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/raid/md.h>
37 #include <linux/sysctl.h>
38 #include <linux/buffer_head.h> /* for invalidate_bdev */
39 #include <linux/poll.h>
40 #include <linux/ctype.h>
41 #include <linux/hdreg.h>
42 #include <linux/proc_fs.h>
43 #include <linux/random.h>
44 #include <linux/reboot.h>
45 #include <linux/file.h>
46 #include <linux/delay.h>
49 /* 63 partitions with the alternate major number (mdp) */
50 #define MdpMinorShift 6
53 #define dprintk(x...) ((void)(DEBUG && printk(x)))
57 static void autostart_arrays(int part);
60 static LIST_HEAD(pers_list);
61 static DEFINE_SPINLOCK(pers_lock);
63 static void md_print_devices(void);
65 static DECLARE_WAIT_QUEUE_HEAD(resync_wait);
67 #define MD_BUG(x...) { printk("md: bug in file %s, line %d\n", __FILE__, __LINE__); md_print_devices(); }
70 * Current RAID-1,4,5 parallel reconstruction 'guaranteed speed limit'
71 * is 1000 KB/sec, so the extra system load does not show up that much.
72 * Increase it if you want to have more _guaranteed_ speed. Note that
73 * the RAID driver will use the maximum available bandwidth if the IO
74 * subsystem is idle. There is also an 'absolute maximum' reconstruction
75 * speed limit - in case reconstruction slows down your system despite
78 * you can change it via /proc/sys/dev/raid/speed_limit_min and _max.
79 * or /sys/block/mdX/md/sync_speed_{min,max}
82 static int sysctl_speed_limit_min = 1000;
83 static int sysctl_speed_limit_max = 200000;
84 static inline int speed_min(mddev_t *mddev)
86 return mddev->sync_speed_min ?
87 mddev->sync_speed_min : sysctl_speed_limit_min;
90 static inline int speed_max(mddev_t *mddev)
92 return mddev->sync_speed_max ?
93 mddev->sync_speed_max : sysctl_speed_limit_max;
96 static struct ctl_table_header *raid_table_header;
98 static ctl_table raid_table[] = {
100 .ctl_name = DEV_RAID_SPEED_LIMIT_MIN,
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 .ctl_name = DEV_RAID_SPEED_LIMIT_MAX,
109 .procname = "speed_limit_max",
110 .data = &sysctl_speed_limit_max,
111 .maxlen = sizeof(int),
112 .mode = S_IRUGO|S_IWUSR,
113 .proc_handler = &proc_dointvec,
118 static ctl_table raid_dir_table[] = {
120 .ctl_name = DEV_RAID,
123 .mode = S_IRUGO|S_IXUGO,
129 static ctl_table raid_root_table[] = {
135 .child = raid_dir_table,
140 static struct block_device_operations md_fops;
142 static int start_readonly;
145 * We have a system wide 'event count' that is incremented
146 * on any 'interesting' event, and readers of /proc/mdstat
147 * can use 'poll' or 'select' to find out when the event
151 * start array, stop array, error, add device, remove device,
152 * start build, activate spare
154 static DECLARE_WAIT_QUEUE_HEAD(md_event_waiters);
155 static atomic_t md_event_count;
156 void md_new_event(mddev_t *mddev)
158 atomic_inc(&md_event_count);
159 wake_up(&md_event_waiters);
161 EXPORT_SYMBOL_GPL(md_new_event);
163 /* Alternate version that can be called from interrupts
164 * when calling sysfs_notify isn't needed.
166 static void md_new_event_inintr(mddev_t *mddev)
168 atomic_inc(&md_event_count);
169 wake_up(&md_event_waiters);
173 * Enables to iterate over all existing md arrays
174 * all_mddevs_lock protects this list.
176 static LIST_HEAD(all_mddevs);
177 static DEFINE_SPINLOCK(all_mddevs_lock);
181 * iterates through all used mddevs in the system.
182 * We take care to grab the all_mddevs_lock whenever navigating
183 * the list, and to always hold a refcount when unlocked.
184 * Any code which breaks out of this loop while own
185 * a reference to the current mddev and must mddev_put it.
187 #define for_each_mddev(mddev,tmp) \
189 for (({ spin_lock(&all_mddevs_lock); \
190 tmp = all_mddevs.next; \
192 ({ if (tmp != &all_mddevs) \
193 mddev_get(list_entry(tmp, mddev_t, all_mddevs));\
194 spin_unlock(&all_mddevs_lock); \
195 if (mddev) mddev_put(mddev); \
196 mddev = list_entry(tmp, mddev_t, all_mddevs); \
197 tmp != &all_mddevs;}); \
198 ({ spin_lock(&all_mddevs_lock); \
203 static int md_fail_request(struct request_queue *q, struct bio *bio)
209 static inline mddev_t *mddev_get(mddev_t *mddev)
211 atomic_inc(&mddev->active);
215 static void mddev_delayed_delete(struct work_struct *ws)
217 mddev_t *mddev = container_of(ws, mddev_t, del_work);
218 kobject_del(&mddev->kobj);
219 kobject_put(&mddev->kobj);
222 static void mddev_put(mddev_t *mddev)
224 if (!atomic_dec_and_lock(&mddev->active, &all_mddevs_lock))
226 if (!mddev->raid_disks && list_empty(&mddev->disks) &&
227 !mddev->hold_active) {
228 list_del(&mddev->all_mddevs);
229 if (mddev->gendisk) {
230 /* we did a probe so need to clean up.
231 * Call schedule_work inside the spinlock
232 * so that flush_scheduled_work() after
233 * mddev_find will succeed in waiting for the
236 INIT_WORK(&mddev->del_work, mddev_delayed_delete);
237 schedule_work(&mddev->del_work);
241 spin_unlock(&all_mddevs_lock);
244 static mddev_t * mddev_find(dev_t unit)
246 mddev_t *mddev, *new = NULL;
249 spin_lock(&all_mddevs_lock);
252 list_for_each_entry(mddev, &all_mddevs, all_mddevs)
253 if (mddev->unit == unit) {
255 spin_unlock(&all_mddevs_lock);
261 list_add(&new->all_mddevs, &all_mddevs);
262 spin_unlock(&all_mddevs_lock);
263 new->hold_active = UNTIL_IOCTL;
267 /* find an unused unit number */
268 static int next_minor = 512;
269 int start = next_minor;
273 dev = MKDEV(MD_MAJOR, next_minor);
275 if (next_minor > MINORMASK)
277 if (next_minor == start) {
278 /* Oh dear, all in use. */
279 spin_unlock(&all_mddevs_lock);
285 list_for_each_entry(mddev, &all_mddevs, all_mddevs)
286 if (mddev->unit == dev) {
292 new->md_minor = MINOR(dev);
293 new->hold_active = UNTIL_STOP;
294 list_add(&new->all_mddevs, &all_mddevs);
295 spin_unlock(&all_mddevs_lock);
298 spin_unlock(&all_mddevs_lock);
300 new = kzalloc(sizeof(*new), GFP_KERNEL);
305 if (MAJOR(unit) == MD_MAJOR)
306 new->md_minor = MINOR(unit);
308 new->md_minor = MINOR(unit) >> MdpMinorShift;
310 mutex_init(&new->reconfig_mutex);
311 INIT_LIST_HEAD(&new->disks);
312 INIT_LIST_HEAD(&new->all_mddevs);
313 init_timer(&new->safemode_timer);
314 atomic_set(&new->active, 1);
315 atomic_set(&new->openers, 0);
316 spin_lock_init(&new->write_lock);
317 init_waitqueue_head(&new->sb_wait);
318 init_waitqueue_head(&new->recovery_wait);
319 new->reshape_position = MaxSector;
321 new->resync_max = MaxSector;
322 new->level = LEVEL_NONE;
327 static inline int mddev_lock(mddev_t * mddev)
329 return mutex_lock_interruptible(&mddev->reconfig_mutex);
332 static inline int mddev_trylock(mddev_t * mddev)
334 return mutex_trylock(&mddev->reconfig_mutex);
337 static inline void mddev_unlock(mddev_t * mddev)
339 mutex_unlock(&mddev->reconfig_mutex);
341 md_wakeup_thread(mddev->thread);
344 static mdk_rdev_t * find_rdev_nr(mddev_t *mddev, int nr)
348 list_for_each_entry(rdev, &mddev->disks, same_set)
349 if (rdev->desc_nr == nr)
355 static mdk_rdev_t * find_rdev(mddev_t * mddev, dev_t dev)
359 list_for_each_entry(rdev, &mddev->disks, same_set)
360 if (rdev->bdev->bd_dev == dev)
366 static struct mdk_personality *find_pers(int level, char *clevel)
368 struct mdk_personality *pers;
369 list_for_each_entry(pers, &pers_list, list) {
370 if (level != LEVEL_NONE && pers->level == level)
372 if (strcmp(pers->name, clevel)==0)
378 /* return the offset of the super block in 512byte sectors */
379 static inline sector_t calc_dev_sboffset(struct block_device *bdev)
381 sector_t num_sectors = bdev->bd_inode->i_size / 512;
382 return MD_NEW_SIZE_SECTORS(num_sectors);
385 static sector_t calc_num_sectors(mdk_rdev_t *rdev, unsigned chunk_size)
387 sector_t num_sectors = rdev->sb_start;
390 num_sectors &= ~((sector_t)chunk_size/512 - 1);
394 static int alloc_disk_sb(mdk_rdev_t * rdev)
399 rdev->sb_page = alloc_page(GFP_KERNEL);
400 if (!rdev->sb_page) {
401 printk(KERN_ALERT "md: out of memory.\n");
408 static void free_disk_sb(mdk_rdev_t * rdev)
411 put_page(rdev->sb_page);
413 rdev->sb_page = NULL;
420 static void super_written(struct bio *bio, int error)
422 mdk_rdev_t *rdev = bio->bi_private;
423 mddev_t *mddev = rdev->mddev;
425 if (error || !test_bit(BIO_UPTODATE, &bio->bi_flags)) {
426 printk("md: super_written gets error=%d, uptodate=%d\n",
427 error, test_bit(BIO_UPTODATE, &bio->bi_flags));
428 WARN_ON(test_bit(BIO_UPTODATE, &bio->bi_flags));
429 md_error(mddev, rdev);
432 if (atomic_dec_and_test(&mddev->pending_writes))
433 wake_up(&mddev->sb_wait);
437 static void super_written_barrier(struct bio *bio, int error)
439 struct bio *bio2 = bio->bi_private;
440 mdk_rdev_t *rdev = bio2->bi_private;
441 mddev_t *mddev = rdev->mddev;
443 if (!test_bit(BIO_UPTODATE, &bio->bi_flags) &&
444 error == -EOPNOTSUPP) {
446 /* barriers don't appear to be supported :-( */
447 set_bit(BarriersNotsupp, &rdev->flags);
448 mddev->barriers_work = 0;
449 spin_lock_irqsave(&mddev->write_lock, flags);
450 bio2->bi_next = mddev->biolist;
451 mddev->biolist = bio2;
452 spin_unlock_irqrestore(&mddev->write_lock, flags);
453 wake_up(&mddev->sb_wait);
457 bio->bi_private = rdev;
458 super_written(bio, error);
462 void md_super_write(mddev_t *mddev, mdk_rdev_t *rdev,
463 sector_t sector, int size, struct page *page)
465 /* write first size bytes of page to sector of rdev
466 * Increment mddev->pending_writes before returning
467 * and decrement it on completion, waking up sb_wait
468 * if zero is reached.
469 * If an error occurred, call md_error
471 * As we might need to resubmit the request if BIO_RW_BARRIER
472 * causes ENOTSUPP, we allocate a spare bio...
474 struct bio *bio = bio_alloc(GFP_NOIO, 1);
475 int rw = (1<<BIO_RW) | (1<<BIO_RW_SYNCIO) | (1<<BIO_RW_UNPLUG);
477 bio->bi_bdev = rdev->bdev;
478 bio->bi_sector = sector;
479 bio_add_page(bio, page, size, 0);
480 bio->bi_private = rdev;
481 bio->bi_end_io = super_written;
484 atomic_inc(&mddev->pending_writes);
485 if (!test_bit(BarriersNotsupp, &rdev->flags)) {
487 rw |= (1<<BIO_RW_BARRIER);
488 rbio = bio_clone(bio, GFP_NOIO);
489 rbio->bi_private = bio;
490 rbio->bi_end_io = super_written_barrier;
491 submit_bio(rw, rbio);
496 void md_super_wait(mddev_t *mddev)
498 /* wait for all superblock writes that were scheduled to complete.
499 * if any had to be retried (due to BARRIER problems), retry them
503 prepare_to_wait(&mddev->sb_wait, &wq, TASK_UNINTERRUPTIBLE);
504 if (atomic_read(&mddev->pending_writes)==0)
506 while (mddev->biolist) {
508 spin_lock_irq(&mddev->write_lock);
509 bio = mddev->biolist;
510 mddev->biolist = bio->bi_next ;
512 spin_unlock_irq(&mddev->write_lock);
513 submit_bio(bio->bi_rw, bio);
517 finish_wait(&mddev->sb_wait, &wq);
520 static void bi_complete(struct bio *bio, int error)
522 complete((struct completion*)bio->bi_private);
525 int sync_page_io(struct block_device *bdev, sector_t sector, int size,
526 struct page *page, int rw)
528 struct bio *bio = bio_alloc(GFP_NOIO, 1);
529 struct completion event;
532 rw |= (1 << BIO_RW_SYNCIO) | (1 << BIO_RW_UNPLUG);
535 bio->bi_sector = sector;
536 bio_add_page(bio, page, size, 0);
537 init_completion(&event);
538 bio->bi_private = &event;
539 bio->bi_end_io = bi_complete;
541 wait_for_completion(&event);
543 ret = test_bit(BIO_UPTODATE, &bio->bi_flags);
547 EXPORT_SYMBOL_GPL(sync_page_io);
549 static int read_disk_sb(mdk_rdev_t * rdev, int size)
551 char b[BDEVNAME_SIZE];
552 if (!rdev->sb_page) {
560 if (!sync_page_io(rdev->bdev, rdev->sb_start, size, rdev->sb_page, READ))
566 printk(KERN_WARNING "md: disabled device %s, could not read superblock.\n",
567 bdevname(rdev->bdev,b));
571 static int uuid_equal(mdp_super_t *sb1, mdp_super_t *sb2)
573 return sb1->set_uuid0 == sb2->set_uuid0 &&
574 sb1->set_uuid1 == sb2->set_uuid1 &&
575 sb1->set_uuid2 == sb2->set_uuid2 &&
576 sb1->set_uuid3 == sb2->set_uuid3;
579 static int sb_equal(mdp_super_t *sb1, mdp_super_t *sb2)
582 mdp_super_t *tmp1, *tmp2;
584 tmp1 = kmalloc(sizeof(*tmp1),GFP_KERNEL);
585 tmp2 = kmalloc(sizeof(*tmp2),GFP_KERNEL);
587 if (!tmp1 || !tmp2) {
589 printk(KERN_INFO "md.c sb_equal(): failed to allocate memory!\n");
597 * nr_disks is not constant
602 ret = (memcmp(tmp1, tmp2, MD_SB_GENERIC_CONSTANT_WORDS * 4) == 0);
610 static u32 md_csum_fold(u32 csum)
612 csum = (csum & 0xffff) + (csum >> 16);
613 return (csum & 0xffff) + (csum >> 16);
616 static unsigned int calc_sb_csum(mdp_super_t * sb)
619 u32 *sb32 = (u32*)sb;
621 unsigned int disk_csum, csum;
623 disk_csum = sb->sb_csum;
626 for (i = 0; i < MD_SB_BYTES/4 ; i++)
628 csum = (newcsum & 0xffffffff) + (newcsum>>32);
632 /* This used to use csum_partial, which was wrong for several
633 * reasons including that different results are returned on
634 * different architectures. It isn't critical that we get exactly
635 * the same return value as before (we always csum_fold before
636 * testing, and that removes any differences). However as we
637 * know that csum_partial always returned a 16bit value on
638 * alphas, do a fold to maximise conformity to previous behaviour.
640 sb->sb_csum = md_csum_fold(disk_csum);
642 sb->sb_csum = disk_csum;
649 * Handle superblock details.
650 * We want to be able to handle multiple superblock formats
651 * so we have a common interface to them all, and an array of
652 * different handlers.
653 * We rely on user-space to write the initial superblock, and support
654 * reading and updating of superblocks.
655 * Interface methods are:
656 * int load_super(mdk_rdev_t *dev, mdk_rdev_t *refdev, int minor_version)
657 * loads and validates a superblock on dev.
658 * if refdev != NULL, compare superblocks on both devices
660 * 0 - dev has a superblock that is compatible with refdev
661 * 1 - dev has a superblock that is compatible and newer than refdev
662 * so dev should be used as the refdev in future
663 * -EINVAL superblock incompatible or invalid
664 * -othererror e.g. -EIO
666 * int validate_super(mddev_t *mddev, mdk_rdev_t *dev)
667 * Verify that dev is acceptable into mddev.
668 * The first time, mddev->raid_disks will be 0, and data from
669 * dev should be merged in. Subsequent calls check that dev
670 * is new enough. Return 0 or -EINVAL
672 * void sync_super(mddev_t *mddev, mdk_rdev_t *dev)
673 * Update the superblock for rdev with data in mddev
674 * This does not write to disc.
680 struct module *owner;
681 int (*load_super)(mdk_rdev_t *rdev, mdk_rdev_t *refdev,
683 int (*validate_super)(mddev_t *mddev, mdk_rdev_t *rdev);
684 void (*sync_super)(mddev_t *mddev, mdk_rdev_t *rdev);
685 unsigned long long (*rdev_size_change)(mdk_rdev_t *rdev,
686 sector_t num_sectors);
690 * load_super for 0.90.0
692 static int super_90_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
694 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
699 * Calculate the position of the superblock (512byte sectors),
700 * it's at the end of the disk.
702 * It also happens to be a multiple of 4Kb.
704 rdev->sb_start = calc_dev_sboffset(rdev->bdev);
706 ret = read_disk_sb(rdev, MD_SB_BYTES);
711 bdevname(rdev->bdev, b);
712 sb = (mdp_super_t*)page_address(rdev->sb_page);
714 if (sb->md_magic != MD_SB_MAGIC) {
715 printk(KERN_ERR "md: invalid raid superblock magic on %s\n",
720 if (sb->major_version != 0 ||
721 sb->minor_version < 90 ||
722 sb->minor_version > 91) {
723 printk(KERN_WARNING "Bad version number %d.%d on %s\n",
724 sb->major_version, sb->minor_version,
729 if (sb->raid_disks <= 0)
732 if (md_csum_fold(calc_sb_csum(sb)) != md_csum_fold(sb->sb_csum)) {
733 printk(KERN_WARNING "md: invalid superblock checksum on %s\n",
738 rdev->preferred_minor = sb->md_minor;
739 rdev->data_offset = 0;
740 rdev->sb_size = MD_SB_BYTES;
742 if (sb->state & (1<<MD_SB_BITMAP_PRESENT)) {
743 if (sb->level != 1 && sb->level != 4
744 && sb->level != 5 && sb->level != 6
745 && sb->level != 10) {
746 /* FIXME use a better test */
748 "md: bitmaps not supported for this level.\n");
753 if (sb->level == LEVEL_MULTIPATH)
756 rdev->desc_nr = sb->this_disk.number;
762 mdp_super_t *refsb = (mdp_super_t*)page_address(refdev->sb_page);
763 if (!uuid_equal(refsb, sb)) {
764 printk(KERN_WARNING "md: %s has different UUID to %s\n",
765 b, bdevname(refdev->bdev,b2));
768 if (!sb_equal(refsb, sb)) {
769 printk(KERN_WARNING "md: %s has same UUID"
770 " but different superblock to %s\n",
771 b, bdevname(refdev->bdev, b2));
775 ev2 = md_event(refsb);
781 rdev->size = calc_num_sectors(rdev, sb->chunk_size) / 2;
783 if (rdev->size < sb->size && sb->level > 1)
784 /* "this cannot possibly happen" ... */
792 * validate_super for 0.90.0
794 static int super_90_validate(mddev_t *mddev, mdk_rdev_t *rdev)
797 mdp_super_t *sb = (mdp_super_t *)page_address(rdev->sb_page);
798 __u64 ev1 = md_event(sb);
800 rdev->raid_disk = -1;
801 clear_bit(Faulty, &rdev->flags);
802 clear_bit(In_sync, &rdev->flags);
803 clear_bit(WriteMostly, &rdev->flags);
804 clear_bit(BarriersNotsupp, &rdev->flags);
806 if (mddev->raid_disks == 0) {
807 mddev->major_version = 0;
808 mddev->minor_version = sb->minor_version;
809 mddev->patch_version = sb->patch_version;
811 mddev->chunk_size = sb->chunk_size;
812 mddev->ctime = sb->ctime;
813 mddev->utime = sb->utime;
814 mddev->level = sb->level;
815 mddev->clevel[0] = 0;
816 mddev->layout = sb->layout;
817 mddev->raid_disks = sb->raid_disks;
818 mddev->size = sb->size;
820 mddev->bitmap_offset = 0;
821 mddev->default_bitmap_offset = MD_SB_BYTES >> 9;
823 if (mddev->minor_version >= 91) {
824 mddev->reshape_position = sb->reshape_position;
825 mddev->delta_disks = sb->delta_disks;
826 mddev->new_level = sb->new_level;
827 mddev->new_layout = sb->new_layout;
828 mddev->new_chunk = sb->new_chunk;
830 mddev->reshape_position = MaxSector;
831 mddev->delta_disks = 0;
832 mddev->new_level = mddev->level;
833 mddev->new_layout = mddev->layout;
834 mddev->new_chunk = mddev->chunk_size;
837 if (sb->state & (1<<MD_SB_CLEAN))
838 mddev->recovery_cp = MaxSector;
840 if (sb->events_hi == sb->cp_events_hi &&
841 sb->events_lo == sb->cp_events_lo) {
842 mddev->recovery_cp = sb->recovery_cp;
844 mddev->recovery_cp = 0;
847 memcpy(mddev->uuid+0, &sb->set_uuid0, 4);
848 memcpy(mddev->uuid+4, &sb->set_uuid1, 4);
849 memcpy(mddev->uuid+8, &sb->set_uuid2, 4);
850 memcpy(mddev->uuid+12,&sb->set_uuid3, 4);
852 mddev->max_disks = MD_SB_DISKS;
854 if (sb->state & (1<<MD_SB_BITMAP_PRESENT) &&
855 mddev->bitmap_file == NULL)
856 mddev->bitmap_offset = mddev->default_bitmap_offset;
858 } else if (mddev->pers == NULL) {
859 /* Insist on good event counter while assembling */
861 if (ev1 < mddev->events)
863 } else if (mddev->bitmap) {
864 /* if adding to array with a bitmap, then we can accept an
865 * older device ... but not too old.
867 if (ev1 < mddev->bitmap->events_cleared)
870 if (ev1 < mddev->events)
871 /* just a hot-add of a new device, leave raid_disk at -1 */
875 if (mddev->level != LEVEL_MULTIPATH) {
876 desc = sb->disks + rdev->desc_nr;
878 if (desc->state & (1<<MD_DISK_FAULTY))
879 set_bit(Faulty, &rdev->flags);
880 else if (desc->state & (1<<MD_DISK_SYNC) /* &&
881 desc->raid_disk < mddev->raid_disks */) {
882 set_bit(In_sync, &rdev->flags);
883 rdev->raid_disk = desc->raid_disk;
885 if (desc->state & (1<<MD_DISK_WRITEMOSTLY))
886 set_bit(WriteMostly, &rdev->flags);
887 } else /* MULTIPATH are always insync */
888 set_bit(In_sync, &rdev->flags);
893 * sync_super for 0.90.0
895 static void super_90_sync(mddev_t *mddev, mdk_rdev_t *rdev)
899 int next_spare = mddev->raid_disks;
902 /* make rdev->sb match mddev data..
905 * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
906 * 3/ any empty disks < next_spare become removed
908 * disks[0] gets initialised to REMOVED because
909 * we cannot be sure from other fields if it has
910 * been initialised or not.
913 int active=0, working=0,failed=0,spare=0,nr_disks=0;
915 rdev->sb_size = MD_SB_BYTES;
917 sb = (mdp_super_t*)page_address(rdev->sb_page);
919 memset(sb, 0, sizeof(*sb));
921 sb->md_magic = MD_SB_MAGIC;
922 sb->major_version = mddev->major_version;
923 sb->patch_version = mddev->patch_version;
924 sb->gvalid_words = 0; /* ignored */
925 memcpy(&sb->set_uuid0, mddev->uuid+0, 4);
926 memcpy(&sb->set_uuid1, mddev->uuid+4, 4);
927 memcpy(&sb->set_uuid2, mddev->uuid+8, 4);
928 memcpy(&sb->set_uuid3, mddev->uuid+12,4);
930 sb->ctime = mddev->ctime;
931 sb->level = mddev->level;
932 sb->size = mddev->size;
933 sb->raid_disks = mddev->raid_disks;
934 sb->md_minor = mddev->md_minor;
935 sb->not_persistent = 0;
936 sb->utime = mddev->utime;
938 sb->events_hi = (mddev->events>>32);
939 sb->events_lo = (u32)mddev->events;
941 if (mddev->reshape_position == MaxSector)
942 sb->minor_version = 90;
944 sb->minor_version = 91;
945 sb->reshape_position = mddev->reshape_position;
946 sb->new_level = mddev->new_level;
947 sb->delta_disks = mddev->delta_disks;
948 sb->new_layout = mddev->new_layout;
949 sb->new_chunk = mddev->new_chunk;
951 mddev->minor_version = sb->minor_version;
954 sb->recovery_cp = mddev->recovery_cp;
955 sb->cp_events_hi = (mddev->events>>32);
956 sb->cp_events_lo = (u32)mddev->events;
957 if (mddev->recovery_cp == MaxSector)
958 sb->state = (1<< MD_SB_CLEAN);
962 sb->layout = mddev->layout;
963 sb->chunk_size = mddev->chunk_size;
965 if (mddev->bitmap && mddev->bitmap_file == NULL)
966 sb->state |= (1<<MD_SB_BITMAP_PRESENT);
968 sb->disks[0].state = (1<<MD_DISK_REMOVED);
969 list_for_each_entry(rdev2, &mddev->disks, same_set) {
972 if (rdev2->raid_disk >= 0 && test_bit(In_sync, &rdev2->flags)
973 && !test_bit(Faulty, &rdev2->flags))
974 desc_nr = rdev2->raid_disk;
976 desc_nr = next_spare++;
977 rdev2->desc_nr = desc_nr;
978 d = &sb->disks[rdev2->desc_nr];
980 d->number = rdev2->desc_nr;
981 d->major = MAJOR(rdev2->bdev->bd_dev);
982 d->minor = MINOR(rdev2->bdev->bd_dev);
983 if (rdev2->raid_disk >= 0 && test_bit(In_sync, &rdev2->flags)
984 && !test_bit(Faulty, &rdev2->flags))
985 d->raid_disk = rdev2->raid_disk;
987 d->raid_disk = rdev2->desc_nr; /* compatibility */
988 if (test_bit(Faulty, &rdev2->flags))
989 d->state = (1<<MD_DISK_FAULTY);
990 else if (test_bit(In_sync, &rdev2->flags)) {
991 d->state = (1<<MD_DISK_ACTIVE);
992 d->state |= (1<<MD_DISK_SYNC);
1000 if (test_bit(WriteMostly, &rdev2->flags))
1001 d->state |= (1<<MD_DISK_WRITEMOSTLY);
1003 /* now set the "removed" and "faulty" bits on any missing devices */
1004 for (i=0 ; i < mddev->raid_disks ; i++) {
1005 mdp_disk_t *d = &sb->disks[i];
1006 if (d->state == 0 && d->number == 0) {
1009 d->state = (1<<MD_DISK_REMOVED);
1010 d->state |= (1<<MD_DISK_FAULTY);
1014 sb->nr_disks = nr_disks;
1015 sb->active_disks = active;
1016 sb->working_disks = working;
1017 sb->failed_disks = failed;
1018 sb->spare_disks = spare;
1020 sb->this_disk = sb->disks[rdev->desc_nr];
1021 sb->sb_csum = calc_sb_csum(sb);
1025 * rdev_size_change for 0.90.0
1027 static unsigned long long
1028 super_90_rdev_size_change(mdk_rdev_t *rdev, sector_t num_sectors)
1030 if (num_sectors && num_sectors < rdev->mddev->size * 2)
1031 return 0; /* component must fit device */
1032 if (rdev->mddev->bitmap_offset)
1033 return 0; /* can't move bitmap */
1034 rdev->sb_start = calc_dev_sboffset(rdev->bdev);
1035 if (!num_sectors || num_sectors > rdev->sb_start)
1036 num_sectors = rdev->sb_start;
1037 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
1039 md_super_wait(rdev->mddev);
1040 return num_sectors / 2; /* kB for sysfs */
1045 * version 1 superblock
1048 static __le32 calc_sb_1_csum(struct mdp_superblock_1 * sb)
1052 unsigned long long newcsum;
1053 int size = 256 + le32_to_cpu(sb->max_dev)*2;
1054 __le32 *isuper = (__le32*)sb;
1057 disk_csum = sb->sb_csum;
1060 for (i=0; size>=4; size -= 4 )
1061 newcsum += le32_to_cpu(*isuper++);
1064 newcsum += le16_to_cpu(*(__le16*) isuper);
1066 csum = (newcsum & 0xffffffff) + (newcsum >> 32);
1067 sb->sb_csum = disk_csum;
1068 return cpu_to_le32(csum);
1071 static int super_1_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
1073 struct mdp_superblock_1 *sb;
1076 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
1080 * Calculate the position of the superblock in 512byte sectors.
1081 * It is always aligned to a 4K boundary and
1082 * depeding on minor_version, it can be:
1083 * 0: At least 8K, but less than 12K, from end of device
1084 * 1: At start of device
1085 * 2: 4K from start of device.
1087 switch(minor_version) {
1089 sb_start = rdev->bdev->bd_inode->i_size >> 9;
1091 sb_start &= ~(sector_t)(4*2-1);
1102 rdev->sb_start = sb_start;
1104 /* superblock is rarely larger than 1K, but it can be larger,
1105 * and it is safe to read 4k, so we do that
1107 ret = read_disk_sb(rdev, 4096);
1108 if (ret) return ret;
1111 sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1113 if (sb->magic != cpu_to_le32(MD_SB_MAGIC) ||
1114 sb->major_version != cpu_to_le32(1) ||
1115 le32_to_cpu(sb->max_dev) > (4096-256)/2 ||
1116 le64_to_cpu(sb->super_offset) != rdev->sb_start ||
1117 (le32_to_cpu(sb->feature_map) & ~MD_FEATURE_ALL) != 0)
1120 if (calc_sb_1_csum(sb) != sb->sb_csum) {
1121 printk("md: invalid superblock checksum on %s\n",
1122 bdevname(rdev->bdev,b));
1125 if (le64_to_cpu(sb->data_size) < 10) {
1126 printk("md: data_size too small on %s\n",
1127 bdevname(rdev->bdev,b));
1130 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET)) {
1131 if (sb->level != cpu_to_le32(1) &&
1132 sb->level != cpu_to_le32(4) &&
1133 sb->level != cpu_to_le32(5) &&
1134 sb->level != cpu_to_le32(6) &&
1135 sb->level != cpu_to_le32(10)) {
1137 "md: bitmaps not supported for this level.\n");
1142 rdev->preferred_minor = 0xffff;
1143 rdev->data_offset = le64_to_cpu(sb->data_offset);
1144 atomic_set(&rdev->corrected_errors, le32_to_cpu(sb->cnt_corrected_read));
1146 rdev->sb_size = le32_to_cpu(sb->max_dev) * 2 + 256;
1147 bmask = queue_hardsect_size(rdev->bdev->bd_disk->queue)-1;
1148 if (rdev->sb_size & bmask)
1149 rdev->sb_size = (rdev->sb_size | bmask) + 1;
1152 && rdev->data_offset < sb_start + (rdev->sb_size/512))
1155 if (sb->level == cpu_to_le32(LEVEL_MULTIPATH))
1158 rdev->desc_nr = le32_to_cpu(sb->dev_number);
1164 struct mdp_superblock_1 *refsb =
1165 (struct mdp_superblock_1*)page_address(refdev->sb_page);
1167 if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 ||
1168 sb->level != refsb->level ||
1169 sb->layout != refsb->layout ||
1170 sb->chunksize != refsb->chunksize) {
1171 printk(KERN_WARNING "md: %s has strangely different"
1172 " superblock to %s\n",
1173 bdevname(rdev->bdev,b),
1174 bdevname(refdev->bdev,b2));
1177 ev1 = le64_to_cpu(sb->events);
1178 ev2 = le64_to_cpu(refsb->events);
1186 rdev->size = ((rdev->bdev->bd_inode->i_size>>9) - le64_to_cpu(sb->data_offset)) / 2;
1188 rdev->size = rdev->sb_start / 2;
1189 if (rdev->size < le64_to_cpu(sb->data_size)/2)
1191 rdev->size = le64_to_cpu(sb->data_size)/2;
1192 if (le32_to_cpu(sb->chunksize))
1193 rdev->size &= ~((sector_t)le32_to_cpu(sb->chunksize)/2 - 1);
1195 if (le64_to_cpu(sb->size) > rdev->size*2)
1200 static int super_1_validate(mddev_t *mddev, mdk_rdev_t *rdev)
1202 struct mdp_superblock_1 *sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1203 __u64 ev1 = le64_to_cpu(sb->events);
1205 rdev->raid_disk = -1;
1206 clear_bit(Faulty, &rdev->flags);
1207 clear_bit(In_sync, &rdev->flags);
1208 clear_bit(WriteMostly, &rdev->flags);
1209 clear_bit(BarriersNotsupp, &rdev->flags);
1211 if (mddev->raid_disks == 0) {
1212 mddev->major_version = 1;
1213 mddev->patch_version = 0;
1214 mddev->external = 0;
1215 mddev->chunk_size = le32_to_cpu(sb->chunksize) << 9;
1216 mddev->ctime = le64_to_cpu(sb->ctime) & ((1ULL << 32)-1);
1217 mddev->utime = le64_to_cpu(sb->utime) & ((1ULL << 32)-1);
1218 mddev->level = le32_to_cpu(sb->level);
1219 mddev->clevel[0] = 0;
1220 mddev->layout = le32_to_cpu(sb->layout);
1221 mddev->raid_disks = le32_to_cpu(sb->raid_disks);
1222 mddev->size = le64_to_cpu(sb->size)/2;
1223 mddev->events = ev1;
1224 mddev->bitmap_offset = 0;
1225 mddev->default_bitmap_offset = 1024 >> 9;
1227 mddev->recovery_cp = le64_to_cpu(sb->resync_offset);
1228 memcpy(mddev->uuid, sb->set_uuid, 16);
1230 mddev->max_disks = (4096-256)/2;
1232 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) &&
1233 mddev->bitmap_file == NULL )
1234 mddev->bitmap_offset = (__s32)le32_to_cpu(sb->bitmap_offset);
1236 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) {
1237 mddev->reshape_position = le64_to_cpu(sb->reshape_position);
1238 mddev->delta_disks = le32_to_cpu(sb->delta_disks);
1239 mddev->new_level = le32_to_cpu(sb->new_level);
1240 mddev->new_layout = le32_to_cpu(sb->new_layout);
1241 mddev->new_chunk = le32_to_cpu(sb->new_chunk)<<9;
1243 mddev->reshape_position = MaxSector;
1244 mddev->delta_disks = 0;
1245 mddev->new_level = mddev->level;
1246 mddev->new_layout = mddev->layout;
1247 mddev->new_chunk = mddev->chunk_size;
1250 } else if (mddev->pers == NULL) {
1251 /* Insist of good event counter while assembling */
1253 if (ev1 < mddev->events)
1255 } else if (mddev->bitmap) {
1256 /* If adding to array with a bitmap, then we can accept an
1257 * older device, but not too old.
1259 if (ev1 < mddev->bitmap->events_cleared)
1262 if (ev1 < mddev->events)
1263 /* just a hot-add of a new device, leave raid_disk at -1 */
1266 if (mddev->level != LEVEL_MULTIPATH) {
1268 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
1270 case 0xffff: /* spare */
1272 case 0xfffe: /* faulty */
1273 set_bit(Faulty, &rdev->flags);
1276 if ((le32_to_cpu(sb->feature_map) &
1277 MD_FEATURE_RECOVERY_OFFSET))
1278 rdev->recovery_offset = le64_to_cpu(sb->recovery_offset);
1280 set_bit(In_sync, &rdev->flags);
1281 rdev->raid_disk = role;
1284 if (sb->devflags & WriteMostly1)
1285 set_bit(WriteMostly, &rdev->flags);
1286 } else /* MULTIPATH are always insync */
1287 set_bit(In_sync, &rdev->flags);
1292 static void super_1_sync(mddev_t *mddev, mdk_rdev_t *rdev)
1294 struct mdp_superblock_1 *sb;
1297 /* make rdev->sb match mddev and rdev data. */
1299 sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1301 sb->feature_map = 0;
1303 sb->recovery_offset = cpu_to_le64(0);
1304 memset(sb->pad1, 0, sizeof(sb->pad1));
1305 memset(sb->pad2, 0, sizeof(sb->pad2));
1306 memset(sb->pad3, 0, sizeof(sb->pad3));
1308 sb->utime = cpu_to_le64((__u64)mddev->utime);
1309 sb->events = cpu_to_le64(mddev->events);
1311 sb->resync_offset = cpu_to_le64(mddev->recovery_cp);
1313 sb->resync_offset = cpu_to_le64(0);
1315 sb->cnt_corrected_read = cpu_to_le32(atomic_read(&rdev->corrected_errors));
1317 sb->raid_disks = cpu_to_le32(mddev->raid_disks);
1318 sb->size = cpu_to_le64(mddev->size<<1);
1320 if (mddev->bitmap && mddev->bitmap_file == NULL) {
1321 sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_offset);
1322 sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET);
1325 if (rdev->raid_disk >= 0 &&
1326 !test_bit(In_sync, &rdev->flags) &&
1327 rdev->recovery_offset > 0) {
1328 sb->feature_map |= cpu_to_le32(MD_FEATURE_RECOVERY_OFFSET);
1329 sb->recovery_offset = cpu_to_le64(rdev->recovery_offset);
1332 if (mddev->reshape_position != MaxSector) {
1333 sb->feature_map |= cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE);
1334 sb->reshape_position = cpu_to_le64(mddev->reshape_position);
1335 sb->new_layout = cpu_to_le32(mddev->new_layout);
1336 sb->delta_disks = cpu_to_le32(mddev->delta_disks);
1337 sb->new_level = cpu_to_le32(mddev->new_level);
1338 sb->new_chunk = cpu_to_le32(mddev->new_chunk>>9);
1342 list_for_each_entry(rdev2, &mddev->disks, same_set)
1343 if (rdev2->desc_nr+1 > max_dev)
1344 max_dev = rdev2->desc_nr+1;
1346 if (max_dev > le32_to_cpu(sb->max_dev))
1347 sb->max_dev = cpu_to_le32(max_dev);
1348 for (i=0; i<max_dev;i++)
1349 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1351 list_for_each_entry(rdev2, &mddev->disks, same_set) {
1353 if (test_bit(Faulty, &rdev2->flags))
1354 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1355 else if (test_bit(In_sync, &rdev2->flags))
1356 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1357 else if (rdev2->raid_disk >= 0 && rdev2->recovery_offset > 0)
1358 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1360 sb->dev_roles[i] = cpu_to_le16(0xffff);
1363 sb->sb_csum = calc_sb_1_csum(sb);
1366 static unsigned long long
1367 super_1_rdev_size_change(mdk_rdev_t *rdev, sector_t num_sectors)
1369 struct mdp_superblock_1 *sb;
1370 sector_t max_sectors;
1371 if (num_sectors && num_sectors < rdev->mddev->size * 2)
1372 return 0; /* component must fit device */
1373 if (rdev->sb_start < rdev->data_offset) {
1374 /* minor versions 1 and 2; superblock before data */
1375 max_sectors = rdev->bdev->bd_inode->i_size >> 9;
1376 max_sectors -= rdev->data_offset;
1377 if (!num_sectors || num_sectors > max_sectors)
1378 num_sectors = max_sectors;
1379 } else if (rdev->mddev->bitmap_offset) {
1380 /* minor version 0 with bitmap we can't move */
1383 /* minor version 0; superblock after data */
1385 sb_start = (rdev->bdev->bd_inode->i_size >> 9) - 8*2;
1386 sb_start &= ~(sector_t)(4*2 - 1);
1387 max_sectors = rdev->size * 2 + sb_start - rdev->sb_start;
1388 if (!num_sectors || num_sectors > max_sectors)
1389 num_sectors = max_sectors;
1390 rdev->sb_start = sb_start;
1392 sb = (struct mdp_superblock_1 *) page_address(rdev->sb_page);
1393 sb->data_size = cpu_to_le64(num_sectors);
1394 sb->super_offset = rdev->sb_start;
1395 sb->sb_csum = calc_sb_1_csum(sb);
1396 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
1398 md_super_wait(rdev->mddev);
1399 return num_sectors / 2; /* kB for sysfs */
1402 static struct super_type super_types[] = {
1405 .owner = THIS_MODULE,
1406 .load_super = super_90_load,
1407 .validate_super = super_90_validate,
1408 .sync_super = super_90_sync,
1409 .rdev_size_change = super_90_rdev_size_change,
1413 .owner = THIS_MODULE,
1414 .load_super = super_1_load,
1415 .validate_super = super_1_validate,
1416 .sync_super = super_1_sync,
1417 .rdev_size_change = super_1_rdev_size_change,
1421 static int match_mddev_units(mddev_t *mddev1, mddev_t *mddev2)
1423 mdk_rdev_t *rdev, *rdev2;
1426 rdev_for_each_rcu(rdev, mddev1)
1427 rdev_for_each_rcu(rdev2, mddev2)
1428 if (rdev->bdev->bd_contains ==
1429 rdev2->bdev->bd_contains) {
1437 static LIST_HEAD(pending_raid_disks);
1439 static void md_integrity_check(mdk_rdev_t *rdev, mddev_t *mddev)
1441 struct mdk_personality *pers = mddev->pers;
1442 struct gendisk *disk = mddev->gendisk;
1443 struct blk_integrity *bi_rdev = bdev_get_integrity(rdev->bdev);
1444 struct blk_integrity *bi_mddev = blk_get_integrity(disk);
1446 /* Data integrity passthrough not supported on RAID 4, 5 and 6 */
1447 if (pers && pers->level >= 4 && pers->level <= 6)
1450 /* If rdev is integrity capable, register profile for mddev */
1451 if (!bi_mddev && bi_rdev) {
1452 if (blk_integrity_register(disk, bi_rdev))
1453 printk(KERN_ERR "%s: %s Could not register integrity!\n",
1454 __func__, disk->disk_name);
1456 printk(KERN_NOTICE "Enabling data integrity on %s\n",
1461 /* Check that mddev and rdev have matching profiles */
1462 if (blk_integrity_compare(disk, rdev->bdev->bd_disk) < 0) {
1463 printk(KERN_ERR "%s: %s/%s integrity mismatch!\n", __func__,
1464 disk->disk_name, rdev->bdev->bd_disk->disk_name);
1465 printk(KERN_NOTICE "Disabling data integrity on %s\n",
1467 blk_integrity_unregister(disk);
1471 static int bind_rdev_to_array(mdk_rdev_t * rdev, mddev_t * mddev)
1473 char b[BDEVNAME_SIZE];
1483 /* prevent duplicates */
1484 if (find_rdev(mddev, rdev->bdev->bd_dev))
1487 /* make sure rdev->size exceeds mddev->size */
1488 if (rdev->size && (mddev->size == 0 || rdev->size < mddev->size)) {
1490 /* Cannot change size, so fail
1491 * If mddev->level <= 0, then we don't care
1492 * about aligning sizes (e.g. linear)
1494 if (mddev->level > 0)
1497 mddev->size = rdev->size;
1500 /* Verify rdev->desc_nr is unique.
1501 * If it is -1, assign a free number, else
1502 * check number is not in use
1504 if (rdev->desc_nr < 0) {
1506 if (mddev->pers) choice = mddev->raid_disks;
1507 while (find_rdev_nr(mddev, choice))
1509 rdev->desc_nr = choice;
1511 if (find_rdev_nr(mddev, rdev->desc_nr))
1514 if (mddev->max_disks && rdev->desc_nr >= mddev->max_disks) {
1515 printk(KERN_WARNING "md: %s: array is limited to %d devices\n",
1516 mdname(mddev), mddev->max_disks);
1519 bdevname(rdev->bdev,b);
1520 while ( (s=strchr(b, '/')) != NULL)
1523 rdev->mddev = mddev;
1524 printk(KERN_INFO "md: bind<%s>\n", b);
1526 if ((err = kobject_add(&rdev->kobj, &mddev->kobj, "dev-%s", b)))
1529 ko = &part_to_dev(rdev->bdev->bd_part)->kobj;
1530 if ((err = sysfs_create_link(&rdev->kobj, ko, "block"))) {
1531 kobject_del(&rdev->kobj);
1534 rdev->sysfs_state = sysfs_get_dirent(rdev->kobj.sd, "state");
1536 list_add_rcu(&rdev->same_set, &mddev->disks);
1537 bd_claim_by_disk(rdev->bdev, rdev->bdev->bd_holder, mddev->gendisk);
1539 /* May as well allow recovery to be retried once */
1540 mddev->recovery_disabled = 0;
1542 md_integrity_check(rdev, mddev);
1546 printk(KERN_WARNING "md: failed to register dev-%s for %s\n",
1551 static void md_delayed_delete(struct work_struct *ws)
1553 mdk_rdev_t *rdev = container_of(ws, mdk_rdev_t, del_work);
1554 kobject_del(&rdev->kobj);
1555 kobject_put(&rdev->kobj);
1558 static void unbind_rdev_from_array(mdk_rdev_t * rdev)
1560 char b[BDEVNAME_SIZE];
1565 bd_release_from_disk(rdev->bdev, rdev->mddev->gendisk);
1566 list_del_rcu(&rdev->same_set);
1567 printk(KERN_INFO "md: unbind<%s>\n", bdevname(rdev->bdev,b));
1569 sysfs_remove_link(&rdev->kobj, "block");
1570 sysfs_put(rdev->sysfs_state);
1571 rdev->sysfs_state = NULL;
1572 /* We need to delay this, otherwise we can deadlock when
1573 * writing to 'remove' to "dev/state". We also need
1574 * to delay it due to rcu usage.
1577 INIT_WORK(&rdev->del_work, md_delayed_delete);
1578 kobject_get(&rdev->kobj);
1579 schedule_work(&rdev->del_work);
1583 * prevent the device from being mounted, repartitioned or
1584 * otherwise reused by a RAID array (or any other kernel
1585 * subsystem), by bd_claiming the device.
1587 static int lock_rdev(mdk_rdev_t *rdev, dev_t dev, int shared)
1590 struct block_device *bdev;
1591 char b[BDEVNAME_SIZE];
1593 bdev = open_by_devnum(dev, FMODE_READ|FMODE_WRITE);
1595 printk(KERN_ERR "md: could not open %s.\n",
1596 __bdevname(dev, b));
1597 return PTR_ERR(bdev);
1599 err = bd_claim(bdev, shared ? (mdk_rdev_t *)lock_rdev : rdev);
1601 printk(KERN_ERR "md: could not bd_claim %s.\n",
1603 blkdev_put(bdev, FMODE_READ|FMODE_WRITE);
1607 set_bit(AllReserved, &rdev->flags);
1612 static void unlock_rdev(mdk_rdev_t *rdev)
1614 struct block_device *bdev = rdev->bdev;
1619 blkdev_put(bdev, FMODE_READ|FMODE_WRITE);
1622 void md_autodetect_dev(dev_t dev);
1624 static void export_rdev(mdk_rdev_t * rdev)
1626 char b[BDEVNAME_SIZE];
1627 printk(KERN_INFO "md: export_rdev(%s)\n",
1628 bdevname(rdev->bdev,b));
1633 if (test_bit(AutoDetected, &rdev->flags))
1634 md_autodetect_dev(rdev->bdev->bd_dev);
1637 kobject_put(&rdev->kobj);
1640 static void kick_rdev_from_array(mdk_rdev_t * rdev)
1642 unbind_rdev_from_array(rdev);
1646 static void export_array(mddev_t *mddev)
1648 mdk_rdev_t *rdev, *tmp;
1650 rdev_for_each(rdev, tmp, mddev) {
1655 kick_rdev_from_array(rdev);
1657 if (!list_empty(&mddev->disks))
1659 mddev->raid_disks = 0;
1660 mddev->major_version = 0;
1663 static void print_desc(mdp_disk_t *desc)
1665 printk(" DISK<N:%d,(%d,%d),R:%d,S:%d>\n", desc->number,
1666 desc->major,desc->minor,desc->raid_disk,desc->state);
1669 static void print_sb_90(mdp_super_t *sb)
1674 "md: SB: (V:%d.%d.%d) ID:<%08x.%08x.%08x.%08x> CT:%08x\n",
1675 sb->major_version, sb->minor_version, sb->patch_version,
1676 sb->set_uuid0, sb->set_uuid1, sb->set_uuid2, sb->set_uuid3,
1678 printk(KERN_INFO "md: L%d S%08d ND:%d RD:%d md%d LO:%d CS:%d\n",
1679 sb->level, sb->size, sb->nr_disks, sb->raid_disks,
1680 sb->md_minor, sb->layout, sb->chunk_size);
1681 printk(KERN_INFO "md: UT:%08x ST:%d AD:%d WD:%d"
1682 " FD:%d SD:%d CSUM:%08x E:%08lx\n",
1683 sb->utime, sb->state, sb->active_disks, sb->working_disks,
1684 sb->failed_disks, sb->spare_disks,
1685 sb->sb_csum, (unsigned long)sb->events_lo);
1688 for (i = 0; i < MD_SB_DISKS; i++) {
1691 desc = sb->disks + i;
1692 if (desc->number || desc->major || desc->minor ||
1693 desc->raid_disk || (desc->state && (desc->state != 4))) {
1694 printk(" D %2d: ", i);
1698 printk(KERN_INFO "md: THIS: ");
1699 print_desc(&sb->this_disk);
1702 static void print_sb_1(struct mdp_superblock_1 *sb)
1706 uuid = sb->set_uuid;
1707 printk(KERN_INFO "md: SB: (V:%u) (F:0x%08x) Array-ID:<%02x%02x%02x%02x"
1708 ":%02x%02x:%02x%02x:%02x%02x:%02x%02x%02x%02x%02x%02x>\n"
1709 KERN_INFO "md: Name: \"%s\" CT:%llu\n",
1710 le32_to_cpu(sb->major_version),
1711 le32_to_cpu(sb->feature_map),
1712 uuid[0], uuid[1], uuid[2], uuid[3],
1713 uuid[4], uuid[5], uuid[6], uuid[7],
1714 uuid[8], uuid[9], uuid[10], uuid[11],
1715 uuid[12], uuid[13], uuid[14], uuid[15],
1717 (unsigned long long)le64_to_cpu(sb->ctime)
1718 & MD_SUPERBLOCK_1_TIME_SEC_MASK);
1720 uuid = sb->device_uuid;
1721 printk(KERN_INFO "md: L%u SZ%llu RD:%u LO:%u CS:%u DO:%llu DS:%llu SO:%llu"
1723 KERN_INFO "md: Dev:%08x UUID: %02x%02x%02x%02x:%02x%02x:%02x%02x:%02x%02x"
1724 ":%02x%02x%02x%02x%02x%02x\n"
1725 KERN_INFO "md: (F:0x%08x) UT:%llu Events:%llu ResyncOffset:%llu CSUM:0x%08x\n"
1726 KERN_INFO "md: (MaxDev:%u) \n",
1727 le32_to_cpu(sb->level),
1728 (unsigned long long)le64_to_cpu(sb->size),
1729 le32_to_cpu(sb->raid_disks),
1730 le32_to_cpu(sb->layout),
1731 le32_to_cpu(sb->chunksize),
1732 (unsigned long long)le64_to_cpu(sb->data_offset),
1733 (unsigned long long)le64_to_cpu(sb->data_size),
1734 (unsigned long long)le64_to_cpu(sb->super_offset),
1735 (unsigned long long)le64_to_cpu(sb->recovery_offset),
1736 le32_to_cpu(sb->dev_number),
1737 uuid[0], uuid[1], uuid[2], uuid[3],
1738 uuid[4], uuid[5], uuid[6], uuid[7],
1739 uuid[8], uuid[9], uuid[10], uuid[11],
1740 uuid[12], uuid[13], uuid[14], uuid[15],
1742 (unsigned long long)le64_to_cpu(sb->utime) & MD_SUPERBLOCK_1_TIME_SEC_MASK,
1743 (unsigned long long)le64_to_cpu(sb->events),
1744 (unsigned long long)le64_to_cpu(sb->resync_offset),
1745 le32_to_cpu(sb->sb_csum),
1746 le32_to_cpu(sb->max_dev)
1750 static void print_rdev(mdk_rdev_t *rdev, int major_version)
1752 char b[BDEVNAME_SIZE];
1753 printk(KERN_INFO "md: rdev %s, SZ:%08llu F:%d S:%d DN:%u\n",
1754 bdevname(rdev->bdev,b), (unsigned long long)rdev->size,
1755 test_bit(Faulty, &rdev->flags), test_bit(In_sync, &rdev->flags),
1757 if (rdev->sb_loaded) {
1758 printk(KERN_INFO "md: rdev superblock (MJ:%d):\n", major_version);
1759 switch (major_version) {
1761 print_sb_90((mdp_super_t*)page_address(rdev->sb_page));
1764 print_sb_1((struct mdp_superblock_1 *)page_address(rdev->sb_page));
1768 printk(KERN_INFO "md: no rdev superblock!\n");
1771 static void md_print_devices(void)
1773 struct list_head *tmp;
1776 char b[BDEVNAME_SIZE];
1779 printk("md: **********************************\n");
1780 printk("md: * <COMPLETE RAID STATE PRINTOUT> *\n");
1781 printk("md: **********************************\n");
1782 for_each_mddev(mddev, tmp) {
1785 bitmap_print_sb(mddev->bitmap);
1787 printk("%s: ", mdname(mddev));
1788 list_for_each_entry(rdev, &mddev->disks, same_set)
1789 printk("<%s>", bdevname(rdev->bdev,b));
1792 list_for_each_entry(rdev, &mddev->disks, same_set)
1793 print_rdev(rdev, mddev->major_version);
1795 printk("md: **********************************\n");
1800 static void sync_sbs(mddev_t * mddev, int nospares)
1802 /* Update each superblock (in-memory image), but
1803 * if we are allowed to, skip spares which already
1804 * have the right event counter, or have one earlier
1805 * (which would mean they aren't being marked as dirty
1806 * with the rest of the array)
1810 list_for_each_entry(rdev, &mddev->disks, same_set) {
1811 if (rdev->sb_events == mddev->events ||
1813 rdev->raid_disk < 0 &&
1814 (rdev->sb_events&1)==0 &&
1815 rdev->sb_events+1 == mddev->events)) {
1816 /* Don't update this superblock */
1817 rdev->sb_loaded = 2;
1819 super_types[mddev->major_version].
1820 sync_super(mddev, rdev);
1821 rdev->sb_loaded = 1;
1826 static void md_update_sb(mddev_t * mddev, int force_change)
1832 if (mddev->external)
1835 spin_lock_irq(&mddev->write_lock);
1837 set_bit(MD_CHANGE_PENDING, &mddev->flags);
1838 if (test_and_clear_bit(MD_CHANGE_DEVS, &mddev->flags))
1840 if (test_and_clear_bit(MD_CHANGE_CLEAN, &mddev->flags))
1841 /* just a clean<-> dirty transition, possibly leave spares alone,
1842 * though if events isn't the right even/odd, we will have to do
1848 if (mddev->degraded)
1849 /* If the array is degraded, then skipping spares is both
1850 * dangerous and fairly pointless.
1851 * Dangerous because a device that was removed from the array
1852 * might have a event_count that still looks up-to-date,
1853 * so it can be re-added without a resync.
1854 * Pointless because if there are any spares to skip,
1855 * then a recovery will happen and soon that array won't
1856 * be degraded any more and the spare can go back to sleep then.
1860 sync_req = mddev->in_sync;
1861 mddev->utime = get_seconds();
1863 /* If this is just a dirty<->clean transition, and the array is clean
1864 * and 'events' is odd, we can roll back to the previous clean state */
1866 && (mddev->in_sync && mddev->recovery_cp == MaxSector)
1867 && (mddev->events & 1)
1868 && mddev->events != 1)
1871 /* otherwise we have to go forward and ... */
1873 if (!mddev->in_sync || mddev->recovery_cp != MaxSector) { /* not clean */
1874 /* .. if the array isn't clean, insist on an odd 'events' */
1875 if ((mddev->events&1)==0) {
1880 /* otherwise insist on an even 'events' (for clean states) */
1881 if ((mddev->events&1)) {
1888 if (!mddev->events) {
1890 * oops, this 64-bit counter should never wrap.
1891 * Either we are in around ~1 trillion A.C., assuming
1892 * 1 reboot per second, or we have a bug:
1899 * do not write anything to disk if using
1900 * nonpersistent superblocks
1902 if (!mddev->persistent) {
1903 if (!mddev->external)
1904 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
1906 spin_unlock_irq(&mddev->write_lock);
1907 wake_up(&mddev->sb_wait);
1910 sync_sbs(mddev, nospares);
1911 spin_unlock_irq(&mddev->write_lock);
1914 "md: updating %s RAID superblock on device (in sync %d)\n",
1915 mdname(mddev),mddev->in_sync);
1917 bitmap_update_sb(mddev->bitmap);
1918 list_for_each_entry(rdev, &mddev->disks, same_set) {
1919 char b[BDEVNAME_SIZE];
1920 dprintk(KERN_INFO "md: ");
1921 if (rdev->sb_loaded != 1)
1922 continue; /* no noise on spare devices */
1923 if (test_bit(Faulty, &rdev->flags))
1924 dprintk("(skipping faulty ");
1926 dprintk("%s ", bdevname(rdev->bdev,b));
1927 if (!test_bit(Faulty, &rdev->flags)) {
1928 md_super_write(mddev,rdev,
1929 rdev->sb_start, rdev->sb_size,
1931 dprintk(KERN_INFO "(write) %s's sb offset: %llu\n",
1932 bdevname(rdev->bdev,b),
1933 (unsigned long long)rdev->sb_start);
1934 rdev->sb_events = mddev->events;
1938 if (mddev->level == LEVEL_MULTIPATH)
1939 /* only need to write one superblock... */
1942 md_super_wait(mddev);
1943 /* if there was a failure, MD_CHANGE_DEVS was set, and we re-write super */
1945 spin_lock_irq(&mddev->write_lock);
1946 if (mddev->in_sync != sync_req ||
1947 test_bit(MD_CHANGE_DEVS, &mddev->flags)) {
1948 /* have to write it out again */
1949 spin_unlock_irq(&mddev->write_lock);
1952 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
1953 spin_unlock_irq(&mddev->write_lock);
1954 wake_up(&mddev->sb_wait);
1958 /* words written to sysfs files may, or may not, be \n terminated.
1959 * We want to accept with case. For this we use cmd_match.
1961 static int cmd_match(const char *cmd, const char *str)
1963 /* See if cmd, written into a sysfs file, matches
1964 * str. They must either be the same, or cmd can
1965 * have a trailing newline
1967 while (*cmd && *str && *cmd == *str) {
1978 struct rdev_sysfs_entry {
1979 struct attribute attr;
1980 ssize_t (*show)(mdk_rdev_t *, char *);
1981 ssize_t (*store)(mdk_rdev_t *, const char *, size_t);
1985 state_show(mdk_rdev_t *rdev, char *page)
1990 if (test_bit(Faulty, &rdev->flags)) {
1991 len+= sprintf(page+len, "%sfaulty",sep);
1994 if (test_bit(In_sync, &rdev->flags)) {
1995 len += sprintf(page+len, "%sin_sync",sep);
1998 if (test_bit(WriteMostly, &rdev->flags)) {
1999 len += sprintf(page+len, "%swrite_mostly",sep);
2002 if (test_bit(Blocked, &rdev->flags)) {
2003 len += sprintf(page+len, "%sblocked", sep);
2006 if (!test_bit(Faulty, &rdev->flags) &&
2007 !test_bit(In_sync, &rdev->flags)) {
2008 len += sprintf(page+len, "%sspare", sep);
2011 return len+sprintf(page+len, "\n");
2015 state_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2018 * faulty - simulates and error
2019 * remove - disconnects the device
2020 * writemostly - sets write_mostly
2021 * -writemostly - clears write_mostly
2022 * blocked - sets the Blocked flag
2023 * -blocked - clears the Blocked flag
2026 if (cmd_match(buf, "faulty") && rdev->mddev->pers) {
2027 md_error(rdev->mddev, rdev);
2029 } else if (cmd_match(buf, "remove")) {
2030 if (rdev->raid_disk >= 0)
2033 mddev_t *mddev = rdev->mddev;
2034 kick_rdev_from_array(rdev);
2036 md_update_sb(mddev, 1);
2037 md_new_event(mddev);
2040 } else if (cmd_match(buf, "writemostly")) {
2041 set_bit(WriteMostly, &rdev->flags);
2043 } else if (cmd_match(buf, "-writemostly")) {
2044 clear_bit(WriteMostly, &rdev->flags);
2046 } else if (cmd_match(buf, "blocked")) {
2047 set_bit(Blocked, &rdev->flags);
2049 } else if (cmd_match(buf, "-blocked")) {
2050 clear_bit(Blocked, &rdev->flags);
2051 wake_up(&rdev->blocked_wait);
2052 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2053 md_wakeup_thread(rdev->mddev->thread);
2057 if (!err && rdev->sysfs_state)
2058 sysfs_notify_dirent(rdev->sysfs_state);
2059 return err ? err : len;
2061 static struct rdev_sysfs_entry rdev_state =
2062 __ATTR(state, S_IRUGO|S_IWUSR, state_show, state_store);
2065 errors_show(mdk_rdev_t *rdev, char *page)
2067 return sprintf(page, "%d\n", atomic_read(&rdev->corrected_errors));
2071 errors_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2074 unsigned long n = simple_strtoul(buf, &e, 10);
2075 if (*buf && (*e == 0 || *e == '\n')) {
2076 atomic_set(&rdev->corrected_errors, n);
2081 static struct rdev_sysfs_entry rdev_errors =
2082 __ATTR(errors, S_IRUGO|S_IWUSR, errors_show, errors_store);
2085 slot_show(mdk_rdev_t *rdev, char *page)
2087 if (rdev->raid_disk < 0)
2088 return sprintf(page, "none\n");
2090 return sprintf(page, "%d\n", rdev->raid_disk);
2094 slot_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2099 int slot = simple_strtoul(buf, &e, 10);
2100 if (strncmp(buf, "none", 4)==0)
2102 else if (e==buf || (*e && *e!= '\n'))
2104 if (rdev->mddev->pers && slot == -1) {
2105 /* Setting 'slot' on an active array requires also
2106 * updating the 'rd%d' link, and communicating
2107 * with the personality with ->hot_*_disk.
2108 * For now we only support removing
2109 * failed/spare devices. This normally happens automatically,
2110 * but not when the metadata is externally managed.
2112 if (rdev->raid_disk == -1)
2114 /* personality does all needed checks */
2115 if (rdev->mddev->pers->hot_add_disk == NULL)
2117 err = rdev->mddev->pers->
2118 hot_remove_disk(rdev->mddev, rdev->raid_disk);
2121 sprintf(nm, "rd%d", rdev->raid_disk);
2122 sysfs_remove_link(&rdev->mddev->kobj, nm);
2123 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2124 md_wakeup_thread(rdev->mddev->thread);
2125 } else if (rdev->mddev->pers) {
2127 /* Activating a spare .. or possibly reactivating
2128 * if we every get bitmaps working here.
2131 if (rdev->raid_disk != -1)
2134 if (rdev->mddev->pers->hot_add_disk == NULL)
2137 list_for_each_entry(rdev2, &rdev->mddev->disks, same_set)
2138 if (rdev2->raid_disk == slot)
2141 rdev->raid_disk = slot;
2142 if (test_bit(In_sync, &rdev->flags))
2143 rdev->saved_raid_disk = slot;
2145 rdev->saved_raid_disk = -1;
2146 err = rdev->mddev->pers->
2147 hot_add_disk(rdev->mddev, rdev);
2149 rdev->raid_disk = -1;
2152 sysfs_notify_dirent(rdev->sysfs_state);
2153 sprintf(nm, "rd%d", rdev->raid_disk);
2154 if (sysfs_create_link(&rdev->mddev->kobj, &rdev->kobj, nm))
2156 "md: cannot register "
2158 nm, mdname(rdev->mddev));
2160 /* don't wakeup anyone, leave that to userspace. */
2162 if (slot >= rdev->mddev->raid_disks)
2164 rdev->raid_disk = slot;
2165 /* assume it is working */
2166 clear_bit(Faulty, &rdev->flags);
2167 clear_bit(WriteMostly, &rdev->flags);
2168 set_bit(In_sync, &rdev->flags);
2169 sysfs_notify_dirent(rdev->sysfs_state);
2175 static struct rdev_sysfs_entry rdev_slot =
2176 __ATTR(slot, S_IRUGO|S_IWUSR, slot_show, slot_store);
2179 offset_show(mdk_rdev_t *rdev, char *page)
2181 return sprintf(page, "%llu\n", (unsigned long long)rdev->data_offset);
2185 offset_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2188 unsigned long long offset = simple_strtoull(buf, &e, 10);
2189 if (e==buf || (*e && *e != '\n'))
2191 if (rdev->mddev->pers && rdev->raid_disk >= 0)
2193 if (rdev->size && rdev->mddev->external)
2194 /* Must set offset before size, so overlap checks
2197 rdev->data_offset = offset;
2201 static struct rdev_sysfs_entry rdev_offset =
2202 __ATTR(offset, S_IRUGO|S_IWUSR, offset_show, offset_store);
2205 rdev_size_show(mdk_rdev_t *rdev, char *page)
2207 return sprintf(page, "%llu\n", (unsigned long long)rdev->size);
2210 static int overlaps(sector_t s1, sector_t l1, sector_t s2, sector_t l2)
2212 /* check if two start/length pairs overlap */
2221 rdev_size_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2223 unsigned long long size;
2224 unsigned long long oldsize = rdev->size;
2225 mddev_t *my_mddev = rdev->mddev;
2227 if (strict_strtoull(buf, 10, &size) < 0)
2229 if (my_mddev->pers && rdev->raid_disk >= 0) {
2230 if (my_mddev->persistent) {
2231 size = super_types[my_mddev->major_version].
2232 rdev_size_change(rdev, size * 2);
2236 size = (rdev->bdev->bd_inode->i_size >> 10);
2237 size -= rdev->data_offset/2;
2240 if (size < my_mddev->size)
2241 return -EINVAL; /* component must fit device */
2244 if (size > oldsize && my_mddev->external) {
2245 /* need to check that all other rdevs with the same ->bdev
2246 * do not overlap. We need to unlock the mddev to avoid
2247 * a deadlock. We have already changed rdev->size, and if
2248 * we have to change it back, we will have the lock again.
2252 struct list_head *tmp;
2254 mddev_unlock(my_mddev);
2255 for_each_mddev(mddev, tmp) {
2259 list_for_each_entry(rdev2, &mddev->disks, same_set)
2260 if (test_bit(AllReserved, &rdev2->flags) ||
2261 (rdev->bdev == rdev2->bdev &&
2263 overlaps(rdev->data_offset, rdev->size * 2,
2265 rdev2->size * 2))) {
2269 mddev_unlock(mddev);
2275 mddev_lock(my_mddev);
2277 /* Someone else could have slipped in a size
2278 * change here, but doing so is just silly.
2279 * We put oldsize back because we *know* it is
2280 * safe, and trust userspace not to race with
2283 rdev->size = oldsize;
2290 static struct rdev_sysfs_entry rdev_size =
2291 __ATTR(size, S_IRUGO|S_IWUSR, rdev_size_show, rdev_size_store);
2293 static struct attribute *rdev_default_attrs[] = {
2302 rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
2304 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
2305 mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
2306 mddev_t *mddev = rdev->mddev;
2312 rv = mddev ? mddev_lock(mddev) : -EBUSY;
2314 if (rdev->mddev == NULL)
2317 rv = entry->show(rdev, page);
2318 mddev_unlock(mddev);
2324 rdev_attr_store(struct kobject *kobj, struct attribute *attr,
2325 const char *page, size_t length)
2327 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
2328 mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
2330 mddev_t *mddev = rdev->mddev;
2334 if (!capable(CAP_SYS_ADMIN))
2336 rv = mddev ? mddev_lock(mddev): -EBUSY;
2338 if (rdev->mddev == NULL)
2341 rv = entry->store(rdev, page, length);
2342 mddev_unlock(mddev);
2347 static void rdev_free(struct kobject *ko)
2349 mdk_rdev_t *rdev = container_of(ko, mdk_rdev_t, kobj);
2352 static struct sysfs_ops rdev_sysfs_ops = {
2353 .show = rdev_attr_show,
2354 .store = rdev_attr_store,
2356 static struct kobj_type rdev_ktype = {
2357 .release = rdev_free,
2358 .sysfs_ops = &rdev_sysfs_ops,
2359 .default_attrs = rdev_default_attrs,
2363 * Import a device. If 'super_format' >= 0, then sanity check the superblock
2365 * mark the device faulty if:
2367 * - the device is nonexistent (zero size)
2368 * - the device has no valid superblock
2370 * a faulty rdev _never_ has rdev->sb set.
2372 static mdk_rdev_t *md_import_device(dev_t newdev, int super_format, int super_minor)
2374 char b[BDEVNAME_SIZE];
2379 rdev = kzalloc(sizeof(*rdev), GFP_KERNEL);
2381 printk(KERN_ERR "md: could not alloc mem for new device!\n");
2382 return ERR_PTR(-ENOMEM);
2385 if ((err = alloc_disk_sb(rdev)))
2388 err = lock_rdev(rdev, newdev, super_format == -2);
2392 kobject_init(&rdev->kobj, &rdev_ktype);
2395 rdev->saved_raid_disk = -1;
2396 rdev->raid_disk = -1;
2398 rdev->data_offset = 0;
2399 rdev->sb_events = 0;
2400 atomic_set(&rdev->nr_pending, 0);
2401 atomic_set(&rdev->read_errors, 0);
2402 atomic_set(&rdev->corrected_errors, 0);
2404 size = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
2407 "md: %s has zero or unknown size, marking faulty!\n",
2408 bdevname(rdev->bdev,b));
2413 if (super_format >= 0) {
2414 err = super_types[super_format].
2415 load_super(rdev, NULL, super_minor);
2416 if (err == -EINVAL) {
2418 "md: %s does not have a valid v%d.%d "
2419 "superblock, not importing!\n",
2420 bdevname(rdev->bdev,b),
2421 super_format, super_minor);
2426 "md: could not read %s's sb, not importing!\n",
2427 bdevname(rdev->bdev,b));
2432 INIT_LIST_HEAD(&rdev->same_set);
2433 init_waitqueue_head(&rdev->blocked_wait);
2438 if (rdev->sb_page) {
2444 return ERR_PTR(err);
2448 * Check a full RAID array for plausibility
2452 static void analyze_sbs(mddev_t * mddev)
2455 mdk_rdev_t *rdev, *freshest, *tmp;
2456 char b[BDEVNAME_SIZE];
2459 rdev_for_each(rdev, tmp, mddev)
2460 switch (super_types[mddev->major_version].
2461 load_super(rdev, freshest, mddev->minor_version)) {
2469 "md: fatal superblock inconsistency in %s"
2470 " -- removing from array\n",
2471 bdevname(rdev->bdev,b));
2472 kick_rdev_from_array(rdev);
2476 super_types[mddev->major_version].
2477 validate_super(mddev, freshest);
2480 rdev_for_each(rdev, tmp, mddev) {
2481 if (rdev->desc_nr >= mddev->max_disks ||
2482 i > mddev->max_disks) {
2484 "md: %s: %s: only %d devices permitted\n",
2485 mdname(mddev), bdevname(rdev->bdev, b),
2487 kick_rdev_from_array(rdev);
2490 if (rdev != freshest)
2491 if (super_types[mddev->major_version].
2492 validate_super(mddev, rdev)) {
2493 printk(KERN_WARNING "md: kicking non-fresh %s"
2495 bdevname(rdev->bdev,b));
2496 kick_rdev_from_array(rdev);
2499 if (mddev->level == LEVEL_MULTIPATH) {
2500 rdev->desc_nr = i++;
2501 rdev->raid_disk = rdev->desc_nr;
2502 set_bit(In_sync, &rdev->flags);
2503 } else if (rdev->raid_disk >= mddev->raid_disks) {
2504 rdev->raid_disk = -1;
2505 clear_bit(In_sync, &rdev->flags);
2511 if (mddev->recovery_cp != MaxSector &&
2513 printk(KERN_ERR "md: %s: raid array is not clean"
2514 " -- starting background reconstruction\n",
2519 static void md_safemode_timeout(unsigned long data);
2522 safe_delay_show(mddev_t *mddev, char *page)
2524 int msec = (mddev->safemode_delay*1000)/HZ;
2525 return sprintf(page, "%d.%03d\n", msec/1000, msec%1000);
2528 safe_delay_store(mddev_t *mddev, const char *cbuf, size_t len)
2536 /* remove a period, and count digits after it */
2537 if (len >= sizeof(buf))
2539 strlcpy(buf, cbuf, sizeof(buf));
2540 for (i=0; i<len; i++) {
2542 if (isdigit(buf[i])) {
2547 } else if (buf[i] == '.') {
2552 if (strict_strtoul(buf, 10, &msec) < 0)
2554 msec = (msec * 1000) / scale;
2556 mddev->safemode_delay = 0;
2558 unsigned long old_delay = mddev->safemode_delay;
2559 mddev->safemode_delay = (msec*HZ)/1000;
2560 if (mddev->safemode_delay == 0)
2561 mddev->safemode_delay = 1;
2562 if (mddev->safemode_delay < old_delay)
2563 md_safemode_timeout((unsigned long)mddev);
2567 static struct md_sysfs_entry md_safe_delay =
2568 __ATTR(safe_mode_delay, S_IRUGO|S_IWUSR,safe_delay_show, safe_delay_store);
2571 level_show(mddev_t *mddev, char *page)
2573 struct mdk_personality *p = mddev->pers;
2575 return sprintf(page, "%s\n", p->name);
2576 else if (mddev->clevel[0])
2577 return sprintf(page, "%s\n", mddev->clevel);
2578 else if (mddev->level != LEVEL_NONE)
2579 return sprintf(page, "%d\n", mddev->level);
2585 level_store(mddev_t *mddev, const char *buf, size_t len)
2592 if (len >= sizeof(mddev->clevel))
2594 strncpy(mddev->clevel, buf, len);
2595 if (mddev->clevel[len-1] == '\n')
2597 mddev->clevel[len] = 0;
2598 mddev->level = LEVEL_NONE;
2602 static struct md_sysfs_entry md_level =
2603 __ATTR(level, S_IRUGO|S_IWUSR, level_show, level_store);
2607 layout_show(mddev_t *mddev, char *page)
2609 /* just a number, not meaningful for all levels */
2610 if (mddev->reshape_position != MaxSector &&
2611 mddev->layout != mddev->new_layout)
2612 return sprintf(page, "%d (%d)\n",
2613 mddev->new_layout, mddev->layout);
2614 return sprintf(page, "%d\n", mddev->layout);
2618 layout_store(mddev_t *mddev, const char *buf, size_t len)
2621 unsigned long n = simple_strtoul(buf, &e, 10);
2623 if (!*buf || (*e && *e != '\n'))
2628 if (mddev->reshape_position != MaxSector)
2629 mddev->new_layout = n;
2634 static struct md_sysfs_entry md_layout =
2635 __ATTR(layout, S_IRUGO|S_IWUSR, layout_show, layout_store);
2639 raid_disks_show(mddev_t *mddev, char *page)
2641 if (mddev->raid_disks == 0)
2643 if (mddev->reshape_position != MaxSector &&
2644 mddev->delta_disks != 0)
2645 return sprintf(page, "%d (%d)\n", mddev->raid_disks,
2646 mddev->raid_disks - mddev->delta_disks);
2647 return sprintf(page, "%d\n", mddev->raid_disks);
2650 static int update_raid_disks(mddev_t *mddev, int raid_disks);
2653 raid_disks_store(mddev_t *mddev, const char *buf, size_t len)
2657 unsigned long n = simple_strtoul(buf, &e, 10);
2659 if (!*buf || (*e && *e != '\n'))
2663 rv = update_raid_disks(mddev, n);
2664 else if (mddev->reshape_position != MaxSector) {
2665 int olddisks = mddev->raid_disks - mddev->delta_disks;
2666 mddev->delta_disks = n - olddisks;
2667 mddev->raid_disks = n;
2669 mddev->raid_disks = n;
2670 return rv ? rv : len;
2672 static struct md_sysfs_entry md_raid_disks =
2673 __ATTR(raid_disks, S_IRUGO|S_IWUSR, raid_disks_show, raid_disks_store);
2676 chunk_size_show(mddev_t *mddev, char *page)
2678 if (mddev->reshape_position != MaxSector &&
2679 mddev->chunk_size != mddev->new_chunk)
2680 return sprintf(page, "%d (%d)\n", mddev->new_chunk,
2682 return sprintf(page, "%d\n", mddev->chunk_size);
2686 chunk_size_store(mddev_t *mddev, const char *buf, size_t len)
2688 /* can only set chunk_size if array is not yet active */
2690 unsigned long n = simple_strtoul(buf, &e, 10);
2692 if (!*buf || (*e && *e != '\n'))
2697 else if (mddev->reshape_position != MaxSector)
2698 mddev->new_chunk = n;
2700 mddev->chunk_size = n;
2703 static struct md_sysfs_entry md_chunk_size =
2704 __ATTR(chunk_size, S_IRUGO|S_IWUSR, chunk_size_show, chunk_size_store);
2707 resync_start_show(mddev_t *mddev, char *page)
2709 return sprintf(page, "%llu\n", (unsigned long long)mddev->recovery_cp);
2713 resync_start_store(mddev_t *mddev, const char *buf, size_t len)
2716 unsigned long long n = simple_strtoull(buf, &e, 10);
2720 if (!*buf || (*e && *e != '\n'))
2723 mddev->recovery_cp = n;
2726 static struct md_sysfs_entry md_resync_start =
2727 __ATTR(resync_start, S_IRUGO|S_IWUSR, resync_start_show, resync_start_store);
2730 * The array state can be:
2733 * No devices, no size, no level
2734 * Equivalent to STOP_ARRAY ioctl
2736 * May have some settings, but array is not active
2737 * all IO results in error
2738 * When written, doesn't tear down array, but just stops it
2739 * suspended (not supported yet)
2740 * All IO requests will block. The array can be reconfigured.
2741 * Writing this, if accepted, will block until array is quiescent
2743 * no resync can happen. no superblocks get written.
2744 * write requests fail
2746 * like readonly, but behaves like 'clean' on a write request.
2748 * clean - no pending writes, but otherwise active.
2749 * When written to inactive array, starts without resync
2750 * If a write request arrives then
2751 * if metadata is known, mark 'dirty' and switch to 'active'.
2752 * if not known, block and switch to write-pending
2753 * If written to an active array that has pending writes, then fails.
2755 * fully active: IO and resync can be happening.
2756 * When written to inactive array, starts with resync
2759 * clean, but writes are blocked waiting for 'active' to be written.
2762 * like active, but no writes have been seen for a while (100msec).
2765 enum array_state { clear, inactive, suspended, readonly, read_auto, clean, active,
2766 write_pending, active_idle, bad_word};
2767 static char *array_states[] = {
2768 "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active",
2769 "write-pending", "active-idle", NULL };
2771 static int match_word(const char *word, char **list)
2774 for (n=0; list[n]; n++)
2775 if (cmd_match(word, list[n]))
2781 array_state_show(mddev_t *mddev, char *page)
2783 enum array_state st = inactive;
2796 else if (test_bit(MD_CHANGE_CLEAN, &mddev->flags))
2798 else if (mddev->safemode)
2804 if (list_empty(&mddev->disks) &&
2805 mddev->raid_disks == 0 &&
2811 return sprintf(page, "%s\n", array_states[st]);
2814 static int do_md_stop(mddev_t * mddev, int ro, int is_open);
2815 static int do_md_run(mddev_t * mddev);
2816 static int restart_array(mddev_t *mddev);
2819 array_state_store(mddev_t *mddev, const char *buf, size_t len)
2822 enum array_state st = match_word(buf, array_states);
2827 /* stopping an active array */
2828 if (atomic_read(&mddev->openers) > 0)
2830 err = do_md_stop(mddev, 0, 0);
2833 /* stopping an active array */
2835 if (atomic_read(&mddev->openers) > 0)
2837 err = do_md_stop(mddev, 2, 0);
2839 err = 0; /* already inactive */
2842 break; /* not supported yet */
2845 err = do_md_stop(mddev, 1, 0);
2848 set_disk_ro(mddev->gendisk, 1);
2849 err = do_md_run(mddev);
2855 err = do_md_stop(mddev, 1, 0);
2856 else if (mddev->ro == 1)
2857 err = restart_array(mddev);
2860 set_disk_ro(mddev->gendisk, 0);
2864 err = do_md_run(mddev);
2869 restart_array(mddev);
2870 spin_lock_irq(&mddev->write_lock);
2871 if (atomic_read(&mddev->writes_pending) == 0) {
2872 if (mddev->in_sync == 0) {
2874 if (mddev->safemode == 1)
2875 mddev->safemode = 0;
2876 if (mddev->persistent)
2877 set_bit(MD_CHANGE_CLEAN,
2883 spin_unlock_irq(&mddev->write_lock);
2886 mddev->recovery_cp = MaxSector;
2887 err = do_md_run(mddev);
2892 restart_array(mddev);
2893 if (mddev->external)
2894 clear_bit(MD_CHANGE_CLEAN, &mddev->flags);
2895 wake_up(&mddev->sb_wait);
2899 set_disk_ro(mddev->gendisk, 0);
2900 err = do_md_run(mddev);
2905 /* these cannot be set */
2911 sysfs_notify_dirent(mddev->sysfs_state);
2915 static struct md_sysfs_entry md_array_state =
2916 __ATTR(array_state, S_IRUGO|S_IWUSR, array_state_show, array_state_store);
2919 null_show(mddev_t *mddev, char *page)
2925 new_dev_store(mddev_t *mddev, const char *buf, size_t len)
2927 /* buf must be %d:%d\n? giving major and minor numbers */
2928 /* The new device is added to the array.
2929 * If the array has a persistent superblock, we read the
2930 * superblock to initialise info and check validity.
2931 * Otherwise, only checking done is that in bind_rdev_to_array,
2932 * which mainly checks size.
2935 int major = simple_strtoul(buf, &e, 10);
2941 if (!*buf || *e != ':' || !e[1] || e[1] == '\n')
2943 minor = simple_strtoul(e+1, &e, 10);
2944 if (*e && *e != '\n')
2946 dev = MKDEV(major, minor);
2947 if (major != MAJOR(dev) ||
2948 minor != MINOR(dev))
2952 if (mddev->persistent) {
2953 rdev = md_import_device(dev, mddev->major_version,
2954 mddev->minor_version);
2955 if (!IS_ERR(rdev) && !list_empty(&mddev->disks)) {
2956 mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
2957 mdk_rdev_t, same_set);
2958 err = super_types[mddev->major_version]
2959 .load_super(rdev, rdev0, mddev->minor_version);
2963 } else if (mddev->external)
2964 rdev = md_import_device(dev, -2, -1);
2966 rdev = md_import_device(dev, -1, -1);
2969 return PTR_ERR(rdev);
2970 err = bind_rdev_to_array(rdev, mddev);
2974 return err ? err : len;
2977 static struct md_sysfs_entry md_new_device =
2978 __ATTR(new_dev, S_IWUSR, null_show, new_dev_store);
2981 bitmap_store(mddev_t *mddev, const char *buf, size_t len)
2984 unsigned long chunk, end_chunk;
2988 /* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */
2990 chunk = end_chunk = simple_strtoul(buf, &end, 0);
2991 if (buf == end) break;
2992 if (*end == '-') { /* range */
2994 end_chunk = simple_strtoul(buf, &end, 0);
2995 if (buf == end) break;
2997 if (*end && !isspace(*end)) break;
2998 bitmap_dirty_bits(mddev->bitmap, chunk, end_chunk);
3000 while (isspace(*buf)) buf++;
3002 bitmap_unplug(mddev->bitmap); /* flush the bits to disk */
3007 static struct md_sysfs_entry md_bitmap =
3008 __ATTR(bitmap_set_bits, S_IWUSR, null_show, bitmap_store);
3011 size_show(mddev_t *mddev, char *page)
3013 return sprintf(page, "%llu\n", (unsigned long long)mddev->size);
3016 static int update_size(mddev_t *mddev, sector_t num_sectors);
3019 size_store(mddev_t *mddev, const char *buf, size_t len)
3021 /* If array is inactive, we can reduce the component size, but
3022 * not increase it (except from 0).
3023 * If array is active, we can try an on-line resize
3027 unsigned long long size = simple_strtoull(buf, &e, 10);
3028 if (!*buf || *buf == '\n' ||
3033 err = update_size(mddev, size * 2);
3034 md_update_sb(mddev, 1);
3036 if (mddev->size == 0 ||
3042 return err ? err : len;
3045 static struct md_sysfs_entry md_size =
3046 __ATTR(component_size, S_IRUGO|S_IWUSR, size_show, size_store);
3051 * 'none' for arrays with no metadata (good luck...)
3052 * 'external' for arrays with externally managed metadata,
3053 * or N.M for internally known formats
3056 metadata_show(mddev_t *mddev, char *page)
3058 if (mddev->persistent)
3059 return sprintf(page, "%d.%d\n",
3060 mddev->major_version, mddev->minor_version);
3061 else if (mddev->external)
3062 return sprintf(page, "external:%s\n", mddev->metadata_type);
3064 return sprintf(page, "none\n");
3068 metadata_store(mddev_t *mddev, const char *buf, size_t len)
3072 /* Changing the details of 'external' metadata is
3073 * always permitted. Otherwise there must be
3074 * no devices attached to the array.
3076 if (mddev->external && strncmp(buf, "external:", 9) == 0)
3078 else if (!list_empty(&mddev->disks))
3081 if (cmd_match(buf, "none")) {
3082 mddev->persistent = 0;
3083 mddev->external = 0;
3084 mddev->major_version = 0;
3085 mddev->minor_version = 90;
3088 if (strncmp(buf, "external:", 9) == 0) {
3089 size_t namelen = len-9;
3090 if (namelen >= sizeof(mddev->metadata_type))
3091 namelen = sizeof(mddev->metadata_type)-1;
3092 strncpy(mddev->metadata_type, buf+9, namelen);
3093 mddev->metadata_type[namelen] = 0;
3094 if (namelen && mddev->metadata_type[namelen-1] == '\n')
3095 mddev->metadata_type[--namelen] = 0;
3096 mddev->persistent = 0;
3097 mddev->external = 1;
3098 mddev->major_version = 0;
3099 mddev->minor_version = 90;
3102 major = simple_strtoul(buf, &e, 10);
3103 if (e==buf || *e != '.')
3106 minor = simple_strtoul(buf, &e, 10);
3107 if (e==buf || (*e && *e != '\n') )
3109 if (major >= ARRAY_SIZE(super_types) || super_types[major].name == NULL)
3111 mddev->major_version = major;
3112 mddev->minor_version = minor;
3113 mddev->persistent = 1;
3114 mddev->external = 0;
3118 static struct md_sysfs_entry md_metadata =
3119 __ATTR(metadata_version, S_IRUGO|S_IWUSR, metadata_show, metadata_store);
3122 action_show(mddev_t *mddev, char *page)
3124 char *type = "idle";
3125 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
3126 (!mddev->ro && test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))) {
3127 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
3129 else if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
3130 if (!test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
3132 else if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
3136 } else if (test_bit(MD_RECOVERY_RECOVER, &mddev->recovery))
3139 return sprintf(page, "%s\n", type);
3143 action_store(mddev_t *mddev, const char *page, size_t len)
3145 if (!mddev->pers || !mddev->pers->sync_request)
3148 if (cmd_match(page, "idle")) {
3149 if (mddev->sync_thread) {
3150 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
3151 md_unregister_thread(mddev->sync_thread);
3152 mddev->sync_thread = NULL;
3153 mddev->recovery = 0;
3155 } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
3156 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
3158 else if (cmd_match(page, "resync"))
3159 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3160 else if (cmd_match(page, "recover")) {
3161 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
3162 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3163 } else if (cmd_match(page, "reshape")) {
3165 if (mddev->pers->start_reshape == NULL)
3167 err = mddev->pers->start_reshape(mddev);
3170 sysfs_notify(&mddev->kobj, NULL, "degraded");
3172 if (cmd_match(page, "check"))
3173 set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
3174 else if (!cmd_match(page, "repair"))
3176 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
3177 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
3179 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3180 md_wakeup_thread(mddev->thread);
3181 sysfs_notify_dirent(mddev->sysfs_action);
3186 mismatch_cnt_show(mddev_t *mddev, char *page)
3188 return sprintf(page, "%llu\n",
3189 (unsigned long long) mddev->resync_mismatches);
3192 static struct md_sysfs_entry md_scan_mode =
3193 __ATTR(sync_action, S_IRUGO|S_IWUSR, action_show, action_store);
3196 static struct md_sysfs_entry md_mismatches = __ATTR_RO(mismatch_cnt);
3199 sync_min_show(mddev_t *mddev, char *page)
3201 return sprintf(page, "%d (%s)\n", speed_min(mddev),
3202 mddev->sync_speed_min ? "local": "system");
3206 sync_min_store(mddev_t *mddev, const char *buf, size_t len)
3210 if (strncmp(buf, "system", 6)==0) {
3211 mddev->sync_speed_min = 0;
3214 min = simple_strtoul(buf, &e, 10);
3215 if (buf == e || (*e && *e != '\n') || min <= 0)
3217 mddev->sync_speed_min = min;
3221 static struct md_sysfs_entry md_sync_min =
3222 __ATTR(sync_speed_min, S_IRUGO|S_IWUSR, sync_min_show, sync_min_store);
3225 sync_max_show(mddev_t *mddev, char *page)
3227 return sprintf(page, "%d (%s)\n", speed_max(mddev),
3228 mddev->sync_speed_max ? "local": "system");
3232 sync_max_store(mddev_t *mddev, const char *buf, size_t len)
3236 if (strncmp(buf, "system", 6)==0) {
3237 mddev->sync_speed_max = 0;
3240 max = simple_strtoul(buf, &e, 10);
3241 if (buf == e || (*e && *e != '\n') || max <= 0)
3243 mddev->sync_speed_max = max;
3247 static struct md_sysfs_entry md_sync_max =
3248 __ATTR(sync_speed_max, S_IRUGO|S_IWUSR, sync_max_show, sync_max_store);
3251 degraded_show(mddev_t *mddev, char *page)
3253 return sprintf(page, "%d\n", mddev->degraded);
3255 static struct md_sysfs_entry md_degraded = __ATTR_RO(degraded);
3258 sync_force_parallel_show(mddev_t *mddev, char *page)
3260 return sprintf(page, "%d\n", mddev->parallel_resync);
3264 sync_force_parallel_store(mddev_t *mddev, const char *buf, size_t len)
3268 if (strict_strtol(buf, 10, &n))
3271 if (n != 0 && n != 1)
3274 mddev->parallel_resync = n;
3276 if (mddev->sync_thread)
3277 wake_up(&resync_wait);
3282 /* force parallel resync, even with shared block devices */
3283 static struct md_sysfs_entry md_sync_force_parallel =
3284 __ATTR(sync_force_parallel, S_IRUGO|S_IWUSR,
3285 sync_force_parallel_show, sync_force_parallel_store);
3288 sync_speed_show(mddev_t *mddev, char *page)
3290 unsigned long resync, dt, db;
3291 resync = mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active);
3292 dt = (jiffies - mddev->resync_mark) / HZ;
3294 db = resync - mddev->resync_mark_cnt;
3295 return sprintf(page, "%lu\n", db/dt/2); /* K/sec */
3298 static struct md_sysfs_entry md_sync_speed = __ATTR_RO(sync_speed);
3301 sync_completed_show(mddev_t *mddev, char *page)
3303 unsigned long max_blocks, resync;
3305 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
3306 max_blocks = mddev->resync_max_sectors;
3308 max_blocks = mddev->size << 1;
3310 resync = (mddev->curr_resync - atomic_read(&mddev->recovery_active));
3311 return sprintf(page, "%lu / %lu\n", resync, max_blocks);
3314 static struct md_sysfs_entry md_sync_completed = __ATTR_RO(sync_completed);
3317 min_sync_show(mddev_t *mddev, char *page)
3319 return sprintf(page, "%llu\n",
3320 (unsigned long long)mddev->resync_min);
3323 min_sync_store(mddev_t *mddev, const char *buf, size_t len)
3325 unsigned long long min;
3326 if (strict_strtoull(buf, 10, &min))
3328 if (min > mddev->resync_max)
3330 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
3333 /* Must be a multiple of chunk_size */
3334 if (mddev->chunk_size) {
3335 if (min & (sector_t)((mddev->chunk_size>>9)-1))
3338 mddev->resync_min = min;
3343 static struct md_sysfs_entry md_min_sync =
3344 __ATTR(sync_min, S_IRUGO|S_IWUSR, min_sync_show, min_sync_store);
3347 max_sync_show(mddev_t *mddev, char *page)
3349 if (mddev->resync_max == MaxSector)
3350 return sprintf(page, "max\n");
3352 return sprintf(page, "%llu\n",
3353 (unsigned long long)mddev->resync_max);
3356 max_sync_store(mddev_t *mddev, const char *buf, size_t len)
3358 if (strncmp(buf, "max", 3) == 0)
3359 mddev->resync_max = MaxSector;
3361 unsigned long long max;
3362 if (strict_strtoull(buf, 10, &max))
3364 if (max < mddev->resync_min)
3366 if (max < mddev->resync_max &&
3367 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
3370 /* Must be a multiple of chunk_size */
3371 if (mddev->chunk_size) {
3372 if (max & (sector_t)((mddev->chunk_size>>9)-1))
3375 mddev->resync_max = max;
3377 wake_up(&mddev->recovery_wait);
3381 static struct md_sysfs_entry md_max_sync =
3382 __ATTR(sync_max, S_IRUGO|S_IWUSR, max_sync_show, max_sync_store);
3385 suspend_lo_show(mddev_t *mddev, char *page)
3387 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_lo);
3391 suspend_lo_store(mddev_t *mddev, const char *buf, size_t len)
3394 unsigned long long new = simple_strtoull(buf, &e, 10);
3396 if (mddev->pers->quiesce == NULL)
3398 if (buf == e || (*e && *e != '\n'))
3400 if (new >= mddev->suspend_hi ||
3401 (new > mddev->suspend_lo && new < mddev->suspend_hi)) {
3402 mddev->suspend_lo = new;
3403 mddev->pers->quiesce(mddev, 2);
3408 static struct md_sysfs_entry md_suspend_lo =
3409 __ATTR(suspend_lo, S_IRUGO|S_IWUSR, suspend_lo_show, suspend_lo_store);
3413 suspend_hi_show(mddev_t *mddev, char *page)
3415 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_hi);
3419 suspend_hi_store(mddev_t *mddev, const char *buf, size_t len)
3422 unsigned long long new = simple_strtoull(buf, &e, 10);
3424 if (mddev->pers->quiesce == NULL)
3426 if (buf == e || (*e && *e != '\n'))
3428 if ((new <= mddev->suspend_lo && mddev->suspend_lo >= mddev->suspend_hi) ||
3429 (new > mddev->suspend_lo && new > mddev->suspend_hi)) {
3430 mddev->suspend_hi = new;
3431 mddev->pers->quiesce(mddev, 1);
3432 mddev->pers->quiesce(mddev, 0);
3437 static struct md_sysfs_entry md_suspend_hi =
3438 __ATTR(suspend_hi, S_IRUGO|S_IWUSR, suspend_hi_show, suspend_hi_store);
3441 reshape_position_show(mddev_t *mddev, char *page)
3443 if (mddev->reshape_position != MaxSector)
3444 return sprintf(page, "%llu\n",
3445 (unsigned long long)mddev->reshape_position);
3446 strcpy(page, "none\n");
3451 reshape_position_store(mddev_t *mddev, const char *buf, size_t len)
3454 unsigned long long new = simple_strtoull(buf, &e, 10);
3457 if (buf == e || (*e && *e != '\n'))
3459 mddev->reshape_position = new;
3460 mddev->delta_disks = 0;
3461 mddev->new_level = mddev->level;
3462 mddev->new_layout = mddev->layout;
3463 mddev->new_chunk = mddev->chunk_size;
3467 static struct md_sysfs_entry md_reshape_position =
3468 __ATTR(reshape_position, S_IRUGO|S_IWUSR, reshape_position_show,
3469 reshape_position_store);
3472 static struct attribute *md_default_attrs[] = {
3475 &md_raid_disks.attr,
3476 &md_chunk_size.attr,
3478 &md_resync_start.attr,
3480 &md_new_device.attr,
3481 &md_safe_delay.attr,
3482 &md_array_state.attr,
3483 &md_reshape_position.attr,
3487 static struct attribute *md_redundancy_attrs[] = {
3489 &md_mismatches.attr,
3492 &md_sync_speed.attr,
3493 &md_sync_force_parallel.attr,
3494 &md_sync_completed.attr,
3497 &md_suspend_lo.attr,
3498 &md_suspend_hi.attr,
3503 static struct attribute_group md_redundancy_group = {
3505 .attrs = md_redundancy_attrs,
3510 md_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
3512 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
3513 mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
3518 rv = mddev_lock(mddev);
3520 rv = entry->show(mddev, page);
3521 mddev_unlock(mddev);
3527 md_attr_store(struct kobject *kobj, struct attribute *attr,
3528 const char *page, size_t length)
3530 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
3531 mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
3536 if (!capable(CAP_SYS_ADMIN))
3538 rv = mddev_lock(mddev);
3539 if (mddev->hold_active == UNTIL_IOCTL)
3540 mddev->hold_active = 0;
3542 rv = entry->store(mddev, page, length);
3543 mddev_unlock(mddev);
3548 static void md_free(struct kobject *ko)
3550 mddev_t *mddev = container_of(ko, mddev_t, kobj);
3552 if (mddev->sysfs_state)
3553 sysfs_put(mddev->sysfs_state);
3555 if (mddev->gendisk) {
3556 del_gendisk(mddev->gendisk);
3557 put_disk(mddev->gendisk);
3560 blk_cleanup_queue(mddev->queue);
3565 static struct sysfs_ops md_sysfs_ops = {
3566 .show = md_attr_show,
3567 .store = md_attr_store,
3569 static struct kobj_type md_ktype = {
3571 .sysfs_ops = &md_sysfs_ops,
3572 .default_attrs = md_default_attrs,
3577 static int md_alloc(dev_t dev, char *name)
3579 static DEFINE_MUTEX(disks_mutex);
3580 mddev_t *mddev = mddev_find(dev);
3581 struct gendisk *disk;
3590 partitioned = (MAJOR(mddev->unit) != MD_MAJOR);
3591 shift = partitioned ? MdpMinorShift : 0;
3592 unit = MINOR(mddev->unit) >> shift;
3594 /* wait for any previous instance if this device
3595 * to be completed removed (mddev_delayed_delete).
3597 flush_scheduled_work();
3599 mutex_lock(&disks_mutex);
3600 if (mddev->gendisk) {
3601 mutex_unlock(&disks_mutex);
3607 /* Need to ensure that 'name' is not a duplicate.
3610 spin_lock(&all_mddevs_lock);
3612 list_for_each_entry(mddev2, &all_mddevs, all_mddevs)
3613 if (mddev2->gendisk &&
3614 strcmp(mddev2->gendisk->disk_name, name) == 0) {
3615 spin_unlock(&all_mddevs_lock);
3618 spin_unlock(&all_mddevs_lock);
3621 mddev->queue = blk_alloc_queue(GFP_KERNEL);
3622 if (!mddev->queue) {
3623 mutex_unlock(&disks_mutex);
3627 /* Can be unlocked because the queue is new: no concurrency */
3628 queue_flag_set_unlocked(QUEUE_FLAG_CLUSTER, mddev->queue);
3630 blk_queue_make_request(mddev->queue, md_fail_request);
3632 disk = alloc_disk(1 << shift);
3634 mutex_unlock(&disks_mutex);
3635 blk_cleanup_queue(mddev->queue);
3636 mddev->queue = NULL;
3640 disk->major = MAJOR(mddev->unit);
3641 disk->first_minor = unit << shift;
3643 strcpy(disk->disk_name, name);
3644 else if (partitioned)
3645 sprintf(disk->disk_name, "md_d%d", unit);
3647 sprintf(disk->disk_name, "md%d", unit);
3648 disk->fops = &md_fops;
3649 disk->private_data = mddev;
3650 disk->queue = mddev->queue;
3651 /* Allow extended partitions. This makes the
3652 * 'mdp' device redundant, but we can't really
3655 disk->flags |= GENHD_FL_EXT_DEVT;
3657 mddev->gendisk = disk;
3658 error = kobject_init_and_add(&mddev->kobj, &md_ktype,
3659 &disk_to_dev(disk)->kobj, "%s", "md");
3660 mutex_unlock(&disks_mutex);
3662 printk(KERN_WARNING "md: cannot register %s/md - name in use\n",
3665 kobject_uevent(&mddev->kobj, KOBJ_ADD);
3666 mddev->sysfs_state = sysfs_get_dirent(mddev->kobj.sd, "array_state");
3672 static struct kobject *md_probe(dev_t dev, int *part, void *data)
3674 md_alloc(dev, NULL);
3678 static int add_named_array(const char *val, struct kernel_param *kp)
3680 /* val must be "md_*" where * is not all digits.
3681 * We allocate an array with a large free minor number, and
3682 * set the name to val. val must not already be an active name.
3684 int len = strlen(val);
3685 char buf[DISK_NAME_LEN];
3687 while (len && val[len-1] == '\n')
3689 if (len >= DISK_NAME_LEN)
3691 strlcpy(buf, val, len+1);
3692 if (strncmp(buf, "md_", 3) != 0)
3694 return md_alloc(0, buf);
3697 static void md_safemode_timeout(unsigned long data)
3699 mddev_t *mddev = (mddev_t *) data;
3701 if (!atomic_read(&mddev->writes_pending)) {
3702 mddev->safemode = 1;
3703 if (mddev->external)
3704 sysfs_notify_dirent(mddev->sysfs_state);
3706 md_wakeup_thread(mddev->thread);
3709 static int start_dirty_degraded;
3711 static int do_md_run(mddev_t * mddev)
3716 struct gendisk *disk;
3717 struct mdk_personality *pers;
3718 char b[BDEVNAME_SIZE];
3720 if (list_empty(&mddev->disks))
3721 /* cannot run an array with no devices.. */
3728 * Analyze all RAID superblock(s)
3730 if (!mddev->raid_disks) {
3731 if (!mddev->persistent)
3736 chunk_size = mddev->chunk_size;
3739 if (chunk_size > MAX_CHUNK_SIZE) {
3740 printk(KERN_ERR "too big chunk_size: %d > %d\n",
3741 chunk_size, MAX_CHUNK_SIZE);
3745 * chunk-size has to be a power of 2
3747 if ( (1 << ffz(~chunk_size)) != chunk_size) {
3748 printk(KERN_ERR "chunk_size of %d not valid\n", chunk_size);
3752 /* devices must have minimum size of one chunk */
3753 list_for_each_entry(rdev, &mddev->disks, same_set) {
3754 if (test_bit(Faulty, &rdev->flags))
3756 if (rdev->size < chunk_size / 1024) {
3758 "md: Dev %s smaller than chunk_size:"
3760 bdevname(rdev->bdev,b),
3761 (unsigned long long)rdev->size,
3768 if (mddev->level != LEVEL_NONE)
3769 request_module("md-level-%d", mddev->level);
3770 else if (mddev->clevel[0])
3771 request_module("md-%s", mddev->clevel);
3774 * Drop all container device buffers, from now on
3775 * the only valid external interface is through the md
3778 list_for_each_entry(rdev, &mddev->disks, same_set) {
3779 if (test_bit(Faulty, &rdev->flags))
3781 sync_blockdev(rdev->bdev);
3782 invalidate_bdev(rdev->bdev);
3784 /* perform some consistency tests on the device.
3785 * We don't want the data to overlap the metadata,
3786 * Internal Bitmap issues has handled elsewhere.
3788 if (rdev->data_offset < rdev->sb_start) {
3790 rdev->data_offset + mddev->size*2
3792 printk("md: %s: data overlaps metadata\n",
3797 if (rdev->sb_start + rdev->sb_size/512
3798 > rdev->data_offset) {
3799 printk("md: %s: metadata overlaps data\n",
3804 sysfs_notify_dirent(rdev->sysfs_state);
3807 md_probe(mddev->unit, NULL, NULL);
3808 disk = mddev->gendisk;
3812 spin_lock(&pers_lock);
3813 pers = find_pers(mddev->level, mddev->clevel);
3814 if (!pers || !try_module_get(pers->owner)) {
3815 spin_unlock(&pers_lock);
3816 if (mddev->level != LEVEL_NONE)
3817 printk(KERN_WARNING "md: personality for level %d is not loaded!\n",
3820 printk(KERN_WARNING "md: personality for level %s is not loaded!\n",
3825 spin_unlock(&pers_lock);
3826 mddev->level = pers->level;
3827 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
3829 if (pers->level >= 4 && pers->level <= 6)
3830 /* Cannot support integrity (yet) */
3831 blk_integrity_unregister(mddev->gendisk);
3833 if (mddev->reshape_position != MaxSector &&
3834 pers->start_reshape == NULL) {
3835 /* This personality cannot handle reshaping... */
3837 module_put(pers->owner);
3841 if (pers->sync_request) {
3842 /* Warn if this is a potentially silly
3845 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
3849 list_for_each_entry(rdev, &mddev->disks, same_set)
3850 list_for_each_entry(rdev2, &mddev->disks, same_set) {
3852 rdev->bdev->bd_contains ==
3853 rdev2->bdev->bd_contains) {
3855 "%s: WARNING: %s appears to be"
3856 " on the same physical disk as"
3859 bdevname(rdev->bdev,b),
3860 bdevname(rdev2->bdev,b2));
3867 "True protection against single-disk"
3868 " failure might be compromised.\n");
3871 mddev->recovery = 0;
3872 mddev->resync_max_sectors = mddev->size << 1; /* may be over-ridden by personality */
3873 mddev->barriers_work = 1;
3874 mddev->ok_start_degraded = start_dirty_degraded;
3877 mddev->ro = 2; /* read-only, but switch on first write */
3879 err = mddev->pers->run(mddev);
3881 printk(KERN_ERR "md: pers->run() failed ...\n");
3882 else if (mddev->pers->sync_request) {
3883 err = bitmap_create(mddev);
3885 printk(KERN_ERR "%s: failed to create bitmap (%d)\n",
3886 mdname(mddev), err);
3887 mddev->pers->stop(mddev);
3891 module_put(mddev->pers->owner);
3893 bitmap_destroy(mddev);
3896 if (mddev->pers->sync_request) {
3897 if (sysfs_create_group(&mddev->kobj, &md_redundancy_group))
3899 "md: cannot register extra attributes for %s\n",
3901 mddev->sysfs_action = sysfs_get_dirent(mddev->kobj.sd, "sync_action");
3902 } else if (mddev->ro == 2) /* auto-readonly not meaningful */
3905 atomic_set(&mddev->writes_pending,0);
3906 mddev->safemode = 0;
3907 mddev->safemode_timer.function = md_safemode_timeout;
3908 mddev->safemode_timer.data = (unsigned long) mddev;
3909 mddev->safemode_delay = (200 * HZ)/1000 +1; /* 200 msec delay */
3912 list_for_each_entry(rdev, &mddev->disks, same_set)
3913 if (rdev->raid_disk >= 0) {
3915 sprintf(nm, "rd%d", rdev->raid_disk);
3916 if (sysfs_create_link(&mddev->kobj, &rdev->kobj, nm))
3917 printk("md: cannot register %s for %s\n",
3921 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3924 md_update_sb(mddev, 0);
3926 set_capacity(disk, mddev->array_sectors);
3928 /* If we call blk_queue_make_request here, it will
3929 * re-initialise max_sectors etc which may have been
3930 * refined inside -> run. So just set the bits we need to set.
3931 * Most initialisation happended when we called
3932 * blk_queue_make_request(..., md_fail_request)
3935 mddev->queue->queuedata = mddev;
3936 mddev->queue->make_request_fn = mddev->pers->make_request;
3938 /* If there is a partially-recovered drive we need to
3939 * start recovery here. If we leave it to md_check_recovery,
3940 * it will remove the drives and not do the right thing
3942 if (mddev->degraded && !mddev->sync_thread) {
3944 list_for_each_entry(rdev, &mddev->disks, same_set)
3945 if (rdev->raid_disk >= 0 &&
3946 !test_bit(In_sync, &rdev->flags) &&
3947 !test_bit(Faulty, &rdev->flags))
3948 /* complete an interrupted recovery */
3950 if (spares && mddev->pers->sync_request) {
3951 mddev->recovery = 0;
3952 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
3953 mddev->sync_thread = md_register_thread(md_do_sync,
3956 if (!mddev->sync_thread) {
3957 printk(KERN_ERR "%s: could not start resync"
3960 /* leave the spares where they are, it shouldn't hurt */
3961 mddev->recovery = 0;
3965 md_wakeup_thread(mddev->thread);
3966 md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
3969 md_new_event(mddev);
3970 sysfs_notify_dirent(mddev->sysfs_state);
3971 if (mddev->sysfs_action)
3972 sysfs_notify_dirent(mddev->sysfs_action);
3973 sysfs_notify(&mddev->kobj, NULL, "degraded");
3974 kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
3978 static int restart_array(mddev_t *mddev)
3980 struct gendisk *disk = mddev->gendisk;
3982 /* Complain if it has no devices */
3983 if (list_empty(&mddev->disks))
3989 mddev->safemode = 0;
3991 set_disk_ro(disk, 0);
3992 printk(KERN_INFO "md: %s switched to read-write mode.\n",
3994 /* Kick recovery or resync if necessary */
3995 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3996 md_wakeup_thread(mddev->thread);
3997 md_wakeup_thread(mddev->sync_thread);
3998 sysfs_notify_dirent(mddev->sysfs_state);
4002 /* similar to deny_write_access, but accounts for our holding a reference
4003 * to the file ourselves */
4004 static int deny_bitmap_write_access(struct file * file)
4006 struct inode *inode = file->f_mapping->host;
4008 spin_lock(&inode->i_lock);
4009 if (atomic_read(&inode->i_writecount) > 1) {
4010 spin_unlock(&inode->i_lock);
4013 atomic_set(&inode->i_writecount, -1);
4014 spin_unlock(&inode->i_lock);
4019 static void restore_bitmap_write_access(struct file *file)
4021 struct inode *inode = file->f_mapping->host;
4023 spin_lock(&inode->i_lock);
4024 atomic_set(&inode->i_writecount, 1);
4025 spin_unlock(&inode->i_lock);
4029 * 0 - completely stop and dis-assemble array
4030 * 1 - switch to readonly
4031 * 2 - stop but do not disassemble array
4033 static int do_md_stop(mddev_t * mddev, int mode, int is_open)
4036 struct gendisk *disk = mddev->gendisk;
4038 if (atomic_read(&mddev->openers) > is_open) {
4039 printk("md: %s still in use.\n",mdname(mddev));
4045 if (mddev->sync_thread) {
4046 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4047 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
4048 md_unregister_thread(mddev->sync_thread);
4049 mddev->sync_thread = NULL;
4052 del_timer_sync(&mddev->safemode_timer);
4055 case 1: /* readonly */
4061 case 0: /* disassemble */
4063 bitmap_flush(mddev);
4064 md_super_wait(mddev);
4066 set_disk_ro(disk, 0);
4067 blk_queue_make_request(mddev->queue, md_fail_request);
4068 mddev->pers->stop(mddev);
4069 mddev->queue->merge_bvec_fn = NULL;
4070 mddev->queue->unplug_fn = NULL;
4071 mddev->queue->backing_dev_info.congested_fn = NULL;
4072 if (mddev->pers->sync_request) {
4073 sysfs_remove_group(&mddev->kobj, &md_redundancy_group);
4074 if (mddev->sysfs_action)
4075 sysfs_put(mddev->sysfs_action);
4076 mddev->sysfs_action = NULL;
4078 module_put(mddev->pers->owner);
4080 /* tell userspace to handle 'inactive' */
4081 sysfs_notify_dirent(mddev->sysfs_state);
4083 set_capacity(disk, 0);
4089 if (!mddev->in_sync || mddev->flags) {
4090 /* mark array as shutdown cleanly */
4092 md_update_sb(mddev, 1);
4095 set_disk_ro(disk, 1);
4096 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4100 * Free resources if final stop
4105 printk(KERN_INFO "md: %s stopped.\n", mdname(mddev));
4107 bitmap_destroy(mddev);
4108 if (mddev->bitmap_file) {
4109 restore_bitmap_write_access(mddev->bitmap_file);
4110 fput(mddev->bitmap_file);
4111 mddev->bitmap_file = NULL;
4113 mddev->bitmap_offset = 0;
4115 list_for_each_entry(rdev, &mddev->disks, same_set)
4116 if (rdev->raid_disk >= 0) {
4118 sprintf(nm, "rd%d", rdev->raid_disk);
4119 sysfs_remove_link(&mddev->kobj, nm);
4122 /* make sure all md_delayed_delete calls have finished */
4123 flush_scheduled_work();
4125 export_array(mddev);
4127 mddev->array_sectors = 0;
4129 mddev->raid_disks = 0;
4130 mddev->recovery_cp = 0;
4131 mddev->resync_min = 0;
4132 mddev->resync_max = MaxSector;
4133 mddev->reshape_position = MaxSector;
4134 mddev->external = 0;
4135 mddev->persistent = 0;
4136 mddev->level = LEVEL_NONE;
4137 mddev->clevel[0] = 0;
4140 mddev->metadata_type[0] = 0;
4141 mddev->chunk_size = 0;
4142 mddev->ctime = mddev->utime = 0;
4144 mddev->max_disks = 0;
4146 mddev->delta_disks = 0;
4147 mddev->new_level = LEVEL_NONE;
4148 mddev->new_layout = 0;
4149 mddev->new_chunk = 0;
4150 mddev->curr_resync = 0;
4151 mddev->resync_mismatches = 0;
4152 mddev->suspend_lo = mddev->suspend_hi = 0;
4153 mddev->sync_speed_min = mddev->sync_speed_max = 0;
4154 mddev->recovery = 0;
4157 mddev->degraded = 0;
4158 mddev->barriers_work = 0;
4159 mddev->safemode = 0;
4160 kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
4161 if (mddev->hold_active == UNTIL_STOP)
4162 mddev->hold_active = 0;
4164 } else if (mddev->pers)
4165 printk(KERN_INFO "md: %s switched to read-only mode.\n",
4168 blk_integrity_unregister(disk);
4169 md_new_event(mddev);
4170 sysfs_notify_dirent(mddev->sysfs_state);
4176 static void autorun_array(mddev_t *mddev)
4181 if (list_empty(&mddev->disks))
4184 printk(KERN_INFO "md: running: ");
4186 list_for_each_entry(rdev, &mddev->disks, same_set) {
4187 char b[BDEVNAME_SIZE];
4188 printk("<%s>", bdevname(rdev->bdev,b));
4192 err = do_md_run(mddev);
4194 printk(KERN_WARNING "md: do_md_run() returned %d\n", err);
4195 do_md_stop(mddev, 0, 0);
4200 * lets try to run arrays based on all disks that have arrived
4201 * until now. (those are in pending_raid_disks)
4203 * the method: pick the first pending disk, collect all disks with
4204 * the same UUID, remove all from the pending list and put them into
4205 * the 'same_array' list. Then order this list based on superblock
4206 * update time (freshest comes first), kick out 'old' disks and
4207 * compare superblocks. If everything's fine then run it.
4209 * If "unit" is allocated, then bump its reference count
4211 static void autorun_devices(int part)
4213 mdk_rdev_t *rdev0, *rdev, *tmp;
4215 char b[BDEVNAME_SIZE];
4217 printk(KERN_INFO "md: autorun ...\n");
4218 while (!list_empty(&pending_raid_disks)) {
4221 LIST_HEAD(candidates);
4222 rdev0 = list_entry(pending_raid_disks.next,
4223 mdk_rdev_t, same_set);
4225 printk(KERN_INFO "md: considering %s ...\n",
4226 bdevname(rdev0->bdev,b));
4227 INIT_LIST_HEAD(&candidates);
4228 rdev_for_each_list(rdev, tmp, &pending_raid_disks)
4229 if (super_90_load(rdev, rdev0, 0) >= 0) {
4230 printk(KERN_INFO "md: adding %s ...\n",
4231 bdevname(rdev->bdev,b));
4232 list_move(&rdev->same_set, &candidates);
4235 * now we have a set of devices, with all of them having
4236 * mostly sane superblocks. It's time to allocate the
4240 dev = MKDEV(mdp_major,
4241 rdev0->preferred_minor << MdpMinorShift);
4242 unit = MINOR(dev) >> MdpMinorShift;
4244 dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
4247 if (rdev0->preferred_minor != unit) {
4248 printk(KERN_INFO "md: unit number in %s is bad: %d\n",
4249 bdevname(rdev0->bdev, b), rdev0->preferred_minor);
4253 md_probe(dev, NULL, NULL);
4254 mddev = mddev_find(dev);
4255 if (!mddev || !mddev->gendisk) {
4259 "md: cannot allocate memory for md drive.\n");
4262 if (mddev_lock(mddev))
4263 printk(KERN_WARNING "md: %s locked, cannot run\n",
4265 else if (mddev->raid_disks || mddev->major_version
4266 || !list_empty(&mddev->disks)) {
4268 "md: %s already running, cannot run %s\n",
4269 mdname(mddev), bdevname(rdev0->bdev,b));
4270 mddev_unlock(mddev);
4272 printk(KERN_INFO "md: created %s\n", mdname(mddev));
4273 mddev->persistent = 1;
4274 rdev_for_each_list(rdev, tmp, &candidates) {
4275 list_del_init(&rdev->same_set);
4276 if (bind_rdev_to_array(rdev, mddev))
4279 autorun_array(mddev);
4280 mddev_unlock(mddev);
4282 /* on success, candidates will be empty, on error
4285 rdev_for_each_list(rdev, tmp, &candidates) {
4286 list_del_init(&rdev->same_set);
4291 printk(KERN_INFO "md: ... autorun DONE.\n");
4293 #endif /* !MODULE */
4295 static int get_version(void __user * arg)
4299 ver.major = MD_MAJOR_VERSION;
4300 ver.minor = MD_MINOR_VERSION;
4301 ver.patchlevel = MD_PATCHLEVEL_VERSION;
4303 if (copy_to_user(arg, &ver, sizeof(ver)))
4309 static int get_array_info(mddev_t * mddev, void __user * arg)
4311 mdu_array_info_t info;
4312 int nr,working,active,failed,spare;
4315 nr=working=active=failed=spare=0;
4316 list_for_each_entry(rdev, &mddev->disks, same_set) {
4318 if (test_bit(Faulty, &rdev->flags))
4322 if (test_bit(In_sync, &rdev->flags))
4329 info.major_version = mddev->major_version;
4330 info.minor_version = mddev->minor_version;
4331 info.patch_version = MD_PATCHLEVEL_VERSION;
4332 info.ctime = mddev->ctime;
4333 info.level = mddev->level;
4334 info.size = mddev->size;
4335 if (info.size != mddev->size) /* overflow */
4338 info.raid_disks = mddev->raid_disks;
4339 info.md_minor = mddev->md_minor;
4340 info.not_persistent= !mddev->persistent;
4342 info.utime = mddev->utime;
4345 info.state = (1<<MD_SB_CLEAN);
4346 if (mddev->bitmap && mddev->bitmap_offset)
4347 info.state = (1<<MD_SB_BITMAP_PRESENT);
4348 info.active_disks = active;
4349 info.working_disks = working;
4350 info.failed_disks = failed;
4351 info.spare_disks = spare;
4353 info.layout = mddev->layout;
4354 info.chunk_size = mddev->chunk_size;
4356 if (copy_to_user(arg, &info, sizeof(info)))
4362 static int get_bitmap_file(mddev_t * mddev, void __user * arg)
4364 mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */
4365 char *ptr, *buf = NULL;
4368 if (md_allow_write(mddev))
4369 file = kmalloc(sizeof(*file), GFP_NOIO);
4371 file = kmalloc(sizeof(*file), GFP_KERNEL);
4376 /* bitmap disabled, zero the first byte and copy out */
4377 if (!mddev->bitmap || !mddev->bitmap->file) {
4378 file->pathname[0] = '\0';
4382 buf = kmalloc(sizeof(file->pathname), GFP_KERNEL);
4386 ptr = d_path(&mddev->bitmap->file->f_path, buf, sizeof(file->pathname));
4390 strcpy(file->pathname, ptr);
4394 if (copy_to_user(arg, file, sizeof(*file)))
4402 static int get_disk_info(mddev_t * mddev, void __user * arg)
4404 mdu_disk_info_t info;
4407 if (copy_from_user(&info, arg, sizeof(info)))
4410 rdev = find_rdev_nr(mddev, info.number);
4412 info.major = MAJOR(rdev->bdev->bd_dev);
4413 info.minor = MINOR(rdev->bdev->bd_dev);
4414 info.raid_disk = rdev->raid_disk;
4416 if (test_bit(Faulty, &rdev->flags))
4417 info.state |= (1<<MD_DISK_FAULTY);
4418 else if (test_bit(In_sync, &rdev->flags)) {
4419 info.state |= (1<<MD_DISK_ACTIVE);
4420 info.state |= (1<<MD_DISK_SYNC);
4422 if (test_bit(WriteMostly, &rdev->flags))
4423 info.state |= (1<<MD_DISK_WRITEMOSTLY);
4425 info.major = info.minor = 0;
4426 info.raid_disk = -1;
4427 info.state = (1<<MD_DISK_REMOVED);
4430 if (copy_to_user(arg, &info, sizeof(info)))
4436 static int add_new_disk(mddev_t * mddev, mdu_disk_info_t *info)
4438 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
4440 dev_t dev = MKDEV(info->major,info->minor);
4442 if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
4445 if (!mddev->raid_disks) {
4447 /* expecting a device which has a superblock */
4448 rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
4451 "md: md_import_device returned %ld\n",
4453 return PTR_ERR(rdev);
4455 if (!list_empty(&mddev->disks)) {
4456 mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
4457 mdk_rdev_t, same_set);
4458 int err = super_types[mddev->major_version]
4459 .load_super(rdev, rdev0, mddev->minor_version);
4462 "md: %s has different UUID to %s\n",
4463 bdevname(rdev->bdev,b),
4464 bdevname(rdev0->bdev,b2));
4469 err = bind_rdev_to_array(rdev, mddev);
4476 * add_new_disk can be used once the array is assembled
4477 * to add "hot spares". They must already have a superblock
4482 if (!mddev->pers->hot_add_disk) {
4484 "%s: personality does not support diskops!\n",
4488 if (mddev->persistent)
4489 rdev = md_import_device(dev, mddev->major_version,
4490 mddev->minor_version);
4492 rdev = md_import_device(dev, -1, -1);
4495 "md: md_import_device returned %ld\n",
4497 return PTR_ERR(rdev);
4499 /* set save_raid_disk if appropriate */
4500 if (!mddev->persistent) {
4501 if (info->state & (1<<MD_DISK_SYNC) &&
4502 info->raid_disk < mddev->raid_disks)
4503 rdev->raid_disk = info->raid_disk;
4505 rdev->raid_disk = -1;
4507 super_types[mddev->major_version].
4508 validate_super(mddev, rdev);
4509 rdev->saved_raid_disk = rdev->raid_disk;
4511 clear_bit(In_sync, &rdev->flags); /* just to be sure */
4512 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
4513 set_bit(WriteMostly, &rdev->flags);
4515 rdev->raid_disk = -1;
4516 err = bind_rdev_to_array(rdev, mddev);
4517 if (!err && !mddev->pers->hot_remove_disk) {
4518 /* If there is hot_add_disk but no hot_remove_disk
4519 * then added disks for geometry changes,
4520 * and should be added immediately.
4522 super_types[mddev->major_version].
4523 validate_super(mddev, rdev);
4524 err = mddev->pers->hot_add_disk(mddev, rdev);
4526 unbind_rdev_from_array(rdev);
4531 sysfs_notify_dirent(rdev->sysfs_state);
4533 md_update_sb(mddev, 1);
4534 if (mddev->degraded)
4535 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
4536 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4537 md_wakeup_thread(mddev->thread);
4541 /* otherwise, add_new_disk is only allowed
4542 * for major_version==0 superblocks
4544 if (mddev->major_version != 0) {
4545 printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n",
4550 if (!(info->state & (1<<MD_DISK_FAULTY))) {
4552 rdev = md_import_device(dev, -1, 0);
4555 "md: error, md_import_device() returned %ld\n",
4557 return PTR_ERR(rdev);
4559 rdev->desc_nr = info->number;
4560 if (info->raid_disk < mddev->raid_disks)
4561 rdev->raid_disk = info->raid_disk;
4563 rdev->raid_disk = -1;
4565 if (rdev->raid_disk < mddev->raid_disks)
4566 if (info->state & (1<<MD_DISK_SYNC))
4567 set_bit(In_sync, &rdev->flags);
4569 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
4570 set_bit(WriteMostly, &rdev->flags);
4572 if (!mddev->persistent) {
4573 printk(KERN_INFO "md: nonpersistent superblock ...\n");
4574 rdev->sb_start = rdev->bdev->bd_inode->i_size / 512;
4576 rdev->sb_start = calc_dev_sboffset(rdev->bdev);
4577 rdev->size = calc_num_sectors(rdev, mddev->chunk_size) / 2;
4579 err = bind_rdev_to_array(rdev, mddev);
4589 static int hot_remove_disk(mddev_t * mddev, dev_t dev)
4591 char b[BDEVNAME_SIZE];
4594 rdev = find_rdev(mddev, dev);
4598 if (rdev->raid_disk >= 0)
4601 kick_rdev_from_array(rdev);
4602 md_update_sb(mddev, 1);
4603 md_new_event(mddev);
4607 printk(KERN_WARNING "md: cannot remove active disk %s from %s ...\n",
4608 bdevname(rdev->bdev,b), mdname(mddev));
4612 static int hot_add_disk(mddev_t * mddev, dev_t dev)
4614 char b[BDEVNAME_SIZE];
4621 if (mddev->major_version != 0) {
4622 printk(KERN_WARNING "%s: HOT_ADD may only be used with"
4623 " version-0 superblocks.\n",
4627 if (!mddev->pers->hot_add_disk) {
4629 "%s: personality does not support diskops!\n",
4634 rdev = md_import_device(dev, -1, 0);
4637 "md: error, md_import_device() returned %ld\n",
4642 if (mddev->persistent)
4643 rdev->sb_start = calc_dev_sboffset(rdev->bdev);
4645 rdev->sb_start = rdev->bdev->bd_inode->i_size / 512;
4647 rdev->size = calc_num_sectors(rdev, mddev->chunk_size) / 2;
4649 if (test_bit(Faulty, &rdev->flags)) {
4651 "md: can not hot-add faulty %s disk to %s!\n",
4652 bdevname(rdev->bdev,b), mdname(mddev));
4656 clear_bit(In_sync, &rdev->flags);
4658 rdev->saved_raid_disk = -1;
4659 err = bind_rdev_to_array(rdev, mddev);
4664 * The rest should better be atomic, we can have disk failures
4665 * noticed in interrupt contexts ...
4668 rdev->raid_disk = -1;
4670 md_update_sb(mddev, 1);
4673 * Kick recovery, maybe this spare has to be added to the
4674 * array immediately.
4676 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4677 md_wakeup_thread(mddev->thread);
4678 md_new_event(mddev);
4686 static int set_bitmap_file(mddev_t *mddev, int fd)
4691 if (!mddev->pers->quiesce)
4693 if (mddev->recovery || mddev->sync_thread)
4695 /* we should be able to change the bitmap.. */
4701 return -EEXIST; /* cannot add when bitmap is present */
4702 mddev->bitmap_file = fget(fd);
4704 if (mddev->bitmap_file == NULL) {
4705 printk(KERN_ERR "%s: error: failed to get bitmap file\n",
4710 err = deny_bitmap_write_access(mddev->bitmap_file);
4712 printk(KERN_ERR "%s: error: bitmap file is already in use\n",
4714 fput(mddev->bitmap_file);
4715 mddev->bitmap_file = NULL;
4718 mddev->bitmap_offset = 0; /* file overrides offset */
4719 } else if (mddev->bitmap == NULL)
4720 return -ENOENT; /* cannot remove what isn't there */
4723 mddev->pers->quiesce(mddev, 1);
4725 err = bitmap_create(mddev);
4726 if (fd < 0 || err) {
4727 bitmap_destroy(mddev);
4728 fd = -1; /* make sure to put the file */
4730 mddev->pers->quiesce(mddev, 0);
4733 if (mddev->bitmap_file) {
4734 restore_bitmap_write_access(mddev->bitmap_file);
4735 fput(mddev->bitmap_file);
4737 mddev->bitmap_file = NULL;
4744 * set_array_info is used two different ways
4745 * The original usage is when creating a new array.
4746 * In this usage, raid_disks is > 0 and it together with
4747 * level, size, not_persistent,layout,chunksize determine the
4748 * shape of the array.
4749 * This will always create an array with a type-0.90.0 superblock.
4750 * The newer usage is when assembling an array.
4751 * In this case raid_disks will be 0, and the major_version field is
4752 * use to determine which style super-blocks are to be found on the devices.
4753 * The minor and patch _version numbers are also kept incase the
4754 * super_block handler wishes to interpret them.
4756 static int set_array_info(mddev_t * mddev, mdu_array_info_t *info)
4759 if (info->raid_disks == 0) {
4760 /* just setting version number for superblock loading */
4761 if (info->major_version < 0 ||
4762 info->major_version >= ARRAY_SIZE(super_types) ||
4763 super_types[info->major_version].name == NULL) {
4764 /* maybe try to auto-load a module? */
4766 "md: superblock version %d not known\n",
4767 info->major_version);
4770 mddev->major_version = info->major_version;
4771 mddev->minor_version = info->minor_version;
4772 mddev->patch_version = info->patch_version;
4773 mddev->persistent = !info->not_persistent;
4776 mddev->major_version = MD_MAJOR_VERSION;
4777 mddev->minor_version = MD_MINOR_VERSION;
4778 mddev->patch_version = MD_PATCHLEVEL_VERSION;
4779 mddev->ctime = get_seconds();
4781 mddev->level = info->level;
4782 mddev->clevel[0] = 0;
4783 mddev->size = info->size;
4784 mddev->raid_disks = info->raid_disks;
4785 /* don't set md_minor, it is determined by which /dev/md* was
4788 if (info->state & (1<<MD_SB_CLEAN))
4789 mddev->recovery_cp = MaxSector;
4791 mddev->recovery_cp = 0;
4792 mddev->persistent = ! info->not_persistent;
4793 mddev->external = 0;
4795 mddev->layout = info->layout;
4796 mddev->chunk_size = info->chunk_size;
4798 mddev->max_disks = MD_SB_DISKS;
4800 if (mddev->persistent)
4802 set_bit(MD_CHANGE_DEVS, &mddev->flags);
4804 mddev->default_bitmap_offset = MD_SB_BYTES >> 9;
4805 mddev->bitmap_offset = 0;
4807 mddev->reshape_position = MaxSector;
4810 * Generate a 128 bit UUID
4812 get_random_bytes(mddev->uuid, 16);
4814 mddev->new_level = mddev->level;
4815 mddev->new_chunk = mddev->chunk_size;
4816 mddev->new_layout = mddev->layout;
4817 mddev->delta_disks = 0;
4822 static int update_size(mddev_t *mddev, sector_t num_sectors)
4826 int fit = (num_sectors == 0);
4828 if (mddev->pers->resize == NULL)
4830 /* The "num_sectors" is the number of sectors of each device that
4831 * is used. This can only make sense for arrays with redundancy.
4832 * linear and raid0 always use whatever space is available. We can only
4833 * consider changing this number if no resync or reconstruction is
4834 * happening, and if the new size is acceptable. It must fit before the
4835 * sb_start or, if that is <data_offset, it must fit before the size
4836 * of each device. If num_sectors is zero, we find the largest size
4840 if (mddev->sync_thread)
4843 /* Sorry, cannot grow a bitmap yet, just remove it,
4847 list_for_each_entry(rdev, &mddev->disks, same_set) {
4849 avail = rdev->size * 2;
4851 if (fit && (num_sectors == 0 || num_sectors > avail))
4852 num_sectors = avail;
4853 if (avail < num_sectors)
4856 rv = mddev->pers->resize(mddev, num_sectors);
4858 struct block_device *bdev;
4860 bdev = bdget_disk(mddev->gendisk, 0);
4862 mutex_lock(&bdev->bd_inode->i_mutex);
4863 i_size_write(bdev->bd_inode,
4864 (loff_t)mddev->array_sectors << 9);
4865 mutex_unlock(&bdev->bd_inode->i_mutex);
4872 static int update_raid_disks(mddev_t *mddev, int raid_disks)
4875 /* change the number of raid disks */
4876 if (mddev->pers->check_reshape == NULL)
4878 if (raid_disks <= 0 ||
4879 raid_disks >= mddev->max_disks)
4881 if (mddev->sync_thread || mddev->reshape_position != MaxSector)
4883 mddev->delta_disks = raid_disks - mddev->raid_disks;
4885 rv = mddev->pers->check_reshape(mddev);
4891 * update_array_info is used to change the configuration of an
4893 * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
4894 * fields in the info are checked against the array.
4895 * Any differences that cannot be handled will cause an error.
4896 * Normally, only one change can be managed at a time.
4898 static int update_array_info(mddev_t *mddev, mdu_array_info_t *info)
4904 /* calculate expected state,ignoring low bits */
4905 if (mddev->bitmap && mddev->bitmap_offset)
4906 state |= (1 << MD_SB_BITMAP_PRESENT);
4908 if (mddev->major_version != info->major_version ||
4909 mddev->minor_version != info->minor_version ||
4910 /* mddev->patch_version != info->patch_version || */
4911 mddev->ctime != info->ctime ||
4912 mddev->level != info->level ||
4913 /* mddev->layout != info->layout || */
4914 !mddev->persistent != info->not_persistent||
4915 mddev->chunk_size != info->chunk_size ||
4916 /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
4917 ((state^info->state) & 0xfffffe00)
4920 /* Check there is only one change */
4921 if (info->size >= 0 && mddev->size != info->size) cnt++;
4922 if (mddev->raid_disks != info->raid_disks) cnt++;
4923 if (mddev->layout != info->layout) cnt++;
4924 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) cnt++;
4925 if (cnt == 0) return 0;
4926 if (cnt > 1) return -EINVAL;
4928 if (mddev->layout != info->layout) {
4930 * we don't need to do anything at the md level, the
4931 * personality will take care of it all.
4933 if (mddev->pers->reconfig == NULL)
4936 return mddev->pers->reconfig(mddev, info->layout, -1);
4938 if (info->size >= 0 && mddev->size != info->size)
4939 rv = update_size(mddev, (sector_t)info->size * 2);
4941 if (mddev->raid_disks != info->raid_disks)
4942 rv = update_raid_disks(mddev, info->raid_disks);
4944 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) {
4945 if (mddev->pers->quiesce == NULL)
4947 if (mddev->recovery || mddev->sync_thread)
4949 if (info->state & (1<<MD_SB_BITMAP_PRESENT)) {
4950 /* add the bitmap */
4953 if (mddev->default_bitmap_offset == 0)
4955 mddev->bitmap_offset = mddev->default_bitmap_offset;
4956 mddev->pers->quiesce(mddev, 1);
4957 rv = bitmap_create(mddev);
4959 bitmap_destroy(mddev);
4960 mddev->pers->quiesce(mddev, 0);
4962 /* remove the bitmap */
4965 if (mddev->bitmap->file)
4967 mddev->pers->quiesce(mddev, 1);
4968 bitmap_destroy(mddev);
4969 mddev->pers->quiesce(mddev, 0);
4970 mddev->bitmap_offset = 0;
4973 md_update_sb(mddev, 1);
4977 static int set_disk_faulty(mddev_t *mddev, dev_t dev)
4981 if (mddev->pers == NULL)
4984 rdev = find_rdev(mddev, dev);
4988 md_error(mddev, rdev);
4993 * We have a problem here : there is no easy way to give a CHS
4994 * virtual geometry. We currently pretend that we have a 2 heads
4995 * 4 sectors (with a BIG number of cylinders...). This drives
4996 * dosfs just mad... ;-)
4998 static int md_getgeo(struct block_device *bdev, struct hd_geometry *geo)
5000 mddev_t *mddev = bdev->bd_disk->private_data;
5004 geo->cylinders = get_capacity(mddev->gendisk) / 8;
5008 static int md_ioctl(struct block_device *bdev, fmode_t mode,
5009 unsigned int cmd, unsigned long arg)
5012 void __user *argp = (void __user *)arg;
5013 mddev_t *mddev = NULL;
5015 if (!capable(CAP_SYS_ADMIN))
5019 * Commands dealing with the RAID driver but not any
5025 err = get_version(argp);
5028 case PRINT_RAID_DEBUG:
5036 autostart_arrays(arg);
5043 * Commands creating/starting a new array:
5046 mddev = bdev->bd_disk->private_data;
5053 err = mddev_lock(mddev);
5056 "md: ioctl lock interrupted, reason %d, cmd %d\n",
5063 case SET_ARRAY_INFO:
5065 mdu_array_info_t info;
5067 memset(&info, 0, sizeof(info));
5068 else if (copy_from_user(&info, argp, sizeof(info))) {
5073 err = update_array_info(mddev, &info);
5075 printk(KERN_WARNING "md: couldn't update"
5076 " array info. %d\n", err);
5081 if (!list_empty(&mddev->disks)) {
5083 "md: array %s already has disks!\n",
5088 if (mddev->raid_disks) {
5090 "md: array %s already initialised!\n",
5095 err = set_array_info(mddev, &info);
5097 printk(KERN_WARNING "md: couldn't set"
5098 " array info. %d\n", err);
5108 * Commands querying/configuring an existing array:
5110 /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
5111 * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */
5112 if ((!mddev->raid_disks && !mddev->external)
5113 && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY
5114 && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE
5115 && cmd != GET_BITMAP_FILE) {
5121 * Commands even a read-only array can execute:
5125 case GET_ARRAY_INFO:
5126 err = get_array_info(mddev, argp);
5129 case GET_BITMAP_FILE:
5130 err = get_bitmap_file(mddev, argp);
5134 err = get_disk_info(mddev, argp);
5137 case RESTART_ARRAY_RW:
5138 err = restart_array(mddev);
5142 err = do_md_stop(mddev, 0, 1);
5146 err = do_md_stop(mddev, 1, 1);
5152 * The remaining ioctls are changing the state of the
5153 * superblock, so we do not allow them on read-only arrays.
5154 * However non-MD ioctls (e.g. get-size) will still come through
5155 * here and hit the 'default' below, so only disallow
5156 * 'md' ioctls, and switch to rw mode if started auto-readonly.
5158 if (_IOC_TYPE(cmd) == MD_MAJOR && mddev->ro && mddev->pers) {
5159 if (mddev->ro == 2) {
5161 sysfs_notify_dirent(mddev->sysfs_state);
5162 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5163 md_wakeup_thread(mddev->thread);
5174 mdu_disk_info_t info;
5175 if (copy_from_user(&info, argp, sizeof(info)))
5178 err = add_new_disk(mddev, &info);
5182 case HOT_REMOVE_DISK:
5183 err = hot_remove_disk(mddev, new_decode_dev(arg));
5187 err = hot_add_disk(mddev, new_decode_dev(arg));
5190 case SET_DISK_FAULTY:
5191 err = set_disk_faulty(mddev, new_decode_dev(arg));
5195 err = do_md_run(mddev);
5198 case SET_BITMAP_FILE:
5199 err = set_bitmap_file(mddev, (int)arg);
5209 if (mddev->hold_active == UNTIL_IOCTL &&
5211 mddev->hold_active = 0;
5212 mddev_unlock(mddev);
5222 static int md_open(struct block_device *bdev, fmode_t mode)
5225 * Succeed if we can lock the mddev, which confirms that
5226 * it isn't being stopped right now.
5228 mddev_t *mddev = mddev_find(bdev->bd_dev);
5231 if (mddev->gendisk != bdev->bd_disk) {
5232 /* we are racing with mddev_put which is discarding this
5236 /* Wait until bdev->bd_disk is definitely gone */
5237 flush_scheduled_work();
5238 /* Then retry the open from the top */
5239 return -ERESTARTSYS;
5241 BUG_ON(mddev != bdev->bd_disk->private_data);
5243 if ((err = mutex_lock_interruptible_nested(&mddev->reconfig_mutex, 1)))
5247 atomic_inc(&mddev->openers);
5248 mddev_unlock(mddev);
5250 check_disk_change(bdev);
5255 static int md_release(struct gendisk *disk, fmode_t mode)
5257 mddev_t *mddev = disk->private_data;
5260 atomic_dec(&mddev->openers);
5266 static int md_media_changed(struct gendisk *disk)
5268 mddev_t *mddev = disk->private_data;
5270 return mddev->changed;
5273 static int md_revalidate(struct gendisk *disk)
5275 mddev_t *mddev = disk->private_data;
5280 static struct block_device_operations md_fops =
5282 .owner = THIS_MODULE,
5284 .release = md_release,
5285 .locked_ioctl = md_ioctl,
5286 .getgeo = md_getgeo,
5287 .media_changed = md_media_changed,
5288 .revalidate_disk= md_revalidate,
5291 static int md_thread(void * arg)
5293 mdk_thread_t *thread = arg;
5296 * md_thread is a 'system-thread', it's priority should be very
5297 * high. We avoid resource deadlocks individually in each
5298 * raid personality. (RAID5 does preallocation) We also use RR and
5299 * the very same RT priority as kswapd, thus we will never get
5300 * into a priority inversion deadlock.
5302 * we definitely have to have equal or higher priority than
5303 * bdflush, otherwise bdflush will deadlock if there are too
5304 * many dirty RAID5 blocks.
5307 allow_signal(SIGKILL);
5308 while (!kthread_should_stop()) {
5310 /* We need to wait INTERRUPTIBLE so that
5311 * we don't add to the load-average.
5312 * That means we need to be sure no signals are
5315 if (signal_pending(current))
5316 flush_signals(current);
5318 wait_event_interruptible_timeout
5320 test_bit(THREAD_WAKEUP, &thread->flags)
5321 || kthread_should_stop(),
5324 clear_bit(THREAD_WAKEUP, &thread->flags);
5326 thread->run(thread->mddev);
5332 void md_wakeup_thread(mdk_thread_t *thread)
5335 dprintk("md: waking up MD thread %s.\n", thread->tsk->comm);
5336 set_bit(THREAD_WAKEUP, &thread->flags);
5337 wake_up(&thread->wqueue);
5341 mdk_thread_t *md_register_thread(void (*run) (mddev_t *), mddev_t *mddev,
5344 mdk_thread_t *thread;
5346 thread = kzalloc(sizeof(mdk_thread_t), GFP_KERNEL);
5350 init_waitqueue_head(&thread->wqueue);
5353 thread->mddev = mddev;
5354 thread->timeout = MAX_SCHEDULE_TIMEOUT;
5355 thread->tsk = kthread_run(md_thread, thread, name, mdname(thread->mddev));
5356 if (IS_ERR(thread->tsk)) {
5363 void md_unregister_thread(mdk_thread_t *thread)
5365 dprintk("interrupting MD-thread pid %d\n", task_pid_nr(thread->tsk));
5367 kthread_stop(thread->tsk);
5371 void md_error(mddev_t *mddev, mdk_rdev_t *rdev)
5378 if (!rdev || test_bit(Faulty, &rdev->flags))
5381 if (mddev->external)
5382 set_bit(Blocked, &rdev->flags);
5384 dprintk("md_error dev:%s, rdev:(%d:%d), (caller: %p,%p,%p,%p).\n",
5386 MAJOR(rdev->bdev->bd_dev), MINOR(rdev->bdev->bd_dev),
5387 __builtin_return_address(0),__builtin_return_address(1),
5388 __builtin_return_address(2),__builtin_return_address(3));
5392 if (!mddev->pers->error_handler)
5394 mddev->pers->error_handler(mddev,rdev);
5395 if (mddev->degraded)
5396 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
5397 set_bit(StateChanged, &rdev->flags);
5398 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5399 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5400 md_wakeup_thread(mddev->thread);
5401 md_new_event_inintr(mddev);
5404 /* seq_file implementation /proc/mdstat */
5406 static void status_unused(struct seq_file *seq)
5411 seq_printf(seq, "unused devices: ");
5413 list_for_each_entry(rdev, &pending_raid_disks, same_set) {
5414 char b[BDEVNAME_SIZE];
5416 seq_printf(seq, "%s ",
5417 bdevname(rdev->bdev,b));
5420 seq_printf(seq, "<none>");
5422 seq_printf(seq, "\n");
5426 static void status_resync(struct seq_file *seq, mddev_t * mddev)
5428 sector_t max_blocks, resync, res;
5429 unsigned long dt, db, rt;
5431 unsigned int per_milli;
5433 resync = (mddev->curr_resync - atomic_read(&mddev->recovery_active))/2;
5435 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
5436 max_blocks = mddev->resync_max_sectors >> 1;
5438 max_blocks = mddev->size;
5441 * Should not happen.
5447 /* Pick 'scale' such that (resync>>scale)*1000 will fit
5448 * in a sector_t, and (max_blocks>>scale) will fit in a
5449 * u32, as those are the requirements for sector_div.
5450 * Thus 'scale' must be at least 10
5453 if (sizeof(sector_t) > sizeof(unsigned long)) {
5454 while ( max_blocks/2 > (1ULL<<(scale+32)))
5457 res = (resync>>scale)*1000;
5458 sector_div(res, (u32)((max_blocks>>scale)+1));
5462 int i, x = per_milli/50, y = 20-x;
5463 seq_printf(seq, "[");
5464 for (i = 0; i < x; i++)
5465 seq_printf(seq, "=");
5466 seq_printf(seq, ">");
5467 for (i = 0; i < y; i++)
5468 seq_printf(seq, ".");
5469 seq_printf(seq, "] ");
5471 seq_printf(seq, " %s =%3u.%u%% (%llu/%llu)",
5472 (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)?
5474 (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)?
5476 (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
5477 "resync" : "recovery"))),
5478 per_milli/10, per_milli % 10,
5479 (unsigned long long) resync,
5480 (unsigned long long) max_blocks);
5483 * We do not want to overflow, so the order of operands and
5484 * the * 100 / 100 trick are important. We do a +1 to be
5485 * safe against division by zero. We only estimate anyway.
5487 * dt: time from mark until now
5488 * db: blocks written from mark until now
5489 * rt: remaining time
5491 dt = ((jiffies - mddev->resync_mark) / HZ);
5493 db = (mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active))
5494 - mddev->resync_mark_cnt;
5495 rt = (dt * ((unsigned long)(max_blocks-resync) / (db/2/100+1)))/100;
5497 seq_printf(seq, " finish=%lu.%lumin", rt / 60, (rt % 60)/6);
5499 seq_printf(seq, " speed=%ldK/sec", db/2/dt);
5502 static void *md_seq_start(struct seq_file *seq, loff_t *pos)
5504 struct list_head *tmp;
5514 spin_lock(&all_mddevs_lock);
5515 list_for_each(tmp,&all_mddevs)
5517 mddev = list_entry(tmp, mddev_t, all_mddevs);
5519 spin_unlock(&all_mddevs_lock);
5522 spin_unlock(&all_mddevs_lock);
5524 return (void*)2;/* tail */
5528 static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
5530 struct list_head *tmp;
5531 mddev_t *next_mddev, *mddev = v;
5537 spin_lock(&all_mddevs_lock);
5539 tmp = all_mddevs.next;
5541 tmp = mddev->all_mddevs.next;
5542 if (tmp != &all_mddevs)
5543 next_mddev = mddev_get(list_entry(tmp,mddev_t,all_mddevs));
5545 next_mddev = (void*)2;
5548 spin_unlock(&all_mddevs_lock);
5556 static void md_seq_stop(struct seq_file *seq, void *v)
5560 if (mddev && v != (void*)1 && v != (void*)2)
5564 struct mdstat_info {
5568 static int md_seq_show(struct seq_file *seq, void *v)
5573 struct mdstat_info *mi = seq->private;
5574 struct bitmap *bitmap;
5576 if (v == (void*)1) {
5577 struct mdk_personality *pers;
5578 seq_printf(seq, "Personalities : ");
5579 spin_lock(&pers_lock);
5580 list_for_each_entry(pers, &pers_list, list)
5581 seq_printf(seq, "[%s] ", pers->name);
5583 spin_unlock(&pers_lock);
5584 seq_printf(seq, "\n");
5585 mi->event = atomic_read(&md_event_count);
5588 if (v == (void*)2) {
5593 if (mddev_lock(mddev) < 0)
5596 if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
5597 seq_printf(seq, "%s : %sactive", mdname(mddev),
5598 mddev->pers ? "" : "in");
5601 seq_printf(seq, " (read-only)");
5603 seq_printf(seq, " (auto-read-only)");
5604 seq_printf(seq, " %s", mddev->pers->name);
5608 list_for_each_entry(rdev, &mddev->disks, same_set) {
5609 char b[BDEVNAME_SIZE];
5610 seq_printf(seq, " %s[%d]",
5611 bdevname(rdev->bdev,b), rdev->desc_nr);
5612 if (test_bit(WriteMostly, &rdev->flags))
5613 seq_printf(seq, "(W)");
5614 if (test_bit(Faulty, &rdev->flags)) {
5615 seq_printf(seq, "(F)");
5617 } else if (rdev->raid_disk < 0)
5618 seq_printf(seq, "(S)"); /* spare */
5622 if (!list_empty(&mddev->disks)) {
5624 seq_printf(seq, "\n %llu blocks",
5625 (unsigned long long)
5626 mddev->array_sectors / 2);
5628 seq_printf(seq, "\n %llu blocks",
5629 (unsigned long long)size);
5631 if (mddev->persistent) {
5632 if (mddev->major_version != 0 ||
5633 mddev->minor_version != 90) {
5634 seq_printf(seq," super %d.%d",
5635 mddev->major_version,
5636 mddev->minor_version);
5638 } else if (mddev->external)
5639 seq_printf(seq, " super external:%s",
5640 mddev->metadata_type);
5642 seq_printf(seq, " super non-persistent");
5645 mddev->pers->status(seq, mddev);
5646 seq_printf(seq, "\n ");
5647 if (mddev->pers->sync_request) {
5648 if (mddev->curr_resync > 2) {
5649 status_resync(seq, mddev);
5650 seq_printf(seq, "\n ");
5651 } else if (mddev->curr_resync == 1 || mddev->curr_resync == 2)
5652 seq_printf(seq, "\tresync=DELAYED\n ");
5653 else if (mddev->recovery_cp < MaxSector)
5654 seq_printf(seq, "\tresync=PENDING\n ");
5657 seq_printf(seq, "\n ");
5659 if ((bitmap = mddev->bitmap)) {
5660 unsigned long chunk_kb;
5661 unsigned long flags;
5662 spin_lock_irqsave(&bitmap->lock, flags);
5663 chunk_kb = bitmap->chunksize >> 10;
5664 seq_printf(seq, "bitmap: %lu/%lu pages [%luKB], "
5666 bitmap->pages - bitmap->missing_pages,
5668 (bitmap->pages - bitmap->missing_pages)
5669 << (PAGE_SHIFT - 10),
5670 chunk_kb ? chunk_kb : bitmap->chunksize,
5671 chunk_kb ? "KB" : "B");
5673 seq_printf(seq, ", file: ");
5674 seq_path(seq, &bitmap->file->f_path, " \t\n");
5677 seq_printf(seq, "\n");
5678 spin_unlock_irqrestore(&bitmap->lock, flags);
5681 seq_printf(seq, "\n");
5683 mddev_unlock(mddev);
5688 static struct seq_operations md_seq_ops = {
5689 .start = md_seq_start,
5690 .next = md_seq_next,
5691 .stop = md_seq_stop,
5692 .show = md_seq_show,
5695 static int md_seq_open(struct inode *inode, struct file *file)
5698 struct mdstat_info *mi = kmalloc(sizeof(*mi), GFP_KERNEL);
5702 error = seq_open(file, &md_seq_ops);
5706 struct seq_file *p = file->private_data;
5708 mi->event = atomic_read(&md_event_count);
5713 static unsigned int mdstat_poll(struct file *filp, poll_table *wait)
5715 struct seq_file *m = filp->private_data;
5716 struct mdstat_info *mi = m->private;
5719 poll_wait(filp, &md_event_waiters, wait);
5721 /* always allow read */
5722 mask = POLLIN | POLLRDNORM;
5724 if (mi->event != atomic_read(&md_event_count))
5725 mask |= POLLERR | POLLPRI;
5729 static const struct file_operations md_seq_fops = {
5730 .owner = THIS_MODULE,
5731 .open = md_seq_open,
5733 .llseek = seq_lseek,
5734 .release = seq_release_private,
5735 .poll = mdstat_poll,
5738 int register_md_personality(struct mdk_personality *p)
5740 spin_lock(&pers_lock);
5741 list_add_tail(&p->list, &pers_list);
5742 printk(KERN_INFO "md: %s personality registered for level %d\n", p->name, p->level);
5743 spin_unlock(&pers_lock);
5747 int unregister_md_personality(struct mdk_personality *p)
5749 printk(KERN_INFO "md: %s personality unregistered\n", p->name);
5750 spin_lock(&pers_lock);
5751 list_del_init(&p->list);
5752 spin_unlock(&pers_lock);
5756 static int is_mddev_idle(mddev_t *mddev, int init)
5764 rdev_for_each_rcu(rdev, mddev) {
5765 struct gendisk *disk = rdev->bdev->bd_contains->bd_disk;
5766 curr_events = (int)part_stat_read(&disk->part0, sectors[0]) +
5767 (int)part_stat_read(&disk->part0, sectors[1]) -
5768 atomic_read(&disk->sync_io);
5769 /* sync IO will cause sync_io to increase before the disk_stats
5770 * as sync_io is counted when a request starts, and
5771 * disk_stats is counted when it completes.
5772 * So resync activity will cause curr_events to be smaller than
5773 * when there was no such activity.
5774 * non-sync IO will cause disk_stat to increase without
5775 * increasing sync_io so curr_events will (eventually)
5776 * be larger than it was before. Once it becomes
5777 * substantially larger, the test below will cause
5778 * the array to appear non-idle, and resync will slow
5780 * If there is a lot of outstanding resync activity when
5781 * we set last_event to curr_events, then all that activity
5782 * completing might cause the array to appear non-idle
5783 * and resync will be slowed down even though there might
5784 * not have been non-resync activity. This will only
5785 * happen once though. 'last_events' will soon reflect
5786 * the state where there is little or no outstanding
5787 * resync requests, and further resync activity will
5788 * always make curr_events less than last_events.
5791 if (init || curr_events - rdev->last_events > 64) {
5792 rdev->last_events = curr_events;
5800 void md_done_sync(mddev_t *mddev, int blocks, int ok)
5802 /* another "blocks" (512byte) blocks have been synced */
5803 atomic_sub(blocks, &mddev->recovery_active);
5804 wake_up(&mddev->recovery_wait);
5806 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5807 md_wakeup_thread(mddev->thread);
5808 // stop recovery, signal do_sync ....
5813 /* md_write_start(mddev, bi)
5814 * If we need to update some array metadata (e.g. 'active' flag
5815 * in superblock) before writing, schedule a superblock update
5816 * and wait for it to complete.
5818 void md_write_start(mddev_t *mddev, struct bio *bi)
5821 if (bio_data_dir(bi) != WRITE)
5824 BUG_ON(mddev->ro == 1);
5825 if (mddev->ro == 2) {
5826 /* need to switch to read/write */
5828 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5829 md_wakeup_thread(mddev->thread);
5830 md_wakeup_thread(mddev->sync_thread);
5833 atomic_inc(&mddev->writes_pending);
5834 if (mddev->safemode == 1)
5835 mddev->safemode = 0;
5836 if (mddev->in_sync) {
5837 spin_lock_irq(&mddev->write_lock);
5838 if (mddev->in_sync) {
5840 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
5841 md_wakeup_thread(mddev->thread);
5844 spin_unlock_irq(&mddev->write_lock);
5847 sysfs_notify_dirent(mddev->sysfs_state);
5848 wait_event(mddev->sb_wait,
5849 !test_bit(MD_CHANGE_CLEAN, &mddev->flags) &&
5850 !test_bit(MD_CHANGE_PENDING, &mddev->flags));
5853 void md_write_end(mddev_t *mddev)
5855 if (atomic_dec_and_test(&mddev->writes_pending)) {
5856 if (mddev->safemode == 2)
5857 md_wakeup_thread(mddev->thread);
5858 else if (mddev->safemode_delay)
5859 mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay);
5863 /* md_allow_write(mddev)
5864 * Calling this ensures that the array is marked 'active' so that writes
5865 * may proceed without blocking. It is important to call this before
5866 * attempting a GFP_KERNEL allocation while holding the mddev lock.
5867 * Must be called with mddev_lock held.
5869 * In the ->external case MD_CHANGE_CLEAN can not be cleared until mddev->lock
5870 * is dropped, so return -EAGAIN after notifying userspace.
5872 int md_allow_write(mddev_t *mddev)
5878 if (!mddev->pers->sync_request)
5881 spin_lock_irq(&mddev->write_lock);
5882 if (mddev->in_sync) {
5884 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
5885 if (mddev->safemode_delay &&
5886 mddev->safemode == 0)
5887 mddev->safemode = 1;
5888 spin_unlock_irq(&mddev->write_lock);
5889 md_update_sb(mddev, 0);
5890 sysfs_notify_dirent(mddev->sysfs_state);
5892 spin_unlock_irq(&mddev->write_lock);
5894 if (test_bit(MD_CHANGE_CLEAN, &mddev->flags))
5899 EXPORT_SYMBOL_GPL(md_allow_write);
5901 #define SYNC_MARKS 10
5902 #define SYNC_MARK_STEP (3*HZ)
5903 void md_do_sync(mddev_t *mddev)
5906 unsigned int currspeed = 0,
5908 sector_t max_sectors,j, io_sectors;
5909 unsigned long mark[SYNC_MARKS];
5910 sector_t mark_cnt[SYNC_MARKS];
5912 struct list_head *tmp;
5913 sector_t last_check;
5918 /* just incase thread restarts... */
5919 if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
5921 if (mddev->ro) /* never try to sync a read-only array */
5924 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
5925 if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
5926 desc = "data-check";
5927 else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
5928 desc = "requested-resync";
5931 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
5936 /* we overload curr_resync somewhat here.
5937 * 0 == not engaged in resync at all
5938 * 2 == checking that there is no conflict with another sync
5939 * 1 == like 2, but have yielded to allow conflicting resync to
5941 * other == active in resync - this many blocks
5943 * Before starting a resync we must have set curr_resync to
5944 * 2, and then checked that every "conflicting" array has curr_resync
5945 * less than ours. When we find one that is the same or higher
5946 * we wait on resync_wait. To avoid deadlock, we reduce curr_resync
5947 * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
5948 * This will mean we have to start checking from the beginning again.
5953 mddev->curr_resync = 2;
5956 if (kthread_should_stop()) {
5957 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5960 for_each_mddev(mddev2, tmp) {
5961 if (mddev2 == mddev)
5963 if (!mddev->parallel_resync
5964 && mddev2->curr_resync
5965 && match_mddev_units(mddev, mddev2)) {
5967 if (mddev < mddev2 && mddev->curr_resync == 2) {
5968 /* arbitrarily yield */
5969 mddev->curr_resync = 1;
5970 wake_up(&resync_wait);
5972 if (mddev > mddev2 && mddev->curr_resync == 1)
5973 /* no need to wait here, we can wait the next
5974 * time 'round when curr_resync == 2
5977 /* We need to wait 'interruptible' so as not to
5978 * contribute to the load average, and not to
5979 * be caught by 'softlockup'
5981 prepare_to_wait(&resync_wait, &wq, TASK_INTERRUPTIBLE);
5982 if (!kthread_should_stop() &&
5983 mddev2->curr_resync >= mddev->curr_resync) {
5984 printk(KERN_INFO "md: delaying %s of %s"
5985 " until %s has finished (they"
5986 " share one or more physical units)\n",
5987 desc, mdname(mddev), mdname(mddev2));
5989 if (signal_pending(current))
5990 flush_signals(current);
5992 finish_wait(&resync_wait, &wq);
5995 finish_wait(&resync_wait, &wq);
5998 } while (mddev->curr_resync < 2);
6001 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
6002 /* resync follows the size requested by the personality,
6003 * which defaults to physical size, but can be virtual size
6005 max_sectors = mddev->resync_max_sectors;
6006 mddev->resync_mismatches = 0;
6007 /* we don't use the checkpoint if there's a bitmap */
6008 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
6009 j = mddev->resync_min;
6010 else if (!mddev->bitmap)
6011 j = mddev->recovery_cp;
6013 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
6014 max_sectors = mddev->size << 1;
6016 /* recovery follows the physical size of devices */
6017 max_sectors = mddev->size << 1;
6019 list_for_each_entry(rdev, &mddev->disks, same_set)
6020 if (rdev->raid_disk >= 0 &&
6021 !test_bit(Faulty, &rdev->flags) &&
6022 !test_bit(In_sync, &rdev->flags) &&
6023 rdev->recovery_offset < j)
6024 j = rdev->recovery_offset;
6027 printk(KERN_INFO "md: %s of RAID array %s\n", desc, mdname(mddev));
6028 printk(KERN_INFO "md: minimum _guaranteed_ speed:"
6029 " %d KB/sec/disk.\n", speed_min(mddev));
6030 printk(KERN_INFO "md: using maximum available idle IO bandwidth "
6031 "(but not more than %d KB/sec) for %s.\n",
6032 speed_max(mddev), desc);
6034 is_mddev_idle(mddev, 1); /* this initializes IO event counters */
6037 for (m = 0; m < SYNC_MARKS; m++) {
6039 mark_cnt[m] = io_sectors;
6042 mddev->resync_mark = mark[last_mark];
6043 mddev->resync_mark_cnt = mark_cnt[last_mark];
6046 * Tune reconstruction:
6048 window = 32*(PAGE_SIZE/512);
6049 printk(KERN_INFO "md: using %dk window, over a total of %llu blocks.\n",
6050 window/2,(unsigned long long) max_sectors/2);
6052 atomic_set(&mddev->recovery_active, 0);
6057 "md: resuming %s of %s from checkpoint.\n",
6058 desc, mdname(mddev));
6059 mddev->curr_resync = j;
6062 while (j < max_sectors) {
6066 if (j >= mddev->resync_max) {
6067 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
6068 wait_event(mddev->recovery_wait,
6069 mddev->resync_max > j
6070 || kthread_should_stop());
6072 if (kthread_should_stop())
6074 sectors = mddev->pers->sync_request(mddev, j, &skipped,
6075 currspeed < speed_min(mddev));
6077 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6081 if (!skipped) { /* actual IO requested */
6082 io_sectors += sectors;
6083 atomic_add(sectors, &mddev->recovery_active);
6087 if (j>1) mddev->curr_resync = j;
6088 mddev->curr_mark_cnt = io_sectors;
6089 if (last_check == 0)
6090 /* this is the earliers that rebuilt will be
6091 * visible in /proc/mdstat
6093 md_new_event(mddev);
6095 if (last_check + window > io_sectors || j == max_sectors)
6098 last_check = io_sectors;
6100 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
6104 if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) {
6106 int next = (last_mark+1) % SYNC_MARKS;
6108 mddev->resync_mark = mark[next];
6109 mddev->resync_mark_cnt = mark_cnt[next];
6110 mark[next] = jiffies;
6111 mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active);
6116 if (kthread_should_stop())
6121 * this loop exits only if either when we are slower than
6122 * the 'hard' speed limit, or the system was IO-idle for
6124 * the system might be non-idle CPU-wise, but we only care
6125 * about not overloading the IO subsystem. (things like an
6126 * e2fsck being done on the RAID array should execute fast)
6128 blk_unplug(mddev->queue);
6131 currspeed = ((unsigned long)(io_sectors-mddev->resync_mark_cnt))/2
6132 /((jiffies-mddev->resync_mark)/HZ +1) +1;
6134 if (currspeed > speed_min(mddev)) {
6135 if ((currspeed > speed_max(mddev)) ||
6136 !is_mddev_idle(mddev, 0)) {
6142 printk(KERN_INFO "md: %s: %s done.\n",mdname(mddev), desc);
6144 * this also signals 'finished resyncing' to md_stop
6147 blk_unplug(mddev->queue);
6149 wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
6151 /* tell personality that we are finished */
6152 mddev->pers->sync_request(mddev, max_sectors, &skipped, 1);
6154 if (!test_bit(MD_RECOVERY_CHECK, &mddev->recovery) &&
6155 mddev->curr_resync > 2) {
6156 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
6157 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
6158 if (mddev->curr_resync >= mddev->recovery_cp) {
6160 "md: checkpointing %s of %s.\n",
6161 desc, mdname(mddev));
6162 mddev->recovery_cp = mddev->curr_resync;
6165 mddev->recovery_cp = MaxSector;
6167 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery))
6168 mddev->curr_resync = MaxSector;
6169 list_for_each_entry(rdev, &mddev->disks, same_set)
6170 if (rdev->raid_disk >= 0 &&
6171 !test_bit(Faulty, &rdev->flags) &&
6172 !test_bit(In_sync, &rdev->flags) &&
6173 rdev->recovery_offset < mddev->curr_resync)
6174 rdev->recovery_offset = mddev->curr_resync;
6177 set_bit(MD_CHANGE_DEVS, &mddev->flags);
6180 mddev->curr_resync = 0;
6181 mddev->resync_min = 0;
6182 mddev->resync_max = MaxSector;
6183 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
6184 wake_up(&resync_wait);
6185 set_bit(MD_RECOVERY_DONE, &mddev->recovery);
6186 md_wakeup_thread(mddev->thread);
6191 * got a signal, exit.
6194 "md: md_do_sync() got signal ... exiting\n");
6195 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6199 EXPORT_SYMBOL_GPL(md_do_sync);
6202 static int remove_and_add_spares(mddev_t *mddev)
6207 list_for_each_entry(rdev, &mddev->disks, same_set)
6208 if (rdev->raid_disk >= 0 &&
6209 !test_bit(Blocked, &rdev->flags) &&
6210 (test_bit(Faulty, &rdev->flags) ||
6211 ! test_bit(In_sync, &rdev->flags)) &&
6212 atomic_read(&rdev->nr_pending)==0) {
6213 if (mddev->pers->hot_remove_disk(
6214 mddev, rdev->raid_disk)==0) {
6216 sprintf(nm,"rd%d", rdev->raid_disk);
6217 sysfs_remove_link(&mddev->kobj, nm);
6218 rdev->raid_disk = -1;
6222 if (mddev->degraded && ! mddev->ro && !mddev->recovery_disabled) {
6223 list_for_each_entry(rdev, &mddev->disks, same_set) {
6224 if (rdev->raid_disk >= 0 &&
6225 !test_bit(In_sync, &rdev->flags) &&
6226 !test_bit(Blocked, &rdev->flags))
6228 if (rdev->raid_disk < 0
6229 && !test_bit(Faulty, &rdev->flags)) {
6230 rdev->recovery_offset = 0;
6232 hot_add_disk(mddev, rdev) == 0) {
6234 sprintf(nm, "rd%d", rdev->raid_disk);
6235 if (sysfs_create_link(&mddev->kobj,
6238 "md: cannot register "
6242 md_new_event(mddev);
6251 * This routine is regularly called by all per-raid-array threads to
6252 * deal with generic issues like resync and super-block update.
6253 * Raid personalities that don't have a thread (linear/raid0) do not
6254 * need this as they never do any recovery or update the superblock.
6256 * It does not do any resync itself, but rather "forks" off other threads
6257 * to do that as needed.
6258 * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
6259 * "->recovery" and create a thread at ->sync_thread.
6260 * When the thread finishes it sets MD_RECOVERY_DONE
6261 * and wakeups up this thread which will reap the thread and finish up.
6262 * This thread also removes any faulty devices (with nr_pending == 0).
6264 * The overall approach is:
6265 * 1/ if the superblock needs updating, update it.
6266 * 2/ If a recovery thread is running, don't do anything else.
6267 * 3/ If recovery has finished, clean up, possibly marking spares active.
6268 * 4/ If there are any faulty devices, remove them.
6269 * 5/ If array is degraded, try to add spares devices
6270 * 6/ If array has spares or is not in-sync, start a resync thread.
6272 void md_check_recovery(mddev_t *mddev)
6278 bitmap_daemon_work(mddev->bitmap);
6283 if (signal_pending(current)) {
6284 if (mddev->pers->sync_request && !mddev->external) {
6285 printk(KERN_INFO "md: %s in immediate safe mode\n",
6287 mddev->safemode = 2;
6289 flush_signals(current);
6292 if (mddev->ro && !test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
6295 (mddev->flags && !mddev->external) ||
6296 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
6297 test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
6298 (mddev->external == 0 && mddev->safemode == 1) ||
6299 (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending)
6300 && !mddev->in_sync && mddev->recovery_cp == MaxSector)
6304 if (mddev_trylock(mddev)) {
6308 /* Only thing we do on a ro array is remove
6311 remove_and_add_spares(mddev);
6312 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6316 if (!mddev->external) {
6318 spin_lock_irq(&mddev->write_lock);
6319 if (mddev->safemode &&
6320 !atomic_read(&mddev->writes_pending) &&
6322 mddev->recovery_cp == MaxSector) {
6325 if (mddev->persistent)
6326 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
6328 if (mddev->safemode == 1)
6329 mddev->safemode = 0;
6330 spin_unlock_irq(&mddev->write_lock);
6332 sysfs_notify_dirent(mddev->sysfs_state);
6336 md_update_sb(mddev, 0);
6338 list_for_each_entry(rdev, &mddev->disks, same_set)
6339 if (test_and_clear_bit(StateChanged, &rdev->flags))
6340 sysfs_notify_dirent(rdev->sysfs_state);
6343 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
6344 !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
6345 /* resync/recovery still happening */
6346 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6349 if (mddev->sync_thread) {
6350 /* resync has finished, collect result */
6351 md_unregister_thread(mddev->sync_thread);
6352 mddev->sync_thread = NULL;
6353 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
6354 !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
6356 /* activate any spares */
6357 if (mddev->pers->spare_active(mddev))
6358 sysfs_notify(&mddev->kobj, NULL,
6361 md_update_sb(mddev, 1);
6363 /* if array is no-longer degraded, then any saved_raid_disk
6364 * information must be scrapped
6366 if (!mddev->degraded)
6367 list_for_each_entry(rdev, &mddev->disks, same_set)
6368 rdev->saved_raid_disk = -1;
6370 mddev->recovery = 0;
6371 /* flag recovery needed just to double check */
6372 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6373 sysfs_notify_dirent(mddev->sysfs_action);
6374 md_new_event(mddev);
6377 /* Set RUNNING before clearing NEEDED to avoid
6378 * any transients in the value of "sync_action".
6380 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
6381 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6382 /* Clear some bits that don't mean anything, but
6385 clear_bit(MD_RECOVERY_INTR, &mddev->recovery);
6386 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
6388 if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
6390 /* no recovery is running.
6391 * remove any failed drives, then
6392 * add spares if possible.
6393 * Spare are also removed and re-added, to allow
6394 * the personality to fail the re-add.
6397 if (mddev->reshape_position != MaxSector) {
6398 if (mddev->pers->check_reshape(mddev) != 0)
6399 /* Cannot proceed */
6401 set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
6402 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
6403 } else if ((spares = remove_and_add_spares(mddev))) {
6404 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
6405 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
6406 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
6407 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
6408 } else if (mddev->recovery_cp < MaxSector) {
6409 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
6410 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
6411 } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
6412 /* nothing to be done ... */
6415 if (mddev->pers->sync_request) {
6416 if (spares && mddev->bitmap && ! mddev->bitmap->file) {
6417 /* We are adding a device or devices to an array
6418 * which has the bitmap stored on all devices.
6419 * So make sure all bitmap pages get written
6421 bitmap_write_all(mddev->bitmap);
6423 mddev->sync_thread = md_register_thread(md_do_sync,
6426 if (!mddev->sync_thread) {
6427 printk(KERN_ERR "%s: could not start resync"
6430 /* leave the spares where they are, it shouldn't hurt */
6431 mddev->recovery = 0;
6433 md_wakeup_thread(mddev->sync_thread);
6434 sysfs_notify_dirent(mddev->sysfs_action);
6435 md_new_event(mddev);
6438 if (!mddev->sync_thread) {
6439 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
6440 if (test_and_clear_bit(MD_RECOVERY_RECOVER,
6442 if (mddev->sysfs_action)
6443 sysfs_notify_dirent(mddev->sysfs_action);
6445 mddev_unlock(mddev);
6449 void md_wait_for_blocked_rdev(mdk_rdev_t *rdev, mddev_t *mddev)
6451 sysfs_notify_dirent(rdev->sysfs_state);
6452 wait_event_timeout(rdev->blocked_wait,
6453 !test_bit(Blocked, &rdev->flags),
6454 msecs_to_jiffies(5000));
6455 rdev_dec_pending(rdev, mddev);
6457 EXPORT_SYMBOL(md_wait_for_blocked_rdev);
6459 static int md_notify_reboot(struct notifier_block *this,
6460 unsigned long code, void *x)
6462 struct list_head *tmp;
6465 if ((code == SYS_DOWN) || (code == SYS_HALT) || (code == SYS_POWER_OFF)) {
6467 printk(KERN_INFO "md: stopping all md devices.\n");
6469 for_each_mddev(mddev, tmp)
6470 if (mddev_trylock(mddev)) {
6471 /* Force a switch to readonly even array
6472 * appears to still be in use. Hence
6475 do_md_stop(mddev, 1, 100);
6476 mddev_unlock(mddev);
6479 * certain more exotic SCSI devices are known to be
6480 * volatile wrt too early system reboots. While the
6481 * right place to handle this issue is the given
6482 * driver, we do want to have a safe RAID driver ...
6489 static struct notifier_block md_notifier = {
6490 .notifier_call = md_notify_reboot,
6492 .priority = INT_MAX, /* before any real devices */
6495 static void md_geninit(void)
6497 dprintk("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
6499 proc_create("mdstat", S_IRUGO, NULL, &md_seq_fops);
6502 static int __init md_init(void)
6504 if (register_blkdev(MD_MAJOR, "md"))
6506 if ((mdp_major=register_blkdev(0, "mdp"))<=0) {
6507 unregister_blkdev(MD_MAJOR, "md");
6510 blk_register_region(MKDEV(MD_MAJOR, 0), 1UL<<MINORBITS, THIS_MODULE,
6511 md_probe, NULL, NULL);
6512 blk_register_region(MKDEV(mdp_major, 0), 1UL<<MINORBITS, THIS_MODULE,
6513 md_probe, NULL, NULL);
6515 register_reboot_notifier(&md_notifier);
6516 raid_table_header = register_sysctl_table(raid_root_table);
6526 * Searches all registered partitions for autorun RAID arrays
6530 static LIST_HEAD(all_detected_devices);
6531 struct detected_devices_node {
6532 struct list_head list;
6536 void md_autodetect_dev(dev_t dev)
6538 struct detected_devices_node *node_detected_dev;
6540 node_detected_dev = kzalloc(sizeof(*node_detected_dev), GFP_KERNEL);
6541 if (node_detected_dev) {
6542 node_detected_dev->dev = dev;
6543 list_add_tail(&node_detected_dev->list, &all_detected_devices);
6545 printk(KERN_CRIT "md: md_autodetect_dev: kzalloc failed"
6546 ", skipping dev(%d,%d)\n", MAJOR(dev), MINOR(dev));
6551 static void autostart_arrays(int part)
6554 struct detected_devices_node *node_detected_dev;
6556 int i_scanned, i_passed;
6561 printk(KERN_INFO "md: Autodetecting RAID arrays.\n");
6563 while (!list_empty(&all_detected_devices) && i_scanned < INT_MAX) {
6565 node_detected_dev = list_entry(all_detected_devices.next,
6566 struct detected_devices_node, list);
6567 list_del(&node_detected_dev->list);
6568 dev = node_detected_dev->dev;
6569 kfree(node_detected_dev);
6570 rdev = md_import_device(dev,0, 90);
6574 if (test_bit(Faulty, &rdev->flags)) {
6578 set_bit(AutoDetected, &rdev->flags);
6579 list_add(&rdev->same_set, &pending_raid_disks);
6583 printk(KERN_INFO "md: Scanned %d and added %d devices.\n",
6584 i_scanned, i_passed);
6586 autorun_devices(part);
6589 #endif /* !MODULE */
6591 static __exit void md_exit(void)
6594 struct list_head *tmp;
6596 blk_unregister_region(MKDEV(MD_MAJOR,0), 1U << MINORBITS);
6597 blk_unregister_region(MKDEV(mdp_major,0), 1U << MINORBITS);
6599 unregister_blkdev(MD_MAJOR,"md");
6600 unregister_blkdev(mdp_major, "mdp");
6601 unregister_reboot_notifier(&md_notifier);
6602 unregister_sysctl_table(raid_table_header);
6603 remove_proc_entry("mdstat", NULL);
6604 for_each_mddev(mddev, tmp) {
6605 export_array(mddev);
6606 mddev->hold_active = 0;
6610 subsys_initcall(md_init);
6611 module_exit(md_exit)
6613 static int get_ro(char *buffer, struct kernel_param *kp)
6615 return sprintf(buffer, "%d", start_readonly);
6617 static int set_ro(const char *val, struct kernel_param *kp)
6620 int num = simple_strtoul(val, &e, 10);
6621 if (*val && (*e == '\0' || *e == '\n')) {
6622 start_readonly = num;
6628 module_param_call(start_ro, set_ro, get_ro, NULL, S_IRUSR|S_IWUSR);
6629 module_param(start_dirty_degraded, int, S_IRUGO|S_IWUSR);
6631 module_param_call(new_array, add_named_array, NULL, NULL, S_IWUSR);
6633 EXPORT_SYMBOL(register_md_personality);
6634 EXPORT_SYMBOL(unregister_md_personality);
6635 EXPORT_SYMBOL(md_error);
6636 EXPORT_SYMBOL(md_done_sync);
6637 EXPORT_SYMBOL(md_write_start);
6638 EXPORT_SYMBOL(md_write_end);
6639 EXPORT_SYMBOL(md_register_thread);
6640 EXPORT_SYMBOL(md_unregister_thread);
6641 EXPORT_SYMBOL(md_wakeup_thread);
6642 EXPORT_SYMBOL(md_check_recovery);
6643 MODULE_LICENSE("GPL");
6645 MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR);