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 int match_mddev_units(mddev_t *mddev1, mddev_t *mddev2)
1302 struct list_head *tmp, *tmp2;
1303 mdk_rdev_t *rdev, *rdev2;
1305 ITERATE_RDEV(mddev1,rdev,tmp)
1306 ITERATE_RDEV(mddev2, rdev2, tmp2)
1307 if (rdev->bdev->bd_contains ==
1308 rdev2->bdev->bd_contains)
1314 static LIST_HEAD(pending_raid_disks);
1316 static int bind_rdev_to_array(mdk_rdev_t * rdev, mddev_t * mddev)
1318 char b[BDEVNAME_SIZE];
1326 /* make sure rdev->size exceeds mddev->size */
1327 if (rdev->size && (mddev->size == 0 || rdev->size < mddev->size)) {
1329 /* Cannot change size, so fail */
1332 mddev->size = rdev->size;
1335 /* Verify rdev->desc_nr is unique.
1336 * If it is -1, assign a free number, else
1337 * check number is not in use
1339 if (rdev->desc_nr < 0) {
1341 if (mddev->pers) choice = mddev->raid_disks;
1342 while (find_rdev_nr(mddev, choice))
1344 rdev->desc_nr = choice;
1346 if (find_rdev_nr(mddev, rdev->desc_nr))
1349 bdevname(rdev->bdev,b);
1350 if (kobject_set_name(&rdev->kobj, "dev-%s", b) < 0)
1352 while ( (s=strchr(rdev->kobj.k_name, '/')) != NULL)
1355 list_add(&rdev->same_set, &mddev->disks);
1356 rdev->mddev = mddev;
1357 printk(KERN_INFO "md: bind<%s>\n", b);
1359 rdev->kobj.parent = &mddev->kobj;
1360 kobject_add(&rdev->kobj);
1362 if (rdev->bdev->bd_part)
1363 ko = &rdev->bdev->bd_part->kobj;
1365 ko = &rdev->bdev->bd_disk->kobj;
1366 sysfs_create_link(&rdev->kobj, ko, "block");
1367 bd_claim_by_disk(rdev->bdev, rdev, mddev->gendisk);
1371 static void unbind_rdev_from_array(mdk_rdev_t * rdev)
1373 char b[BDEVNAME_SIZE];
1378 bd_release_from_disk(rdev->bdev, rdev->mddev->gendisk);
1379 list_del_init(&rdev->same_set);
1380 printk(KERN_INFO "md: unbind<%s>\n", bdevname(rdev->bdev,b));
1382 sysfs_remove_link(&rdev->kobj, "block");
1383 kobject_del(&rdev->kobj);
1387 * prevent the device from being mounted, repartitioned or
1388 * otherwise reused by a RAID array (or any other kernel
1389 * subsystem), by bd_claiming the device.
1391 static int lock_rdev(mdk_rdev_t *rdev, dev_t dev)
1394 struct block_device *bdev;
1395 char b[BDEVNAME_SIZE];
1397 bdev = open_by_devnum(dev, FMODE_READ|FMODE_WRITE);
1399 printk(KERN_ERR "md: could not open %s.\n",
1400 __bdevname(dev, b));
1401 return PTR_ERR(bdev);
1403 err = bd_claim(bdev, rdev);
1405 printk(KERN_ERR "md: could not bd_claim %s.\n",
1414 static void unlock_rdev(mdk_rdev_t *rdev)
1416 struct block_device *bdev = rdev->bdev;
1424 void md_autodetect_dev(dev_t dev);
1426 static void export_rdev(mdk_rdev_t * rdev)
1428 char b[BDEVNAME_SIZE];
1429 printk(KERN_INFO "md: export_rdev(%s)\n",
1430 bdevname(rdev->bdev,b));
1434 list_del_init(&rdev->same_set);
1436 md_autodetect_dev(rdev->bdev->bd_dev);
1439 kobject_put(&rdev->kobj);
1442 static void kick_rdev_from_array(mdk_rdev_t * rdev)
1444 unbind_rdev_from_array(rdev);
1448 static void export_array(mddev_t *mddev)
1450 struct list_head *tmp;
1453 ITERATE_RDEV(mddev,rdev,tmp) {
1458 kick_rdev_from_array(rdev);
1460 if (!list_empty(&mddev->disks))
1462 mddev->raid_disks = 0;
1463 mddev->major_version = 0;
1466 static void print_desc(mdp_disk_t *desc)
1468 printk(" DISK<N:%d,(%d,%d),R:%d,S:%d>\n", desc->number,
1469 desc->major,desc->minor,desc->raid_disk,desc->state);
1472 static void print_sb(mdp_super_t *sb)
1477 "md: SB: (V:%d.%d.%d) ID:<%08x.%08x.%08x.%08x> CT:%08x\n",
1478 sb->major_version, sb->minor_version, sb->patch_version,
1479 sb->set_uuid0, sb->set_uuid1, sb->set_uuid2, sb->set_uuid3,
1481 printk(KERN_INFO "md: L%d S%08d ND:%d RD:%d md%d LO:%d CS:%d\n",
1482 sb->level, sb->size, sb->nr_disks, sb->raid_disks,
1483 sb->md_minor, sb->layout, sb->chunk_size);
1484 printk(KERN_INFO "md: UT:%08x ST:%d AD:%d WD:%d"
1485 " FD:%d SD:%d CSUM:%08x E:%08lx\n",
1486 sb->utime, sb->state, sb->active_disks, sb->working_disks,
1487 sb->failed_disks, sb->spare_disks,
1488 sb->sb_csum, (unsigned long)sb->events_lo);
1491 for (i = 0; i < MD_SB_DISKS; i++) {
1494 desc = sb->disks + i;
1495 if (desc->number || desc->major || desc->minor ||
1496 desc->raid_disk || (desc->state && (desc->state != 4))) {
1497 printk(" D %2d: ", i);
1501 printk(KERN_INFO "md: THIS: ");
1502 print_desc(&sb->this_disk);
1506 static void print_rdev(mdk_rdev_t *rdev)
1508 char b[BDEVNAME_SIZE];
1509 printk(KERN_INFO "md: rdev %s, SZ:%08llu F:%d S:%d DN:%u\n",
1510 bdevname(rdev->bdev,b), (unsigned long long)rdev->size,
1511 test_bit(Faulty, &rdev->flags), test_bit(In_sync, &rdev->flags),
1513 if (rdev->sb_loaded) {
1514 printk(KERN_INFO "md: rdev superblock:\n");
1515 print_sb((mdp_super_t*)page_address(rdev->sb_page));
1517 printk(KERN_INFO "md: no rdev superblock!\n");
1520 static void md_print_devices(void)
1522 struct list_head *tmp, *tmp2;
1525 char b[BDEVNAME_SIZE];
1528 printk("md: **********************************\n");
1529 printk("md: * <COMPLETE RAID STATE PRINTOUT> *\n");
1530 printk("md: **********************************\n");
1531 ITERATE_MDDEV(mddev,tmp) {
1534 bitmap_print_sb(mddev->bitmap);
1536 printk("%s: ", mdname(mddev));
1537 ITERATE_RDEV(mddev,rdev,tmp2)
1538 printk("<%s>", bdevname(rdev->bdev,b));
1541 ITERATE_RDEV(mddev,rdev,tmp2)
1544 printk("md: **********************************\n");
1549 static void sync_sbs(mddev_t * mddev, int nospares)
1551 /* Update each superblock (in-memory image), but
1552 * if we are allowed to, skip spares which already
1553 * have the right event counter, or have one earlier
1554 * (which would mean they aren't being marked as dirty
1555 * with the rest of the array)
1558 struct list_head *tmp;
1560 ITERATE_RDEV(mddev,rdev,tmp) {
1561 if (rdev->sb_events == mddev->events ||
1563 rdev->raid_disk < 0 &&
1564 (rdev->sb_events&1)==0 &&
1565 rdev->sb_events+1 == mddev->events)) {
1566 /* Don't update this superblock */
1567 rdev->sb_loaded = 2;
1569 super_types[mddev->major_version].
1570 sync_super(mddev, rdev);
1571 rdev->sb_loaded = 1;
1576 static void md_update_sb(mddev_t * mddev, int force_change)
1579 struct list_head *tmp;
1585 spin_lock_irq(&mddev->write_lock);
1587 set_bit(MD_CHANGE_PENDING, &mddev->flags);
1588 if (test_and_clear_bit(MD_CHANGE_DEVS, &mddev->flags))
1590 if (test_and_clear_bit(MD_CHANGE_CLEAN, &mddev->flags))
1591 /* just a clean<-> dirty transition, possibly leave spares alone,
1592 * though if events isn't the right even/odd, we will have to do
1598 if (mddev->degraded)
1599 /* If the array is degraded, then skipping spares is both
1600 * dangerous and fairly pointless.
1601 * Dangerous because a device that was removed from the array
1602 * might have a event_count that still looks up-to-date,
1603 * so it can be re-added without a resync.
1604 * Pointless because if there are any spares to skip,
1605 * then a recovery will happen and soon that array won't
1606 * be degraded any more and the spare can go back to sleep then.
1610 sync_req = mddev->in_sync;
1611 mddev->utime = get_seconds();
1613 /* If this is just a dirty<->clean transition, and the array is clean
1614 * and 'events' is odd, we can roll back to the previous clean state */
1616 && (mddev->in_sync && mddev->recovery_cp == MaxSector)
1617 && (mddev->events & 1)
1618 && mddev->events != 1)
1621 /* otherwise we have to go forward and ... */
1623 if (!mddev->in_sync || mddev->recovery_cp != MaxSector) { /* not clean */
1624 /* .. if the array isn't clean, insist on an odd 'events' */
1625 if ((mddev->events&1)==0) {
1630 /* otherwise insist on an even 'events' (for clean states) */
1631 if ((mddev->events&1)) {
1638 if (!mddev->events) {
1640 * oops, this 64-bit counter should never wrap.
1641 * Either we are in around ~1 trillion A.C., assuming
1642 * 1 reboot per second, or we have a bug:
1647 sync_sbs(mddev, nospares);
1650 * do not write anything to disk if using
1651 * nonpersistent superblocks
1653 if (!mddev->persistent) {
1654 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
1655 spin_unlock_irq(&mddev->write_lock);
1656 wake_up(&mddev->sb_wait);
1659 spin_unlock_irq(&mddev->write_lock);
1662 "md: updating %s RAID superblock on device (in sync %d)\n",
1663 mdname(mddev),mddev->in_sync);
1665 err = bitmap_update_sb(mddev->bitmap);
1666 ITERATE_RDEV(mddev,rdev,tmp) {
1667 char b[BDEVNAME_SIZE];
1668 dprintk(KERN_INFO "md: ");
1669 if (rdev->sb_loaded != 1)
1670 continue; /* no noise on spare devices */
1671 if (test_bit(Faulty, &rdev->flags))
1672 dprintk("(skipping faulty ");
1674 dprintk("%s ", bdevname(rdev->bdev,b));
1675 if (!test_bit(Faulty, &rdev->flags)) {
1676 md_super_write(mddev,rdev,
1677 rdev->sb_offset<<1, rdev->sb_size,
1679 dprintk(KERN_INFO "(write) %s's sb offset: %llu\n",
1680 bdevname(rdev->bdev,b),
1681 (unsigned long long)rdev->sb_offset);
1682 rdev->sb_events = mddev->events;
1686 if (mddev->level == LEVEL_MULTIPATH)
1687 /* only need to write one superblock... */
1690 md_super_wait(mddev);
1691 /* if there was a failure, MD_CHANGE_DEVS was set, and we re-write super */
1693 spin_lock_irq(&mddev->write_lock);
1694 if (mddev->in_sync != sync_req ||
1695 test_bit(MD_CHANGE_DEVS, &mddev->flags)) {
1696 /* have to write it out again */
1697 spin_unlock_irq(&mddev->write_lock);
1700 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
1701 spin_unlock_irq(&mddev->write_lock);
1702 wake_up(&mddev->sb_wait);
1706 /* words written to sysfs files may, or my not, be \n terminated.
1707 * We want to accept with case. For this we use cmd_match.
1709 static int cmd_match(const char *cmd, const char *str)
1711 /* See if cmd, written into a sysfs file, matches
1712 * str. They must either be the same, or cmd can
1713 * have a trailing newline
1715 while (*cmd && *str && *cmd == *str) {
1726 struct rdev_sysfs_entry {
1727 struct attribute attr;
1728 ssize_t (*show)(mdk_rdev_t *, char *);
1729 ssize_t (*store)(mdk_rdev_t *, const char *, size_t);
1733 state_show(mdk_rdev_t *rdev, char *page)
1738 if (test_bit(Faulty, &rdev->flags)) {
1739 len+= sprintf(page+len, "%sfaulty",sep);
1742 if (test_bit(In_sync, &rdev->flags)) {
1743 len += sprintf(page+len, "%sin_sync",sep);
1746 if (test_bit(WriteMostly, &rdev->flags)) {
1747 len += sprintf(page+len, "%swrite_mostly",sep);
1750 if (!test_bit(Faulty, &rdev->flags) &&
1751 !test_bit(In_sync, &rdev->flags)) {
1752 len += sprintf(page+len, "%sspare", sep);
1755 return len+sprintf(page+len, "\n");
1759 state_store(mdk_rdev_t *rdev, const char *buf, size_t len)
1762 * faulty - simulates and error
1763 * remove - disconnects the device
1764 * writemostly - sets write_mostly
1765 * -writemostly - clears write_mostly
1768 if (cmd_match(buf, "faulty") && rdev->mddev->pers) {
1769 md_error(rdev->mddev, rdev);
1771 } else if (cmd_match(buf, "remove")) {
1772 if (rdev->raid_disk >= 0)
1775 mddev_t *mddev = rdev->mddev;
1776 kick_rdev_from_array(rdev);
1778 md_update_sb(mddev, 1);
1779 md_new_event(mddev);
1782 } else if (cmd_match(buf, "writemostly")) {
1783 set_bit(WriteMostly, &rdev->flags);
1785 } else if (cmd_match(buf, "-writemostly")) {
1786 clear_bit(WriteMostly, &rdev->flags);
1789 return err ? err : len;
1791 static struct rdev_sysfs_entry rdev_state =
1792 __ATTR(state, S_IRUGO|S_IWUSR, state_show, state_store);
1795 super_show(mdk_rdev_t *rdev, char *page)
1797 if (rdev->sb_loaded && rdev->sb_size) {
1798 memcpy(page, page_address(rdev->sb_page), rdev->sb_size);
1799 return rdev->sb_size;
1803 static struct rdev_sysfs_entry rdev_super = __ATTR_RO(super);
1806 errors_show(mdk_rdev_t *rdev, char *page)
1808 return sprintf(page, "%d\n", atomic_read(&rdev->corrected_errors));
1812 errors_store(mdk_rdev_t *rdev, const char *buf, size_t len)
1815 unsigned long n = simple_strtoul(buf, &e, 10);
1816 if (*buf && (*e == 0 || *e == '\n')) {
1817 atomic_set(&rdev->corrected_errors, n);
1822 static struct rdev_sysfs_entry rdev_errors =
1823 __ATTR(errors, S_IRUGO|S_IWUSR, errors_show, errors_store);
1826 slot_show(mdk_rdev_t *rdev, char *page)
1828 if (rdev->raid_disk < 0)
1829 return sprintf(page, "none\n");
1831 return sprintf(page, "%d\n", rdev->raid_disk);
1835 slot_store(mdk_rdev_t *rdev, const char *buf, size_t len)
1838 int slot = simple_strtoul(buf, &e, 10);
1839 if (strncmp(buf, "none", 4)==0)
1841 else if (e==buf || (*e && *e!= '\n'))
1843 if (rdev->mddev->pers)
1844 /* Cannot set slot in active array (yet) */
1846 if (slot >= rdev->mddev->raid_disks)
1848 rdev->raid_disk = slot;
1849 /* assume it is working */
1851 set_bit(In_sync, &rdev->flags);
1856 static struct rdev_sysfs_entry rdev_slot =
1857 __ATTR(slot, S_IRUGO|S_IWUSR, slot_show, slot_store);
1860 offset_show(mdk_rdev_t *rdev, char *page)
1862 return sprintf(page, "%llu\n", (unsigned long long)rdev->data_offset);
1866 offset_store(mdk_rdev_t *rdev, const char *buf, size_t len)
1869 unsigned long long offset = simple_strtoull(buf, &e, 10);
1870 if (e==buf || (*e && *e != '\n'))
1872 if (rdev->mddev->pers)
1874 rdev->data_offset = offset;
1878 static struct rdev_sysfs_entry rdev_offset =
1879 __ATTR(offset, S_IRUGO|S_IWUSR, offset_show, offset_store);
1882 rdev_size_show(mdk_rdev_t *rdev, char *page)
1884 return sprintf(page, "%llu\n", (unsigned long long)rdev->size);
1888 rdev_size_store(mdk_rdev_t *rdev, const char *buf, size_t len)
1891 unsigned long long size = simple_strtoull(buf, &e, 10);
1892 if (e==buf || (*e && *e != '\n'))
1894 if (rdev->mddev->pers)
1897 if (size < rdev->mddev->size || rdev->mddev->size == 0)
1898 rdev->mddev->size = size;
1902 static struct rdev_sysfs_entry rdev_size =
1903 __ATTR(size, S_IRUGO|S_IWUSR, rdev_size_show, rdev_size_store);
1905 static struct attribute *rdev_default_attrs[] = {
1915 rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
1917 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
1918 mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
1922 return entry->show(rdev, page);
1926 rdev_attr_store(struct kobject *kobj, struct attribute *attr,
1927 const char *page, size_t length)
1929 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
1930 mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
1934 if (!capable(CAP_SYS_ADMIN))
1936 return entry->store(rdev, page, length);
1939 static void rdev_free(struct kobject *ko)
1941 mdk_rdev_t *rdev = container_of(ko, mdk_rdev_t, kobj);
1944 static struct sysfs_ops rdev_sysfs_ops = {
1945 .show = rdev_attr_show,
1946 .store = rdev_attr_store,
1948 static struct kobj_type rdev_ktype = {
1949 .release = rdev_free,
1950 .sysfs_ops = &rdev_sysfs_ops,
1951 .default_attrs = rdev_default_attrs,
1955 * Import a device. If 'super_format' >= 0, then sanity check the superblock
1957 * mark the device faulty if:
1959 * - the device is nonexistent (zero size)
1960 * - the device has no valid superblock
1962 * a faulty rdev _never_ has rdev->sb set.
1964 static mdk_rdev_t *md_import_device(dev_t newdev, int super_format, int super_minor)
1966 char b[BDEVNAME_SIZE];
1971 rdev = kzalloc(sizeof(*rdev), GFP_KERNEL);
1973 printk(KERN_ERR "md: could not alloc mem for new device!\n");
1974 return ERR_PTR(-ENOMEM);
1977 if ((err = alloc_disk_sb(rdev)))
1980 err = lock_rdev(rdev, newdev);
1984 rdev->kobj.parent = NULL;
1985 rdev->kobj.ktype = &rdev_ktype;
1986 kobject_init(&rdev->kobj);
1989 rdev->saved_raid_disk = -1;
1990 rdev->raid_disk = -1;
1992 rdev->data_offset = 0;
1993 rdev->sb_events = 0;
1994 atomic_set(&rdev->nr_pending, 0);
1995 atomic_set(&rdev->read_errors, 0);
1996 atomic_set(&rdev->corrected_errors, 0);
1998 size = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
2001 "md: %s has zero or unknown size, marking faulty!\n",
2002 bdevname(rdev->bdev,b));
2007 if (super_format >= 0) {
2008 err = super_types[super_format].
2009 load_super(rdev, NULL, super_minor);
2010 if (err == -EINVAL) {
2012 "md: %s has invalid sb, not importing!\n",
2013 bdevname(rdev->bdev,b));
2018 "md: could not read %s's sb, not importing!\n",
2019 bdevname(rdev->bdev,b));
2023 INIT_LIST_HEAD(&rdev->same_set);
2028 if (rdev->sb_page) {
2034 return ERR_PTR(err);
2038 * Check a full RAID array for plausibility
2042 static void analyze_sbs(mddev_t * mddev)
2045 struct list_head *tmp;
2046 mdk_rdev_t *rdev, *freshest;
2047 char b[BDEVNAME_SIZE];
2050 ITERATE_RDEV(mddev,rdev,tmp)
2051 switch (super_types[mddev->major_version].
2052 load_super(rdev, freshest, mddev->minor_version)) {
2060 "md: fatal superblock inconsistency in %s"
2061 " -- removing from array\n",
2062 bdevname(rdev->bdev,b));
2063 kick_rdev_from_array(rdev);
2067 super_types[mddev->major_version].
2068 validate_super(mddev, freshest);
2071 ITERATE_RDEV(mddev,rdev,tmp) {
2072 if (rdev != freshest)
2073 if (super_types[mddev->major_version].
2074 validate_super(mddev, rdev)) {
2075 printk(KERN_WARNING "md: kicking non-fresh %s"
2077 bdevname(rdev->bdev,b));
2078 kick_rdev_from_array(rdev);
2081 if (mddev->level == LEVEL_MULTIPATH) {
2082 rdev->desc_nr = i++;
2083 rdev->raid_disk = rdev->desc_nr;
2084 set_bit(In_sync, &rdev->flags);
2090 if (mddev->recovery_cp != MaxSector &&
2092 printk(KERN_ERR "md: %s: raid array is not clean"
2093 " -- starting background reconstruction\n",
2099 safe_delay_show(mddev_t *mddev, char *page)
2101 int msec = (mddev->safemode_delay*1000)/HZ;
2102 return sprintf(page, "%d.%03d\n", msec/1000, msec%1000);
2105 safe_delay_store(mddev_t *mddev, const char *cbuf, size_t len)
2113 /* remove a period, and count digits after it */
2114 if (len >= sizeof(buf))
2116 strlcpy(buf, cbuf, len);
2118 for (i=0; i<len; i++) {
2120 if (isdigit(buf[i])) {
2125 } else if (buf[i] == '.') {
2130 msec = simple_strtoul(buf, &e, 10);
2131 if (e == buf || (*e && *e != '\n'))
2133 msec = (msec * 1000) / scale;
2135 mddev->safemode_delay = 0;
2137 mddev->safemode_delay = (msec*HZ)/1000;
2138 if (mddev->safemode_delay == 0)
2139 mddev->safemode_delay = 1;
2143 static struct md_sysfs_entry md_safe_delay =
2144 __ATTR(safe_mode_delay, S_IRUGO|S_IWUSR,safe_delay_show, safe_delay_store);
2147 level_show(mddev_t *mddev, char *page)
2149 struct mdk_personality *p = mddev->pers;
2151 return sprintf(page, "%s\n", p->name);
2152 else if (mddev->clevel[0])
2153 return sprintf(page, "%s\n", mddev->clevel);
2154 else if (mddev->level != LEVEL_NONE)
2155 return sprintf(page, "%d\n", mddev->level);
2161 level_store(mddev_t *mddev, const char *buf, size_t len)
2168 if (len >= sizeof(mddev->clevel))
2170 strncpy(mddev->clevel, buf, len);
2171 if (mddev->clevel[len-1] == '\n')
2173 mddev->clevel[len] = 0;
2174 mddev->level = LEVEL_NONE;
2178 static struct md_sysfs_entry md_level =
2179 __ATTR(level, S_IRUGO|S_IWUSR, level_show, level_store);
2183 layout_show(mddev_t *mddev, char *page)
2185 /* just a number, not meaningful for all levels */
2186 return sprintf(page, "%d\n", mddev->layout);
2190 layout_store(mddev_t *mddev, const char *buf, size_t len)
2193 unsigned long n = simple_strtoul(buf, &e, 10);
2197 if (!*buf || (*e && *e != '\n'))
2203 static struct md_sysfs_entry md_layout =
2204 __ATTR(layout, S_IRUGO|S_IWUSR, layout_show, layout_store);
2208 raid_disks_show(mddev_t *mddev, char *page)
2210 if (mddev->raid_disks == 0)
2212 return sprintf(page, "%d\n", mddev->raid_disks);
2215 static int update_raid_disks(mddev_t *mddev, int raid_disks);
2218 raid_disks_store(mddev_t *mddev, const char *buf, size_t len)
2222 unsigned long n = simple_strtoul(buf, &e, 10);
2224 if (!*buf || (*e && *e != '\n'))
2228 rv = update_raid_disks(mddev, n);
2230 mddev->raid_disks = n;
2231 return rv ? rv : len;
2233 static struct md_sysfs_entry md_raid_disks =
2234 __ATTR(raid_disks, S_IRUGO|S_IWUSR, raid_disks_show, raid_disks_store);
2237 chunk_size_show(mddev_t *mddev, char *page)
2239 return sprintf(page, "%d\n", mddev->chunk_size);
2243 chunk_size_store(mddev_t *mddev, const char *buf, size_t len)
2245 /* can only set chunk_size if array is not yet active */
2247 unsigned long n = simple_strtoul(buf, &e, 10);
2251 if (!*buf || (*e && *e != '\n'))
2254 mddev->chunk_size = n;
2257 static struct md_sysfs_entry md_chunk_size =
2258 __ATTR(chunk_size, S_IRUGO|S_IWUSR, chunk_size_show, chunk_size_store);
2261 resync_start_show(mddev_t *mddev, char *page)
2263 return sprintf(page, "%llu\n", (unsigned long long)mddev->recovery_cp);
2267 resync_start_store(mddev_t *mddev, const char *buf, size_t len)
2269 /* can only set chunk_size if array is not yet active */
2271 unsigned long long n = simple_strtoull(buf, &e, 10);
2275 if (!*buf || (*e && *e != '\n'))
2278 mddev->recovery_cp = n;
2281 static struct md_sysfs_entry md_resync_start =
2282 __ATTR(resync_start, S_IRUGO|S_IWUSR, resync_start_show, resync_start_store);
2285 * The array state can be:
2288 * No devices, no size, no level
2289 * Equivalent to STOP_ARRAY ioctl
2291 * May have some settings, but array is not active
2292 * all IO results in error
2293 * When written, doesn't tear down array, but just stops it
2294 * suspended (not supported yet)
2295 * All IO requests will block. The array can be reconfigured.
2296 * Writing this, if accepted, will block until array is quiessent
2298 * no resync can happen. no superblocks get written.
2299 * write requests fail
2301 * like readonly, but behaves like 'clean' on a write request.
2303 * clean - no pending writes, but otherwise active.
2304 * When written to inactive array, starts without resync
2305 * If a write request arrives then
2306 * if metadata is known, mark 'dirty' and switch to 'active'.
2307 * if not known, block and switch to write-pending
2308 * If written to an active array that has pending writes, then fails.
2310 * fully active: IO and resync can be happening.
2311 * When written to inactive array, starts with resync
2314 * clean, but writes are blocked waiting for 'active' to be written.
2317 * like active, but no writes have been seen for a while (100msec).
2320 enum array_state { clear, inactive, suspended, readonly, read_auto, clean, active,
2321 write_pending, active_idle, bad_word};
2322 static char *array_states[] = {
2323 "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active",
2324 "write-pending", "active-idle", NULL };
2326 static int match_word(const char *word, char **list)
2329 for (n=0; list[n]; n++)
2330 if (cmd_match(word, list[n]))
2336 array_state_show(mddev_t *mddev, char *page)
2338 enum array_state st = inactive;
2351 else if (mddev->safemode)
2357 if (list_empty(&mddev->disks) &&
2358 mddev->raid_disks == 0 &&
2364 return sprintf(page, "%s\n", array_states[st]);
2367 static int do_md_stop(mddev_t * mddev, int ro);
2368 static int do_md_run(mddev_t * mddev);
2369 static int restart_array(mddev_t *mddev);
2372 array_state_store(mddev_t *mddev, const char *buf, size_t len)
2375 enum array_state st = match_word(buf, array_states);
2380 /* stopping an active array */
2382 if (atomic_read(&mddev->active) > 1)
2384 err = do_md_stop(mddev, 0);
2388 /* stopping an active array */
2390 if (atomic_read(&mddev->active) > 1)
2392 err = do_md_stop(mddev, 2);
2396 break; /* not supported yet */
2399 err = do_md_stop(mddev, 1);
2402 err = do_md_run(mddev);
2406 /* stopping an active array */
2408 err = do_md_stop(mddev, 1);
2410 mddev->ro = 2; /* FIXME mark devices writable */
2413 err = do_md_run(mddev);
2418 restart_array(mddev);
2419 spin_lock_irq(&mddev->write_lock);
2420 if (atomic_read(&mddev->writes_pending) == 0) {
2422 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
2424 spin_unlock_irq(&mddev->write_lock);
2427 mddev->recovery_cp = MaxSector;
2428 err = do_md_run(mddev);
2433 restart_array(mddev);
2434 clear_bit(MD_CHANGE_CLEAN, &mddev->flags);
2435 wake_up(&mddev->sb_wait);
2439 err = do_md_run(mddev);
2444 /* these cannot be set */
2452 static struct md_sysfs_entry md_array_state =
2453 __ATTR(array_state, S_IRUGO|S_IWUSR, array_state_show, array_state_store);
2456 null_show(mddev_t *mddev, char *page)
2462 new_dev_store(mddev_t *mddev, const char *buf, size_t len)
2464 /* buf must be %d:%d\n? giving major and minor numbers */
2465 /* The new device is added to the array.
2466 * If the array has a persistent superblock, we read the
2467 * superblock to initialise info and check validity.
2468 * Otherwise, only checking done is that in bind_rdev_to_array,
2469 * which mainly checks size.
2472 int major = simple_strtoul(buf, &e, 10);
2478 if (!*buf || *e != ':' || !e[1] || e[1] == '\n')
2480 minor = simple_strtoul(e+1, &e, 10);
2481 if (*e && *e != '\n')
2483 dev = MKDEV(major, minor);
2484 if (major != MAJOR(dev) ||
2485 minor != MINOR(dev))
2489 if (mddev->persistent) {
2490 rdev = md_import_device(dev, mddev->major_version,
2491 mddev->minor_version);
2492 if (!IS_ERR(rdev) && !list_empty(&mddev->disks)) {
2493 mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
2494 mdk_rdev_t, same_set);
2495 err = super_types[mddev->major_version]
2496 .load_super(rdev, rdev0, mddev->minor_version);
2501 rdev = md_import_device(dev, -1, -1);
2504 return PTR_ERR(rdev);
2505 err = bind_rdev_to_array(rdev, mddev);
2509 return err ? err : len;
2512 static struct md_sysfs_entry md_new_device =
2513 __ATTR(new_dev, S_IWUSR, null_show, new_dev_store);
2516 bitmap_store(mddev_t *mddev, const char *buf, size_t len)
2519 unsigned long chunk, end_chunk;
2523 /* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */
2525 chunk = end_chunk = simple_strtoul(buf, &end, 0);
2526 if (buf == end) break;
2527 if (*end == '-') { /* range */
2529 end_chunk = simple_strtoul(buf, &end, 0);
2530 if (buf == end) break;
2532 if (*end && !isspace(*end)) break;
2533 bitmap_dirty_bits(mddev->bitmap, chunk, end_chunk);
2535 while (isspace(*buf)) buf++;
2537 bitmap_unplug(mddev->bitmap); /* flush the bits to disk */
2542 static struct md_sysfs_entry md_bitmap =
2543 __ATTR(bitmap_set_bits, S_IWUSR, null_show, bitmap_store);
2546 size_show(mddev_t *mddev, char *page)
2548 return sprintf(page, "%llu\n", (unsigned long long)mddev->size);
2551 static int update_size(mddev_t *mddev, unsigned long size);
2554 size_store(mddev_t *mddev, const char *buf, size_t len)
2556 /* If array is inactive, we can reduce the component size, but
2557 * not increase it (except from 0).
2558 * If array is active, we can try an on-line resize
2562 unsigned long long size = simple_strtoull(buf, &e, 10);
2563 if (!*buf || *buf == '\n' ||
2568 err = update_size(mddev, size);
2569 md_update_sb(mddev, 1);
2571 if (mddev->size == 0 ||
2577 return err ? err : len;
2580 static struct md_sysfs_entry md_size =
2581 __ATTR(component_size, S_IRUGO|S_IWUSR, size_show, size_store);
2585 * This is either 'none' for arrays with externally managed metadata,
2586 * or N.M for internally known formats
2589 metadata_show(mddev_t *mddev, char *page)
2591 if (mddev->persistent)
2592 return sprintf(page, "%d.%d\n",
2593 mddev->major_version, mddev->minor_version);
2595 return sprintf(page, "none\n");
2599 metadata_store(mddev_t *mddev, const char *buf, size_t len)
2603 if (!list_empty(&mddev->disks))
2606 if (cmd_match(buf, "none")) {
2607 mddev->persistent = 0;
2608 mddev->major_version = 0;
2609 mddev->minor_version = 90;
2612 major = simple_strtoul(buf, &e, 10);
2613 if (e==buf || *e != '.')
2616 minor = simple_strtoul(buf, &e, 10);
2617 if (e==buf || (*e && *e != '\n') )
2619 if (major >= sizeof(super_types)/sizeof(super_types[0]) ||
2620 super_types[major].name == NULL)
2622 mddev->major_version = major;
2623 mddev->minor_version = minor;
2624 mddev->persistent = 1;
2628 static struct md_sysfs_entry md_metadata =
2629 __ATTR(metadata_version, S_IRUGO|S_IWUSR, metadata_show, metadata_store);
2632 action_show(mddev_t *mddev, char *page)
2634 char *type = "idle";
2635 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
2636 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery)) {
2637 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
2639 else if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
2640 if (!test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
2642 else if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
2649 return sprintf(page, "%s\n", type);
2653 action_store(mddev_t *mddev, const char *page, size_t len)
2655 if (!mddev->pers || !mddev->pers->sync_request)
2658 if (cmd_match(page, "idle")) {
2659 if (mddev->sync_thread) {
2660 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
2661 md_unregister_thread(mddev->sync_thread);
2662 mddev->sync_thread = NULL;
2663 mddev->recovery = 0;
2665 } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
2666 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
2668 else if (cmd_match(page, "resync") || cmd_match(page, "recover"))
2669 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2670 else if (cmd_match(page, "reshape")) {
2672 if (mddev->pers->start_reshape == NULL)
2674 err = mddev->pers->start_reshape(mddev);
2678 if (cmd_match(page, "check"))
2679 set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
2680 else if (!cmd_match(page, "repair"))
2682 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
2683 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
2685 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2686 md_wakeup_thread(mddev->thread);
2691 mismatch_cnt_show(mddev_t *mddev, char *page)
2693 return sprintf(page, "%llu\n",
2694 (unsigned long long) mddev->resync_mismatches);
2697 static struct md_sysfs_entry md_scan_mode =
2698 __ATTR(sync_action, S_IRUGO|S_IWUSR, action_show, action_store);
2701 static struct md_sysfs_entry md_mismatches = __ATTR_RO(mismatch_cnt);
2704 sync_min_show(mddev_t *mddev, char *page)
2706 return sprintf(page, "%d (%s)\n", speed_min(mddev),
2707 mddev->sync_speed_min ? "local": "system");
2711 sync_min_store(mddev_t *mddev, const char *buf, size_t len)
2715 if (strncmp(buf, "system", 6)==0) {
2716 mddev->sync_speed_min = 0;
2719 min = simple_strtoul(buf, &e, 10);
2720 if (buf == e || (*e && *e != '\n') || min <= 0)
2722 mddev->sync_speed_min = min;
2726 static struct md_sysfs_entry md_sync_min =
2727 __ATTR(sync_speed_min, S_IRUGO|S_IWUSR, sync_min_show, sync_min_store);
2730 sync_max_show(mddev_t *mddev, char *page)
2732 return sprintf(page, "%d (%s)\n", speed_max(mddev),
2733 mddev->sync_speed_max ? "local": "system");
2737 sync_max_store(mddev_t *mddev, const char *buf, size_t len)
2741 if (strncmp(buf, "system", 6)==0) {
2742 mddev->sync_speed_max = 0;
2745 max = simple_strtoul(buf, &e, 10);
2746 if (buf == e || (*e && *e != '\n') || max <= 0)
2748 mddev->sync_speed_max = max;
2752 static struct md_sysfs_entry md_sync_max =
2753 __ATTR(sync_speed_max, S_IRUGO|S_IWUSR, sync_max_show, sync_max_store);
2757 sync_speed_show(mddev_t *mddev, char *page)
2759 unsigned long resync, dt, db;
2760 resync = (mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active));
2761 dt = ((jiffies - mddev->resync_mark) / HZ);
2763 db = resync - (mddev->resync_mark_cnt);
2764 return sprintf(page, "%ld\n", db/dt/2); /* K/sec */
2767 static struct md_sysfs_entry md_sync_speed = __ATTR_RO(sync_speed);
2770 sync_completed_show(mddev_t *mddev, char *page)
2772 unsigned long max_blocks, resync;
2774 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
2775 max_blocks = mddev->resync_max_sectors;
2777 max_blocks = mddev->size << 1;
2779 resync = (mddev->curr_resync - atomic_read(&mddev->recovery_active));
2780 return sprintf(page, "%lu / %lu\n", resync, max_blocks);
2783 static struct md_sysfs_entry md_sync_completed = __ATTR_RO(sync_completed);
2786 suspend_lo_show(mddev_t *mddev, char *page)
2788 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_lo);
2792 suspend_lo_store(mddev_t *mddev, const char *buf, size_t len)
2795 unsigned long long new = simple_strtoull(buf, &e, 10);
2797 if (mddev->pers->quiesce == NULL)
2799 if (buf == e || (*e && *e != '\n'))
2801 if (new >= mddev->suspend_hi ||
2802 (new > mddev->suspend_lo && new < mddev->suspend_hi)) {
2803 mddev->suspend_lo = new;
2804 mddev->pers->quiesce(mddev, 2);
2809 static struct md_sysfs_entry md_suspend_lo =
2810 __ATTR(suspend_lo, S_IRUGO|S_IWUSR, suspend_lo_show, suspend_lo_store);
2814 suspend_hi_show(mddev_t *mddev, char *page)
2816 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_hi);
2820 suspend_hi_store(mddev_t *mddev, const char *buf, size_t len)
2823 unsigned long long new = simple_strtoull(buf, &e, 10);
2825 if (mddev->pers->quiesce == NULL)
2827 if (buf == e || (*e && *e != '\n'))
2829 if ((new <= mddev->suspend_lo && mddev->suspend_lo >= mddev->suspend_hi) ||
2830 (new > mddev->suspend_lo && new > mddev->suspend_hi)) {
2831 mddev->suspend_hi = new;
2832 mddev->pers->quiesce(mddev, 1);
2833 mddev->pers->quiesce(mddev, 0);
2838 static struct md_sysfs_entry md_suspend_hi =
2839 __ATTR(suspend_hi, S_IRUGO|S_IWUSR, suspend_hi_show, suspend_hi_store);
2842 static struct attribute *md_default_attrs[] = {
2845 &md_raid_disks.attr,
2846 &md_chunk_size.attr,
2848 &md_resync_start.attr,
2850 &md_new_device.attr,
2851 &md_safe_delay.attr,
2852 &md_array_state.attr,
2856 static struct attribute *md_redundancy_attrs[] = {
2858 &md_mismatches.attr,
2861 &md_sync_speed.attr,
2862 &md_sync_completed.attr,
2863 &md_suspend_lo.attr,
2864 &md_suspend_hi.attr,
2868 static struct attribute_group md_redundancy_group = {
2870 .attrs = md_redundancy_attrs,
2875 md_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
2877 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
2878 mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
2883 rv = mddev_lock(mddev);
2885 rv = entry->show(mddev, page);
2886 mddev_unlock(mddev);
2892 md_attr_store(struct kobject *kobj, struct attribute *attr,
2893 const char *page, size_t length)
2895 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
2896 mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
2901 if (!capable(CAP_SYS_ADMIN))
2903 rv = mddev_lock(mddev);
2905 rv = entry->store(mddev, page, length);
2906 mddev_unlock(mddev);
2911 static void md_free(struct kobject *ko)
2913 mddev_t *mddev = container_of(ko, mddev_t, kobj);
2917 static struct sysfs_ops md_sysfs_ops = {
2918 .show = md_attr_show,
2919 .store = md_attr_store,
2921 static struct kobj_type md_ktype = {
2923 .sysfs_ops = &md_sysfs_ops,
2924 .default_attrs = md_default_attrs,
2929 static struct kobject *md_probe(dev_t dev, int *part, void *data)
2931 static DEFINE_MUTEX(disks_mutex);
2932 mddev_t *mddev = mddev_find(dev);
2933 struct gendisk *disk;
2934 int partitioned = (MAJOR(dev) != MD_MAJOR);
2935 int shift = partitioned ? MdpMinorShift : 0;
2936 int unit = MINOR(dev) >> shift;
2941 mutex_lock(&disks_mutex);
2942 if (mddev->gendisk) {
2943 mutex_unlock(&disks_mutex);
2947 disk = alloc_disk(1 << shift);
2949 mutex_unlock(&disks_mutex);
2953 disk->major = MAJOR(dev);
2954 disk->first_minor = unit << shift;
2956 sprintf(disk->disk_name, "md_d%d", unit);
2958 sprintf(disk->disk_name, "md%d", unit);
2959 disk->fops = &md_fops;
2960 disk->private_data = mddev;
2961 disk->queue = mddev->queue;
2963 mddev->gendisk = disk;
2964 mutex_unlock(&disks_mutex);
2965 mddev->kobj.parent = &disk->kobj;
2966 mddev->kobj.k_name = NULL;
2967 snprintf(mddev->kobj.name, KOBJ_NAME_LEN, "%s", "md");
2968 mddev->kobj.ktype = &md_ktype;
2969 kobject_register(&mddev->kobj);
2973 static void md_safemode_timeout(unsigned long data)
2975 mddev_t *mddev = (mddev_t *) data;
2977 mddev->safemode = 1;
2978 md_wakeup_thread(mddev->thread);
2981 static int start_dirty_degraded;
2983 static int do_md_run(mddev_t * mddev)
2987 struct list_head *tmp;
2989 struct gendisk *disk;
2990 struct mdk_personality *pers;
2991 char b[BDEVNAME_SIZE];
2993 if (list_empty(&mddev->disks))
2994 /* cannot run an array with no devices.. */
3001 * Analyze all RAID superblock(s)
3003 if (!mddev->raid_disks)
3006 chunk_size = mddev->chunk_size;
3009 if (chunk_size > MAX_CHUNK_SIZE) {
3010 printk(KERN_ERR "too big chunk_size: %d > %d\n",
3011 chunk_size, MAX_CHUNK_SIZE);
3015 * chunk-size has to be a power of 2 and multiples of PAGE_SIZE
3017 if ( (1 << ffz(~chunk_size)) != chunk_size) {
3018 printk(KERN_ERR "chunk_size of %d not valid\n", chunk_size);
3021 if (chunk_size < PAGE_SIZE) {
3022 printk(KERN_ERR "too small chunk_size: %d < %ld\n",
3023 chunk_size, PAGE_SIZE);
3027 /* devices must have minimum size of one chunk */
3028 ITERATE_RDEV(mddev,rdev,tmp) {
3029 if (test_bit(Faulty, &rdev->flags))
3031 if (rdev->size < chunk_size / 1024) {
3033 "md: Dev %s smaller than chunk_size:"
3035 bdevname(rdev->bdev,b),
3036 (unsigned long long)rdev->size,
3044 if (mddev->level != LEVEL_NONE)
3045 request_module("md-level-%d", mddev->level);
3046 else if (mddev->clevel[0])
3047 request_module("md-%s", mddev->clevel);
3051 * Drop all container device buffers, from now on
3052 * the only valid external interface is through the md
3054 * Also find largest hardsector size
3056 ITERATE_RDEV(mddev,rdev,tmp) {
3057 if (test_bit(Faulty, &rdev->flags))
3059 sync_blockdev(rdev->bdev);
3060 invalidate_bdev(rdev->bdev, 0);
3063 md_probe(mddev->unit, NULL, NULL);
3064 disk = mddev->gendisk;
3068 spin_lock(&pers_lock);
3069 pers = find_pers(mddev->level, mddev->clevel);
3070 if (!pers || !try_module_get(pers->owner)) {
3071 spin_unlock(&pers_lock);
3072 if (mddev->level != LEVEL_NONE)
3073 printk(KERN_WARNING "md: personality for level %d is not loaded!\n",
3076 printk(KERN_WARNING "md: personality for level %s is not loaded!\n",
3081 spin_unlock(&pers_lock);
3082 mddev->level = pers->level;
3083 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
3085 if (mddev->reshape_position != MaxSector &&
3086 pers->start_reshape == NULL) {
3087 /* This personality cannot handle reshaping... */
3089 module_put(pers->owner);
3093 if (pers->sync_request) {
3094 /* Warn if this is a potentially silly
3097 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
3099 struct list_head *tmp2;
3101 ITERATE_RDEV(mddev, rdev, tmp) {
3102 ITERATE_RDEV(mddev, rdev2, tmp2) {
3104 rdev->bdev->bd_contains ==
3105 rdev2->bdev->bd_contains) {
3107 "%s: WARNING: %s appears to be"
3108 " on the same physical disk as"
3111 bdevname(rdev->bdev,b),
3112 bdevname(rdev2->bdev,b2));
3119 "True protection against single-disk"
3120 " failure might be compromised.\n");
3123 mddev->recovery = 0;
3124 mddev->resync_max_sectors = mddev->size << 1; /* may be over-ridden by personality */
3125 mddev->barriers_work = 1;
3126 mddev->ok_start_degraded = start_dirty_degraded;
3129 mddev->ro = 2; /* read-only, but switch on first write */
3131 err = mddev->pers->run(mddev);
3132 if (!err && mddev->pers->sync_request) {
3133 err = bitmap_create(mddev);
3135 printk(KERN_ERR "%s: failed to create bitmap (%d)\n",
3136 mdname(mddev), err);
3137 mddev->pers->stop(mddev);
3141 printk(KERN_ERR "md: pers->run() failed ...\n");
3142 module_put(mddev->pers->owner);
3144 bitmap_destroy(mddev);
3147 if (mddev->pers->sync_request)
3148 sysfs_create_group(&mddev->kobj, &md_redundancy_group);
3149 else if (mddev->ro == 2) /* auto-readonly not meaningful */
3152 atomic_set(&mddev->writes_pending,0);
3153 mddev->safemode = 0;
3154 mddev->safemode_timer.function = md_safemode_timeout;
3155 mddev->safemode_timer.data = (unsigned long) mddev;
3156 mddev->safemode_delay = (200 * HZ)/1000 +1; /* 200 msec delay */
3159 ITERATE_RDEV(mddev,rdev,tmp)
3160 if (rdev->raid_disk >= 0) {
3162 sprintf(nm, "rd%d", rdev->raid_disk);
3163 sysfs_create_link(&mddev->kobj, &rdev->kobj, nm);
3166 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3169 md_update_sb(mddev, 0);
3171 set_capacity(disk, mddev->array_size<<1);
3173 /* If we call blk_queue_make_request here, it will
3174 * re-initialise max_sectors etc which may have been
3175 * refined inside -> run. So just set the bits we need to set.
3176 * Most initialisation happended when we called
3177 * blk_queue_make_request(..., md_fail_request)
3180 mddev->queue->queuedata = mddev;
3181 mddev->queue->make_request_fn = mddev->pers->make_request;
3183 /* If there is a partially-recovered drive we need to
3184 * start recovery here. If we leave it to md_check_recovery,
3185 * it will remove the drives and not do the right thing
3187 if (mddev->degraded && !mddev->sync_thread) {
3188 struct list_head *rtmp;
3190 ITERATE_RDEV(mddev,rdev,rtmp)
3191 if (rdev->raid_disk >= 0 &&
3192 !test_bit(In_sync, &rdev->flags) &&
3193 !test_bit(Faulty, &rdev->flags))
3194 /* complete an interrupted recovery */
3196 if (spares && mddev->pers->sync_request) {
3197 mddev->recovery = 0;
3198 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
3199 mddev->sync_thread = md_register_thread(md_do_sync,
3202 if (!mddev->sync_thread) {
3203 printk(KERN_ERR "%s: could not start resync"
3206 /* leave the spares where they are, it shouldn't hurt */
3207 mddev->recovery = 0;
3211 md_wakeup_thread(mddev->thread);
3212 md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
3215 md_new_event(mddev);
3216 kobject_uevent(&mddev->gendisk->kobj, KOBJ_CHANGE);
3220 static int restart_array(mddev_t *mddev)
3222 struct gendisk *disk = mddev->gendisk;
3226 * Complain if it has no devices
3229 if (list_empty(&mddev->disks))
3237 mddev->safemode = 0;
3239 set_disk_ro(disk, 0);
3241 printk(KERN_INFO "md: %s switched to read-write mode.\n",
3244 * Kick recovery or resync if necessary
3246 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3247 md_wakeup_thread(mddev->thread);
3248 md_wakeup_thread(mddev->sync_thread);
3257 /* similar to deny_write_access, but accounts for our holding a reference
3258 * to the file ourselves */
3259 static int deny_bitmap_write_access(struct file * file)
3261 struct inode *inode = file->f_mapping->host;
3263 spin_lock(&inode->i_lock);
3264 if (atomic_read(&inode->i_writecount) > 1) {
3265 spin_unlock(&inode->i_lock);
3268 atomic_set(&inode->i_writecount, -1);
3269 spin_unlock(&inode->i_lock);
3274 static void restore_bitmap_write_access(struct file *file)
3276 struct inode *inode = file->f_mapping->host;
3278 spin_lock(&inode->i_lock);
3279 atomic_set(&inode->i_writecount, 1);
3280 spin_unlock(&inode->i_lock);
3284 * 0 - completely stop and dis-assemble array
3285 * 1 - switch to readonly
3286 * 2 - stop but do not disassemble array
3288 static int do_md_stop(mddev_t * mddev, int mode)
3291 struct gendisk *disk = mddev->gendisk;
3294 if (atomic_read(&mddev->active)>2) {
3295 printk("md: %s still in use.\n",mdname(mddev));
3299 if (mddev->sync_thread) {
3300 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
3301 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
3302 md_unregister_thread(mddev->sync_thread);
3303 mddev->sync_thread = NULL;
3306 del_timer_sync(&mddev->safemode_timer);
3308 invalidate_partition(disk, 0);
3311 case 1: /* readonly */
3317 case 0: /* disassemble */
3319 bitmap_flush(mddev);
3320 md_super_wait(mddev);
3322 set_disk_ro(disk, 0);
3323 blk_queue_make_request(mddev->queue, md_fail_request);
3324 mddev->pers->stop(mddev);
3325 mddev->queue->merge_bvec_fn = NULL;
3326 mddev->queue->unplug_fn = NULL;
3327 mddev->queue->issue_flush_fn = NULL;
3328 if (mddev->pers->sync_request)
3329 sysfs_remove_group(&mddev->kobj, &md_redundancy_group);
3331 module_put(mddev->pers->owner);
3334 set_capacity(disk, 0);
3340 if (!mddev->in_sync || mddev->flags) {
3341 /* mark array as shutdown cleanly */
3343 md_update_sb(mddev, 1);
3346 set_disk_ro(disk, 1);
3347 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
3351 * Free resources if final stop
3355 struct list_head *tmp;
3357 printk(KERN_INFO "md: %s stopped.\n", mdname(mddev));
3359 bitmap_destroy(mddev);
3360 if (mddev->bitmap_file) {
3361 restore_bitmap_write_access(mddev->bitmap_file);
3362 fput(mddev->bitmap_file);
3363 mddev->bitmap_file = NULL;
3365 mddev->bitmap_offset = 0;
3367 ITERATE_RDEV(mddev,rdev,tmp)
3368 if (rdev->raid_disk >= 0) {
3370 sprintf(nm, "rd%d", rdev->raid_disk);
3371 sysfs_remove_link(&mddev->kobj, nm);
3374 export_array(mddev);
3376 mddev->array_size = 0;
3378 mddev->raid_disks = 0;
3379 mddev->recovery_cp = 0;
3381 } else if (mddev->pers)
3382 printk(KERN_INFO "md: %s switched to read-only mode.\n",
3385 md_new_event(mddev);
3391 static void autorun_array(mddev_t *mddev)
3394 struct list_head *tmp;
3397 if (list_empty(&mddev->disks))
3400 printk(KERN_INFO "md: running: ");
3402 ITERATE_RDEV(mddev,rdev,tmp) {
3403 char b[BDEVNAME_SIZE];
3404 printk("<%s>", bdevname(rdev->bdev,b));
3408 err = do_md_run (mddev);
3410 printk(KERN_WARNING "md: do_md_run() returned %d\n", err);
3411 do_md_stop (mddev, 0);
3416 * lets try to run arrays based on all disks that have arrived
3417 * until now. (those are in pending_raid_disks)
3419 * the method: pick the first pending disk, collect all disks with
3420 * the same UUID, remove all from the pending list and put them into
3421 * the 'same_array' list. Then order this list based on superblock
3422 * update time (freshest comes first), kick out 'old' disks and
3423 * compare superblocks. If everything's fine then run it.
3425 * If "unit" is allocated, then bump its reference count
3427 static void autorun_devices(int part)
3429 struct list_head *tmp;
3430 mdk_rdev_t *rdev0, *rdev;
3432 char b[BDEVNAME_SIZE];
3434 printk(KERN_INFO "md: autorun ...\n");
3435 while (!list_empty(&pending_raid_disks)) {
3438 LIST_HEAD(candidates);
3439 rdev0 = list_entry(pending_raid_disks.next,
3440 mdk_rdev_t, same_set);
3442 printk(KERN_INFO "md: considering %s ...\n",
3443 bdevname(rdev0->bdev,b));
3444 INIT_LIST_HEAD(&candidates);
3445 ITERATE_RDEV_PENDING(rdev,tmp)
3446 if (super_90_load(rdev, rdev0, 0) >= 0) {
3447 printk(KERN_INFO "md: adding %s ...\n",
3448 bdevname(rdev->bdev,b));
3449 list_move(&rdev->same_set, &candidates);
3452 * now we have a set of devices, with all of them having
3453 * mostly sane superblocks. It's time to allocate the
3457 dev = MKDEV(mdp_major,
3458 rdev0->preferred_minor << MdpMinorShift);
3459 unit = MINOR(dev) >> MdpMinorShift;
3461 dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
3464 if (rdev0->preferred_minor != unit) {
3465 printk(KERN_INFO "md: unit number in %s is bad: %d\n",
3466 bdevname(rdev0->bdev, b), rdev0->preferred_minor);
3470 md_probe(dev, NULL, NULL);
3471 mddev = mddev_find(dev);
3474 "md: cannot allocate memory for md drive.\n");
3477 if (mddev_lock(mddev))
3478 printk(KERN_WARNING "md: %s locked, cannot run\n",
3480 else if (mddev->raid_disks || mddev->major_version
3481 || !list_empty(&mddev->disks)) {
3483 "md: %s already running, cannot run %s\n",
3484 mdname(mddev), bdevname(rdev0->bdev,b));
3485 mddev_unlock(mddev);
3487 printk(KERN_INFO "md: created %s\n", mdname(mddev));
3488 ITERATE_RDEV_GENERIC(candidates,rdev,tmp) {
3489 list_del_init(&rdev->same_set);
3490 if (bind_rdev_to_array(rdev, mddev))
3493 autorun_array(mddev);
3494 mddev_unlock(mddev);
3496 /* on success, candidates will be empty, on error
3499 ITERATE_RDEV_GENERIC(candidates,rdev,tmp)
3503 printk(KERN_INFO "md: ... autorun DONE.\n");
3505 #endif /* !MODULE */
3507 static int get_version(void __user * arg)
3511 ver.major = MD_MAJOR_VERSION;
3512 ver.minor = MD_MINOR_VERSION;
3513 ver.patchlevel = MD_PATCHLEVEL_VERSION;
3515 if (copy_to_user(arg, &ver, sizeof(ver)))
3521 static int get_array_info(mddev_t * mddev, void __user * arg)
3523 mdu_array_info_t info;
3524 int nr,working,active,failed,spare;
3526 struct list_head *tmp;
3528 nr=working=active=failed=spare=0;
3529 ITERATE_RDEV(mddev,rdev,tmp) {
3531 if (test_bit(Faulty, &rdev->flags))
3535 if (test_bit(In_sync, &rdev->flags))
3542 info.major_version = mddev->major_version;
3543 info.minor_version = mddev->minor_version;
3544 info.patch_version = MD_PATCHLEVEL_VERSION;
3545 info.ctime = mddev->ctime;
3546 info.level = mddev->level;
3547 info.size = mddev->size;
3548 if (info.size != mddev->size) /* overflow */
3551 info.raid_disks = mddev->raid_disks;
3552 info.md_minor = mddev->md_minor;
3553 info.not_persistent= !mddev->persistent;
3555 info.utime = mddev->utime;
3558 info.state = (1<<MD_SB_CLEAN);
3559 if (mddev->bitmap && mddev->bitmap_offset)
3560 info.state = (1<<MD_SB_BITMAP_PRESENT);
3561 info.active_disks = active;
3562 info.working_disks = working;
3563 info.failed_disks = failed;
3564 info.spare_disks = spare;
3566 info.layout = mddev->layout;
3567 info.chunk_size = mddev->chunk_size;
3569 if (copy_to_user(arg, &info, sizeof(info)))
3575 static int get_bitmap_file(mddev_t * mddev, void __user * arg)
3577 mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */
3578 char *ptr, *buf = NULL;
3581 md_allow_write(mddev);
3583 file = kmalloc(sizeof(*file), GFP_KERNEL);
3587 /* bitmap disabled, zero the first byte and copy out */
3588 if (!mddev->bitmap || !mddev->bitmap->file) {
3589 file->pathname[0] = '\0';
3593 buf = kmalloc(sizeof(file->pathname), GFP_KERNEL);
3597 ptr = file_path(mddev->bitmap->file, buf, sizeof(file->pathname));
3601 strcpy(file->pathname, ptr);
3605 if (copy_to_user(arg, file, sizeof(*file)))
3613 static int get_disk_info(mddev_t * mddev, void __user * arg)
3615 mdu_disk_info_t info;
3619 if (copy_from_user(&info, arg, sizeof(info)))
3624 rdev = find_rdev_nr(mddev, nr);
3626 info.major = MAJOR(rdev->bdev->bd_dev);
3627 info.minor = MINOR(rdev->bdev->bd_dev);
3628 info.raid_disk = rdev->raid_disk;
3630 if (test_bit(Faulty, &rdev->flags))
3631 info.state |= (1<<MD_DISK_FAULTY);
3632 else if (test_bit(In_sync, &rdev->flags)) {
3633 info.state |= (1<<MD_DISK_ACTIVE);
3634 info.state |= (1<<MD_DISK_SYNC);
3636 if (test_bit(WriteMostly, &rdev->flags))
3637 info.state |= (1<<MD_DISK_WRITEMOSTLY);
3639 info.major = info.minor = 0;
3640 info.raid_disk = -1;
3641 info.state = (1<<MD_DISK_REMOVED);
3644 if (copy_to_user(arg, &info, sizeof(info)))
3650 static int add_new_disk(mddev_t * mddev, mdu_disk_info_t *info)
3652 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
3654 dev_t dev = MKDEV(info->major,info->minor);
3656 if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
3659 if (!mddev->raid_disks) {
3661 /* expecting a device which has a superblock */
3662 rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
3665 "md: md_import_device returned %ld\n",
3667 return PTR_ERR(rdev);
3669 if (!list_empty(&mddev->disks)) {
3670 mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
3671 mdk_rdev_t, same_set);
3672 int err = super_types[mddev->major_version]
3673 .load_super(rdev, rdev0, mddev->minor_version);
3676 "md: %s has different UUID to %s\n",
3677 bdevname(rdev->bdev,b),
3678 bdevname(rdev0->bdev,b2));
3683 err = bind_rdev_to_array(rdev, mddev);
3690 * add_new_disk can be used once the array is assembled
3691 * to add "hot spares". They must already have a superblock
3696 if (!mddev->pers->hot_add_disk) {
3698 "%s: personality does not support diskops!\n",
3702 if (mddev->persistent)
3703 rdev = md_import_device(dev, mddev->major_version,
3704 mddev->minor_version);
3706 rdev = md_import_device(dev, -1, -1);
3709 "md: md_import_device returned %ld\n",
3711 return PTR_ERR(rdev);
3713 /* set save_raid_disk if appropriate */
3714 if (!mddev->persistent) {
3715 if (info->state & (1<<MD_DISK_SYNC) &&
3716 info->raid_disk < mddev->raid_disks)
3717 rdev->raid_disk = info->raid_disk;
3719 rdev->raid_disk = -1;
3721 super_types[mddev->major_version].
3722 validate_super(mddev, rdev);
3723 rdev->saved_raid_disk = rdev->raid_disk;
3725 clear_bit(In_sync, &rdev->flags); /* just to be sure */
3726 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
3727 set_bit(WriteMostly, &rdev->flags);
3729 rdev->raid_disk = -1;
3730 err = bind_rdev_to_array(rdev, mddev);
3731 if (!err && !mddev->pers->hot_remove_disk) {
3732 /* If there is hot_add_disk but no hot_remove_disk
3733 * then added disks for geometry changes,
3734 * and should be added immediately.
3736 super_types[mddev->major_version].
3737 validate_super(mddev, rdev);
3738 err = mddev->pers->hot_add_disk(mddev, rdev);
3740 unbind_rdev_from_array(rdev);
3745 md_update_sb(mddev, 1);
3746 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3747 md_wakeup_thread(mddev->thread);
3751 /* otherwise, add_new_disk is only allowed
3752 * for major_version==0 superblocks
3754 if (mddev->major_version != 0) {
3755 printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n",
3760 if (!(info->state & (1<<MD_DISK_FAULTY))) {
3762 rdev = md_import_device (dev, -1, 0);
3765 "md: error, md_import_device() returned %ld\n",
3767 return PTR_ERR(rdev);
3769 rdev->desc_nr = info->number;
3770 if (info->raid_disk < mddev->raid_disks)
3771 rdev->raid_disk = info->raid_disk;
3773 rdev->raid_disk = -1;
3777 if (rdev->raid_disk < mddev->raid_disks)
3778 if (info->state & (1<<MD_DISK_SYNC))
3779 set_bit(In_sync, &rdev->flags);
3781 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
3782 set_bit(WriteMostly, &rdev->flags);
3784 if (!mddev->persistent) {
3785 printk(KERN_INFO "md: nonpersistent superblock ...\n");
3786 rdev->sb_offset = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
3788 rdev->sb_offset = calc_dev_sboffset(rdev->bdev);
3789 rdev->size = calc_dev_size(rdev, mddev->chunk_size);
3791 err = bind_rdev_to_array(rdev, mddev);
3801 static int hot_remove_disk(mddev_t * mddev, dev_t dev)
3803 char b[BDEVNAME_SIZE];
3809 rdev = find_rdev(mddev, dev);
3813 if (rdev->raid_disk >= 0)
3816 kick_rdev_from_array(rdev);
3817 md_update_sb(mddev, 1);
3818 md_new_event(mddev);
3822 printk(KERN_WARNING "md: cannot remove active disk %s from %s ... \n",
3823 bdevname(rdev->bdev,b), mdname(mddev));
3827 static int hot_add_disk(mddev_t * mddev, dev_t dev)
3829 char b[BDEVNAME_SIZE];
3837 if (mddev->major_version != 0) {
3838 printk(KERN_WARNING "%s: HOT_ADD may only be used with"
3839 " version-0 superblocks.\n",
3843 if (!mddev->pers->hot_add_disk) {
3845 "%s: personality does not support diskops!\n",
3850 rdev = md_import_device (dev, -1, 0);
3853 "md: error, md_import_device() returned %ld\n",
3858 if (mddev->persistent)
3859 rdev->sb_offset = calc_dev_sboffset(rdev->bdev);
3862 rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
3864 size = calc_dev_size(rdev, mddev->chunk_size);
3867 if (test_bit(Faulty, &rdev->flags)) {
3869 "md: can not hot-add faulty %s disk to %s!\n",
3870 bdevname(rdev->bdev,b), mdname(mddev));
3874 clear_bit(In_sync, &rdev->flags);
3876 rdev->saved_raid_disk = -1;
3877 err = bind_rdev_to_array(rdev, mddev);
3882 * The rest should better be atomic, we can have disk failures
3883 * noticed in interrupt contexts ...
3886 if (rdev->desc_nr == mddev->max_disks) {
3887 printk(KERN_WARNING "%s: can not hot-add to full array!\n",
3890 goto abort_unbind_export;
3893 rdev->raid_disk = -1;
3895 md_update_sb(mddev, 1);
3898 * Kick recovery, maybe this spare has to be added to the
3899 * array immediately.
3901 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3902 md_wakeup_thread(mddev->thread);
3903 md_new_event(mddev);
3906 abort_unbind_export:
3907 unbind_rdev_from_array(rdev);
3914 static int set_bitmap_file(mddev_t *mddev, int fd)
3919 if (!mddev->pers->quiesce)
3921 if (mddev->recovery || mddev->sync_thread)
3923 /* we should be able to change the bitmap.. */
3929 return -EEXIST; /* cannot add when bitmap is present */
3930 mddev->bitmap_file = fget(fd);
3932 if (mddev->bitmap_file == NULL) {
3933 printk(KERN_ERR "%s: error: failed to get bitmap file\n",
3938 err = deny_bitmap_write_access(mddev->bitmap_file);
3940 printk(KERN_ERR "%s: error: bitmap file is already in use\n",
3942 fput(mddev->bitmap_file);
3943 mddev->bitmap_file = NULL;
3946 mddev->bitmap_offset = 0; /* file overrides offset */
3947 } else if (mddev->bitmap == NULL)
3948 return -ENOENT; /* cannot remove what isn't there */
3951 mddev->pers->quiesce(mddev, 1);
3953 err = bitmap_create(mddev);
3954 if (fd < 0 || err) {
3955 bitmap_destroy(mddev);
3956 fd = -1; /* make sure to put the file */
3958 mddev->pers->quiesce(mddev, 0);
3961 if (mddev->bitmap_file) {
3962 restore_bitmap_write_access(mddev->bitmap_file);
3963 fput(mddev->bitmap_file);
3965 mddev->bitmap_file = NULL;
3972 * set_array_info is used two different ways
3973 * The original usage is when creating a new array.
3974 * In this usage, raid_disks is > 0 and it together with
3975 * level, size, not_persistent,layout,chunksize determine the
3976 * shape of the array.
3977 * This will always create an array with a type-0.90.0 superblock.
3978 * The newer usage is when assembling an array.
3979 * In this case raid_disks will be 0, and the major_version field is
3980 * use to determine which style super-blocks are to be found on the devices.
3981 * The minor and patch _version numbers are also kept incase the
3982 * super_block handler wishes to interpret them.
3984 static int set_array_info(mddev_t * mddev, mdu_array_info_t *info)
3987 if (info->raid_disks == 0) {
3988 /* just setting version number for superblock loading */
3989 if (info->major_version < 0 ||
3990 info->major_version >= sizeof(super_types)/sizeof(super_types[0]) ||
3991 super_types[info->major_version].name == NULL) {
3992 /* maybe try to auto-load a module? */
3994 "md: superblock version %d not known\n",
3995 info->major_version);
3998 mddev->major_version = info->major_version;
3999 mddev->minor_version = info->minor_version;
4000 mddev->patch_version = info->patch_version;
4001 mddev->persistent = !info->not_persistent;
4004 mddev->major_version = MD_MAJOR_VERSION;
4005 mddev->minor_version = MD_MINOR_VERSION;
4006 mddev->patch_version = MD_PATCHLEVEL_VERSION;
4007 mddev->ctime = get_seconds();
4009 mddev->level = info->level;
4010 mddev->clevel[0] = 0;
4011 mddev->size = info->size;
4012 mddev->raid_disks = info->raid_disks;
4013 /* don't set md_minor, it is determined by which /dev/md* was
4016 if (info->state & (1<<MD_SB_CLEAN))
4017 mddev->recovery_cp = MaxSector;
4019 mddev->recovery_cp = 0;
4020 mddev->persistent = ! info->not_persistent;
4022 mddev->layout = info->layout;
4023 mddev->chunk_size = info->chunk_size;
4025 mddev->max_disks = MD_SB_DISKS;
4028 set_bit(MD_CHANGE_DEVS, &mddev->flags);
4030 mddev->default_bitmap_offset = MD_SB_BYTES >> 9;
4031 mddev->bitmap_offset = 0;
4033 mddev->reshape_position = MaxSector;
4036 * Generate a 128 bit UUID
4038 get_random_bytes(mddev->uuid, 16);
4040 mddev->new_level = mddev->level;
4041 mddev->new_chunk = mddev->chunk_size;
4042 mddev->new_layout = mddev->layout;
4043 mddev->delta_disks = 0;
4048 static int update_size(mddev_t *mddev, unsigned long size)
4052 struct list_head *tmp;
4053 int fit = (size == 0);
4055 if (mddev->pers->resize == NULL)
4057 /* The "size" is the amount of each device that is used.
4058 * This can only make sense for arrays with redundancy.
4059 * linear and raid0 always use whatever space is available
4060 * We can only consider changing the size if no resync
4061 * or reconstruction is happening, and if the new size
4062 * is acceptable. It must fit before the sb_offset or,
4063 * if that is <data_offset, it must fit before the
4064 * size of each device.
4065 * If size is zero, we find the largest size that fits.
4067 if (mddev->sync_thread)
4069 ITERATE_RDEV(mddev,rdev,tmp) {
4071 avail = rdev->size * 2;
4073 if (fit && (size == 0 || size > avail/2))
4075 if (avail < ((sector_t)size << 1))
4078 rv = mddev->pers->resize(mddev, (sector_t)size *2);
4080 struct block_device *bdev;
4082 bdev = bdget_disk(mddev->gendisk, 0);
4084 mutex_lock(&bdev->bd_inode->i_mutex);
4085 i_size_write(bdev->bd_inode, (loff_t)mddev->array_size << 10);
4086 mutex_unlock(&bdev->bd_inode->i_mutex);
4093 static int update_raid_disks(mddev_t *mddev, int raid_disks)
4096 /* change the number of raid disks */
4097 if (mddev->pers->check_reshape == NULL)
4099 if (raid_disks <= 0 ||
4100 raid_disks >= mddev->max_disks)
4102 if (mddev->sync_thread || mddev->reshape_position != MaxSector)
4104 mddev->delta_disks = raid_disks - mddev->raid_disks;
4106 rv = mddev->pers->check_reshape(mddev);
4112 * update_array_info is used to change the configuration of an
4114 * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
4115 * fields in the info are checked against the array.
4116 * Any differences that cannot be handled will cause an error.
4117 * Normally, only one change can be managed at a time.
4119 static int update_array_info(mddev_t *mddev, mdu_array_info_t *info)
4125 /* calculate expected state,ignoring low bits */
4126 if (mddev->bitmap && mddev->bitmap_offset)
4127 state |= (1 << MD_SB_BITMAP_PRESENT);
4129 if (mddev->major_version != info->major_version ||
4130 mddev->minor_version != info->minor_version ||
4131 /* mddev->patch_version != info->patch_version || */
4132 mddev->ctime != info->ctime ||
4133 mddev->level != info->level ||
4134 /* mddev->layout != info->layout || */
4135 !mddev->persistent != info->not_persistent||
4136 mddev->chunk_size != info->chunk_size ||
4137 /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
4138 ((state^info->state) & 0xfffffe00)
4141 /* Check there is only one change */
4142 if (info->size >= 0 && mddev->size != info->size) cnt++;
4143 if (mddev->raid_disks != info->raid_disks) cnt++;
4144 if (mddev->layout != info->layout) cnt++;
4145 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) cnt++;
4146 if (cnt == 0) return 0;
4147 if (cnt > 1) return -EINVAL;
4149 if (mddev->layout != info->layout) {
4151 * we don't need to do anything at the md level, the
4152 * personality will take care of it all.
4154 if (mddev->pers->reconfig == NULL)
4157 return mddev->pers->reconfig(mddev, info->layout, -1);
4159 if (info->size >= 0 && mddev->size != info->size)
4160 rv = update_size(mddev, info->size);
4162 if (mddev->raid_disks != info->raid_disks)
4163 rv = update_raid_disks(mddev, info->raid_disks);
4165 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) {
4166 if (mddev->pers->quiesce == NULL)
4168 if (mddev->recovery || mddev->sync_thread)
4170 if (info->state & (1<<MD_SB_BITMAP_PRESENT)) {
4171 /* add the bitmap */
4174 if (mddev->default_bitmap_offset == 0)
4176 mddev->bitmap_offset = mddev->default_bitmap_offset;
4177 mddev->pers->quiesce(mddev, 1);
4178 rv = bitmap_create(mddev);
4180 bitmap_destroy(mddev);
4181 mddev->pers->quiesce(mddev, 0);
4183 /* remove the bitmap */
4186 if (mddev->bitmap->file)
4188 mddev->pers->quiesce(mddev, 1);
4189 bitmap_destroy(mddev);
4190 mddev->pers->quiesce(mddev, 0);
4191 mddev->bitmap_offset = 0;
4194 md_update_sb(mddev, 1);
4198 static int set_disk_faulty(mddev_t *mddev, dev_t dev)
4202 if (mddev->pers == NULL)
4205 rdev = find_rdev(mddev, dev);
4209 md_error(mddev, rdev);
4213 static int md_getgeo(struct block_device *bdev, struct hd_geometry *geo)
4215 mddev_t *mddev = bdev->bd_disk->private_data;
4219 geo->cylinders = get_capacity(mddev->gendisk) / 8;
4223 static int md_ioctl(struct inode *inode, struct file *file,
4224 unsigned int cmd, unsigned long arg)
4227 void __user *argp = (void __user *)arg;
4228 mddev_t *mddev = NULL;
4230 if (!capable(CAP_SYS_ADMIN))
4234 * Commands dealing with the RAID driver but not any
4240 err = get_version(argp);
4243 case PRINT_RAID_DEBUG:
4251 autostart_arrays(arg);
4258 * Commands creating/starting a new array:
4261 mddev = inode->i_bdev->bd_disk->private_data;
4268 err = mddev_lock(mddev);
4271 "md: ioctl lock interrupted, reason %d, cmd %d\n",
4278 case SET_ARRAY_INFO:
4280 mdu_array_info_t info;
4282 memset(&info, 0, sizeof(info));
4283 else if (copy_from_user(&info, argp, sizeof(info))) {
4288 err = update_array_info(mddev, &info);
4290 printk(KERN_WARNING "md: couldn't update"
4291 " array info. %d\n", err);
4296 if (!list_empty(&mddev->disks)) {
4298 "md: array %s already has disks!\n",
4303 if (mddev->raid_disks) {
4305 "md: array %s already initialised!\n",
4310 err = set_array_info(mddev, &info);
4312 printk(KERN_WARNING "md: couldn't set"
4313 " array info. %d\n", err);
4323 * Commands querying/configuring an existing array:
4325 /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
4326 * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */
4327 if (!mddev->raid_disks && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY
4328 && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE
4329 && cmd != GET_BITMAP_FILE) {
4335 * Commands even a read-only array can execute:
4339 case GET_ARRAY_INFO:
4340 err = get_array_info(mddev, argp);
4343 case GET_BITMAP_FILE:
4344 err = get_bitmap_file(mddev, argp);
4348 err = get_disk_info(mddev, argp);
4351 case RESTART_ARRAY_RW:
4352 err = restart_array(mddev);
4356 err = do_md_stop (mddev, 0);
4360 err = do_md_stop (mddev, 1);
4364 * We have a problem here : there is no easy way to give a CHS
4365 * virtual geometry. We currently pretend that we have a 2 heads
4366 * 4 sectors (with a BIG number of cylinders...). This drives
4367 * dosfs just mad... ;-)
4372 * The remaining ioctls are changing the state of the
4373 * superblock, so we do not allow them on read-only arrays.
4374 * However non-MD ioctls (e.g. get-size) will still come through
4375 * here and hit the 'default' below, so only disallow
4376 * 'md' ioctls, and switch to rw mode if started auto-readonly.
4378 if (_IOC_TYPE(cmd) == MD_MAJOR &&
4379 mddev->ro && mddev->pers) {
4380 if (mddev->ro == 2) {
4382 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4383 md_wakeup_thread(mddev->thread);
4395 mdu_disk_info_t info;
4396 if (copy_from_user(&info, argp, sizeof(info)))
4399 err = add_new_disk(mddev, &info);
4403 case HOT_REMOVE_DISK:
4404 err = hot_remove_disk(mddev, new_decode_dev(arg));
4408 err = hot_add_disk(mddev, new_decode_dev(arg));
4411 case SET_DISK_FAULTY:
4412 err = set_disk_faulty(mddev, new_decode_dev(arg));
4416 err = do_md_run (mddev);
4419 case SET_BITMAP_FILE:
4420 err = set_bitmap_file(mddev, (int)arg);
4430 mddev_unlock(mddev);
4440 static int md_open(struct inode *inode, struct file *file)
4443 * Succeed if we can lock the mddev, which confirms that
4444 * it isn't being stopped right now.
4446 mddev_t *mddev = inode->i_bdev->bd_disk->private_data;
4449 if ((err = mutex_lock_interruptible_nested(&mddev->reconfig_mutex, 1)))
4454 mddev_unlock(mddev);
4456 check_disk_change(inode->i_bdev);
4461 static int md_release(struct inode *inode, struct file * file)
4463 mddev_t *mddev = inode->i_bdev->bd_disk->private_data;
4471 static int md_media_changed(struct gendisk *disk)
4473 mddev_t *mddev = disk->private_data;
4475 return mddev->changed;
4478 static int md_revalidate(struct gendisk *disk)
4480 mddev_t *mddev = disk->private_data;
4485 static struct block_device_operations md_fops =
4487 .owner = THIS_MODULE,
4489 .release = md_release,
4491 .getgeo = md_getgeo,
4492 .media_changed = md_media_changed,
4493 .revalidate_disk= md_revalidate,
4496 static int md_thread(void * arg)
4498 mdk_thread_t *thread = arg;
4501 * md_thread is a 'system-thread', it's priority should be very
4502 * high. We avoid resource deadlocks individually in each
4503 * raid personality. (RAID5 does preallocation) We also use RR and
4504 * the very same RT priority as kswapd, thus we will never get
4505 * into a priority inversion deadlock.
4507 * we definitely have to have equal or higher priority than
4508 * bdflush, otherwise bdflush will deadlock if there are too
4509 * many dirty RAID5 blocks.
4512 current->flags |= PF_NOFREEZE;
4513 allow_signal(SIGKILL);
4514 while (!kthread_should_stop()) {
4516 /* We need to wait INTERRUPTIBLE so that
4517 * we don't add to the load-average.
4518 * That means we need to be sure no signals are
4521 if (signal_pending(current))
4522 flush_signals(current);
4524 wait_event_interruptible_timeout
4526 test_bit(THREAD_WAKEUP, &thread->flags)
4527 || kthread_should_stop(),
4530 clear_bit(THREAD_WAKEUP, &thread->flags);
4532 thread->run(thread->mddev);
4538 void md_wakeup_thread(mdk_thread_t *thread)
4541 dprintk("md: waking up MD thread %s.\n", thread->tsk->comm);
4542 set_bit(THREAD_WAKEUP, &thread->flags);
4543 wake_up(&thread->wqueue);
4547 mdk_thread_t *md_register_thread(void (*run) (mddev_t *), mddev_t *mddev,
4550 mdk_thread_t *thread;
4552 thread = kzalloc(sizeof(mdk_thread_t), GFP_KERNEL);
4556 init_waitqueue_head(&thread->wqueue);
4559 thread->mddev = mddev;
4560 thread->timeout = MAX_SCHEDULE_TIMEOUT;
4561 thread->tsk = kthread_run(md_thread, thread, name, mdname(thread->mddev));
4562 if (IS_ERR(thread->tsk)) {
4569 void md_unregister_thread(mdk_thread_t *thread)
4571 dprintk("interrupting MD-thread pid %d\n", thread->tsk->pid);
4573 kthread_stop(thread->tsk);
4577 void md_error(mddev_t *mddev, mdk_rdev_t *rdev)
4584 if (!rdev || test_bit(Faulty, &rdev->flags))
4587 dprintk("md_error dev:%s, rdev:(%d:%d), (caller: %p,%p,%p,%p).\n",
4589 MAJOR(rdev->bdev->bd_dev), MINOR(rdev->bdev->bd_dev),
4590 __builtin_return_address(0),__builtin_return_address(1),
4591 __builtin_return_address(2),__builtin_return_address(3));
4595 if (!mddev->pers->error_handler)
4597 mddev->pers->error_handler(mddev,rdev);
4598 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
4599 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4600 md_wakeup_thread(mddev->thread);
4601 md_new_event_inintr(mddev);
4604 /* seq_file implementation /proc/mdstat */
4606 static void status_unused(struct seq_file *seq)
4610 struct list_head *tmp;
4612 seq_printf(seq, "unused devices: ");
4614 ITERATE_RDEV_PENDING(rdev,tmp) {
4615 char b[BDEVNAME_SIZE];
4617 seq_printf(seq, "%s ",
4618 bdevname(rdev->bdev,b));
4621 seq_printf(seq, "<none>");
4623 seq_printf(seq, "\n");
4627 static void status_resync(struct seq_file *seq, mddev_t * mddev)
4629 sector_t max_blocks, resync, res;
4630 unsigned long dt, db, rt;
4632 unsigned int per_milli;
4634 resync = (mddev->curr_resync - atomic_read(&mddev->recovery_active))/2;
4636 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
4637 max_blocks = mddev->resync_max_sectors >> 1;
4639 max_blocks = mddev->size;
4642 * Should not happen.
4648 /* Pick 'scale' such that (resync>>scale)*1000 will fit
4649 * in a sector_t, and (max_blocks>>scale) will fit in a
4650 * u32, as those are the requirements for sector_div.
4651 * Thus 'scale' must be at least 10
4654 if (sizeof(sector_t) > sizeof(unsigned long)) {
4655 while ( max_blocks/2 > (1ULL<<(scale+32)))
4658 res = (resync>>scale)*1000;
4659 sector_div(res, (u32)((max_blocks>>scale)+1));
4663 int i, x = per_milli/50, y = 20-x;
4664 seq_printf(seq, "[");
4665 for (i = 0; i < x; i++)
4666 seq_printf(seq, "=");
4667 seq_printf(seq, ">");
4668 for (i = 0; i < y; i++)
4669 seq_printf(seq, ".");
4670 seq_printf(seq, "] ");
4672 seq_printf(seq, " %s =%3u.%u%% (%llu/%llu)",
4673 (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)?
4675 (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)?
4677 (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
4678 "resync" : "recovery"))),
4679 per_milli/10, per_milli % 10,
4680 (unsigned long long) resync,
4681 (unsigned long long) max_blocks);
4684 * We do not want to overflow, so the order of operands and
4685 * the * 100 / 100 trick are important. We do a +1 to be
4686 * safe against division by zero. We only estimate anyway.
4688 * dt: time from mark until now
4689 * db: blocks written from mark until now
4690 * rt: remaining time
4692 dt = ((jiffies - mddev->resync_mark) / HZ);
4694 db = (mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active))
4695 - mddev->resync_mark_cnt;
4696 rt = (dt * ((unsigned long)(max_blocks-resync) / (db/2/100+1)))/100;
4698 seq_printf(seq, " finish=%lu.%lumin", rt / 60, (rt % 60)/6);
4700 seq_printf(seq, " speed=%ldK/sec", db/2/dt);
4703 static void *md_seq_start(struct seq_file *seq, loff_t *pos)
4705 struct list_head *tmp;
4715 spin_lock(&all_mddevs_lock);
4716 list_for_each(tmp,&all_mddevs)
4718 mddev = list_entry(tmp, mddev_t, all_mddevs);
4720 spin_unlock(&all_mddevs_lock);
4723 spin_unlock(&all_mddevs_lock);
4725 return (void*)2;/* tail */
4729 static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4731 struct list_head *tmp;
4732 mddev_t *next_mddev, *mddev = v;
4738 spin_lock(&all_mddevs_lock);
4740 tmp = all_mddevs.next;
4742 tmp = mddev->all_mddevs.next;
4743 if (tmp != &all_mddevs)
4744 next_mddev = mddev_get(list_entry(tmp,mddev_t,all_mddevs));
4746 next_mddev = (void*)2;
4749 spin_unlock(&all_mddevs_lock);
4757 static void md_seq_stop(struct seq_file *seq, void *v)
4761 if (mddev && v != (void*)1 && v != (void*)2)
4765 struct mdstat_info {
4769 static int md_seq_show(struct seq_file *seq, void *v)
4773 struct list_head *tmp2;
4775 struct mdstat_info *mi = seq->private;
4776 struct bitmap *bitmap;
4778 if (v == (void*)1) {
4779 struct mdk_personality *pers;
4780 seq_printf(seq, "Personalities : ");
4781 spin_lock(&pers_lock);
4782 list_for_each_entry(pers, &pers_list, list)
4783 seq_printf(seq, "[%s] ", pers->name);
4785 spin_unlock(&pers_lock);
4786 seq_printf(seq, "\n");
4787 mi->event = atomic_read(&md_event_count);
4790 if (v == (void*)2) {
4795 if (mddev_lock(mddev) < 0)
4798 if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
4799 seq_printf(seq, "%s : %sactive", mdname(mddev),
4800 mddev->pers ? "" : "in");
4803 seq_printf(seq, " (read-only)");
4805 seq_printf(seq, "(auto-read-only)");
4806 seq_printf(seq, " %s", mddev->pers->name);
4810 ITERATE_RDEV(mddev,rdev,tmp2) {
4811 char b[BDEVNAME_SIZE];
4812 seq_printf(seq, " %s[%d]",
4813 bdevname(rdev->bdev,b), rdev->desc_nr);
4814 if (test_bit(WriteMostly, &rdev->flags))
4815 seq_printf(seq, "(W)");
4816 if (test_bit(Faulty, &rdev->flags)) {
4817 seq_printf(seq, "(F)");
4819 } else if (rdev->raid_disk < 0)
4820 seq_printf(seq, "(S)"); /* spare */
4824 if (!list_empty(&mddev->disks)) {
4826 seq_printf(seq, "\n %llu blocks",
4827 (unsigned long long)mddev->array_size);
4829 seq_printf(seq, "\n %llu blocks",
4830 (unsigned long long)size);
4832 if (mddev->persistent) {
4833 if (mddev->major_version != 0 ||
4834 mddev->minor_version != 90) {
4835 seq_printf(seq," super %d.%d",
4836 mddev->major_version,
4837 mddev->minor_version);
4840 seq_printf(seq, " super non-persistent");
4843 mddev->pers->status (seq, mddev);
4844 seq_printf(seq, "\n ");
4845 if (mddev->pers->sync_request) {
4846 if (mddev->curr_resync > 2) {
4847 status_resync (seq, mddev);
4848 seq_printf(seq, "\n ");
4849 } else if (mddev->curr_resync == 1 || mddev->curr_resync == 2)
4850 seq_printf(seq, "\tresync=DELAYED\n ");
4851 else if (mddev->recovery_cp < MaxSector)
4852 seq_printf(seq, "\tresync=PENDING\n ");
4855 seq_printf(seq, "\n ");
4857 if ((bitmap = mddev->bitmap)) {
4858 unsigned long chunk_kb;
4859 unsigned long flags;
4860 spin_lock_irqsave(&bitmap->lock, flags);
4861 chunk_kb = bitmap->chunksize >> 10;
4862 seq_printf(seq, "bitmap: %lu/%lu pages [%luKB], "
4864 bitmap->pages - bitmap->missing_pages,
4866 (bitmap->pages - bitmap->missing_pages)
4867 << (PAGE_SHIFT - 10),
4868 chunk_kb ? chunk_kb : bitmap->chunksize,
4869 chunk_kb ? "KB" : "B");
4871 seq_printf(seq, ", file: ");
4872 seq_path(seq, bitmap->file->f_path.mnt,
4873 bitmap->file->f_path.dentry," \t\n");
4876 seq_printf(seq, "\n");
4877 spin_unlock_irqrestore(&bitmap->lock, flags);
4880 seq_printf(seq, "\n");
4882 mddev_unlock(mddev);
4887 static struct seq_operations md_seq_ops = {
4888 .start = md_seq_start,
4889 .next = md_seq_next,
4890 .stop = md_seq_stop,
4891 .show = md_seq_show,
4894 static int md_seq_open(struct inode *inode, struct file *file)
4897 struct mdstat_info *mi = kmalloc(sizeof(*mi), GFP_KERNEL);
4901 error = seq_open(file, &md_seq_ops);
4905 struct seq_file *p = file->private_data;
4907 mi->event = atomic_read(&md_event_count);
4912 static int md_seq_release(struct inode *inode, struct file *file)
4914 struct seq_file *m = file->private_data;
4915 struct mdstat_info *mi = m->private;
4918 return seq_release(inode, file);
4921 static unsigned int mdstat_poll(struct file *filp, poll_table *wait)
4923 struct seq_file *m = filp->private_data;
4924 struct mdstat_info *mi = m->private;
4927 poll_wait(filp, &md_event_waiters, wait);
4929 /* always allow read */
4930 mask = POLLIN | POLLRDNORM;
4932 if (mi->event != atomic_read(&md_event_count))
4933 mask |= POLLERR | POLLPRI;
4937 static const struct file_operations md_seq_fops = {
4938 .owner = THIS_MODULE,
4939 .open = md_seq_open,
4941 .llseek = seq_lseek,
4942 .release = md_seq_release,
4943 .poll = mdstat_poll,
4946 int register_md_personality(struct mdk_personality *p)
4948 spin_lock(&pers_lock);
4949 list_add_tail(&p->list, &pers_list);
4950 printk(KERN_INFO "md: %s personality registered for level %d\n", p->name, p->level);
4951 spin_unlock(&pers_lock);
4955 int unregister_md_personality(struct mdk_personality *p)
4957 printk(KERN_INFO "md: %s personality unregistered\n", p->name);
4958 spin_lock(&pers_lock);
4959 list_del_init(&p->list);
4960 spin_unlock(&pers_lock);
4964 static int is_mddev_idle(mddev_t *mddev)
4967 struct list_head *tmp;
4969 unsigned long curr_events;
4972 ITERATE_RDEV(mddev,rdev,tmp) {
4973 struct gendisk *disk = rdev->bdev->bd_contains->bd_disk;
4974 curr_events = disk_stat_read(disk, sectors[0]) +
4975 disk_stat_read(disk, sectors[1]) -
4976 atomic_read(&disk->sync_io);
4977 /* The difference between curr_events and last_events
4978 * will be affected by any new non-sync IO (making
4979 * curr_events bigger) and any difference in the amount of
4980 * in-flight syncio (making current_events bigger or smaller)
4981 * The amount in-flight is currently limited to
4982 * 32*64K in raid1/10 and 256*PAGE_SIZE in raid5/6
4983 * which is at most 4096 sectors.
4984 * These numbers are fairly fragile and should be made
4985 * more robust, probably by enforcing the
4986 * 'window size' that md_do_sync sort-of uses.
4988 * Note: the following is an unsigned comparison.
4990 if ((curr_events - rdev->last_events + 4096) > 8192) {
4991 rdev->last_events = curr_events;
4998 void md_done_sync(mddev_t *mddev, int blocks, int ok)
5000 /* another "blocks" (512byte) blocks have been synced */
5001 atomic_sub(blocks, &mddev->recovery_active);
5002 wake_up(&mddev->recovery_wait);
5004 set_bit(MD_RECOVERY_ERR, &mddev->recovery);
5005 md_wakeup_thread(mddev->thread);
5006 // stop recovery, signal do_sync ....
5011 /* md_write_start(mddev, bi)
5012 * If we need to update some array metadata (e.g. 'active' flag
5013 * in superblock) before writing, schedule a superblock update
5014 * and wait for it to complete.
5016 void md_write_start(mddev_t *mddev, struct bio *bi)
5018 if (bio_data_dir(bi) != WRITE)
5021 BUG_ON(mddev->ro == 1);
5022 if (mddev->ro == 2) {
5023 /* need to switch to read/write */
5025 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5026 md_wakeup_thread(mddev->thread);
5028 atomic_inc(&mddev->writes_pending);
5029 if (mddev->in_sync) {
5030 spin_lock_irq(&mddev->write_lock);
5031 if (mddev->in_sync) {
5033 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
5034 md_wakeup_thread(mddev->thread);
5036 spin_unlock_irq(&mddev->write_lock);
5038 wait_event(mddev->sb_wait, mddev->flags==0);
5041 void md_write_end(mddev_t *mddev)
5043 if (atomic_dec_and_test(&mddev->writes_pending)) {
5044 if (mddev->safemode == 2)
5045 md_wakeup_thread(mddev->thread);
5046 else if (mddev->safemode_delay)
5047 mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay);
5051 /* md_allow_write(mddev)
5052 * Calling this ensures that the array is marked 'active' so that writes
5053 * may proceed without blocking. It is important to call this before
5054 * attempting a GFP_KERNEL allocation while holding the mddev lock.
5055 * Must be called with mddev_lock held.
5057 void md_allow_write(mddev_t *mddev)
5064 spin_lock_irq(&mddev->write_lock);
5065 if (mddev->in_sync) {
5067 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
5068 if (mddev->safemode_delay &&
5069 mddev->safemode == 0)
5070 mddev->safemode = 1;
5071 spin_unlock_irq(&mddev->write_lock);
5072 md_update_sb(mddev, 0);
5074 spin_unlock_irq(&mddev->write_lock);
5076 EXPORT_SYMBOL_GPL(md_allow_write);
5078 static DECLARE_WAIT_QUEUE_HEAD(resync_wait);
5080 #define SYNC_MARKS 10
5081 #define SYNC_MARK_STEP (3*HZ)
5082 void md_do_sync(mddev_t *mddev)
5085 unsigned int currspeed = 0,
5087 sector_t max_sectors,j, io_sectors;
5088 unsigned long mark[SYNC_MARKS];
5089 sector_t mark_cnt[SYNC_MARKS];
5091 struct list_head *tmp;
5092 sector_t last_check;
5094 struct list_head *rtmp;
5098 /* just incase thread restarts... */
5099 if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
5101 if (mddev->ro) /* never try to sync a read-only array */
5104 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
5105 if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
5106 desc = "data-check";
5107 else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
5108 desc = "requested-resync";
5111 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
5116 /* we overload curr_resync somewhat here.
5117 * 0 == not engaged in resync at all
5118 * 2 == checking that there is no conflict with another sync
5119 * 1 == like 2, but have yielded to allow conflicting resync to
5121 * other == active in resync - this many blocks
5123 * Before starting a resync we must have set curr_resync to
5124 * 2, and then checked that every "conflicting" array has curr_resync
5125 * less than ours. When we find one that is the same or higher
5126 * we wait on resync_wait. To avoid deadlock, we reduce curr_resync
5127 * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
5128 * This will mean we have to start checking from the beginning again.
5133 mddev->curr_resync = 2;
5136 if (kthread_should_stop()) {
5137 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5140 ITERATE_MDDEV(mddev2,tmp) {
5141 if (mddev2 == mddev)
5143 if (mddev2->curr_resync &&
5144 match_mddev_units(mddev,mddev2)) {
5146 if (mddev < mddev2 && mddev->curr_resync == 2) {
5147 /* arbitrarily yield */
5148 mddev->curr_resync = 1;
5149 wake_up(&resync_wait);
5151 if (mddev > mddev2 && mddev->curr_resync == 1)
5152 /* no need to wait here, we can wait the next
5153 * time 'round when curr_resync == 2
5156 prepare_to_wait(&resync_wait, &wq, TASK_UNINTERRUPTIBLE);
5157 if (!kthread_should_stop() &&
5158 mddev2->curr_resync >= mddev->curr_resync) {
5159 printk(KERN_INFO "md: delaying %s of %s"
5160 " until %s has finished (they"
5161 " share one or more physical units)\n",
5162 desc, mdname(mddev), mdname(mddev2));
5165 finish_wait(&resync_wait, &wq);
5168 finish_wait(&resync_wait, &wq);
5171 } while (mddev->curr_resync < 2);
5174 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
5175 /* resync follows the size requested by the personality,
5176 * which defaults to physical size, but can be virtual size
5178 max_sectors = mddev->resync_max_sectors;
5179 mddev->resync_mismatches = 0;
5180 /* we don't use the checkpoint if there's a bitmap */
5181 if (!mddev->bitmap &&
5182 !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
5183 j = mddev->recovery_cp;
5184 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
5185 max_sectors = mddev->size << 1;
5187 /* recovery follows the physical size of devices */
5188 max_sectors = mddev->size << 1;
5190 ITERATE_RDEV(mddev,rdev,rtmp)
5191 if (rdev->raid_disk >= 0 &&
5192 !test_bit(Faulty, &rdev->flags) &&
5193 !test_bit(In_sync, &rdev->flags) &&
5194 rdev->recovery_offset < j)
5195 j = rdev->recovery_offset;
5198 printk(KERN_INFO "md: %s of RAID array %s\n", desc, mdname(mddev));
5199 printk(KERN_INFO "md: minimum _guaranteed_ speed:"
5200 " %d KB/sec/disk.\n", speed_min(mddev));
5201 printk(KERN_INFO "md: using maximum available idle IO bandwidth "
5202 "(but not more than %d KB/sec) for %s.\n",
5203 speed_max(mddev), desc);
5205 is_mddev_idle(mddev); /* this also initializes IO event counters */
5208 for (m = 0; m < SYNC_MARKS; m++) {
5210 mark_cnt[m] = io_sectors;
5213 mddev->resync_mark = mark[last_mark];
5214 mddev->resync_mark_cnt = mark_cnt[last_mark];
5217 * Tune reconstruction:
5219 window = 32*(PAGE_SIZE/512);
5220 printk(KERN_INFO "md: using %dk window, over a total of %llu blocks.\n",
5221 window/2,(unsigned long long) max_sectors/2);
5223 atomic_set(&mddev->recovery_active, 0);
5224 init_waitqueue_head(&mddev->recovery_wait);
5229 "md: resuming %s of %s from checkpoint.\n",
5230 desc, mdname(mddev));
5231 mddev->curr_resync = j;
5234 while (j < max_sectors) {
5238 sectors = mddev->pers->sync_request(mddev, j, &skipped,
5239 currspeed < speed_min(mddev));
5241 set_bit(MD_RECOVERY_ERR, &mddev->recovery);
5245 if (!skipped) { /* actual IO requested */
5246 io_sectors += sectors;
5247 atomic_add(sectors, &mddev->recovery_active);
5251 if (j>1) mddev->curr_resync = j;
5252 mddev->curr_mark_cnt = io_sectors;
5253 if (last_check == 0)
5254 /* this is the earliers that rebuilt will be
5255 * visible in /proc/mdstat
5257 md_new_event(mddev);
5259 if (last_check + window > io_sectors || j == max_sectors)
5262 last_check = io_sectors;
5264 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery) ||
5265 test_bit(MD_RECOVERY_ERR, &mddev->recovery))
5269 if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) {
5271 int next = (last_mark+1) % SYNC_MARKS;
5273 mddev->resync_mark = mark[next];
5274 mddev->resync_mark_cnt = mark_cnt[next];
5275 mark[next] = jiffies;
5276 mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active);
5281 if (kthread_should_stop()) {
5283 * got a signal, exit.
5286 "md: md_do_sync() got signal ... exiting\n");
5287 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5292 * this loop exits only if either when we are slower than
5293 * the 'hard' speed limit, or the system was IO-idle for
5295 * the system might be non-idle CPU-wise, but we only care
5296 * about not overloading the IO subsystem. (things like an
5297 * e2fsck being done on the RAID array should execute fast)
5299 mddev->queue->unplug_fn(mddev->queue);
5302 currspeed = ((unsigned long)(io_sectors-mddev->resync_mark_cnt))/2
5303 /((jiffies-mddev->resync_mark)/HZ +1) +1;
5305 if (currspeed > speed_min(mddev)) {
5306 if ((currspeed > speed_max(mddev)) ||
5307 !is_mddev_idle(mddev)) {
5313 printk(KERN_INFO "md: %s: %s done.\n",mdname(mddev), desc);
5315 * this also signals 'finished resyncing' to md_stop
5318 mddev->queue->unplug_fn(mddev->queue);
5320 wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
5322 /* tell personality that we are finished */
5323 mddev->pers->sync_request(mddev, max_sectors, &skipped, 1);
5325 if (!test_bit(MD_RECOVERY_ERR, &mddev->recovery) &&
5326 !test_bit(MD_RECOVERY_CHECK, &mddev->recovery) &&
5327 mddev->curr_resync > 2) {
5328 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
5329 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
5330 if (mddev->curr_resync >= mddev->recovery_cp) {
5332 "md: checkpointing %s of %s.\n",
5333 desc, mdname(mddev));
5334 mddev->recovery_cp = mddev->curr_resync;
5337 mddev->recovery_cp = MaxSector;
5339 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery))
5340 mddev->curr_resync = MaxSector;
5341 ITERATE_RDEV(mddev,rdev,rtmp)
5342 if (rdev->raid_disk >= 0 &&
5343 !test_bit(Faulty, &rdev->flags) &&
5344 !test_bit(In_sync, &rdev->flags) &&
5345 rdev->recovery_offset < mddev->curr_resync)
5346 rdev->recovery_offset = mddev->curr_resync;
5349 set_bit(MD_CHANGE_DEVS, &mddev->flags);
5352 mddev->curr_resync = 0;
5353 wake_up(&resync_wait);
5354 set_bit(MD_RECOVERY_DONE, &mddev->recovery);
5355 md_wakeup_thread(mddev->thread);
5357 EXPORT_SYMBOL_GPL(md_do_sync);
5360 static int remove_and_add_spares(mddev_t *mddev)
5363 struct list_head *rtmp;
5366 ITERATE_RDEV(mddev,rdev,rtmp)
5367 if (rdev->raid_disk >= 0 &&
5368 (test_bit(Faulty, &rdev->flags) ||
5369 ! test_bit(In_sync, &rdev->flags)) &&
5370 atomic_read(&rdev->nr_pending)==0) {
5371 if (mddev->pers->hot_remove_disk(
5372 mddev, rdev->raid_disk)==0) {
5374 sprintf(nm,"rd%d", rdev->raid_disk);
5375 sysfs_remove_link(&mddev->kobj, nm);
5376 rdev->raid_disk = -1;
5380 if (mddev->degraded) {
5381 ITERATE_RDEV(mddev,rdev,rtmp)
5382 if (rdev->raid_disk < 0
5383 && !test_bit(Faulty, &rdev->flags)) {
5384 rdev->recovery_offset = 0;
5385 if (mddev->pers->hot_add_disk(mddev,rdev)) {
5387 sprintf(nm, "rd%d", rdev->raid_disk);
5388 sysfs_create_link(&mddev->kobj,
5391 md_new_event(mddev);
5399 * This routine is regularly called by all per-raid-array threads to
5400 * deal with generic issues like resync and super-block update.
5401 * Raid personalities that don't have a thread (linear/raid0) do not
5402 * need this as they never do any recovery or update the superblock.
5404 * It does not do any resync itself, but rather "forks" off other threads
5405 * to do that as needed.
5406 * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
5407 * "->recovery" and create a thread at ->sync_thread.
5408 * When the thread finishes it sets MD_RECOVERY_DONE (and might set MD_RECOVERY_ERR)
5409 * and wakeups up this thread which will reap the thread and finish up.
5410 * This thread also removes any faulty devices (with nr_pending == 0).
5412 * The overall approach is:
5413 * 1/ if the superblock needs updating, update it.
5414 * 2/ If a recovery thread is running, don't do anything else.
5415 * 3/ If recovery has finished, clean up, possibly marking spares active.
5416 * 4/ If there are any faulty devices, remove them.
5417 * 5/ If array is degraded, try to add spares devices
5418 * 6/ If array has spares or is not in-sync, start a resync thread.
5420 void md_check_recovery(mddev_t *mddev)
5423 struct list_head *rtmp;
5427 bitmap_daemon_work(mddev->bitmap);
5432 if (signal_pending(current)) {
5433 if (mddev->pers->sync_request) {
5434 printk(KERN_INFO "md: %s in immediate safe mode\n",
5436 mddev->safemode = 2;
5438 flush_signals(current);
5443 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
5444 test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
5445 (mddev->safemode == 1) ||
5446 (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending)
5447 && !mddev->in_sync && mddev->recovery_cp == MaxSector)
5451 if (mddev_trylock(mddev)) {
5454 spin_lock_irq(&mddev->write_lock);
5455 if (mddev->safemode && !atomic_read(&mddev->writes_pending) &&
5456 !mddev->in_sync && mddev->recovery_cp == MaxSector) {
5458 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
5460 if (mddev->safemode == 1)
5461 mddev->safemode = 0;
5462 spin_unlock_irq(&mddev->write_lock);
5465 md_update_sb(mddev, 0);
5468 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
5469 !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
5470 /* resync/recovery still happening */
5471 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5474 if (mddev->sync_thread) {
5475 /* resync has finished, collect result */
5476 md_unregister_thread(mddev->sync_thread);
5477 mddev->sync_thread = NULL;
5478 if (!test_bit(MD_RECOVERY_ERR, &mddev->recovery) &&
5479 !test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
5481 /* activate any spares */
5482 mddev->pers->spare_active(mddev);
5484 md_update_sb(mddev, 1);
5486 /* if array is no-longer degraded, then any saved_raid_disk
5487 * information must be scrapped
5489 if (!mddev->degraded)
5490 ITERATE_RDEV(mddev,rdev,rtmp)
5491 rdev->saved_raid_disk = -1;
5493 mddev->recovery = 0;
5494 /* flag recovery needed just to double check */
5495 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5496 md_new_event(mddev);
5499 /* Clear some bits that don't mean anything, but
5502 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5503 clear_bit(MD_RECOVERY_ERR, &mddev->recovery);
5504 clear_bit(MD_RECOVERY_INTR, &mddev->recovery);
5505 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
5507 if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
5509 /* no recovery is running.
5510 * remove any failed drives, then
5511 * add spares if possible.
5512 * Spare are also removed and re-added, to allow
5513 * the personality to fail the re-add.
5516 if (mddev->reshape_position != MaxSector) {
5517 if (mddev->pers->check_reshape(mddev) != 0)
5518 /* Cannot proceed */
5520 set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
5521 } else if ((spares = remove_and_add_spares(mddev))) {
5522 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
5523 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
5524 } else if (mddev->recovery_cp < MaxSector) {
5525 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
5526 } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
5527 /* nothing to be done ... */
5530 if (mddev->pers->sync_request) {
5531 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
5532 if (spares && mddev->bitmap && ! mddev->bitmap->file) {
5533 /* We are adding a device or devices to an array
5534 * which has the bitmap stored on all devices.
5535 * So make sure all bitmap pages get written
5537 bitmap_write_all(mddev->bitmap);
5539 mddev->sync_thread = md_register_thread(md_do_sync,
5542 if (!mddev->sync_thread) {
5543 printk(KERN_ERR "%s: could not start resync"
5546 /* leave the spares where they are, it shouldn't hurt */
5547 mddev->recovery = 0;
5549 md_wakeup_thread(mddev->sync_thread);
5550 md_new_event(mddev);
5553 mddev_unlock(mddev);
5557 static int md_notify_reboot(struct notifier_block *this,
5558 unsigned long code, void *x)
5560 struct list_head *tmp;
5563 if ((code == SYS_DOWN) || (code == SYS_HALT) || (code == SYS_POWER_OFF)) {
5565 printk(KERN_INFO "md: stopping all md devices.\n");
5567 ITERATE_MDDEV(mddev,tmp)
5568 if (mddev_trylock(mddev)) {
5569 do_md_stop (mddev, 1);
5570 mddev_unlock(mddev);
5573 * certain more exotic SCSI devices are known to be
5574 * volatile wrt too early system reboots. While the
5575 * right place to handle this issue is the given
5576 * driver, we do want to have a safe RAID driver ...
5583 static struct notifier_block md_notifier = {
5584 .notifier_call = md_notify_reboot,
5586 .priority = INT_MAX, /* before any real devices */
5589 static void md_geninit(void)
5591 struct proc_dir_entry *p;
5593 dprintk("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
5595 p = create_proc_entry("mdstat", S_IRUGO, NULL);
5597 p->proc_fops = &md_seq_fops;
5600 static int __init md_init(void)
5602 if (register_blkdev(MAJOR_NR, "md"))
5604 if ((mdp_major=register_blkdev(0, "mdp"))<=0) {
5605 unregister_blkdev(MAJOR_NR, "md");
5608 blk_register_region(MKDEV(MAJOR_NR, 0), 1UL<<MINORBITS, THIS_MODULE,
5609 md_probe, NULL, NULL);
5610 blk_register_region(MKDEV(mdp_major, 0), 1UL<<MINORBITS, THIS_MODULE,
5611 md_probe, NULL, NULL);
5613 register_reboot_notifier(&md_notifier);
5614 raid_table_header = register_sysctl_table(raid_root_table);
5624 * Searches all registered partitions for autorun RAID arrays
5627 static dev_t detected_devices[128];
5630 void md_autodetect_dev(dev_t dev)
5632 if (dev_cnt >= 0 && dev_cnt < 127)
5633 detected_devices[dev_cnt++] = dev;
5637 static void autostart_arrays(int part)
5642 printk(KERN_INFO "md: Autodetecting RAID arrays.\n");
5644 for (i = 0; i < dev_cnt; i++) {
5645 dev_t dev = detected_devices[i];
5647 rdev = md_import_device(dev,0, 0);
5651 if (test_bit(Faulty, &rdev->flags)) {
5655 list_add(&rdev->same_set, &pending_raid_disks);
5659 autorun_devices(part);
5662 #endif /* !MODULE */
5664 static __exit void md_exit(void)
5667 struct list_head *tmp;
5669 blk_unregister_region(MKDEV(MAJOR_NR,0), 1U << MINORBITS);
5670 blk_unregister_region(MKDEV(mdp_major,0), 1U << MINORBITS);
5672 unregister_blkdev(MAJOR_NR,"md");
5673 unregister_blkdev(mdp_major, "mdp");
5674 unregister_reboot_notifier(&md_notifier);
5675 unregister_sysctl_table(raid_table_header);
5676 remove_proc_entry("mdstat", NULL);
5677 ITERATE_MDDEV(mddev,tmp) {
5678 struct gendisk *disk = mddev->gendisk;
5681 export_array(mddev);
5684 mddev->gendisk = NULL;
5689 module_init(md_init)
5690 module_exit(md_exit)
5692 static int get_ro(char *buffer, struct kernel_param *kp)
5694 return sprintf(buffer, "%d", start_readonly);
5696 static int set_ro(const char *val, struct kernel_param *kp)
5699 int num = simple_strtoul(val, &e, 10);
5700 if (*val && (*e == '\0' || *e == '\n')) {
5701 start_readonly = num;
5707 module_param_call(start_ro, set_ro, get_ro, NULL, S_IRUSR|S_IWUSR);
5708 module_param(start_dirty_degraded, int, S_IRUGO|S_IWUSR);
5711 EXPORT_SYMBOL(register_md_personality);
5712 EXPORT_SYMBOL(unregister_md_personality);
5713 EXPORT_SYMBOL(md_error);
5714 EXPORT_SYMBOL(md_done_sync);
5715 EXPORT_SYMBOL(md_write_start);
5716 EXPORT_SYMBOL(md_write_end);
5717 EXPORT_SYMBOL(md_register_thread);
5718 EXPORT_SYMBOL(md_unregister_thread);
5719 EXPORT_SYMBOL(md_wakeup_thread);
5720 EXPORT_SYMBOL(md_check_recovery);
5721 MODULE_LICENSE("GPL");
5723 MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR);