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/module.h>
36 #include <linux/kthread.h>
37 #include <linux/linkage.h>
38 #include <linux/raid/md.h>
39 #include <linux/raid/bitmap.h>
40 #include <linux/sysctl.h>
41 #include <linux/buffer_head.h> /* for invalidate_bdev */
42 #include <linux/poll.h>
43 #include <linux/mutex.h>
44 #include <linux/ctype.h>
45 #include <linux/freezer.h>
47 #include <linux/init.h>
49 #include <linux/file.h>
52 #include <linux/kmod.h>
55 #include <asm/unaligned.h>
57 #define MAJOR_NR MD_MAJOR
60 /* 63 partitions with the alternate major number (mdp) */
61 #define MdpMinorShift 6
64 #define dprintk(x...) ((void)(DEBUG && printk(x)))
68 static void autostart_arrays (int part);
71 static LIST_HEAD(pers_list);
72 static DEFINE_SPINLOCK(pers_lock);
74 static void md_print_devices(void);
76 #define MD_BUG(x...) { printk("md: bug in file %s, line %d\n", __FILE__, __LINE__); md_print_devices(); }
79 * Current RAID-1,4,5 parallel reconstruction 'guaranteed speed limit'
80 * is 1000 KB/sec, so the extra system load does not show up that much.
81 * Increase it if you want to have more _guaranteed_ speed. Note that
82 * the RAID driver will use the maximum available bandwidth if the IO
83 * subsystem is idle. There is also an 'absolute maximum' reconstruction
84 * speed limit - in case reconstruction slows down your system despite
87 * you can change it via /proc/sys/dev/raid/speed_limit_min and _max.
88 * or /sys/block/mdX/md/sync_speed_{min,max}
91 static int sysctl_speed_limit_min = 1000;
92 static int sysctl_speed_limit_max = 200000;
93 static inline int speed_min(mddev_t *mddev)
95 return mddev->sync_speed_min ?
96 mddev->sync_speed_min : sysctl_speed_limit_min;
99 static inline int speed_max(mddev_t *mddev)
101 return mddev->sync_speed_max ?
102 mddev->sync_speed_max : sysctl_speed_limit_max;
105 static struct ctl_table_header *raid_table_header;
107 static ctl_table raid_table[] = {
109 .ctl_name = DEV_RAID_SPEED_LIMIT_MIN,
110 .procname = "speed_limit_min",
111 .data = &sysctl_speed_limit_min,
112 .maxlen = sizeof(int),
113 .mode = S_IRUGO|S_IWUSR,
114 .proc_handler = &proc_dointvec,
117 .ctl_name = DEV_RAID_SPEED_LIMIT_MAX,
118 .procname = "speed_limit_max",
119 .data = &sysctl_speed_limit_max,
120 .maxlen = sizeof(int),
121 .mode = S_IRUGO|S_IWUSR,
122 .proc_handler = &proc_dointvec,
127 static ctl_table raid_dir_table[] = {
129 .ctl_name = DEV_RAID,
132 .mode = S_IRUGO|S_IXUGO,
138 static ctl_table raid_root_table[] = {
144 .child = raid_dir_table,
149 static struct block_device_operations md_fops;
151 static int start_readonly;
154 * We have a system wide 'event count' that is incremented
155 * on any 'interesting' event, and readers of /proc/mdstat
156 * can use 'poll' or 'select' to find out when the event
160 * start array, stop array, error, add device, remove device,
161 * start build, activate spare
163 static DECLARE_WAIT_QUEUE_HEAD(md_event_waiters);
164 static atomic_t md_event_count;
165 void md_new_event(mddev_t *mddev)
167 atomic_inc(&md_event_count);
168 wake_up(&md_event_waiters);
169 sysfs_notify(&mddev->kobj, NULL, "sync_action");
171 EXPORT_SYMBOL_GPL(md_new_event);
173 /* Alternate version that can be called from interrupts
174 * when calling sysfs_notify isn't needed.
176 static void md_new_event_inintr(mddev_t *mddev)
178 atomic_inc(&md_event_count);
179 wake_up(&md_event_waiters);
183 * Enables to iterate over all existing md arrays
184 * all_mddevs_lock protects this list.
186 static LIST_HEAD(all_mddevs);
187 static DEFINE_SPINLOCK(all_mddevs_lock);
191 * iterates through all used mddevs in the system.
192 * We take care to grab the all_mddevs_lock whenever navigating
193 * the list, and to always hold a refcount when unlocked.
194 * Any code which breaks out of this loop while own
195 * a reference to the current mddev and must mddev_put it.
197 #define ITERATE_MDDEV(mddev,tmp) \
199 for (({ spin_lock(&all_mddevs_lock); \
200 tmp = all_mddevs.next; \
202 ({ if (tmp != &all_mddevs) \
203 mddev_get(list_entry(tmp, mddev_t, all_mddevs));\
204 spin_unlock(&all_mddevs_lock); \
205 if (mddev) mddev_put(mddev); \
206 mddev = list_entry(tmp, mddev_t, all_mddevs); \
207 tmp != &all_mddevs;}); \
208 ({ spin_lock(&all_mddevs_lock); \
213 static int md_fail_request (request_queue_t *q, struct bio *bio)
215 bio_io_error(bio, bio->bi_size);
219 static inline mddev_t *mddev_get(mddev_t *mddev)
221 atomic_inc(&mddev->active);
225 static void mddev_put(mddev_t *mddev)
227 if (!atomic_dec_and_lock(&mddev->active, &all_mddevs_lock))
229 if (!mddev->raid_disks && list_empty(&mddev->disks)) {
230 list_del(&mddev->all_mddevs);
231 spin_unlock(&all_mddevs_lock);
232 blk_cleanup_queue(mddev->queue);
233 kobject_unregister(&mddev->kobj);
235 spin_unlock(&all_mddevs_lock);
238 static mddev_t * mddev_find(dev_t unit)
240 mddev_t *mddev, *new = NULL;
243 spin_lock(&all_mddevs_lock);
244 list_for_each_entry(mddev, &all_mddevs, all_mddevs)
245 if (mddev->unit == unit) {
247 spin_unlock(&all_mddevs_lock);
253 list_add(&new->all_mddevs, &all_mddevs);
254 spin_unlock(&all_mddevs_lock);
257 spin_unlock(&all_mddevs_lock);
259 new = kzalloc(sizeof(*new), GFP_KERNEL);
264 if (MAJOR(unit) == MD_MAJOR)
265 new->md_minor = MINOR(unit);
267 new->md_minor = MINOR(unit) >> MdpMinorShift;
269 mutex_init(&new->reconfig_mutex);
270 INIT_LIST_HEAD(&new->disks);
271 INIT_LIST_HEAD(&new->all_mddevs);
272 init_timer(&new->safemode_timer);
273 atomic_set(&new->active, 1);
274 spin_lock_init(&new->write_lock);
275 init_waitqueue_head(&new->sb_wait);
277 new->queue = blk_alloc_queue(GFP_KERNEL);
282 set_bit(QUEUE_FLAG_CLUSTER, &new->queue->queue_flags);
284 blk_queue_make_request(new->queue, md_fail_request);
289 static inline int mddev_lock(mddev_t * mddev)
291 return mutex_lock_interruptible(&mddev->reconfig_mutex);
294 static inline int mddev_trylock(mddev_t * mddev)
296 return mutex_trylock(&mddev->reconfig_mutex);
299 static inline void mddev_unlock(mddev_t * mddev)
301 mutex_unlock(&mddev->reconfig_mutex);
303 md_wakeup_thread(mddev->thread);
306 static mdk_rdev_t * find_rdev_nr(mddev_t *mddev, int nr)
309 struct list_head *tmp;
311 ITERATE_RDEV(mddev,rdev,tmp) {
312 if (rdev->desc_nr == nr)
318 static mdk_rdev_t * find_rdev(mddev_t * mddev, dev_t dev)
320 struct list_head *tmp;
323 ITERATE_RDEV(mddev,rdev,tmp) {
324 if (rdev->bdev->bd_dev == dev)
330 static struct mdk_personality *find_pers(int level, char *clevel)
332 struct mdk_personality *pers;
333 list_for_each_entry(pers, &pers_list, list) {
334 if (level != LEVEL_NONE && pers->level == level)
336 if (strcmp(pers->name, clevel)==0)
342 static inline sector_t calc_dev_sboffset(struct block_device *bdev)
344 sector_t size = bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
345 return MD_NEW_SIZE_BLOCKS(size);
348 static sector_t calc_dev_size(mdk_rdev_t *rdev, unsigned chunk_size)
352 size = rdev->sb_offset;
355 size &= ~((sector_t)chunk_size/1024 - 1);
359 static int alloc_disk_sb(mdk_rdev_t * rdev)
364 rdev->sb_page = alloc_page(GFP_KERNEL);
365 if (!rdev->sb_page) {
366 printk(KERN_ALERT "md: out of memory.\n");
373 static void free_disk_sb(mdk_rdev_t * rdev)
376 put_page(rdev->sb_page);
378 rdev->sb_page = NULL;
385 static int super_written(struct bio *bio, unsigned int bytes_done, int error)
387 mdk_rdev_t *rdev = bio->bi_private;
388 mddev_t *mddev = rdev->mddev;
392 if (error || !test_bit(BIO_UPTODATE, &bio->bi_flags)) {
393 printk("md: super_written gets error=%d, uptodate=%d\n",
394 error, test_bit(BIO_UPTODATE, &bio->bi_flags));
395 WARN_ON(test_bit(BIO_UPTODATE, &bio->bi_flags));
396 md_error(mddev, rdev);
399 if (atomic_dec_and_test(&mddev->pending_writes))
400 wake_up(&mddev->sb_wait);
405 static int super_written_barrier(struct bio *bio, unsigned int bytes_done, int error)
407 struct bio *bio2 = bio->bi_private;
408 mdk_rdev_t *rdev = bio2->bi_private;
409 mddev_t *mddev = rdev->mddev;
413 if (!test_bit(BIO_UPTODATE, &bio->bi_flags) &&
414 error == -EOPNOTSUPP) {
416 /* barriers don't appear to be supported :-( */
417 set_bit(BarriersNotsupp, &rdev->flags);
418 mddev->barriers_work = 0;
419 spin_lock_irqsave(&mddev->write_lock, flags);
420 bio2->bi_next = mddev->biolist;
421 mddev->biolist = bio2;
422 spin_unlock_irqrestore(&mddev->write_lock, flags);
423 wake_up(&mddev->sb_wait);
428 bio->bi_private = rdev;
429 return super_written(bio, bytes_done, error);
432 void md_super_write(mddev_t *mddev, mdk_rdev_t *rdev,
433 sector_t sector, int size, struct page *page)
435 /* write first size bytes of page to sector of rdev
436 * Increment mddev->pending_writes before returning
437 * and decrement it on completion, waking up sb_wait
438 * if zero is reached.
439 * If an error occurred, call md_error
441 * As we might need to resubmit the request if BIO_RW_BARRIER
442 * causes ENOTSUPP, we allocate a spare bio...
444 struct bio *bio = bio_alloc(GFP_NOIO, 1);
445 int rw = (1<<BIO_RW) | (1<<BIO_RW_SYNC);
447 bio->bi_bdev = rdev->bdev;
448 bio->bi_sector = sector;
449 bio_add_page(bio, page, size, 0);
450 bio->bi_private = rdev;
451 bio->bi_end_io = super_written;
454 atomic_inc(&mddev->pending_writes);
455 if (!test_bit(BarriersNotsupp, &rdev->flags)) {
457 rw |= (1<<BIO_RW_BARRIER);
458 rbio = bio_clone(bio, GFP_NOIO);
459 rbio->bi_private = bio;
460 rbio->bi_end_io = super_written_barrier;
461 submit_bio(rw, rbio);
466 void md_super_wait(mddev_t *mddev)
468 /* wait for all superblock writes that were scheduled to complete.
469 * if any had to be retried (due to BARRIER problems), retry them
473 prepare_to_wait(&mddev->sb_wait, &wq, TASK_UNINTERRUPTIBLE);
474 if (atomic_read(&mddev->pending_writes)==0)
476 while (mddev->biolist) {
478 spin_lock_irq(&mddev->write_lock);
479 bio = mddev->biolist;
480 mddev->biolist = bio->bi_next ;
482 spin_unlock_irq(&mddev->write_lock);
483 submit_bio(bio->bi_rw, bio);
487 finish_wait(&mddev->sb_wait, &wq);
490 static int bi_complete(struct bio *bio, unsigned int bytes_done, int error)
495 complete((struct completion*)bio->bi_private);
499 int sync_page_io(struct block_device *bdev, sector_t sector, int size,
500 struct page *page, int rw)
502 struct bio *bio = bio_alloc(GFP_NOIO, 1);
503 struct completion event;
506 rw |= (1 << BIO_RW_SYNC);
509 bio->bi_sector = sector;
510 bio_add_page(bio, page, size, 0);
511 init_completion(&event);
512 bio->bi_private = &event;
513 bio->bi_end_io = bi_complete;
515 wait_for_completion(&event);
517 ret = test_bit(BIO_UPTODATE, &bio->bi_flags);
521 EXPORT_SYMBOL_GPL(sync_page_io);
523 static int read_disk_sb(mdk_rdev_t * rdev, int size)
525 char b[BDEVNAME_SIZE];
526 if (!rdev->sb_page) {
534 if (!sync_page_io(rdev->bdev, rdev->sb_offset<<1, size, rdev->sb_page, READ))
540 printk(KERN_WARNING "md: disabled device %s, could not read superblock.\n",
541 bdevname(rdev->bdev,b));
545 static int uuid_equal(mdp_super_t *sb1, mdp_super_t *sb2)
547 if ( (sb1->set_uuid0 == sb2->set_uuid0) &&
548 (sb1->set_uuid1 == sb2->set_uuid1) &&
549 (sb1->set_uuid2 == sb2->set_uuid2) &&
550 (sb1->set_uuid3 == sb2->set_uuid3))
558 static int sb_equal(mdp_super_t *sb1, mdp_super_t *sb2)
561 mdp_super_t *tmp1, *tmp2;
563 tmp1 = kmalloc(sizeof(*tmp1),GFP_KERNEL);
564 tmp2 = kmalloc(sizeof(*tmp2),GFP_KERNEL);
566 if (!tmp1 || !tmp2) {
568 printk(KERN_INFO "md.c: sb1 is not equal to sb2!\n");
576 * nr_disks is not constant
581 if (memcmp(tmp1, tmp2, MD_SB_GENERIC_CONSTANT_WORDS * 4))
592 static unsigned int calc_sb_csum(mdp_super_t * sb)
594 unsigned int disk_csum, csum;
596 disk_csum = sb->sb_csum;
598 csum = csum_partial((void *)sb, MD_SB_BYTES, 0);
599 sb->sb_csum = disk_csum;
605 * Handle superblock details.
606 * We want to be able to handle multiple superblock formats
607 * so we have a common interface to them all, and an array of
608 * different handlers.
609 * We rely on user-space to write the initial superblock, and support
610 * reading and updating of superblocks.
611 * Interface methods are:
612 * int load_super(mdk_rdev_t *dev, mdk_rdev_t *refdev, int minor_version)
613 * loads and validates a superblock on dev.
614 * if refdev != NULL, compare superblocks on both devices
616 * 0 - dev has a superblock that is compatible with refdev
617 * 1 - dev has a superblock that is compatible and newer than refdev
618 * so dev should be used as the refdev in future
619 * -EINVAL superblock incompatible or invalid
620 * -othererror e.g. -EIO
622 * int validate_super(mddev_t *mddev, mdk_rdev_t *dev)
623 * Verify that dev is acceptable into mddev.
624 * The first time, mddev->raid_disks will be 0, and data from
625 * dev should be merged in. Subsequent calls check that dev
626 * is new enough. Return 0 or -EINVAL
628 * void sync_super(mddev_t *mddev, mdk_rdev_t *dev)
629 * Update the superblock for rdev with data in mddev
630 * This does not write to disc.
636 struct module *owner;
637 int (*load_super)(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version);
638 int (*validate_super)(mddev_t *mddev, mdk_rdev_t *rdev);
639 void (*sync_super)(mddev_t *mddev, mdk_rdev_t *rdev);
643 * load_super for 0.90.0
645 static int super_90_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
647 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
653 * Calculate the position of the superblock,
654 * it's at the end of the disk.
656 * It also happens to be a multiple of 4Kb.
658 sb_offset = calc_dev_sboffset(rdev->bdev);
659 rdev->sb_offset = sb_offset;
661 ret = read_disk_sb(rdev, MD_SB_BYTES);
666 bdevname(rdev->bdev, b);
667 sb = (mdp_super_t*)page_address(rdev->sb_page);
669 if (sb->md_magic != MD_SB_MAGIC) {
670 printk(KERN_ERR "md: invalid raid superblock magic on %s\n",
675 if (sb->major_version != 0 ||
676 sb->minor_version < 90 ||
677 sb->minor_version > 91) {
678 printk(KERN_WARNING "Bad version number %d.%d on %s\n",
679 sb->major_version, sb->minor_version,
684 if (sb->raid_disks <= 0)
687 if (csum_fold(calc_sb_csum(sb)) != csum_fold(sb->sb_csum)) {
688 printk(KERN_WARNING "md: invalid superblock checksum on %s\n",
693 rdev->preferred_minor = sb->md_minor;
694 rdev->data_offset = 0;
695 rdev->sb_size = MD_SB_BYTES;
697 if (sb->level == LEVEL_MULTIPATH)
700 rdev->desc_nr = sb->this_disk.number;
706 mdp_super_t *refsb = (mdp_super_t*)page_address(refdev->sb_page);
707 if (!uuid_equal(refsb, sb)) {
708 printk(KERN_WARNING "md: %s has different UUID to %s\n",
709 b, bdevname(refdev->bdev,b2));
712 if (!sb_equal(refsb, sb)) {
713 printk(KERN_WARNING "md: %s has same UUID"
714 " but different superblock to %s\n",
715 b, bdevname(refdev->bdev, b2));
719 ev2 = md_event(refsb);
725 rdev->size = calc_dev_size(rdev, sb->chunk_size);
727 if (rdev->size < sb->size && sb->level > 1)
728 /* "this cannot possibly happen" ... */
736 * validate_super for 0.90.0
738 static int super_90_validate(mddev_t *mddev, mdk_rdev_t *rdev)
741 mdp_super_t *sb = (mdp_super_t *)page_address(rdev->sb_page);
742 __u64 ev1 = md_event(sb);
744 rdev->raid_disk = -1;
746 if (mddev->raid_disks == 0) {
747 mddev->major_version = 0;
748 mddev->minor_version = sb->minor_version;
749 mddev->patch_version = sb->patch_version;
750 mddev->persistent = ! sb->not_persistent;
751 mddev->chunk_size = sb->chunk_size;
752 mddev->ctime = sb->ctime;
753 mddev->utime = sb->utime;
754 mddev->level = sb->level;
755 mddev->clevel[0] = 0;
756 mddev->layout = sb->layout;
757 mddev->raid_disks = sb->raid_disks;
758 mddev->size = sb->size;
760 mddev->bitmap_offset = 0;
761 mddev->default_bitmap_offset = MD_SB_BYTES >> 9;
763 if (mddev->minor_version >= 91) {
764 mddev->reshape_position = sb->reshape_position;
765 mddev->delta_disks = sb->delta_disks;
766 mddev->new_level = sb->new_level;
767 mddev->new_layout = sb->new_layout;
768 mddev->new_chunk = sb->new_chunk;
770 mddev->reshape_position = MaxSector;
771 mddev->delta_disks = 0;
772 mddev->new_level = mddev->level;
773 mddev->new_layout = mddev->layout;
774 mddev->new_chunk = mddev->chunk_size;
777 if (sb->state & (1<<MD_SB_CLEAN))
778 mddev->recovery_cp = MaxSector;
780 if (sb->events_hi == sb->cp_events_hi &&
781 sb->events_lo == sb->cp_events_lo) {
782 mddev->recovery_cp = sb->recovery_cp;
784 mddev->recovery_cp = 0;
787 memcpy(mddev->uuid+0, &sb->set_uuid0, 4);
788 memcpy(mddev->uuid+4, &sb->set_uuid1, 4);
789 memcpy(mddev->uuid+8, &sb->set_uuid2, 4);
790 memcpy(mddev->uuid+12,&sb->set_uuid3, 4);
792 mddev->max_disks = MD_SB_DISKS;
794 if (sb->state & (1<<MD_SB_BITMAP_PRESENT) &&
795 mddev->bitmap_file == NULL) {
796 if (mddev->level != 1 && mddev->level != 4
797 && mddev->level != 5 && mddev->level != 6
798 && mddev->level != 10) {
799 /* FIXME use a better test */
800 printk(KERN_WARNING "md: bitmaps not supported for this level.\n");
803 mddev->bitmap_offset = mddev->default_bitmap_offset;
806 } else if (mddev->pers == NULL) {
807 /* Insist on good event counter while assembling */
809 if (ev1 < mddev->events)
811 } else if (mddev->bitmap) {
812 /* if adding to array with a bitmap, then we can accept an
813 * older device ... but not too old.
815 if (ev1 < mddev->bitmap->events_cleared)
818 if (ev1 < mddev->events)
819 /* just a hot-add of a new device, leave raid_disk at -1 */
823 if (mddev->level != LEVEL_MULTIPATH) {
824 desc = sb->disks + rdev->desc_nr;
826 if (desc->state & (1<<MD_DISK_FAULTY))
827 set_bit(Faulty, &rdev->flags);
828 else if (desc->state & (1<<MD_DISK_SYNC) /* &&
829 desc->raid_disk < mddev->raid_disks */) {
830 set_bit(In_sync, &rdev->flags);
831 rdev->raid_disk = desc->raid_disk;
833 if (desc->state & (1<<MD_DISK_WRITEMOSTLY))
834 set_bit(WriteMostly, &rdev->flags);
835 } else /* MULTIPATH are always insync */
836 set_bit(In_sync, &rdev->flags);
841 * sync_super for 0.90.0
843 static void super_90_sync(mddev_t *mddev, mdk_rdev_t *rdev)
846 struct list_head *tmp;
848 int next_spare = mddev->raid_disks;
851 /* make rdev->sb match mddev data..
854 * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
855 * 3/ any empty disks < next_spare become removed
857 * disks[0] gets initialised to REMOVED because
858 * we cannot be sure from other fields if it has
859 * been initialised or not.
862 int active=0, working=0,failed=0,spare=0,nr_disks=0;
864 rdev->sb_size = MD_SB_BYTES;
866 sb = (mdp_super_t*)page_address(rdev->sb_page);
868 memset(sb, 0, sizeof(*sb));
870 sb->md_magic = MD_SB_MAGIC;
871 sb->major_version = mddev->major_version;
872 sb->patch_version = mddev->patch_version;
873 sb->gvalid_words = 0; /* ignored */
874 memcpy(&sb->set_uuid0, mddev->uuid+0, 4);
875 memcpy(&sb->set_uuid1, mddev->uuid+4, 4);
876 memcpy(&sb->set_uuid2, mddev->uuid+8, 4);
877 memcpy(&sb->set_uuid3, mddev->uuid+12,4);
879 sb->ctime = mddev->ctime;
880 sb->level = mddev->level;
881 sb->size = mddev->size;
882 sb->raid_disks = mddev->raid_disks;
883 sb->md_minor = mddev->md_minor;
884 sb->not_persistent = !mddev->persistent;
885 sb->utime = mddev->utime;
887 sb->events_hi = (mddev->events>>32);
888 sb->events_lo = (u32)mddev->events;
890 if (mddev->reshape_position == MaxSector)
891 sb->minor_version = 90;
893 sb->minor_version = 91;
894 sb->reshape_position = mddev->reshape_position;
895 sb->new_level = mddev->new_level;
896 sb->delta_disks = mddev->delta_disks;
897 sb->new_layout = mddev->new_layout;
898 sb->new_chunk = mddev->new_chunk;
900 mddev->minor_version = sb->minor_version;
903 sb->recovery_cp = mddev->recovery_cp;
904 sb->cp_events_hi = (mddev->events>>32);
905 sb->cp_events_lo = (u32)mddev->events;
906 if (mddev->recovery_cp == MaxSector)
907 sb->state = (1<< MD_SB_CLEAN);
911 sb->layout = mddev->layout;
912 sb->chunk_size = mddev->chunk_size;
914 if (mddev->bitmap && mddev->bitmap_file == NULL)
915 sb->state |= (1<<MD_SB_BITMAP_PRESENT);
917 sb->disks[0].state = (1<<MD_DISK_REMOVED);
918 ITERATE_RDEV(mddev,rdev2,tmp) {
921 if (rdev2->raid_disk >= 0 && test_bit(In_sync, &rdev2->flags)
922 && !test_bit(Faulty, &rdev2->flags))
923 desc_nr = rdev2->raid_disk;
925 desc_nr = next_spare++;
926 rdev2->desc_nr = desc_nr;
927 d = &sb->disks[rdev2->desc_nr];
929 d->number = rdev2->desc_nr;
930 d->major = MAJOR(rdev2->bdev->bd_dev);
931 d->minor = MINOR(rdev2->bdev->bd_dev);
932 if (rdev2->raid_disk >= 0 && test_bit(In_sync, &rdev2->flags)
933 && !test_bit(Faulty, &rdev2->flags))
934 d->raid_disk = rdev2->raid_disk;
936 d->raid_disk = rdev2->desc_nr; /* compatibility */
937 if (test_bit(Faulty, &rdev2->flags))
938 d->state = (1<<MD_DISK_FAULTY);
939 else if (test_bit(In_sync, &rdev2->flags)) {
940 d->state = (1<<MD_DISK_ACTIVE);
941 d->state |= (1<<MD_DISK_SYNC);
949 if (test_bit(WriteMostly, &rdev2->flags))
950 d->state |= (1<<MD_DISK_WRITEMOSTLY);
952 /* now set the "removed" and "faulty" bits on any missing devices */
953 for (i=0 ; i < mddev->raid_disks ; i++) {
954 mdp_disk_t *d = &sb->disks[i];
955 if (d->state == 0 && d->number == 0) {
958 d->state = (1<<MD_DISK_REMOVED);
959 d->state |= (1<<MD_DISK_FAULTY);
963 sb->nr_disks = nr_disks;
964 sb->active_disks = active;
965 sb->working_disks = working;
966 sb->failed_disks = failed;
967 sb->spare_disks = spare;
969 sb->this_disk = sb->disks[rdev->desc_nr];
970 sb->sb_csum = calc_sb_csum(sb);
974 * version 1 superblock
977 static __le32 calc_sb_1_csum(struct mdp_superblock_1 * sb)
981 unsigned long long newcsum;
982 int size = 256 + le32_to_cpu(sb->max_dev)*2;
983 __le32 *isuper = (__le32*)sb;
986 disk_csum = sb->sb_csum;
989 for (i=0; size>=4; size -= 4 )
990 newcsum += le32_to_cpu(*isuper++);
993 newcsum += le16_to_cpu(*(__le16*) isuper);
995 csum = (newcsum & 0xffffffff) + (newcsum >> 32);
996 sb->sb_csum = disk_csum;
997 return cpu_to_le32(csum);
1000 static int super_1_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
1002 struct mdp_superblock_1 *sb;
1005 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
1009 * Calculate the position of the superblock.
1010 * It is always aligned to a 4K boundary and
1011 * depeding on minor_version, it can be:
1012 * 0: At least 8K, but less than 12K, from end of device
1013 * 1: At start of device
1014 * 2: 4K from start of device.
1016 switch(minor_version) {
1018 sb_offset = rdev->bdev->bd_inode->i_size >> 9;
1020 sb_offset &= ~(sector_t)(4*2-1);
1021 /* convert from sectors to K */
1033 rdev->sb_offset = sb_offset;
1035 /* superblock is rarely larger than 1K, but it can be larger,
1036 * and it is safe to read 4k, so we do that
1038 ret = read_disk_sb(rdev, 4096);
1039 if (ret) return ret;
1042 sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1044 if (sb->magic != cpu_to_le32(MD_SB_MAGIC) ||
1045 sb->major_version != cpu_to_le32(1) ||
1046 le32_to_cpu(sb->max_dev) > (4096-256)/2 ||
1047 le64_to_cpu(sb->super_offset) != (rdev->sb_offset<<1) ||
1048 (le32_to_cpu(sb->feature_map) & ~MD_FEATURE_ALL) != 0)
1051 if (calc_sb_1_csum(sb) != sb->sb_csum) {
1052 printk("md: invalid superblock checksum on %s\n",
1053 bdevname(rdev->bdev,b));
1056 if (le64_to_cpu(sb->data_size) < 10) {
1057 printk("md: data_size too small on %s\n",
1058 bdevname(rdev->bdev,b));
1061 rdev->preferred_minor = 0xffff;
1062 rdev->data_offset = le64_to_cpu(sb->data_offset);
1063 atomic_set(&rdev->corrected_errors, le32_to_cpu(sb->cnt_corrected_read));
1065 rdev->sb_size = le32_to_cpu(sb->max_dev) * 2 + 256;
1066 bmask = queue_hardsect_size(rdev->bdev->bd_disk->queue)-1;
1067 if (rdev->sb_size & bmask)
1068 rdev-> sb_size = (rdev->sb_size | bmask)+1;
1070 if (sb->level == cpu_to_le32(LEVEL_MULTIPATH))
1073 rdev->desc_nr = le32_to_cpu(sb->dev_number);
1079 struct mdp_superblock_1 *refsb =
1080 (struct mdp_superblock_1*)page_address(refdev->sb_page);
1082 if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 ||
1083 sb->level != refsb->level ||
1084 sb->layout != refsb->layout ||
1085 sb->chunksize != refsb->chunksize) {
1086 printk(KERN_WARNING "md: %s has strangely different"
1087 " superblock to %s\n",
1088 bdevname(rdev->bdev,b),
1089 bdevname(refdev->bdev,b2));
1092 ev1 = le64_to_cpu(sb->events);
1093 ev2 = le64_to_cpu(refsb->events);
1101 rdev->size = ((rdev->bdev->bd_inode->i_size>>9) - le64_to_cpu(sb->data_offset)) / 2;
1103 rdev->size = rdev->sb_offset;
1104 if (rdev->size < le64_to_cpu(sb->data_size)/2)
1106 rdev->size = le64_to_cpu(sb->data_size)/2;
1107 if (le32_to_cpu(sb->chunksize))
1108 rdev->size &= ~((sector_t)le32_to_cpu(sb->chunksize)/2 - 1);
1110 if (le64_to_cpu(sb->size) > rdev->size*2)
1115 static int super_1_validate(mddev_t *mddev, mdk_rdev_t *rdev)
1117 struct mdp_superblock_1 *sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1118 __u64 ev1 = le64_to_cpu(sb->events);
1120 rdev->raid_disk = -1;
1122 if (mddev->raid_disks == 0) {
1123 mddev->major_version = 1;
1124 mddev->patch_version = 0;
1125 mddev->persistent = 1;
1126 mddev->chunk_size = le32_to_cpu(sb->chunksize) << 9;
1127 mddev->ctime = le64_to_cpu(sb->ctime) & ((1ULL << 32)-1);
1128 mddev->utime = le64_to_cpu(sb->utime) & ((1ULL << 32)-1);
1129 mddev->level = le32_to_cpu(sb->level);
1130 mddev->clevel[0] = 0;
1131 mddev->layout = le32_to_cpu(sb->layout);
1132 mddev->raid_disks = le32_to_cpu(sb->raid_disks);
1133 mddev->size = le64_to_cpu(sb->size)/2;
1134 mddev->events = ev1;
1135 mddev->bitmap_offset = 0;
1136 mddev->default_bitmap_offset = 1024 >> 9;
1138 mddev->recovery_cp = le64_to_cpu(sb->resync_offset);
1139 memcpy(mddev->uuid, sb->set_uuid, 16);
1141 mddev->max_disks = (4096-256)/2;
1143 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) &&
1144 mddev->bitmap_file == NULL ) {
1145 if (mddev->level != 1 && mddev->level != 5 && mddev->level != 6
1146 && mddev->level != 10) {
1147 printk(KERN_WARNING "md: bitmaps not supported for this level.\n");
1150 mddev->bitmap_offset = (__s32)le32_to_cpu(sb->bitmap_offset);
1152 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) {
1153 mddev->reshape_position = le64_to_cpu(sb->reshape_position);
1154 mddev->delta_disks = le32_to_cpu(sb->delta_disks);
1155 mddev->new_level = le32_to_cpu(sb->new_level);
1156 mddev->new_layout = le32_to_cpu(sb->new_layout);
1157 mddev->new_chunk = le32_to_cpu(sb->new_chunk)<<9;
1159 mddev->reshape_position = MaxSector;
1160 mddev->delta_disks = 0;
1161 mddev->new_level = mddev->level;
1162 mddev->new_layout = mddev->layout;
1163 mddev->new_chunk = mddev->chunk_size;
1166 } else if (mddev->pers == NULL) {
1167 /* Insist of good event counter while assembling */
1169 if (ev1 < mddev->events)
1171 } else if (mddev->bitmap) {
1172 /* If adding to array with a bitmap, then we can accept an
1173 * older device, but not too old.
1175 if (ev1 < mddev->bitmap->events_cleared)
1178 if (ev1 < mddev->events)
1179 /* just a hot-add of a new device, leave raid_disk at -1 */
1182 if (mddev->level != LEVEL_MULTIPATH) {
1184 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
1186 case 0xffff: /* spare */
1188 case 0xfffe: /* faulty */
1189 set_bit(Faulty, &rdev->flags);
1192 if ((le32_to_cpu(sb->feature_map) &
1193 MD_FEATURE_RECOVERY_OFFSET))
1194 rdev->recovery_offset = le64_to_cpu(sb->recovery_offset);
1196 set_bit(In_sync, &rdev->flags);
1197 rdev->raid_disk = role;
1200 if (sb->devflags & WriteMostly1)
1201 set_bit(WriteMostly, &rdev->flags);
1202 } else /* MULTIPATH are always insync */
1203 set_bit(In_sync, &rdev->flags);
1208 static void super_1_sync(mddev_t *mddev, mdk_rdev_t *rdev)
1210 struct mdp_superblock_1 *sb;
1211 struct list_head *tmp;
1214 /* make rdev->sb match mddev and rdev data. */
1216 sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1218 sb->feature_map = 0;
1220 sb->recovery_offset = cpu_to_le64(0);
1221 memset(sb->pad1, 0, sizeof(sb->pad1));
1222 memset(sb->pad2, 0, sizeof(sb->pad2));
1223 memset(sb->pad3, 0, sizeof(sb->pad3));
1225 sb->utime = cpu_to_le64((__u64)mddev->utime);
1226 sb->events = cpu_to_le64(mddev->events);
1228 sb->resync_offset = cpu_to_le64(mddev->recovery_cp);
1230 sb->resync_offset = cpu_to_le64(0);
1232 sb->cnt_corrected_read = cpu_to_le32(atomic_read(&rdev->corrected_errors));
1234 sb->raid_disks = cpu_to_le32(mddev->raid_disks);
1235 sb->size = cpu_to_le64(mddev->size<<1);
1237 if (mddev->bitmap && mddev->bitmap_file == NULL) {
1238 sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_offset);
1239 sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET);
1242 if (rdev->raid_disk >= 0 &&
1243 !test_bit(In_sync, &rdev->flags) &&
1244 rdev->recovery_offset > 0) {
1245 sb->feature_map |= cpu_to_le32(MD_FEATURE_RECOVERY_OFFSET);
1246 sb->recovery_offset = cpu_to_le64(rdev->recovery_offset);
1249 if (mddev->reshape_position != MaxSector) {
1250 sb->feature_map |= cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE);
1251 sb->reshape_position = cpu_to_le64(mddev->reshape_position);
1252 sb->new_layout = cpu_to_le32(mddev->new_layout);
1253 sb->delta_disks = cpu_to_le32(mddev->delta_disks);
1254 sb->new_level = cpu_to_le32(mddev->new_level);
1255 sb->new_chunk = cpu_to_le32(mddev->new_chunk>>9);
1259 ITERATE_RDEV(mddev,rdev2,tmp)
1260 if (rdev2->desc_nr+1 > max_dev)
1261 max_dev = rdev2->desc_nr+1;
1263 sb->max_dev = cpu_to_le32(max_dev);
1264 for (i=0; i<max_dev;i++)
1265 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1267 ITERATE_RDEV(mddev,rdev2,tmp) {
1269 if (test_bit(Faulty, &rdev2->flags))
1270 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1271 else if (test_bit(In_sync, &rdev2->flags))
1272 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1273 else if (rdev2->raid_disk >= 0 && rdev2->recovery_offset > 0)
1274 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1276 sb->dev_roles[i] = cpu_to_le16(0xffff);
1279 sb->sb_csum = calc_sb_1_csum(sb);
1283 static struct super_type super_types[] = {
1286 .owner = THIS_MODULE,
1287 .load_super = super_90_load,
1288 .validate_super = super_90_validate,
1289 .sync_super = super_90_sync,
1293 .owner = THIS_MODULE,
1294 .load_super = super_1_load,
1295 .validate_super = super_1_validate,
1296 .sync_super = super_1_sync,
1300 static mdk_rdev_t * match_dev_unit(mddev_t *mddev, mdk_rdev_t *dev)
1302 struct list_head *tmp;
1305 ITERATE_RDEV(mddev,rdev,tmp)
1306 if (rdev->bdev->bd_contains == dev->bdev->bd_contains)
1312 static int match_mddev_units(mddev_t *mddev1, mddev_t *mddev2)
1314 struct list_head *tmp;
1317 ITERATE_RDEV(mddev1,rdev,tmp)
1318 if (match_dev_unit(mddev2, rdev))
1324 static LIST_HEAD(pending_raid_disks);
1326 static int bind_rdev_to_array(mdk_rdev_t * rdev, mddev_t * mddev)
1328 mdk_rdev_t *same_pdev;
1329 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
1337 /* make sure rdev->size exceeds mddev->size */
1338 if (rdev->size && (mddev->size == 0 || rdev->size < mddev->size)) {
1340 /* Cannot change size, so fail */
1343 mddev->size = rdev->size;
1345 same_pdev = match_dev_unit(mddev, rdev);
1348 "%s: WARNING: %s appears to be on the same physical"
1349 " disk as %s. True\n protection against single-disk"
1350 " failure might be compromised.\n",
1351 mdname(mddev), bdevname(rdev->bdev,b),
1352 bdevname(same_pdev->bdev,b2));
1354 /* Verify rdev->desc_nr is unique.
1355 * If it is -1, assign a free number, else
1356 * check number is not in use
1358 if (rdev->desc_nr < 0) {
1360 if (mddev->pers) choice = mddev->raid_disks;
1361 while (find_rdev_nr(mddev, choice))
1363 rdev->desc_nr = choice;
1365 if (find_rdev_nr(mddev, rdev->desc_nr))
1368 bdevname(rdev->bdev,b);
1369 if (kobject_set_name(&rdev->kobj, "dev-%s", b) < 0)
1371 while ( (s=strchr(rdev->kobj.k_name, '/')) != NULL)
1374 list_add(&rdev->same_set, &mddev->disks);
1375 rdev->mddev = mddev;
1376 printk(KERN_INFO "md: bind<%s>\n", b);
1378 rdev->kobj.parent = &mddev->kobj;
1379 kobject_add(&rdev->kobj);
1381 if (rdev->bdev->bd_part)
1382 ko = &rdev->bdev->bd_part->kobj;
1384 ko = &rdev->bdev->bd_disk->kobj;
1385 sysfs_create_link(&rdev->kobj, ko, "block");
1386 bd_claim_by_disk(rdev->bdev, rdev, mddev->gendisk);
1390 static void unbind_rdev_from_array(mdk_rdev_t * rdev)
1392 char b[BDEVNAME_SIZE];
1397 bd_release_from_disk(rdev->bdev, rdev->mddev->gendisk);
1398 list_del_init(&rdev->same_set);
1399 printk(KERN_INFO "md: unbind<%s>\n", bdevname(rdev->bdev,b));
1401 sysfs_remove_link(&rdev->kobj, "block");
1402 kobject_del(&rdev->kobj);
1406 * prevent the device from being mounted, repartitioned or
1407 * otherwise reused by a RAID array (or any other kernel
1408 * subsystem), by bd_claiming the device.
1410 static int lock_rdev(mdk_rdev_t *rdev, dev_t dev)
1413 struct block_device *bdev;
1414 char b[BDEVNAME_SIZE];
1416 bdev = open_by_devnum(dev, FMODE_READ|FMODE_WRITE);
1418 printk(KERN_ERR "md: could not open %s.\n",
1419 __bdevname(dev, b));
1420 return PTR_ERR(bdev);
1422 err = bd_claim(bdev, rdev);
1424 printk(KERN_ERR "md: could not bd_claim %s.\n",
1433 static void unlock_rdev(mdk_rdev_t *rdev)
1435 struct block_device *bdev = rdev->bdev;
1443 void md_autodetect_dev(dev_t dev);
1445 static void export_rdev(mdk_rdev_t * rdev)
1447 char b[BDEVNAME_SIZE];
1448 printk(KERN_INFO "md: export_rdev(%s)\n",
1449 bdevname(rdev->bdev,b));
1453 list_del_init(&rdev->same_set);
1455 md_autodetect_dev(rdev->bdev->bd_dev);
1458 kobject_put(&rdev->kobj);
1461 static void kick_rdev_from_array(mdk_rdev_t * rdev)
1463 unbind_rdev_from_array(rdev);
1467 static void export_array(mddev_t *mddev)
1469 struct list_head *tmp;
1472 ITERATE_RDEV(mddev,rdev,tmp) {
1477 kick_rdev_from_array(rdev);
1479 if (!list_empty(&mddev->disks))
1481 mddev->raid_disks = 0;
1482 mddev->major_version = 0;
1485 static void print_desc(mdp_disk_t *desc)
1487 printk(" DISK<N:%d,(%d,%d),R:%d,S:%d>\n", desc->number,
1488 desc->major,desc->minor,desc->raid_disk,desc->state);
1491 static void print_sb(mdp_super_t *sb)
1496 "md: SB: (V:%d.%d.%d) ID:<%08x.%08x.%08x.%08x> CT:%08x\n",
1497 sb->major_version, sb->minor_version, sb->patch_version,
1498 sb->set_uuid0, sb->set_uuid1, sb->set_uuid2, sb->set_uuid3,
1500 printk(KERN_INFO "md: L%d S%08d ND:%d RD:%d md%d LO:%d CS:%d\n",
1501 sb->level, sb->size, sb->nr_disks, sb->raid_disks,
1502 sb->md_minor, sb->layout, sb->chunk_size);
1503 printk(KERN_INFO "md: UT:%08x ST:%d AD:%d WD:%d"
1504 " FD:%d SD:%d CSUM:%08x E:%08lx\n",
1505 sb->utime, sb->state, sb->active_disks, sb->working_disks,
1506 sb->failed_disks, sb->spare_disks,
1507 sb->sb_csum, (unsigned long)sb->events_lo);
1510 for (i = 0; i < MD_SB_DISKS; i++) {
1513 desc = sb->disks + i;
1514 if (desc->number || desc->major || desc->minor ||
1515 desc->raid_disk || (desc->state && (desc->state != 4))) {
1516 printk(" D %2d: ", i);
1520 printk(KERN_INFO "md: THIS: ");
1521 print_desc(&sb->this_disk);
1525 static void print_rdev(mdk_rdev_t *rdev)
1527 char b[BDEVNAME_SIZE];
1528 printk(KERN_INFO "md: rdev %s, SZ:%08llu F:%d S:%d DN:%u\n",
1529 bdevname(rdev->bdev,b), (unsigned long long)rdev->size,
1530 test_bit(Faulty, &rdev->flags), test_bit(In_sync, &rdev->flags),
1532 if (rdev->sb_loaded) {
1533 printk(KERN_INFO "md: rdev superblock:\n");
1534 print_sb((mdp_super_t*)page_address(rdev->sb_page));
1536 printk(KERN_INFO "md: no rdev superblock!\n");
1539 static void md_print_devices(void)
1541 struct list_head *tmp, *tmp2;
1544 char b[BDEVNAME_SIZE];
1547 printk("md: **********************************\n");
1548 printk("md: * <COMPLETE RAID STATE PRINTOUT> *\n");
1549 printk("md: **********************************\n");
1550 ITERATE_MDDEV(mddev,tmp) {
1553 bitmap_print_sb(mddev->bitmap);
1555 printk("%s: ", mdname(mddev));
1556 ITERATE_RDEV(mddev,rdev,tmp2)
1557 printk("<%s>", bdevname(rdev->bdev,b));
1560 ITERATE_RDEV(mddev,rdev,tmp2)
1563 printk("md: **********************************\n");
1568 static void sync_sbs(mddev_t * mddev, int nospares)
1570 /* Update each superblock (in-memory image), but
1571 * if we are allowed to, skip spares which already
1572 * have the right event counter, or have one earlier
1573 * (which would mean they aren't being marked as dirty
1574 * with the rest of the array)
1577 struct list_head *tmp;
1579 ITERATE_RDEV(mddev,rdev,tmp) {
1580 if (rdev->sb_events == mddev->events ||
1582 rdev->raid_disk < 0 &&
1583 (rdev->sb_events&1)==0 &&
1584 rdev->sb_events+1 == mddev->events)) {
1585 /* Don't update this superblock */
1586 rdev->sb_loaded = 2;
1588 super_types[mddev->major_version].
1589 sync_super(mddev, rdev);
1590 rdev->sb_loaded = 1;
1595 static void md_update_sb(mddev_t * mddev, int force_change)
1598 struct list_head *tmp;
1604 spin_lock_irq(&mddev->write_lock);
1606 set_bit(MD_CHANGE_PENDING, &mddev->flags);
1607 if (test_and_clear_bit(MD_CHANGE_DEVS, &mddev->flags))
1609 if (test_and_clear_bit(MD_CHANGE_CLEAN, &mddev->flags))
1610 /* just a clean<-> dirty transition, possibly leave spares alone,
1611 * though if events isn't the right even/odd, we will have to do
1617 if (mddev->degraded)
1618 /* If the array is degraded, then skipping spares is both
1619 * dangerous and fairly pointless.
1620 * Dangerous because a device that was removed from the array
1621 * might have a event_count that still looks up-to-date,
1622 * so it can be re-added without a resync.
1623 * Pointless because if there are any spares to skip,
1624 * then a recovery will happen and soon that array won't
1625 * be degraded any more and the spare can go back to sleep then.
1629 sync_req = mddev->in_sync;
1630 mddev->utime = get_seconds();
1632 /* If this is just a dirty<->clean transition, and the array is clean
1633 * and 'events' is odd, we can roll back to the previous clean state */
1635 && (mddev->in_sync && mddev->recovery_cp == MaxSector)
1636 && (mddev->events & 1)
1637 && mddev->events != 1)
1640 /* otherwise we have to go forward and ... */
1642 if (!mddev->in_sync || mddev->recovery_cp != MaxSector) { /* not clean */
1643 /* .. if the array isn't clean, insist on an odd 'events' */
1644 if ((mddev->events&1)==0) {
1649 /* otherwise insist on an even 'events' (for clean states) */
1650 if ((mddev->events&1)) {
1657 if (!mddev->events) {
1659 * oops, this 64-bit counter should never wrap.
1660 * Either we are in around ~1 trillion A.C., assuming
1661 * 1 reboot per second, or we have a bug:
1666 sync_sbs(mddev, nospares);
1669 * do not write anything to disk if using
1670 * nonpersistent superblocks
1672 if (!mddev->persistent) {
1673 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
1674 spin_unlock_irq(&mddev->write_lock);
1675 wake_up(&mddev->sb_wait);
1678 spin_unlock_irq(&mddev->write_lock);
1681 "md: updating %s RAID superblock on device (in sync %d)\n",
1682 mdname(mddev),mddev->in_sync);
1684 err = bitmap_update_sb(mddev->bitmap);
1685 ITERATE_RDEV(mddev,rdev,tmp) {
1686 char b[BDEVNAME_SIZE];
1687 dprintk(KERN_INFO "md: ");
1688 if (rdev->sb_loaded != 1)
1689 continue; /* no noise on spare devices */
1690 if (test_bit(Faulty, &rdev->flags))
1691 dprintk("(skipping faulty ");
1693 dprintk("%s ", bdevname(rdev->bdev,b));
1694 if (!test_bit(Faulty, &rdev->flags)) {
1695 md_super_write(mddev,rdev,
1696 rdev->sb_offset<<1, rdev->sb_size,
1698 dprintk(KERN_INFO "(write) %s's sb offset: %llu\n",
1699 bdevname(rdev->bdev,b),
1700 (unsigned long long)rdev->sb_offset);
1701 rdev->sb_events = mddev->events;
1705 if (mddev->level == LEVEL_MULTIPATH)
1706 /* only need to write one superblock... */
1709 md_super_wait(mddev);
1710 /* if there was a failure, MD_CHANGE_DEVS was set, and we re-write super */
1712 spin_lock_irq(&mddev->write_lock);
1713 if (mddev->in_sync != sync_req ||
1714 test_bit(MD_CHANGE_DEVS, &mddev->flags)) {
1715 /* have to write it out again */
1716 spin_unlock_irq(&mddev->write_lock);
1719 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
1720 spin_unlock_irq(&mddev->write_lock);
1721 wake_up(&mddev->sb_wait);
1725 /* words written to sysfs files may, or my not, be \n terminated.
1726 * We want to accept with case. For this we use cmd_match.
1728 static int cmd_match(const char *cmd, const char *str)
1730 /* See if cmd, written into a sysfs file, matches
1731 * str. They must either be the same, or cmd can
1732 * have a trailing newline
1734 while (*cmd && *str && *cmd == *str) {
1745 struct rdev_sysfs_entry {
1746 struct attribute attr;
1747 ssize_t (*show)(mdk_rdev_t *, char *);
1748 ssize_t (*store)(mdk_rdev_t *, const char *, size_t);
1752 state_show(mdk_rdev_t *rdev, char *page)
1757 if (test_bit(Faulty, &rdev->flags)) {
1758 len+= sprintf(page+len, "%sfaulty",sep);
1761 if (test_bit(In_sync, &rdev->flags)) {
1762 len += sprintf(page+len, "%sin_sync",sep);
1765 if (test_bit(WriteMostly, &rdev->flags)) {
1766 len += sprintf(page+len, "%swrite_mostly",sep);
1769 if (!test_bit(Faulty, &rdev->flags) &&
1770 !test_bit(In_sync, &rdev->flags)) {
1771 len += sprintf(page+len, "%sspare", sep);
1774 return len+sprintf(page+len, "\n");
1778 state_store(mdk_rdev_t *rdev, const char *buf, size_t len)
1781 * faulty - simulates and error
1782 * remove - disconnects the device
1783 * writemostly - sets write_mostly
1784 * -writemostly - clears write_mostly
1787 if (cmd_match(buf, "faulty") && rdev->mddev->pers) {
1788 md_error(rdev->mddev, rdev);
1790 } else if (cmd_match(buf, "remove")) {
1791 if (rdev->raid_disk >= 0)
1794 mddev_t *mddev = rdev->mddev;
1795 kick_rdev_from_array(rdev);
1797 md_update_sb(mddev, 1);
1798 md_new_event(mddev);
1801 } else if (cmd_match(buf, "writemostly")) {
1802 set_bit(WriteMostly, &rdev->flags);
1804 } else if (cmd_match(buf, "-writemostly")) {
1805 clear_bit(WriteMostly, &rdev->flags);
1808 return err ? err : len;
1810 static struct rdev_sysfs_entry rdev_state =
1811 __ATTR(state, S_IRUGO|S_IWUSR, state_show, state_store);
1814 super_show(mdk_rdev_t *rdev, char *page)
1816 if (rdev->sb_loaded && rdev->sb_size) {
1817 memcpy(page, page_address(rdev->sb_page), rdev->sb_size);
1818 return rdev->sb_size;
1822 static struct rdev_sysfs_entry rdev_super = __ATTR_RO(super);
1825 errors_show(mdk_rdev_t *rdev, char *page)
1827 return sprintf(page, "%d\n", atomic_read(&rdev->corrected_errors));
1831 errors_store(mdk_rdev_t *rdev, const char *buf, size_t len)
1834 unsigned long n = simple_strtoul(buf, &e, 10);
1835 if (*buf && (*e == 0 || *e == '\n')) {
1836 atomic_set(&rdev->corrected_errors, n);
1841 static struct rdev_sysfs_entry rdev_errors =
1842 __ATTR(errors, S_IRUGO|S_IWUSR, errors_show, errors_store);
1845 slot_show(mdk_rdev_t *rdev, char *page)
1847 if (rdev->raid_disk < 0)
1848 return sprintf(page, "none\n");
1850 return sprintf(page, "%d\n", rdev->raid_disk);
1854 slot_store(mdk_rdev_t *rdev, const char *buf, size_t len)
1857 int slot = simple_strtoul(buf, &e, 10);
1858 if (strncmp(buf, "none", 4)==0)
1860 else if (e==buf || (*e && *e!= '\n'))
1862 if (rdev->mddev->pers)
1863 /* Cannot set slot in active array (yet) */
1865 if (slot >= rdev->mddev->raid_disks)
1867 rdev->raid_disk = slot;
1868 /* assume it is working */
1870 set_bit(In_sync, &rdev->flags);
1875 static struct rdev_sysfs_entry rdev_slot =
1876 __ATTR(slot, S_IRUGO|S_IWUSR, slot_show, slot_store);
1879 offset_show(mdk_rdev_t *rdev, char *page)
1881 return sprintf(page, "%llu\n", (unsigned long long)rdev->data_offset);
1885 offset_store(mdk_rdev_t *rdev, const char *buf, size_t len)
1888 unsigned long long offset = simple_strtoull(buf, &e, 10);
1889 if (e==buf || (*e && *e != '\n'))
1891 if (rdev->mddev->pers)
1893 rdev->data_offset = offset;
1897 static struct rdev_sysfs_entry rdev_offset =
1898 __ATTR(offset, S_IRUGO|S_IWUSR, offset_show, offset_store);
1901 rdev_size_show(mdk_rdev_t *rdev, char *page)
1903 return sprintf(page, "%llu\n", (unsigned long long)rdev->size);
1907 rdev_size_store(mdk_rdev_t *rdev, const char *buf, size_t len)
1910 unsigned long long size = simple_strtoull(buf, &e, 10);
1911 if (e==buf || (*e && *e != '\n'))
1913 if (rdev->mddev->pers)
1916 if (size < rdev->mddev->size || rdev->mddev->size == 0)
1917 rdev->mddev->size = size;
1921 static struct rdev_sysfs_entry rdev_size =
1922 __ATTR(size, S_IRUGO|S_IWUSR, rdev_size_show, rdev_size_store);
1924 static struct attribute *rdev_default_attrs[] = {
1934 rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
1936 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
1937 mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
1941 return entry->show(rdev, page);
1945 rdev_attr_store(struct kobject *kobj, struct attribute *attr,
1946 const char *page, size_t length)
1948 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
1949 mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
1953 if (!capable(CAP_SYS_ADMIN))
1955 return entry->store(rdev, page, length);
1958 static void rdev_free(struct kobject *ko)
1960 mdk_rdev_t *rdev = container_of(ko, mdk_rdev_t, kobj);
1963 static struct sysfs_ops rdev_sysfs_ops = {
1964 .show = rdev_attr_show,
1965 .store = rdev_attr_store,
1967 static struct kobj_type rdev_ktype = {
1968 .release = rdev_free,
1969 .sysfs_ops = &rdev_sysfs_ops,
1970 .default_attrs = rdev_default_attrs,
1974 * Import a device. If 'super_format' >= 0, then sanity check the superblock
1976 * mark the device faulty if:
1978 * - the device is nonexistent (zero size)
1979 * - the device has no valid superblock
1981 * a faulty rdev _never_ has rdev->sb set.
1983 static mdk_rdev_t *md_import_device(dev_t newdev, int super_format, int super_minor)
1985 char b[BDEVNAME_SIZE];
1990 rdev = kzalloc(sizeof(*rdev), GFP_KERNEL);
1992 printk(KERN_ERR "md: could not alloc mem for new device!\n");
1993 return ERR_PTR(-ENOMEM);
1996 if ((err = alloc_disk_sb(rdev)))
1999 err = lock_rdev(rdev, newdev);
2003 rdev->kobj.parent = NULL;
2004 rdev->kobj.ktype = &rdev_ktype;
2005 kobject_init(&rdev->kobj);
2008 rdev->saved_raid_disk = -1;
2009 rdev->raid_disk = -1;
2011 rdev->data_offset = 0;
2012 rdev->sb_events = 0;
2013 atomic_set(&rdev->nr_pending, 0);
2014 atomic_set(&rdev->read_errors, 0);
2015 atomic_set(&rdev->corrected_errors, 0);
2017 size = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
2020 "md: %s has zero or unknown size, marking faulty!\n",
2021 bdevname(rdev->bdev,b));
2026 if (super_format >= 0) {
2027 err = super_types[super_format].
2028 load_super(rdev, NULL, super_minor);
2029 if (err == -EINVAL) {
2031 "md: %s has invalid sb, not importing!\n",
2032 bdevname(rdev->bdev,b));
2037 "md: could not read %s's sb, not importing!\n",
2038 bdevname(rdev->bdev,b));
2042 INIT_LIST_HEAD(&rdev->same_set);
2047 if (rdev->sb_page) {
2053 return ERR_PTR(err);
2057 * Check a full RAID array for plausibility
2061 static void analyze_sbs(mddev_t * mddev)
2064 struct list_head *tmp;
2065 mdk_rdev_t *rdev, *freshest;
2066 char b[BDEVNAME_SIZE];
2069 ITERATE_RDEV(mddev,rdev,tmp)
2070 switch (super_types[mddev->major_version].
2071 load_super(rdev, freshest, mddev->minor_version)) {
2079 "md: fatal superblock inconsistency in %s"
2080 " -- removing from array\n",
2081 bdevname(rdev->bdev,b));
2082 kick_rdev_from_array(rdev);
2086 super_types[mddev->major_version].
2087 validate_super(mddev, freshest);
2090 ITERATE_RDEV(mddev,rdev,tmp) {
2091 if (rdev != freshest)
2092 if (super_types[mddev->major_version].
2093 validate_super(mddev, rdev)) {
2094 printk(KERN_WARNING "md: kicking non-fresh %s"
2096 bdevname(rdev->bdev,b));
2097 kick_rdev_from_array(rdev);
2100 if (mddev->level == LEVEL_MULTIPATH) {
2101 rdev->desc_nr = i++;
2102 rdev->raid_disk = rdev->desc_nr;
2103 set_bit(In_sync, &rdev->flags);
2109 if (mddev->recovery_cp != MaxSector &&
2111 printk(KERN_ERR "md: %s: raid array is not clean"
2112 " -- starting background reconstruction\n",
2118 safe_delay_show(mddev_t *mddev, char *page)
2120 int msec = (mddev->safemode_delay*1000)/HZ;
2121 return sprintf(page, "%d.%03d\n", msec/1000, msec%1000);
2124 safe_delay_store(mddev_t *mddev, const char *cbuf, size_t len)
2132 /* remove a period, and count digits after it */
2133 if (len >= sizeof(buf))
2135 strlcpy(buf, cbuf, len);
2137 for (i=0; i<len; i++) {
2139 if (isdigit(buf[i])) {
2144 } else if (buf[i] == '.') {
2149 msec = simple_strtoul(buf, &e, 10);
2150 if (e == buf || (*e && *e != '\n'))
2152 msec = (msec * 1000) / scale;
2154 mddev->safemode_delay = 0;
2156 mddev->safemode_delay = (msec*HZ)/1000;
2157 if (mddev->safemode_delay == 0)
2158 mddev->safemode_delay = 1;
2162 static struct md_sysfs_entry md_safe_delay =
2163 __ATTR(safe_mode_delay, S_IRUGO|S_IWUSR,safe_delay_show, safe_delay_store);
2166 level_show(mddev_t *mddev, char *page)
2168 struct mdk_personality *p = mddev->pers;
2170 return sprintf(page, "%s\n", p->name);
2171 else if (mddev->clevel[0])
2172 return sprintf(page, "%s\n", mddev->clevel);
2173 else if (mddev->level != LEVEL_NONE)
2174 return sprintf(page, "%d\n", mddev->level);
2180 level_store(mddev_t *mddev, const char *buf, size_t len)
2187 if (len >= sizeof(mddev->clevel))
2189 strncpy(mddev->clevel, buf, len);
2190 if (mddev->clevel[len-1] == '\n')
2192 mddev->clevel[len] = 0;
2193 mddev->level = LEVEL_NONE;
2197 static struct md_sysfs_entry md_level =
2198 __ATTR(level, S_IRUGO|S_IWUSR, level_show, level_store);
2202 layout_show(mddev_t *mddev, char *page)
2204 /* just a number, not meaningful for all levels */
2205 return sprintf(page, "%d\n", mddev->layout);
2209 layout_store(mddev_t *mddev, const char *buf, size_t len)
2212 unsigned long n = simple_strtoul(buf, &e, 10);
2216 if (!*buf || (*e && *e != '\n'))
2222 static struct md_sysfs_entry md_layout =
2223 __ATTR(layout, S_IRUGO|S_IWUSR, layout_show, layout_store);
2227 raid_disks_show(mddev_t *mddev, char *page)
2229 if (mddev->raid_disks == 0)
2231 return sprintf(page, "%d\n", mddev->raid_disks);
2234 static int update_raid_disks(mddev_t *mddev, int raid_disks);
2237 raid_disks_store(mddev_t *mddev, const char *buf, size_t len)
2241 unsigned long n = simple_strtoul(buf, &e, 10);
2243 if (!*buf || (*e && *e != '\n'))
2247 rv = update_raid_disks(mddev, n);
2249 mddev->raid_disks = n;
2250 return rv ? rv : len;
2252 static struct md_sysfs_entry md_raid_disks =
2253 __ATTR(raid_disks, S_IRUGO|S_IWUSR, raid_disks_show, raid_disks_store);
2256 chunk_size_show(mddev_t *mddev, char *page)
2258 return sprintf(page, "%d\n", mddev->chunk_size);
2262 chunk_size_store(mddev_t *mddev, const char *buf, size_t len)
2264 /* can only set chunk_size if array is not yet active */
2266 unsigned long n = simple_strtoul(buf, &e, 10);
2270 if (!*buf || (*e && *e != '\n'))
2273 mddev->chunk_size = n;
2276 static struct md_sysfs_entry md_chunk_size =
2277 __ATTR(chunk_size, S_IRUGO|S_IWUSR, chunk_size_show, chunk_size_store);
2280 resync_start_show(mddev_t *mddev, char *page)
2282 return sprintf(page, "%llu\n", (unsigned long long)mddev->recovery_cp);
2286 resync_start_store(mddev_t *mddev, const char *buf, size_t len)
2288 /* can only set chunk_size if array is not yet active */
2290 unsigned long long n = simple_strtoull(buf, &e, 10);
2294 if (!*buf || (*e && *e != '\n'))
2297 mddev->recovery_cp = n;
2300 static struct md_sysfs_entry md_resync_start =
2301 __ATTR(resync_start, S_IRUGO|S_IWUSR, resync_start_show, resync_start_store);
2304 * The array state can be:
2307 * No devices, no size, no level
2308 * Equivalent to STOP_ARRAY ioctl
2310 * May have some settings, but array is not active
2311 * all IO results in error
2312 * When written, doesn't tear down array, but just stops it
2313 * suspended (not supported yet)
2314 * All IO requests will block. The array can be reconfigured.
2315 * Writing this, if accepted, will block until array is quiessent
2317 * no resync can happen. no superblocks get written.
2318 * write requests fail
2320 * like readonly, but behaves like 'clean' on a write request.
2322 * clean - no pending writes, but otherwise active.
2323 * When written to inactive array, starts without resync
2324 * If a write request arrives then
2325 * if metadata is known, mark 'dirty' and switch to 'active'.
2326 * if not known, block and switch to write-pending
2327 * If written to an active array that has pending writes, then fails.
2329 * fully active: IO and resync can be happening.
2330 * When written to inactive array, starts with resync
2333 * clean, but writes are blocked waiting for 'active' to be written.
2336 * like active, but no writes have been seen for a while (100msec).
2339 enum array_state { clear, inactive, suspended, readonly, read_auto, clean, active,
2340 write_pending, active_idle, bad_word};
2341 static char *array_states[] = {
2342 "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active",
2343 "write-pending", "active-idle", NULL };
2345 static int match_word(const char *word, char **list)
2348 for (n=0; list[n]; n++)
2349 if (cmd_match(word, list[n]))
2355 array_state_show(mddev_t *mddev, char *page)
2357 enum array_state st = inactive;
2370 else if (mddev->safemode)
2376 if (list_empty(&mddev->disks) &&
2377 mddev->raid_disks == 0 &&
2383 return sprintf(page, "%s\n", array_states[st]);
2386 static int do_md_stop(mddev_t * mddev, int ro);
2387 static int do_md_run(mddev_t * mddev);
2388 static int restart_array(mddev_t *mddev);
2391 array_state_store(mddev_t *mddev, const char *buf, size_t len)
2394 enum array_state st = match_word(buf, array_states);
2399 /* stopping an active array */
2401 if (atomic_read(&mddev->active) > 1)
2403 err = do_md_stop(mddev, 0);
2407 /* stopping an active array */
2409 if (atomic_read(&mddev->active) > 1)
2411 err = do_md_stop(mddev, 2);
2415 break; /* not supported yet */
2418 err = do_md_stop(mddev, 1);
2421 err = do_md_run(mddev);
2425 /* stopping an active array */
2427 err = do_md_stop(mddev, 1);
2429 mddev->ro = 2; /* FIXME mark devices writable */
2432 err = do_md_run(mddev);
2437 restart_array(mddev);
2438 spin_lock_irq(&mddev->write_lock);
2439 if (atomic_read(&mddev->writes_pending) == 0) {
2441 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
2443 spin_unlock_irq(&mddev->write_lock);
2446 mddev->recovery_cp = MaxSector;
2447 err = do_md_run(mddev);
2452 restart_array(mddev);
2453 clear_bit(MD_CHANGE_CLEAN, &mddev->flags);
2454 wake_up(&mddev->sb_wait);
2458 err = do_md_run(mddev);
2463 /* these cannot be set */
2471 static struct md_sysfs_entry md_array_state =
2472 __ATTR(array_state, S_IRUGO|S_IWUSR, array_state_show, array_state_store);
2475 null_show(mddev_t *mddev, char *page)
2481 new_dev_store(mddev_t *mddev, const char *buf, size_t len)
2483 /* buf must be %d:%d\n? giving major and minor numbers */
2484 /* The new device is added to the array.
2485 * If the array has a persistent superblock, we read the
2486 * superblock to initialise info and check validity.
2487 * Otherwise, only checking done is that in bind_rdev_to_array,
2488 * which mainly checks size.
2491 int major = simple_strtoul(buf, &e, 10);
2497 if (!*buf || *e != ':' || !e[1] || e[1] == '\n')
2499 minor = simple_strtoul(e+1, &e, 10);
2500 if (*e && *e != '\n')
2502 dev = MKDEV(major, minor);
2503 if (major != MAJOR(dev) ||
2504 minor != MINOR(dev))
2508 if (mddev->persistent) {
2509 rdev = md_import_device(dev, mddev->major_version,
2510 mddev->minor_version);
2511 if (!IS_ERR(rdev) && !list_empty(&mddev->disks)) {
2512 mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
2513 mdk_rdev_t, same_set);
2514 err = super_types[mddev->major_version]
2515 .load_super(rdev, rdev0, mddev->minor_version);
2520 rdev = md_import_device(dev, -1, -1);
2523 return PTR_ERR(rdev);
2524 err = bind_rdev_to_array(rdev, mddev);
2528 return err ? err : len;
2531 static struct md_sysfs_entry md_new_device =
2532 __ATTR(new_dev, S_IWUSR, null_show, new_dev_store);
2535 bitmap_store(mddev_t *mddev, const char *buf, size_t len)
2538 unsigned long chunk, end_chunk;
2542 /* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */
2544 chunk = end_chunk = simple_strtoul(buf, &end, 0);
2545 if (buf == end) break;
2546 if (*end == '-') { /* range */
2548 end_chunk = simple_strtoul(buf, &end, 0);
2549 if (buf == end) break;
2551 if (*end && !isspace(*end)) break;
2552 bitmap_dirty_bits(mddev->bitmap, chunk, end_chunk);
2554 while (isspace(*buf)) buf++;
2556 bitmap_unplug(mddev->bitmap); /* flush the bits to disk */
2561 static struct md_sysfs_entry md_bitmap =
2562 __ATTR(bitmap_set_bits, S_IWUSR, null_show, bitmap_store);
2565 size_show(mddev_t *mddev, char *page)
2567 return sprintf(page, "%llu\n", (unsigned long long)mddev->size);
2570 static int update_size(mddev_t *mddev, unsigned long size);
2573 size_store(mddev_t *mddev, const char *buf, size_t len)
2575 /* If array is inactive, we can reduce the component size, but
2576 * not increase it (except from 0).
2577 * If array is active, we can try an on-line resize
2581 unsigned long long size = simple_strtoull(buf, &e, 10);
2582 if (!*buf || *buf == '\n' ||
2587 err = update_size(mddev, size);
2588 md_update_sb(mddev, 1);
2590 if (mddev->size == 0 ||
2596 return err ? err : len;
2599 static struct md_sysfs_entry md_size =
2600 __ATTR(component_size, S_IRUGO|S_IWUSR, size_show, size_store);
2604 * This is either 'none' for arrays with externally managed metadata,
2605 * or N.M for internally known formats
2608 metadata_show(mddev_t *mddev, char *page)
2610 if (mddev->persistent)
2611 return sprintf(page, "%d.%d\n",
2612 mddev->major_version, mddev->minor_version);
2614 return sprintf(page, "none\n");
2618 metadata_store(mddev_t *mddev, const char *buf, size_t len)
2622 if (!list_empty(&mddev->disks))
2625 if (cmd_match(buf, "none")) {
2626 mddev->persistent = 0;
2627 mddev->major_version = 0;
2628 mddev->minor_version = 90;
2631 major = simple_strtoul(buf, &e, 10);
2632 if (e==buf || *e != '.')
2635 minor = simple_strtoul(buf, &e, 10);
2636 if (e==buf || (*e && *e != '\n') )
2638 if (major >= sizeof(super_types)/sizeof(super_types[0]) ||
2639 super_types[major].name == NULL)
2641 mddev->major_version = major;
2642 mddev->minor_version = minor;
2643 mddev->persistent = 1;
2647 static struct md_sysfs_entry md_metadata =
2648 __ATTR(metadata_version, S_IRUGO|S_IWUSR, metadata_show, metadata_store);
2651 action_show(mddev_t *mddev, char *page)
2653 char *type = "idle";
2654 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
2655 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery)) {
2656 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
2658 else if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
2659 if (!test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
2661 else if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
2668 return sprintf(page, "%s\n", type);
2672 action_store(mddev_t *mddev, const char *page, size_t len)
2674 if (!mddev->pers || !mddev->pers->sync_request)
2677 if (cmd_match(page, "idle")) {
2678 if (mddev->sync_thread) {
2679 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
2680 md_unregister_thread(mddev->sync_thread);
2681 mddev->sync_thread = NULL;
2682 mddev->recovery = 0;
2684 } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
2685 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
2687 else if (cmd_match(page, "resync") || cmd_match(page, "recover"))
2688 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2689 else if (cmd_match(page, "reshape")) {
2691 if (mddev->pers->start_reshape == NULL)
2693 err = mddev->pers->start_reshape(mddev);
2697 if (cmd_match(page, "check"))
2698 set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
2699 else if (!cmd_match(page, "repair"))
2701 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
2702 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
2704 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2705 md_wakeup_thread(mddev->thread);
2710 mismatch_cnt_show(mddev_t *mddev, char *page)
2712 return sprintf(page, "%llu\n",
2713 (unsigned long long) mddev->resync_mismatches);
2716 static struct md_sysfs_entry md_scan_mode =
2717 __ATTR(sync_action, S_IRUGO|S_IWUSR, action_show, action_store);
2720 static struct md_sysfs_entry md_mismatches = __ATTR_RO(mismatch_cnt);
2723 sync_min_show(mddev_t *mddev, char *page)
2725 return sprintf(page, "%d (%s)\n", speed_min(mddev),
2726 mddev->sync_speed_min ? "local": "system");
2730 sync_min_store(mddev_t *mddev, const char *buf, size_t len)
2734 if (strncmp(buf, "system", 6)==0) {
2735 mddev->sync_speed_min = 0;
2738 min = simple_strtoul(buf, &e, 10);
2739 if (buf == e || (*e && *e != '\n') || min <= 0)
2741 mddev->sync_speed_min = min;
2745 static struct md_sysfs_entry md_sync_min =
2746 __ATTR(sync_speed_min, S_IRUGO|S_IWUSR, sync_min_show, sync_min_store);
2749 sync_max_show(mddev_t *mddev, char *page)
2751 return sprintf(page, "%d (%s)\n", speed_max(mddev),
2752 mddev->sync_speed_max ? "local": "system");
2756 sync_max_store(mddev_t *mddev, const char *buf, size_t len)
2760 if (strncmp(buf, "system", 6)==0) {
2761 mddev->sync_speed_max = 0;
2764 max = simple_strtoul(buf, &e, 10);
2765 if (buf == e || (*e && *e != '\n') || max <= 0)
2767 mddev->sync_speed_max = max;
2771 static struct md_sysfs_entry md_sync_max =
2772 __ATTR(sync_speed_max, S_IRUGO|S_IWUSR, sync_max_show, sync_max_store);
2776 sync_speed_show(mddev_t *mddev, char *page)
2778 unsigned long resync, dt, db;
2779 resync = (mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active));
2780 dt = ((jiffies - mddev->resync_mark) / HZ);
2782 db = resync - (mddev->resync_mark_cnt);
2783 return sprintf(page, "%ld\n", db/dt/2); /* K/sec */
2786 static struct md_sysfs_entry md_sync_speed = __ATTR_RO(sync_speed);
2789 sync_completed_show(mddev_t *mddev, char *page)
2791 unsigned long max_blocks, resync;
2793 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
2794 max_blocks = mddev->resync_max_sectors;
2796 max_blocks = mddev->size << 1;
2798 resync = (mddev->curr_resync - atomic_read(&mddev->recovery_active));
2799 return sprintf(page, "%lu / %lu\n", resync, max_blocks);
2802 static struct md_sysfs_entry md_sync_completed = __ATTR_RO(sync_completed);
2805 suspend_lo_show(mddev_t *mddev, char *page)
2807 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_lo);
2811 suspend_lo_store(mddev_t *mddev, const char *buf, size_t len)
2814 unsigned long long new = simple_strtoull(buf, &e, 10);
2816 if (mddev->pers->quiesce == NULL)
2818 if (buf == e || (*e && *e != '\n'))
2820 if (new >= mddev->suspend_hi ||
2821 (new > mddev->suspend_lo && new < mddev->suspend_hi)) {
2822 mddev->suspend_lo = new;
2823 mddev->pers->quiesce(mddev, 2);
2828 static struct md_sysfs_entry md_suspend_lo =
2829 __ATTR(suspend_lo, S_IRUGO|S_IWUSR, suspend_lo_show, suspend_lo_store);
2833 suspend_hi_show(mddev_t *mddev, char *page)
2835 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_hi);
2839 suspend_hi_store(mddev_t *mddev, const char *buf, size_t len)
2842 unsigned long long new = simple_strtoull(buf, &e, 10);
2844 if (mddev->pers->quiesce == NULL)
2846 if (buf == e || (*e && *e != '\n'))
2848 if ((new <= mddev->suspend_lo && mddev->suspend_lo >= mddev->suspend_hi) ||
2849 (new > mddev->suspend_lo && new > mddev->suspend_hi)) {
2850 mddev->suspend_hi = new;
2851 mddev->pers->quiesce(mddev, 1);
2852 mddev->pers->quiesce(mddev, 0);
2857 static struct md_sysfs_entry md_suspend_hi =
2858 __ATTR(suspend_hi, S_IRUGO|S_IWUSR, suspend_hi_show, suspend_hi_store);
2861 static struct attribute *md_default_attrs[] = {
2864 &md_raid_disks.attr,
2865 &md_chunk_size.attr,
2867 &md_resync_start.attr,
2869 &md_new_device.attr,
2870 &md_safe_delay.attr,
2871 &md_array_state.attr,
2875 static struct attribute *md_redundancy_attrs[] = {
2877 &md_mismatches.attr,
2880 &md_sync_speed.attr,
2881 &md_sync_completed.attr,
2882 &md_suspend_lo.attr,
2883 &md_suspend_hi.attr,
2887 static struct attribute_group md_redundancy_group = {
2889 .attrs = md_redundancy_attrs,
2894 md_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
2896 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
2897 mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
2902 rv = mddev_lock(mddev);
2904 rv = entry->show(mddev, page);
2905 mddev_unlock(mddev);
2911 md_attr_store(struct kobject *kobj, struct attribute *attr,
2912 const char *page, size_t length)
2914 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
2915 mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
2920 if (!capable(CAP_SYS_ADMIN))
2922 rv = mddev_lock(mddev);
2924 rv = entry->store(mddev, page, length);
2925 mddev_unlock(mddev);
2930 static void md_free(struct kobject *ko)
2932 mddev_t *mddev = container_of(ko, mddev_t, kobj);
2936 static struct sysfs_ops md_sysfs_ops = {
2937 .show = md_attr_show,
2938 .store = md_attr_store,
2940 static struct kobj_type md_ktype = {
2942 .sysfs_ops = &md_sysfs_ops,
2943 .default_attrs = md_default_attrs,
2948 static struct kobject *md_probe(dev_t dev, int *part, void *data)
2950 static DEFINE_MUTEX(disks_mutex);
2951 mddev_t *mddev = mddev_find(dev);
2952 struct gendisk *disk;
2953 int partitioned = (MAJOR(dev) != MD_MAJOR);
2954 int shift = partitioned ? MdpMinorShift : 0;
2955 int unit = MINOR(dev) >> shift;
2960 mutex_lock(&disks_mutex);
2961 if (mddev->gendisk) {
2962 mutex_unlock(&disks_mutex);
2966 disk = alloc_disk(1 << shift);
2968 mutex_unlock(&disks_mutex);
2972 disk->major = MAJOR(dev);
2973 disk->first_minor = unit << shift;
2975 sprintf(disk->disk_name, "md_d%d", unit);
2977 sprintf(disk->disk_name, "md%d", unit);
2978 disk->fops = &md_fops;
2979 disk->private_data = mddev;
2980 disk->queue = mddev->queue;
2982 mddev->gendisk = disk;
2983 mutex_unlock(&disks_mutex);
2984 mddev->kobj.parent = &disk->kobj;
2985 mddev->kobj.k_name = NULL;
2986 snprintf(mddev->kobj.name, KOBJ_NAME_LEN, "%s", "md");
2987 mddev->kobj.ktype = &md_ktype;
2988 kobject_register(&mddev->kobj);
2992 static void md_safemode_timeout(unsigned long data)
2994 mddev_t *mddev = (mddev_t *) data;
2996 mddev->safemode = 1;
2997 md_wakeup_thread(mddev->thread);
3000 static int start_dirty_degraded;
3002 static int do_md_run(mddev_t * mddev)
3006 struct list_head *tmp;
3008 struct gendisk *disk;
3009 struct mdk_personality *pers;
3010 char b[BDEVNAME_SIZE];
3012 if (list_empty(&mddev->disks))
3013 /* cannot run an array with no devices.. */
3020 * Analyze all RAID superblock(s)
3022 if (!mddev->raid_disks)
3025 chunk_size = mddev->chunk_size;
3028 if (chunk_size > MAX_CHUNK_SIZE) {
3029 printk(KERN_ERR "too big chunk_size: %d > %d\n",
3030 chunk_size, MAX_CHUNK_SIZE);
3034 * chunk-size has to be a power of 2 and multiples of PAGE_SIZE
3036 if ( (1 << ffz(~chunk_size)) != chunk_size) {
3037 printk(KERN_ERR "chunk_size of %d not valid\n", chunk_size);
3040 if (chunk_size < PAGE_SIZE) {
3041 printk(KERN_ERR "too small chunk_size: %d < %ld\n",
3042 chunk_size, PAGE_SIZE);
3046 /* devices must have minimum size of one chunk */
3047 ITERATE_RDEV(mddev,rdev,tmp) {
3048 if (test_bit(Faulty, &rdev->flags))
3050 if (rdev->size < chunk_size / 1024) {
3052 "md: Dev %s smaller than chunk_size:"
3054 bdevname(rdev->bdev,b),
3055 (unsigned long long)rdev->size,
3063 if (mddev->level != LEVEL_NONE)
3064 request_module("md-level-%d", mddev->level);
3065 else if (mddev->clevel[0])
3066 request_module("md-%s", mddev->clevel);
3070 * Drop all container device buffers, from now on
3071 * the only valid external interface is through the md
3073 * Also find largest hardsector size
3075 ITERATE_RDEV(mddev,rdev,tmp) {
3076 if (test_bit(Faulty, &rdev->flags))
3078 sync_blockdev(rdev->bdev);
3079 invalidate_bdev(rdev->bdev, 0);
3082 md_probe(mddev->unit, NULL, NULL);
3083 disk = mddev->gendisk;
3087 spin_lock(&pers_lock);
3088 pers = find_pers(mddev->level, mddev->clevel);
3089 if (!pers || !try_module_get(pers->owner)) {
3090 spin_unlock(&pers_lock);
3091 if (mddev->level != LEVEL_NONE)
3092 printk(KERN_WARNING "md: personality for level %d is not loaded!\n",
3095 printk(KERN_WARNING "md: personality for level %s is not loaded!\n",
3100 spin_unlock(&pers_lock);
3101 mddev->level = pers->level;
3102 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
3104 if (mddev->reshape_position != MaxSector &&
3105 pers->start_reshape == NULL) {
3106 /* This personality cannot handle reshaping... */
3108 module_put(pers->owner);
3112 mddev->recovery = 0;
3113 mddev->resync_max_sectors = mddev->size << 1; /* may be over-ridden by personality */
3114 mddev->barriers_work = 1;
3115 mddev->ok_start_degraded = start_dirty_degraded;
3118 mddev->ro = 2; /* read-only, but switch on first write */
3120 err = mddev->pers->run(mddev);
3121 if (!err && mddev->pers->sync_request) {
3122 err = bitmap_create(mddev);
3124 printk(KERN_ERR "%s: failed to create bitmap (%d)\n",
3125 mdname(mddev), err);
3126 mddev->pers->stop(mddev);
3130 printk(KERN_ERR "md: pers->run() failed ...\n");
3131 module_put(mddev->pers->owner);
3133 bitmap_destroy(mddev);
3136 if (mddev->pers->sync_request)
3137 sysfs_create_group(&mddev->kobj, &md_redundancy_group);
3138 else if (mddev->ro == 2) /* auto-readonly not meaningful */
3141 atomic_set(&mddev->writes_pending,0);
3142 mddev->safemode = 0;
3143 mddev->safemode_timer.function = md_safemode_timeout;
3144 mddev->safemode_timer.data = (unsigned long) mddev;
3145 mddev->safemode_delay = (200 * HZ)/1000 +1; /* 200 msec delay */
3148 ITERATE_RDEV(mddev,rdev,tmp)
3149 if (rdev->raid_disk >= 0) {
3151 sprintf(nm, "rd%d", rdev->raid_disk);
3152 sysfs_create_link(&mddev->kobj, &rdev->kobj, nm);
3155 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3158 md_update_sb(mddev, 0);
3160 set_capacity(disk, mddev->array_size<<1);
3162 /* If we call blk_queue_make_request here, it will
3163 * re-initialise max_sectors etc which may have been
3164 * refined inside -> run. So just set the bits we need to set.
3165 * Most initialisation happended when we called
3166 * blk_queue_make_request(..., md_fail_request)
3169 mddev->queue->queuedata = mddev;
3170 mddev->queue->make_request_fn = mddev->pers->make_request;
3172 /* If there is a partially-recovered drive we need to
3173 * start recovery here. If we leave it to md_check_recovery,
3174 * it will remove the drives and not do the right thing
3176 if (mddev->degraded && !mddev->sync_thread) {
3177 struct list_head *rtmp;
3179 ITERATE_RDEV(mddev,rdev,rtmp)
3180 if (rdev->raid_disk >= 0 &&
3181 !test_bit(In_sync, &rdev->flags) &&
3182 !test_bit(Faulty, &rdev->flags))
3183 /* complete an interrupted recovery */
3185 if (spares && mddev->pers->sync_request) {
3186 mddev->recovery = 0;
3187 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
3188 mddev->sync_thread = md_register_thread(md_do_sync,
3191 if (!mddev->sync_thread) {
3192 printk(KERN_ERR "%s: could not start resync"
3195 /* leave the spares where they are, it shouldn't hurt */
3196 mddev->recovery = 0;
3200 md_wakeup_thread(mddev->thread);
3201 md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
3204 md_new_event(mddev);
3205 kobject_uevent(&mddev->gendisk->kobj, KOBJ_CHANGE);
3209 static int restart_array(mddev_t *mddev)
3211 struct gendisk *disk = mddev->gendisk;
3215 * Complain if it has no devices
3218 if (list_empty(&mddev->disks))
3226 mddev->safemode = 0;
3228 set_disk_ro(disk, 0);
3230 printk(KERN_INFO "md: %s switched to read-write mode.\n",
3233 * Kick recovery or resync if necessary
3235 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3236 md_wakeup_thread(mddev->thread);
3237 md_wakeup_thread(mddev->sync_thread);
3246 /* similar to deny_write_access, but accounts for our holding a reference
3247 * to the file ourselves */
3248 static int deny_bitmap_write_access(struct file * file)
3250 struct inode *inode = file->f_mapping->host;
3252 spin_lock(&inode->i_lock);
3253 if (atomic_read(&inode->i_writecount) > 1) {
3254 spin_unlock(&inode->i_lock);
3257 atomic_set(&inode->i_writecount, -1);
3258 spin_unlock(&inode->i_lock);
3263 static void restore_bitmap_write_access(struct file *file)
3265 struct inode *inode = file->f_mapping->host;
3267 spin_lock(&inode->i_lock);
3268 atomic_set(&inode->i_writecount, 1);
3269 spin_unlock(&inode->i_lock);
3273 * 0 - completely stop and dis-assemble array
3274 * 1 - switch to readonly
3275 * 2 - stop but do not disassemble array
3277 static int do_md_stop(mddev_t * mddev, int mode)
3280 struct gendisk *disk = mddev->gendisk;
3283 if (atomic_read(&mddev->active)>2) {
3284 printk("md: %s still in use.\n",mdname(mddev));
3288 if (mddev->sync_thread) {
3289 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
3290 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
3291 md_unregister_thread(mddev->sync_thread);
3292 mddev->sync_thread = NULL;
3295 del_timer_sync(&mddev->safemode_timer);
3297 invalidate_partition(disk, 0);
3300 case 1: /* readonly */
3306 case 0: /* disassemble */
3308 bitmap_flush(mddev);
3309 md_super_wait(mddev);
3311 set_disk_ro(disk, 0);
3312 blk_queue_make_request(mddev->queue, md_fail_request);
3313 mddev->pers->stop(mddev);
3314 if (mddev->pers->sync_request)
3315 sysfs_remove_group(&mddev->kobj, &md_redundancy_group);
3317 module_put(mddev->pers->owner);
3320 set_capacity(disk, 0);
3326 if (!mddev->in_sync || mddev->flags) {
3327 /* mark array as shutdown cleanly */
3329 md_update_sb(mddev, 1);
3332 set_disk_ro(disk, 1);
3333 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
3337 * Free resources if final stop
3341 struct list_head *tmp;
3343 printk(KERN_INFO "md: %s stopped.\n", mdname(mddev));
3345 bitmap_destroy(mddev);
3346 if (mddev->bitmap_file) {
3347 restore_bitmap_write_access(mddev->bitmap_file);
3348 fput(mddev->bitmap_file);
3349 mddev->bitmap_file = NULL;
3351 mddev->bitmap_offset = 0;
3353 ITERATE_RDEV(mddev,rdev,tmp)
3354 if (rdev->raid_disk >= 0) {
3356 sprintf(nm, "rd%d", rdev->raid_disk);
3357 sysfs_remove_link(&mddev->kobj, nm);
3360 export_array(mddev);
3362 mddev->array_size = 0;
3364 mddev->raid_disks = 0;
3365 mddev->recovery_cp = 0;
3367 } else if (mddev->pers)
3368 printk(KERN_INFO "md: %s switched to read-only mode.\n",
3371 md_new_event(mddev);
3377 static void autorun_array(mddev_t *mddev)
3380 struct list_head *tmp;
3383 if (list_empty(&mddev->disks))
3386 printk(KERN_INFO "md: running: ");
3388 ITERATE_RDEV(mddev,rdev,tmp) {
3389 char b[BDEVNAME_SIZE];
3390 printk("<%s>", bdevname(rdev->bdev,b));
3394 err = do_md_run (mddev);
3396 printk(KERN_WARNING "md: do_md_run() returned %d\n", err);
3397 do_md_stop (mddev, 0);
3402 * lets try to run arrays based on all disks that have arrived
3403 * until now. (those are in pending_raid_disks)
3405 * the method: pick the first pending disk, collect all disks with
3406 * the same UUID, remove all from the pending list and put them into
3407 * the 'same_array' list. Then order this list based on superblock
3408 * update time (freshest comes first), kick out 'old' disks and
3409 * compare superblocks. If everything's fine then run it.
3411 * If "unit" is allocated, then bump its reference count
3413 static void autorun_devices(int part)
3415 struct list_head *tmp;
3416 mdk_rdev_t *rdev0, *rdev;
3418 char b[BDEVNAME_SIZE];
3420 printk(KERN_INFO "md: autorun ...\n");
3421 while (!list_empty(&pending_raid_disks)) {
3424 LIST_HEAD(candidates);
3425 rdev0 = list_entry(pending_raid_disks.next,
3426 mdk_rdev_t, same_set);
3428 printk(KERN_INFO "md: considering %s ...\n",
3429 bdevname(rdev0->bdev,b));
3430 INIT_LIST_HEAD(&candidates);
3431 ITERATE_RDEV_PENDING(rdev,tmp)
3432 if (super_90_load(rdev, rdev0, 0) >= 0) {
3433 printk(KERN_INFO "md: adding %s ...\n",
3434 bdevname(rdev->bdev,b));
3435 list_move(&rdev->same_set, &candidates);
3438 * now we have a set of devices, with all of them having
3439 * mostly sane superblocks. It's time to allocate the
3443 dev = MKDEV(mdp_major,
3444 rdev0->preferred_minor << MdpMinorShift);
3445 unit = MINOR(dev) >> MdpMinorShift;
3447 dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
3450 if (rdev0->preferred_minor != unit) {
3451 printk(KERN_INFO "md: unit number in %s is bad: %d\n",
3452 bdevname(rdev0->bdev, b), rdev0->preferred_minor);
3456 md_probe(dev, NULL, NULL);
3457 mddev = mddev_find(dev);
3460 "md: cannot allocate memory for md drive.\n");
3463 if (mddev_lock(mddev))
3464 printk(KERN_WARNING "md: %s locked, cannot run\n",
3466 else if (mddev->raid_disks || mddev->major_version
3467 || !list_empty(&mddev->disks)) {
3469 "md: %s already running, cannot run %s\n",
3470 mdname(mddev), bdevname(rdev0->bdev,b));
3471 mddev_unlock(mddev);
3473 printk(KERN_INFO "md: created %s\n", mdname(mddev));
3474 ITERATE_RDEV_GENERIC(candidates,rdev,tmp) {
3475 list_del_init(&rdev->same_set);
3476 if (bind_rdev_to_array(rdev, mddev))
3479 autorun_array(mddev);
3480 mddev_unlock(mddev);
3482 /* on success, candidates will be empty, on error
3485 ITERATE_RDEV_GENERIC(candidates,rdev,tmp)
3489 printk(KERN_INFO "md: ... autorun DONE.\n");
3491 #endif /* !MODULE */
3494 * import RAID devices based on one partition
3495 * if possible, the array gets run as well.
3498 static int autostart_array(dev_t startdev)
3500 char b[BDEVNAME_SIZE];
3501 int err = -EINVAL, i;
3502 mdp_super_t *sb = NULL;
3503 mdk_rdev_t *start_rdev = NULL, *rdev;
3505 start_rdev = md_import_device(startdev, 0, 0);
3506 if (IS_ERR(start_rdev))
3510 /* NOTE: this can only work for 0.90.0 superblocks */
3511 sb = (mdp_super_t*)page_address(start_rdev->sb_page);
3512 if (sb->major_version != 0 ||
3513 sb->minor_version != 90 ) {
3514 printk(KERN_WARNING "md: can only autostart 0.90.0 arrays\n");
3515 export_rdev(start_rdev);
3519 if (test_bit(Faulty, &start_rdev->flags)) {
3521 "md: can not autostart based on faulty %s!\n",
3522 bdevname(start_rdev->bdev,b));
3523 export_rdev(start_rdev);
3526 list_add(&start_rdev->same_set, &pending_raid_disks);
3528 for (i = 0; i < MD_SB_DISKS; i++) {
3529 mdp_disk_t *desc = sb->disks + i;
3530 dev_t dev = MKDEV(desc->major, desc->minor);
3534 if (dev == startdev)
3536 if (MAJOR(dev) != desc->major || MINOR(dev) != desc->minor)
3538 rdev = md_import_device(dev, 0, 0);
3542 list_add(&rdev->same_set, &pending_raid_disks);
3546 * possibly return codes
3554 static int get_version(void __user * arg)
3558 ver.major = MD_MAJOR_VERSION;
3559 ver.minor = MD_MINOR_VERSION;
3560 ver.patchlevel = MD_PATCHLEVEL_VERSION;
3562 if (copy_to_user(arg, &ver, sizeof(ver)))
3568 static int get_array_info(mddev_t * mddev, void __user * arg)
3570 mdu_array_info_t info;
3571 int nr,working,active,failed,spare;
3573 struct list_head *tmp;
3575 nr=working=active=failed=spare=0;
3576 ITERATE_RDEV(mddev,rdev,tmp) {
3578 if (test_bit(Faulty, &rdev->flags))
3582 if (test_bit(In_sync, &rdev->flags))
3589 info.major_version = mddev->major_version;
3590 info.minor_version = mddev->minor_version;
3591 info.patch_version = MD_PATCHLEVEL_VERSION;
3592 info.ctime = mddev->ctime;
3593 info.level = mddev->level;
3594 info.size = mddev->size;
3595 if (info.size != mddev->size) /* overflow */
3598 info.raid_disks = mddev->raid_disks;
3599 info.md_minor = mddev->md_minor;
3600 info.not_persistent= !mddev->persistent;
3602 info.utime = mddev->utime;
3605 info.state = (1<<MD_SB_CLEAN);
3606 if (mddev->bitmap && mddev->bitmap_offset)
3607 info.state = (1<<MD_SB_BITMAP_PRESENT);
3608 info.active_disks = active;
3609 info.working_disks = working;
3610 info.failed_disks = failed;
3611 info.spare_disks = spare;
3613 info.layout = mddev->layout;
3614 info.chunk_size = mddev->chunk_size;
3616 if (copy_to_user(arg, &info, sizeof(info)))
3622 static int get_bitmap_file(mddev_t * mddev, void __user * arg)
3624 mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */
3625 char *ptr, *buf = NULL;
3628 md_allow_write(mddev);
3630 file = kmalloc(sizeof(*file), GFP_KERNEL);
3634 /* bitmap disabled, zero the first byte and copy out */
3635 if (!mddev->bitmap || !mddev->bitmap->file) {
3636 file->pathname[0] = '\0';
3640 buf = kmalloc(sizeof(file->pathname), GFP_KERNEL);
3644 ptr = file_path(mddev->bitmap->file, buf, sizeof(file->pathname));
3648 strcpy(file->pathname, ptr);
3652 if (copy_to_user(arg, file, sizeof(*file)))
3660 static int get_disk_info(mddev_t * mddev, void __user * arg)
3662 mdu_disk_info_t info;
3666 if (copy_from_user(&info, arg, sizeof(info)))
3671 rdev = find_rdev_nr(mddev, nr);
3673 info.major = MAJOR(rdev->bdev->bd_dev);
3674 info.minor = MINOR(rdev->bdev->bd_dev);
3675 info.raid_disk = rdev->raid_disk;
3677 if (test_bit(Faulty, &rdev->flags))
3678 info.state |= (1<<MD_DISK_FAULTY);
3679 else if (test_bit(In_sync, &rdev->flags)) {
3680 info.state |= (1<<MD_DISK_ACTIVE);
3681 info.state |= (1<<MD_DISK_SYNC);
3683 if (test_bit(WriteMostly, &rdev->flags))
3684 info.state |= (1<<MD_DISK_WRITEMOSTLY);
3686 info.major = info.minor = 0;
3687 info.raid_disk = -1;
3688 info.state = (1<<MD_DISK_REMOVED);
3691 if (copy_to_user(arg, &info, sizeof(info)))
3697 static int add_new_disk(mddev_t * mddev, mdu_disk_info_t *info)
3699 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
3701 dev_t dev = MKDEV(info->major,info->minor);
3703 if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
3706 if (!mddev->raid_disks) {
3708 /* expecting a device which has a superblock */
3709 rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
3712 "md: md_import_device returned %ld\n",
3714 return PTR_ERR(rdev);
3716 if (!list_empty(&mddev->disks)) {
3717 mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
3718 mdk_rdev_t, same_set);
3719 int err = super_types[mddev->major_version]
3720 .load_super(rdev, rdev0, mddev->minor_version);
3723 "md: %s has different UUID to %s\n",
3724 bdevname(rdev->bdev,b),
3725 bdevname(rdev0->bdev,b2));
3730 err = bind_rdev_to_array(rdev, mddev);
3737 * add_new_disk can be used once the array is assembled
3738 * to add "hot spares". They must already have a superblock
3743 if (!mddev->pers->hot_add_disk) {
3745 "%s: personality does not support diskops!\n",
3749 if (mddev->persistent)
3750 rdev = md_import_device(dev, mddev->major_version,
3751 mddev->minor_version);
3753 rdev = md_import_device(dev, -1, -1);
3756 "md: md_import_device returned %ld\n",
3758 return PTR_ERR(rdev);
3760 /* set save_raid_disk if appropriate */
3761 if (!mddev->persistent) {
3762 if (info->state & (1<<MD_DISK_SYNC) &&
3763 info->raid_disk < mddev->raid_disks)
3764 rdev->raid_disk = info->raid_disk;
3766 rdev->raid_disk = -1;
3768 super_types[mddev->major_version].
3769 validate_super(mddev, rdev);
3770 rdev->saved_raid_disk = rdev->raid_disk;
3772 clear_bit(In_sync, &rdev->flags); /* just to be sure */
3773 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
3774 set_bit(WriteMostly, &rdev->flags);
3776 rdev->raid_disk = -1;
3777 err = bind_rdev_to_array(rdev, mddev);
3778 if (!err && !mddev->pers->hot_remove_disk) {
3779 /* If there is hot_add_disk but no hot_remove_disk
3780 * then added disks for geometry changes,
3781 * and should be added immediately.
3783 super_types[mddev->major_version].
3784 validate_super(mddev, rdev);
3785 err = mddev->pers->hot_add_disk(mddev, rdev);
3787 unbind_rdev_from_array(rdev);
3792 md_update_sb(mddev, 1);
3793 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3794 md_wakeup_thread(mddev->thread);
3798 /* otherwise, add_new_disk is only allowed
3799 * for major_version==0 superblocks
3801 if (mddev->major_version != 0) {
3802 printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n",
3807 if (!(info->state & (1<<MD_DISK_FAULTY))) {
3809 rdev = md_import_device (dev, -1, 0);
3812 "md: error, md_import_device() returned %ld\n",
3814 return PTR_ERR(rdev);
3816 rdev->desc_nr = info->number;
3817 if (info->raid_disk < mddev->raid_disks)
3818 rdev->raid_disk = info->raid_disk;
3820 rdev->raid_disk = -1;
3824 if (rdev->raid_disk < mddev->raid_disks)
3825 if (info->state & (1<<MD_DISK_SYNC))
3826 set_bit(In_sync, &rdev->flags);
3828 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
3829 set_bit(WriteMostly, &rdev->flags);
3831 if (!mddev->persistent) {
3832 printk(KERN_INFO "md: nonpersistent superblock ...\n");
3833 rdev->sb_offset = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
3835 rdev->sb_offset = calc_dev_sboffset(rdev->bdev);
3836 rdev->size = calc_dev_size(rdev, mddev->chunk_size);
3838 err = bind_rdev_to_array(rdev, mddev);
3848 static int hot_remove_disk(mddev_t * mddev, dev_t dev)
3850 char b[BDEVNAME_SIZE];
3856 rdev = find_rdev(mddev, dev);
3860 if (rdev->raid_disk >= 0)
3863 kick_rdev_from_array(rdev);
3864 md_update_sb(mddev, 1);
3865 md_new_event(mddev);
3869 printk(KERN_WARNING "md: cannot remove active disk %s from %s ... \n",
3870 bdevname(rdev->bdev,b), mdname(mddev));
3874 static int hot_add_disk(mddev_t * mddev, dev_t dev)
3876 char b[BDEVNAME_SIZE];
3884 if (mddev->major_version != 0) {
3885 printk(KERN_WARNING "%s: HOT_ADD may only be used with"
3886 " version-0 superblocks.\n",
3890 if (!mddev->pers->hot_add_disk) {
3892 "%s: personality does not support diskops!\n",
3897 rdev = md_import_device (dev, -1, 0);
3900 "md: error, md_import_device() returned %ld\n",
3905 if (mddev->persistent)
3906 rdev->sb_offset = calc_dev_sboffset(rdev->bdev);
3909 rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
3911 size = calc_dev_size(rdev, mddev->chunk_size);
3914 if (test_bit(Faulty, &rdev->flags)) {
3916 "md: can not hot-add faulty %s disk to %s!\n",
3917 bdevname(rdev->bdev,b), mdname(mddev));
3921 clear_bit(In_sync, &rdev->flags);
3923 rdev->saved_raid_disk = -1;
3924 err = bind_rdev_to_array(rdev, mddev);
3929 * The rest should better be atomic, we can have disk failures
3930 * noticed in interrupt contexts ...
3933 if (rdev->desc_nr == mddev->max_disks) {
3934 printk(KERN_WARNING "%s: can not hot-add to full array!\n",
3937 goto abort_unbind_export;
3940 rdev->raid_disk = -1;
3942 md_update_sb(mddev, 1);
3945 * Kick recovery, maybe this spare has to be added to the
3946 * array immediately.
3948 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3949 md_wakeup_thread(mddev->thread);
3950 md_new_event(mddev);
3953 abort_unbind_export:
3954 unbind_rdev_from_array(rdev);
3961 static int set_bitmap_file(mddev_t *mddev, int fd)
3966 if (!mddev->pers->quiesce)
3968 if (mddev->recovery || mddev->sync_thread)
3970 /* we should be able to change the bitmap.. */
3976 return -EEXIST; /* cannot add when bitmap is present */
3977 mddev->bitmap_file = fget(fd);
3979 if (mddev->bitmap_file == NULL) {
3980 printk(KERN_ERR "%s: error: failed to get bitmap file\n",
3985 err = deny_bitmap_write_access(mddev->bitmap_file);
3987 printk(KERN_ERR "%s: error: bitmap file is already in use\n",
3989 fput(mddev->bitmap_file);
3990 mddev->bitmap_file = NULL;
3993 mddev->bitmap_offset = 0; /* file overrides offset */
3994 } else if (mddev->bitmap == NULL)
3995 return -ENOENT; /* cannot remove what isn't there */
3998 mddev->pers->quiesce(mddev, 1);
4000 err = bitmap_create(mddev);
4001 if (fd < 0 || err) {
4002 bitmap_destroy(mddev);
4003 fd = -1; /* make sure to put the file */
4005 mddev->pers->quiesce(mddev, 0);
4008 if (mddev->bitmap_file) {
4009 restore_bitmap_write_access(mddev->bitmap_file);
4010 fput(mddev->bitmap_file);
4012 mddev->bitmap_file = NULL;
4019 * set_array_info is used two different ways
4020 * The original usage is when creating a new array.
4021 * In this usage, raid_disks is > 0 and it together with
4022 * level, size, not_persistent,layout,chunksize determine the
4023 * shape of the array.
4024 * This will always create an array with a type-0.90.0 superblock.
4025 * The newer usage is when assembling an array.
4026 * In this case raid_disks will be 0, and the major_version field is
4027 * use to determine which style super-blocks are to be found on the devices.
4028 * The minor and patch _version numbers are also kept incase the
4029 * super_block handler wishes to interpret them.
4031 static int set_array_info(mddev_t * mddev, mdu_array_info_t *info)
4034 if (info->raid_disks == 0) {
4035 /* just setting version number for superblock loading */
4036 if (info->major_version < 0 ||
4037 info->major_version >= sizeof(super_types)/sizeof(super_types[0]) ||
4038 super_types[info->major_version].name == NULL) {
4039 /* maybe try to auto-load a module? */
4041 "md: superblock version %d not known\n",
4042 info->major_version);
4045 mddev->major_version = info->major_version;
4046 mddev->minor_version = info->minor_version;
4047 mddev->patch_version = info->patch_version;
4048 mddev->persistent = !info->not_persistent;
4051 mddev->major_version = MD_MAJOR_VERSION;
4052 mddev->minor_version = MD_MINOR_VERSION;
4053 mddev->patch_version = MD_PATCHLEVEL_VERSION;
4054 mddev->ctime = get_seconds();
4056 mddev->level = info->level;
4057 mddev->clevel[0] = 0;
4058 mddev->size = info->size;
4059 mddev->raid_disks = info->raid_disks;
4060 /* don't set md_minor, it is determined by which /dev/md* was
4063 if (info->state & (1<<MD_SB_CLEAN))
4064 mddev->recovery_cp = MaxSector;
4066 mddev->recovery_cp = 0;
4067 mddev->persistent = ! info->not_persistent;
4069 mddev->layout = info->layout;
4070 mddev->chunk_size = info->chunk_size;
4072 mddev->max_disks = MD_SB_DISKS;
4075 set_bit(MD_CHANGE_DEVS, &mddev->flags);
4077 mddev->default_bitmap_offset = MD_SB_BYTES >> 9;
4078 mddev->bitmap_offset = 0;
4080 mddev->reshape_position = MaxSector;
4083 * Generate a 128 bit UUID
4085 get_random_bytes(mddev->uuid, 16);
4087 mddev->new_level = mddev->level;
4088 mddev->new_chunk = mddev->chunk_size;
4089 mddev->new_layout = mddev->layout;
4090 mddev->delta_disks = 0;
4095 static int update_size(mddev_t *mddev, unsigned long size)
4099 struct list_head *tmp;
4100 int fit = (size == 0);
4102 if (mddev->pers->resize == NULL)
4104 /* The "size" is the amount of each device that is used.
4105 * This can only make sense for arrays with redundancy.
4106 * linear and raid0 always use whatever space is available
4107 * We can only consider changing the size if no resync
4108 * or reconstruction is happening, and if the new size
4109 * is acceptable. It must fit before the sb_offset or,
4110 * if that is <data_offset, it must fit before the
4111 * size of each device.
4112 * If size is zero, we find the largest size that fits.
4114 if (mddev->sync_thread)
4116 ITERATE_RDEV(mddev,rdev,tmp) {
4118 avail = rdev->size * 2;
4120 if (fit && (size == 0 || size > avail/2))
4122 if (avail < ((sector_t)size << 1))
4125 rv = mddev->pers->resize(mddev, (sector_t)size *2);
4127 struct block_device *bdev;
4129 bdev = bdget_disk(mddev->gendisk, 0);
4131 mutex_lock(&bdev->bd_inode->i_mutex);
4132 i_size_write(bdev->bd_inode, (loff_t)mddev->array_size << 10);
4133 mutex_unlock(&bdev->bd_inode->i_mutex);
4140 static int update_raid_disks(mddev_t *mddev, int raid_disks)
4143 /* change the number of raid disks */
4144 if (mddev->pers->check_reshape == NULL)
4146 if (raid_disks <= 0 ||
4147 raid_disks >= mddev->max_disks)
4149 if (mddev->sync_thread || mddev->reshape_position != MaxSector)
4151 mddev->delta_disks = raid_disks - mddev->raid_disks;
4153 rv = mddev->pers->check_reshape(mddev);
4159 * update_array_info is used to change the configuration of an
4161 * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
4162 * fields in the info are checked against the array.
4163 * Any differences that cannot be handled will cause an error.
4164 * Normally, only one change can be managed at a time.
4166 static int update_array_info(mddev_t *mddev, mdu_array_info_t *info)
4172 /* calculate expected state,ignoring low bits */
4173 if (mddev->bitmap && mddev->bitmap_offset)
4174 state |= (1 << MD_SB_BITMAP_PRESENT);
4176 if (mddev->major_version != info->major_version ||
4177 mddev->minor_version != info->minor_version ||
4178 /* mddev->patch_version != info->patch_version || */
4179 mddev->ctime != info->ctime ||
4180 mddev->level != info->level ||
4181 /* mddev->layout != info->layout || */
4182 !mddev->persistent != info->not_persistent||
4183 mddev->chunk_size != info->chunk_size ||
4184 /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
4185 ((state^info->state) & 0xfffffe00)
4188 /* Check there is only one change */
4189 if (info->size >= 0 && mddev->size != info->size) cnt++;
4190 if (mddev->raid_disks != info->raid_disks) cnt++;
4191 if (mddev->layout != info->layout) cnt++;
4192 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) cnt++;
4193 if (cnt == 0) return 0;
4194 if (cnt > 1) return -EINVAL;
4196 if (mddev->layout != info->layout) {
4198 * we don't need to do anything at the md level, the
4199 * personality will take care of it all.
4201 if (mddev->pers->reconfig == NULL)
4204 return mddev->pers->reconfig(mddev, info->layout, -1);
4206 if (info->size >= 0 && mddev->size != info->size)
4207 rv = update_size(mddev, info->size);
4209 if (mddev->raid_disks != info->raid_disks)
4210 rv = update_raid_disks(mddev, info->raid_disks);
4212 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) {
4213 if (mddev->pers->quiesce == NULL)
4215 if (mddev->recovery || mddev->sync_thread)
4217 if (info->state & (1<<MD_SB_BITMAP_PRESENT)) {
4218 /* add the bitmap */
4221 if (mddev->default_bitmap_offset == 0)
4223 mddev->bitmap_offset = mddev->default_bitmap_offset;
4224 mddev->pers->quiesce(mddev, 1);
4225 rv = bitmap_create(mddev);
4227 bitmap_destroy(mddev);
4228 mddev->pers->quiesce(mddev, 0);
4230 /* remove the bitmap */
4233 if (mddev->bitmap->file)
4235 mddev->pers->quiesce(mddev, 1);
4236 bitmap_destroy(mddev);
4237 mddev->pers->quiesce(mddev, 0);
4238 mddev->bitmap_offset = 0;
4241 md_update_sb(mddev, 1);
4245 static int set_disk_faulty(mddev_t *mddev, dev_t dev)
4249 if (mddev->pers == NULL)
4252 rdev = find_rdev(mddev, dev);
4256 md_error(mddev, rdev);
4260 static int md_getgeo(struct block_device *bdev, struct hd_geometry *geo)
4262 mddev_t *mddev = bdev->bd_disk->private_data;
4266 geo->cylinders = get_capacity(mddev->gendisk) / 8;
4270 static int md_ioctl(struct inode *inode, struct file *file,
4271 unsigned int cmd, unsigned long arg)
4274 void __user *argp = (void __user *)arg;
4275 mddev_t *mddev = NULL;
4277 if (!capable(CAP_SYS_ADMIN))
4281 * Commands dealing with the RAID driver but not any
4287 err = get_version(argp);
4290 case PRINT_RAID_DEBUG:
4298 autostart_arrays(arg);
4305 * Commands creating/starting a new array:
4308 mddev = inode->i_bdev->bd_disk->private_data;
4316 if (cmd == START_ARRAY) {
4317 /* START_ARRAY doesn't need to lock the array as autostart_array
4318 * does the locking, and it could even be a different array
4323 "md: %s(pid %d) used deprecated START_ARRAY ioctl. "
4324 "This will not be supported beyond July 2006\n",
4325 current->comm, current->pid);
4328 err = autostart_array(new_decode_dev(arg));
4330 printk(KERN_WARNING "md: autostart failed!\n");
4336 err = mddev_lock(mddev);
4339 "md: ioctl lock interrupted, reason %d, cmd %d\n",
4346 case SET_ARRAY_INFO:
4348 mdu_array_info_t info;
4350 memset(&info, 0, sizeof(info));
4351 else if (copy_from_user(&info, argp, sizeof(info))) {
4356 err = update_array_info(mddev, &info);
4358 printk(KERN_WARNING "md: couldn't update"
4359 " array info. %d\n", err);
4364 if (!list_empty(&mddev->disks)) {
4366 "md: array %s already has disks!\n",
4371 if (mddev->raid_disks) {
4373 "md: array %s already initialised!\n",
4378 err = set_array_info(mddev, &info);
4380 printk(KERN_WARNING "md: couldn't set"
4381 " array info. %d\n", err);
4391 * Commands querying/configuring an existing array:
4393 /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
4394 * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */
4395 if (!mddev->raid_disks && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY
4396 && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE
4397 && cmd != GET_BITMAP_FILE) {
4403 * Commands even a read-only array can execute:
4407 case GET_ARRAY_INFO:
4408 err = get_array_info(mddev, argp);
4411 case GET_BITMAP_FILE:
4412 err = get_bitmap_file(mddev, argp);
4416 err = get_disk_info(mddev, argp);
4419 case RESTART_ARRAY_RW:
4420 err = restart_array(mddev);
4424 err = do_md_stop (mddev, 0);
4428 err = do_md_stop (mddev, 1);
4432 * We have a problem here : there is no easy way to give a CHS
4433 * virtual geometry. We currently pretend that we have a 2 heads
4434 * 4 sectors (with a BIG number of cylinders...). This drives
4435 * dosfs just mad... ;-)
4440 * The remaining ioctls are changing the state of the
4441 * superblock, so we do not allow them on read-only arrays.
4442 * However non-MD ioctls (e.g. get-size) will still come through
4443 * here and hit the 'default' below, so only disallow
4444 * 'md' ioctls, and switch to rw mode if started auto-readonly.
4446 if (_IOC_TYPE(cmd) == MD_MAJOR &&
4447 mddev->ro && mddev->pers) {
4448 if (mddev->ro == 2) {
4450 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4451 md_wakeup_thread(mddev->thread);
4463 mdu_disk_info_t info;
4464 if (copy_from_user(&info, argp, sizeof(info)))
4467 err = add_new_disk(mddev, &info);
4471 case HOT_REMOVE_DISK:
4472 err = hot_remove_disk(mddev, new_decode_dev(arg));
4476 err = hot_add_disk(mddev, new_decode_dev(arg));
4479 case SET_DISK_FAULTY:
4480 err = set_disk_faulty(mddev, new_decode_dev(arg));
4484 err = do_md_run (mddev);
4487 case SET_BITMAP_FILE:
4488 err = set_bitmap_file(mddev, (int)arg);
4498 mddev_unlock(mddev);
4508 static int md_open(struct inode *inode, struct file *file)
4511 * Succeed if we can lock the mddev, which confirms that
4512 * it isn't being stopped right now.
4514 mddev_t *mddev = inode->i_bdev->bd_disk->private_data;
4517 if ((err = mutex_lock_interruptible_nested(&mddev->reconfig_mutex, 1)))
4522 mddev_unlock(mddev);
4524 check_disk_change(inode->i_bdev);
4529 static int md_release(struct inode *inode, struct file * file)
4531 mddev_t *mddev = inode->i_bdev->bd_disk->private_data;
4539 static int md_media_changed(struct gendisk *disk)
4541 mddev_t *mddev = disk->private_data;
4543 return mddev->changed;
4546 static int md_revalidate(struct gendisk *disk)
4548 mddev_t *mddev = disk->private_data;
4553 static struct block_device_operations md_fops =
4555 .owner = THIS_MODULE,
4557 .release = md_release,
4559 .getgeo = md_getgeo,
4560 .media_changed = md_media_changed,
4561 .revalidate_disk= md_revalidate,
4564 static int md_thread(void * arg)
4566 mdk_thread_t *thread = arg;
4569 * md_thread is a 'system-thread', it's priority should be very
4570 * high. We avoid resource deadlocks individually in each
4571 * raid personality. (RAID5 does preallocation) We also use RR and
4572 * the very same RT priority as kswapd, thus we will never get
4573 * into a priority inversion deadlock.
4575 * we definitely have to have equal or higher priority than
4576 * bdflush, otherwise bdflush will deadlock if there are too
4577 * many dirty RAID5 blocks.
4580 current->flags |= PF_NOFREEZE;
4581 allow_signal(SIGKILL);
4582 while (!kthread_should_stop()) {
4584 /* We need to wait INTERRUPTIBLE so that
4585 * we don't add to the load-average.
4586 * That means we need to be sure no signals are
4589 if (signal_pending(current))
4590 flush_signals(current);
4592 wait_event_interruptible_timeout
4594 test_bit(THREAD_WAKEUP, &thread->flags)
4595 || kthread_should_stop(),
4598 clear_bit(THREAD_WAKEUP, &thread->flags);
4600 thread->run(thread->mddev);
4606 void md_wakeup_thread(mdk_thread_t *thread)
4609 dprintk("md: waking up MD thread %s.\n", thread->tsk->comm);
4610 set_bit(THREAD_WAKEUP, &thread->flags);
4611 wake_up(&thread->wqueue);
4615 mdk_thread_t *md_register_thread(void (*run) (mddev_t *), mddev_t *mddev,
4618 mdk_thread_t *thread;
4620 thread = kzalloc(sizeof(mdk_thread_t), GFP_KERNEL);
4624 init_waitqueue_head(&thread->wqueue);
4627 thread->mddev = mddev;
4628 thread->timeout = MAX_SCHEDULE_TIMEOUT;
4629 thread->tsk = kthread_run(md_thread, thread, name, mdname(thread->mddev));
4630 if (IS_ERR(thread->tsk)) {
4637 void md_unregister_thread(mdk_thread_t *thread)
4639 dprintk("interrupting MD-thread pid %d\n", thread->tsk->pid);
4641 kthread_stop(thread->tsk);
4645 void md_error(mddev_t *mddev, mdk_rdev_t *rdev)
4652 if (!rdev || test_bit(Faulty, &rdev->flags))
4655 dprintk("md_error dev:%s, rdev:(%d:%d), (caller: %p,%p,%p,%p).\n",
4657 MAJOR(rdev->bdev->bd_dev), MINOR(rdev->bdev->bd_dev),
4658 __builtin_return_address(0),__builtin_return_address(1),
4659 __builtin_return_address(2),__builtin_return_address(3));
4663 if (!mddev->pers->error_handler)
4665 mddev->pers->error_handler(mddev,rdev);
4666 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
4667 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4668 md_wakeup_thread(mddev->thread);
4669 md_new_event_inintr(mddev);
4672 /* seq_file implementation /proc/mdstat */
4674 static void status_unused(struct seq_file *seq)
4678 struct list_head *tmp;
4680 seq_printf(seq, "unused devices: ");
4682 ITERATE_RDEV_PENDING(rdev,tmp) {
4683 char b[BDEVNAME_SIZE];
4685 seq_printf(seq, "%s ",
4686 bdevname(rdev->bdev,b));
4689 seq_printf(seq, "<none>");
4691 seq_printf(seq, "\n");
4695 static void status_resync(struct seq_file *seq, mddev_t * mddev)
4697 sector_t max_blocks, resync, res;
4698 unsigned long dt, db, rt;
4700 unsigned int per_milli;
4702 resync = (mddev->curr_resync - atomic_read(&mddev->recovery_active))/2;
4704 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
4705 max_blocks = mddev->resync_max_sectors >> 1;
4707 max_blocks = mddev->size;
4710 * Should not happen.
4716 /* Pick 'scale' such that (resync>>scale)*1000 will fit
4717 * in a sector_t, and (max_blocks>>scale) will fit in a
4718 * u32, as those are the requirements for sector_div.
4719 * Thus 'scale' must be at least 10
4722 if (sizeof(sector_t) > sizeof(unsigned long)) {
4723 while ( max_blocks/2 > (1ULL<<(scale+32)))
4726 res = (resync>>scale)*1000;
4727 sector_div(res, (u32)((max_blocks>>scale)+1));
4731 int i, x = per_milli/50, y = 20-x;
4732 seq_printf(seq, "[");
4733 for (i = 0; i < x; i++)
4734 seq_printf(seq, "=");
4735 seq_printf(seq, ">");
4736 for (i = 0; i < y; i++)
4737 seq_printf(seq, ".");
4738 seq_printf(seq, "] ");
4740 seq_printf(seq, " %s =%3u.%u%% (%llu/%llu)",
4741 (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)?
4743 (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)?
4745 (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
4746 "resync" : "recovery"))),
4747 per_milli/10, per_milli % 10,
4748 (unsigned long long) resync,
4749 (unsigned long long) max_blocks);
4752 * We do not want to overflow, so the order of operands and
4753 * the * 100 / 100 trick are important. We do a +1 to be
4754 * safe against division by zero. We only estimate anyway.
4756 * dt: time from mark until now
4757 * db: blocks written from mark until now
4758 * rt: remaining time
4760 dt = ((jiffies - mddev->resync_mark) / HZ);
4762 db = (mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active))
4763 - mddev->resync_mark_cnt;
4764 rt = (dt * ((unsigned long)(max_blocks-resync) / (db/2/100+1)))/100;
4766 seq_printf(seq, " finish=%lu.%lumin", rt / 60, (rt % 60)/6);
4768 seq_printf(seq, " speed=%ldK/sec", db/2/dt);
4771 static void *md_seq_start(struct seq_file *seq, loff_t *pos)
4773 struct list_head *tmp;
4783 spin_lock(&all_mddevs_lock);
4784 list_for_each(tmp,&all_mddevs)
4786 mddev = list_entry(tmp, mddev_t, all_mddevs);
4788 spin_unlock(&all_mddevs_lock);
4791 spin_unlock(&all_mddevs_lock);
4793 return (void*)2;/* tail */
4797 static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4799 struct list_head *tmp;
4800 mddev_t *next_mddev, *mddev = v;
4806 spin_lock(&all_mddevs_lock);
4808 tmp = all_mddevs.next;
4810 tmp = mddev->all_mddevs.next;
4811 if (tmp != &all_mddevs)
4812 next_mddev = mddev_get(list_entry(tmp,mddev_t,all_mddevs));
4814 next_mddev = (void*)2;
4817 spin_unlock(&all_mddevs_lock);
4825 static void md_seq_stop(struct seq_file *seq, void *v)
4829 if (mddev && v != (void*)1 && v != (void*)2)
4833 struct mdstat_info {
4837 static int md_seq_show(struct seq_file *seq, void *v)
4841 struct list_head *tmp2;
4843 struct mdstat_info *mi = seq->private;
4844 struct bitmap *bitmap;
4846 if (v == (void*)1) {
4847 struct mdk_personality *pers;
4848 seq_printf(seq, "Personalities : ");
4849 spin_lock(&pers_lock);
4850 list_for_each_entry(pers, &pers_list, list)
4851 seq_printf(seq, "[%s] ", pers->name);
4853 spin_unlock(&pers_lock);
4854 seq_printf(seq, "\n");
4855 mi->event = atomic_read(&md_event_count);
4858 if (v == (void*)2) {
4863 if (mddev_lock(mddev) < 0)
4866 if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
4867 seq_printf(seq, "%s : %sactive", mdname(mddev),
4868 mddev->pers ? "" : "in");
4871 seq_printf(seq, " (read-only)");
4873 seq_printf(seq, "(auto-read-only)");
4874 seq_printf(seq, " %s", mddev->pers->name);
4878 ITERATE_RDEV(mddev,rdev,tmp2) {
4879 char b[BDEVNAME_SIZE];
4880 seq_printf(seq, " %s[%d]",
4881 bdevname(rdev->bdev,b), rdev->desc_nr);
4882 if (test_bit(WriteMostly, &rdev->flags))
4883 seq_printf(seq, "(W)");
4884 if (test_bit(Faulty, &rdev->flags)) {
4885 seq_printf(seq, "(F)");
4887 } else if (rdev->raid_disk < 0)
4888 seq_printf(seq, "(S)"); /* spare */
4892 if (!list_empty(&mddev->disks)) {
4894 seq_printf(seq, "\n %llu blocks",
4895 (unsigned long long)mddev->array_size);
4897 seq_printf(seq, "\n %llu blocks",
4898 (unsigned long long)size);
4900 if (mddev->persistent) {
4901 if (mddev->major_version != 0 ||
4902 mddev->minor_version != 90) {
4903 seq_printf(seq," super %d.%d",
4904 mddev->major_version,
4905 mddev->minor_version);
4908 seq_printf(seq, " super non-persistent");
4911 mddev->pers->status (seq, mddev);
4912 seq_printf(seq, "\n ");
4913 if (mddev->pers->sync_request) {
4914 if (mddev->curr_resync > 2) {
4915 status_resync (seq, mddev);
4916 seq_printf(seq, "\n ");
4917 } else if (mddev->curr_resync == 1 || mddev->curr_resync == 2)
4918 seq_printf(seq, "\tresync=DELAYED\n ");
4919 else if (mddev->recovery_cp < MaxSector)
4920 seq_printf(seq, "\tresync=PENDING\n ");
4923 seq_printf(seq, "\n ");
4925 if ((bitmap = mddev->bitmap)) {
4926 unsigned long chunk_kb;
4927 unsigned long flags;
4928 spin_lock_irqsave(&bitmap->lock, flags);
4929 chunk_kb = bitmap->chunksize >> 10;
4930 seq_printf(seq, "bitmap: %lu/%lu pages [%luKB], "
4932 bitmap->pages - bitmap->missing_pages,
4934 (bitmap->pages - bitmap->missing_pages)
4935 << (PAGE_SHIFT - 10),
4936 chunk_kb ? chunk_kb : bitmap->chunksize,
4937 chunk_kb ? "KB" : "B");
4939 seq_printf(seq, ", file: ");
4940 seq_path(seq, bitmap->file->f_path.mnt,
4941 bitmap->file->f_path.dentry," \t\n");
4944 seq_printf(seq, "\n");
4945 spin_unlock_irqrestore(&bitmap->lock, flags);
4948 seq_printf(seq, "\n");
4950 mddev_unlock(mddev);
4955 static struct seq_operations md_seq_ops = {
4956 .start = md_seq_start,
4957 .next = md_seq_next,
4958 .stop = md_seq_stop,
4959 .show = md_seq_show,
4962 static int md_seq_open(struct inode *inode, struct file *file)
4965 struct mdstat_info *mi = kmalloc(sizeof(*mi), GFP_KERNEL);
4969 error = seq_open(file, &md_seq_ops);
4973 struct seq_file *p = file->private_data;
4975 mi->event = atomic_read(&md_event_count);
4980 static int md_seq_release(struct inode *inode, struct file *file)
4982 struct seq_file *m = file->private_data;
4983 struct mdstat_info *mi = m->private;
4986 return seq_release(inode, file);
4989 static unsigned int mdstat_poll(struct file *filp, poll_table *wait)
4991 struct seq_file *m = filp->private_data;
4992 struct mdstat_info *mi = m->private;
4995 poll_wait(filp, &md_event_waiters, wait);
4997 /* always allow read */
4998 mask = POLLIN | POLLRDNORM;
5000 if (mi->event != atomic_read(&md_event_count))
5001 mask |= POLLERR | POLLPRI;
5005 static struct file_operations md_seq_fops = {
5006 .owner = THIS_MODULE,
5007 .open = md_seq_open,
5009 .llseek = seq_lseek,
5010 .release = md_seq_release,
5011 .poll = mdstat_poll,
5014 int register_md_personality(struct mdk_personality *p)
5016 spin_lock(&pers_lock);
5017 list_add_tail(&p->list, &pers_list);
5018 printk(KERN_INFO "md: %s personality registered for level %d\n", p->name, p->level);
5019 spin_unlock(&pers_lock);
5023 int unregister_md_personality(struct mdk_personality *p)
5025 printk(KERN_INFO "md: %s personality unregistered\n", p->name);
5026 spin_lock(&pers_lock);
5027 list_del_init(&p->list);
5028 spin_unlock(&pers_lock);
5032 static int is_mddev_idle(mddev_t *mddev)
5035 struct list_head *tmp;
5037 unsigned long curr_events;
5040 ITERATE_RDEV(mddev,rdev,tmp) {
5041 struct gendisk *disk = rdev->bdev->bd_contains->bd_disk;
5042 curr_events = disk_stat_read(disk, sectors[0]) +
5043 disk_stat_read(disk, sectors[1]) -
5044 atomic_read(&disk->sync_io);
5045 /* The difference between curr_events and last_events
5046 * will be affected by any new non-sync IO (making
5047 * curr_events bigger) and any difference in the amount of
5048 * in-flight syncio (making current_events bigger or smaller)
5049 * The amount in-flight is currently limited to
5050 * 32*64K in raid1/10 and 256*PAGE_SIZE in raid5/6
5051 * which is at most 4096 sectors.
5052 * These numbers are fairly fragile and should be made
5053 * more robust, probably by enforcing the
5054 * 'window size' that md_do_sync sort-of uses.
5056 * Note: the following is an unsigned comparison.
5058 if ((curr_events - rdev->last_events + 4096) > 8192) {
5059 rdev->last_events = curr_events;
5066 void md_done_sync(mddev_t *mddev, int blocks, int ok)
5068 /* another "blocks" (512byte) blocks have been synced */
5069 atomic_sub(blocks, &mddev->recovery_active);
5070 wake_up(&mddev->recovery_wait);
5072 set_bit(MD_RECOVERY_ERR, &mddev->recovery);
5073 md_wakeup_thread(mddev->thread);
5074 // stop recovery, signal do_sync ....
5079 /* md_write_start(mddev, bi)
5080 * If we need to update some array metadata (e.g. 'active' flag
5081 * in superblock) before writing, schedule a superblock update
5082 * and wait for it to complete.
5084 void md_write_start(mddev_t *mddev, struct bio *bi)
5086 if (bio_data_dir(bi) != WRITE)
5089 BUG_ON(mddev->ro == 1);
5090 if (mddev->ro == 2) {
5091 /* need to switch to read/write */
5093 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5094 md_wakeup_thread(mddev->thread);
5096 atomic_inc(&mddev->writes_pending);
5097 if (mddev->in_sync) {
5098 spin_lock_irq(&mddev->write_lock);
5099 if (mddev->in_sync) {
5101 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
5102 md_wakeup_thread(mddev->thread);
5104 spin_unlock_irq(&mddev->write_lock);
5106 wait_event(mddev->sb_wait, mddev->flags==0);
5109 void md_write_end(mddev_t *mddev)
5111 if (atomic_dec_and_test(&mddev->writes_pending)) {
5112 if (mddev->safemode == 2)
5113 md_wakeup_thread(mddev->thread);
5114 else if (mddev->safemode_delay)
5115 mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay);
5119 /* md_allow_write(mddev)
5120 * Calling this ensures that the array is marked 'active' so that writes
5121 * may proceed without blocking. It is important to call this before
5122 * attempting a GFP_KERNEL allocation while holding the mddev lock.
5123 * Must be called with mddev_lock held.
5125 void md_allow_write(mddev_t *mddev)
5132 spin_lock_irq(&mddev->write_lock);
5133 if (mddev->in_sync) {
5135 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
5136 if (mddev->safemode_delay &&
5137 mddev->safemode == 0)
5138 mddev->safemode = 1;
5139 spin_unlock_irq(&mddev->write_lock);
5140 md_update_sb(mddev, 0);
5142 spin_unlock_irq(&mddev->write_lock);
5144 EXPORT_SYMBOL_GPL(md_allow_write);
5146 static DECLARE_WAIT_QUEUE_HEAD(resync_wait);
5148 #define SYNC_MARKS 10
5149 #define SYNC_MARK_STEP (3*HZ)
5150 void md_do_sync(mddev_t *mddev)
5153 unsigned int currspeed = 0,
5155 sector_t max_sectors,j, io_sectors;
5156 unsigned long mark[SYNC_MARKS];
5157 sector_t mark_cnt[SYNC_MARKS];
5159 struct list_head *tmp;
5160 sector_t last_check;
5162 struct list_head *rtmp;
5166 /* just incase thread restarts... */
5167 if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
5169 if (mddev->ro) /* never try to sync a read-only array */
5172 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
5173 if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
5174 desc = "data-check";
5175 else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
5176 desc = "requested-resync";
5179 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
5184 /* we overload curr_resync somewhat here.
5185 * 0 == not engaged in resync at all
5186 * 2 == checking that there is no conflict with another sync
5187 * 1 == like 2, but have yielded to allow conflicting resync to
5189 * other == active in resync - this many blocks
5191 * Before starting a resync we must have set curr_resync to
5192 * 2, and then checked that every "conflicting" array has curr_resync
5193 * less than ours. When we find one that is the same or higher
5194 * we wait on resync_wait. To avoid deadlock, we reduce curr_resync
5195 * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
5196 * This will mean we have to start checking from the beginning again.
5201 mddev->curr_resync = 2;
5204 if (kthread_should_stop()) {
5205 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5208 ITERATE_MDDEV(mddev2,tmp) {
5209 if (mddev2 == mddev)
5211 if (mddev2->curr_resync &&
5212 match_mddev_units(mddev,mddev2)) {
5214 if (mddev < mddev2 && mddev->curr_resync == 2) {
5215 /* arbitrarily yield */
5216 mddev->curr_resync = 1;
5217 wake_up(&resync_wait);
5219 if (mddev > mddev2 && mddev->curr_resync == 1)
5220 /* no need to wait here, we can wait the next
5221 * time 'round when curr_resync == 2
5224 prepare_to_wait(&resync_wait, &wq, TASK_UNINTERRUPTIBLE);
5225 if (!kthread_should_stop() &&
5226 mddev2->curr_resync >= mddev->curr_resync) {
5227 printk(KERN_INFO "md: delaying %s of %s"
5228 " until %s has finished (they"
5229 " share one or more physical units)\n",
5230 desc, mdname(mddev), mdname(mddev2));
5233 finish_wait(&resync_wait, &wq);
5236 finish_wait(&resync_wait, &wq);
5239 } while (mddev->curr_resync < 2);
5242 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
5243 /* resync follows the size requested by the personality,
5244 * which defaults to physical size, but can be virtual size
5246 max_sectors = mddev->resync_max_sectors;
5247 mddev->resync_mismatches = 0;
5248 /* we don't use the checkpoint if there's a bitmap */
5249 if (!mddev->bitmap &&
5250 !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
5251 j = mddev->recovery_cp;
5252 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
5253 max_sectors = mddev->size << 1;
5255 /* recovery follows the physical size of devices */
5256 max_sectors = mddev->size << 1;
5258 ITERATE_RDEV(mddev,rdev,rtmp)
5259 if (rdev->raid_disk >= 0 &&
5260 !test_bit(Faulty, &rdev->flags) &&
5261 !test_bit(In_sync, &rdev->flags) &&
5262 rdev->recovery_offset < j)
5263 j = rdev->recovery_offset;
5266 printk(KERN_INFO "md: %s of RAID array %s\n", desc, mdname(mddev));
5267 printk(KERN_INFO "md: minimum _guaranteed_ speed:"
5268 " %d KB/sec/disk.\n", speed_min(mddev));
5269 printk(KERN_INFO "md: using maximum available idle IO bandwidth "
5270 "(but not more than %d KB/sec) for %s.\n",
5271 speed_max(mddev), desc);
5273 is_mddev_idle(mddev); /* this also initializes IO event counters */
5276 for (m = 0; m < SYNC_MARKS; m++) {
5278 mark_cnt[m] = io_sectors;
5281 mddev->resync_mark = mark[last_mark];
5282 mddev->resync_mark_cnt = mark_cnt[last_mark];
5285 * Tune reconstruction:
5287 window = 32*(PAGE_SIZE/512);
5288 printk(KERN_INFO "md: using %dk window, over a total of %llu blocks.\n",
5289 window/2,(unsigned long long) max_sectors/2);
5291 atomic_set(&mddev->recovery_active, 0);
5292 init_waitqueue_head(&mddev->recovery_wait);
5297 "md: resuming %s of %s from checkpoint.\n",
5298 desc, mdname(mddev));
5299 mddev->curr_resync = j;
5302 while (j < max_sectors) {
5306 sectors = mddev->pers->sync_request(mddev, j, &skipped,
5307 currspeed < speed_min(mddev));
5309 set_bit(MD_RECOVERY_ERR, &mddev->recovery);
5313 if (!skipped) { /* actual IO requested */
5314 io_sectors += sectors;
5315 atomic_add(sectors, &mddev->recovery_active);
5319 if (j>1) mddev->curr_resync = j;
5320 mddev->curr_mark_cnt = io_sectors;
5321 if (last_check == 0)
5322 /* this is the earliers that rebuilt will be
5323 * visible in /proc/mdstat
5325 md_new_event(mddev);
5327 if (last_check + window > io_sectors || j == max_sectors)
5330 last_check = io_sectors;
5332 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery) ||
5333 test_bit(MD_RECOVERY_ERR, &mddev->recovery))
5337 if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) {
5339 int next = (last_mark+1) % SYNC_MARKS;
5341 mddev->resync_mark = mark[next];
5342 mddev->resync_mark_cnt = mark_cnt[next];
5343 mark[next] = jiffies;
5344 mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active);
5349 if (kthread_should_stop()) {
5351 * got a signal, exit.
5354 "md: md_do_sync() got signal ... exiting\n");
5355 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5360 * this loop exits only if either when we are slower than
5361 * the 'hard' speed limit, or the system was IO-idle for
5363 * the system might be non-idle CPU-wise, but we only care
5364 * about not overloading the IO subsystem. (things like an
5365 * e2fsck being done on the RAID array should execute fast)
5367 mddev->queue->unplug_fn(mddev->queue);
5370 currspeed = ((unsigned long)(io_sectors-mddev->resync_mark_cnt))/2
5371 /((jiffies-mddev->resync_mark)/HZ +1) +1;
5373 if (currspeed > speed_min(mddev)) {
5374 if ((currspeed > speed_max(mddev)) ||
5375 !is_mddev_idle(mddev)) {
5381 printk(KERN_INFO "md: %s: %s done.\n",mdname(mddev), desc);
5383 * this also signals 'finished resyncing' to md_stop
5386 mddev->queue->unplug_fn(mddev->queue);
5388 wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
5390 /* tell personality that we are finished */
5391 mddev->pers->sync_request(mddev, max_sectors, &skipped, 1);
5393 if (!test_bit(MD_RECOVERY_ERR, &mddev->recovery) &&
5394 !test_bit(MD_RECOVERY_CHECK, &mddev->recovery) &&
5395 mddev->curr_resync > 2) {
5396 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
5397 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
5398 if (mddev->curr_resync >= mddev->recovery_cp) {
5400 "md: checkpointing %s of %s.\n",
5401 desc, mdname(mddev));
5402 mddev->recovery_cp = mddev->curr_resync;
5405 mddev->recovery_cp = MaxSector;
5407 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery))
5408 mddev->curr_resync = MaxSector;
5409 ITERATE_RDEV(mddev,rdev,rtmp)
5410 if (rdev->raid_disk >= 0 &&
5411 !test_bit(Faulty, &rdev->flags) &&
5412 !test_bit(In_sync, &rdev->flags) &&
5413 rdev->recovery_offset < mddev->curr_resync)
5414 rdev->recovery_offset = mddev->curr_resync;
5417 set_bit(MD_CHANGE_DEVS, &mddev->flags);
5420 mddev->curr_resync = 0;
5421 wake_up(&resync_wait);
5422 set_bit(MD_RECOVERY_DONE, &mddev->recovery);
5423 md_wakeup_thread(mddev->thread);
5425 EXPORT_SYMBOL_GPL(md_do_sync);
5429 * This routine is regularly called by all per-raid-array threads to
5430 * deal with generic issues like resync and super-block update.
5431 * Raid personalities that don't have a thread (linear/raid0) do not
5432 * need this as they never do any recovery or update the superblock.
5434 * It does not do any resync itself, but rather "forks" off other threads
5435 * to do that as needed.
5436 * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
5437 * "->recovery" and create a thread at ->sync_thread.
5438 * When the thread finishes it sets MD_RECOVERY_DONE (and might set MD_RECOVERY_ERR)
5439 * and wakeups up this thread which will reap the thread and finish up.
5440 * This thread also removes any faulty devices (with nr_pending == 0).
5442 * The overall approach is:
5443 * 1/ if the superblock needs updating, update it.
5444 * 2/ If a recovery thread is running, don't do anything else.
5445 * 3/ If recovery has finished, clean up, possibly marking spares active.
5446 * 4/ If there are any faulty devices, remove them.
5447 * 5/ If array is degraded, try to add spares devices
5448 * 6/ If array has spares or is not in-sync, start a resync thread.
5450 void md_check_recovery(mddev_t *mddev)
5453 struct list_head *rtmp;
5457 bitmap_daemon_work(mddev->bitmap);
5462 if (signal_pending(current)) {
5463 if (mddev->pers->sync_request) {
5464 printk(KERN_INFO "md: %s in immediate safe mode\n",
5466 mddev->safemode = 2;
5468 flush_signals(current);
5473 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
5474 test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
5475 (mddev->safemode == 1) ||
5476 (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending)
5477 && !mddev->in_sync && mddev->recovery_cp == MaxSector)
5481 if (mddev_trylock(mddev)) {
5484 spin_lock_irq(&mddev->write_lock);
5485 if (mddev->safemode && !atomic_read(&mddev->writes_pending) &&
5486 !mddev->in_sync && mddev->recovery_cp == MaxSector) {
5488 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
5490 if (mddev->safemode == 1)
5491 mddev->safemode = 0;
5492 spin_unlock_irq(&mddev->write_lock);
5495 md_update_sb(mddev, 0);
5498 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
5499 !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
5500 /* resync/recovery still happening */
5501 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5504 if (mddev->sync_thread) {
5505 /* resync has finished, collect result */
5506 md_unregister_thread(mddev->sync_thread);
5507 mddev->sync_thread = NULL;
5508 if (!test_bit(MD_RECOVERY_ERR, &mddev->recovery) &&
5509 !test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
5511 /* activate any spares */
5512 mddev->pers->spare_active(mddev);
5514 md_update_sb(mddev, 1);
5516 /* if array is no-longer degraded, then any saved_raid_disk
5517 * information must be scrapped
5519 if (!mddev->degraded)
5520 ITERATE_RDEV(mddev,rdev,rtmp)
5521 rdev->saved_raid_disk = -1;
5523 mddev->recovery = 0;
5524 /* flag recovery needed just to double check */
5525 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5526 md_new_event(mddev);
5529 /* Clear some bits that don't mean anything, but
5532 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5533 clear_bit(MD_RECOVERY_ERR, &mddev->recovery);
5534 clear_bit(MD_RECOVERY_INTR, &mddev->recovery);
5535 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
5537 if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
5539 /* no recovery is running.
5540 * remove any failed drives, then
5541 * add spares if possible.
5542 * Spare are also removed and re-added, to allow
5543 * the personality to fail the re-add.
5545 ITERATE_RDEV(mddev,rdev,rtmp)
5546 if (rdev->raid_disk >= 0 &&
5547 (test_bit(Faulty, &rdev->flags) || ! test_bit(In_sync, &rdev->flags)) &&
5548 atomic_read(&rdev->nr_pending)==0) {
5549 if (mddev->pers->hot_remove_disk(mddev, rdev->raid_disk)==0) {
5551 sprintf(nm,"rd%d", rdev->raid_disk);
5552 sysfs_remove_link(&mddev->kobj, nm);
5553 rdev->raid_disk = -1;
5557 if (mddev->degraded) {
5558 ITERATE_RDEV(mddev,rdev,rtmp)
5559 if (rdev->raid_disk < 0
5560 && !test_bit(Faulty, &rdev->flags)) {
5561 rdev->recovery_offset = 0;
5562 if (mddev->pers->hot_add_disk(mddev,rdev)) {
5564 sprintf(nm, "rd%d", rdev->raid_disk);
5565 sysfs_create_link(&mddev->kobj, &rdev->kobj, nm);
5567 md_new_event(mddev);
5574 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
5575 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
5576 } else if (mddev->recovery_cp < MaxSector) {
5577 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
5578 } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
5579 /* nothing to be done ... */
5582 if (mddev->pers->sync_request) {
5583 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
5584 if (spares && mddev->bitmap && ! mddev->bitmap->file) {
5585 /* We are adding a device or devices to an array
5586 * which has the bitmap stored on all devices.
5587 * So make sure all bitmap pages get written
5589 bitmap_write_all(mddev->bitmap);
5591 mddev->sync_thread = md_register_thread(md_do_sync,
5594 if (!mddev->sync_thread) {
5595 printk(KERN_ERR "%s: could not start resync"
5598 /* leave the spares where they are, it shouldn't hurt */
5599 mddev->recovery = 0;
5601 md_wakeup_thread(mddev->sync_thread);
5602 md_new_event(mddev);
5605 mddev_unlock(mddev);
5609 static int md_notify_reboot(struct notifier_block *this,
5610 unsigned long code, void *x)
5612 struct list_head *tmp;
5615 if ((code == SYS_DOWN) || (code == SYS_HALT) || (code == SYS_POWER_OFF)) {
5617 printk(KERN_INFO "md: stopping all md devices.\n");
5619 ITERATE_MDDEV(mddev,tmp)
5620 if (mddev_trylock(mddev)) {
5621 do_md_stop (mddev, 1);
5622 mddev_unlock(mddev);
5625 * certain more exotic SCSI devices are known to be
5626 * volatile wrt too early system reboots. While the
5627 * right place to handle this issue is the given
5628 * driver, we do want to have a safe RAID driver ...
5635 static struct notifier_block md_notifier = {
5636 .notifier_call = md_notify_reboot,
5638 .priority = INT_MAX, /* before any real devices */
5641 static void md_geninit(void)
5643 struct proc_dir_entry *p;
5645 dprintk("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
5647 p = create_proc_entry("mdstat", S_IRUGO, NULL);
5649 p->proc_fops = &md_seq_fops;
5652 static int __init md_init(void)
5654 if (register_blkdev(MAJOR_NR, "md"))
5656 if ((mdp_major=register_blkdev(0, "mdp"))<=0) {
5657 unregister_blkdev(MAJOR_NR, "md");
5660 blk_register_region(MKDEV(MAJOR_NR, 0), 1UL<<MINORBITS, THIS_MODULE,
5661 md_probe, NULL, NULL);
5662 blk_register_region(MKDEV(mdp_major, 0), 1UL<<MINORBITS, THIS_MODULE,
5663 md_probe, NULL, NULL);
5665 register_reboot_notifier(&md_notifier);
5666 raid_table_header = register_sysctl_table(raid_root_table, 1);
5676 * Searches all registered partitions for autorun RAID arrays
5679 static dev_t detected_devices[128];
5682 void md_autodetect_dev(dev_t dev)
5684 if (dev_cnt >= 0 && dev_cnt < 127)
5685 detected_devices[dev_cnt++] = dev;
5689 static void autostart_arrays(int part)
5694 printk(KERN_INFO "md: Autodetecting RAID arrays.\n");
5696 for (i = 0; i < dev_cnt; i++) {
5697 dev_t dev = detected_devices[i];
5699 rdev = md_import_device(dev,0, 0);
5703 if (test_bit(Faulty, &rdev->flags)) {
5707 list_add(&rdev->same_set, &pending_raid_disks);
5711 autorun_devices(part);
5714 #endif /* !MODULE */
5716 static __exit void md_exit(void)
5719 struct list_head *tmp;
5721 blk_unregister_region(MKDEV(MAJOR_NR,0), 1U << MINORBITS);
5722 blk_unregister_region(MKDEV(mdp_major,0), 1U << MINORBITS);
5724 unregister_blkdev(MAJOR_NR,"md");
5725 unregister_blkdev(mdp_major, "mdp");
5726 unregister_reboot_notifier(&md_notifier);
5727 unregister_sysctl_table(raid_table_header);
5728 remove_proc_entry("mdstat", NULL);
5729 ITERATE_MDDEV(mddev,tmp) {
5730 struct gendisk *disk = mddev->gendisk;
5733 export_array(mddev);
5736 mddev->gendisk = NULL;
5741 module_init(md_init)
5742 module_exit(md_exit)
5744 static int get_ro(char *buffer, struct kernel_param *kp)
5746 return sprintf(buffer, "%d", start_readonly);
5748 static int set_ro(const char *val, struct kernel_param *kp)
5751 int num = simple_strtoul(val, &e, 10);
5752 if (*val && (*e == '\0' || *e == '\n')) {
5753 start_readonly = num;
5759 module_param_call(start_ro, set_ro, get_ro, NULL, S_IRUSR|S_IWUSR);
5760 module_param(start_dirty_degraded, int, S_IRUGO|S_IWUSR);
5763 EXPORT_SYMBOL(register_md_personality);
5764 EXPORT_SYMBOL(unregister_md_personality);
5765 EXPORT_SYMBOL(md_error);
5766 EXPORT_SYMBOL(md_done_sync);
5767 EXPORT_SYMBOL(md_write_start);
5768 EXPORT_SYMBOL(md_write_end);
5769 EXPORT_SYMBOL(md_register_thread);
5770 EXPORT_SYMBOL(md_unregister_thread);
5771 EXPORT_SYMBOL(md_wakeup_thread);
5772 EXPORT_SYMBOL(md_check_recovery);
5773 MODULE_LICENSE("GPL");
5775 MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR);