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/delay.h>
48 #include <linux/raid/md_p.h>
49 #include <linux/raid/md_u.h>
54 #define dprintk(x...) ((void)(DEBUG && printk(x)))
58 static void autostart_arrays(int part);
61 static LIST_HEAD(pers_list);
62 static DEFINE_SPINLOCK(pers_lock);
64 static void md_print_devices(void);
66 static DECLARE_WAIT_QUEUE_HEAD(resync_wait);
68 #define MD_BUG(x...) { printk("md: bug in file %s, line %d\n", __FILE__, __LINE__); md_print_devices(); }
71 * Current RAID-1,4,5 parallel reconstruction 'guaranteed speed limit'
72 * is 1000 KB/sec, so the extra system load does not show up that much.
73 * Increase it if you want to have more _guaranteed_ speed. Note that
74 * the RAID driver will use the maximum available bandwidth if the IO
75 * subsystem is idle. There is also an 'absolute maximum' reconstruction
76 * speed limit - in case reconstruction slows down your system despite
79 * you can change it via /proc/sys/dev/raid/speed_limit_min and _max.
80 * or /sys/block/mdX/md/sync_speed_{min,max}
83 static int sysctl_speed_limit_min = 1000;
84 static int sysctl_speed_limit_max = 200000;
85 static inline int speed_min(mddev_t *mddev)
87 return mddev->sync_speed_min ?
88 mddev->sync_speed_min : sysctl_speed_limit_min;
91 static inline int speed_max(mddev_t *mddev)
93 return mddev->sync_speed_max ?
94 mddev->sync_speed_max : sysctl_speed_limit_max;
97 static struct ctl_table_header *raid_table_header;
99 static ctl_table raid_table[] = {
101 .procname = "speed_limit_min",
102 .data = &sysctl_speed_limit_min,
103 .maxlen = sizeof(int),
104 .mode = S_IRUGO|S_IWUSR,
105 .proc_handler = proc_dointvec,
108 .procname = "speed_limit_max",
109 .data = &sysctl_speed_limit_max,
110 .maxlen = sizeof(int),
111 .mode = S_IRUGO|S_IWUSR,
112 .proc_handler = proc_dointvec,
117 static ctl_table raid_dir_table[] = {
121 .mode = S_IRUGO|S_IXUGO,
127 static ctl_table raid_root_table[] = {
132 .child = raid_dir_table,
137 static const struct block_device_operations md_fops;
139 static int start_readonly;
142 * We have a system wide 'event count' that is incremented
143 * on any 'interesting' event, and readers of /proc/mdstat
144 * can use 'poll' or 'select' to find out when the event
148 * start array, stop array, error, add device, remove device,
149 * start build, activate spare
151 static DECLARE_WAIT_QUEUE_HEAD(md_event_waiters);
152 static atomic_t md_event_count;
153 void md_new_event(mddev_t *mddev)
155 atomic_inc(&md_event_count);
156 wake_up(&md_event_waiters);
158 EXPORT_SYMBOL_GPL(md_new_event);
160 /* Alternate version that can be called from interrupts
161 * when calling sysfs_notify isn't needed.
163 static void md_new_event_inintr(mddev_t *mddev)
165 atomic_inc(&md_event_count);
166 wake_up(&md_event_waiters);
170 * Enables to iterate over all existing md arrays
171 * all_mddevs_lock protects this list.
173 static LIST_HEAD(all_mddevs);
174 static DEFINE_SPINLOCK(all_mddevs_lock);
178 * iterates through all used mddevs in the system.
179 * We take care to grab the all_mddevs_lock whenever navigating
180 * the list, and to always hold a refcount when unlocked.
181 * Any code which breaks out of this loop while own
182 * a reference to the current mddev and must mddev_put it.
184 #define for_each_mddev(mddev,tmp) \
186 for (({ spin_lock(&all_mddevs_lock); \
187 tmp = all_mddevs.next; \
189 ({ if (tmp != &all_mddevs) \
190 mddev_get(list_entry(tmp, mddev_t, all_mddevs));\
191 spin_unlock(&all_mddevs_lock); \
192 if (mddev) mddev_put(mddev); \
193 mddev = list_entry(tmp, mddev_t, all_mddevs); \
194 tmp != &all_mddevs;}); \
195 ({ spin_lock(&all_mddevs_lock); \
200 /* Rather than calling directly into the personality make_request function,
201 * IO requests come here first so that we can check if the device is
202 * being suspended pending a reconfiguration.
203 * We hold a refcount over the call to ->make_request. By the time that
204 * call has finished, the bio has been linked into some internal structure
205 * and so is visible to ->quiesce(), so we don't need the refcount any more.
207 static int md_make_request(struct request_queue *q, struct bio *bio)
209 mddev_t *mddev = q->queuedata;
211 if (mddev == NULL || mddev->pers == NULL) {
216 if (mddev->suspended) {
219 prepare_to_wait(&mddev->sb_wait, &__wait,
220 TASK_UNINTERRUPTIBLE);
221 if (!mddev->suspended)
227 finish_wait(&mddev->sb_wait, &__wait);
229 atomic_inc(&mddev->active_io);
231 rv = mddev->pers->make_request(q, bio);
232 if (atomic_dec_and_test(&mddev->active_io) && mddev->suspended)
233 wake_up(&mddev->sb_wait);
238 static void mddev_suspend(mddev_t *mddev)
240 BUG_ON(mddev->suspended);
241 mddev->suspended = 1;
243 wait_event(mddev->sb_wait, atomic_read(&mddev->active_io) == 0);
244 mddev->pers->quiesce(mddev, 1);
245 md_unregister_thread(mddev->thread);
246 mddev->thread = NULL;
247 /* we now know that no code is executing in the personality module,
248 * except possibly the tail end of a ->bi_end_io function, but that
249 * is certain to complete before the module has a chance to get
254 static void mddev_resume(mddev_t *mddev)
256 mddev->suspended = 0;
257 wake_up(&mddev->sb_wait);
258 mddev->pers->quiesce(mddev, 0);
261 int mddev_congested(mddev_t *mddev, int bits)
263 return mddev->suspended;
265 EXPORT_SYMBOL(mddev_congested);
268 static inline mddev_t *mddev_get(mddev_t *mddev)
270 atomic_inc(&mddev->active);
274 static void mddev_delayed_delete(struct work_struct *ws);
276 static void mddev_put(mddev_t *mddev)
278 if (!atomic_dec_and_lock(&mddev->active, &all_mddevs_lock))
280 if (!mddev->raid_disks && list_empty(&mddev->disks) &&
281 !mddev->hold_active) {
282 list_del(&mddev->all_mddevs);
283 if (mddev->gendisk) {
284 /* we did a probe so need to clean up.
285 * Call schedule_work inside the spinlock
286 * so that flush_scheduled_work() after
287 * mddev_find will succeed in waiting for the
290 INIT_WORK(&mddev->del_work, mddev_delayed_delete);
291 schedule_work(&mddev->del_work);
295 spin_unlock(&all_mddevs_lock);
298 static mddev_t * mddev_find(dev_t unit)
300 mddev_t *mddev, *new = NULL;
303 spin_lock(&all_mddevs_lock);
306 list_for_each_entry(mddev, &all_mddevs, all_mddevs)
307 if (mddev->unit == unit) {
309 spin_unlock(&all_mddevs_lock);
315 list_add(&new->all_mddevs, &all_mddevs);
316 spin_unlock(&all_mddevs_lock);
317 new->hold_active = UNTIL_IOCTL;
321 /* find an unused unit number */
322 static int next_minor = 512;
323 int start = next_minor;
327 dev = MKDEV(MD_MAJOR, next_minor);
329 if (next_minor > MINORMASK)
331 if (next_minor == start) {
332 /* Oh dear, all in use. */
333 spin_unlock(&all_mddevs_lock);
339 list_for_each_entry(mddev, &all_mddevs, all_mddevs)
340 if (mddev->unit == dev) {
346 new->md_minor = MINOR(dev);
347 new->hold_active = UNTIL_STOP;
348 list_add(&new->all_mddevs, &all_mddevs);
349 spin_unlock(&all_mddevs_lock);
352 spin_unlock(&all_mddevs_lock);
354 new = kzalloc(sizeof(*new), GFP_KERNEL);
359 if (MAJOR(unit) == MD_MAJOR)
360 new->md_minor = MINOR(unit);
362 new->md_minor = MINOR(unit) >> MdpMinorShift;
364 mutex_init(&new->open_mutex);
365 mutex_init(&new->reconfig_mutex);
366 mutex_init(&new->bitmap_mutex);
367 INIT_LIST_HEAD(&new->disks);
368 INIT_LIST_HEAD(&new->all_mddevs);
369 init_timer(&new->safemode_timer);
370 atomic_set(&new->active, 1);
371 atomic_set(&new->openers, 0);
372 atomic_set(&new->active_io, 0);
373 spin_lock_init(&new->write_lock);
374 init_waitqueue_head(&new->sb_wait);
375 init_waitqueue_head(&new->recovery_wait);
376 new->reshape_position = MaxSector;
378 new->resync_max = MaxSector;
379 new->level = LEVEL_NONE;
384 static inline int mddev_lock(mddev_t * mddev)
386 return mutex_lock_interruptible(&mddev->reconfig_mutex);
389 static inline int mddev_is_locked(mddev_t *mddev)
391 return mutex_is_locked(&mddev->reconfig_mutex);
394 static inline int mddev_trylock(mddev_t * mddev)
396 return mutex_trylock(&mddev->reconfig_mutex);
399 static inline void mddev_unlock(mddev_t * mddev)
401 mutex_unlock(&mddev->reconfig_mutex);
403 md_wakeup_thread(mddev->thread);
406 static mdk_rdev_t * find_rdev_nr(mddev_t *mddev, int nr)
410 list_for_each_entry(rdev, &mddev->disks, same_set)
411 if (rdev->desc_nr == nr)
417 static mdk_rdev_t * find_rdev(mddev_t * mddev, dev_t dev)
421 list_for_each_entry(rdev, &mddev->disks, same_set)
422 if (rdev->bdev->bd_dev == dev)
428 static struct mdk_personality *find_pers(int level, char *clevel)
430 struct mdk_personality *pers;
431 list_for_each_entry(pers, &pers_list, list) {
432 if (level != LEVEL_NONE && pers->level == level)
434 if (strcmp(pers->name, clevel)==0)
440 /* return the offset of the super block in 512byte sectors */
441 static inline sector_t calc_dev_sboffset(struct block_device *bdev)
443 sector_t num_sectors = bdev->bd_inode->i_size / 512;
444 return MD_NEW_SIZE_SECTORS(num_sectors);
447 static int alloc_disk_sb(mdk_rdev_t * rdev)
452 rdev->sb_page = alloc_page(GFP_KERNEL);
453 if (!rdev->sb_page) {
454 printk(KERN_ALERT "md: out of memory.\n");
461 static void free_disk_sb(mdk_rdev_t * rdev)
464 put_page(rdev->sb_page);
466 rdev->sb_page = NULL;
473 static void super_written(struct bio *bio, int error)
475 mdk_rdev_t *rdev = bio->bi_private;
476 mddev_t *mddev = rdev->mddev;
478 if (error || !test_bit(BIO_UPTODATE, &bio->bi_flags)) {
479 printk("md: super_written gets error=%d, uptodate=%d\n",
480 error, test_bit(BIO_UPTODATE, &bio->bi_flags));
481 WARN_ON(test_bit(BIO_UPTODATE, &bio->bi_flags));
482 md_error(mddev, rdev);
485 if (atomic_dec_and_test(&mddev->pending_writes))
486 wake_up(&mddev->sb_wait);
490 static void super_written_barrier(struct bio *bio, int error)
492 struct bio *bio2 = bio->bi_private;
493 mdk_rdev_t *rdev = bio2->bi_private;
494 mddev_t *mddev = rdev->mddev;
496 if (!test_bit(BIO_UPTODATE, &bio->bi_flags) &&
497 error == -EOPNOTSUPP) {
499 /* barriers don't appear to be supported :-( */
500 set_bit(BarriersNotsupp, &rdev->flags);
501 mddev->barriers_work = 0;
502 spin_lock_irqsave(&mddev->write_lock, flags);
503 bio2->bi_next = mddev->biolist;
504 mddev->biolist = bio2;
505 spin_unlock_irqrestore(&mddev->write_lock, flags);
506 wake_up(&mddev->sb_wait);
510 bio->bi_private = rdev;
511 super_written(bio, error);
515 void md_super_write(mddev_t *mddev, mdk_rdev_t *rdev,
516 sector_t sector, int size, struct page *page)
518 /* write first size bytes of page to sector of rdev
519 * Increment mddev->pending_writes before returning
520 * and decrement it on completion, waking up sb_wait
521 * if zero is reached.
522 * If an error occurred, call md_error
524 * As we might need to resubmit the request if BIO_RW_BARRIER
525 * causes ENOTSUPP, we allocate a spare bio...
527 struct bio *bio = bio_alloc(GFP_NOIO, 1);
528 int rw = (1<<BIO_RW) | (1<<BIO_RW_SYNCIO) | (1<<BIO_RW_UNPLUG);
530 bio->bi_bdev = rdev->bdev;
531 bio->bi_sector = sector;
532 bio_add_page(bio, page, size, 0);
533 bio->bi_private = rdev;
534 bio->bi_end_io = super_written;
537 atomic_inc(&mddev->pending_writes);
538 if (!test_bit(BarriersNotsupp, &rdev->flags)) {
540 rw |= (1<<BIO_RW_BARRIER);
541 rbio = bio_clone(bio, GFP_NOIO);
542 rbio->bi_private = bio;
543 rbio->bi_end_io = super_written_barrier;
544 submit_bio(rw, rbio);
549 void md_super_wait(mddev_t *mddev)
551 /* wait for all superblock writes that were scheduled to complete.
552 * if any had to be retried (due to BARRIER problems), retry them
556 prepare_to_wait(&mddev->sb_wait, &wq, TASK_UNINTERRUPTIBLE);
557 if (atomic_read(&mddev->pending_writes)==0)
559 while (mddev->biolist) {
561 spin_lock_irq(&mddev->write_lock);
562 bio = mddev->biolist;
563 mddev->biolist = bio->bi_next ;
565 spin_unlock_irq(&mddev->write_lock);
566 submit_bio(bio->bi_rw, bio);
570 finish_wait(&mddev->sb_wait, &wq);
573 static void bi_complete(struct bio *bio, int error)
575 complete((struct completion*)bio->bi_private);
578 int sync_page_io(struct block_device *bdev, sector_t sector, int size,
579 struct page *page, int rw)
581 struct bio *bio = bio_alloc(GFP_NOIO, 1);
582 struct completion event;
585 rw |= (1 << BIO_RW_SYNCIO) | (1 << BIO_RW_UNPLUG);
588 bio->bi_sector = sector;
589 bio_add_page(bio, page, size, 0);
590 init_completion(&event);
591 bio->bi_private = &event;
592 bio->bi_end_io = bi_complete;
594 wait_for_completion(&event);
596 ret = test_bit(BIO_UPTODATE, &bio->bi_flags);
600 EXPORT_SYMBOL_GPL(sync_page_io);
602 static int read_disk_sb(mdk_rdev_t * rdev, int size)
604 char b[BDEVNAME_SIZE];
605 if (!rdev->sb_page) {
613 if (!sync_page_io(rdev->bdev, rdev->sb_start, size, rdev->sb_page, READ))
619 printk(KERN_WARNING "md: disabled device %s, could not read superblock.\n",
620 bdevname(rdev->bdev,b));
624 static int uuid_equal(mdp_super_t *sb1, mdp_super_t *sb2)
626 return sb1->set_uuid0 == sb2->set_uuid0 &&
627 sb1->set_uuid1 == sb2->set_uuid1 &&
628 sb1->set_uuid2 == sb2->set_uuid2 &&
629 sb1->set_uuid3 == sb2->set_uuid3;
632 static int sb_equal(mdp_super_t *sb1, mdp_super_t *sb2)
635 mdp_super_t *tmp1, *tmp2;
637 tmp1 = kmalloc(sizeof(*tmp1),GFP_KERNEL);
638 tmp2 = kmalloc(sizeof(*tmp2),GFP_KERNEL);
640 if (!tmp1 || !tmp2) {
642 printk(KERN_INFO "md.c sb_equal(): failed to allocate memory!\n");
650 * nr_disks is not constant
655 ret = (memcmp(tmp1, tmp2, MD_SB_GENERIC_CONSTANT_WORDS * 4) == 0);
663 static u32 md_csum_fold(u32 csum)
665 csum = (csum & 0xffff) + (csum >> 16);
666 return (csum & 0xffff) + (csum >> 16);
669 static unsigned int calc_sb_csum(mdp_super_t * sb)
672 u32 *sb32 = (u32*)sb;
674 unsigned int disk_csum, csum;
676 disk_csum = sb->sb_csum;
679 for (i = 0; i < MD_SB_BYTES/4 ; i++)
681 csum = (newcsum & 0xffffffff) + (newcsum>>32);
685 /* This used to use csum_partial, which was wrong for several
686 * reasons including that different results are returned on
687 * different architectures. It isn't critical that we get exactly
688 * the same return value as before (we always csum_fold before
689 * testing, and that removes any differences). However as we
690 * know that csum_partial always returned a 16bit value on
691 * alphas, do a fold to maximise conformity to previous behaviour.
693 sb->sb_csum = md_csum_fold(disk_csum);
695 sb->sb_csum = disk_csum;
702 * Handle superblock details.
703 * We want to be able to handle multiple superblock formats
704 * so we have a common interface to them all, and an array of
705 * different handlers.
706 * We rely on user-space to write the initial superblock, and support
707 * reading and updating of superblocks.
708 * Interface methods are:
709 * int load_super(mdk_rdev_t *dev, mdk_rdev_t *refdev, int minor_version)
710 * loads and validates a superblock on dev.
711 * if refdev != NULL, compare superblocks on both devices
713 * 0 - dev has a superblock that is compatible with refdev
714 * 1 - dev has a superblock that is compatible and newer than refdev
715 * so dev should be used as the refdev in future
716 * -EINVAL superblock incompatible or invalid
717 * -othererror e.g. -EIO
719 * int validate_super(mddev_t *mddev, mdk_rdev_t *dev)
720 * Verify that dev is acceptable into mddev.
721 * The first time, mddev->raid_disks will be 0, and data from
722 * dev should be merged in. Subsequent calls check that dev
723 * is new enough. Return 0 or -EINVAL
725 * void sync_super(mddev_t *mddev, mdk_rdev_t *dev)
726 * Update the superblock for rdev with data in mddev
727 * This does not write to disc.
733 struct module *owner;
734 int (*load_super)(mdk_rdev_t *rdev, mdk_rdev_t *refdev,
736 int (*validate_super)(mddev_t *mddev, mdk_rdev_t *rdev);
737 void (*sync_super)(mddev_t *mddev, mdk_rdev_t *rdev);
738 unsigned long long (*rdev_size_change)(mdk_rdev_t *rdev,
739 sector_t num_sectors);
743 * Check that the given mddev has no bitmap.
745 * This function is called from the run method of all personalities that do not
746 * support bitmaps. It prints an error message and returns non-zero if mddev
747 * has a bitmap. Otherwise, it returns 0.
750 int md_check_no_bitmap(mddev_t *mddev)
752 if (!mddev->bitmap_file && !mddev->bitmap_offset)
754 printk(KERN_ERR "%s: bitmaps are not supported for %s\n",
755 mdname(mddev), mddev->pers->name);
758 EXPORT_SYMBOL(md_check_no_bitmap);
761 * load_super for 0.90.0
763 static int super_90_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
765 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
770 * Calculate the position of the superblock (512byte sectors),
771 * it's at the end of the disk.
773 * It also happens to be a multiple of 4Kb.
775 rdev->sb_start = calc_dev_sboffset(rdev->bdev);
777 ret = read_disk_sb(rdev, MD_SB_BYTES);
782 bdevname(rdev->bdev, b);
783 sb = (mdp_super_t*)page_address(rdev->sb_page);
785 if (sb->md_magic != MD_SB_MAGIC) {
786 printk(KERN_ERR "md: invalid raid superblock magic on %s\n",
791 if (sb->major_version != 0 ||
792 sb->minor_version < 90 ||
793 sb->minor_version > 91) {
794 printk(KERN_WARNING "Bad version number %d.%d on %s\n",
795 sb->major_version, sb->minor_version,
800 if (sb->raid_disks <= 0)
803 if (md_csum_fold(calc_sb_csum(sb)) != md_csum_fold(sb->sb_csum)) {
804 printk(KERN_WARNING "md: invalid superblock checksum on %s\n",
809 rdev->preferred_minor = sb->md_minor;
810 rdev->data_offset = 0;
811 rdev->sb_size = MD_SB_BYTES;
813 if (sb->level == LEVEL_MULTIPATH)
816 rdev->desc_nr = sb->this_disk.number;
822 mdp_super_t *refsb = (mdp_super_t*)page_address(refdev->sb_page);
823 if (!uuid_equal(refsb, sb)) {
824 printk(KERN_WARNING "md: %s has different UUID to %s\n",
825 b, bdevname(refdev->bdev,b2));
828 if (!sb_equal(refsb, sb)) {
829 printk(KERN_WARNING "md: %s has same UUID"
830 " but different superblock to %s\n",
831 b, bdevname(refdev->bdev, b2));
835 ev2 = md_event(refsb);
841 rdev->sectors = rdev->sb_start;
843 if (rdev->sectors < sb->size * 2 && sb->level > 1)
844 /* "this cannot possibly happen" ... */
852 * validate_super for 0.90.0
854 static int super_90_validate(mddev_t *mddev, mdk_rdev_t *rdev)
857 mdp_super_t *sb = (mdp_super_t *)page_address(rdev->sb_page);
858 __u64 ev1 = md_event(sb);
860 rdev->raid_disk = -1;
861 clear_bit(Faulty, &rdev->flags);
862 clear_bit(In_sync, &rdev->flags);
863 clear_bit(WriteMostly, &rdev->flags);
864 clear_bit(BarriersNotsupp, &rdev->flags);
866 if (mddev->raid_disks == 0) {
867 mddev->major_version = 0;
868 mddev->minor_version = sb->minor_version;
869 mddev->patch_version = sb->patch_version;
871 mddev->chunk_sectors = sb->chunk_size >> 9;
872 mddev->ctime = sb->ctime;
873 mddev->utime = sb->utime;
874 mddev->level = sb->level;
875 mddev->clevel[0] = 0;
876 mddev->layout = sb->layout;
877 mddev->raid_disks = sb->raid_disks;
878 mddev->dev_sectors = sb->size * 2;
880 mddev->bitmap_offset = 0;
881 mddev->default_bitmap_offset = MD_SB_BYTES >> 9;
883 if (mddev->minor_version >= 91) {
884 mddev->reshape_position = sb->reshape_position;
885 mddev->delta_disks = sb->delta_disks;
886 mddev->new_level = sb->new_level;
887 mddev->new_layout = sb->new_layout;
888 mddev->new_chunk_sectors = sb->new_chunk >> 9;
890 mddev->reshape_position = MaxSector;
891 mddev->delta_disks = 0;
892 mddev->new_level = mddev->level;
893 mddev->new_layout = mddev->layout;
894 mddev->new_chunk_sectors = mddev->chunk_sectors;
897 if (sb->state & (1<<MD_SB_CLEAN))
898 mddev->recovery_cp = MaxSector;
900 if (sb->events_hi == sb->cp_events_hi &&
901 sb->events_lo == sb->cp_events_lo) {
902 mddev->recovery_cp = sb->recovery_cp;
904 mddev->recovery_cp = 0;
907 memcpy(mddev->uuid+0, &sb->set_uuid0, 4);
908 memcpy(mddev->uuid+4, &sb->set_uuid1, 4);
909 memcpy(mddev->uuid+8, &sb->set_uuid2, 4);
910 memcpy(mddev->uuid+12,&sb->set_uuid3, 4);
912 mddev->max_disks = MD_SB_DISKS;
914 if (sb->state & (1<<MD_SB_BITMAP_PRESENT) &&
915 mddev->bitmap_file == NULL)
916 mddev->bitmap_offset = mddev->default_bitmap_offset;
918 } else if (mddev->pers == NULL) {
919 /* Insist on good event counter while assembling */
921 if (ev1 < mddev->events)
923 } else if (mddev->bitmap) {
924 /* if adding to array with a bitmap, then we can accept an
925 * older device ... but not too old.
927 if (ev1 < mddev->bitmap->events_cleared)
930 if (ev1 < mddev->events)
931 /* just a hot-add of a new device, leave raid_disk at -1 */
935 if (mddev->level != LEVEL_MULTIPATH) {
936 desc = sb->disks + rdev->desc_nr;
938 if (desc->state & (1<<MD_DISK_FAULTY))
939 set_bit(Faulty, &rdev->flags);
940 else if (desc->state & (1<<MD_DISK_SYNC) /* &&
941 desc->raid_disk < mddev->raid_disks */) {
942 set_bit(In_sync, &rdev->flags);
943 rdev->raid_disk = desc->raid_disk;
944 } else if (desc->state & (1<<MD_DISK_ACTIVE)) {
945 /* active but not in sync implies recovery up to
946 * reshape position. We don't know exactly where
947 * that is, so set to zero for now */
948 if (mddev->minor_version >= 91) {
949 rdev->recovery_offset = 0;
950 rdev->raid_disk = desc->raid_disk;
953 if (desc->state & (1<<MD_DISK_WRITEMOSTLY))
954 set_bit(WriteMostly, &rdev->flags);
955 } else /* MULTIPATH are always insync */
956 set_bit(In_sync, &rdev->flags);
961 * sync_super for 0.90.0
963 static void super_90_sync(mddev_t *mddev, mdk_rdev_t *rdev)
967 int next_spare = mddev->raid_disks;
970 /* make rdev->sb match mddev data..
973 * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
974 * 3/ any empty disks < next_spare become removed
976 * disks[0] gets initialised to REMOVED because
977 * we cannot be sure from other fields if it has
978 * been initialised or not.
981 int active=0, working=0,failed=0,spare=0,nr_disks=0;
983 rdev->sb_size = MD_SB_BYTES;
985 sb = (mdp_super_t*)page_address(rdev->sb_page);
987 memset(sb, 0, sizeof(*sb));
989 sb->md_magic = MD_SB_MAGIC;
990 sb->major_version = mddev->major_version;
991 sb->patch_version = mddev->patch_version;
992 sb->gvalid_words = 0; /* ignored */
993 memcpy(&sb->set_uuid0, mddev->uuid+0, 4);
994 memcpy(&sb->set_uuid1, mddev->uuid+4, 4);
995 memcpy(&sb->set_uuid2, mddev->uuid+8, 4);
996 memcpy(&sb->set_uuid3, mddev->uuid+12,4);
998 sb->ctime = mddev->ctime;
999 sb->level = mddev->level;
1000 sb->size = mddev->dev_sectors / 2;
1001 sb->raid_disks = mddev->raid_disks;
1002 sb->md_minor = mddev->md_minor;
1003 sb->not_persistent = 0;
1004 sb->utime = mddev->utime;
1006 sb->events_hi = (mddev->events>>32);
1007 sb->events_lo = (u32)mddev->events;
1009 if (mddev->reshape_position == MaxSector)
1010 sb->minor_version = 90;
1012 sb->minor_version = 91;
1013 sb->reshape_position = mddev->reshape_position;
1014 sb->new_level = mddev->new_level;
1015 sb->delta_disks = mddev->delta_disks;
1016 sb->new_layout = mddev->new_layout;
1017 sb->new_chunk = mddev->new_chunk_sectors << 9;
1019 mddev->minor_version = sb->minor_version;
1022 sb->recovery_cp = mddev->recovery_cp;
1023 sb->cp_events_hi = (mddev->events>>32);
1024 sb->cp_events_lo = (u32)mddev->events;
1025 if (mddev->recovery_cp == MaxSector)
1026 sb->state = (1<< MD_SB_CLEAN);
1028 sb->recovery_cp = 0;
1030 sb->layout = mddev->layout;
1031 sb->chunk_size = mddev->chunk_sectors << 9;
1033 if (mddev->bitmap && mddev->bitmap_file == NULL)
1034 sb->state |= (1<<MD_SB_BITMAP_PRESENT);
1036 sb->disks[0].state = (1<<MD_DISK_REMOVED);
1037 list_for_each_entry(rdev2, &mddev->disks, same_set) {
1040 int is_active = test_bit(In_sync, &rdev2->flags);
1042 if (rdev2->raid_disk >= 0 &&
1043 sb->minor_version >= 91)
1044 /* we have nowhere to store the recovery_offset,
1045 * but if it is not below the reshape_position,
1046 * we can piggy-back on that.
1049 if (rdev2->raid_disk < 0 ||
1050 test_bit(Faulty, &rdev2->flags))
1053 desc_nr = rdev2->raid_disk;
1055 desc_nr = next_spare++;
1056 rdev2->desc_nr = desc_nr;
1057 d = &sb->disks[rdev2->desc_nr];
1059 d->number = rdev2->desc_nr;
1060 d->major = MAJOR(rdev2->bdev->bd_dev);
1061 d->minor = MINOR(rdev2->bdev->bd_dev);
1063 d->raid_disk = rdev2->raid_disk;
1065 d->raid_disk = rdev2->desc_nr; /* compatibility */
1066 if (test_bit(Faulty, &rdev2->flags))
1067 d->state = (1<<MD_DISK_FAULTY);
1068 else if (is_active) {
1069 d->state = (1<<MD_DISK_ACTIVE);
1070 if (test_bit(In_sync, &rdev2->flags))
1071 d->state |= (1<<MD_DISK_SYNC);
1079 if (test_bit(WriteMostly, &rdev2->flags))
1080 d->state |= (1<<MD_DISK_WRITEMOSTLY);
1082 /* now set the "removed" and "faulty" bits on any missing devices */
1083 for (i=0 ; i < mddev->raid_disks ; i++) {
1084 mdp_disk_t *d = &sb->disks[i];
1085 if (d->state == 0 && d->number == 0) {
1088 d->state = (1<<MD_DISK_REMOVED);
1089 d->state |= (1<<MD_DISK_FAULTY);
1093 sb->nr_disks = nr_disks;
1094 sb->active_disks = active;
1095 sb->working_disks = working;
1096 sb->failed_disks = failed;
1097 sb->spare_disks = spare;
1099 sb->this_disk = sb->disks[rdev->desc_nr];
1100 sb->sb_csum = calc_sb_csum(sb);
1104 * rdev_size_change for 0.90.0
1106 static unsigned long long
1107 super_90_rdev_size_change(mdk_rdev_t *rdev, sector_t num_sectors)
1109 if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
1110 return 0; /* component must fit device */
1111 if (rdev->mddev->bitmap_offset)
1112 return 0; /* can't move bitmap */
1113 rdev->sb_start = calc_dev_sboffset(rdev->bdev);
1114 if (!num_sectors || num_sectors > rdev->sb_start)
1115 num_sectors = rdev->sb_start;
1116 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
1118 md_super_wait(rdev->mddev);
1119 return num_sectors / 2; /* kB for sysfs */
1124 * version 1 superblock
1127 static __le32 calc_sb_1_csum(struct mdp_superblock_1 * sb)
1131 unsigned long long newcsum;
1132 int size = 256 + le32_to_cpu(sb->max_dev)*2;
1133 __le32 *isuper = (__le32*)sb;
1136 disk_csum = sb->sb_csum;
1139 for (i=0; size>=4; size -= 4 )
1140 newcsum += le32_to_cpu(*isuper++);
1143 newcsum += le16_to_cpu(*(__le16*) isuper);
1145 csum = (newcsum & 0xffffffff) + (newcsum >> 32);
1146 sb->sb_csum = disk_csum;
1147 return cpu_to_le32(csum);
1150 static int super_1_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
1152 struct mdp_superblock_1 *sb;
1155 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
1159 * Calculate the position of the superblock in 512byte sectors.
1160 * It is always aligned to a 4K boundary and
1161 * depeding on minor_version, it can be:
1162 * 0: At least 8K, but less than 12K, from end of device
1163 * 1: At start of device
1164 * 2: 4K from start of device.
1166 switch(minor_version) {
1168 sb_start = rdev->bdev->bd_inode->i_size >> 9;
1170 sb_start &= ~(sector_t)(4*2-1);
1181 rdev->sb_start = sb_start;
1183 /* superblock is rarely larger than 1K, but it can be larger,
1184 * and it is safe to read 4k, so we do that
1186 ret = read_disk_sb(rdev, 4096);
1187 if (ret) return ret;
1190 sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1192 if (sb->magic != cpu_to_le32(MD_SB_MAGIC) ||
1193 sb->major_version != cpu_to_le32(1) ||
1194 le32_to_cpu(sb->max_dev) > (4096-256)/2 ||
1195 le64_to_cpu(sb->super_offset) != rdev->sb_start ||
1196 (le32_to_cpu(sb->feature_map) & ~MD_FEATURE_ALL) != 0)
1199 if (calc_sb_1_csum(sb) != sb->sb_csum) {
1200 printk("md: invalid superblock checksum on %s\n",
1201 bdevname(rdev->bdev,b));
1204 if (le64_to_cpu(sb->data_size) < 10) {
1205 printk("md: data_size too small on %s\n",
1206 bdevname(rdev->bdev,b));
1210 rdev->preferred_minor = 0xffff;
1211 rdev->data_offset = le64_to_cpu(sb->data_offset);
1212 atomic_set(&rdev->corrected_errors, le32_to_cpu(sb->cnt_corrected_read));
1214 rdev->sb_size = le32_to_cpu(sb->max_dev) * 2 + 256;
1215 bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
1216 if (rdev->sb_size & bmask)
1217 rdev->sb_size = (rdev->sb_size | bmask) + 1;
1220 && rdev->data_offset < sb_start + (rdev->sb_size/512))
1223 if (sb->level == cpu_to_le32(LEVEL_MULTIPATH))
1226 rdev->desc_nr = le32_to_cpu(sb->dev_number);
1232 struct mdp_superblock_1 *refsb =
1233 (struct mdp_superblock_1*)page_address(refdev->sb_page);
1235 if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 ||
1236 sb->level != refsb->level ||
1237 sb->layout != refsb->layout ||
1238 sb->chunksize != refsb->chunksize) {
1239 printk(KERN_WARNING "md: %s has strangely different"
1240 " superblock to %s\n",
1241 bdevname(rdev->bdev,b),
1242 bdevname(refdev->bdev,b2));
1245 ev1 = le64_to_cpu(sb->events);
1246 ev2 = le64_to_cpu(refsb->events);
1254 rdev->sectors = (rdev->bdev->bd_inode->i_size >> 9) -
1255 le64_to_cpu(sb->data_offset);
1257 rdev->sectors = rdev->sb_start;
1258 if (rdev->sectors < le64_to_cpu(sb->data_size))
1260 rdev->sectors = le64_to_cpu(sb->data_size);
1261 if (le64_to_cpu(sb->size) > rdev->sectors)
1266 static int super_1_validate(mddev_t *mddev, mdk_rdev_t *rdev)
1268 struct mdp_superblock_1 *sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1269 __u64 ev1 = le64_to_cpu(sb->events);
1271 rdev->raid_disk = -1;
1272 clear_bit(Faulty, &rdev->flags);
1273 clear_bit(In_sync, &rdev->flags);
1274 clear_bit(WriteMostly, &rdev->flags);
1275 clear_bit(BarriersNotsupp, &rdev->flags);
1277 if (mddev->raid_disks == 0) {
1278 mddev->major_version = 1;
1279 mddev->patch_version = 0;
1280 mddev->external = 0;
1281 mddev->chunk_sectors = le32_to_cpu(sb->chunksize);
1282 mddev->ctime = le64_to_cpu(sb->ctime) & ((1ULL << 32)-1);
1283 mddev->utime = le64_to_cpu(sb->utime) & ((1ULL << 32)-1);
1284 mddev->level = le32_to_cpu(sb->level);
1285 mddev->clevel[0] = 0;
1286 mddev->layout = le32_to_cpu(sb->layout);
1287 mddev->raid_disks = le32_to_cpu(sb->raid_disks);
1288 mddev->dev_sectors = le64_to_cpu(sb->size);
1289 mddev->events = ev1;
1290 mddev->bitmap_offset = 0;
1291 mddev->default_bitmap_offset = 1024 >> 9;
1293 mddev->recovery_cp = le64_to_cpu(sb->resync_offset);
1294 memcpy(mddev->uuid, sb->set_uuid, 16);
1296 mddev->max_disks = (4096-256)/2;
1298 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) &&
1299 mddev->bitmap_file == NULL )
1300 mddev->bitmap_offset = (__s32)le32_to_cpu(sb->bitmap_offset);
1302 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) {
1303 mddev->reshape_position = le64_to_cpu(sb->reshape_position);
1304 mddev->delta_disks = le32_to_cpu(sb->delta_disks);
1305 mddev->new_level = le32_to_cpu(sb->new_level);
1306 mddev->new_layout = le32_to_cpu(sb->new_layout);
1307 mddev->new_chunk_sectors = le32_to_cpu(sb->new_chunk);
1309 mddev->reshape_position = MaxSector;
1310 mddev->delta_disks = 0;
1311 mddev->new_level = mddev->level;
1312 mddev->new_layout = mddev->layout;
1313 mddev->new_chunk_sectors = mddev->chunk_sectors;
1316 } else if (mddev->pers == NULL) {
1317 /* Insist of good event counter while assembling */
1319 if (ev1 < mddev->events)
1321 } else if (mddev->bitmap) {
1322 /* If adding to array with a bitmap, then we can accept an
1323 * older device, but not too old.
1325 if (ev1 < mddev->bitmap->events_cleared)
1328 if (ev1 < mddev->events)
1329 /* just a hot-add of a new device, leave raid_disk at -1 */
1332 if (mddev->level != LEVEL_MULTIPATH) {
1334 if (rdev->desc_nr < 0 ||
1335 rdev->desc_nr >= le32_to_cpu(sb->max_dev)) {
1339 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
1341 case 0xffff: /* spare */
1343 case 0xfffe: /* faulty */
1344 set_bit(Faulty, &rdev->flags);
1347 if ((le32_to_cpu(sb->feature_map) &
1348 MD_FEATURE_RECOVERY_OFFSET))
1349 rdev->recovery_offset = le64_to_cpu(sb->recovery_offset);
1351 set_bit(In_sync, &rdev->flags);
1352 rdev->raid_disk = role;
1355 if (sb->devflags & WriteMostly1)
1356 set_bit(WriteMostly, &rdev->flags);
1357 } else /* MULTIPATH are always insync */
1358 set_bit(In_sync, &rdev->flags);
1363 static void super_1_sync(mddev_t *mddev, mdk_rdev_t *rdev)
1365 struct mdp_superblock_1 *sb;
1368 /* make rdev->sb match mddev and rdev data. */
1370 sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1372 sb->feature_map = 0;
1374 sb->recovery_offset = cpu_to_le64(0);
1375 memset(sb->pad1, 0, sizeof(sb->pad1));
1376 memset(sb->pad2, 0, sizeof(sb->pad2));
1377 memset(sb->pad3, 0, sizeof(sb->pad3));
1379 sb->utime = cpu_to_le64((__u64)mddev->utime);
1380 sb->events = cpu_to_le64(mddev->events);
1382 sb->resync_offset = cpu_to_le64(mddev->recovery_cp);
1384 sb->resync_offset = cpu_to_le64(0);
1386 sb->cnt_corrected_read = cpu_to_le32(atomic_read(&rdev->corrected_errors));
1388 sb->raid_disks = cpu_to_le32(mddev->raid_disks);
1389 sb->size = cpu_to_le64(mddev->dev_sectors);
1390 sb->chunksize = cpu_to_le32(mddev->chunk_sectors);
1391 sb->level = cpu_to_le32(mddev->level);
1392 sb->layout = cpu_to_le32(mddev->layout);
1394 if (mddev->bitmap && mddev->bitmap_file == NULL) {
1395 sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_offset);
1396 sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET);
1399 if (rdev->raid_disk >= 0 &&
1400 !test_bit(In_sync, &rdev->flags)) {
1401 if (rdev->recovery_offset > 0) {
1403 cpu_to_le32(MD_FEATURE_RECOVERY_OFFSET);
1404 sb->recovery_offset =
1405 cpu_to_le64(rdev->recovery_offset);
1409 if (mddev->reshape_position != MaxSector) {
1410 sb->feature_map |= cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE);
1411 sb->reshape_position = cpu_to_le64(mddev->reshape_position);
1412 sb->new_layout = cpu_to_le32(mddev->new_layout);
1413 sb->delta_disks = cpu_to_le32(mddev->delta_disks);
1414 sb->new_level = cpu_to_le32(mddev->new_level);
1415 sb->new_chunk = cpu_to_le32(mddev->new_chunk_sectors);
1419 list_for_each_entry(rdev2, &mddev->disks, same_set)
1420 if (rdev2->desc_nr+1 > max_dev)
1421 max_dev = rdev2->desc_nr+1;
1423 if (max_dev > le32_to_cpu(sb->max_dev)) {
1425 sb->max_dev = cpu_to_le32(max_dev);
1426 rdev->sb_size = max_dev * 2 + 256;
1427 bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
1428 if (rdev->sb_size & bmask)
1429 rdev->sb_size = (rdev->sb_size | bmask) + 1;
1431 for (i=0; i<max_dev;i++)
1432 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1434 list_for_each_entry(rdev2, &mddev->disks, same_set) {
1436 if (test_bit(Faulty, &rdev2->flags))
1437 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1438 else if (test_bit(In_sync, &rdev2->flags))
1439 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1440 else if (rdev2->raid_disk >= 0 && rdev2->recovery_offset > 0)
1441 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1443 sb->dev_roles[i] = cpu_to_le16(0xffff);
1446 sb->sb_csum = calc_sb_1_csum(sb);
1449 static unsigned long long
1450 super_1_rdev_size_change(mdk_rdev_t *rdev, sector_t num_sectors)
1452 struct mdp_superblock_1 *sb;
1453 sector_t max_sectors;
1454 if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
1455 return 0; /* component must fit device */
1456 if (rdev->sb_start < rdev->data_offset) {
1457 /* minor versions 1 and 2; superblock before data */
1458 max_sectors = rdev->bdev->bd_inode->i_size >> 9;
1459 max_sectors -= rdev->data_offset;
1460 if (!num_sectors || num_sectors > max_sectors)
1461 num_sectors = max_sectors;
1462 } else if (rdev->mddev->bitmap_offset) {
1463 /* minor version 0 with bitmap we can't move */
1466 /* minor version 0; superblock after data */
1468 sb_start = (rdev->bdev->bd_inode->i_size >> 9) - 8*2;
1469 sb_start &= ~(sector_t)(4*2 - 1);
1470 max_sectors = rdev->sectors + sb_start - rdev->sb_start;
1471 if (!num_sectors || num_sectors > max_sectors)
1472 num_sectors = max_sectors;
1473 rdev->sb_start = sb_start;
1475 sb = (struct mdp_superblock_1 *) page_address(rdev->sb_page);
1476 sb->data_size = cpu_to_le64(num_sectors);
1477 sb->super_offset = rdev->sb_start;
1478 sb->sb_csum = calc_sb_1_csum(sb);
1479 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
1481 md_super_wait(rdev->mddev);
1482 return num_sectors / 2; /* kB for sysfs */
1485 static struct super_type super_types[] = {
1488 .owner = THIS_MODULE,
1489 .load_super = super_90_load,
1490 .validate_super = super_90_validate,
1491 .sync_super = super_90_sync,
1492 .rdev_size_change = super_90_rdev_size_change,
1496 .owner = THIS_MODULE,
1497 .load_super = super_1_load,
1498 .validate_super = super_1_validate,
1499 .sync_super = super_1_sync,
1500 .rdev_size_change = super_1_rdev_size_change,
1504 static int match_mddev_units(mddev_t *mddev1, mddev_t *mddev2)
1506 mdk_rdev_t *rdev, *rdev2;
1509 rdev_for_each_rcu(rdev, mddev1)
1510 rdev_for_each_rcu(rdev2, mddev2)
1511 if (rdev->bdev->bd_contains ==
1512 rdev2->bdev->bd_contains) {
1520 static LIST_HEAD(pending_raid_disks);
1523 * Try to register data integrity profile for an mddev
1525 * This is called when an array is started and after a disk has been kicked
1526 * from the array. It only succeeds if all working and active component devices
1527 * are integrity capable with matching profiles.
1529 int md_integrity_register(mddev_t *mddev)
1531 mdk_rdev_t *rdev, *reference = NULL;
1533 if (list_empty(&mddev->disks))
1534 return 0; /* nothing to do */
1535 if (blk_get_integrity(mddev->gendisk))
1536 return 0; /* already registered */
1537 list_for_each_entry(rdev, &mddev->disks, same_set) {
1538 /* skip spares and non-functional disks */
1539 if (test_bit(Faulty, &rdev->flags))
1541 if (rdev->raid_disk < 0)
1544 * If at least one rdev is not integrity capable, we can not
1545 * enable data integrity for the md device.
1547 if (!bdev_get_integrity(rdev->bdev))
1550 /* Use the first rdev as the reference */
1554 /* does this rdev's profile match the reference profile? */
1555 if (blk_integrity_compare(reference->bdev->bd_disk,
1556 rdev->bdev->bd_disk) < 0)
1560 * All component devices are integrity capable and have matching
1561 * profiles, register the common profile for the md device.
1563 if (blk_integrity_register(mddev->gendisk,
1564 bdev_get_integrity(reference->bdev)) != 0) {
1565 printk(KERN_ERR "md: failed to register integrity for %s\n",
1569 printk(KERN_NOTICE "md: data integrity on %s enabled\n",
1573 EXPORT_SYMBOL(md_integrity_register);
1575 /* Disable data integrity if non-capable/non-matching disk is being added */
1576 void md_integrity_add_rdev(mdk_rdev_t *rdev, mddev_t *mddev)
1578 struct blk_integrity *bi_rdev = bdev_get_integrity(rdev->bdev);
1579 struct blk_integrity *bi_mddev = blk_get_integrity(mddev->gendisk);
1581 if (!bi_mddev) /* nothing to do */
1583 if (rdev->raid_disk < 0) /* skip spares */
1585 if (bi_rdev && blk_integrity_compare(mddev->gendisk,
1586 rdev->bdev->bd_disk) >= 0)
1588 printk(KERN_NOTICE "disabling data integrity on %s\n", mdname(mddev));
1589 blk_integrity_unregister(mddev->gendisk);
1591 EXPORT_SYMBOL(md_integrity_add_rdev);
1593 static int bind_rdev_to_array(mdk_rdev_t * rdev, mddev_t * mddev)
1595 char b[BDEVNAME_SIZE];
1605 /* prevent duplicates */
1606 if (find_rdev(mddev, rdev->bdev->bd_dev))
1609 /* make sure rdev->sectors exceeds mddev->dev_sectors */
1610 if (rdev->sectors && (mddev->dev_sectors == 0 ||
1611 rdev->sectors < mddev->dev_sectors)) {
1613 /* Cannot change size, so fail
1614 * If mddev->level <= 0, then we don't care
1615 * about aligning sizes (e.g. linear)
1617 if (mddev->level > 0)
1620 mddev->dev_sectors = rdev->sectors;
1623 /* Verify rdev->desc_nr is unique.
1624 * If it is -1, assign a free number, else
1625 * check number is not in use
1627 if (rdev->desc_nr < 0) {
1629 if (mddev->pers) choice = mddev->raid_disks;
1630 while (find_rdev_nr(mddev, choice))
1632 rdev->desc_nr = choice;
1634 if (find_rdev_nr(mddev, rdev->desc_nr))
1637 if (mddev->max_disks && rdev->desc_nr >= mddev->max_disks) {
1638 printk(KERN_WARNING "md: %s: array is limited to %d devices\n",
1639 mdname(mddev), mddev->max_disks);
1642 bdevname(rdev->bdev,b);
1643 while ( (s=strchr(b, '/')) != NULL)
1646 rdev->mddev = mddev;
1647 printk(KERN_INFO "md: bind<%s>\n", b);
1649 if ((err = kobject_add(&rdev->kobj, &mddev->kobj, "dev-%s", b)))
1652 ko = &part_to_dev(rdev->bdev->bd_part)->kobj;
1653 if ((err = sysfs_create_link(&rdev->kobj, ko, "block"))) {
1654 kobject_del(&rdev->kobj);
1657 rdev->sysfs_state = sysfs_get_dirent(rdev->kobj.sd, "state");
1659 list_add_rcu(&rdev->same_set, &mddev->disks);
1660 bd_claim_by_disk(rdev->bdev, rdev->bdev->bd_holder, mddev->gendisk);
1662 /* May as well allow recovery to be retried once */
1663 mddev->recovery_disabled = 0;
1668 printk(KERN_WARNING "md: failed to register dev-%s for %s\n",
1673 static void md_delayed_delete(struct work_struct *ws)
1675 mdk_rdev_t *rdev = container_of(ws, mdk_rdev_t, del_work);
1676 kobject_del(&rdev->kobj);
1677 kobject_put(&rdev->kobj);
1680 static void unbind_rdev_from_array(mdk_rdev_t * rdev)
1682 char b[BDEVNAME_SIZE];
1687 bd_release_from_disk(rdev->bdev, rdev->mddev->gendisk);
1688 list_del_rcu(&rdev->same_set);
1689 printk(KERN_INFO "md: unbind<%s>\n", bdevname(rdev->bdev,b));
1691 sysfs_remove_link(&rdev->kobj, "block");
1692 sysfs_put(rdev->sysfs_state);
1693 rdev->sysfs_state = NULL;
1694 /* We need to delay this, otherwise we can deadlock when
1695 * writing to 'remove' to "dev/state". We also need
1696 * to delay it due to rcu usage.
1699 INIT_WORK(&rdev->del_work, md_delayed_delete);
1700 kobject_get(&rdev->kobj);
1701 schedule_work(&rdev->del_work);
1705 * prevent the device from being mounted, repartitioned or
1706 * otherwise reused by a RAID array (or any other kernel
1707 * subsystem), by bd_claiming the device.
1709 static int lock_rdev(mdk_rdev_t *rdev, dev_t dev, int shared)
1712 struct block_device *bdev;
1713 char b[BDEVNAME_SIZE];
1715 bdev = open_by_devnum(dev, FMODE_READ|FMODE_WRITE);
1717 printk(KERN_ERR "md: could not open %s.\n",
1718 __bdevname(dev, b));
1719 return PTR_ERR(bdev);
1721 err = bd_claim(bdev, shared ? (mdk_rdev_t *)lock_rdev : rdev);
1723 printk(KERN_ERR "md: could not bd_claim %s.\n",
1725 blkdev_put(bdev, FMODE_READ|FMODE_WRITE);
1729 set_bit(AllReserved, &rdev->flags);
1734 static void unlock_rdev(mdk_rdev_t *rdev)
1736 struct block_device *bdev = rdev->bdev;
1741 blkdev_put(bdev, FMODE_READ|FMODE_WRITE);
1744 void md_autodetect_dev(dev_t dev);
1746 static void export_rdev(mdk_rdev_t * rdev)
1748 char b[BDEVNAME_SIZE];
1749 printk(KERN_INFO "md: export_rdev(%s)\n",
1750 bdevname(rdev->bdev,b));
1755 if (test_bit(AutoDetected, &rdev->flags))
1756 md_autodetect_dev(rdev->bdev->bd_dev);
1759 kobject_put(&rdev->kobj);
1762 static void kick_rdev_from_array(mdk_rdev_t * rdev)
1764 unbind_rdev_from_array(rdev);
1768 static void export_array(mddev_t *mddev)
1770 mdk_rdev_t *rdev, *tmp;
1772 rdev_for_each(rdev, tmp, mddev) {
1777 kick_rdev_from_array(rdev);
1779 if (!list_empty(&mddev->disks))
1781 mddev->raid_disks = 0;
1782 mddev->major_version = 0;
1785 static void print_desc(mdp_disk_t *desc)
1787 printk(" DISK<N:%d,(%d,%d),R:%d,S:%d>\n", desc->number,
1788 desc->major,desc->minor,desc->raid_disk,desc->state);
1791 static void print_sb_90(mdp_super_t *sb)
1796 "md: SB: (V:%d.%d.%d) ID:<%08x.%08x.%08x.%08x> CT:%08x\n",
1797 sb->major_version, sb->minor_version, sb->patch_version,
1798 sb->set_uuid0, sb->set_uuid1, sb->set_uuid2, sb->set_uuid3,
1800 printk(KERN_INFO "md: L%d S%08d ND:%d RD:%d md%d LO:%d CS:%d\n",
1801 sb->level, sb->size, sb->nr_disks, sb->raid_disks,
1802 sb->md_minor, sb->layout, sb->chunk_size);
1803 printk(KERN_INFO "md: UT:%08x ST:%d AD:%d WD:%d"
1804 " FD:%d SD:%d CSUM:%08x E:%08lx\n",
1805 sb->utime, sb->state, sb->active_disks, sb->working_disks,
1806 sb->failed_disks, sb->spare_disks,
1807 sb->sb_csum, (unsigned long)sb->events_lo);
1810 for (i = 0; i < MD_SB_DISKS; i++) {
1813 desc = sb->disks + i;
1814 if (desc->number || desc->major || desc->minor ||
1815 desc->raid_disk || (desc->state && (desc->state != 4))) {
1816 printk(" D %2d: ", i);
1820 printk(KERN_INFO "md: THIS: ");
1821 print_desc(&sb->this_disk);
1824 static void print_sb_1(struct mdp_superblock_1 *sb)
1828 uuid = sb->set_uuid;
1830 "md: SB: (V:%u) (F:0x%08x) Array-ID:<%02x%02x%02x%02x"
1831 ":%02x%02x:%02x%02x:%02x%02x:%02x%02x%02x%02x%02x%02x>\n"
1832 "md: Name: \"%s\" CT:%llu\n",
1833 le32_to_cpu(sb->major_version),
1834 le32_to_cpu(sb->feature_map),
1835 uuid[0], uuid[1], uuid[2], uuid[3],
1836 uuid[4], uuid[5], uuid[6], uuid[7],
1837 uuid[8], uuid[9], uuid[10], uuid[11],
1838 uuid[12], uuid[13], uuid[14], uuid[15],
1840 (unsigned long long)le64_to_cpu(sb->ctime)
1841 & MD_SUPERBLOCK_1_TIME_SEC_MASK);
1843 uuid = sb->device_uuid;
1845 "md: L%u SZ%llu RD:%u LO:%u CS:%u DO:%llu DS:%llu SO:%llu"
1847 "md: Dev:%08x UUID: %02x%02x%02x%02x:%02x%02x:%02x%02x:%02x%02x"
1848 ":%02x%02x%02x%02x%02x%02x\n"
1849 "md: (F:0x%08x) UT:%llu Events:%llu ResyncOffset:%llu CSUM:0x%08x\n"
1850 "md: (MaxDev:%u) \n",
1851 le32_to_cpu(sb->level),
1852 (unsigned long long)le64_to_cpu(sb->size),
1853 le32_to_cpu(sb->raid_disks),
1854 le32_to_cpu(sb->layout),
1855 le32_to_cpu(sb->chunksize),
1856 (unsigned long long)le64_to_cpu(sb->data_offset),
1857 (unsigned long long)le64_to_cpu(sb->data_size),
1858 (unsigned long long)le64_to_cpu(sb->super_offset),
1859 (unsigned long long)le64_to_cpu(sb->recovery_offset),
1860 le32_to_cpu(sb->dev_number),
1861 uuid[0], uuid[1], uuid[2], uuid[3],
1862 uuid[4], uuid[5], uuid[6], uuid[7],
1863 uuid[8], uuid[9], uuid[10], uuid[11],
1864 uuid[12], uuid[13], uuid[14], uuid[15],
1866 (unsigned long long)le64_to_cpu(sb->utime) & MD_SUPERBLOCK_1_TIME_SEC_MASK,
1867 (unsigned long long)le64_to_cpu(sb->events),
1868 (unsigned long long)le64_to_cpu(sb->resync_offset),
1869 le32_to_cpu(sb->sb_csum),
1870 le32_to_cpu(sb->max_dev)
1874 static void print_rdev(mdk_rdev_t *rdev, int major_version)
1876 char b[BDEVNAME_SIZE];
1877 printk(KERN_INFO "md: rdev %s, Sect:%08llu F:%d S:%d DN:%u\n",
1878 bdevname(rdev->bdev, b), (unsigned long long)rdev->sectors,
1879 test_bit(Faulty, &rdev->flags), test_bit(In_sync, &rdev->flags),
1881 if (rdev->sb_loaded) {
1882 printk(KERN_INFO "md: rdev superblock (MJ:%d):\n", major_version);
1883 switch (major_version) {
1885 print_sb_90((mdp_super_t*)page_address(rdev->sb_page));
1888 print_sb_1((struct mdp_superblock_1 *)page_address(rdev->sb_page));
1892 printk(KERN_INFO "md: no rdev superblock!\n");
1895 static void md_print_devices(void)
1897 struct list_head *tmp;
1900 char b[BDEVNAME_SIZE];
1903 printk("md: **********************************\n");
1904 printk("md: * <COMPLETE RAID STATE PRINTOUT> *\n");
1905 printk("md: **********************************\n");
1906 for_each_mddev(mddev, tmp) {
1909 bitmap_print_sb(mddev->bitmap);
1911 printk("%s: ", mdname(mddev));
1912 list_for_each_entry(rdev, &mddev->disks, same_set)
1913 printk("<%s>", bdevname(rdev->bdev,b));
1916 list_for_each_entry(rdev, &mddev->disks, same_set)
1917 print_rdev(rdev, mddev->major_version);
1919 printk("md: **********************************\n");
1924 static void sync_sbs(mddev_t * mddev, int nospares)
1926 /* Update each superblock (in-memory image), but
1927 * if we are allowed to, skip spares which already
1928 * have the right event counter, or have one earlier
1929 * (which would mean they aren't being marked as dirty
1930 * with the rest of the array)
1934 /* First make sure individual recovery_offsets are correct */
1935 list_for_each_entry(rdev, &mddev->disks, same_set) {
1936 if (rdev->raid_disk >= 0 &&
1937 !test_bit(In_sync, &rdev->flags) &&
1938 mddev->curr_resync_completed > rdev->recovery_offset)
1939 rdev->recovery_offset = mddev->curr_resync_completed;
1942 list_for_each_entry(rdev, &mddev->disks, same_set) {
1943 if (rdev->sb_events == mddev->events ||
1945 rdev->raid_disk < 0 &&
1946 (rdev->sb_events&1)==0 &&
1947 rdev->sb_events+1 == mddev->events)) {
1948 /* Don't update this superblock */
1949 rdev->sb_loaded = 2;
1951 super_types[mddev->major_version].
1952 sync_super(mddev, rdev);
1953 rdev->sb_loaded = 1;
1958 static void md_update_sb(mddev_t * mddev, int force_change)
1964 mddev->utime = get_seconds();
1965 if (mddev->external)
1968 spin_lock_irq(&mddev->write_lock);
1970 set_bit(MD_CHANGE_PENDING, &mddev->flags);
1971 if (test_and_clear_bit(MD_CHANGE_DEVS, &mddev->flags))
1973 if (test_and_clear_bit(MD_CHANGE_CLEAN, &mddev->flags))
1974 /* just a clean<-> dirty transition, possibly leave spares alone,
1975 * though if events isn't the right even/odd, we will have to do
1981 if (mddev->degraded)
1982 /* If the array is degraded, then skipping spares is both
1983 * dangerous and fairly pointless.
1984 * Dangerous because a device that was removed from the array
1985 * might have a event_count that still looks up-to-date,
1986 * so it can be re-added without a resync.
1987 * Pointless because if there are any spares to skip,
1988 * then a recovery will happen and soon that array won't
1989 * be degraded any more and the spare can go back to sleep then.
1993 sync_req = mddev->in_sync;
1995 /* If this is just a dirty<->clean transition, and the array is clean
1996 * and 'events' is odd, we can roll back to the previous clean state */
1998 && (mddev->in_sync && mddev->recovery_cp == MaxSector)
1999 && (mddev->events & 1)
2000 && mddev->events != 1)
2003 /* otherwise we have to go forward and ... */
2005 if (!mddev->in_sync || mddev->recovery_cp != MaxSector) { /* not clean */
2006 /* .. if the array isn't clean, an 'even' event must also go
2008 if ((mddev->events&1)==0)
2011 /* otherwise an 'odd' event must go to spares */
2012 if ((mddev->events&1))
2017 if (!mddev->events) {
2019 * oops, this 64-bit counter should never wrap.
2020 * Either we are in around ~1 trillion A.C., assuming
2021 * 1 reboot per second, or we have a bug:
2028 * do not write anything to disk if using
2029 * nonpersistent superblocks
2031 if (!mddev->persistent) {
2032 if (!mddev->external)
2033 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
2035 spin_unlock_irq(&mddev->write_lock);
2036 wake_up(&mddev->sb_wait);
2039 sync_sbs(mddev, nospares);
2040 spin_unlock_irq(&mddev->write_lock);
2043 "md: updating %s RAID superblock on device (in sync %d)\n",
2044 mdname(mddev),mddev->in_sync);
2046 bitmap_update_sb(mddev->bitmap);
2047 list_for_each_entry(rdev, &mddev->disks, same_set) {
2048 char b[BDEVNAME_SIZE];
2049 dprintk(KERN_INFO "md: ");
2050 if (rdev->sb_loaded != 1)
2051 continue; /* no noise on spare devices */
2052 if (test_bit(Faulty, &rdev->flags))
2053 dprintk("(skipping faulty ");
2055 dprintk("%s ", bdevname(rdev->bdev,b));
2056 if (!test_bit(Faulty, &rdev->flags)) {
2057 md_super_write(mddev,rdev,
2058 rdev->sb_start, rdev->sb_size,
2060 dprintk(KERN_INFO "(write) %s's sb offset: %llu\n",
2061 bdevname(rdev->bdev,b),
2062 (unsigned long long)rdev->sb_start);
2063 rdev->sb_events = mddev->events;
2067 if (mddev->level == LEVEL_MULTIPATH)
2068 /* only need to write one superblock... */
2071 md_super_wait(mddev);
2072 /* if there was a failure, MD_CHANGE_DEVS was set, and we re-write super */
2074 spin_lock_irq(&mddev->write_lock);
2075 if (mddev->in_sync != sync_req ||
2076 test_bit(MD_CHANGE_DEVS, &mddev->flags)) {
2077 /* have to write it out again */
2078 spin_unlock_irq(&mddev->write_lock);
2081 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
2082 spin_unlock_irq(&mddev->write_lock);
2083 wake_up(&mddev->sb_wait);
2084 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
2085 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
2089 /* words written to sysfs files may, or may not, be \n terminated.
2090 * We want to accept with case. For this we use cmd_match.
2092 static int cmd_match(const char *cmd, const char *str)
2094 /* See if cmd, written into a sysfs file, matches
2095 * str. They must either be the same, or cmd can
2096 * have a trailing newline
2098 while (*cmd && *str && *cmd == *str) {
2109 struct rdev_sysfs_entry {
2110 struct attribute attr;
2111 ssize_t (*show)(mdk_rdev_t *, char *);
2112 ssize_t (*store)(mdk_rdev_t *, const char *, size_t);
2116 state_show(mdk_rdev_t *rdev, char *page)
2121 if (test_bit(Faulty, &rdev->flags)) {
2122 len+= sprintf(page+len, "%sfaulty",sep);
2125 if (test_bit(In_sync, &rdev->flags)) {
2126 len += sprintf(page+len, "%sin_sync",sep);
2129 if (test_bit(WriteMostly, &rdev->flags)) {
2130 len += sprintf(page+len, "%swrite_mostly",sep);
2133 if (test_bit(Blocked, &rdev->flags)) {
2134 len += sprintf(page+len, "%sblocked", sep);
2137 if (!test_bit(Faulty, &rdev->flags) &&
2138 !test_bit(In_sync, &rdev->flags)) {
2139 len += sprintf(page+len, "%sspare", sep);
2142 return len+sprintf(page+len, "\n");
2146 state_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2149 * faulty - simulates and error
2150 * remove - disconnects the device
2151 * writemostly - sets write_mostly
2152 * -writemostly - clears write_mostly
2153 * blocked - sets the Blocked flag
2154 * -blocked - clears the Blocked flag
2155 * insync - sets Insync providing device isn't active
2158 if (cmd_match(buf, "faulty") && rdev->mddev->pers) {
2159 md_error(rdev->mddev, rdev);
2161 } else if (cmd_match(buf, "remove")) {
2162 if (rdev->raid_disk >= 0)
2165 mddev_t *mddev = rdev->mddev;
2166 kick_rdev_from_array(rdev);
2168 md_update_sb(mddev, 1);
2169 md_new_event(mddev);
2172 } else if (cmd_match(buf, "writemostly")) {
2173 set_bit(WriteMostly, &rdev->flags);
2175 } else if (cmd_match(buf, "-writemostly")) {
2176 clear_bit(WriteMostly, &rdev->flags);
2178 } else if (cmd_match(buf, "blocked")) {
2179 set_bit(Blocked, &rdev->flags);
2181 } else if (cmd_match(buf, "-blocked")) {
2182 clear_bit(Blocked, &rdev->flags);
2183 wake_up(&rdev->blocked_wait);
2184 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2185 md_wakeup_thread(rdev->mddev->thread);
2188 } else if (cmd_match(buf, "insync") && rdev->raid_disk == -1) {
2189 set_bit(In_sync, &rdev->flags);
2192 if (!err && rdev->sysfs_state)
2193 sysfs_notify_dirent(rdev->sysfs_state);
2194 return err ? err : len;
2196 static struct rdev_sysfs_entry rdev_state =
2197 __ATTR(state, S_IRUGO|S_IWUSR, state_show, state_store);
2200 errors_show(mdk_rdev_t *rdev, char *page)
2202 return sprintf(page, "%d\n", atomic_read(&rdev->corrected_errors));
2206 errors_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2209 unsigned long n = simple_strtoul(buf, &e, 10);
2210 if (*buf && (*e == 0 || *e == '\n')) {
2211 atomic_set(&rdev->corrected_errors, n);
2216 static struct rdev_sysfs_entry rdev_errors =
2217 __ATTR(errors, S_IRUGO|S_IWUSR, errors_show, errors_store);
2220 slot_show(mdk_rdev_t *rdev, char *page)
2222 if (rdev->raid_disk < 0)
2223 return sprintf(page, "none\n");
2225 return sprintf(page, "%d\n", rdev->raid_disk);
2229 slot_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2234 int slot = simple_strtoul(buf, &e, 10);
2235 if (strncmp(buf, "none", 4)==0)
2237 else if (e==buf || (*e && *e!= '\n'))
2239 if (rdev->mddev->pers && slot == -1) {
2240 /* Setting 'slot' on an active array requires also
2241 * updating the 'rd%d' link, and communicating
2242 * with the personality with ->hot_*_disk.
2243 * For now we only support removing
2244 * failed/spare devices. This normally happens automatically,
2245 * but not when the metadata is externally managed.
2247 if (rdev->raid_disk == -1)
2249 /* personality does all needed checks */
2250 if (rdev->mddev->pers->hot_add_disk == NULL)
2252 err = rdev->mddev->pers->
2253 hot_remove_disk(rdev->mddev, rdev->raid_disk);
2256 sprintf(nm, "rd%d", rdev->raid_disk);
2257 sysfs_remove_link(&rdev->mddev->kobj, nm);
2258 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2259 md_wakeup_thread(rdev->mddev->thread);
2260 } else if (rdev->mddev->pers) {
2262 /* Activating a spare .. or possibly reactivating
2263 * if we ever get bitmaps working here.
2266 if (rdev->raid_disk != -1)
2269 if (rdev->mddev->pers->hot_add_disk == NULL)
2272 list_for_each_entry(rdev2, &rdev->mddev->disks, same_set)
2273 if (rdev2->raid_disk == slot)
2276 rdev->raid_disk = slot;
2277 if (test_bit(In_sync, &rdev->flags))
2278 rdev->saved_raid_disk = slot;
2280 rdev->saved_raid_disk = -1;
2281 err = rdev->mddev->pers->
2282 hot_add_disk(rdev->mddev, rdev);
2284 rdev->raid_disk = -1;
2287 sysfs_notify_dirent(rdev->sysfs_state);
2288 sprintf(nm, "rd%d", rdev->raid_disk);
2289 if (sysfs_create_link(&rdev->mddev->kobj, &rdev->kobj, nm))
2291 "md: cannot register "
2293 nm, mdname(rdev->mddev));
2295 /* don't wakeup anyone, leave that to userspace. */
2297 if (slot >= rdev->mddev->raid_disks)
2299 rdev->raid_disk = slot;
2300 /* assume it is working */
2301 clear_bit(Faulty, &rdev->flags);
2302 clear_bit(WriteMostly, &rdev->flags);
2303 set_bit(In_sync, &rdev->flags);
2304 sysfs_notify_dirent(rdev->sysfs_state);
2310 static struct rdev_sysfs_entry rdev_slot =
2311 __ATTR(slot, S_IRUGO|S_IWUSR, slot_show, slot_store);
2314 offset_show(mdk_rdev_t *rdev, char *page)
2316 return sprintf(page, "%llu\n", (unsigned long long)rdev->data_offset);
2320 offset_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2323 unsigned long long offset = simple_strtoull(buf, &e, 10);
2324 if (e==buf || (*e && *e != '\n'))
2326 if (rdev->mddev->pers && rdev->raid_disk >= 0)
2328 if (rdev->sectors && rdev->mddev->external)
2329 /* Must set offset before size, so overlap checks
2332 rdev->data_offset = offset;
2336 static struct rdev_sysfs_entry rdev_offset =
2337 __ATTR(offset, S_IRUGO|S_IWUSR, offset_show, offset_store);
2340 rdev_size_show(mdk_rdev_t *rdev, char *page)
2342 return sprintf(page, "%llu\n", (unsigned long long)rdev->sectors / 2);
2345 static int overlaps(sector_t s1, sector_t l1, sector_t s2, sector_t l2)
2347 /* check if two start/length pairs overlap */
2355 static int strict_blocks_to_sectors(const char *buf, sector_t *sectors)
2357 unsigned long long blocks;
2360 if (strict_strtoull(buf, 10, &blocks) < 0)
2363 if (blocks & 1ULL << (8 * sizeof(blocks) - 1))
2364 return -EINVAL; /* sector conversion overflow */
2367 if (new != blocks * 2)
2368 return -EINVAL; /* unsigned long long to sector_t overflow */
2375 rdev_size_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2377 mddev_t *my_mddev = rdev->mddev;
2378 sector_t oldsectors = rdev->sectors;
2381 if (strict_blocks_to_sectors(buf, §ors) < 0)
2383 if (my_mddev->pers && rdev->raid_disk >= 0) {
2384 if (my_mddev->persistent) {
2385 sectors = super_types[my_mddev->major_version].
2386 rdev_size_change(rdev, sectors);
2389 } else if (!sectors)
2390 sectors = (rdev->bdev->bd_inode->i_size >> 9) -
2393 if (sectors < my_mddev->dev_sectors)
2394 return -EINVAL; /* component must fit device */
2396 rdev->sectors = sectors;
2397 if (sectors > oldsectors && my_mddev->external) {
2398 /* need to check that all other rdevs with the same ->bdev
2399 * do not overlap. We need to unlock the mddev to avoid
2400 * a deadlock. We have already changed rdev->sectors, and if
2401 * we have to change it back, we will have the lock again.
2405 struct list_head *tmp;
2407 mddev_unlock(my_mddev);
2408 for_each_mddev(mddev, tmp) {
2412 list_for_each_entry(rdev2, &mddev->disks, same_set)
2413 if (test_bit(AllReserved, &rdev2->flags) ||
2414 (rdev->bdev == rdev2->bdev &&
2416 overlaps(rdev->data_offset, rdev->sectors,
2422 mddev_unlock(mddev);
2428 mddev_lock(my_mddev);
2430 /* Someone else could have slipped in a size
2431 * change here, but doing so is just silly.
2432 * We put oldsectors back because we *know* it is
2433 * safe, and trust userspace not to race with
2436 rdev->sectors = oldsectors;
2443 static struct rdev_sysfs_entry rdev_size =
2444 __ATTR(size, S_IRUGO|S_IWUSR, rdev_size_show, rdev_size_store);
2446 static struct attribute *rdev_default_attrs[] = {
2455 rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
2457 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
2458 mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
2459 mddev_t *mddev = rdev->mddev;
2465 rv = mddev ? mddev_lock(mddev) : -EBUSY;
2467 if (rdev->mddev == NULL)
2470 rv = entry->show(rdev, page);
2471 mddev_unlock(mddev);
2477 rdev_attr_store(struct kobject *kobj, struct attribute *attr,
2478 const char *page, size_t length)
2480 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
2481 mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
2483 mddev_t *mddev = rdev->mddev;
2487 if (!capable(CAP_SYS_ADMIN))
2489 rv = mddev ? mddev_lock(mddev): -EBUSY;
2491 if (rdev->mddev == NULL)
2494 rv = entry->store(rdev, page, length);
2495 mddev_unlock(mddev);
2500 static void rdev_free(struct kobject *ko)
2502 mdk_rdev_t *rdev = container_of(ko, mdk_rdev_t, kobj);
2505 static struct sysfs_ops rdev_sysfs_ops = {
2506 .show = rdev_attr_show,
2507 .store = rdev_attr_store,
2509 static struct kobj_type rdev_ktype = {
2510 .release = rdev_free,
2511 .sysfs_ops = &rdev_sysfs_ops,
2512 .default_attrs = rdev_default_attrs,
2516 * Import a device. If 'super_format' >= 0, then sanity check the superblock
2518 * mark the device faulty if:
2520 * - the device is nonexistent (zero size)
2521 * - the device has no valid superblock
2523 * a faulty rdev _never_ has rdev->sb set.
2525 static mdk_rdev_t *md_import_device(dev_t newdev, int super_format, int super_minor)
2527 char b[BDEVNAME_SIZE];
2532 rdev = kzalloc(sizeof(*rdev), GFP_KERNEL);
2534 printk(KERN_ERR "md: could not alloc mem for new device!\n");
2535 return ERR_PTR(-ENOMEM);
2538 if ((err = alloc_disk_sb(rdev)))
2541 err = lock_rdev(rdev, newdev, super_format == -2);
2545 kobject_init(&rdev->kobj, &rdev_ktype);
2548 rdev->saved_raid_disk = -1;
2549 rdev->raid_disk = -1;
2551 rdev->data_offset = 0;
2552 rdev->sb_events = 0;
2553 atomic_set(&rdev->nr_pending, 0);
2554 atomic_set(&rdev->read_errors, 0);
2555 atomic_set(&rdev->corrected_errors, 0);
2557 size = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
2560 "md: %s has zero or unknown size, marking faulty!\n",
2561 bdevname(rdev->bdev,b));
2566 if (super_format >= 0) {
2567 err = super_types[super_format].
2568 load_super(rdev, NULL, super_minor);
2569 if (err == -EINVAL) {
2571 "md: %s does not have a valid v%d.%d "
2572 "superblock, not importing!\n",
2573 bdevname(rdev->bdev,b),
2574 super_format, super_minor);
2579 "md: could not read %s's sb, not importing!\n",
2580 bdevname(rdev->bdev,b));
2585 INIT_LIST_HEAD(&rdev->same_set);
2586 init_waitqueue_head(&rdev->blocked_wait);
2591 if (rdev->sb_page) {
2597 return ERR_PTR(err);
2601 * Check a full RAID array for plausibility
2605 static void analyze_sbs(mddev_t * mddev)
2608 mdk_rdev_t *rdev, *freshest, *tmp;
2609 char b[BDEVNAME_SIZE];
2612 rdev_for_each(rdev, tmp, mddev)
2613 switch (super_types[mddev->major_version].
2614 load_super(rdev, freshest, mddev->minor_version)) {
2622 "md: fatal superblock inconsistency in %s"
2623 " -- removing from array\n",
2624 bdevname(rdev->bdev,b));
2625 kick_rdev_from_array(rdev);
2629 super_types[mddev->major_version].
2630 validate_super(mddev, freshest);
2633 rdev_for_each(rdev, tmp, mddev) {
2634 if (rdev->desc_nr >= mddev->max_disks ||
2635 i > mddev->max_disks) {
2637 "md: %s: %s: only %d devices permitted\n",
2638 mdname(mddev), bdevname(rdev->bdev, b),
2640 kick_rdev_from_array(rdev);
2643 if (rdev != freshest)
2644 if (super_types[mddev->major_version].
2645 validate_super(mddev, rdev)) {
2646 printk(KERN_WARNING "md: kicking non-fresh %s"
2648 bdevname(rdev->bdev,b));
2649 kick_rdev_from_array(rdev);
2652 if (mddev->level == LEVEL_MULTIPATH) {
2653 rdev->desc_nr = i++;
2654 rdev->raid_disk = rdev->desc_nr;
2655 set_bit(In_sync, &rdev->flags);
2656 } else if (rdev->raid_disk >= (mddev->raid_disks - min(0, mddev->delta_disks))) {
2657 rdev->raid_disk = -1;
2658 clear_bit(In_sync, &rdev->flags);
2663 static void md_safemode_timeout(unsigned long data);
2666 safe_delay_show(mddev_t *mddev, char *page)
2668 int msec = (mddev->safemode_delay*1000)/HZ;
2669 return sprintf(page, "%d.%03d\n", msec/1000, msec%1000);
2672 safe_delay_store(mddev_t *mddev, const char *cbuf, size_t len)
2680 /* remove a period, and count digits after it */
2681 if (len >= sizeof(buf))
2683 strlcpy(buf, cbuf, sizeof(buf));
2684 for (i=0; i<len; i++) {
2686 if (isdigit(buf[i])) {
2691 } else if (buf[i] == '.') {
2696 if (strict_strtoul(buf, 10, &msec) < 0)
2698 msec = (msec * 1000) / scale;
2700 mddev->safemode_delay = 0;
2702 unsigned long old_delay = mddev->safemode_delay;
2703 mddev->safemode_delay = (msec*HZ)/1000;
2704 if (mddev->safemode_delay == 0)
2705 mddev->safemode_delay = 1;
2706 if (mddev->safemode_delay < old_delay)
2707 md_safemode_timeout((unsigned long)mddev);
2711 static struct md_sysfs_entry md_safe_delay =
2712 __ATTR(safe_mode_delay, S_IRUGO|S_IWUSR,safe_delay_show, safe_delay_store);
2715 level_show(mddev_t *mddev, char *page)
2717 struct mdk_personality *p = mddev->pers;
2719 return sprintf(page, "%s\n", p->name);
2720 else if (mddev->clevel[0])
2721 return sprintf(page, "%s\n", mddev->clevel);
2722 else if (mddev->level != LEVEL_NONE)
2723 return sprintf(page, "%d\n", mddev->level);
2729 level_store(mddev_t *mddev, const char *buf, size_t len)
2733 struct mdk_personality *pers;
2737 if (mddev->pers == NULL) {
2740 if (len >= sizeof(mddev->clevel))
2742 strncpy(mddev->clevel, buf, len);
2743 if (mddev->clevel[len-1] == '\n')
2745 mddev->clevel[len] = 0;
2746 mddev->level = LEVEL_NONE;
2750 /* request to change the personality. Need to ensure:
2751 * - array is not engaged in resync/recovery/reshape
2752 * - old personality can be suspended
2753 * - new personality will access other array.
2756 if (mddev->sync_thread || mddev->reshape_position != MaxSector)
2759 if (!mddev->pers->quiesce) {
2760 printk(KERN_WARNING "md: %s: %s does not support online personality change\n",
2761 mdname(mddev), mddev->pers->name);
2765 /* Now find the new personality */
2766 if (len == 0 || len >= sizeof(level))
2768 strncpy(level, buf, len);
2769 if (level[len-1] == '\n')
2773 request_module("md-%s", level);
2774 spin_lock(&pers_lock);
2775 pers = find_pers(LEVEL_NONE, level);
2776 if (!pers || !try_module_get(pers->owner)) {
2777 spin_unlock(&pers_lock);
2778 printk(KERN_WARNING "md: personality %s not loaded\n", level);
2781 spin_unlock(&pers_lock);
2783 if (pers == mddev->pers) {
2784 /* Nothing to do! */
2785 module_put(pers->owner);
2788 if (!pers->takeover) {
2789 module_put(pers->owner);
2790 printk(KERN_WARNING "md: %s: %s does not support personality takeover\n",
2791 mdname(mddev), level);
2795 /* ->takeover must set new_* and/or delta_disks
2796 * if it succeeds, and may set them when it fails.
2798 priv = pers->takeover(mddev);
2800 mddev->new_level = mddev->level;
2801 mddev->new_layout = mddev->layout;
2802 mddev->new_chunk_sectors = mddev->chunk_sectors;
2803 mddev->raid_disks -= mddev->delta_disks;
2804 mddev->delta_disks = 0;
2805 module_put(pers->owner);
2806 printk(KERN_WARNING "md: %s: %s would not accept array\n",
2807 mdname(mddev), level);
2808 return PTR_ERR(priv);
2811 /* Looks like we have a winner */
2812 mddev_suspend(mddev);
2813 mddev->pers->stop(mddev);
2814 module_put(mddev->pers->owner);
2815 /* Invalidate devices that are now superfluous */
2816 list_for_each_entry(rdev, &mddev->disks, same_set)
2817 if (rdev->raid_disk >= mddev->raid_disks) {
2818 rdev->raid_disk = -1;
2819 clear_bit(In_sync, &rdev->flags);
2822 mddev->private = priv;
2823 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
2824 mddev->level = mddev->new_level;
2825 mddev->layout = mddev->new_layout;
2826 mddev->chunk_sectors = mddev->new_chunk_sectors;
2827 mddev->delta_disks = 0;
2829 mddev_resume(mddev);
2830 set_bit(MD_CHANGE_DEVS, &mddev->flags);
2831 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2832 md_wakeup_thread(mddev->thread);
2836 static struct md_sysfs_entry md_level =
2837 __ATTR(level, S_IRUGO|S_IWUSR, level_show, level_store);
2841 layout_show(mddev_t *mddev, char *page)
2843 /* just a number, not meaningful for all levels */
2844 if (mddev->reshape_position != MaxSector &&
2845 mddev->layout != mddev->new_layout)
2846 return sprintf(page, "%d (%d)\n",
2847 mddev->new_layout, mddev->layout);
2848 return sprintf(page, "%d\n", mddev->layout);
2852 layout_store(mddev_t *mddev, const char *buf, size_t len)
2855 unsigned long n = simple_strtoul(buf, &e, 10);
2857 if (!*buf || (*e && *e != '\n'))
2862 if (mddev->pers->check_reshape == NULL)
2864 mddev->new_layout = n;
2865 err = mddev->pers->check_reshape(mddev);
2867 mddev->new_layout = mddev->layout;
2871 mddev->new_layout = n;
2872 if (mddev->reshape_position == MaxSector)
2877 static struct md_sysfs_entry md_layout =
2878 __ATTR(layout, S_IRUGO|S_IWUSR, layout_show, layout_store);
2882 raid_disks_show(mddev_t *mddev, char *page)
2884 if (mddev->raid_disks == 0)
2886 if (mddev->reshape_position != MaxSector &&
2887 mddev->delta_disks != 0)
2888 return sprintf(page, "%d (%d)\n", mddev->raid_disks,
2889 mddev->raid_disks - mddev->delta_disks);
2890 return sprintf(page, "%d\n", mddev->raid_disks);
2893 static int update_raid_disks(mddev_t *mddev, int raid_disks);
2896 raid_disks_store(mddev_t *mddev, const char *buf, size_t len)
2900 unsigned long n = simple_strtoul(buf, &e, 10);
2902 if (!*buf || (*e && *e != '\n'))
2906 rv = update_raid_disks(mddev, n);
2907 else if (mddev->reshape_position != MaxSector) {
2908 int olddisks = mddev->raid_disks - mddev->delta_disks;
2909 mddev->delta_disks = n - olddisks;
2910 mddev->raid_disks = n;
2912 mddev->raid_disks = n;
2913 return rv ? rv : len;
2915 static struct md_sysfs_entry md_raid_disks =
2916 __ATTR(raid_disks, S_IRUGO|S_IWUSR, raid_disks_show, raid_disks_store);
2919 chunk_size_show(mddev_t *mddev, char *page)
2921 if (mddev->reshape_position != MaxSector &&
2922 mddev->chunk_sectors != mddev->new_chunk_sectors)
2923 return sprintf(page, "%d (%d)\n",
2924 mddev->new_chunk_sectors << 9,
2925 mddev->chunk_sectors << 9);
2926 return sprintf(page, "%d\n", mddev->chunk_sectors << 9);
2930 chunk_size_store(mddev_t *mddev, const char *buf, size_t len)
2933 unsigned long n = simple_strtoul(buf, &e, 10);
2935 if (!*buf || (*e && *e != '\n'))
2940 if (mddev->pers->check_reshape == NULL)
2942 mddev->new_chunk_sectors = n >> 9;
2943 err = mddev->pers->check_reshape(mddev);
2945 mddev->new_chunk_sectors = mddev->chunk_sectors;
2949 mddev->new_chunk_sectors = n >> 9;
2950 if (mddev->reshape_position == MaxSector)
2951 mddev->chunk_sectors = n >> 9;
2955 static struct md_sysfs_entry md_chunk_size =
2956 __ATTR(chunk_size, S_IRUGO|S_IWUSR, chunk_size_show, chunk_size_store);
2959 resync_start_show(mddev_t *mddev, char *page)
2961 if (mddev->recovery_cp == MaxSector)
2962 return sprintf(page, "none\n");
2963 return sprintf(page, "%llu\n", (unsigned long long)mddev->recovery_cp);
2967 resync_start_store(mddev_t *mddev, const char *buf, size_t len)
2970 unsigned long long n = simple_strtoull(buf, &e, 10);
2974 if (!*buf || (*e && *e != '\n'))
2977 mddev->recovery_cp = n;
2980 static struct md_sysfs_entry md_resync_start =
2981 __ATTR(resync_start, S_IRUGO|S_IWUSR, resync_start_show, resync_start_store);
2984 * The array state can be:
2987 * No devices, no size, no level
2988 * Equivalent to STOP_ARRAY ioctl
2990 * May have some settings, but array is not active
2991 * all IO results in error
2992 * When written, doesn't tear down array, but just stops it
2993 * suspended (not supported yet)
2994 * All IO requests will block. The array can be reconfigured.
2995 * Writing this, if accepted, will block until array is quiescent
2997 * no resync can happen. no superblocks get written.
2998 * write requests fail
3000 * like readonly, but behaves like 'clean' on a write request.
3002 * clean - no pending writes, but otherwise active.
3003 * When written to inactive array, starts without resync
3004 * If a write request arrives then
3005 * if metadata is known, mark 'dirty' and switch to 'active'.
3006 * if not known, block and switch to write-pending
3007 * If written to an active array that has pending writes, then fails.
3009 * fully active: IO and resync can be happening.
3010 * When written to inactive array, starts with resync
3013 * clean, but writes are blocked waiting for 'active' to be written.
3016 * like active, but no writes have been seen for a while (100msec).
3019 enum array_state { clear, inactive, suspended, readonly, read_auto, clean, active,
3020 write_pending, active_idle, bad_word};
3021 static char *array_states[] = {
3022 "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active",
3023 "write-pending", "active-idle", NULL };
3025 static int match_word(const char *word, char **list)
3028 for (n=0; list[n]; n++)
3029 if (cmd_match(word, list[n]))
3035 array_state_show(mddev_t *mddev, char *page)
3037 enum array_state st = inactive;
3050 else if (test_bit(MD_CHANGE_CLEAN, &mddev->flags))
3052 else if (mddev->safemode)
3058 if (list_empty(&mddev->disks) &&
3059 mddev->raid_disks == 0 &&
3060 mddev->dev_sectors == 0)
3065 return sprintf(page, "%s\n", array_states[st]);
3068 static int do_md_stop(mddev_t * mddev, int ro, int is_open);
3069 static int do_md_run(mddev_t * mddev);
3070 static int restart_array(mddev_t *mddev);
3073 array_state_store(mddev_t *mddev, const char *buf, size_t len)
3076 enum array_state st = match_word(buf, array_states);
3081 /* stopping an active array */
3082 if (atomic_read(&mddev->openers) > 0)
3084 err = do_md_stop(mddev, 0, 0);
3087 /* stopping an active array */
3089 if (atomic_read(&mddev->openers) > 0)
3091 err = do_md_stop(mddev, 2, 0);
3093 err = 0; /* already inactive */
3096 break; /* not supported yet */
3099 err = do_md_stop(mddev, 1, 0);
3102 set_disk_ro(mddev->gendisk, 1);
3103 err = do_md_run(mddev);
3109 err = do_md_stop(mddev, 1, 0);
3110 else if (mddev->ro == 1)
3111 err = restart_array(mddev);
3114 set_disk_ro(mddev->gendisk, 0);
3118 err = do_md_run(mddev);
3123 restart_array(mddev);
3124 spin_lock_irq(&mddev->write_lock);
3125 if (atomic_read(&mddev->writes_pending) == 0) {
3126 if (mddev->in_sync == 0) {
3128 if (mddev->safemode == 1)
3129 mddev->safemode = 0;
3130 if (mddev->persistent)
3131 set_bit(MD_CHANGE_CLEAN,
3137 spin_unlock_irq(&mddev->write_lock);
3143 restart_array(mddev);
3144 if (mddev->external)
3145 clear_bit(MD_CHANGE_CLEAN, &mddev->flags);
3146 wake_up(&mddev->sb_wait);
3150 set_disk_ro(mddev->gendisk, 0);
3151 err = do_md_run(mddev);
3156 /* these cannot be set */
3162 sysfs_notify_dirent(mddev->sysfs_state);
3166 static struct md_sysfs_entry md_array_state =
3167 __ATTR(array_state, S_IRUGO|S_IWUSR, array_state_show, array_state_store);
3170 null_show(mddev_t *mddev, char *page)
3176 new_dev_store(mddev_t *mddev, const char *buf, size_t len)
3178 /* buf must be %d:%d\n? giving major and minor numbers */
3179 /* The new device is added to the array.
3180 * If the array has a persistent superblock, we read the
3181 * superblock to initialise info and check validity.
3182 * Otherwise, only checking done is that in bind_rdev_to_array,
3183 * which mainly checks size.
3186 int major = simple_strtoul(buf, &e, 10);
3192 if (!*buf || *e != ':' || !e[1] || e[1] == '\n')
3194 minor = simple_strtoul(e+1, &e, 10);
3195 if (*e && *e != '\n')
3197 dev = MKDEV(major, minor);
3198 if (major != MAJOR(dev) ||
3199 minor != MINOR(dev))
3203 if (mddev->persistent) {
3204 rdev = md_import_device(dev, mddev->major_version,
3205 mddev->minor_version);
3206 if (!IS_ERR(rdev) && !list_empty(&mddev->disks)) {
3207 mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
3208 mdk_rdev_t, same_set);
3209 err = super_types[mddev->major_version]
3210 .load_super(rdev, rdev0, mddev->minor_version);
3214 } else if (mddev->external)
3215 rdev = md_import_device(dev, -2, -1);
3217 rdev = md_import_device(dev, -1, -1);
3220 return PTR_ERR(rdev);
3221 err = bind_rdev_to_array(rdev, mddev);
3225 return err ? err : len;
3228 static struct md_sysfs_entry md_new_device =
3229 __ATTR(new_dev, S_IWUSR, null_show, new_dev_store);
3232 bitmap_store(mddev_t *mddev, const char *buf, size_t len)
3235 unsigned long chunk, end_chunk;
3239 /* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */
3241 chunk = end_chunk = simple_strtoul(buf, &end, 0);
3242 if (buf == end) break;
3243 if (*end == '-') { /* range */
3245 end_chunk = simple_strtoul(buf, &end, 0);
3246 if (buf == end) break;
3248 if (*end && !isspace(*end)) break;
3249 bitmap_dirty_bits(mddev->bitmap, chunk, end_chunk);
3251 while (isspace(*buf)) buf++;
3253 bitmap_unplug(mddev->bitmap); /* flush the bits to disk */
3258 static struct md_sysfs_entry md_bitmap =
3259 __ATTR(bitmap_set_bits, S_IWUSR, null_show, bitmap_store);
3262 size_show(mddev_t *mddev, char *page)
3264 return sprintf(page, "%llu\n",
3265 (unsigned long long)mddev->dev_sectors / 2);
3268 static int update_size(mddev_t *mddev, sector_t num_sectors);
3271 size_store(mddev_t *mddev, const char *buf, size_t len)
3273 /* If array is inactive, we can reduce the component size, but
3274 * not increase it (except from 0).
3275 * If array is active, we can try an on-line resize
3278 int err = strict_blocks_to_sectors(buf, §ors);
3283 err = update_size(mddev, sectors);
3284 md_update_sb(mddev, 1);
3286 if (mddev->dev_sectors == 0 ||
3287 mddev->dev_sectors > sectors)
3288 mddev->dev_sectors = sectors;
3292 return err ? err : len;
3295 static struct md_sysfs_entry md_size =
3296 __ATTR(component_size, S_IRUGO|S_IWUSR, size_show, size_store);
3301 * 'none' for arrays with no metadata (good luck...)
3302 * 'external' for arrays with externally managed metadata,
3303 * or N.M for internally known formats
3306 metadata_show(mddev_t *mddev, char *page)
3308 if (mddev->persistent)
3309 return sprintf(page, "%d.%d\n",
3310 mddev->major_version, mddev->minor_version);
3311 else if (mddev->external)
3312 return sprintf(page, "external:%s\n", mddev->metadata_type);
3314 return sprintf(page, "none\n");
3318 metadata_store(mddev_t *mddev, const char *buf, size_t len)
3322 /* Changing the details of 'external' metadata is
3323 * always permitted. Otherwise there must be
3324 * no devices attached to the array.
3326 if (mddev->external && strncmp(buf, "external:", 9) == 0)
3328 else if (!list_empty(&mddev->disks))
3331 if (cmd_match(buf, "none")) {
3332 mddev->persistent = 0;
3333 mddev->external = 0;
3334 mddev->major_version = 0;
3335 mddev->minor_version = 90;
3338 if (strncmp(buf, "external:", 9) == 0) {
3339 size_t namelen = len-9;
3340 if (namelen >= sizeof(mddev->metadata_type))
3341 namelen = sizeof(mddev->metadata_type)-1;
3342 strncpy(mddev->metadata_type, buf+9, namelen);
3343 mddev->metadata_type[namelen] = 0;
3344 if (namelen && mddev->metadata_type[namelen-1] == '\n')
3345 mddev->metadata_type[--namelen] = 0;
3346 mddev->persistent = 0;
3347 mddev->external = 1;
3348 mddev->major_version = 0;
3349 mddev->minor_version = 90;
3352 major = simple_strtoul(buf, &e, 10);
3353 if (e==buf || *e != '.')
3356 minor = simple_strtoul(buf, &e, 10);
3357 if (e==buf || (*e && *e != '\n') )
3359 if (major >= ARRAY_SIZE(super_types) || super_types[major].name == NULL)
3361 mddev->major_version = major;
3362 mddev->minor_version = minor;
3363 mddev->persistent = 1;
3364 mddev->external = 0;
3368 static struct md_sysfs_entry md_metadata =
3369 __ATTR(metadata_version, S_IRUGO|S_IWUSR, metadata_show, metadata_store);
3372 action_show(mddev_t *mddev, char *page)
3374 char *type = "idle";
3375 if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
3377 else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
3378 (!mddev->ro && test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))) {
3379 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
3381 else if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
3382 if (!test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
3384 else if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
3388 } else if (test_bit(MD_RECOVERY_RECOVER, &mddev->recovery))
3391 return sprintf(page, "%s\n", type);
3395 action_store(mddev_t *mddev, const char *page, size_t len)
3397 if (!mddev->pers || !mddev->pers->sync_request)
3400 if (cmd_match(page, "frozen"))
3401 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
3403 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
3405 if (cmd_match(page, "idle") || cmd_match(page, "frozen")) {
3406 if (mddev->sync_thread) {
3407 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
3408 md_unregister_thread(mddev->sync_thread);
3409 mddev->sync_thread = NULL;
3410 mddev->recovery = 0;
3412 } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
3413 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
3415 else if (cmd_match(page, "resync"))
3416 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3417 else if (cmd_match(page, "recover")) {
3418 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
3419 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3420 } else if (cmd_match(page, "reshape")) {
3422 if (mddev->pers->start_reshape == NULL)
3424 err = mddev->pers->start_reshape(mddev);
3427 sysfs_notify(&mddev->kobj, NULL, "degraded");
3429 if (cmd_match(page, "check"))
3430 set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
3431 else if (!cmd_match(page, "repair"))
3433 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
3434 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
3436 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3437 md_wakeup_thread(mddev->thread);
3438 sysfs_notify_dirent(mddev->sysfs_action);
3443 mismatch_cnt_show(mddev_t *mddev, char *page)
3445 return sprintf(page, "%llu\n",
3446 (unsigned long long) mddev->resync_mismatches);
3449 static struct md_sysfs_entry md_scan_mode =
3450 __ATTR(sync_action, S_IRUGO|S_IWUSR, action_show, action_store);
3453 static struct md_sysfs_entry md_mismatches = __ATTR_RO(mismatch_cnt);
3456 sync_min_show(mddev_t *mddev, char *page)
3458 return sprintf(page, "%d (%s)\n", speed_min(mddev),
3459 mddev->sync_speed_min ? "local": "system");
3463 sync_min_store(mddev_t *mddev, const char *buf, size_t len)
3467 if (strncmp(buf, "system", 6)==0) {
3468 mddev->sync_speed_min = 0;
3471 min = simple_strtoul(buf, &e, 10);
3472 if (buf == e || (*e && *e != '\n') || min <= 0)
3474 mddev->sync_speed_min = min;
3478 static struct md_sysfs_entry md_sync_min =
3479 __ATTR(sync_speed_min, S_IRUGO|S_IWUSR, sync_min_show, sync_min_store);
3482 sync_max_show(mddev_t *mddev, char *page)
3484 return sprintf(page, "%d (%s)\n", speed_max(mddev),
3485 mddev->sync_speed_max ? "local": "system");
3489 sync_max_store(mddev_t *mddev, const char *buf, size_t len)
3493 if (strncmp(buf, "system", 6)==0) {
3494 mddev->sync_speed_max = 0;
3497 max = simple_strtoul(buf, &e, 10);
3498 if (buf == e || (*e && *e != '\n') || max <= 0)
3500 mddev->sync_speed_max = max;
3504 static struct md_sysfs_entry md_sync_max =
3505 __ATTR(sync_speed_max, S_IRUGO|S_IWUSR, sync_max_show, sync_max_store);
3508 degraded_show(mddev_t *mddev, char *page)
3510 return sprintf(page, "%d\n", mddev->degraded);
3512 static struct md_sysfs_entry md_degraded = __ATTR_RO(degraded);
3515 sync_force_parallel_show(mddev_t *mddev, char *page)
3517 return sprintf(page, "%d\n", mddev->parallel_resync);
3521 sync_force_parallel_store(mddev_t *mddev, const char *buf, size_t len)
3525 if (strict_strtol(buf, 10, &n))
3528 if (n != 0 && n != 1)
3531 mddev->parallel_resync = n;
3533 if (mddev->sync_thread)
3534 wake_up(&resync_wait);
3539 /* force parallel resync, even with shared block devices */
3540 static struct md_sysfs_entry md_sync_force_parallel =
3541 __ATTR(sync_force_parallel, S_IRUGO|S_IWUSR,
3542 sync_force_parallel_show, sync_force_parallel_store);
3545 sync_speed_show(mddev_t *mddev, char *page)
3547 unsigned long resync, dt, db;
3548 if (mddev->curr_resync == 0)
3549 return sprintf(page, "none\n");
3550 resync = mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active);
3551 dt = (jiffies - mddev->resync_mark) / HZ;
3553 db = resync - mddev->resync_mark_cnt;
3554 return sprintf(page, "%lu\n", db/dt/2); /* K/sec */
3557 static struct md_sysfs_entry md_sync_speed = __ATTR_RO(sync_speed);
3560 sync_completed_show(mddev_t *mddev, char *page)
3562 unsigned long max_sectors, resync;
3564 if (!test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
3565 return sprintf(page, "none\n");
3567 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
3568 max_sectors = mddev->resync_max_sectors;
3570 max_sectors = mddev->dev_sectors;
3572 resync = mddev->curr_resync_completed;
3573 return sprintf(page, "%lu / %lu\n", resync, max_sectors);
3576 static struct md_sysfs_entry md_sync_completed = __ATTR_RO(sync_completed);
3579 min_sync_show(mddev_t *mddev, char *page)
3581 return sprintf(page, "%llu\n",
3582 (unsigned long long)mddev->resync_min);
3585 min_sync_store(mddev_t *mddev, const char *buf, size_t len)
3587 unsigned long long min;
3588 if (strict_strtoull(buf, 10, &min))
3590 if (min > mddev->resync_max)
3592 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
3595 /* Must be a multiple of chunk_size */
3596 if (mddev->chunk_sectors) {
3597 sector_t temp = min;
3598 if (sector_div(temp, mddev->chunk_sectors))
3601 mddev->resync_min = min;
3606 static struct md_sysfs_entry md_min_sync =
3607 __ATTR(sync_min, S_IRUGO|S_IWUSR, min_sync_show, min_sync_store);
3610 max_sync_show(mddev_t *mddev, char *page)
3612 if (mddev->resync_max == MaxSector)
3613 return sprintf(page, "max\n");
3615 return sprintf(page, "%llu\n",
3616 (unsigned long long)mddev->resync_max);
3619 max_sync_store(mddev_t *mddev, const char *buf, size_t len)
3621 if (strncmp(buf, "max", 3) == 0)
3622 mddev->resync_max = MaxSector;
3624 unsigned long long max;
3625 if (strict_strtoull(buf, 10, &max))
3627 if (max < mddev->resync_min)
3629 if (max < mddev->resync_max &&
3631 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
3634 /* Must be a multiple of chunk_size */
3635 if (mddev->chunk_sectors) {
3636 sector_t temp = max;
3637 if (sector_div(temp, mddev->chunk_sectors))
3640 mddev->resync_max = max;
3642 wake_up(&mddev->recovery_wait);
3646 static struct md_sysfs_entry md_max_sync =
3647 __ATTR(sync_max, S_IRUGO|S_IWUSR, max_sync_show, max_sync_store);
3650 suspend_lo_show(mddev_t *mddev, char *page)
3652 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_lo);
3656 suspend_lo_store(mddev_t *mddev, const char *buf, size_t len)
3659 unsigned long long new = simple_strtoull(buf, &e, 10);
3661 if (mddev->pers == NULL ||
3662 mddev->pers->quiesce == NULL)
3664 if (buf == e || (*e && *e != '\n'))
3666 if (new >= mddev->suspend_hi ||
3667 (new > mddev->suspend_lo && new < mddev->suspend_hi)) {
3668 mddev->suspend_lo = new;
3669 mddev->pers->quiesce(mddev, 2);
3674 static struct md_sysfs_entry md_suspend_lo =
3675 __ATTR(suspend_lo, S_IRUGO|S_IWUSR, suspend_lo_show, suspend_lo_store);
3679 suspend_hi_show(mddev_t *mddev, char *page)
3681 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_hi);
3685 suspend_hi_store(mddev_t *mddev, const char *buf, size_t len)
3688 unsigned long long new = simple_strtoull(buf, &e, 10);
3690 if (mddev->pers == NULL ||
3691 mddev->pers->quiesce == NULL)
3693 if (buf == e || (*e && *e != '\n'))
3695 if ((new <= mddev->suspend_lo && mddev->suspend_lo >= mddev->suspend_hi) ||
3696 (new > mddev->suspend_lo && new > mddev->suspend_hi)) {
3697 mddev->suspend_hi = new;
3698 mddev->pers->quiesce(mddev, 1);
3699 mddev->pers->quiesce(mddev, 0);
3704 static struct md_sysfs_entry md_suspend_hi =
3705 __ATTR(suspend_hi, S_IRUGO|S_IWUSR, suspend_hi_show, suspend_hi_store);
3708 reshape_position_show(mddev_t *mddev, char *page)
3710 if (mddev->reshape_position != MaxSector)
3711 return sprintf(page, "%llu\n",
3712 (unsigned long long)mddev->reshape_position);
3713 strcpy(page, "none\n");
3718 reshape_position_store(mddev_t *mddev, const char *buf, size_t len)
3721 unsigned long long new = simple_strtoull(buf, &e, 10);
3724 if (buf == e || (*e && *e != '\n'))
3726 mddev->reshape_position = new;
3727 mddev->delta_disks = 0;
3728 mddev->new_level = mddev->level;
3729 mddev->new_layout = mddev->layout;
3730 mddev->new_chunk_sectors = mddev->chunk_sectors;
3734 static struct md_sysfs_entry md_reshape_position =
3735 __ATTR(reshape_position, S_IRUGO|S_IWUSR, reshape_position_show,
3736 reshape_position_store);
3739 array_size_show(mddev_t *mddev, char *page)
3741 if (mddev->external_size)
3742 return sprintf(page, "%llu\n",
3743 (unsigned long long)mddev->array_sectors/2);
3745 return sprintf(page, "default\n");
3749 array_size_store(mddev_t *mddev, const char *buf, size_t len)
3753 if (strncmp(buf, "default", 7) == 0) {
3755 sectors = mddev->pers->size(mddev, 0, 0);
3757 sectors = mddev->array_sectors;
3759 mddev->external_size = 0;
3761 if (strict_blocks_to_sectors(buf, §ors) < 0)
3763 if (mddev->pers && mddev->pers->size(mddev, 0, 0) < sectors)
3766 mddev->external_size = 1;
3769 mddev->array_sectors = sectors;
3770 set_capacity(mddev->gendisk, mddev->array_sectors);
3772 revalidate_disk(mddev->gendisk);
3777 static struct md_sysfs_entry md_array_size =
3778 __ATTR(array_size, S_IRUGO|S_IWUSR, array_size_show,
3781 static struct attribute *md_default_attrs[] = {
3784 &md_raid_disks.attr,
3785 &md_chunk_size.attr,
3787 &md_resync_start.attr,
3789 &md_new_device.attr,
3790 &md_safe_delay.attr,
3791 &md_array_state.attr,
3792 &md_reshape_position.attr,
3793 &md_array_size.attr,
3797 static struct attribute *md_redundancy_attrs[] = {
3799 &md_mismatches.attr,
3802 &md_sync_speed.attr,
3803 &md_sync_force_parallel.attr,
3804 &md_sync_completed.attr,
3807 &md_suspend_lo.attr,
3808 &md_suspend_hi.attr,
3813 static struct attribute_group md_redundancy_group = {
3815 .attrs = md_redundancy_attrs,
3820 md_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
3822 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
3823 mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
3828 rv = mddev_lock(mddev);
3830 rv = entry->show(mddev, page);
3831 mddev_unlock(mddev);
3837 md_attr_store(struct kobject *kobj, struct attribute *attr,
3838 const char *page, size_t length)
3840 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
3841 mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
3846 if (!capable(CAP_SYS_ADMIN))
3848 rv = mddev_lock(mddev);
3849 if (mddev->hold_active == UNTIL_IOCTL)
3850 mddev->hold_active = 0;
3852 rv = entry->store(mddev, page, length);
3853 mddev_unlock(mddev);
3858 static void md_free(struct kobject *ko)
3860 mddev_t *mddev = container_of(ko, mddev_t, kobj);
3862 if (mddev->sysfs_state)
3863 sysfs_put(mddev->sysfs_state);
3865 if (mddev->gendisk) {
3866 del_gendisk(mddev->gendisk);
3867 put_disk(mddev->gendisk);
3870 blk_cleanup_queue(mddev->queue);
3875 static struct sysfs_ops md_sysfs_ops = {
3876 .show = md_attr_show,
3877 .store = md_attr_store,
3879 static struct kobj_type md_ktype = {
3881 .sysfs_ops = &md_sysfs_ops,
3882 .default_attrs = md_default_attrs,
3887 static void mddev_delayed_delete(struct work_struct *ws)
3889 mddev_t *mddev = container_of(ws, mddev_t, del_work);
3891 if (mddev->private == &md_redundancy_group) {
3892 sysfs_remove_group(&mddev->kobj, &md_redundancy_group);
3893 if (mddev->sysfs_action)
3894 sysfs_put(mddev->sysfs_action);
3895 mddev->sysfs_action = NULL;
3896 mddev->private = NULL;
3898 kobject_del(&mddev->kobj);
3899 kobject_put(&mddev->kobj);
3902 static int md_alloc(dev_t dev, char *name)
3904 static DEFINE_MUTEX(disks_mutex);
3905 mddev_t *mddev = mddev_find(dev);
3906 struct gendisk *disk;
3915 partitioned = (MAJOR(mddev->unit) != MD_MAJOR);
3916 shift = partitioned ? MdpMinorShift : 0;
3917 unit = MINOR(mddev->unit) >> shift;
3919 /* wait for any previous instance if this device
3920 * to be completed removed (mddev_delayed_delete).
3922 flush_scheduled_work();
3924 mutex_lock(&disks_mutex);
3930 /* Need to ensure that 'name' is not a duplicate.
3933 spin_lock(&all_mddevs_lock);
3935 list_for_each_entry(mddev2, &all_mddevs, all_mddevs)
3936 if (mddev2->gendisk &&
3937 strcmp(mddev2->gendisk->disk_name, name) == 0) {
3938 spin_unlock(&all_mddevs_lock);
3941 spin_unlock(&all_mddevs_lock);
3945 mddev->queue = blk_alloc_queue(GFP_KERNEL);
3948 mddev->queue->queuedata = mddev;
3950 /* Can be unlocked because the queue is new: no concurrency */
3951 queue_flag_set_unlocked(QUEUE_FLAG_CLUSTER, mddev->queue);
3953 blk_queue_make_request(mddev->queue, md_make_request);
3955 disk = alloc_disk(1 << shift);
3957 blk_cleanup_queue(mddev->queue);
3958 mddev->queue = NULL;
3961 disk->major = MAJOR(mddev->unit);
3962 disk->first_minor = unit << shift;
3964 strcpy(disk->disk_name, name);
3965 else if (partitioned)
3966 sprintf(disk->disk_name, "md_d%d", unit);
3968 sprintf(disk->disk_name, "md%d", unit);
3969 disk->fops = &md_fops;
3970 disk->private_data = mddev;
3971 disk->queue = mddev->queue;
3972 /* Allow extended partitions. This makes the
3973 * 'mdp' device redundant, but we can't really
3976 disk->flags |= GENHD_FL_EXT_DEVT;
3978 mddev->gendisk = disk;
3979 error = kobject_init_and_add(&mddev->kobj, &md_ktype,
3980 &disk_to_dev(disk)->kobj, "%s", "md");
3982 /* This isn't possible, but as kobject_init_and_add is marked
3983 * __must_check, we must do something with the result
3985 printk(KERN_WARNING "md: cannot register %s/md - name in use\n",
3990 mutex_unlock(&disks_mutex);
3992 kobject_uevent(&mddev->kobj, KOBJ_ADD);
3993 mddev->sysfs_state = sysfs_get_dirent(mddev->kobj.sd, "array_state");
3999 static struct kobject *md_probe(dev_t dev, int *part, void *data)
4001 md_alloc(dev, NULL);
4005 static int add_named_array(const char *val, struct kernel_param *kp)
4007 /* val must be "md_*" where * is not all digits.
4008 * We allocate an array with a large free minor number, and
4009 * set the name to val. val must not already be an active name.
4011 int len = strlen(val);
4012 char buf[DISK_NAME_LEN];
4014 while (len && val[len-1] == '\n')
4016 if (len >= DISK_NAME_LEN)
4018 strlcpy(buf, val, len+1);
4019 if (strncmp(buf, "md_", 3) != 0)
4021 return md_alloc(0, buf);
4024 static void md_safemode_timeout(unsigned long data)
4026 mddev_t *mddev = (mddev_t *) data;
4028 if (!atomic_read(&mddev->writes_pending)) {
4029 mddev->safemode = 1;
4030 if (mddev->external)
4031 sysfs_notify_dirent(mddev->sysfs_state);
4033 md_wakeup_thread(mddev->thread);
4036 static int start_dirty_degraded;
4038 static int do_md_run(mddev_t * mddev)
4042 struct gendisk *disk;
4043 struct mdk_personality *pers;
4045 if (list_empty(&mddev->disks))
4046 /* cannot run an array with no devices.. */
4053 * Analyze all RAID superblock(s)
4055 if (!mddev->raid_disks) {
4056 if (!mddev->persistent)
4061 if (mddev->level != LEVEL_NONE)
4062 request_module("md-level-%d", mddev->level);
4063 else if (mddev->clevel[0])
4064 request_module("md-%s", mddev->clevel);
4067 * Drop all container device buffers, from now on
4068 * the only valid external interface is through the md
4071 list_for_each_entry(rdev, &mddev->disks, same_set) {
4072 if (test_bit(Faulty, &rdev->flags))
4074 sync_blockdev(rdev->bdev);
4075 invalidate_bdev(rdev->bdev);
4077 /* perform some consistency tests on the device.
4078 * We don't want the data to overlap the metadata,
4079 * Internal Bitmap issues have been handled elsewhere.
4081 if (rdev->data_offset < rdev->sb_start) {
4082 if (mddev->dev_sectors &&
4083 rdev->data_offset + mddev->dev_sectors
4085 printk("md: %s: data overlaps metadata\n",
4090 if (rdev->sb_start + rdev->sb_size/512
4091 > rdev->data_offset) {
4092 printk("md: %s: metadata overlaps data\n",
4097 sysfs_notify_dirent(rdev->sysfs_state);
4100 md_probe(mddev->unit, NULL, NULL);
4101 disk = mddev->gendisk;
4105 spin_lock(&pers_lock);
4106 pers = find_pers(mddev->level, mddev->clevel);
4107 if (!pers || !try_module_get(pers->owner)) {
4108 spin_unlock(&pers_lock);
4109 if (mddev->level != LEVEL_NONE)
4110 printk(KERN_WARNING "md: personality for level %d is not loaded!\n",
4113 printk(KERN_WARNING "md: personality for level %s is not loaded!\n",
4118 spin_unlock(&pers_lock);
4119 if (mddev->level != pers->level) {
4120 mddev->level = pers->level;
4121 mddev->new_level = pers->level;
4123 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
4125 if (mddev->reshape_position != MaxSector &&
4126 pers->start_reshape == NULL) {
4127 /* This personality cannot handle reshaping... */
4129 module_put(pers->owner);
4133 if (pers->sync_request) {
4134 /* Warn if this is a potentially silly
4137 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
4141 list_for_each_entry(rdev, &mddev->disks, same_set)
4142 list_for_each_entry(rdev2, &mddev->disks, same_set) {
4144 rdev->bdev->bd_contains ==
4145 rdev2->bdev->bd_contains) {
4147 "%s: WARNING: %s appears to be"
4148 " on the same physical disk as"
4151 bdevname(rdev->bdev,b),
4152 bdevname(rdev2->bdev,b2));
4159 "True protection against single-disk"
4160 " failure might be compromised.\n");
4163 mddev->recovery = 0;
4164 /* may be over-ridden by personality */
4165 mddev->resync_max_sectors = mddev->dev_sectors;
4167 mddev->barriers_work = 1;
4168 mddev->ok_start_degraded = start_dirty_degraded;
4171 mddev->ro = 2; /* read-only, but switch on first write */
4173 err = mddev->pers->run(mddev);
4175 printk(KERN_ERR "md: pers->run() failed ...\n");
4176 else if (mddev->pers->size(mddev, 0, 0) < mddev->array_sectors) {
4177 WARN_ONCE(!mddev->external_size, "%s: default size too small,"
4178 " but 'external_size' not in effect?\n", __func__);
4180 "md: invalid array_size %llu > default size %llu\n",
4181 (unsigned long long)mddev->array_sectors / 2,
4182 (unsigned long long)mddev->pers->size(mddev, 0, 0) / 2);
4184 mddev->pers->stop(mddev);
4186 if (err == 0 && mddev->pers->sync_request) {
4187 err = bitmap_create(mddev);
4189 printk(KERN_ERR "%s: failed to create bitmap (%d)\n",
4190 mdname(mddev), err);
4191 mddev->pers->stop(mddev);
4195 module_put(mddev->pers->owner);
4197 bitmap_destroy(mddev);
4200 if (mddev->pers->sync_request) {
4201 if (sysfs_create_group(&mddev->kobj, &md_redundancy_group))
4203 "md: cannot register extra attributes for %s\n",
4205 mddev->sysfs_action = sysfs_get_dirent(mddev->kobj.sd, "sync_action");
4206 } else if (mddev->ro == 2) /* auto-readonly not meaningful */
4209 atomic_set(&mddev->writes_pending,0);
4210 mddev->safemode = 0;
4211 mddev->safemode_timer.function = md_safemode_timeout;
4212 mddev->safemode_timer.data = (unsigned long) mddev;
4213 mddev->safemode_delay = (200 * HZ)/1000 +1; /* 200 msec delay */
4216 list_for_each_entry(rdev, &mddev->disks, same_set)
4217 if (rdev->raid_disk >= 0) {
4219 sprintf(nm, "rd%d", rdev->raid_disk);
4220 if (sysfs_create_link(&mddev->kobj, &rdev->kobj, nm))
4221 printk("md: cannot register %s for %s\n",
4225 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4228 md_update_sb(mddev, 0);
4230 set_capacity(disk, mddev->array_sectors);
4232 /* If there is a partially-recovered drive we need to
4233 * start recovery here. If we leave it to md_check_recovery,
4234 * it will remove the drives and not do the right thing
4236 if (mddev->degraded && !mddev->sync_thread) {
4238 list_for_each_entry(rdev, &mddev->disks, same_set)
4239 if (rdev->raid_disk >= 0 &&
4240 !test_bit(In_sync, &rdev->flags) &&
4241 !test_bit(Faulty, &rdev->flags))
4242 /* complete an interrupted recovery */
4244 if (spares && mddev->pers->sync_request) {
4245 mddev->recovery = 0;
4246 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
4247 mddev->sync_thread = md_register_thread(md_do_sync,
4250 if (!mddev->sync_thread) {
4251 printk(KERN_ERR "%s: could not start resync"
4254 /* leave the spares where they are, it shouldn't hurt */
4255 mddev->recovery = 0;
4259 md_wakeup_thread(mddev->thread);
4260 md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
4262 revalidate_disk(mddev->gendisk);
4264 md_new_event(mddev);
4265 sysfs_notify_dirent(mddev->sysfs_state);
4266 if (mddev->sysfs_action)
4267 sysfs_notify_dirent(mddev->sysfs_action);
4268 sysfs_notify(&mddev->kobj, NULL, "degraded");
4269 kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
4273 static int restart_array(mddev_t *mddev)
4275 struct gendisk *disk = mddev->gendisk;
4277 /* Complain if it has no devices */
4278 if (list_empty(&mddev->disks))
4284 mddev->safemode = 0;
4286 set_disk_ro(disk, 0);
4287 printk(KERN_INFO "md: %s switched to read-write mode.\n",
4289 /* Kick recovery or resync if necessary */
4290 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4291 md_wakeup_thread(mddev->thread);
4292 md_wakeup_thread(mddev->sync_thread);
4293 sysfs_notify_dirent(mddev->sysfs_state);
4297 /* similar to deny_write_access, but accounts for our holding a reference
4298 * to the file ourselves */
4299 static int deny_bitmap_write_access(struct file * file)
4301 struct inode *inode = file->f_mapping->host;
4303 spin_lock(&inode->i_lock);
4304 if (atomic_read(&inode->i_writecount) > 1) {
4305 spin_unlock(&inode->i_lock);
4308 atomic_set(&inode->i_writecount, -1);
4309 spin_unlock(&inode->i_lock);
4314 static void restore_bitmap_write_access(struct file *file)
4316 struct inode *inode = file->f_mapping->host;
4318 spin_lock(&inode->i_lock);
4319 atomic_set(&inode->i_writecount, 1);
4320 spin_unlock(&inode->i_lock);
4324 * 0 - completely stop and dis-assemble array
4325 * 1 - switch to readonly
4326 * 2 - stop but do not disassemble array
4328 static int do_md_stop(mddev_t * mddev, int mode, int is_open)
4331 struct gendisk *disk = mddev->gendisk;
4334 mutex_lock(&mddev->open_mutex);
4335 if (atomic_read(&mddev->openers) > is_open) {
4336 printk("md: %s still in use.\n",mdname(mddev));
4338 } else if (mddev->pers) {
4340 if (mddev->sync_thread) {
4341 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4342 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
4343 md_unregister_thread(mddev->sync_thread);
4344 mddev->sync_thread = NULL;
4347 del_timer_sync(&mddev->safemode_timer);
4350 case 1: /* readonly */
4356 case 0: /* disassemble */
4358 bitmap_flush(mddev);
4359 md_super_wait(mddev);
4361 set_disk_ro(disk, 0);
4363 mddev->pers->stop(mddev);
4364 mddev->queue->merge_bvec_fn = NULL;
4365 mddev->queue->unplug_fn = NULL;
4366 mddev->queue->backing_dev_info.congested_fn = NULL;
4367 module_put(mddev->pers->owner);
4368 if (mddev->pers->sync_request)
4369 mddev->private = &md_redundancy_group;
4371 /* tell userspace to handle 'inactive' */
4372 sysfs_notify_dirent(mddev->sysfs_state);
4374 list_for_each_entry(rdev, &mddev->disks, same_set)
4375 if (rdev->raid_disk >= 0) {
4377 sprintf(nm, "rd%d", rdev->raid_disk);
4378 sysfs_remove_link(&mddev->kobj, nm);
4381 set_capacity(disk, 0);
4387 if (!mddev->in_sync || mddev->flags) {
4388 /* mark array as shutdown cleanly */
4390 md_update_sb(mddev, 1);
4393 set_disk_ro(disk, 1);
4394 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4398 mutex_unlock(&mddev->open_mutex);
4402 * Free resources if final stop
4406 printk(KERN_INFO "md: %s stopped.\n", mdname(mddev));
4408 bitmap_destroy(mddev);
4409 if (mddev->bitmap_file) {
4410 restore_bitmap_write_access(mddev->bitmap_file);
4411 fput(mddev->bitmap_file);
4412 mddev->bitmap_file = NULL;
4414 mddev->bitmap_offset = 0;
4416 /* make sure all md_delayed_delete calls have finished */
4417 flush_scheduled_work();
4419 export_array(mddev);
4421 mddev->array_sectors = 0;
4422 mddev->external_size = 0;
4423 mddev->dev_sectors = 0;
4424 mddev->raid_disks = 0;
4425 mddev->recovery_cp = 0;
4426 mddev->resync_min = 0;
4427 mddev->resync_max = MaxSector;
4428 mddev->reshape_position = MaxSector;
4429 mddev->external = 0;
4430 mddev->persistent = 0;
4431 mddev->level = LEVEL_NONE;
4432 mddev->clevel[0] = 0;
4435 mddev->metadata_type[0] = 0;
4436 mddev->chunk_sectors = 0;
4437 mddev->ctime = mddev->utime = 0;
4439 mddev->max_disks = 0;
4441 mddev->delta_disks = 0;
4442 mddev->new_level = LEVEL_NONE;
4443 mddev->new_layout = 0;
4444 mddev->new_chunk_sectors = 0;
4445 mddev->curr_resync = 0;
4446 mddev->resync_mismatches = 0;
4447 mddev->suspend_lo = mddev->suspend_hi = 0;
4448 mddev->sync_speed_min = mddev->sync_speed_max = 0;
4449 mddev->recovery = 0;
4452 mddev->degraded = 0;
4453 mddev->barriers_work = 0;
4454 mddev->safemode = 0;
4455 kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
4456 if (mddev->hold_active == UNTIL_STOP)
4457 mddev->hold_active = 0;
4459 } else if (mddev->pers)
4460 printk(KERN_INFO "md: %s switched to read-only mode.\n",
4463 blk_integrity_unregister(disk);
4464 md_new_event(mddev);
4465 sysfs_notify_dirent(mddev->sysfs_state);
4470 static void autorun_array(mddev_t *mddev)
4475 if (list_empty(&mddev->disks))
4478 printk(KERN_INFO "md: running: ");
4480 list_for_each_entry(rdev, &mddev->disks, same_set) {
4481 char b[BDEVNAME_SIZE];
4482 printk("<%s>", bdevname(rdev->bdev,b));
4486 err = do_md_run(mddev);
4488 printk(KERN_WARNING "md: do_md_run() returned %d\n", err);
4489 do_md_stop(mddev, 0, 0);
4494 * lets try to run arrays based on all disks that have arrived
4495 * until now. (those are in pending_raid_disks)
4497 * the method: pick the first pending disk, collect all disks with
4498 * the same UUID, remove all from the pending list and put them into
4499 * the 'same_array' list. Then order this list based on superblock
4500 * update time (freshest comes first), kick out 'old' disks and
4501 * compare superblocks. If everything's fine then run it.
4503 * If "unit" is allocated, then bump its reference count
4505 static void autorun_devices(int part)
4507 mdk_rdev_t *rdev0, *rdev, *tmp;
4509 char b[BDEVNAME_SIZE];
4511 printk(KERN_INFO "md: autorun ...\n");
4512 while (!list_empty(&pending_raid_disks)) {
4515 LIST_HEAD(candidates);
4516 rdev0 = list_entry(pending_raid_disks.next,
4517 mdk_rdev_t, same_set);
4519 printk(KERN_INFO "md: considering %s ...\n",
4520 bdevname(rdev0->bdev,b));
4521 INIT_LIST_HEAD(&candidates);
4522 rdev_for_each_list(rdev, tmp, &pending_raid_disks)
4523 if (super_90_load(rdev, rdev0, 0) >= 0) {
4524 printk(KERN_INFO "md: adding %s ...\n",
4525 bdevname(rdev->bdev,b));
4526 list_move(&rdev->same_set, &candidates);
4529 * now we have a set of devices, with all of them having
4530 * mostly sane superblocks. It's time to allocate the
4534 dev = MKDEV(mdp_major,
4535 rdev0->preferred_minor << MdpMinorShift);
4536 unit = MINOR(dev) >> MdpMinorShift;
4538 dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
4541 if (rdev0->preferred_minor != unit) {
4542 printk(KERN_INFO "md: unit number in %s is bad: %d\n",
4543 bdevname(rdev0->bdev, b), rdev0->preferred_minor);
4547 md_probe(dev, NULL, NULL);
4548 mddev = mddev_find(dev);
4549 if (!mddev || !mddev->gendisk) {
4553 "md: cannot allocate memory for md drive.\n");
4556 if (mddev_lock(mddev))
4557 printk(KERN_WARNING "md: %s locked, cannot run\n",
4559 else if (mddev->raid_disks || mddev->major_version
4560 || !list_empty(&mddev->disks)) {
4562 "md: %s already running, cannot run %s\n",
4563 mdname(mddev), bdevname(rdev0->bdev,b));
4564 mddev_unlock(mddev);
4566 printk(KERN_INFO "md: created %s\n", mdname(mddev));
4567 mddev->persistent = 1;
4568 rdev_for_each_list(rdev, tmp, &candidates) {
4569 list_del_init(&rdev->same_set);
4570 if (bind_rdev_to_array(rdev, mddev))
4573 autorun_array(mddev);
4574 mddev_unlock(mddev);
4576 /* on success, candidates will be empty, on error
4579 rdev_for_each_list(rdev, tmp, &candidates) {
4580 list_del_init(&rdev->same_set);
4585 printk(KERN_INFO "md: ... autorun DONE.\n");
4587 #endif /* !MODULE */
4589 static int get_version(void __user * arg)
4593 ver.major = MD_MAJOR_VERSION;
4594 ver.minor = MD_MINOR_VERSION;
4595 ver.patchlevel = MD_PATCHLEVEL_VERSION;
4597 if (copy_to_user(arg, &ver, sizeof(ver)))
4603 static int get_array_info(mddev_t * mddev, void __user * arg)
4605 mdu_array_info_t info;
4606 int nr,working,insync,failed,spare;
4609 nr=working=insync=failed=spare=0;
4610 list_for_each_entry(rdev, &mddev->disks, same_set) {
4612 if (test_bit(Faulty, &rdev->flags))
4616 if (test_bit(In_sync, &rdev->flags))
4623 info.major_version = mddev->major_version;
4624 info.minor_version = mddev->minor_version;
4625 info.patch_version = MD_PATCHLEVEL_VERSION;
4626 info.ctime = mddev->ctime;
4627 info.level = mddev->level;
4628 info.size = mddev->dev_sectors / 2;
4629 if (info.size != mddev->dev_sectors / 2) /* overflow */
4632 info.raid_disks = mddev->raid_disks;
4633 info.md_minor = mddev->md_minor;
4634 info.not_persistent= !mddev->persistent;
4636 info.utime = mddev->utime;
4639 info.state = (1<<MD_SB_CLEAN);
4640 if (mddev->bitmap && mddev->bitmap_offset)
4641 info.state = (1<<MD_SB_BITMAP_PRESENT);
4642 info.active_disks = insync;
4643 info.working_disks = working;
4644 info.failed_disks = failed;
4645 info.spare_disks = spare;
4647 info.layout = mddev->layout;
4648 info.chunk_size = mddev->chunk_sectors << 9;
4650 if (copy_to_user(arg, &info, sizeof(info)))
4656 static int get_bitmap_file(mddev_t * mddev, void __user * arg)
4658 mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */
4659 char *ptr, *buf = NULL;
4662 if (md_allow_write(mddev))
4663 file = kmalloc(sizeof(*file), GFP_NOIO);
4665 file = kmalloc(sizeof(*file), GFP_KERNEL);
4670 /* bitmap disabled, zero the first byte and copy out */
4671 if (!mddev->bitmap || !mddev->bitmap->file) {
4672 file->pathname[0] = '\0';
4676 buf = kmalloc(sizeof(file->pathname), GFP_KERNEL);
4680 ptr = d_path(&mddev->bitmap->file->f_path, buf, sizeof(file->pathname));
4684 strcpy(file->pathname, ptr);
4688 if (copy_to_user(arg, file, sizeof(*file)))
4696 static int get_disk_info(mddev_t * mddev, void __user * arg)
4698 mdu_disk_info_t info;
4701 if (copy_from_user(&info, arg, sizeof(info)))
4704 rdev = find_rdev_nr(mddev, info.number);
4706 info.major = MAJOR(rdev->bdev->bd_dev);
4707 info.minor = MINOR(rdev->bdev->bd_dev);
4708 info.raid_disk = rdev->raid_disk;
4710 if (test_bit(Faulty, &rdev->flags))
4711 info.state |= (1<<MD_DISK_FAULTY);
4712 else if (test_bit(In_sync, &rdev->flags)) {
4713 info.state |= (1<<MD_DISK_ACTIVE);
4714 info.state |= (1<<MD_DISK_SYNC);
4716 if (test_bit(WriteMostly, &rdev->flags))
4717 info.state |= (1<<MD_DISK_WRITEMOSTLY);
4719 info.major = info.minor = 0;
4720 info.raid_disk = -1;
4721 info.state = (1<<MD_DISK_REMOVED);
4724 if (copy_to_user(arg, &info, sizeof(info)))
4730 static int add_new_disk(mddev_t * mddev, mdu_disk_info_t *info)
4732 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
4734 dev_t dev = MKDEV(info->major,info->minor);
4736 if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
4739 if (!mddev->raid_disks) {
4741 /* expecting a device which has a superblock */
4742 rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
4745 "md: md_import_device returned %ld\n",
4747 return PTR_ERR(rdev);
4749 if (!list_empty(&mddev->disks)) {
4750 mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
4751 mdk_rdev_t, same_set);
4752 err = super_types[mddev->major_version]
4753 .load_super(rdev, rdev0, mddev->minor_version);
4756 "md: %s has different UUID to %s\n",
4757 bdevname(rdev->bdev,b),
4758 bdevname(rdev0->bdev,b2));
4763 err = bind_rdev_to_array(rdev, mddev);
4770 * add_new_disk can be used once the array is assembled
4771 * to add "hot spares". They must already have a superblock
4776 if (!mddev->pers->hot_add_disk) {
4778 "%s: personality does not support diskops!\n",
4782 if (mddev->persistent)
4783 rdev = md_import_device(dev, mddev->major_version,
4784 mddev->minor_version);
4786 rdev = md_import_device(dev, -1, -1);
4789 "md: md_import_device returned %ld\n",
4791 return PTR_ERR(rdev);
4793 /* set save_raid_disk if appropriate */
4794 if (!mddev->persistent) {
4795 if (info->state & (1<<MD_DISK_SYNC) &&
4796 info->raid_disk < mddev->raid_disks)
4797 rdev->raid_disk = info->raid_disk;
4799 rdev->raid_disk = -1;
4801 super_types[mddev->major_version].
4802 validate_super(mddev, rdev);
4803 rdev->saved_raid_disk = rdev->raid_disk;
4805 clear_bit(In_sync, &rdev->flags); /* just to be sure */
4806 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
4807 set_bit(WriteMostly, &rdev->flags);
4809 clear_bit(WriteMostly, &rdev->flags);
4811 rdev->raid_disk = -1;
4812 err = bind_rdev_to_array(rdev, mddev);
4813 if (!err && !mddev->pers->hot_remove_disk) {
4814 /* If there is hot_add_disk but no hot_remove_disk
4815 * then added disks for geometry changes,
4816 * and should be added immediately.
4818 super_types[mddev->major_version].
4819 validate_super(mddev, rdev);
4820 err = mddev->pers->hot_add_disk(mddev, rdev);
4822 unbind_rdev_from_array(rdev);
4827 sysfs_notify_dirent(rdev->sysfs_state);
4829 md_update_sb(mddev, 1);
4830 if (mddev->degraded)
4831 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
4832 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4833 md_wakeup_thread(mddev->thread);
4837 /* otherwise, add_new_disk is only allowed
4838 * for major_version==0 superblocks
4840 if (mddev->major_version != 0) {
4841 printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n",
4846 if (!(info->state & (1<<MD_DISK_FAULTY))) {
4848 rdev = md_import_device(dev, -1, 0);
4851 "md: error, md_import_device() returned %ld\n",
4853 return PTR_ERR(rdev);
4855 rdev->desc_nr = info->number;
4856 if (info->raid_disk < mddev->raid_disks)
4857 rdev->raid_disk = info->raid_disk;
4859 rdev->raid_disk = -1;
4861 if (rdev->raid_disk < mddev->raid_disks)
4862 if (info->state & (1<<MD_DISK_SYNC))
4863 set_bit(In_sync, &rdev->flags);
4865 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
4866 set_bit(WriteMostly, &rdev->flags);
4868 if (!mddev->persistent) {
4869 printk(KERN_INFO "md: nonpersistent superblock ...\n");
4870 rdev->sb_start = rdev->bdev->bd_inode->i_size / 512;
4872 rdev->sb_start = calc_dev_sboffset(rdev->bdev);
4873 rdev->sectors = rdev->sb_start;
4875 err = bind_rdev_to_array(rdev, mddev);
4885 static int hot_remove_disk(mddev_t * mddev, dev_t dev)
4887 char b[BDEVNAME_SIZE];
4890 rdev = find_rdev(mddev, dev);
4894 if (rdev->raid_disk >= 0)
4897 kick_rdev_from_array(rdev);
4898 md_update_sb(mddev, 1);
4899 md_new_event(mddev);
4903 printk(KERN_WARNING "md: cannot remove active disk %s from %s ...\n",
4904 bdevname(rdev->bdev,b), mdname(mddev));
4908 static int hot_add_disk(mddev_t * mddev, dev_t dev)
4910 char b[BDEVNAME_SIZE];
4917 if (mddev->major_version != 0) {
4918 printk(KERN_WARNING "%s: HOT_ADD may only be used with"
4919 " version-0 superblocks.\n",
4923 if (!mddev->pers->hot_add_disk) {
4925 "%s: personality does not support diskops!\n",
4930 rdev = md_import_device(dev, -1, 0);
4933 "md: error, md_import_device() returned %ld\n",
4938 if (mddev->persistent)
4939 rdev->sb_start = calc_dev_sboffset(rdev->bdev);
4941 rdev->sb_start = rdev->bdev->bd_inode->i_size / 512;
4943 rdev->sectors = rdev->sb_start;
4945 if (test_bit(Faulty, &rdev->flags)) {
4947 "md: can not hot-add faulty %s disk to %s!\n",
4948 bdevname(rdev->bdev,b), mdname(mddev));
4952 clear_bit(In_sync, &rdev->flags);
4954 rdev->saved_raid_disk = -1;
4955 err = bind_rdev_to_array(rdev, mddev);
4960 * The rest should better be atomic, we can have disk failures
4961 * noticed in interrupt contexts ...
4964 rdev->raid_disk = -1;
4966 md_update_sb(mddev, 1);
4969 * Kick recovery, maybe this spare has to be added to the
4970 * array immediately.
4972 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4973 md_wakeup_thread(mddev->thread);
4974 md_new_event(mddev);
4982 static int set_bitmap_file(mddev_t *mddev, int fd)
4987 if (!mddev->pers->quiesce)
4989 if (mddev->recovery || mddev->sync_thread)
4991 /* we should be able to change the bitmap.. */
4997 return -EEXIST; /* cannot add when bitmap is present */
4998 mddev->bitmap_file = fget(fd);
5000 if (mddev->bitmap_file == NULL) {
5001 printk(KERN_ERR "%s: error: failed to get bitmap file\n",
5006 err = deny_bitmap_write_access(mddev->bitmap_file);
5008 printk(KERN_ERR "%s: error: bitmap file is already in use\n",
5010 fput(mddev->bitmap_file);
5011 mddev->bitmap_file = NULL;
5014 mddev->bitmap_offset = 0; /* file overrides offset */
5015 } else if (mddev->bitmap == NULL)
5016 return -ENOENT; /* cannot remove what isn't there */
5019 mddev->pers->quiesce(mddev, 1);
5021 err = bitmap_create(mddev);
5022 if (fd < 0 || err) {
5023 bitmap_destroy(mddev);
5024 fd = -1; /* make sure to put the file */
5026 mddev->pers->quiesce(mddev, 0);
5029 if (mddev->bitmap_file) {
5030 restore_bitmap_write_access(mddev->bitmap_file);
5031 fput(mddev->bitmap_file);
5033 mddev->bitmap_file = NULL;
5040 * set_array_info is used two different ways
5041 * The original usage is when creating a new array.
5042 * In this usage, raid_disks is > 0 and it together with
5043 * level, size, not_persistent,layout,chunksize determine the
5044 * shape of the array.
5045 * This will always create an array with a type-0.90.0 superblock.
5046 * The newer usage is when assembling an array.
5047 * In this case raid_disks will be 0, and the major_version field is
5048 * use to determine which style super-blocks are to be found on the devices.
5049 * The minor and patch _version numbers are also kept incase the
5050 * super_block handler wishes to interpret them.
5052 static int set_array_info(mddev_t * mddev, mdu_array_info_t *info)
5055 if (info->raid_disks == 0) {
5056 /* just setting version number for superblock loading */
5057 if (info->major_version < 0 ||
5058 info->major_version >= ARRAY_SIZE(super_types) ||
5059 super_types[info->major_version].name == NULL) {
5060 /* maybe try to auto-load a module? */
5062 "md: superblock version %d not known\n",
5063 info->major_version);
5066 mddev->major_version = info->major_version;
5067 mddev->minor_version = info->minor_version;
5068 mddev->patch_version = info->patch_version;
5069 mddev->persistent = !info->not_persistent;
5072 mddev->major_version = MD_MAJOR_VERSION;
5073 mddev->minor_version = MD_MINOR_VERSION;
5074 mddev->patch_version = MD_PATCHLEVEL_VERSION;
5075 mddev->ctime = get_seconds();
5077 mddev->level = info->level;
5078 mddev->clevel[0] = 0;
5079 mddev->dev_sectors = 2 * (sector_t)info->size;
5080 mddev->raid_disks = info->raid_disks;
5081 /* don't set md_minor, it is determined by which /dev/md* was
5084 if (info->state & (1<<MD_SB_CLEAN))
5085 mddev->recovery_cp = MaxSector;
5087 mddev->recovery_cp = 0;
5088 mddev->persistent = ! info->not_persistent;
5089 mddev->external = 0;
5091 mddev->layout = info->layout;
5092 mddev->chunk_sectors = info->chunk_size >> 9;
5094 mddev->max_disks = MD_SB_DISKS;
5096 if (mddev->persistent)
5098 set_bit(MD_CHANGE_DEVS, &mddev->flags);
5100 mddev->default_bitmap_offset = MD_SB_BYTES >> 9;
5101 mddev->bitmap_offset = 0;
5103 mddev->reshape_position = MaxSector;
5106 * Generate a 128 bit UUID
5108 get_random_bytes(mddev->uuid, 16);
5110 mddev->new_level = mddev->level;
5111 mddev->new_chunk_sectors = mddev->chunk_sectors;
5112 mddev->new_layout = mddev->layout;
5113 mddev->delta_disks = 0;
5118 void md_set_array_sectors(mddev_t *mddev, sector_t array_sectors)
5120 WARN(!mddev_is_locked(mddev), "%s: unlocked mddev!\n", __func__);
5122 if (mddev->external_size)
5125 mddev->array_sectors = array_sectors;
5127 EXPORT_SYMBOL(md_set_array_sectors);
5129 static int update_size(mddev_t *mddev, sector_t num_sectors)
5133 int fit = (num_sectors == 0);
5135 if (mddev->pers->resize == NULL)
5137 /* The "num_sectors" is the number of sectors of each device that
5138 * is used. This can only make sense for arrays with redundancy.
5139 * linear and raid0 always use whatever space is available. We can only
5140 * consider changing this number if no resync or reconstruction is
5141 * happening, and if the new size is acceptable. It must fit before the
5142 * sb_start or, if that is <data_offset, it must fit before the size
5143 * of each device. If num_sectors is zero, we find the largest size
5147 if (mddev->sync_thread)
5150 /* Sorry, cannot grow a bitmap yet, just remove it,
5154 list_for_each_entry(rdev, &mddev->disks, same_set) {
5155 sector_t avail = rdev->sectors;
5157 if (fit && (num_sectors == 0 || num_sectors > avail))
5158 num_sectors = avail;
5159 if (avail < num_sectors)
5162 rv = mddev->pers->resize(mddev, num_sectors);
5164 revalidate_disk(mddev->gendisk);
5168 static int update_raid_disks(mddev_t *mddev, int raid_disks)
5171 /* change the number of raid disks */
5172 if (mddev->pers->check_reshape == NULL)
5174 if (raid_disks <= 0 ||
5175 raid_disks >= mddev->max_disks)
5177 if (mddev->sync_thread || mddev->reshape_position != MaxSector)
5179 mddev->delta_disks = raid_disks - mddev->raid_disks;
5181 rv = mddev->pers->check_reshape(mddev);
5187 * update_array_info is used to change the configuration of an
5189 * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
5190 * fields in the info are checked against the array.
5191 * Any differences that cannot be handled will cause an error.
5192 * Normally, only one change can be managed at a time.
5194 static int update_array_info(mddev_t *mddev, mdu_array_info_t *info)
5200 /* calculate expected state,ignoring low bits */
5201 if (mddev->bitmap && mddev->bitmap_offset)
5202 state |= (1 << MD_SB_BITMAP_PRESENT);
5204 if (mddev->major_version != info->major_version ||
5205 mddev->minor_version != info->minor_version ||
5206 /* mddev->patch_version != info->patch_version || */
5207 mddev->ctime != info->ctime ||
5208 mddev->level != info->level ||
5209 /* mddev->layout != info->layout || */
5210 !mddev->persistent != info->not_persistent||
5211 mddev->chunk_sectors != info->chunk_size >> 9 ||
5212 /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
5213 ((state^info->state) & 0xfffffe00)
5216 /* Check there is only one change */
5217 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
5219 if (mddev->raid_disks != info->raid_disks)
5221 if (mddev->layout != info->layout)
5223 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT))
5230 if (mddev->layout != info->layout) {
5232 * we don't need to do anything at the md level, the
5233 * personality will take care of it all.
5235 if (mddev->pers->check_reshape == NULL)
5238 mddev->new_layout = info->layout;
5239 rv = mddev->pers->check_reshape(mddev);
5241 mddev->new_layout = mddev->layout;
5245 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
5246 rv = update_size(mddev, (sector_t)info->size * 2);
5248 if (mddev->raid_disks != info->raid_disks)
5249 rv = update_raid_disks(mddev, info->raid_disks);
5251 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) {
5252 if (mddev->pers->quiesce == NULL)
5254 if (mddev->recovery || mddev->sync_thread)
5256 if (info->state & (1<<MD_SB_BITMAP_PRESENT)) {
5257 /* add the bitmap */
5260 if (mddev->default_bitmap_offset == 0)
5262 mddev->bitmap_offset = mddev->default_bitmap_offset;
5263 mddev->pers->quiesce(mddev, 1);
5264 rv = bitmap_create(mddev);
5266 bitmap_destroy(mddev);
5267 mddev->pers->quiesce(mddev, 0);
5269 /* remove the bitmap */
5272 if (mddev->bitmap->file)
5274 mddev->pers->quiesce(mddev, 1);
5275 bitmap_destroy(mddev);
5276 mddev->pers->quiesce(mddev, 0);
5277 mddev->bitmap_offset = 0;
5280 md_update_sb(mddev, 1);
5284 static int set_disk_faulty(mddev_t *mddev, dev_t dev)
5288 if (mddev->pers == NULL)
5291 rdev = find_rdev(mddev, dev);
5295 md_error(mddev, rdev);
5300 * We have a problem here : there is no easy way to give a CHS
5301 * virtual geometry. We currently pretend that we have a 2 heads
5302 * 4 sectors (with a BIG number of cylinders...). This drives
5303 * dosfs just mad... ;-)
5305 static int md_getgeo(struct block_device *bdev, struct hd_geometry *geo)
5307 mddev_t *mddev = bdev->bd_disk->private_data;
5311 geo->cylinders = get_capacity(mddev->gendisk) / 8;
5315 static int md_ioctl(struct block_device *bdev, fmode_t mode,
5316 unsigned int cmd, unsigned long arg)
5319 void __user *argp = (void __user *)arg;
5320 mddev_t *mddev = NULL;
5322 if (!capable(CAP_SYS_ADMIN))
5326 * Commands dealing with the RAID driver but not any
5332 err = get_version(argp);
5335 case PRINT_RAID_DEBUG:
5343 autostart_arrays(arg);
5350 * Commands creating/starting a new array:
5353 mddev = bdev->bd_disk->private_data;
5360 err = mddev_lock(mddev);
5363 "md: ioctl lock interrupted, reason %d, cmd %d\n",
5370 case SET_ARRAY_INFO:
5372 mdu_array_info_t info;
5374 memset(&info, 0, sizeof(info));
5375 else if (copy_from_user(&info, argp, sizeof(info))) {
5380 err = update_array_info(mddev, &info);
5382 printk(KERN_WARNING "md: couldn't update"
5383 " array info. %d\n", err);
5388 if (!list_empty(&mddev->disks)) {
5390 "md: array %s already has disks!\n",
5395 if (mddev->raid_disks) {
5397 "md: array %s already initialised!\n",
5402 err = set_array_info(mddev, &info);
5404 printk(KERN_WARNING "md: couldn't set"
5405 " array info. %d\n", err);
5415 * Commands querying/configuring an existing array:
5417 /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
5418 * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */
5419 if ((!mddev->raid_disks && !mddev->external)
5420 && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY
5421 && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE
5422 && cmd != GET_BITMAP_FILE) {
5428 * Commands even a read-only array can execute:
5432 case GET_ARRAY_INFO:
5433 err = get_array_info(mddev, argp);
5436 case GET_BITMAP_FILE:
5437 err = get_bitmap_file(mddev, argp);
5441 err = get_disk_info(mddev, argp);
5444 case RESTART_ARRAY_RW:
5445 err = restart_array(mddev);
5449 err = do_md_stop(mddev, 0, 1);
5453 err = do_md_stop(mddev, 1, 1);
5459 * The remaining ioctls are changing the state of the
5460 * superblock, so we do not allow them on read-only arrays.
5461 * However non-MD ioctls (e.g. get-size) will still come through
5462 * here and hit the 'default' below, so only disallow
5463 * 'md' ioctls, and switch to rw mode if started auto-readonly.
5465 if (_IOC_TYPE(cmd) == MD_MAJOR && mddev->ro && mddev->pers) {
5466 if (mddev->ro == 2) {
5468 sysfs_notify_dirent(mddev->sysfs_state);
5469 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5470 md_wakeup_thread(mddev->thread);
5481 mdu_disk_info_t info;
5482 if (copy_from_user(&info, argp, sizeof(info)))
5485 err = add_new_disk(mddev, &info);
5489 case HOT_REMOVE_DISK:
5490 err = hot_remove_disk(mddev, new_decode_dev(arg));
5494 err = hot_add_disk(mddev, new_decode_dev(arg));
5497 case SET_DISK_FAULTY:
5498 err = set_disk_faulty(mddev, new_decode_dev(arg));
5502 err = do_md_run(mddev);
5505 case SET_BITMAP_FILE:
5506 err = set_bitmap_file(mddev, (int)arg);
5516 if (mddev->hold_active == UNTIL_IOCTL &&
5518 mddev->hold_active = 0;
5519 mddev_unlock(mddev);
5529 static int md_open(struct block_device *bdev, fmode_t mode)
5532 * Succeed if we can lock the mddev, which confirms that
5533 * it isn't being stopped right now.
5535 mddev_t *mddev = mddev_find(bdev->bd_dev);
5538 if (mddev->gendisk != bdev->bd_disk) {
5539 /* we are racing with mddev_put which is discarding this
5543 /* Wait until bdev->bd_disk is definitely gone */
5544 flush_scheduled_work();
5545 /* Then retry the open from the top */
5546 return -ERESTARTSYS;
5548 BUG_ON(mddev != bdev->bd_disk->private_data);
5550 if ((err = mutex_lock_interruptible(&mddev->open_mutex)))
5554 atomic_inc(&mddev->openers);
5555 mutex_unlock(&mddev->open_mutex);
5557 check_disk_change(bdev);
5562 static int md_release(struct gendisk *disk, fmode_t mode)
5564 mddev_t *mddev = disk->private_data;
5567 atomic_dec(&mddev->openers);
5573 static int md_media_changed(struct gendisk *disk)
5575 mddev_t *mddev = disk->private_data;
5577 return mddev->changed;
5580 static int md_revalidate(struct gendisk *disk)
5582 mddev_t *mddev = disk->private_data;
5587 static const struct block_device_operations md_fops =
5589 .owner = THIS_MODULE,
5591 .release = md_release,
5593 .getgeo = md_getgeo,
5594 .media_changed = md_media_changed,
5595 .revalidate_disk= md_revalidate,
5598 static int md_thread(void * arg)
5600 mdk_thread_t *thread = arg;
5603 * md_thread is a 'system-thread', it's priority should be very
5604 * high. We avoid resource deadlocks individually in each
5605 * raid personality. (RAID5 does preallocation) We also use RR and
5606 * the very same RT priority as kswapd, thus we will never get
5607 * into a priority inversion deadlock.
5609 * we definitely have to have equal or higher priority than
5610 * bdflush, otherwise bdflush will deadlock if there are too
5611 * many dirty RAID5 blocks.
5614 allow_signal(SIGKILL);
5615 while (!kthread_should_stop()) {
5617 /* We need to wait INTERRUPTIBLE so that
5618 * we don't add to the load-average.
5619 * That means we need to be sure no signals are
5622 if (signal_pending(current))
5623 flush_signals(current);
5625 wait_event_interruptible_timeout
5627 test_bit(THREAD_WAKEUP, &thread->flags)
5628 || kthread_should_stop(),
5631 clear_bit(THREAD_WAKEUP, &thread->flags);
5633 thread->run(thread->mddev);
5639 void md_wakeup_thread(mdk_thread_t *thread)
5642 dprintk("md: waking up MD thread %s.\n", thread->tsk->comm);
5643 set_bit(THREAD_WAKEUP, &thread->flags);
5644 wake_up(&thread->wqueue);
5648 mdk_thread_t *md_register_thread(void (*run) (mddev_t *), mddev_t *mddev,
5651 mdk_thread_t *thread;
5653 thread = kzalloc(sizeof(mdk_thread_t), GFP_KERNEL);
5657 init_waitqueue_head(&thread->wqueue);
5660 thread->mddev = mddev;
5661 thread->timeout = MAX_SCHEDULE_TIMEOUT;
5662 thread->tsk = kthread_run(md_thread, thread,
5664 mdname(thread->mddev),
5665 name ?: mddev->pers->name);
5666 if (IS_ERR(thread->tsk)) {
5673 void md_unregister_thread(mdk_thread_t *thread)
5677 dprintk("interrupting MD-thread pid %d\n", task_pid_nr(thread->tsk));
5679 kthread_stop(thread->tsk);
5683 void md_error(mddev_t *mddev, mdk_rdev_t *rdev)
5690 if (!rdev || test_bit(Faulty, &rdev->flags))
5693 if (mddev->external)
5694 set_bit(Blocked, &rdev->flags);
5696 dprintk("md_error dev:%s, rdev:(%d:%d), (caller: %p,%p,%p,%p).\n",
5698 MAJOR(rdev->bdev->bd_dev), MINOR(rdev->bdev->bd_dev),
5699 __builtin_return_address(0),__builtin_return_address(1),
5700 __builtin_return_address(2),__builtin_return_address(3));
5704 if (!mddev->pers->error_handler)
5706 mddev->pers->error_handler(mddev,rdev);
5707 if (mddev->degraded)
5708 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
5709 set_bit(StateChanged, &rdev->flags);
5710 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5711 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5712 md_wakeup_thread(mddev->thread);
5713 md_new_event_inintr(mddev);
5716 /* seq_file implementation /proc/mdstat */
5718 static void status_unused(struct seq_file *seq)
5723 seq_printf(seq, "unused devices: ");
5725 list_for_each_entry(rdev, &pending_raid_disks, same_set) {
5726 char b[BDEVNAME_SIZE];
5728 seq_printf(seq, "%s ",
5729 bdevname(rdev->bdev,b));
5732 seq_printf(seq, "<none>");
5734 seq_printf(seq, "\n");
5738 static void status_resync(struct seq_file *seq, mddev_t * mddev)
5740 sector_t max_sectors, resync, res;
5741 unsigned long dt, db;
5744 unsigned int per_milli;
5746 resync = mddev->curr_resync - atomic_read(&mddev->recovery_active);
5748 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
5749 max_sectors = mddev->resync_max_sectors;
5751 max_sectors = mddev->dev_sectors;
5754 * Should not happen.
5760 /* Pick 'scale' such that (resync>>scale)*1000 will fit
5761 * in a sector_t, and (max_sectors>>scale) will fit in a
5762 * u32, as those are the requirements for sector_div.
5763 * Thus 'scale' must be at least 10
5766 if (sizeof(sector_t) > sizeof(unsigned long)) {
5767 while ( max_sectors/2 > (1ULL<<(scale+32)))
5770 res = (resync>>scale)*1000;
5771 sector_div(res, (u32)((max_sectors>>scale)+1));
5775 int i, x = per_milli/50, y = 20-x;
5776 seq_printf(seq, "[");
5777 for (i = 0; i < x; i++)
5778 seq_printf(seq, "=");
5779 seq_printf(seq, ">");
5780 for (i = 0; i < y; i++)
5781 seq_printf(seq, ".");
5782 seq_printf(seq, "] ");
5784 seq_printf(seq, " %s =%3u.%u%% (%llu/%llu)",
5785 (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)?
5787 (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)?
5789 (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
5790 "resync" : "recovery"))),
5791 per_milli/10, per_milli % 10,
5792 (unsigned long long) resync/2,
5793 (unsigned long long) max_sectors/2);
5796 * dt: time from mark until now
5797 * db: blocks written from mark until now
5798 * rt: remaining time
5800 * rt is a sector_t, so could be 32bit or 64bit.
5801 * So we divide before multiply in case it is 32bit and close
5803 * We scale the divisor (db) by 32 to avoid loosing precision
5804 * near the end of resync when the number of remaining sectors
5806 * We then divide rt by 32 after multiplying by db to compensate.
5807 * The '+1' avoids division by zero if db is very small.
5809 dt = ((jiffies - mddev->resync_mark) / HZ);
5811 db = (mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active))
5812 - mddev->resync_mark_cnt;
5814 rt = max_sectors - resync; /* number of remaining sectors */
5815 sector_div(rt, db/32+1);
5819 seq_printf(seq, " finish=%lu.%lumin", (unsigned long)rt / 60,
5820 ((unsigned long)rt % 60)/6);
5822 seq_printf(seq, " speed=%ldK/sec", db/2/dt);
5825 static void *md_seq_start(struct seq_file *seq, loff_t *pos)
5827 struct list_head *tmp;
5837 spin_lock(&all_mddevs_lock);
5838 list_for_each(tmp,&all_mddevs)
5840 mddev = list_entry(tmp, mddev_t, all_mddevs);
5842 spin_unlock(&all_mddevs_lock);
5845 spin_unlock(&all_mddevs_lock);
5847 return (void*)2;/* tail */
5851 static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
5853 struct list_head *tmp;
5854 mddev_t *next_mddev, *mddev = v;
5860 spin_lock(&all_mddevs_lock);
5862 tmp = all_mddevs.next;
5864 tmp = mddev->all_mddevs.next;
5865 if (tmp != &all_mddevs)
5866 next_mddev = mddev_get(list_entry(tmp,mddev_t,all_mddevs));
5868 next_mddev = (void*)2;
5871 spin_unlock(&all_mddevs_lock);
5879 static void md_seq_stop(struct seq_file *seq, void *v)
5883 if (mddev && v != (void*)1 && v != (void*)2)
5887 struct mdstat_info {
5891 static int md_seq_show(struct seq_file *seq, void *v)
5896 struct mdstat_info *mi = seq->private;
5897 struct bitmap *bitmap;
5899 if (v == (void*)1) {
5900 struct mdk_personality *pers;
5901 seq_printf(seq, "Personalities : ");
5902 spin_lock(&pers_lock);
5903 list_for_each_entry(pers, &pers_list, list)
5904 seq_printf(seq, "[%s] ", pers->name);
5906 spin_unlock(&pers_lock);
5907 seq_printf(seq, "\n");
5908 mi->event = atomic_read(&md_event_count);
5911 if (v == (void*)2) {
5916 if (mddev_lock(mddev) < 0)
5919 if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
5920 seq_printf(seq, "%s : %sactive", mdname(mddev),
5921 mddev->pers ? "" : "in");
5924 seq_printf(seq, " (read-only)");
5926 seq_printf(seq, " (auto-read-only)");
5927 seq_printf(seq, " %s", mddev->pers->name);
5931 list_for_each_entry(rdev, &mddev->disks, same_set) {
5932 char b[BDEVNAME_SIZE];
5933 seq_printf(seq, " %s[%d]",
5934 bdevname(rdev->bdev,b), rdev->desc_nr);
5935 if (test_bit(WriteMostly, &rdev->flags))
5936 seq_printf(seq, "(W)");
5937 if (test_bit(Faulty, &rdev->flags)) {
5938 seq_printf(seq, "(F)");
5940 } else if (rdev->raid_disk < 0)
5941 seq_printf(seq, "(S)"); /* spare */
5942 sectors += rdev->sectors;
5945 if (!list_empty(&mddev->disks)) {
5947 seq_printf(seq, "\n %llu blocks",
5948 (unsigned long long)
5949 mddev->array_sectors / 2);
5951 seq_printf(seq, "\n %llu blocks",
5952 (unsigned long long)sectors / 2);
5954 if (mddev->persistent) {
5955 if (mddev->major_version != 0 ||
5956 mddev->minor_version != 90) {
5957 seq_printf(seq," super %d.%d",
5958 mddev->major_version,
5959 mddev->minor_version);
5961 } else if (mddev->external)
5962 seq_printf(seq, " super external:%s",
5963 mddev->metadata_type);
5965 seq_printf(seq, " super non-persistent");
5968 mddev->pers->status(seq, mddev);
5969 seq_printf(seq, "\n ");
5970 if (mddev->pers->sync_request) {
5971 if (mddev->curr_resync > 2) {
5972 status_resync(seq, mddev);
5973 seq_printf(seq, "\n ");
5974 } else if (mddev->curr_resync == 1 || mddev->curr_resync == 2)
5975 seq_printf(seq, "\tresync=DELAYED\n ");
5976 else if (mddev->recovery_cp < MaxSector)
5977 seq_printf(seq, "\tresync=PENDING\n ");
5980 seq_printf(seq, "\n ");
5982 if ((bitmap = mddev->bitmap)) {
5983 unsigned long chunk_kb;
5984 unsigned long flags;
5985 spin_lock_irqsave(&bitmap->lock, flags);
5986 chunk_kb = bitmap->chunksize >> 10;
5987 seq_printf(seq, "bitmap: %lu/%lu pages [%luKB], "
5989 bitmap->pages - bitmap->missing_pages,
5991 (bitmap->pages - bitmap->missing_pages)
5992 << (PAGE_SHIFT - 10),
5993 chunk_kb ? chunk_kb : bitmap->chunksize,
5994 chunk_kb ? "KB" : "B");
5996 seq_printf(seq, ", file: ");
5997 seq_path(seq, &bitmap->file->f_path, " \t\n");
6000 seq_printf(seq, "\n");
6001 spin_unlock_irqrestore(&bitmap->lock, flags);
6004 seq_printf(seq, "\n");
6006 mddev_unlock(mddev);
6011 static const struct seq_operations md_seq_ops = {
6012 .start = md_seq_start,
6013 .next = md_seq_next,
6014 .stop = md_seq_stop,
6015 .show = md_seq_show,
6018 static int md_seq_open(struct inode *inode, struct file *file)
6021 struct mdstat_info *mi = kmalloc(sizeof(*mi), GFP_KERNEL);
6025 error = seq_open(file, &md_seq_ops);
6029 struct seq_file *p = file->private_data;
6031 mi->event = atomic_read(&md_event_count);
6036 static unsigned int mdstat_poll(struct file *filp, poll_table *wait)
6038 struct seq_file *m = filp->private_data;
6039 struct mdstat_info *mi = m->private;
6042 poll_wait(filp, &md_event_waiters, wait);
6044 /* always allow read */
6045 mask = POLLIN | POLLRDNORM;
6047 if (mi->event != atomic_read(&md_event_count))
6048 mask |= POLLERR | POLLPRI;
6052 static const struct file_operations md_seq_fops = {
6053 .owner = THIS_MODULE,
6054 .open = md_seq_open,
6056 .llseek = seq_lseek,
6057 .release = seq_release_private,
6058 .poll = mdstat_poll,
6061 int register_md_personality(struct mdk_personality *p)
6063 spin_lock(&pers_lock);
6064 list_add_tail(&p->list, &pers_list);
6065 printk(KERN_INFO "md: %s personality registered for level %d\n", p->name, p->level);
6066 spin_unlock(&pers_lock);
6070 int unregister_md_personality(struct mdk_personality *p)
6072 printk(KERN_INFO "md: %s personality unregistered\n", p->name);
6073 spin_lock(&pers_lock);
6074 list_del_init(&p->list);
6075 spin_unlock(&pers_lock);
6079 static int is_mddev_idle(mddev_t *mddev, int init)
6087 rdev_for_each_rcu(rdev, mddev) {
6088 struct gendisk *disk = rdev->bdev->bd_contains->bd_disk;
6089 curr_events = (int)part_stat_read(&disk->part0, sectors[0]) +
6090 (int)part_stat_read(&disk->part0, sectors[1]) -
6091 atomic_read(&disk->sync_io);
6092 /* sync IO will cause sync_io to increase before the disk_stats
6093 * as sync_io is counted when a request starts, and
6094 * disk_stats is counted when it completes.
6095 * So resync activity will cause curr_events to be smaller than
6096 * when there was no such activity.
6097 * non-sync IO will cause disk_stat to increase without
6098 * increasing sync_io so curr_events will (eventually)
6099 * be larger than it was before. Once it becomes
6100 * substantially larger, the test below will cause
6101 * the array to appear non-idle, and resync will slow
6103 * If there is a lot of outstanding resync activity when
6104 * we set last_event to curr_events, then all that activity
6105 * completing might cause the array to appear non-idle
6106 * and resync will be slowed down even though there might
6107 * not have been non-resync activity. This will only
6108 * happen once though. 'last_events' will soon reflect
6109 * the state where there is little or no outstanding
6110 * resync requests, and further resync activity will
6111 * always make curr_events less than last_events.
6114 if (init || curr_events - rdev->last_events > 64) {
6115 rdev->last_events = curr_events;
6123 void md_done_sync(mddev_t *mddev, int blocks, int ok)
6125 /* another "blocks" (512byte) blocks have been synced */
6126 atomic_sub(blocks, &mddev->recovery_active);
6127 wake_up(&mddev->recovery_wait);
6129 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6130 md_wakeup_thread(mddev->thread);
6131 // stop recovery, signal do_sync ....
6136 /* md_write_start(mddev, bi)
6137 * If we need to update some array metadata (e.g. 'active' flag
6138 * in superblock) before writing, schedule a superblock update
6139 * and wait for it to complete.
6141 void md_write_start(mddev_t *mddev, struct bio *bi)
6144 if (bio_data_dir(bi) != WRITE)
6147 BUG_ON(mddev->ro == 1);
6148 if (mddev->ro == 2) {
6149 /* need to switch to read/write */
6151 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6152 md_wakeup_thread(mddev->thread);
6153 md_wakeup_thread(mddev->sync_thread);
6156 atomic_inc(&mddev->writes_pending);
6157 if (mddev->safemode == 1)
6158 mddev->safemode = 0;
6159 if (mddev->in_sync) {
6160 spin_lock_irq(&mddev->write_lock);
6161 if (mddev->in_sync) {
6163 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
6164 md_wakeup_thread(mddev->thread);
6167 spin_unlock_irq(&mddev->write_lock);
6170 sysfs_notify_dirent(mddev->sysfs_state);
6171 wait_event(mddev->sb_wait,
6172 !test_bit(MD_CHANGE_CLEAN, &mddev->flags) &&
6173 !test_bit(MD_CHANGE_PENDING, &mddev->flags));
6176 void md_write_end(mddev_t *mddev)
6178 if (atomic_dec_and_test(&mddev->writes_pending)) {
6179 if (mddev->safemode == 2)
6180 md_wakeup_thread(mddev->thread);
6181 else if (mddev->safemode_delay)
6182 mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay);
6186 /* md_allow_write(mddev)
6187 * Calling this ensures that the array is marked 'active' so that writes
6188 * may proceed without blocking. It is important to call this before
6189 * attempting a GFP_KERNEL allocation while holding the mddev lock.
6190 * Must be called with mddev_lock held.
6192 * In the ->external case MD_CHANGE_CLEAN can not be cleared until mddev->lock
6193 * is dropped, so return -EAGAIN after notifying userspace.
6195 int md_allow_write(mddev_t *mddev)
6201 if (!mddev->pers->sync_request)
6204 spin_lock_irq(&mddev->write_lock);
6205 if (mddev->in_sync) {
6207 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
6208 if (mddev->safemode_delay &&
6209 mddev->safemode == 0)
6210 mddev->safemode = 1;
6211 spin_unlock_irq(&mddev->write_lock);
6212 md_update_sb(mddev, 0);
6213 sysfs_notify_dirent(mddev->sysfs_state);
6215 spin_unlock_irq(&mddev->write_lock);
6217 if (test_bit(MD_CHANGE_CLEAN, &mddev->flags))
6222 EXPORT_SYMBOL_GPL(md_allow_write);
6224 #define SYNC_MARKS 10
6225 #define SYNC_MARK_STEP (3*HZ)
6226 void md_do_sync(mddev_t *mddev)
6229 unsigned int currspeed = 0,
6231 sector_t max_sectors,j, io_sectors;
6232 unsigned long mark[SYNC_MARKS];
6233 sector_t mark_cnt[SYNC_MARKS];
6235 struct list_head *tmp;
6236 sector_t last_check;
6241 /* just incase thread restarts... */
6242 if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
6244 if (mddev->ro) /* never try to sync a read-only array */
6247 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
6248 if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
6249 desc = "data-check";
6250 else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
6251 desc = "requested-resync";
6254 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
6259 /* we overload curr_resync somewhat here.
6260 * 0 == not engaged in resync at all
6261 * 2 == checking that there is no conflict with another sync
6262 * 1 == like 2, but have yielded to allow conflicting resync to
6264 * other == active in resync - this many blocks
6266 * Before starting a resync we must have set curr_resync to
6267 * 2, and then checked that every "conflicting" array has curr_resync
6268 * less than ours. When we find one that is the same or higher
6269 * we wait on resync_wait. To avoid deadlock, we reduce curr_resync
6270 * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
6271 * This will mean we have to start checking from the beginning again.
6276 mddev->curr_resync = 2;
6279 if (kthread_should_stop()) {
6280 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6283 for_each_mddev(mddev2, tmp) {
6284 if (mddev2 == mddev)
6286 if (!mddev->parallel_resync
6287 && mddev2->curr_resync
6288 && match_mddev_units(mddev, mddev2)) {
6290 if (mddev < mddev2 && mddev->curr_resync == 2) {
6291 /* arbitrarily yield */
6292 mddev->curr_resync = 1;
6293 wake_up(&resync_wait);
6295 if (mddev > mddev2 && mddev->curr_resync == 1)
6296 /* no need to wait here, we can wait the next
6297 * time 'round when curr_resync == 2
6300 /* We need to wait 'interruptible' so as not to
6301 * contribute to the load average, and not to
6302 * be caught by 'softlockup'
6304 prepare_to_wait(&resync_wait, &wq, TASK_INTERRUPTIBLE);
6305 if (!kthread_should_stop() &&
6306 mddev2->curr_resync >= mddev->curr_resync) {
6307 printk(KERN_INFO "md: delaying %s of %s"
6308 " until %s has finished (they"
6309 " share one or more physical units)\n",
6310 desc, mdname(mddev), mdname(mddev2));
6312 if (signal_pending(current))
6313 flush_signals(current);
6315 finish_wait(&resync_wait, &wq);
6318 finish_wait(&resync_wait, &wq);
6321 } while (mddev->curr_resync < 2);
6324 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
6325 /* resync follows the size requested by the personality,
6326 * which defaults to physical size, but can be virtual size
6328 max_sectors = mddev->resync_max_sectors;
6329 mddev->resync_mismatches = 0;
6330 /* we don't use the checkpoint if there's a bitmap */
6331 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
6332 j = mddev->resync_min;
6333 else if (!mddev->bitmap)
6334 j = mddev->recovery_cp;
6336 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
6337 max_sectors = mddev->dev_sectors;
6339 /* recovery follows the physical size of devices */
6340 max_sectors = mddev->dev_sectors;
6342 list_for_each_entry(rdev, &mddev->disks, same_set)
6343 if (rdev->raid_disk >= 0 &&
6344 !test_bit(Faulty, &rdev->flags) &&
6345 !test_bit(In_sync, &rdev->flags) &&
6346 rdev->recovery_offset < j)
6347 j = rdev->recovery_offset;
6350 printk(KERN_INFO "md: %s of RAID array %s\n", desc, mdname(mddev));
6351 printk(KERN_INFO "md: minimum _guaranteed_ speed:"
6352 " %d KB/sec/disk.\n", speed_min(mddev));
6353 printk(KERN_INFO "md: using maximum available idle IO bandwidth "
6354 "(but not more than %d KB/sec) for %s.\n",
6355 speed_max(mddev), desc);
6357 is_mddev_idle(mddev, 1); /* this initializes IO event counters */
6360 for (m = 0; m < SYNC_MARKS; m++) {
6362 mark_cnt[m] = io_sectors;
6365 mddev->resync_mark = mark[last_mark];
6366 mddev->resync_mark_cnt = mark_cnt[last_mark];
6369 * Tune reconstruction:
6371 window = 32*(PAGE_SIZE/512);
6372 printk(KERN_INFO "md: using %dk window, over a total of %llu blocks.\n",
6373 window/2,(unsigned long long) max_sectors/2);
6375 atomic_set(&mddev->recovery_active, 0);
6380 "md: resuming %s of %s from checkpoint.\n",
6381 desc, mdname(mddev));
6382 mddev->curr_resync = j;
6385 while (j < max_sectors) {
6390 if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
6391 ((mddev->curr_resync > mddev->curr_resync_completed &&
6392 (mddev->curr_resync - mddev->curr_resync_completed)
6393 > (max_sectors >> 4)) ||
6394 (j - mddev->curr_resync_completed)*2
6395 >= mddev->resync_max - mddev->curr_resync_completed
6397 /* time to update curr_resync_completed */
6398 blk_unplug(mddev->queue);
6399 wait_event(mddev->recovery_wait,
6400 atomic_read(&mddev->recovery_active) == 0);
6401 mddev->curr_resync_completed =
6403 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
6404 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
6407 while (j >= mddev->resync_max && !kthread_should_stop()) {
6408 /* As this condition is controlled by user-space,
6409 * we can block indefinitely, so use '_interruptible'
6410 * to avoid triggering warnings.
6412 flush_signals(current); /* just in case */
6413 wait_event_interruptible(mddev->recovery_wait,
6414 mddev->resync_max > j
6415 || kthread_should_stop());
6418 if (kthread_should_stop())
6421 sectors = mddev->pers->sync_request(mddev, j, &skipped,
6422 currspeed < speed_min(mddev));
6424 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6428 if (!skipped) { /* actual IO requested */
6429 io_sectors += sectors;
6430 atomic_add(sectors, &mddev->recovery_active);
6434 if (j>1) mddev->curr_resync = j;
6435 mddev->curr_mark_cnt = io_sectors;
6436 if (last_check == 0)
6437 /* this is the earliers that rebuilt will be
6438 * visible in /proc/mdstat
6440 md_new_event(mddev);
6442 if (last_check + window > io_sectors || j == max_sectors)
6445 last_check = io_sectors;
6447 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
6451 if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) {
6453 int next = (last_mark+1) % SYNC_MARKS;
6455 mddev->resync_mark = mark[next];
6456 mddev->resync_mark_cnt = mark_cnt[next];
6457 mark[next] = jiffies;
6458 mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active);
6463 if (kthread_should_stop())
6468 * this loop exits only if either when we are slower than
6469 * the 'hard' speed limit, or the system was IO-idle for
6471 * the system might be non-idle CPU-wise, but we only care
6472 * about not overloading the IO subsystem. (things like an
6473 * e2fsck being done on the RAID array should execute fast)
6475 blk_unplug(mddev->queue);
6478 currspeed = ((unsigned long)(io_sectors-mddev->resync_mark_cnt))/2
6479 /((jiffies-mddev->resync_mark)/HZ +1) +1;
6481 if (currspeed > speed_min(mddev)) {
6482 if ((currspeed > speed_max(mddev)) ||
6483 !is_mddev_idle(mddev, 0)) {
6489 printk(KERN_INFO "md: %s: %s done.\n",mdname(mddev), desc);
6491 * this also signals 'finished resyncing' to md_stop
6494 blk_unplug(mddev->queue);
6496 wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
6498 /* tell personality that we are finished */
6499 mddev->pers->sync_request(mddev, max_sectors, &skipped, 1);
6501 if (!test_bit(MD_RECOVERY_CHECK, &mddev->recovery) &&
6502 mddev->curr_resync > 2) {
6503 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
6504 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
6505 if (mddev->curr_resync >= mddev->recovery_cp) {
6507 "md: checkpointing %s of %s.\n",
6508 desc, mdname(mddev));
6509 mddev->recovery_cp = mddev->curr_resync;
6512 mddev->recovery_cp = MaxSector;
6514 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery))
6515 mddev->curr_resync = MaxSector;
6516 list_for_each_entry(rdev, &mddev->disks, same_set)
6517 if (rdev->raid_disk >= 0 &&
6518 !test_bit(Faulty, &rdev->flags) &&
6519 !test_bit(In_sync, &rdev->flags) &&
6520 rdev->recovery_offset < mddev->curr_resync)
6521 rdev->recovery_offset = mddev->curr_resync;
6524 set_bit(MD_CHANGE_DEVS, &mddev->flags);
6527 mddev->curr_resync = 0;
6528 mddev->curr_resync_completed = 0;
6529 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery))
6530 /* We completed so max setting can be forgotten. */
6531 mddev->resync_max = MaxSector;
6532 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
6533 wake_up(&resync_wait);
6534 set_bit(MD_RECOVERY_DONE, &mddev->recovery);
6535 md_wakeup_thread(mddev->thread);
6540 * got a signal, exit.
6543 "md: md_do_sync() got signal ... exiting\n");
6544 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6548 EXPORT_SYMBOL_GPL(md_do_sync);
6551 static int remove_and_add_spares(mddev_t *mddev)
6556 mddev->curr_resync_completed = 0;
6558 list_for_each_entry(rdev, &mddev->disks, same_set)
6559 if (rdev->raid_disk >= 0 &&
6560 !test_bit(Blocked, &rdev->flags) &&
6561 (test_bit(Faulty, &rdev->flags) ||
6562 ! test_bit(In_sync, &rdev->flags)) &&
6563 atomic_read(&rdev->nr_pending)==0) {
6564 if (mddev->pers->hot_remove_disk(
6565 mddev, rdev->raid_disk)==0) {
6567 sprintf(nm,"rd%d", rdev->raid_disk);
6568 sysfs_remove_link(&mddev->kobj, nm);
6569 rdev->raid_disk = -1;
6573 if (mddev->degraded && ! mddev->ro && !mddev->recovery_disabled) {
6574 list_for_each_entry(rdev, &mddev->disks, same_set) {
6575 if (rdev->raid_disk >= 0 &&
6576 !test_bit(In_sync, &rdev->flags) &&
6577 !test_bit(Blocked, &rdev->flags))
6579 if (rdev->raid_disk < 0
6580 && !test_bit(Faulty, &rdev->flags)) {
6581 rdev->recovery_offset = 0;
6583 hot_add_disk(mddev, rdev) == 0) {
6585 sprintf(nm, "rd%d", rdev->raid_disk);
6586 if (sysfs_create_link(&mddev->kobj,
6589 "md: cannot register "
6593 md_new_event(mddev);
6602 * This routine is regularly called by all per-raid-array threads to
6603 * deal with generic issues like resync and super-block update.
6604 * Raid personalities that don't have a thread (linear/raid0) do not
6605 * need this as they never do any recovery or update the superblock.
6607 * It does not do any resync itself, but rather "forks" off other threads
6608 * to do that as needed.
6609 * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
6610 * "->recovery" and create a thread at ->sync_thread.
6611 * When the thread finishes it sets MD_RECOVERY_DONE
6612 * and wakeups up this thread which will reap the thread and finish up.
6613 * This thread also removes any faulty devices (with nr_pending == 0).
6615 * The overall approach is:
6616 * 1/ if the superblock needs updating, update it.
6617 * 2/ If a recovery thread is running, don't do anything else.
6618 * 3/ If recovery has finished, clean up, possibly marking spares active.
6619 * 4/ If there are any faulty devices, remove them.
6620 * 5/ If array is degraded, try to add spares devices
6621 * 6/ If array has spares or is not in-sync, start a resync thread.
6623 void md_check_recovery(mddev_t *mddev)
6629 bitmap_daemon_work(mddev);
6634 if (signal_pending(current)) {
6635 if (mddev->pers->sync_request && !mddev->external) {
6636 printk(KERN_INFO "md: %s in immediate safe mode\n",
6638 mddev->safemode = 2;
6640 flush_signals(current);
6643 if (mddev->ro && !test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
6646 (mddev->flags && !mddev->external) ||
6647 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
6648 test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
6649 (mddev->external == 0 && mddev->safemode == 1) ||
6650 (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending)
6651 && !mddev->in_sync && mddev->recovery_cp == MaxSector)
6655 if (mddev_trylock(mddev)) {
6659 /* Only thing we do on a ro array is remove
6662 remove_and_add_spares(mddev);
6663 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6667 if (!mddev->external) {
6669 spin_lock_irq(&mddev->write_lock);
6670 if (mddev->safemode &&
6671 !atomic_read(&mddev->writes_pending) &&
6673 mddev->recovery_cp == MaxSector) {
6676 if (mddev->persistent)
6677 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
6679 if (mddev->safemode == 1)
6680 mddev->safemode = 0;
6681 spin_unlock_irq(&mddev->write_lock);
6683 sysfs_notify_dirent(mddev->sysfs_state);
6687 md_update_sb(mddev, 0);
6689 list_for_each_entry(rdev, &mddev->disks, same_set)
6690 if (test_and_clear_bit(StateChanged, &rdev->flags))
6691 sysfs_notify_dirent(rdev->sysfs_state);
6694 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
6695 !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
6696 /* resync/recovery still happening */
6697 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6700 if (mddev->sync_thread) {
6701 /* resync has finished, collect result */
6702 md_unregister_thread(mddev->sync_thread);
6703 mddev->sync_thread = NULL;
6704 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
6705 !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
6707 /* activate any spares */
6708 if (mddev->pers->spare_active(mddev))
6709 sysfs_notify(&mddev->kobj, NULL,
6712 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
6713 mddev->pers->finish_reshape)
6714 mddev->pers->finish_reshape(mddev);
6715 md_update_sb(mddev, 1);
6717 /* if array is no-longer degraded, then any saved_raid_disk
6718 * information must be scrapped
6720 if (!mddev->degraded)
6721 list_for_each_entry(rdev, &mddev->disks, same_set)
6722 rdev->saved_raid_disk = -1;
6724 mddev->recovery = 0;
6725 /* flag recovery needed just to double check */
6726 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6727 sysfs_notify_dirent(mddev->sysfs_action);
6728 md_new_event(mddev);
6731 /* Set RUNNING before clearing NEEDED to avoid
6732 * any transients in the value of "sync_action".
6734 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
6735 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6736 /* Clear some bits that don't mean anything, but
6739 clear_bit(MD_RECOVERY_INTR, &mddev->recovery);
6740 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
6742 if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
6744 /* no recovery is running.
6745 * remove any failed drives, then
6746 * add spares if possible.
6747 * Spare are also removed and re-added, to allow
6748 * the personality to fail the re-add.
6751 if (mddev->reshape_position != MaxSector) {
6752 if (mddev->pers->check_reshape == NULL ||
6753 mddev->pers->check_reshape(mddev) != 0)
6754 /* Cannot proceed */
6756 set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
6757 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
6758 } else if ((spares = remove_and_add_spares(mddev))) {
6759 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
6760 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
6761 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
6762 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
6763 } else if (mddev->recovery_cp < MaxSector) {
6764 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
6765 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
6766 } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
6767 /* nothing to be done ... */
6770 if (mddev->pers->sync_request) {
6771 if (spares && mddev->bitmap && ! mddev->bitmap->file) {
6772 /* We are adding a device or devices to an array
6773 * which has the bitmap stored on all devices.
6774 * So make sure all bitmap pages get written
6776 bitmap_write_all(mddev->bitmap);
6778 mddev->sync_thread = md_register_thread(md_do_sync,
6781 if (!mddev->sync_thread) {
6782 printk(KERN_ERR "%s: could not start resync"
6785 /* leave the spares where they are, it shouldn't hurt */
6786 mddev->recovery = 0;
6788 md_wakeup_thread(mddev->sync_thread);
6789 sysfs_notify_dirent(mddev->sysfs_action);
6790 md_new_event(mddev);
6793 if (!mddev->sync_thread) {
6794 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
6795 if (test_and_clear_bit(MD_RECOVERY_RECOVER,
6797 if (mddev->sysfs_action)
6798 sysfs_notify_dirent(mddev->sysfs_action);
6800 mddev_unlock(mddev);
6804 void md_wait_for_blocked_rdev(mdk_rdev_t *rdev, mddev_t *mddev)
6806 sysfs_notify_dirent(rdev->sysfs_state);
6807 wait_event_timeout(rdev->blocked_wait,
6808 !test_bit(Blocked, &rdev->flags),
6809 msecs_to_jiffies(5000));
6810 rdev_dec_pending(rdev, mddev);
6812 EXPORT_SYMBOL(md_wait_for_blocked_rdev);
6814 static int md_notify_reboot(struct notifier_block *this,
6815 unsigned long code, void *x)
6817 struct list_head *tmp;
6820 if ((code == SYS_DOWN) || (code == SYS_HALT) || (code == SYS_POWER_OFF)) {
6822 printk(KERN_INFO "md: stopping all md devices.\n");
6824 for_each_mddev(mddev, tmp)
6825 if (mddev_trylock(mddev)) {
6826 /* Force a switch to readonly even array
6827 * appears to still be in use. Hence
6830 do_md_stop(mddev, 1, 100);
6831 mddev_unlock(mddev);
6834 * certain more exotic SCSI devices are known to be
6835 * volatile wrt too early system reboots. While the
6836 * right place to handle this issue is the given
6837 * driver, we do want to have a safe RAID driver ...
6844 static struct notifier_block md_notifier = {
6845 .notifier_call = md_notify_reboot,
6847 .priority = INT_MAX, /* before any real devices */
6850 static void md_geninit(void)
6852 dprintk("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
6854 proc_create("mdstat", S_IRUGO, NULL, &md_seq_fops);
6857 static int __init md_init(void)
6859 if (register_blkdev(MD_MAJOR, "md"))
6861 if ((mdp_major=register_blkdev(0, "mdp"))<=0) {
6862 unregister_blkdev(MD_MAJOR, "md");
6865 blk_register_region(MKDEV(MD_MAJOR, 0), 1UL<<MINORBITS, THIS_MODULE,
6866 md_probe, NULL, NULL);
6867 blk_register_region(MKDEV(mdp_major, 0), 1UL<<MINORBITS, THIS_MODULE,
6868 md_probe, NULL, NULL);
6870 register_reboot_notifier(&md_notifier);
6871 raid_table_header = register_sysctl_table(raid_root_table);
6881 * Searches all registered partitions for autorun RAID arrays
6885 static LIST_HEAD(all_detected_devices);
6886 struct detected_devices_node {
6887 struct list_head list;
6891 void md_autodetect_dev(dev_t dev)
6893 struct detected_devices_node *node_detected_dev;
6895 node_detected_dev = kzalloc(sizeof(*node_detected_dev), GFP_KERNEL);
6896 if (node_detected_dev) {
6897 node_detected_dev->dev = dev;
6898 list_add_tail(&node_detected_dev->list, &all_detected_devices);
6900 printk(KERN_CRIT "md: md_autodetect_dev: kzalloc failed"
6901 ", skipping dev(%d,%d)\n", MAJOR(dev), MINOR(dev));
6906 static void autostart_arrays(int part)
6909 struct detected_devices_node *node_detected_dev;
6911 int i_scanned, i_passed;
6916 printk(KERN_INFO "md: Autodetecting RAID arrays.\n");
6918 while (!list_empty(&all_detected_devices) && i_scanned < INT_MAX) {
6920 node_detected_dev = list_entry(all_detected_devices.next,
6921 struct detected_devices_node, list);
6922 list_del(&node_detected_dev->list);
6923 dev = node_detected_dev->dev;
6924 kfree(node_detected_dev);
6925 rdev = md_import_device(dev,0, 90);
6929 if (test_bit(Faulty, &rdev->flags)) {
6933 set_bit(AutoDetected, &rdev->flags);
6934 list_add(&rdev->same_set, &pending_raid_disks);
6938 printk(KERN_INFO "md: Scanned %d and added %d devices.\n",
6939 i_scanned, i_passed);
6941 autorun_devices(part);
6944 #endif /* !MODULE */
6946 static __exit void md_exit(void)
6949 struct list_head *tmp;
6951 blk_unregister_region(MKDEV(MD_MAJOR,0), 1U << MINORBITS);
6952 blk_unregister_region(MKDEV(mdp_major,0), 1U << MINORBITS);
6954 unregister_blkdev(MD_MAJOR,"md");
6955 unregister_blkdev(mdp_major, "mdp");
6956 unregister_reboot_notifier(&md_notifier);
6957 unregister_sysctl_table(raid_table_header);
6958 remove_proc_entry("mdstat", NULL);
6959 for_each_mddev(mddev, tmp) {
6960 export_array(mddev);
6961 mddev->hold_active = 0;
6965 subsys_initcall(md_init);
6966 module_exit(md_exit)
6968 static int get_ro(char *buffer, struct kernel_param *kp)
6970 return sprintf(buffer, "%d", start_readonly);
6972 static int set_ro(const char *val, struct kernel_param *kp)
6975 int num = simple_strtoul(val, &e, 10);
6976 if (*val && (*e == '\0' || *e == '\n')) {
6977 start_readonly = num;
6983 module_param_call(start_ro, set_ro, get_ro, NULL, S_IRUSR|S_IWUSR);
6984 module_param(start_dirty_degraded, int, S_IRUGO|S_IWUSR);
6986 module_param_call(new_array, add_named_array, NULL, NULL, S_IWUSR);
6988 EXPORT_SYMBOL(register_md_personality);
6989 EXPORT_SYMBOL(unregister_md_personality);
6990 EXPORT_SYMBOL(md_error);
6991 EXPORT_SYMBOL(md_done_sync);
6992 EXPORT_SYMBOL(md_write_start);
6993 EXPORT_SYMBOL(md_write_end);
6994 EXPORT_SYMBOL(md_register_thread);
6995 EXPORT_SYMBOL(md_unregister_thread);
6996 EXPORT_SYMBOL(md_wakeup_thread);
6997 EXPORT_SYMBOL(md_check_recovery);
6998 MODULE_LICENSE("GPL");
7000 MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR);