddev->curr_resync_completed =
6632 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
6633 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
6636 while (j >= mddev->resync_max && !kthread_should_stop()) {
6637 /* As this condition is controlled by user-space,
6638 * we can block indefinitely, so use '_interruptible'
6639 * to avoid triggering warnings.
6641 flush_signals(current); /* just in case */
6642 wait_event_interruptible(mddev->recovery_wait,
6643 mddev->resync_max > j
6644 || kthread_should_stop());
6647 if (kthread_should_stop())
6650 sectors = mddev->pers->sync_request(mddev, j, &skipped,
6651 currspeed < speed_min(mddev));
6653 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6657 if (!skipped) { /* actual IO requested */
6658 io_sectors += sectors;
6659 atomic_add(sectors, &mddev->recovery_active);
6663 if (j>1) mddev->curr_resync = j;
6664 mddev->curr_mark_cnt = io_sectors;
6665 if (last_check == 0)
6666 /* this is the earliers that rebuilt will be
6667 * visible in /proc/mdstat
6669 md_new_event(mddev);
6671 if (last_check + window > io_sectors || j == max_sectors)
6674 last_check = io_sectors;
6676 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
6680 if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) {
6682 int next = (last_mark+1) % SYNC_MARKS;
6684 mddev->resync_mark = mark[next];
6685 mddev->resync_mark_cnt = mark_cnt[next];
6686 mark[next] = jiffies;
6687 mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active);
6692 if (kthread_should_stop())
6697 * this loop exits only if either when we are slower than
6698 * the 'hard' speed limit, or the system was IO-idle for
6700 * the system might be non-idle CPU-wise, but we only care
6701 * about not overloading the IO subsystem. (things like an
6702 * e2fsck being done on the RAID array should execute fast)
6704 blk_unplug(mddev->queue);
6707 currspeed = ((unsigned long)(io_sectors-mddev->resync_mark_cnt))/2
6708 /((jiffies-mddev->resync_mark)/HZ +1) +1;
6710 if (currspeed > speed_min(mddev)) {
6711 if ((currspeed > speed_max(mddev)) ||
6712 !is_mddev_idle(mddev, 0)) {
6718 printk(KERN_INFO "md: %s: %s done.\n",mdname(mddev), desc);
6720 * this also signals 'finished resyncing' to md_stop
6723 blk_unplug(mddev->queue);
6725 wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
6727 /* tell personality that we are finished */
6728 mddev->pers->sync_request(mddev, max_sectors, &skipped, 1);
6730 if (!test_bit(MD_RECOVERY_CHECK, &mddev->recovery) &&
6731 mddev->curr_resync > 2) {
6732 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
6733 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
6734 if (mddev->curr_resync >= mddev->recovery_cp) {
6736 "md: checkpointing %s of %s.\n",
6737 desc, mdname(mddev));
6738 mddev->recovery_cp = mddev->curr_resync;
6741 mddev->recovery_cp = MaxSector;
6743 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery))
6744 mddev->curr_resync = MaxSector;
6746 list_for_each_entry_rcu(rdev, &mddev->disks, same_set)
6747 if (rdev->raid_disk >= 0 &&
6748 !test_bit(Faulty, &rdev->flags) &&
6749 !test_bit(In_sync, &rdev->flags) &&
6750 rdev->recovery_offset < mddev->curr_resync)
6751 rdev->recovery_offset = mddev->curr_resync;
6755 set_bit(MD_CHANGE_DEVS, &mddev->flags);
6758 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
6759 /* We completed so min/max setting can be forgotten if used. */
6760 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
6761 mddev->resync_min = 0;
6762 mddev->resync_max = MaxSector;
6763 } else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
6764 mddev->resync_min = mddev->curr_resync_completed;
6765 mddev->curr_resync = 0;
6766 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery))
6767 mddev->curr_resync_completed = 0;
6768 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
6769 wake_up(&resync_wait);
6770 set_bit(MD_RECOVERY_DONE, &mddev->recovery);
6771 md_wakeup_thread(mddev->thread);
6776 * got a signal, exit.
6779 "md: md_do_sync() got signal ... exiting\n");
6780 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6784 EXPORT_SYMBOL_GPL(md_do_sync);
6787 static int remove_and_add_spares(mddev_t *mddev)
6792 mddev->curr_resync_completed = 0;
6794 list_for_each_entry(rdev, &mddev->disks, same_set)
6795 if (rdev->raid_disk >= 0 &&
6796 !test_bit(Blocked, &rdev->flags) &&
6797 (test_bit(Faulty, &rdev->flags) ||
6798 ! test_bit(In_sync, &rdev->flags)) &&
6799 atomic_read(&rdev->nr_pending)==0) {
6800 if (mddev->pers->hot_remove_disk(
6801 mddev, rdev->raid_disk)==0) {
6803 sprintf(nm,"rd%d", rdev->raid_disk);
6804 sysfs_remove_link(&mddev->kobj, nm);
6805 rdev->raid_disk = -1;
6809 if (mddev->degraded && ! mddev->ro && !mddev->recovery_disabled) {
6810 list_for_each_entry(rdev, &mddev->disks, same_set) {
6811 if (rdev->raid_disk >= 0 &&
6812 !test_bit(In_sync, &rdev->flags) &&
6813 !test_bit(Blocked, &rdev->flags))
6815 if (rdev->raid_disk < 0
6816 && !test_bit(Faulty, &rdev->flags)) {
6817 rdev->recovery_offset = 0;
6819 hot_add_disk(mddev, rdev) == 0) {
6821 sprintf(nm, "rd%d", rdev->raid_disk);
6822 if (sysfs_create_link(&mddev->kobj,
6825 "md: cannot register "
6829 md_new_event(mddev);
6830 set_bit(MD_CHANGE_DEVS, &mddev->flags);
6839 * This routine is regularly called by all per-raid-array threads to
6840 * deal with generic issues like resync and super-block update.
6841 * Raid personalities that don't have a thread (linear/raid0) do not
6842 * need this as they never do any recovery or update the superblock.
6844 * It does not do any resync itself, but rather "forks" off other threads
6845 * to do that as needed.
6846 * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
6847 * "->recovery" and create a thread at ->sync_thread.
6848 * When the thread finishes it sets MD_RECOVERY_DONE
6849 * and wakeups up this thread which will reap the thread and finish up.
6850 * This thread also removes any faulty devices (with nr_pending == 0).
6852 * The overall approach is:
6853 * 1/ if the superblock needs updating, update it.
6854 * 2/ If a recovery thread is running, don't do anything else.
6855 * 3/ If recovery has finished, clean up, possibly marking spares active.
6856 * 4/ If there are any faulty devices, remove them.
6857 * 5/ If array is degraded, try to add spares devices
6858 * 6/ If array has spares or is not in-sync, start a resync thread.
6860 void md_check_recovery(mddev_t *mddev)
6866 bitmap_daemon_work(mddev);
6871 if (signal_pending(current)) {
6872 if (mddev->pers->sync_request && !mddev->external) {
6873 printk(KERN_INFO "md: %s in immediate safe mode\n",
6875 mddev->safemode = 2;
6877 flush_signals(current);
6880 if (mddev->ro && !test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
6883 (mddev->flags && !mddev->external) ||
6884 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
6885 test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
6886 (mddev->external == 0 && mddev->safemode == 1) ||
6887 (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending)
6888 && !mddev->in_sync && mddev->recovery_cp == MaxSector)
6892 if (mddev_trylock(mddev)) {
6896 /* Only thing we do on a ro array is remove
6899 remove_and_add_spares(mddev);
6900 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6904 if (!mddev->external) {
6906 spin_lock_irq(&mddev->write_lock);
6907 if (mddev->safemode &&
6908 !atomic_read(&mddev->writes_pending) &&
6910 mddev->recovery_cp == MaxSector) {
6913 if (mddev->persistent)
6914 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
6916 if (mddev->safemode == 1)
6917 mddev->safemode = 0;
6918 spin_unlock_irq(&mddev->write_lock);
6920 sysfs_notify_dirent(mddev->sysfs_state);
6924 md_update_sb(mddev, 0);
6926 list_for_each_entry(rdev, &mddev->disks, same_set)
6927 if (test_and_clear_bit(StateChanged, &rdev->flags))
6928 sysfs_notify_dirent(rdev->sysfs_state);
6931 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
6932 !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
6933 /* resync/recovery still happening */
6934 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6937 if (mddev->sync_thread) {
6938 /* resync has finished, collect result */
6939 md_unregister_thread(mddev->sync_thread);
6940 mddev->sync_thread = NULL;
6941 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
6942 !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
6944 /* activate any spares */
6945 if (mddev->pers->spare_active(mddev))
6946 sysfs_notify(&mddev->kobj, NULL,
6949 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
6950 mddev->pers->finish_reshape)
6951 mddev->pers->finish_reshape(mddev);
6952 md_update_sb(mddev, 1);
6954 /* if array is no-longer degraded, then any saved_raid_disk
6955 * information must be scrapped
6957 if (!mddev->degraded)
6958 list_for_each_entry(rdev, &mddev->disks, same_set)
6959 rdev->saved_raid_disk = -1;
6961 mddev->recovery = 0;
6962 /* flag recovery needed just to double check */
6963 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6964 sysfs_notify_dirent(mddev->sysfs_action);
6965 md_new_event(mddev);
6968 /* Set RUNNING before clearing NEEDED to avoid
6969 * any transients in the value of "sync_action".
6971 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
6972 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6973 /* Clear some bits that don't mean anything, but
6976 clear_bit(MD_RECOVERY_INTR, &mddev->recovery);
6977 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
6979 if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
6981 /* no recovery is running.
6982 * remove any failed drives, then
6983 * add spares if possible.
6984 * Spare are also removed and re-added, to allow
6985 * the personality to fail the re-add.
6988 if (mddev->reshape_position != MaxSector) {
6989 if (mddev->pers->check_reshape == NULL ||
6990 mddev->pers->check_reshape(mddev) != 0)
6991 /* Cannot proceed */
6993 set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
6994 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
6995 } else if ((spares = remove_and_add_spares(mddev))) {
6996 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
6997 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
6998 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
6999 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
7000 } else if (mddev->recovery_cp < MaxSector) {
7001 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
7002 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
7003 } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
7004 /* nothing to be done ... */
7007 if (mddev->pers->sync_request) {
7008 if (spares && mddev->bitmap && ! mddev->bitmap->file) {
7009 /* We are adding a device or devices to an array
7010 * which has the bitmap stored on all devices.
7011 * So make sure all bitmap pages get written
7013 bitmap_write_all(mddev->bitmap);
7015 mddev->sync_thread = md_register_thread(md_do_sync,
7018 if (!mddev->sync_thread) {
7019 printk(KERN_ERR "%s: could not start resync"
7022 /* leave the spares where they are, it shouldn't hurt */
7023 mddev->recovery = 0;
7025 md_wakeup_thread(mddev->sync_thread);
7026 sysfs_notify_dirent(mddev->sysfs_action);
7027 md_new_event(mddev);
7030 if (!mddev->sync_thread) {
7031 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
7032 if (test_and_clear_bit(MD_RECOVERY_RECOVER,
7034 if (mddev->sysfs_action)
7035 sysfs_notify_dirent(mddev->sysfs_action);
7037 mddev_unlock(mddev);
7041 void md_wait_for_blocked_rdev(mdk_rdev_t *rdev, mddev_t *mddev)
7043 sysfs_notify_dirent(rdev->sysfs_state);
7044 wait_event_timeout(rdev->blocked_wait,
7045 !test_bit(Blocked, &rdev->flags),
7046 msecs_to_jiffies(5000));
7047 rdev_dec_pending(rdev, mddev);
7049 EXPORT_SYMBOL(md_wait_for_blocked_rdev);
7051 static int md_notify_reboot(struct notifier_block *this,
7052 unsigned long code, void *x)
7054 struct list_head *tmp;
7057 if ((code == SYS_DOWN) || (code == SYS_HALT) || (code == SYS_POWER_OFF)) {
7059 printk(KERN_INFO "md: stopping all md devices.\n");
7061 for_each_mddev(mddev, tmp)
7062 if (mddev_trylock(mddev)) {
7063 /* Force a switch to readonly even array
7064 * appears to still be in use. Hence
7067 do_md_stop(mddev, 1, 100);
7068 mddev_unlock(mddev);
7071 * certain more exotic SCSI devices are known to be
7072 * volatile wrt too early system reboots. While the
7073 * right place to handle this issue is the given
7074 * driver, we do want to have a safe RAID driver ...
7081 static struct notifier_block md_notifier = {
7082 .notifier_call = md_notify_reboot,
7084 .priority = INT_MAX, /* before any real devices */
7087 static void md_geninit(void)
7089 dprintk("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
7091 proc_create("mdstat", S_IRUGO, NULL, &md_seq_fops);
7094 static int __init md_init(void)
7096 if (register_blkdev(MD_MAJOR, "md"))
7098 if ((mdp_major=register_blkdev(0, "mdp"))<=0) {
7099 unregister_blkdev(MD_MAJOR, "md");
7102 blk_register_region(MKDEV(MD_MAJOR, 0), 1UL<<MINORBITS, THIS_MODULE,
7103 md_probe, NULL, NULL);
7104 blk_register_region(MKDEV(mdp_major, 0), 1UL<<MINORBITS, THIS_MODULE,
7105 md_probe, NULL, NULL);
7107 register_reboot_notifier(&md_notifier);
7108 raid_table_header = register_sysctl_table(raid_root_table);
7118 * Searches all registered partitions for autorun RAID arrays
7122 static LIST_HEAD(all_detected_devices);
7123 struct detected_devices_node {
7124 struct list_head list;
7128 void md_autodetect_dev(dev_t dev)
7130 struct detected_devices_node *node_detected_dev;
7132 node_detected_dev = kzalloc(sizeof(*node_detected_dev), GFP_KERNEL);
7133 if (node_detected_dev) {
7134 node_detected_dev->dev = dev;
7135 list_add_tail(&node_detected_dev->list, &all_detected_devices);
7137 printk(KERN_CRIT "md: md_autodetect_dev: kzalloc failed"
7138 ", skipping dev(%d,%d)\n", MAJOR(dev), MINOR(dev));
7143 static void autostart_arrays(int part)
7146 struct detected_devices_node *node_detected_dev;
7148 int i_scanned, i_passed;
7153 printk(KERN_INFO "md: Autodetecting RAID arrays.\n");
7155 while (!list_empty(&all_detected_devices) && i_scanned < INT_MAX) {
7157 node_detected_dev = list_entry(all_detected_devices.next,
7158 struct detected_devices_node, list);
7159 list_del(&node_detected_dev->list);
7160 dev = node_detected_dev->dev;
7161 kfree(node_detected_dev);
7162 rdev = md_import_device(dev,0, 90);
7166 if (test_bit(Faulty, &rdev->flags)) {
7170 set_bit(AutoDetected, &rdev->flags);
7171 list_add(&rdev->same_set, &pending_raid_disks);
7175 printk(KERN_INFO "md: Scanned %d and added %d devices.\n",
7176 i_scanned, i_passed);
7178 autorun_devices(part);
7181 #endif /* !MODULE */
7183 static __exit void md_exit(void)
7186 struct list_head *tmp;
7188 blk_unregister_region(MKDEV(MD_MAJOR,0), 1U << MINORBITS);
7189 blk_unregister_region(MKDEV(mdp_major,0), 1U << MINORBITS);
7191 unregister_blkdev(MD_MAJOR,"md");
7192 unregister_blkdev(mdp_major, "mdp");
7193 unregister_reboot_notifier(&md_notifier);
7194 unregister_sysctl_table(raid_table_header);
7195 remove_proc_entry("mdstat", NULL);
7196 for_each_mddev(mddev, tmp) {
7197 export_array(mddev);
7198 mddev->hold_active = 0;
7202 subsys_initcall(md_init);
7203 module_exit(md_exit)
7205 static int get_ro(char *buffer, struct kernel_param *kp)
7207 return sprintf(buffer, "%d", start_readonly);
7209 static int set_ro(const char *val, struct kernel_param *kp)
7212 int num = simple_strtoul(val, &e, 10);
7213 if (*val && (*e == '\0' || *e == '\n')) {
7214 start_readonly = num;
7220 module_param_call(start_ro, set_ro, get_ro, NULL, S_IRUSR|S_IWUSR);
7221 module_param(start_dirty_degraded, int, S_IRUGO|S_IWUSR);
7223 module_param_call(new_array, add_named_array, NULL, NULL, S_IWUSR);
7225 EXPORT_SYMBOL(register_md_personality);
7226 EXPORT_SYMBOL(unregister_md_personality);
7227 EXPORT_SYMBOL(md_error);
7228 EXPORT_SYMBOL(md_done_sync);
7229 EXPORT_SYMBOL(md_write_start);
7230 EXPORT_SYMBOL(md_write_end);
7231 EXPORT_SYMBOL(md_register_thread);
7232 EXPORT_SYMBOL(md_unregister_thread);
7233 EXPORT_SYMBOL(md_wakeup_thread);
7234 EXPORT_SYMBOL(md_check_recovery);
7235 MODULE_LICENSE("GPL");
7236 MODULE_DESCRIPTION("MD RAID framework");
7238 MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR);