2 md.c : Multiple Devices driver for Linux
3 Copyright (C) 1998, 1999, 2000 Ingo Molnar
5 completely rewritten, based on the MD driver code from Marc Zyngier
9 - RAID-1/RAID-5 extensions by Miguel de Icaza, Gadi Oxman, Ingo Molnar
10 - RAID-6 extensions by H. Peter Anvin <hpa@zytor.com>
11 - boot support for linear and striped mode by Harald Hoyer <HarryH@Royal.Net>
12 - kerneld support by Boris Tobotras <boris@xtalk.msk.su>
13 - kmod support by: Cyrus Durgin
14 - RAID0 bugfixes: Mark Anthony Lisher <markal@iname.com>
15 - Devfs support by Richard Gooch <rgooch@atnf.csiro.au>
17 - lots of fixes and improvements to the RAID1/RAID5 and generic
18 RAID code (such as request based resynchronization):
20 Neil Brown <neilb@cse.unsw.edu.au>.
22 - persistent bitmap code
23 Copyright (C) 2003-2004, Paul Clements, SteelEye Technology, Inc.
25 This program is free software; you can redistribute it and/or modify
26 it under the terms of the GNU General Public License as published by
27 the Free Software Foundation; either version 2, or (at your option)
30 You should have received a copy of the GNU General Public License
31 (for example /usr/src/linux/COPYING); if not, write to the Free
32 Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
35 #include <linux/kthread.h>
36 #include <linux/blkdev.h>
37 #include <linux/sysctl.h>
38 #include <linux/seq_file.h>
39 #include <linux/mutex.h>
40 #include <linux/buffer_head.h> /* for invalidate_bdev */
41 #include <linux/poll.h>
42 #include <linux/ctype.h>
43 #include <linux/string.h>
44 #include <linux/hdreg.h>
45 #include <linux/proc_fs.h>
46 #include <linux/random.h>
47 #include <linux/reboot.h>
48 #include <linux/file.h>
49 #include <linux/compat.h>
50 #include <linux/delay.h>
51 #include <linux/raid/md_p.h>
52 #include <linux/raid/md_u.h>
53 #include <linux/slab.h>
58 #define dprintk(x...) ((void)(DEBUG && printk(x)))
61 static void autostart_arrays(int part);
64 static LIST_HEAD(pers_list);
65 static DEFINE_SPINLOCK(pers_lock);
67 static void md_print_devices(void);
69 static DECLARE_WAIT_QUEUE_HEAD(resync_wait);
70 static struct workqueue_struct *md_wq;
71 static struct workqueue_struct *md_misc_wq;
73 #define MD_BUG(x...) { printk("md: bug in file %s, line %d\n", __FILE__, __LINE__); md_print_devices(); }
76 * Default number of read corrections we'll attempt on an rdev
77 * before ejecting it from the array. We divide the read error
78 * count by 2 for every hour elapsed between read errors.
80 #define MD_DEFAULT_MAX_CORRECTED_READ_ERRORS 20
82 * Current RAID-1,4,5 parallel reconstruction 'guaranteed speed limit'
83 * is 1000 KB/sec, so the extra system load does not show up that much.
84 * Increase it if you want to have more _guaranteed_ speed. Note that
85 * the RAID driver will use the maximum available bandwidth if the IO
86 * subsystem is idle. There is also an 'absolute maximum' reconstruction
87 * speed limit - in case reconstruction slows down your system despite
90 * you can change it via /proc/sys/dev/raid/speed_limit_min and _max.
91 * or /sys/block/mdX/md/sync_speed_{min,max}
94 static int sysctl_speed_limit_min = 1000;
95 static int sysctl_speed_limit_max = 200000;
96 static inline int speed_min(mddev_t *mddev)
98 return mddev->sync_speed_min ?
99 mddev->sync_speed_min : sysctl_speed_limit_min;
102 static inline int speed_max(mddev_t *mddev)
104 return mddev->sync_speed_max ?
105 mddev->sync_speed_max : sysctl_speed_limit_max;
108 static struct ctl_table_header *raid_table_header;
110 static ctl_table raid_table[] = {
112 .procname = "speed_limit_min",
113 .data = &sysctl_speed_limit_min,
114 .maxlen = sizeof(int),
115 .mode = S_IRUGO|S_IWUSR,
116 .proc_handler = proc_dointvec,
119 .procname = "speed_limit_max",
120 .data = &sysctl_speed_limit_max,
121 .maxlen = sizeof(int),
122 .mode = S_IRUGO|S_IWUSR,
123 .proc_handler = proc_dointvec,
128 static ctl_table raid_dir_table[] = {
132 .mode = S_IRUGO|S_IXUGO,
138 static ctl_table raid_root_table[] = {
143 .child = raid_dir_table,
148 static const struct block_device_operations md_fops;
150 static int start_readonly;
153 * like bio_clone, but with a local bio set
156 static void mddev_bio_destructor(struct bio *bio)
158 mddev_t *mddev, **mddevp;
163 bio_free(bio, mddev->bio_set);
166 struct bio *bio_alloc_mddev(gfp_t gfp_mask, int nr_iovecs,
172 if (!mddev || !mddev->bio_set)
173 return bio_alloc(gfp_mask, nr_iovecs);
175 b = bio_alloc_bioset(gfp_mask, nr_iovecs,
181 b->bi_destructor = mddev_bio_destructor;
184 EXPORT_SYMBOL_GPL(bio_alloc_mddev);
186 struct bio *bio_clone_mddev(struct bio *bio, gfp_t gfp_mask,
192 if (!mddev || !mddev->bio_set)
193 return bio_clone(bio, gfp_mask);
195 b = bio_alloc_bioset(gfp_mask, bio->bi_max_vecs,
201 b->bi_destructor = mddev_bio_destructor;
203 if (bio_integrity(bio)) {
206 ret = bio_integrity_clone(b, bio, gfp_mask, mddev->bio_set);
216 EXPORT_SYMBOL_GPL(bio_clone_mddev);
219 * We have a system wide 'event count' that is incremented
220 * on any 'interesting' event, and readers of /proc/mdstat
221 * can use 'poll' or 'select' to find out when the event
225 * start array, stop array, error, add device, remove device,
226 * start build, activate spare
228 static DECLARE_WAIT_QUEUE_HEAD(md_event_waiters);
229 static atomic_t md_event_count;
230 void md_new_event(mddev_t *mddev)
232 atomic_inc(&md_event_count);
233 wake_up(&md_event_waiters);
235 EXPORT_SYMBOL_GPL(md_new_event);
237 /* Alternate version that can be called from interrupts
238 * when calling sysfs_notify isn't needed.
240 static void md_new_event_inintr(mddev_t *mddev)
242 atomic_inc(&md_event_count);
243 wake_up(&md_event_waiters);
247 * Enables to iterate over all existing md arrays
248 * all_mddevs_lock protects this list.
250 static LIST_HEAD(all_mddevs);
251 static DEFINE_SPINLOCK(all_mddevs_lock);
255 * iterates through all used mddevs in the system.
256 * We take care to grab the all_mddevs_lock whenever navigating
257 * the list, and to always hold a refcount when unlocked.
258 * Any code which breaks out of this loop while own
259 * a reference to the current mddev and must mddev_put it.
261 #define for_each_mddev(mddev,tmp) \
263 for (({ spin_lock(&all_mddevs_lock); \
264 tmp = all_mddevs.next; \
266 ({ if (tmp != &all_mddevs) \
267 mddev_get(list_entry(tmp, mddev_t, all_mddevs));\
268 spin_unlock(&all_mddevs_lock); \
269 if (mddev) mddev_put(mddev); \
270 mddev = list_entry(tmp, mddev_t, all_mddevs); \
271 tmp != &all_mddevs;}); \
272 ({ spin_lock(&all_mddevs_lock); \
277 /* Rather than calling directly into the personality make_request function,
278 * IO requests come here first so that we can check if the device is
279 * being suspended pending a reconfiguration.
280 * We hold a refcount over the call to ->make_request. By the time that
281 * call has finished, the bio has been linked into some internal structure
282 * and so is visible to ->quiesce(), so we don't need the refcount any more.
284 static int md_make_request(struct request_queue *q, struct bio *bio)
286 const int rw = bio_data_dir(bio);
287 mddev_t *mddev = q->queuedata;
291 if (mddev == NULL || mddev->pers == NULL
296 smp_rmb(); /* Ensure implications of 'active' are visible */
298 if (mddev->suspended) {
301 prepare_to_wait(&mddev->sb_wait, &__wait,
302 TASK_UNINTERRUPTIBLE);
303 if (!mddev->suspended)
309 finish_wait(&mddev->sb_wait, &__wait);
311 atomic_inc(&mddev->active_io);
314 rv = mddev->pers->make_request(mddev, bio);
316 cpu = part_stat_lock();
317 part_stat_inc(cpu, &mddev->gendisk->part0, ios[rw]);
318 part_stat_add(cpu, &mddev->gendisk->part0, sectors[rw],
322 if (atomic_dec_and_test(&mddev->active_io) && mddev->suspended)
323 wake_up(&mddev->sb_wait);
328 /* mddev_suspend makes sure no new requests are submitted
329 * to the device, and that any requests that have been submitted
330 * are completely handled.
331 * Once ->stop is called and completes, the module will be completely
334 void mddev_suspend(mddev_t *mddev)
336 BUG_ON(mddev->suspended);
337 mddev->suspended = 1;
339 wait_event(mddev->sb_wait, atomic_read(&mddev->active_io) == 0);
340 mddev->pers->quiesce(mddev, 1);
342 EXPORT_SYMBOL_GPL(mddev_suspend);
344 void mddev_resume(mddev_t *mddev)
346 mddev->suspended = 0;
347 wake_up(&mddev->sb_wait);
348 mddev->pers->quiesce(mddev, 0);
350 EXPORT_SYMBOL_GPL(mddev_resume);
352 int mddev_congested(mddev_t *mddev, int bits)
354 return mddev->suspended;
356 EXPORT_SYMBOL(mddev_congested);
359 * Generic flush handling for md
362 static void md_end_flush(struct bio *bio, int err)
364 mdk_rdev_t *rdev = bio->bi_private;
365 mddev_t *mddev = rdev->mddev;
367 rdev_dec_pending(rdev, mddev);
369 if (atomic_dec_and_test(&mddev->flush_pending)) {
370 /* The pre-request flush has finished */
371 queue_work(md_wq, &mddev->flush_work);
376 static void md_submit_flush_data(struct work_struct *ws);
378 static void submit_flushes(struct work_struct *ws)
380 mddev_t *mddev = container_of(ws, mddev_t, flush_work);
383 INIT_WORK(&mddev->flush_work, md_submit_flush_data);
384 atomic_set(&mddev->flush_pending, 1);
386 list_for_each_entry_rcu(rdev, &mddev->disks, same_set)
387 if (rdev->raid_disk >= 0 &&
388 !test_bit(Faulty, &rdev->flags)) {
389 /* Take two references, one is dropped
390 * when request finishes, one after
391 * we reclaim rcu_read_lock
394 atomic_inc(&rdev->nr_pending);
395 atomic_inc(&rdev->nr_pending);
397 bi = bio_alloc_mddev(GFP_KERNEL, 0, mddev);
398 bi->bi_end_io = md_end_flush;
399 bi->bi_private = rdev;
400 bi->bi_bdev = rdev->bdev;
401 atomic_inc(&mddev->flush_pending);
402 submit_bio(WRITE_FLUSH, bi);
404 rdev_dec_pending(rdev, mddev);
407 if (atomic_dec_and_test(&mddev->flush_pending))
408 queue_work(md_wq, &mddev->flush_work);
411 static void md_submit_flush_data(struct work_struct *ws)
413 mddev_t *mddev = container_of(ws, mddev_t, flush_work);
414 struct bio *bio = mddev->flush_bio;
416 if (bio->bi_size == 0)
417 /* an empty barrier - all done */
420 bio->bi_rw &= ~REQ_FLUSH;
421 if (mddev->pers->make_request(mddev, bio))
422 generic_make_request(bio);
425 mddev->flush_bio = NULL;
426 wake_up(&mddev->sb_wait);
429 void md_flush_request(mddev_t *mddev, struct bio *bio)
431 spin_lock_irq(&mddev->write_lock);
432 wait_event_lock_irq(mddev->sb_wait,
434 mddev->write_lock, /*nothing*/);
435 mddev->flush_bio = bio;
436 spin_unlock_irq(&mddev->write_lock);
438 INIT_WORK(&mddev->flush_work, submit_flushes);
439 queue_work(md_wq, &mddev->flush_work);
441 EXPORT_SYMBOL(md_flush_request);
443 /* Support for plugging.
444 * This mirrors the plugging support in request_queue, but does not
445 * require having a whole queue
447 static void plugger_work(struct work_struct *work)
449 struct plug_handle *plug =
450 container_of(work, struct plug_handle, unplug_work);
451 plug->unplug_fn(plug);
453 static void plugger_timeout(unsigned long data)
455 struct plug_handle *plug = (void *)data;
456 kblockd_schedule_work(NULL, &plug->unplug_work);
458 void plugger_init(struct plug_handle *plug,
459 void (*unplug_fn)(struct plug_handle *))
461 plug->unplug_flag = 0;
462 plug->unplug_fn = unplug_fn;
463 init_timer(&plug->unplug_timer);
464 plug->unplug_timer.function = plugger_timeout;
465 plug->unplug_timer.data = (unsigned long)plug;
466 INIT_WORK(&plug->unplug_work, plugger_work);
468 EXPORT_SYMBOL_GPL(plugger_init);
470 void plugger_set_plug(struct plug_handle *plug)
472 if (!test_and_set_bit(PLUGGED_FLAG, &plug->unplug_flag))
473 mod_timer(&plug->unplug_timer, jiffies + msecs_to_jiffies(3)+1);
475 EXPORT_SYMBOL_GPL(plugger_set_plug);
477 int plugger_remove_plug(struct plug_handle *plug)
479 if (test_and_clear_bit(PLUGGED_FLAG, &plug->unplug_flag)) {
480 del_timer(&plug->unplug_timer);
485 EXPORT_SYMBOL_GPL(plugger_remove_plug);
488 static inline mddev_t *mddev_get(mddev_t *mddev)
490 atomic_inc(&mddev->active);
494 static void mddev_delayed_delete(struct work_struct *ws);
496 static void mddev_put(mddev_t *mddev)
498 struct bio_set *bs = NULL;
500 if (!atomic_dec_and_lock(&mddev->active, &all_mddevs_lock))
502 if (!mddev->raid_disks && list_empty(&mddev->disks) &&
503 mddev->ctime == 0 && !mddev->hold_active) {
504 /* Array is not configured at all, and not held active,
506 list_del(&mddev->all_mddevs);
508 mddev->bio_set = NULL;
509 if (mddev->gendisk) {
510 /* We did a probe so need to clean up. Call
511 * queue_work inside the spinlock so that
512 * flush_workqueue() after mddev_find will
513 * succeed in waiting for the work to be done.
515 INIT_WORK(&mddev->del_work, mddev_delayed_delete);
516 queue_work(md_misc_wq, &mddev->del_work);
520 spin_unlock(&all_mddevs_lock);
525 void mddev_init(mddev_t *mddev)
527 mutex_init(&mddev->open_mutex);
528 mutex_init(&mddev->reconfig_mutex);
529 mutex_init(&mddev->bitmap_info.mutex);
530 INIT_LIST_HEAD(&mddev->disks);
531 INIT_LIST_HEAD(&mddev->all_mddevs);
532 init_timer(&mddev->safemode_timer);
533 atomic_set(&mddev->active, 1);
534 atomic_set(&mddev->openers, 0);
535 atomic_set(&mddev->active_io, 0);
536 spin_lock_init(&mddev->write_lock);
537 atomic_set(&mddev->flush_pending, 0);
538 init_waitqueue_head(&mddev->sb_wait);
539 init_waitqueue_head(&mddev->recovery_wait);
540 mddev->reshape_position = MaxSector;
541 mddev->resync_min = 0;
542 mddev->resync_max = MaxSector;
543 mddev->level = LEVEL_NONE;
545 EXPORT_SYMBOL_GPL(mddev_init);
547 static mddev_t * mddev_find(dev_t unit)
549 mddev_t *mddev, *new = NULL;
552 spin_lock(&all_mddevs_lock);
555 list_for_each_entry(mddev, &all_mddevs, all_mddevs)
556 if (mddev->unit == unit) {
558 spin_unlock(&all_mddevs_lock);
564 list_add(&new->all_mddevs, &all_mddevs);
565 spin_unlock(&all_mddevs_lock);
566 new->hold_active = UNTIL_IOCTL;
570 /* find an unused unit number */
571 static int next_minor = 512;
572 int start = next_minor;
576 dev = MKDEV(MD_MAJOR, next_minor);
578 if (next_minor > MINORMASK)
580 if (next_minor == start) {
581 /* Oh dear, all in use. */
582 spin_unlock(&all_mddevs_lock);
588 list_for_each_entry(mddev, &all_mddevs, all_mddevs)
589 if (mddev->unit == dev) {
595 new->md_minor = MINOR(dev);
596 new->hold_active = UNTIL_STOP;
597 list_add(&new->all_mddevs, &all_mddevs);
598 spin_unlock(&all_mddevs_lock);
601 spin_unlock(&all_mddevs_lock);
603 new = kzalloc(sizeof(*new), GFP_KERNEL);
608 if (MAJOR(unit) == MD_MAJOR)
609 new->md_minor = MINOR(unit);
611 new->md_minor = MINOR(unit) >> MdpMinorShift;
618 static inline int mddev_lock(mddev_t * mddev)
620 return mutex_lock_interruptible(&mddev->reconfig_mutex);
623 static inline int mddev_is_locked(mddev_t *mddev)
625 return mutex_is_locked(&mddev->reconfig_mutex);
628 static inline int mddev_trylock(mddev_t * mddev)
630 return mutex_trylock(&mddev->reconfig_mutex);
633 static struct attribute_group md_redundancy_group;
635 static void mddev_unlock(mddev_t * mddev)
637 if (mddev->to_remove) {
638 /* These cannot be removed under reconfig_mutex as
639 * an access to the files will try to take reconfig_mutex
640 * while holding the file unremovable, which leads to
642 * So hold set sysfs_active while the remove in happeing,
643 * and anything else which might set ->to_remove or my
644 * otherwise change the sysfs namespace will fail with
645 * -EBUSY if sysfs_active is still set.
646 * We set sysfs_active under reconfig_mutex and elsewhere
647 * test it under the same mutex to ensure its correct value
650 struct attribute_group *to_remove = mddev->to_remove;
651 mddev->to_remove = NULL;
652 mddev->sysfs_active = 1;
653 mutex_unlock(&mddev->reconfig_mutex);
655 if (mddev->kobj.sd) {
656 if (to_remove != &md_redundancy_group)
657 sysfs_remove_group(&mddev->kobj, to_remove);
658 if (mddev->pers == NULL ||
659 mddev->pers->sync_request == NULL) {
660 sysfs_remove_group(&mddev->kobj, &md_redundancy_group);
661 if (mddev->sysfs_action)
662 sysfs_put(mddev->sysfs_action);
663 mddev->sysfs_action = NULL;
666 mddev->sysfs_active = 0;
668 mutex_unlock(&mddev->reconfig_mutex);
670 md_wakeup_thread(mddev->thread);
673 static mdk_rdev_t * find_rdev_nr(mddev_t *mddev, int nr)
677 list_for_each_entry(rdev, &mddev->disks, same_set)
678 if (rdev->desc_nr == nr)
684 static mdk_rdev_t * find_rdev(mddev_t * mddev, dev_t dev)
688 list_for_each_entry(rdev, &mddev->disks, same_set)
689 if (rdev->bdev->bd_dev == dev)
695 static struct mdk_personality *find_pers(int level, char *clevel)
697 struct mdk_personality *pers;
698 list_for_each_entry(pers, &pers_list, list) {
699 if (level != LEVEL_NONE && pers->level == level)
701 if (strcmp(pers->name, clevel)==0)
707 /* return the offset of the super block in 512byte sectors */
708 static inline sector_t calc_dev_sboffset(mdk_rdev_t *rdev)
710 sector_t num_sectors = i_size_read(rdev->bdev->bd_inode) / 512;
711 return MD_NEW_SIZE_SECTORS(num_sectors);
714 static int alloc_disk_sb(mdk_rdev_t * rdev)
719 rdev->sb_page = alloc_page(GFP_KERNEL);
720 if (!rdev->sb_page) {
721 printk(KERN_ALERT "md: out of memory.\n");
728 static void free_disk_sb(mdk_rdev_t * rdev)
731 put_page(rdev->sb_page);
733 rdev->sb_page = NULL;
740 static void super_written(struct bio *bio, int error)
742 mdk_rdev_t *rdev = bio->bi_private;
743 mddev_t *mddev = rdev->mddev;
745 if (error || !test_bit(BIO_UPTODATE, &bio->bi_flags)) {
746 printk("md: super_written gets error=%d, uptodate=%d\n",
747 error, test_bit(BIO_UPTODATE, &bio->bi_flags));
748 WARN_ON(test_bit(BIO_UPTODATE, &bio->bi_flags));
749 md_error(mddev, rdev);
752 if (atomic_dec_and_test(&mddev->pending_writes))
753 wake_up(&mddev->sb_wait);
757 void md_super_write(mddev_t *mddev, mdk_rdev_t *rdev,
758 sector_t sector, int size, struct page *page)
760 /* write first size bytes of page to sector of rdev
761 * Increment mddev->pending_writes before returning
762 * and decrement it on completion, waking up sb_wait
763 * if zero is reached.
764 * If an error occurred, call md_error
766 struct bio *bio = bio_alloc_mddev(GFP_NOIO, 1, mddev);
768 bio->bi_bdev = rdev->meta_bdev ? rdev->meta_bdev : rdev->bdev;
769 bio->bi_sector = sector;
770 bio_add_page(bio, page, size, 0);
771 bio->bi_private = rdev;
772 bio->bi_end_io = super_written;
774 atomic_inc(&mddev->pending_writes);
775 submit_bio(REQ_WRITE | REQ_SYNC | REQ_UNPLUG | REQ_FLUSH | REQ_FUA,
779 void md_super_wait(mddev_t *mddev)
781 /* wait for all superblock writes that were scheduled to complete */
784 prepare_to_wait(&mddev->sb_wait, &wq, TASK_UNINTERRUPTIBLE);
785 if (atomic_read(&mddev->pending_writes)==0)
789 finish_wait(&mddev->sb_wait, &wq);
792 static void bi_complete(struct bio *bio, int error)
794 complete((struct completion*)bio->bi_private);
797 int sync_page_io(mdk_rdev_t *rdev, sector_t sector, int size,
798 struct page *page, int rw, bool metadata_op)
800 struct bio *bio = bio_alloc_mddev(GFP_NOIO, 1, rdev->mddev);
801 struct completion event;
804 rw |= REQ_SYNC | REQ_UNPLUG;
806 bio->bi_bdev = (metadata_op && rdev->meta_bdev) ?
807 rdev->meta_bdev : rdev->bdev;
809 bio->bi_sector = sector + rdev->sb_start;
811 bio->bi_sector = sector + rdev->data_offset;
812 bio_add_page(bio, page, size, 0);
813 init_completion(&event);
814 bio->bi_private = &event;
815 bio->bi_end_io = bi_complete;
817 wait_for_completion(&event);
819 ret = test_bit(BIO_UPTODATE, &bio->bi_flags);
823 EXPORT_SYMBOL_GPL(sync_page_io);
825 static int read_disk_sb(mdk_rdev_t * rdev, int size)
827 char b[BDEVNAME_SIZE];
828 if (!rdev->sb_page) {
836 if (!sync_page_io(rdev, 0, size, rdev->sb_page, READ, true))
842 printk(KERN_WARNING "md: disabled device %s, could not read superblock.\n",
843 bdevname(rdev->bdev,b));
847 static int uuid_equal(mdp_super_t *sb1, mdp_super_t *sb2)
849 return sb1->set_uuid0 == sb2->set_uuid0 &&
850 sb1->set_uuid1 == sb2->set_uuid1 &&
851 sb1->set_uuid2 == sb2->set_uuid2 &&
852 sb1->set_uuid3 == sb2->set_uuid3;
855 static int sb_equal(mdp_super_t *sb1, mdp_super_t *sb2)
858 mdp_super_t *tmp1, *tmp2;
860 tmp1 = kmalloc(sizeof(*tmp1),GFP_KERNEL);
861 tmp2 = kmalloc(sizeof(*tmp2),GFP_KERNEL);
863 if (!tmp1 || !tmp2) {
865 printk(KERN_INFO "md.c sb_equal(): failed to allocate memory!\n");
873 * nr_disks is not constant
878 ret = (memcmp(tmp1, tmp2, MD_SB_GENERIC_CONSTANT_WORDS * 4) == 0);
886 static u32 md_csum_fold(u32 csum)
888 csum = (csum & 0xffff) + (csum >> 16);
889 return (csum & 0xffff) + (csum >> 16);
892 static unsigned int calc_sb_csum(mdp_super_t * sb)
895 u32 *sb32 = (u32*)sb;
897 unsigned int disk_csum, csum;
899 disk_csum = sb->sb_csum;
902 for (i = 0; i < MD_SB_BYTES/4 ; i++)
904 csum = (newcsum & 0xffffffff) + (newcsum>>32);
908 /* This used to use csum_partial, which was wrong for several
909 * reasons including that different results are returned on
910 * different architectures. It isn't critical that we get exactly
911 * the same return value as before (we always csum_fold before
912 * testing, and that removes any differences). However as we
913 * know that csum_partial always returned a 16bit value on
914 * alphas, do a fold to maximise conformity to previous behaviour.
916 sb->sb_csum = md_csum_fold(disk_csum);
918 sb->sb_csum = disk_csum;
925 * Handle superblock details.
926 * We want to be able to handle multiple superblock formats
927 * so we have a common interface to them all, and an array of
928 * different handlers.
929 * We rely on user-space to write the initial superblock, and support
930 * reading and updating of superblocks.
931 * Interface methods are:
932 * int load_super(mdk_rdev_t *dev, mdk_rdev_t *refdev, int minor_version)
933 * loads and validates a superblock on dev.
934 * if refdev != NULL, compare superblocks on both devices
936 * 0 - dev has a superblock that is compatible with refdev
937 * 1 - dev has a superblock that is compatible and newer than refdev
938 * so dev should be used as the refdev in future
939 * -EINVAL superblock incompatible or invalid
940 * -othererror e.g. -EIO
942 * int validate_super(mddev_t *mddev, mdk_rdev_t *dev)
943 * Verify that dev is acceptable into mddev.
944 * The first time, mddev->raid_disks will be 0, and data from
945 * dev should be merged in. Subsequent calls check that dev
946 * is new enough. Return 0 or -EINVAL
948 * void sync_super(mddev_t *mddev, mdk_rdev_t *dev)
949 * Update the superblock for rdev with data in mddev
950 * This does not write to disc.
956 struct module *owner;
957 int (*load_super)(mdk_rdev_t *rdev, mdk_rdev_t *refdev,
959 int (*validate_super)(mddev_t *mddev, mdk_rdev_t *rdev);
960 void (*sync_super)(mddev_t *mddev, mdk_rdev_t *rdev);
961 unsigned long long (*rdev_size_change)(mdk_rdev_t *rdev,
962 sector_t num_sectors);
966 * Check that the given mddev has no bitmap.
968 * This function is called from the run method of all personalities that do not
969 * support bitmaps. It prints an error message and returns non-zero if mddev
970 * has a bitmap. Otherwise, it returns 0.
973 int md_check_no_bitmap(mddev_t *mddev)
975 if (!mddev->bitmap_info.file && !mddev->bitmap_info.offset)
977 printk(KERN_ERR "%s: bitmaps are not supported for %s\n",
978 mdname(mddev), mddev->pers->name);
981 EXPORT_SYMBOL(md_check_no_bitmap);
984 * load_super for 0.90.0
986 static int super_90_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
988 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
993 * Calculate the position of the superblock (512byte sectors),
994 * it's at the end of the disk.
996 * It also happens to be a multiple of 4Kb.
998 rdev->sb_start = calc_dev_sboffset(rdev);
1000 ret = read_disk_sb(rdev, MD_SB_BYTES);
1001 if (ret) return ret;
1005 bdevname(rdev->bdev, b);
1006 sb = (mdp_super_t*)page_address(rdev->sb_page);
1008 if (sb->md_magic != MD_SB_MAGIC) {
1009 printk(KERN_ERR "md: invalid raid superblock magic on %s\n",
1014 if (sb->major_version != 0 ||
1015 sb->minor_version < 90 ||
1016 sb->minor_version > 91) {
1017 printk(KERN_WARNING "Bad version number %d.%d on %s\n",
1018 sb->major_version, sb->minor_version,
1023 if (sb->raid_disks <= 0)
1026 if (md_csum_fold(calc_sb_csum(sb)) != md_csum_fold(sb->sb_csum)) {
1027 printk(KERN_WARNING "md: invalid superblock checksum on %s\n",
1032 rdev->preferred_minor = sb->md_minor;
1033 rdev->data_offset = 0;
1034 rdev->sb_size = MD_SB_BYTES;
1036 if (sb->level == LEVEL_MULTIPATH)
1039 rdev->desc_nr = sb->this_disk.number;
1045 mdp_super_t *refsb = (mdp_super_t*)page_address(refdev->sb_page);
1046 if (!uuid_equal(refsb, sb)) {
1047 printk(KERN_WARNING "md: %s has different UUID to %s\n",
1048 b, bdevname(refdev->bdev,b2));
1051 if (!sb_equal(refsb, sb)) {
1052 printk(KERN_WARNING "md: %s has same UUID"
1053 " but different superblock to %s\n",
1054 b, bdevname(refdev->bdev, b2));
1058 ev2 = md_event(refsb);
1064 rdev->sectors = rdev->sb_start;
1066 if (rdev->sectors < sb->size * 2 && sb->level > 1)
1067 /* "this cannot possibly happen" ... */
1075 * validate_super for 0.90.0
1077 static int super_90_validate(mddev_t *mddev, mdk_rdev_t *rdev)
1080 mdp_super_t *sb = (mdp_super_t *)page_address(rdev->sb_page);
1081 __u64 ev1 = md_event(sb);
1083 rdev->raid_disk = -1;
1084 clear_bit(Faulty, &rdev->flags);
1085 clear_bit(In_sync, &rdev->flags);
1086 clear_bit(WriteMostly, &rdev->flags);
1088 if (mddev->raid_disks == 0) {
1089 mddev->major_version = 0;
1090 mddev->minor_version = sb->minor_version;
1091 mddev->patch_version = sb->patch_version;
1092 mddev->external = 0;
1093 mddev->chunk_sectors = sb->chunk_size >> 9;
1094 mddev->ctime = sb->ctime;
1095 mddev->utime = sb->utime;
1096 mddev->level = sb->level;
1097 mddev->clevel[0] = 0;
1098 mddev->layout = sb->layout;
1099 mddev->raid_disks = sb->raid_disks;
1100 mddev->dev_sectors = sb->size * 2;
1101 mddev->events = ev1;
1102 mddev->bitmap_info.offset = 0;
1103 mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9;
1105 if (mddev->minor_version >= 91) {
1106 mddev->reshape_position = sb->reshape_position;
1107 mddev->delta_disks = sb->delta_disks;
1108 mddev->new_level = sb->new_level;
1109 mddev->new_layout = sb->new_layout;
1110 mddev->new_chunk_sectors = sb->new_chunk >> 9;
1112 mddev->reshape_position = MaxSector;
1113 mddev->delta_disks = 0;
1114 mddev->new_level = mddev->level;
1115 mddev->new_layout = mddev->layout;
1116 mddev->new_chunk_sectors = mddev->chunk_sectors;
1119 if (sb->state & (1<<MD_SB_CLEAN))
1120 mddev->recovery_cp = MaxSector;
1122 if (sb->events_hi == sb->cp_events_hi &&
1123 sb->events_lo == sb->cp_events_lo) {
1124 mddev->recovery_cp = sb->recovery_cp;
1126 mddev->recovery_cp = 0;
1129 memcpy(mddev->uuid+0, &sb->set_uuid0, 4);
1130 memcpy(mddev->uuid+4, &sb->set_uuid1, 4);
1131 memcpy(mddev->uuid+8, &sb->set_uuid2, 4);
1132 memcpy(mddev->uuid+12,&sb->set_uuid3, 4);
1134 mddev->max_disks = MD_SB_DISKS;
1136 if (sb->state & (1<<MD_SB_BITMAP_PRESENT) &&
1137 mddev->bitmap_info.file == NULL)
1138 mddev->bitmap_info.offset =
1139 mddev->bitmap_info.default_offset;
1141 } else if (mddev->pers == NULL) {
1142 /* Insist on good event counter while assembling, except
1143 * for spares (which don't need an event count) */
1145 if (sb->disks[rdev->desc_nr].state & (
1146 (1<<MD_DISK_SYNC) | (1 << MD_DISK_ACTIVE)))
1147 if (ev1 < mddev->events)
1149 } else if (mddev->bitmap) {
1150 /* if adding to array with a bitmap, then we can accept an
1151 * older device ... but not too old.
1153 if (ev1 < mddev->bitmap->events_cleared)
1156 if (ev1 < mddev->events)
1157 /* just a hot-add of a new device, leave raid_disk at -1 */
1161 if (mddev->level != LEVEL_MULTIPATH) {
1162 desc = sb->disks + rdev->desc_nr;
1164 if (desc->state & (1<<MD_DISK_FAULTY))
1165 set_bit(Faulty, &rdev->flags);
1166 else if (desc->state & (1<<MD_DISK_SYNC) /* &&
1167 desc->raid_disk < mddev->raid_disks */) {
1168 set_bit(In_sync, &rdev->flags);
1169 rdev->raid_disk = desc->raid_disk;
1170 } else if (desc->state & (1<<MD_DISK_ACTIVE)) {
1171 /* active but not in sync implies recovery up to
1172 * reshape position. We don't know exactly where
1173 * that is, so set to zero for now */
1174 if (mddev->minor_version >= 91) {
1175 rdev->recovery_offset = 0;
1176 rdev->raid_disk = desc->raid_disk;
1179 if (desc->state & (1<<MD_DISK_WRITEMOSTLY))
1180 set_bit(WriteMostly, &rdev->flags);
1181 } else /* MULTIPATH are always insync */
1182 set_bit(In_sync, &rdev->flags);
1187 * sync_super for 0.90.0
1189 static void super_90_sync(mddev_t *mddev, mdk_rdev_t *rdev)
1193 int next_spare = mddev->raid_disks;
1196 /* make rdev->sb match mddev data..
1199 * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
1200 * 3/ any empty disks < next_spare become removed
1202 * disks[0] gets initialised to REMOVED because
1203 * we cannot be sure from other fields if it has
1204 * been initialised or not.
1207 int active=0, working=0,failed=0,spare=0,nr_disks=0;
1209 rdev->sb_size = MD_SB_BYTES;
1211 sb = (mdp_super_t*)page_address(rdev->sb_page);
1213 memset(sb, 0, sizeof(*sb));
1215 sb->md_magic = MD_SB_MAGIC;
1216 sb->major_version = mddev->major_version;
1217 sb->patch_version = mddev->patch_version;
1218 sb->gvalid_words = 0; /* ignored */
1219 memcpy(&sb->set_uuid0, mddev->uuid+0, 4);
1220 memcpy(&sb->set_uuid1, mddev->uuid+4, 4);
1221 memcpy(&sb->set_uuid2, mddev->uuid+8, 4);
1222 memcpy(&sb->set_uuid3, mddev->uuid+12,4);
1224 sb->ctime = mddev->ctime;
1225 sb->level = mddev->level;
1226 sb->size = mddev->dev_sectors / 2;
1227 sb->raid_disks = mddev->raid_disks;
1228 sb->md_minor = mddev->md_minor;
1229 sb->not_persistent = 0;
1230 sb->utime = mddev->utime;
1232 sb->events_hi = (mddev->events>>32);
1233 sb->events_lo = (u32)mddev->events;
1235 if (mddev->reshape_position == MaxSector)
1236 sb->minor_version = 90;
1238 sb->minor_version = 91;
1239 sb->reshape_position = mddev->reshape_position;
1240 sb->new_level = mddev->new_level;
1241 sb->delta_disks = mddev->delta_disks;
1242 sb->new_layout = mddev->new_layout;
1243 sb->new_chunk = mddev->new_chunk_sectors << 9;
1245 mddev->minor_version = sb->minor_version;
1248 sb->recovery_cp = mddev->recovery_cp;
1249 sb->cp_events_hi = (mddev->events>>32);
1250 sb->cp_events_lo = (u32)mddev->events;
1251 if (mddev->recovery_cp == MaxSector)
1252 sb->state = (1<< MD_SB_CLEAN);
1254 sb->recovery_cp = 0;
1256 sb->layout = mddev->layout;
1257 sb->chunk_size = mddev->chunk_sectors << 9;
1259 if (mddev->bitmap && mddev->bitmap_info.file == NULL)
1260 sb->state |= (1<<MD_SB_BITMAP_PRESENT);
1262 sb->disks[0].state = (1<<MD_DISK_REMOVED);
1263 list_for_each_entry(rdev2, &mddev->disks, same_set) {
1266 int is_active = test_bit(In_sync, &rdev2->flags);
1268 if (rdev2->raid_disk >= 0 &&
1269 sb->minor_version >= 91)
1270 /* we have nowhere to store the recovery_offset,
1271 * but if it is not below the reshape_position,
1272 * we can piggy-back on that.
1275 if (rdev2->raid_disk < 0 ||
1276 test_bit(Faulty, &rdev2->flags))
1279 desc_nr = rdev2->raid_disk;
1281 desc_nr = next_spare++;
1282 rdev2->desc_nr = desc_nr;
1283 d = &sb->disks[rdev2->desc_nr];
1285 d->number = rdev2->desc_nr;
1286 d->major = MAJOR(rdev2->bdev->bd_dev);
1287 d->minor = MINOR(rdev2->bdev->bd_dev);
1289 d->raid_disk = rdev2->raid_disk;
1291 d->raid_disk = rdev2->desc_nr; /* compatibility */
1292 if (test_bit(Faulty, &rdev2->flags))
1293 d->state = (1<<MD_DISK_FAULTY);
1294 else if (is_active) {
1295 d->state = (1<<MD_DISK_ACTIVE);
1296 if (test_bit(In_sync, &rdev2->flags))
1297 d->state |= (1<<MD_DISK_SYNC);
1305 if (test_bit(WriteMostly, &rdev2->flags))
1306 d->state |= (1<<MD_DISK_WRITEMOSTLY);
1308 /* now set the "removed" and "faulty" bits on any missing devices */
1309 for (i=0 ; i < mddev->raid_disks ; i++) {
1310 mdp_disk_t *d = &sb->disks[i];
1311 if (d->state == 0 && d->number == 0) {
1314 d->state = (1<<MD_DISK_REMOVED);
1315 d->state |= (1<<MD_DISK_FAULTY);
1319 sb->nr_disks = nr_disks;
1320 sb->active_disks = active;
1321 sb->working_disks = working;
1322 sb->failed_disks = failed;
1323 sb->spare_disks = spare;
1325 sb->this_disk = sb->disks[rdev->desc_nr];
1326 sb->sb_csum = calc_sb_csum(sb);
1330 * rdev_size_change for 0.90.0
1332 static unsigned long long
1333 super_90_rdev_size_change(mdk_rdev_t *rdev, sector_t num_sectors)
1335 if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
1336 return 0; /* component must fit device */
1337 if (rdev->mddev->bitmap_info.offset)
1338 return 0; /* can't move bitmap */
1339 rdev->sb_start = calc_dev_sboffset(rdev);
1340 if (!num_sectors || num_sectors > rdev->sb_start)
1341 num_sectors = rdev->sb_start;
1342 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
1344 md_super_wait(rdev->mddev);
1350 * version 1 superblock
1353 static __le32 calc_sb_1_csum(struct mdp_superblock_1 * sb)
1357 unsigned long long newcsum;
1358 int size = 256 + le32_to_cpu(sb->max_dev)*2;
1359 __le32 *isuper = (__le32*)sb;
1362 disk_csum = sb->sb_csum;
1365 for (i=0; size>=4; size -= 4 )
1366 newcsum += le32_to_cpu(*isuper++);
1369 newcsum += le16_to_cpu(*(__le16*) isuper);
1371 csum = (newcsum & 0xffffffff) + (newcsum >> 32);
1372 sb->sb_csum = disk_csum;
1373 return cpu_to_le32(csum);
1376 static int super_1_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
1378 struct mdp_superblock_1 *sb;
1381 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
1385 * Calculate the position of the superblock in 512byte sectors.
1386 * It is always aligned to a 4K boundary and
1387 * depeding on minor_version, it can be:
1388 * 0: At least 8K, but less than 12K, from end of device
1389 * 1: At start of device
1390 * 2: 4K from start of device.
1392 switch(minor_version) {
1394 sb_start = i_size_read(rdev->bdev->bd_inode) >> 9;
1396 sb_start &= ~(sector_t)(4*2-1);
1407 rdev->sb_start = sb_start;
1409 /* superblock is rarely larger than 1K, but it can be larger,
1410 * and it is safe to read 4k, so we do that
1412 ret = read_disk_sb(rdev, 4096);
1413 if (ret) return ret;
1416 sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1418 if (sb->magic != cpu_to_le32(MD_SB_MAGIC) ||
1419 sb->major_version != cpu_to_le32(1) ||
1420 le32_to_cpu(sb->max_dev) > (4096-256)/2 ||
1421 le64_to_cpu(sb->super_offset) != rdev->sb_start ||
1422 (le32_to_cpu(sb->feature_map) & ~MD_FEATURE_ALL) != 0)
1425 if (calc_sb_1_csum(sb) != sb->sb_csum) {
1426 printk("md: invalid superblock checksum on %s\n",
1427 bdevname(rdev->bdev,b));
1430 if (le64_to_cpu(sb->data_size) < 10) {
1431 printk("md: data_size too small on %s\n",
1432 bdevname(rdev->bdev,b));
1436 rdev->preferred_minor = 0xffff;
1437 rdev->data_offset = le64_to_cpu(sb->data_offset);
1438 atomic_set(&rdev->corrected_errors, le32_to_cpu(sb->cnt_corrected_read));
1440 rdev->sb_size = le32_to_cpu(sb->max_dev) * 2 + 256;
1441 bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
1442 if (rdev->sb_size & bmask)
1443 rdev->sb_size = (rdev->sb_size | bmask) + 1;
1446 && rdev->data_offset < sb_start + (rdev->sb_size/512))
1449 if (sb->level == cpu_to_le32(LEVEL_MULTIPATH))
1452 rdev->desc_nr = le32_to_cpu(sb->dev_number);
1458 struct mdp_superblock_1 *refsb =
1459 (struct mdp_superblock_1*)page_address(refdev->sb_page);
1461 if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 ||
1462 sb->level != refsb->level ||
1463 sb->layout != refsb->layout ||
1464 sb->chunksize != refsb->chunksize) {
1465 printk(KERN_WARNING "md: %s has strangely different"
1466 " superblock to %s\n",
1467 bdevname(rdev->bdev,b),
1468 bdevname(refdev->bdev,b2));
1471 ev1 = le64_to_cpu(sb->events);
1472 ev2 = le64_to_cpu(refsb->events);
1480 rdev->sectors = (i_size_read(rdev->bdev->bd_inode) >> 9) -
1481 le64_to_cpu(sb->data_offset);
1483 rdev->sectors = rdev->sb_start;
1484 if (rdev->sectors < le64_to_cpu(sb->data_size))
1486 rdev->sectors = le64_to_cpu(sb->data_size);
1487 if (le64_to_cpu(sb->size) > rdev->sectors)
1492 static int super_1_validate(mddev_t *mddev, mdk_rdev_t *rdev)
1494 struct mdp_superblock_1 *sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1495 __u64 ev1 = le64_to_cpu(sb->events);
1497 rdev->raid_disk = -1;
1498 clear_bit(Faulty, &rdev->flags);
1499 clear_bit(In_sync, &rdev->flags);
1500 clear_bit(WriteMostly, &rdev->flags);
1502 if (mddev->raid_disks == 0) {
1503 mddev->major_version = 1;
1504 mddev->patch_version = 0;
1505 mddev->external = 0;
1506 mddev->chunk_sectors = le32_to_cpu(sb->chunksize);
1507 mddev->ctime = le64_to_cpu(sb->ctime) & ((1ULL << 32)-1);
1508 mddev->utime = le64_to_cpu(sb->utime) & ((1ULL << 32)-1);
1509 mddev->level = le32_to_cpu(sb->level);
1510 mddev->clevel[0] = 0;
1511 mddev->layout = le32_to_cpu(sb->layout);
1512 mddev->raid_disks = le32_to_cpu(sb->raid_disks);
1513 mddev->dev_sectors = le64_to_cpu(sb->size);
1514 mddev->events = ev1;
1515 mddev->bitmap_info.offset = 0;
1516 mddev->bitmap_info.default_offset = 1024 >> 9;
1518 mddev->recovery_cp = le64_to_cpu(sb->resync_offset);
1519 memcpy(mddev->uuid, sb->set_uuid, 16);
1521 mddev->max_disks = (4096-256)/2;
1523 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) &&
1524 mddev->bitmap_info.file == NULL )
1525 mddev->bitmap_info.offset =
1526 (__s32)le32_to_cpu(sb->bitmap_offset);
1528 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) {
1529 mddev->reshape_position = le64_to_cpu(sb->reshape_position);
1530 mddev->delta_disks = le32_to_cpu(sb->delta_disks);
1531 mddev->new_level = le32_to_cpu(sb->new_level);
1532 mddev->new_layout = le32_to_cpu(sb->new_layout);
1533 mddev->new_chunk_sectors = le32_to_cpu(sb->new_chunk);
1535 mddev->reshape_position = MaxSector;
1536 mddev->delta_disks = 0;
1537 mddev->new_level = mddev->level;
1538 mddev->new_layout = mddev->layout;
1539 mddev->new_chunk_sectors = mddev->chunk_sectors;
1542 } else if (mddev->pers == NULL) {
1543 /* Insist of good event counter while assembling, except for
1544 * spares (which don't need an event count) */
1546 if (rdev->desc_nr >= 0 &&
1547 rdev->desc_nr < le32_to_cpu(sb->max_dev) &&
1548 le16_to_cpu(sb->dev_roles[rdev->desc_nr]) < 0xfffe)
1549 if (ev1 < mddev->events)
1551 } else if (mddev->bitmap) {
1552 /* If adding to array with a bitmap, then we can accept an
1553 * older device, but not too old.
1555 if (ev1 < mddev->bitmap->events_cleared)
1558 if (ev1 < mddev->events)
1559 /* just a hot-add of a new device, leave raid_disk at -1 */
1562 if (mddev->level != LEVEL_MULTIPATH) {
1564 if (rdev->desc_nr < 0 ||
1565 rdev->desc_nr >= le32_to_cpu(sb->max_dev)) {
1569 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
1571 case 0xffff: /* spare */
1573 case 0xfffe: /* faulty */
1574 set_bit(Faulty, &rdev->flags);
1577 if ((le32_to_cpu(sb->feature_map) &
1578 MD_FEATURE_RECOVERY_OFFSET))
1579 rdev->recovery_offset = le64_to_cpu(sb->recovery_offset);
1581 set_bit(In_sync, &rdev->flags);
1582 rdev->raid_disk = role;
1585 if (sb->devflags & WriteMostly1)
1586 set_bit(WriteMostly, &rdev->flags);
1587 } else /* MULTIPATH are always insync */
1588 set_bit(In_sync, &rdev->flags);
1593 static void super_1_sync(mddev_t *mddev, mdk_rdev_t *rdev)
1595 struct mdp_superblock_1 *sb;
1598 /* make rdev->sb match mddev and rdev data. */
1600 sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1602 sb->feature_map = 0;
1604 sb->recovery_offset = cpu_to_le64(0);
1605 memset(sb->pad1, 0, sizeof(sb->pad1));
1606 memset(sb->pad2, 0, sizeof(sb->pad2));
1607 memset(sb->pad3, 0, sizeof(sb->pad3));
1609 sb->utime = cpu_to_le64((__u64)mddev->utime);
1610 sb->events = cpu_to_le64(mddev->events);
1612 sb->resync_offset = cpu_to_le64(mddev->recovery_cp);
1614 sb->resync_offset = cpu_to_le64(0);
1616 sb->cnt_corrected_read = cpu_to_le32(atomic_read(&rdev->corrected_errors));
1618 sb->raid_disks = cpu_to_le32(mddev->raid_disks);
1619 sb->size = cpu_to_le64(mddev->dev_sectors);
1620 sb->chunksize = cpu_to_le32(mddev->chunk_sectors);
1621 sb->level = cpu_to_le32(mddev->level);
1622 sb->layout = cpu_to_le32(mddev->layout);
1624 if (mddev->bitmap && mddev->bitmap_info.file == NULL) {
1625 sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_info.offset);
1626 sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET);
1629 if (rdev->raid_disk >= 0 &&
1630 !test_bit(In_sync, &rdev->flags)) {
1632 cpu_to_le32(MD_FEATURE_RECOVERY_OFFSET);
1633 sb->recovery_offset =
1634 cpu_to_le64(rdev->recovery_offset);
1637 if (mddev->reshape_position != MaxSector) {
1638 sb->feature_map |= cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE);
1639 sb->reshape_position = cpu_to_le64(mddev->reshape_position);
1640 sb->new_layout = cpu_to_le32(mddev->new_layout);
1641 sb->delta_disks = cpu_to_le32(mddev->delta_disks);
1642 sb->new_level = cpu_to_le32(mddev->new_level);
1643 sb->new_chunk = cpu_to_le32(mddev->new_chunk_sectors);
1647 list_for_each_entry(rdev2, &mddev->disks, same_set)
1648 if (rdev2->desc_nr+1 > max_dev)
1649 max_dev = rdev2->desc_nr+1;
1651 if (max_dev > le32_to_cpu(sb->max_dev)) {
1653 sb->max_dev = cpu_to_le32(max_dev);
1654 rdev->sb_size = max_dev * 2 + 256;
1655 bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
1656 if (rdev->sb_size & bmask)
1657 rdev->sb_size = (rdev->sb_size | bmask) + 1;
1659 max_dev = le32_to_cpu(sb->max_dev);
1661 for (i=0; i<max_dev;i++)
1662 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1664 list_for_each_entry(rdev2, &mddev->disks, same_set) {
1666 if (test_bit(Faulty, &rdev2->flags))
1667 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1668 else if (test_bit(In_sync, &rdev2->flags))
1669 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1670 else if (rdev2->raid_disk >= 0)
1671 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1673 sb->dev_roles[i] = cpu_to_le16(0xffff);
1676 sb->sb_csum = calc_sb_1_csum(sb);
1679 static unsigned long long
1680 super_1_rdev_size_change(mdk_rdev_t *rdev, sector_t num_sectors)
1682 struct mdp_superblock_1 *sb;
1683 sector_t max_sectors;
1684 if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
1685 return 0; /* component must fit device */
1686 if (rdev->sb_start < rdev->data_offset) {
1687 /* minor versions 1 and 2; superblock before data */
1688 max_sectors = i_size_read(rdev->bdev->bd_inode) >> 9;
1689 max_sectors -= rdev->data_offset;
1690 if (!num_sectors || num_sectors > max_sectors)
1691 num_sectors = max_sectors;
1692 } else if (rdev->mddev->bitmap_info.offset) {
1693 /* minor version 0 with bitmap we can't move */
1696 /* minor version 0; superblock after data */
1698 sb_start = (i_size_read(rdev->bdev->bd_inode) >> 9) - 8*2;
1699 sb_start &= ~(sector_t)(4*2 - 1);
1700 max_sectors = rdev->sectors + sb_start - rdev->sb_start;
1701 if (!num_sectors || num_sectors > max_sectors)
1702 num_sectors = max_sectors;
1703 rdev->sb_start = sb_start;
1705 sb = (struct mdp_superblock_1 *) page_address(rdev->sb_page);
1706 sb->data_size = cpu_to_le64(num_sectors);
1707 sb->super_offset = rdev->sb_start;
1708 sb->sb_csum = calc_sb_1_csum(sb);
1709 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
1711 md_super_wait(rdev->mddev);
1715 static struct super_type super_types[] = {
1718 .owner = THIS_MODULE,
1719 .load_super = super_90_load,
1720 .validate_super = super_90_validate,
1721 .sync_super = super_90_sync,
1722 .rdev_size_change = super_90_rdev_size_change,
1726 .owner = THIS_MODULE,
1727 .load_super = super_1_load,
1728 .validate_super = super_1_validate,
1729 .sync_super = super_1_sync,
1730 .rdev_size_change = super_1_rdev_size_change,
1734 static int match_mddev_units(mddev_t *mddev1, mddev_t *mddev2)
1736 mdk_rdev_t *rdev, *rdev2;
1739 rdev_for_each_rcu(rdev, mddev1)
1740 rdev_for_each_rcu(rdev2, mddev2)
1741 if (rdev->bdev->bd_contains ==
1742 rdev2->bdev->bd_contains) {
1750 static LIST_HEAD(pending_raid_disks);
1753 * Try to register data integrity profile for an mddev
1755 * This is called when an array is started and after a disk has been kicked
1756 * from the array. It only succeeds if all working and active component devices
1757 * are integrity capable with matching profiles.
1759 int md_integrity_register(mddev_t *mddev)
1761 mdk_rdev_t *rdev, *reference = NULL;
1763 if (list_empty(&mddev->disks))
1764 return 0; /* nothing to do */
1765 if (blk_get_integrity(mddev->gendisk))
1766 return 0; /* already registered */
1767 list_for_each_entry(rdev, &mddev->disks, same_set) {
1768 /* skip spares and non-functional disks */
1769 if (test_bit(Faulty, &rdev->flags))
1771 if (rdev->raid_disk < 0)
1774 * If at least one rdev is not integrity capable, we can not
1775 * enable data integrity for the md device.
1777 if (!bdev_get_integrity(rdev->bdev))
1780 /* Use the first rdev as the reference */
1784 /* does this rdev's profile match the reference profile? */
1785 if (blk_integrity_compare(reference->bdev->bd_disk,
1786 rdev->bdev->bd_disk) < 0)
1790 * All component devices are integrity capable and have matching
1791 * profiles, register the common profile for the md device.
1793 if (blk_integrity_register(mddev->gendisk,
1794 bdev_get_integrity(reference->bdev)) != 0) {
1795 printk(KERN_ERR "md: failed to register integrity for %s\n",
1799 printk(KERN_NOTICE "md: data integrity on %s enabled\n",
1803 EXPORT_SYMBOL(md_integrity_register);
1805 /* Disable data integrity if non-capable/non-matching disk is being added */
1806 void md_integrity_add_rdev(mdk_rdev_t *rdev, mddev_t *mddev)
1808 struct blk_integrity *bi_rdev = bdev_get_integrity(rdev->bdev);
1809 struct blk_integrity *bi_mddev = blk_get_integrity(mddev->gendisk);
1811 if (!bi_mddev) /* nothing to do */
1813 if (rdev->raid_disk < 0) /* skip spares */
1815 if (bi_rdev && blk_integrity_compare(mddev->gendisk,
1816 rdev->bdev->bd_disk) >= 0)
1818 printk(KERN_NOTICE "disabling data integrity on %s\n", mdname(mddev));
1819 blk_integrity_unregister(mddev->gendisk);
1821 EXPORT_SYMBOL(md_integrity_add_rdev);
1823 static int bind_rdev_to_array(mdk_rdev_t * rdev, mddev_t * mddev)
1825 char b[BDEVNAME_SIZE];
1835 /* prevent duplicates */
1836 if (find_rdev(mddev, rdev->bdev->bd_dev))
1839 /* make sure rdev->sectors exceeds mddev->dev_sectors */
1840 if (rdev->sectors && (mddev->dev_sectors == 0 ||
1841 rdev->sectors < mddev->dev_sectors)) {
1843 /* Cannot change size, so fail
1844 * If mddev->level <= 0, then we don't care
1845 * about aligning sizes (e.g. linear)
1847 if (mddev->level > 0)
1850 mddev->dev_sectors = rdev->sectors;
1853 /* Verify rdev->desc_nr is unique.
1854 * If it is -1, assign a free number, else
1855 * check number is not in use
1857 if (rdev->desc_nr < 0) {
1859 if (mddev->pers) choice = mddev->raid_disks;
1860 while (find_rdev_nr(mddev, choice))
1862 rdev->desc_nr = choice;
1864 if (find_rdev_nr(mddev, rdev->desc_nr))
1867 if (mddev->max_disks && rdev->desc_nr >= mddev->max_disks) {
1868 printk(KERN_WARNING "md: %s: array is limited to %d devices\n",
1869 mdname(mddev), mddev->max_disks);
1872 bdevname(rdev->bdev,b);
1873 while ( (s=strchr(b, '/')) != NULL)
1876 rdev->mddev = mddev;
1877 printk(KERN_INFO "md: bind<%s>\n", b);
1879 if ((err = kobject_add(&rdev->kobj, &mddev->kobj, "dev-%s", b)))
1882 ko = &part_to_dev(rdev->bdev->bd_part)->kobj;
1883 if (sysfs_create_link(&rdev->kobj, ko, "block"))
1884 /* failure here is OK */;
1885 rdev->sysfs_state = sysfs_get_dirent_safe(rdev->kobj.sd, "state");
1887 list_add_rcu(&rdev->same_set, &mddev->disks);
1888 bd_claim_by_disk(rdev->bdev, rdev->bdev->bd_holder, mddev->gendisk);
1890 /* May as well allow recovery to be retried once */
1891 mddev->recovery_disabled = 0;
1896 printk(KERN_WARNING "md: failed to register dev-%s for %s\n",
1901 static void md_delayed_delete(struct work_struct *ws)
1903 mdk_rdev_t *rdev = container_of(ws, mdk_rdev_t, del_work);
1904 kobject_del(&rdev->kobj);
1905 kobject_put(&rdev->kobj);
1908 static void unbind_rdev_from_array(mdk_rdev_t * rdev)
1910 char b[BDEVNAME_SIZE];
1915 bd_release_from_disk(rdev->bdev, rdev->mddev->gendisk);
1916 list_del_rcu(&rdev->same_set);
1917 printk(KERN_INFO "md: unbind<%s>\n", bdevname(rdev->bdev,b));
1919 sysfs_remove_link(&rdev->kobj, "block");
1920 sysfs_put(rdev->sysfs_state);
1921 rdev->sysfs_state = NULL;
1922 /* We need to delay this, otherwise we can deadlock when
1923 * writing to 'remove' to "dev/state". We also need
1924 * to delay it due to rcu usage.
1927 INIT_WORK(&rdev->del_work, md_delayed_delete);
1928 kobject_get(&rdev->kobj);
1929 queue_work(md_misc_wq, &rdev->del_work);
1933 * prevent the device from being mounted, repartitioned or
1934 * otherwise reused by a RAID array (or any other kernel
1935 * subsystem), by bd_claiming the device.
1937 static int lock_rdev(mdk_rdev_t *rdev, dev_t dev, int shared)
1940 struct block_device *bdev;
1941 char b[BDEVNAME_SIZE];
1943 bdev = open_by_devnum(dev, FMODE_READ|FMODE_WRITE);
1945 printk(KERN_ERR "md: could not open %s.\n",
1946 __bdevname(dev, b));
1947 return PTR_ERR(bdev);
1949 err = bd_claim(bdev, shared ? (mdk_rdev_t *)lock_rdev : rdev);
1951 printk(KERN_ERR "md: could not bd_claim %s.\n",
1953 blkdev_put(bdev, FMODE_READ|FMODE_WRITE);
1957 set_bit(AllReserved, &rdev->flags);
1962 static void unlock_rdev(mdk_rdev_t *rdev)
1964 struct block_device *bdev = rdev->bdev;
1969 blkdev_put(bdev, FMODE_READ|FMODE_WRITE);
1972 void md_autodetect_dev(dev_t dev);
1974 static void export_rdev(mdk_rdev_t * rdev)
1976 char b[BDEVNAME_SIZE];
1977 printk(KERN_INFO "md: export_rdev(%s)\n",
1978 bdevname(rdev->bdev,b));
1983 if (test_bit(AutoDetected, &rdev->flags))
1984 md_autodetect_dev(rdev->bdev->bd_dev);
1987 kobject_put(&rdev->kobj);
1990 static void kick_rdev_from_array(mdk_rdev_t * rdev)
1992 unbind_rdev_from_array(rdev);
1996 static void export_array(mddev_t *mddev)
1998 mdk_rdev_t *rdev, *tmp;
2000 rdev_for_each(rdev, tmp, mddev) {
2005 kick_rdev_from_array(rdev);
2007 if (!list_empty(&mddev->disks))
2009 mddev->raid_disks = 0;
2010 mddev->major_version = 0;
2013 static void print_desc(mdp_disk_t *desc)
2015 printk(" DISK<N:%d,(%d,%d),R:%d,S:%d>\n", desc->number,
2016 desc->major,desc->minor,desc->raid_disk,desc->state);
2019 static void print_sb_90(mdp_super_t *sb)
2024 "md: SB: (V:%d.%d.%d) ID:<%08x.%08x.%08x.%08x> CT:%08x\n",
2025 sb->major_version, sb->minor_version, sb->patch_version,
2026 sb->set_uuid0, sb->set_uuid1, sb->set_uuid2, sb->set_uuid3,
2028 printk(KERN_INFO "md: L%d S%08d ND:%d RD:%d md%d LO:%d CS:%d\n",
2029 sb->level, sb->size, sb->nr_disks, sb->raid_disks,
2030 sb->md_minor, sb->layout, sb->chunk_size);
2031 printk(KERN_INFO "md: UT:%08x ST:%d AD:%d WD:%d"
2032 " FD:%d SD:%d CSUM:%08x E:%08lx\n",
2033 sb->utime, sb->state, sb->active_disks, sb->working_disks,
2034 sb->failed_disks, sb->spare_disks,
2035 sb->sb_csum, (unsigned long)sb->events_lo);
2038 for (i = 0; i < MD_SB_DISKS; i++) {
2041 desc = sb->disks + i;
2042 if (desc->number || desc->major || desc->minor ||
2043 desc->raid_disk || (desc->state && (desc->state != 4))) {
2044 printk(" D %2d: ", i);
2048 printk(KERN_INFO "md: THIS: ");
2049 print_desc(&sb->this_disk);
2052 static void print_sb_1(struct mdp_superblock_1 *sb)
2056 uuid = sb->set_uuid;
2058 "md: SB: (V:%u) (F:0x%08x) Array-ID:<%pU>\n"
2059 "md: Name: \"%s\" CT:%llu\n",
2060 le32_to_cpu(sb->major_version),
2061 le32_to_cpu(sb->feature_map),
2064 (unsigned long long)le64_to_cpu(sb->ctime)
2065 & MD_SUPERBLOCK_1_TIME_SEC_MASK);
2067 uuid = sb->device_uuid;
2069 "md: L%u SZ%llu RD:%u LO:%u CS:%u DO:%llu DS:%llu SO:%llu"
2071 "md: Dev:%08x UUID: %pU\n"
2072 "md: (F:0x%08x) UT:%llu Events:%llu ResyncOffset:%llu CSUM:0x%08x\n"
2073 "md: (MaxDev:%u) \n",
2074 le32_to_cpu(sb->level),
2075 (unsigned long long)le64_to_cpu(sb->size),
2076 le32_to_cpu(sb->raid_disks),
2077 le32_to_cpu(sb->layout),
2078 le32_to_cpu(sb->chunksize),
2079 (unsigned long long)le64_to_cpu(sb->data_offset),
2080 (unsigned long long)le64_to_cpu(sb->data_size),
2081 (unsigned long long)le64_to_cpu(sb->super_offset),
2082 (unsigned long long)le64_to_cpu(sb->recovery_offset),
2083 le32_to_cpu(sb->dev_number),
2086 (unsigned long long)le64_to_cpu(sb->utime) & MD_SUPERBLOCK_1_TIME_SEC_MASK,
2087 (unsigned long long)le64_to_cpu(sb->events),
2088 (unsigned long long)le64_to_cpu(sb->resync_offset),
2089 le32_to_cpu(sb->sb_csum),
2090 le32_to_cpu(sb->max_dev)
2094 static void print_rdev(mdk_rdev_t *rdev, int major_version)
2096 char b[BDEVNAME_SIZE];
2097 printk(KERN_INFO "md: rdev %s, Sect:%08llu F:%d S:%d DN:%u\n",
2098 bdevname(rdev->bdev, b), (unsigned long long)rdev->sectors,
2099 test_bit(Faulty, &rdev->flags), test_bit(In_sync, &rdev->flags),
2101 if (rdev->sb_loaded) {
2102 printk(KERN_INFO "md: rdev superblock (MJ:%d):\n", major_version);
2103 switch (major_version) {
2105 print_sb_90((mdp_super_t*)page_address(rdev->sb_page));
2108 print_sb_1((struct mdp_superblock_1 *)page_address(rdev->sb_page));
2112 printk(KERN_INFO "md: no rdev superblock!\n");
2115 static void md_print_devices(void)
2117 struct list_head *tmp;
2120 char b[BDEVNAME_SIZE];
2123 printk("md: **********************************\n");
2124 printk("md: * <COMPLETE RAID STATE PRINTOUT> *\n");
2125 printk("md: **********************************\n");
2126 for_each_mddev(mddev, tmp) {
2129 bitmap_print_sb(mddev->bitmap);
2131 printk("%s: ", mdname(mddev));
2132 list_for_each_entry(rdev, &mddev->disks, same_set)
2133 printk("<%s>", bdevname(rdev->bdev,b));
2136 list_for_each_entry(rdev, &mddev->disks, same_set)
2137 print_rdev(rdev, mddev->major_version);
2139 printk("md: **********************************\n");
2144 static void sync_sbs(mddev_t * mddev, int nospares)
2146 /* Update each superblock (in-memory image), but
2147 * if we are allowed to, skip spares which already
2148 * have the right event counter, or have one earlier
2149 * (which would mean they aren't being marked as dirty
2150 * with the rest of the array)
2153 list_for_each_entry(rdev, &mddev->disks, same_set) {
2154 if (rdev->sb_events == mddev->events ||
2156 rdev->raid_disk < 0 &&
2157 rdev->sb_events+1 == mddev->events)) {
2158 /* Don't update this superblock */
2159 rdev->sb_loaded = 2;
2161 super_types[mddev->major_version].
2162 sync_super(mddev, rdev);
2163 rdev->sb_loaded = 1;
2168 static void md_update_sb(mddev_t * mddev, int force_change)
2175 /* First make sure individual recovery_offsets are correct */
2176 list_for_each_entry(rdev, &mddev->disks, same_set) {
2177 if (rdev->raid_disk >= 0 &&
2178 mddev->delta_disks >= 0 &&
2179 !test_bit(In_sync, &rdev->flags) &&
2180 mddev->curr_resync_completed > rdev->recovery_offset)
2181 rdev->recovery_offset = mddev->curr_resync_completed;
2184 if (!mddev->persistent) {
2185 clear_bit(MD_CHANGE_CLEAN, &mddev->flags);
2186 clear_bit(MD_CHANGE_DEVS, &mddev->flags);
2187 if (!mddev->external)
2188 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
2189 wake_up(&mddev->sb_wait);
2193 spin_lock_irq(&mddev->write_lock);
2195 mddev->utime = get_seconds();
2197 if (test_and_clear_bit(MD_CHANGE_DEVS, &mddev->flags))
2199 if (test_and_clear_bit(MD_CHANGE_CLEAN, &mddev->flags))
2200 /* just a clean<-> dirty transition, possibly leave spares alone,
2201 * though if events isn't the right even/odd, we will have to do
2207 if (mddev->degraded)
2208 /* If the array is degraded, then skipping spares is both
2209 * dangerous and fairly pointless.
2210 * Dangerous because a device that was removed from the array
2211 * might have a event_count that still looks up-to-date,
2212 * so it can be re-added without a resync.
2213 * Pointless because if there are any spares to skip,
2214 * then a recovery will happen and soon that array won't
2215 * be degraded any more and the spare can go back to sleep then.
2219 sync_req = mddev->in_sync;
2221 /* If this is just a dirty<->clean transition, and the array is clean
2222 * and 'events' is odd, we can roll back to the previous clean state */
2224 && (mddev->in_sync && mddev->recovery_cp == MaxSector)
2225 && mddev->can_decrease_events
2226 && mddev->events != 1) {
2228 mddev->can_decrease_events = 0;
2230 /* otherwise we have to go forward and ... */
2232 mddev->can_decrease_events = nospares;
2235 if (!mddev->events) {
2237 * oops, this 64-bit counter should never wrap.
2238 * Either we are in around ~1 trillion A.C., assuming
2239 * 1 reboot per second, or we have a bug:
2244 sync_sbs(mddev, nospares);
2245 spin_unlock_irq(&mddev->write_lock);
2248 "md: updating %s RAID superblock on device (in sync %d)\n",
2249 mdname(mddev),mddev->in_sync);
2251 bitmap_update_sb(mddev->bitmap);
2252 list_for_each_entry(rdev, &mddev->disks, same_set) {
2253 char b[BDEVNAME_SIZE];
2254 dprintk(KERN_INFO "md: ");
2255 if (rdev->sb_loaded != 1)
2256 continue; /* no noise on spare devices */
2257 if (test_bit(Faulty, &rdev->flags))
2258 dprintk("(skipping faulty ");
2260 dprintk("%s ", bdevname(rdev->bdev,b));
2261 if (!test_bit(Faulty, &rdev->flags)) {
2262 md_super_write(mddev,rdev,
2263 rdev->sb_start, rdev->sb_size,
2265 dprintk(KERN_INFO "(write) %s's sb offset: %llu\n",
2266 bdevname(rdev->bdev,b),
2267 (unsigned long long)rdev->sb_start);
2268 rdev->sb_events = mddev->events;
2272 if (mddev->level == LEVEL_MULTIPATH)
2273 /* only need to write one superblock... */
2276 md_super_wait(mddev);
2277 /* if there was a failure, MD_CHANGE_DEVS was set, and we re-write super */
2279 spin_lock_irq(&mddev->write_lock);
2280 if (mddev->in_sync != sync_req ||
2281 test_bit(MD_CHANGE_DEVS, &mddev->flags)) {
2282 /* have to write it out again */
2283 spin_unlock_irq(&mddev->write_lock);
2286 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
2287 spin_unlock_irq(&mddev->write_lock);
2288 wake_up(&mddev->sb_wait);
2289 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
2290 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
2294 /* words written to sysfs files may, or may not, be \n terminated.
2295 * We want to accept with case. For this we use cmd_match.
2297 static int cmd_match(const char *cmd, const char *str)
2299 /* See if cmd, written into a sysfs file, matches
2300 * str. They must either be the same, or cmd can
2301 * have a trailing newline
2303 while (*cmd && *str && *cmd == *str) {
2314 struct rdev_sysfs_entry {
2315 struct attribute attr;
2316 ssize_t (*show)(mdk_rdev_t *, char *);
2317 ssize_t (*store)(mdk_rdev_t *, const char *, size_t);
2321 state_show(mdk_rdev_t *rdev, char *page)
2326 if (test_bit(Faulty, &rdev->flags)) {
2327 len+= sprintf(page+len, "%sfaulty",sep);
2330 if (test_bit(In_sync, &rdev->flags)) {
2331 len += sprintf(page+len, "%sin_sync",sep);
2334 if (test_bit(WriteMostly, &rdev->flags)) {
2335 len += sprintf(page+len, "%swrite_mostly",sep);
2338 if (test_bit(Blocked, &rdev->flags)) {
2339 len += sprintf(page+len, "%sblocked", sep);
2342 if (!test_bit(Faulty, &rdev->flags) &&
2343 !test_bit(In_sync, &rdev->flags)) {
2344 len += sprintf(page+len, "%sspare", sep);
2347 return len+sprintf(page+len, "\n");
2351 state_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2354 * faulty - simulates and error
2355 * remove - disconnects the device
2356 * writemostly - sets write_mostly
2357 * -writemostly - clears write_mostly
2358 * blocked - sets the Blocked flag
2359 * -blocked - clears the Blocked flag
2360 * insync - sets Insync providing device isn't active
2363 if (cmd_match(buf, "faulty") && rdev->mddev->pers) {
2364 md_error(rdev->mddev, rdev);
2366 } else if (cmd_match(buf, "remove")) {
2367 if (rdev->raid_disk >= 0)
2370 mddev_t *mddev = rdev->mddev;
2371 kick_rdev_from_array(rdev);
2373 md_update_sb(mddev, 1);
2374 md_new_event(mddev);
2377 } else if (cmd_match(buf, "writemostly")) {
2378 set_bit(WriteMostly, &rdev->flags);
2380 } else if (cmd_match(buf, "-writemostly")) {
2381 clear_bit(WriteMostly, &rdev->flags);
2383 } else if (cmd_match(buf, "blocked")) {
2384 set_bit(Blocked, &rdev->flags);
2386 } else if (cmd_match(buf, "-blocked")) {
2387 clear_bit(Blocked, &rdev->flags);
2388 wake_up(&rdev->blocked_wait);
2389 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2390 md_wakeup_thread(rdev->mddev->thread);
2393 } else if (cmd_match(buf, "insync") && rdev->raid_disk == -1) {
2394 set_bit(In_sync, &rdev->flags);
2398 sysfs_notify_dirent_safe(rdev->sysfs_state);
2399 return err ? err : len;
2401 static struct rdev_sysfs_entry rdev_state =
2402 __ATTR(state, S_IRUGO|S_IWUSR, state_show, state_store);
2405 errors_show(mdk_rdev_t *rdev, char *page)
2407 return sprintf(page, "%d\n", atomic_read(&rdev->corrected_errors));
2411 errors_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2414 unsigned long n = simple_strtoul(buf, &e, 10);
2415 if (*buf && (*e == 0 || *e == '\n')) {
2416 atomic_set(&rdev->corrected_errors, n);
2421 static struct rdev_sysfs_entry rdev_errors =
2422 __ATTR(errors, S_IRUGO|S_IWUSR, errors_show, errors_store);
2425 slot_show(mdk_rdev_t *rdev, char *page)
2427 if (rdev->raid_disk < 0)
2428 return sprintf(page, "none\n");
2430 return sprintf(page, "%d\n", rdev->raid_disk);
2434 slot_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2439 int slot = simple_strtoul(buf, &e, 10);
2440 if (strncmp(buf, "none", 4)==0)
2442 else if (e==buf || (*e && *e!= '\n'))
2444 if (rdev->mddev->pers && slot == -1) {
2445 /* Setting 'slot' on an active array requires also
2446 * updating the 'rd%d' link, and communicating
2447 * with the personality with ->hot_*_disk.
2448 * For now we only support removing
2449 * failed/spare devices. This normally happens automatically,
2450 * but not when the metadata is externally managed.
2452 if (rdev->raid_disk == -1)
2454 /* personality does all needed checks */
2455 if (rdev->mddev->pers->hot_add_disk == NULL)
2457 err = rdev->mddev->pers->
2458 hot_remove_disk(rdev->mddev, rdev->raid_disk);
2461 sprintf(nm, "rd%d", rdev->raid_disk);
2462 sysfs_remove_link(&rdev->mddev->kobj, nm);
2463 rdev->raid_disk = -1;
2464 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2465 md_wakeup_thread(rdev->mddev->thread);
2466 } else if (rdev->mddev->pers) {
2468 /* Activating a spare .. or possibly reactivating
2469 * if we ever get bitmaps working here.
2472 if (rdev->raid_disk != -1)
2475 if (rdev->mddev->pers->hot_add_disk == NULL)
2478 list_for_each_entry(rdev2, &rdev->mddev->disks, same_set)
2479 if (rdev2->raid_disk == slot)
2482 rdev->raid_disk = slot;
2483 if (test_bit(In_sync, &rdev->flags))
2484 rdev->saved_raid_disk = slot;
2486 rdev->saved_raid_disk = -1;
2487 err = rdev->mddev->pers->
2488 hot_add_disk(rdev->mddev, rdev);
2490 rdev->raid_disk = -1;
2493 sysfs_notify_dirent_safe(rdev->sysfs_state);
2494 sprintf(nm, "rd%d", rdev->raid_disk);
2495 if (sysfs_create_link(&rdev->mddev->kobj, &rdev->kobj, nm))
2496 /* failure here is OK */;
2497 /* don't wakeup anyone, leave that to userspace. */
2499 if (slot >= rdev->mddev->raid_disks)
2501 rdev->raid_disk = slot;
2502 /* assume it is working */
2503 clear_bit(Faulty, &rdev->flags);
2504 clear_bit(WriteMostly, &rdev->flags);
2505 set_bit(In_sync, &rdev->flags);
2506 sysfs_notify_dirent_safe(rdev->sysfs_state);
2512 static struct rdev_sysfs_entry rdev_slot =
2513 __ATTR(slot, S_IRUGO|S_IWUSR, slot_show, slot_store);
2516 offset_show(mdk_rdev_t *rdev, char *page)
2518 return sprintf(page, "%llu\n", (unsigned long long)rdev->data_offset);
2522 offset_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2525 unsigned long long offset = simple_strtoull(buf, &e, 10);
2526 if (e==buf || (*e && *e != '\n'))
2528 if (rdev->mddev->pers && rdev->raid_disk >= 0)
2530 if (rdev->sectors && rdev->mddev->external)
2531 /* Must set offset before size, so overlap checks
2534 rdev->data_offset = offset;
2538 static struct rdev_sysfs_entry rdev_offset =
2539 __ATTR(offset, S_IRUGO|S_IWUSR, offset_show, offset_store);
2542 rdev_size_show(mdk_rdev_t *rdev, char *page)
2544 return sprintf(page, "%llu\n", (unsigned long long)rdev->sectors / 2);
2547 static int overlaps(sector_t s1, sector_t l1, sector_t s2, sector_t l2)
2549 /* check if two start/length pairs overlap */
2557 static int strict_blocks_to_sectors(const char *buf, sector_t *sectors)
2559 unsigned long long blocks;
2562 if (strict_strtoull(buf, 10, &blocks) < 0)
2565 if (blocks & 1ULL << (8 * sizeof(blocks) - 1))
2566 return -EINVAL; /* sector conversion overflow */
2569 if (new != blocks * 2)
2570 return -EINVAL; /* unsigned long long to sector_t overflow */
2577 rdev_size_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2579 mddev_t *my_mddev = rdev->mddev;
2580 sector_t oldsectors = rdev->sectors;
2583 if (strict_blocks_to_sectors(buf, §ors) < 0)
2585 if (my_mddev->pers && rdev->raid_disk >= 0) {
2586 if (my_mddev->persistent) {
2587 sectors = super_types[my_mddev->major_version].
2588 rdev_size_change(rdev, sectors);
2591 } else if (!sectors)
2592 sectors = (i_size_read(rdev->bdev->bd_inode) >> 9) -
2595 if (sectors < my_mddev->dev_sectors)
2596 return -EINVAL; /* component must fit device */
2598 rdev->sectors = sectors;
2599 if (sectors > oldsectors && my_mddev->external) {
2600 /* need to check that all other rdevs with the same ->bdev
2601 * do not overlap. We need to unlock the mddev to avoid
2602 * a deadlock. We have already changed rdev->sectors, and if
2603 * we have to change it back, we will have the lock again.
2607 struct list_head *tmp;
2609 mddev_unlock(my_mddev);
2610 for_each_mddev(mddev, tmp) {
2614 list_for_each_entry(rdev2, &mddev->disks, same_set)
2615 if (test_bit(AllReserved, &rdev2->flags) ||
2616 (rdev->bdev == rdev2->bdev &&
2618 overlaps(rdev->data_offset, rdev->sectors,
2624 mddev_unlock(mddev);
2630 mddev_lock(my_mddev);
2632 /* Someone else could have slipped in a size
2633 * change here, but doing so is just silly.
2634 * We put oldsectors back because we *know* it is
2635 * safe, and trust userspace not to race with
2638 rdev->sectors = oldsectors;
2645 static struct rdev_sysfs_entry rdev_size =
2646 __ATTR(size, S_IRUGO|S_IWUSR, rdev_size_show, rdev_size_store);
2649 static ssize_t recovery_start_show(mdk_rdev_t *rdev, char *page)
2651 unsigned long long recovery_start = rdev->recovery_offset;
2653 if (test_bit(In_sync, &rdev->flags) ||
2654 recovery_start == MaxSector)
2655 return sprintf(page, "none\n");
2657 return sprintf(page, "%llu\n", recovery_start);
2660 static ssize_t recovery_start_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2662 unsigned long long recovery_start;
2664 if (cmd_match(buf, "none"))
2665 recovery_start = MaxSector;
2666 else if (strict_strtoull(buf, 10, &recovery_start))
2669 if (rdev->mddev->pers &&
2670 rdev->raid_disk >= 0)
2673 rdev->recovery_offset = recovery_start;
2674 if (recovery_start == MaxSector)
2675 set_bit(In_sync, &rdev->flags);
2677 clear_bit(In_sync, &rdev->flags);
2681 static struct rdev_sysfs_entry rdev_recovery_start =
2682 __ATTR(recovery_start, S_IRUGO|S_IWUSR, recovery_start_show, recovery_start_store);
2684 static struct attribute *rdev_default_attrs[] = {
2690 &rdev_recovery_start.attr,
2694 rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
2696 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
2697 mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
2698 mddev_t *mddev = rdev->mddev;
2704 rv = mddev ? mddev_lock(mddev) : -EBUSY;
2706 if (rdev->mddev == NULL)
2709 rv = entry->show(rdev, page);
2710 mddev_unlock(mddev);
2716 rdev_attr_store(struct kobject *kobj, struct attribute *attr,
2717 const char *page, size_t length)
2719 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
2720 mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
2722 mddev_t *mddev = rdev->mddev;
2726 if (!capable(CAP_SYS_ADMIN))
2728 rv = mddev ? mddev_lock(mddev): -EBUSY;
2730 if (rdev->mddev == NULL)
2733 rv = entry->store(rdev, page, length);
2734 mddev_unlock(mddev);
2739 static void rdev_free(struct kobject *ko)
2741 mdk_rdev_t *rdev = container_of(ko, mdk_rdev_t, kobj);
2744 static const struct sysfs_ops rdev_sysfs_ops = {
2745 .show = rdev_attr_show,
2746 .store = rdev_attr_store,
2748 static struct kobj_type rdev_ktype = {
2749 .release = rdev_free,
2750 .sysfs_ops = &rdev_sysfs_ops,
2751 .default_attrs = rdev_default_attrs,
2754 void md_rdev_init(mdk_rdev_t *rdev)
2757 rdev->saved_raid_disk = -1;
2758 rdev->raid_disk = -1;
2760 rdev->data_offset = 0;
2761 rdev->sb_events = 0;
2762 rdev->last_read_error.tv_sec = 0;
2763 rdev->last_read_error.tv_nsec = 0;
2764 atomic_set(&rdev->nr_pending, 0);
2765 atomic_set(&rdev->read_errors, 0);
2766 atomic_set(&rdev->corrected_errors, 0);
2768 INIT_LIST_HEAD(&rdev->same_set);
2769 init_waitqueue_head(&rdev->blocked_wait);
2771 EXPORT_SYMBOL_GPL(md_rdev_init);
2773 * Import a device. If 'super_format' >= 0, then sanity check the superblock
2775 * mark the device faulty if:
2777 * - the device is nonexistent (zero size)
2778 * - the device has no valid superblock
2780 * a faulty rdev _never_ has rdev->sb set.
2782 static mdk_rdev_t *md_import_device(dev_t newdev, int super_format, int super_minor)
2784 char b[BDEVNAME_SIZE];
2789 rdev = kzalloc(sizeof(*rdev), GFP_KERNEL);
2791 printk(KERN_ERR "md: could not alloc mem for new device!\n");
2792 return ERR_PTR(-ENOMEM);
2796 if ((err = alloc_disk_sb(rdev)))
2799 err = lock_rdev(rdev, newdev, super_format == -2);
2803 kobject_init(&rdev->kobj, &rdev_ktype);
2805 size = i_size_read(rdev->bdev->bd_inode) >> BLOCK_SIZE_BITS;
2808 "md: %s has zero or unknown size, marking faulty!\n",
2809 bdevname(rdev->bdev,b));
2814 if (super_format >= 0) {
2815 err = super_types[super_format].
2816 load_super(rdev, NULL, super_minor);
2817 if (err == -EINVAL) {
2819 "md: %s does not have a valid v%d.%d "
2820 "superblock, not importing!\n",
2821 bdevname(rdev->bdev,b),
2822 super_format, super_minor);
2827 "md: could not read %s's sb, not importing!\n",
2828 bdevname(rdev->bdev,b));
2836 if (rdev->sb_page) {
2842 return ERR_PTR(err);
2846 * Check a full RAID array for plausibility
2850 static void analyze_sbs(mddev_t * mddev)
2853 mdk_rdev_t *rdev, *freshest, *tmp;
2854 char b[BDEVNAME_SIZE];
2857 rdev_for_each(rdev, tmp, mddev)
2858 switch (super_types[mddev->major_version].
2859 load_super(rdev, freshest, mddev->minor_version)) {
2867 "md: fatal superblock inconsistency in %s"
2868 " -- removing from array\n",
2869 bdevname(rdev->bdev,b));
2870 kick_rdev_from_array(rdev);
2874 super_types[mddev->major_version].
2875 validate_super(mddev, freshest);
2878 rdev_for_each(rdev, tmp, mddev) {
2879 if (mddev->max_disks &&
2880 (rdev->desc_nr >= mddev->max_disks ||
2881 i > mddev->max_disks)) {
2883 "md: %s: %s: only %d devices permitted\n",
2884 mdname(mddev), bdevname(rdev->bdev, b),
2886 kick_rdev_from_array(rdev);
2889 if (rdev != freshest)
2890 if (super_types[mddev->major_version].
2891 validate_super(mddev, rdev)) {
2892 printk(KERN_WARNING "md: kicking non-fresh %s"
2894 bdevname(rdev->bdev,b));
2895 kick_rdev_from_array(rdev);
2898 if (mddev->level == LEVEL_MULTIPATH) {
2899 rdev->desc_nr = i++;
2900 rdev->raid_disk = rdev->desc_nr;
2901 set_bit(In_sync, &rdev->flags);
2902 } else if (rdev->raid_disk >= (mddev->raid_disks - min(0, mddev->delta_disks))) {
2903 rdev->raid_disk = -1;
2904 clear_bit(In_sync, &rdev->flags);
2909 /* Read a fixed-point number.
2910 * Numbers in sysfs attributes should be in "standard" units where
2911 * possible, so time should be in seconds.
2912 * However we internally use a a much smaller unit such as
2913 * milliseconds or jiffies.
2914 * This function takes a decimal number with a possible fractional
2915 * component, and produces an integer which is the result of
2916 * multiplying that number by 10^'scale'.
2917 * all without any floating-point arithmetic.
2919 int strict_strtoul_scaled(const char *cp, unsigned long *res, int scale)
2921 unsigned long result = 0;
2923 while (isdigit(*cp) || (*cp == '.' && decimals < 0)) {
2926 else if (decimals < scale) {
2929 result = result * 10 + value;
2941 while (decimals < scale) {
2950 static void md_safemode_timeout(unsigned long data);
2953 safe_delay_show(mddev_t *mddev, char *page)
2955 int msec = (mddev->safemode_delay*1000)/HZ;
2956 return sprintf(page, "%d.%03d\n", msec/1000, msec%1000);
2959 safe_delay_store(mddev_t *mddev, const char *cbuf, size_t len)
2963 if (strict_strtoul_scaled(cbuf, &msec, 3) < 0)
2966 mddev->safemode_delay = 0;
2968 unsigned long old_delay = mddev->safemode_delay;
2969 mddev->safemode_delay = (msec*HZ)/1000;
2970 if (mddev->safemode_delay == 0)
2971 mddev->safemode_delay = 1;
2972 if (mddev->safemode_delay < old_delay)
2973 md_safemode_timeout((unsigned long)mddev);
2977 static struct md_sysfs_entry md_safe_delay =
2978 __ATTR(safe_mode_delay, S_IRUGO|S_IWUSR,safe_delay_show, safe_delay_store);
2981 level_show(mddev_t *mddev, char *page)
2983 struct mdk_personality *p = mddev->pers;
2985 return sprintf(page, "%s\n", p->name);
2986 else if (mddev->clevel[0])
2987 return sprintf(page, "%s\n", mddev->clevel);
2988 else if (mddev->level != LEVEL_NONE)
2989 return sprintf(page, "%d\n", mddev->level);
2995 level_store(mddev_t *mddev, const char *buf, size_t len)
2999 struct mdk_personality *pers;
3004 if (mddev->pers == NULL) {
3007 if (len >= sizeof(mddev->clevel))
3009 strncpy(mddev->clevel, buf, len);
3010 if (mddev->clevel[len-1] == '\n')
3012 mddev->clevel[len] = 0;
3013 mddev->level = LEVEL_NONE;
3017 /* request to change the personality. Need to ensure:
3018 * - array is not engaged in resync/recovery/reshape
3019 * - old personality can be suspended
3020 * - new personality will access other array.
3023 if (mddev->sync_thread ||
3024 mddev->reshape_position != MaxSector ||
3025 mddev->sysfs_active)
3028 if (!mddev->pers->quiesce) {
3029 printk(KERN_WARNING "md: %s: %s does not support online personality change\n",
3030 mdname(mddev), mddev->pers->name);
3034 /* Now find the new personality */
3035 if (len == 0 || len >= sizeof(clevel))
3037 strncpy(clevel, buf, len);
3038 if (clevel[len-1] == '\n')
3041 if (strict_strtol(clevel, 10, &level))
3044 if (request_module("md-%s", clevel) != 0)
3045 request_module("md-level-%s", clevel);
3046 spin_lock(&pers_lock);
3047 pers = find_pers(level, clevel);
3048 if (!pers || !try_module_get(pers->owner)) {
3049 spin_unlock(&pers_lock);
3050 printk(KERN_WARNING "md: personality %s not loaded\n", clevel);
3053 spin_unlock(&pers_lock);
3055 if (pers == mddev->pers) {
3056 /* Nothing to do! */
3057 module_put(pers->owner);
3060 if (!pers->takeover) {
3061 module_put(pers->owner);
3062 printk(KERN_WARNING "md: %s: %s does not support personality takeover\n",
3063 mdname(mddev), clevel);
3067 list_for_each_entry(rdev, &mddev->disks, same_set)
3068 rdev->new_raid_disk = rdev->raid_disk;
3070 /* ->takeover must set new_* and/or delta_disks
3071 * if it succeeds, and may set them when it fails.
3073 priv = pers->takeover(mddev);
3075 mddev->new_level = mddev->level;
3076 mddev->new_layout = mddev->layout;
3077 mddev->new_chunk_sectors = mddev->chunk_sectors;
3078 mddev->raid_disks -= mddev->delta_disks;
3079 mddev->delta_disks = 0;
3080 module_put(pers->owner);
3081 printk(KERN_WARNING "md: %s: %s would not accept array\n",
3082 mdname(mddev), clevel);
3083 return PTR_ERR(priv);
3086 /* Looks like we have a winner */
3087 mddev_suspend(mddev);
3088 mddev->pers->stop(mddev);
3090 if (mddev->pers->sync_request == NULL &&
3091 pers->sync_request != NULL) {
3092 /* need to add the md_redundancy_group */
3093 if (sysfs_create_group(&mddev->kobj, &md_redundancy_group))
3095 "md: cannot register extra attributes for %s\n",
3097 mddev->sysfs_action = sysfs_get_dirent(mddev->kobj.sd, NULL, "sync_action");
3099 if (mddev->pers->sync_request != NULL &&
3100 pers->sync_request == NULL) {
3101 /* need to remove the md_redundancy_group */
3102 if (mddev->to_remove == NULL)
3103 mddev->to_remove = &md_redundancy_group;
3106 if (mddev->pers->sync_request == NULL &&
3108 /* We are converting from a no-redundancy array
3109 * to a redundancy array and metadata is managed
3110 * externally so we need to be sure that writes
3111 * won't block due to a need to transition
3113 * until external management is started.
3116 mddev->safemode_delay = 0;
3117 mddev->safemode = 0;
3120 list_for_each_entry(rdev, &mddev->disks, same_set) {
3122 if (rdev->raid_disk < 0)
3124 if (rdev->new_raid_disk > mddev->raid_disks)
3125 rdev->new_raid_disk = -1;
3126 if (rdev->new_raid_disk == rdev->raid_disk)
3128 sprintf(nm, "rd%d", rdev->raid_disk);
3129 sysfs_remove_link(&mddev->kobj, nm);
3131 list_for_each_entry(rdev, &mddev->disks, same_set) {
3132 if (rdev->raid_disk < 0)
3134 if (rdev->new_raid_disk == rdev->raid_disk)
3136 rdev->raid_disk = rdev->new_raid_disk;
3137 if (rdev->raid_disk < 0)
3138 clear_bit(In_sync, &rdev->flags);
3141 sprintf(nm, "rd%d", rdev->raid_disk);
3142 if(sysfs_create_link(&mddev->kobj, &rdev->kobj, nm))
3143 printk("md: cannot register %s for %s after level change\n",
3148 module_put(mddev->pers->owner);
3150 mddev->private = priv;
3151 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
3152 mddev->level = mddev->new_level;
3153 mddev->layout = mddev->new_layout;
3154 mddev->chunk_sectors = mddev->new_chunk_sectors;
3155 mddev->delta_disks = 0;
3156 if (mddev->pers->sync_request == NULL) {
3157 /* this is now an array without redundancy, so
3158 * it must always be in_sync
3161 del_timer_sync(&mddev->safemode_timer);
3164 mddev_resume(mddev);
3165 set_bit(MD_CHANGE_DEVS, &mddev->flags);
3166 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3167 md_wakeup_thread(mddev->thread);
3168 sysfs_notify(&mddev->kobj, NULL, "level");
3169 md_new_event(mddev);
3173 static struct md_sysfs_entry md_level =
3174 __ATTR(level, S_IRUGO|S_IWUSR, level_show, level_store);
3178 layout_show(mddev_t *mddev, char *page)
3180 /* just a number, not meaningful for all levels */
3181 if (mddev->reshape_position != MaxSector &&
3182 mddev->layout != mddev->new_layout)
3183 return sprintf(page, "%d (%d)\n",
3184 mddev->new_layout, mddev->layout);
3185 return sprintf(page, "%d\n", mddev->layout);
3189 layout_store(mddev_t *mddev, const char *buf, size_t len)
3192 unsigned long n = simple_strtoul(buf, &e, 10);
3194 if (!*buf || (*e && *e != '\n'))
3199 if (mddev->pers->check_reshape == NULL)
3201 mddev->new_layout = n;
3202 err = mddev->pers->check_reshape(mddev);
3204 mddev->new_layout = mddev->layout;
3208 mddev->new_layout = n;
3209 if (mddev->reshape_position == MaxSector)
3214 static struct md_sysfs_entry md_layout =
3215 __ATTR(layout, S_IRUGO|S_IWUSR, layout_show, layout_store);
3219 raid_disks_show(mddev_t *mddev, char *page)
3221 if (mddev->raid_disks == 0)
3223 if (mddev->reshape_position != MaxSector &&
3224 mddev->delta_disks != 0)
3225 return sprintf(page, "%d (%d)\n", mddev->raid_disks,
3226 mddev->raid_disks - mddev->delta_disks);
3227 return sprintf(page, "%d\n", mddev->raid_disks);
3230 static int update_raid_disks(mddev_t *mddev, int raid_disks);
3233 raid_disks_store(mddev_t *mddev, const char *buf, size_t len)
3237 unsigned long n = simple_strtoul(buf, &e, 10);
3239 if (!*buf || (*e && *e != '\n'))
3243 rv = update_raid_disks(mddev, n);
3244 else if (mddev->reshape_position != MaxSector) {
3245 int olddisks = mddev->raid_disks - mddev->delta_disks;
3246 mddev->delta_disks = n - olddisks;
3247 mddev->raid_disks = n;
3249 mddev->raid_disks = n;
3250 return rv ? rv : len;
3252 static struct md_sysfs_entry md_raid_disks =
3253 __ATTR(raid_disks, S_IRUGO|S_IWUSR, raid_disks_show, raid_disks_store);
3256 chunk_size_show(mddev_t *mddev, char *page)
3258 if (mddev->reshape_position != MaxSector &&
3259 mddev->chunk_sectors != mddev->new_chunk_sectors)
3260 return sprintf(page, "%d (%d)\n",
3261 mddev->new_chunk_sectors << 9,
3262 mddev->chunk_sectors << 9);
3263 return sprintf(page, "%d\n", mddev->chunk_sectors << 9);
3267 chunk_size_store(mddev_t *mddev, const char *buf, size_t len)
3270 unsigned long n = simple_strtoul(buf, &e, 10);
3272 if (!*buf || (*e && *e != '\n'))
3277 if (mddev->pers->check_reshape == NULL)
3279 mddev->new_chunk_sectors = n >> 9;
3280 err = mddev->pers->check_reshape(mddev);
3282 mddev->new_chunk_sectors = mddev->chunk_sectors;
3286 mddev->new_chunk_sectors = n >> 9;
3287 if (mddev->reshape_position == MaxSector)
3288 mddev->chunk_sectors = n >> 9;
3292 static struct md_sysfs_entry md_chunk_size =
3293 __ATTR(chunk_size, S_IRUGO|S_IWUSR, chunk_size_show, chunk_size_store);
3296 resync_start_show(mddev_t *mddev, char *page)
3298 if (mddev->recovery_cp == MaxSector)
3299 return sprintf(page, "none\n");
3300 return sprintf(page, "%llu\n", (unsigned long long)mddev->recovery_cp);
3304 resync_start_store(mddev_t *mddev, const char *buf, size_t len)
3307 unsigned long long n = simple_strtoull(buf, &e, 10);
3311 if (cmd_match(buf, "none"))
3313 else if (!*buf || (*e && *e != '\n'))
3316 mddev->recovery_cp = n;
3319 static struct md_sysfs_entry md_resync_start =
3320 __ATTR(resync_start, S_IRUGO|S_IWUSR, resync_start_show, resync_start_store);
3323 * The array state can be:
3326 * No devices, no size, no level
3327 * Equivalent to STOP_ARRAY ioctl
3329 * May have some settings, but array is not active
3330 * all IO results in error
3331 * When written, doesn't tear down array, but just stops it
3332 * suspended (not supported yet)
3333 * All IO requests will block. The array can be reconfigured.
3334 * Writing this, if accepted, will block until array is quiescent
3336 * no resync can happen. no superblocks get written.
3337 * write requests fail
3339 * like readonly, but behaves like 'clean' on a write request.
3341 * clean - no pending writes, but otherwise active.
3342 * When written to inactive array, starts without resync
3343 * If a write request arrives then
3344 * if metadata is known, mark 'dirty' and switch to 'active'.
3345 * if not known, block and switch to write-pending
3346 * If written to an active array that has pending writes, then fails.
3348 * fully active: IO and resync can be happening.
3349 * When written to inactive array, starts with resync
3352 * clean, but writes are blocked waiting for 'active' to be written.
3355 * like active, but no writes have been seen for a while (100msec).
3358 enum array_state { clear, inactive, suspended, readonly, read_auto, clean, active,
3359 write_pending, active_idle, bad_word};
3360 static char *array_states[] = {
3361 "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active",
3362 "write-pending", "active-idle", NULL };
3364 static int match_word(const char *word, char **list)
3367 for (n=0; list[n]; n++)
3368 if (cmd_match(word, list[n]))
3374 array_state_show(mddev_t *mddev, char *page)
3376 enum array_state st = inactive;
3389 else if (test_bit(MD_CHANGE_PENDING, &mddev->flags))
3391 else if (mddev->safemode)
3397 if (list_empty(&mddev->disks) &&
3398 mddev->raid_disks == 0 &&
3399 mddev->dev_sectors == 0)
3404 return sprintf(page, "%s\n", array_states[st]);
3407 static int do_md_stop(mddev_t * mddev, int ro, int is_open);
3408 static int md_set_readonly(mddev_t * mddev, int is_open);
3409 static int do_md_run(mddev_t * mddev);
3410 static int restart_array(mddev_t *mddev);
3413 array_state_store(mddev_t *mddev, const char *buf, size_t len)
3416 enum array_state st = match_word(buf, array_states);
3421 /* stopping an active array */
3422 if (atomic_read(&mddev->openers) > 0)
3424 err = do_md_stop(mddev, 0, 0);
3427 /* stopping an active array */
3429 if (atomic_read(&mddev->openers) > 0)
3431 err = do_md_stop(mddev, 2, 0);
3433 err = 0; /* already inactive */
3436 break; /* not supported yet */
3439 err = md_set_readonly(mddev, 0);
3442 set_disk_ro(mddev->gendisk, 1);
3443 err = do_md_run(mddev);
3449 err = md_set_readonly(mddev, 0);
3450 else if (mddev->ro == 1)
3451 err = restart_array(mddev);
3454 set_disk_ro(mddev->gendisk, 0);
3458 err = do_md_run(mddev);
3463 restart_array(mddev);
3464 spin_lock_irq(&mddev->write_lock);
3465 if (atomic_read(&mddev->writes_pending) == 0) {
3466 if (mddev->in_sync == 0) {
3468 if (mddev->safemode == 1)
3469 mddev->safemode = 0;
3470 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
3475 spin_unlock_irq(&mddev->write_lock);
3481 restart_array(mddev);
3482 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
3483 wake_up(&mddev->sb_wait);
3487 set_disk_ro(mddev->gendisk, 0);
3488 err = do_md_run(mddev);
3493 /* these cannot be set */
3499 sysfs_notify_dirent_safe(mddev->sysfs_state);
3503 static struct md_sysfs_entry md_array_state =
3504 __ATTR(array_state, S_IRUGO|S_IWUSR, array_state_show, array_state_store);
3507 max_corrected_read_errors_show(mddev_t *mddev, char *page) {
3508 return sprintf(page, "%d\n",
3509 atomic_read(&mddev->max_corr_read_errors));
3513 max_corrected_read_errors_store(mddev_t *mddev, const char *buf, size_t len)
3516 unsigned long n = simple_strtoul(buf, &e, 10);
3518 if (*buf && (*e == 0 || *e == '\n')) {
3519 atomic_set(&mddev->max_corr_read_errors, n);
3525 static struct md_sysfs_entry max_corr_read_errors =
3526 __ATTR(max_read_errors, S_IRUGO|S_IWUSR, max_corrected_read_errors_show,
3527 max_corrected_read_errors_store);
3530 null_show(mddev_t *mddev, char *page)
3536 new_dev_store(mddev_t *mddev, const char *buf, size_t len)
3538 /* buf must be %d:%d\n? giving major and minor numbers */
3539 /* The new device is added to the array.
3540 * If the array has a persistent superblock, we read the
3541 * superblock to initialise info and check validity.
3542 * Otherwise, only checking done is that in bind_rdev_to_array,
3543 * which mainly checks size.
3546 int major = simple_strtoul(buf, &e, 10);
3552 if (!*buf || *e != ':' || !e[1] || e[1] == '\n')
3554 minor = simple_strtoul(e+1, &e, 10);
3555 if (*e && *e != '\n')
3557 dev = MKDEV(major, minor);
3558 if (major != MAJOR(dev) ||
3559 minor != MINOR(dev))
3563 if (mddev->persistent) {
3564 rdev = md_import_device(dev, mddev->major_version,
3565 mddev->minor_version);
3566 if (!IS_ERR(rdev) && !list_empty(&mddev->disks)) {
3567 mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
3568 mdk_rdev_t, same_set);
3569 err = super_types[mddev->major_version]
3570 .load_super(rdev, rdev0, mddev->minor_version);
3574 } else if (mddev->external)
3575 rdev = md_import_device(dev, -2, -1);
3577 rdev = md_import_device(dev, -1, -1);
3580 return PTR_ERR(rdev);
3581 err = bind_rdev_to_array(rdev, mddev);
3585 return err ? err : len;
3588 static struct md_sysfs_entry md_new_device =
3589 __ATTR(new_dev, S_IWUSR, null_show, new_dev_store);
3592 bitmap_store(mddev_t *mddev, const char *buf, size_t len)
3595 unsigned long chunk, end_chunk;
3599 /* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */
3601 chunk = end_chunk = simple_strtoul(buf, &end, 0);
3602 if (buf == end) break;
3603 if (*end == '-') { /* range */
3605 end_chunk = simple_strtoul(buf, &end, 0);
3606 if (buf == end) break;
3608 if (*end && !isspace(*end)) break;
3609 bitmap_dirty_bits(mddev->bitmap, chunk, end_chunk);
3610 buf = skip_spaces(end);
3612 bitmap_unplug(mddev->bitmap); /* flush the bits to disk */
3617 static struct md_sysfs_entry md_bitmap =
3618 __ATTR(bitmap_set_bits, S_IWUSR, null_show, bitmap_store);
3621 size_show(mddev_t *mddev, char *page)
3623 return sprintf(page, "%llu\n",
3624 (unsigned long long)mddev->dev_sectors / 2);
3627 static int update_size(mddev_t *mddev, sector_t num_sectors);
3630 size_store(mddev_t *mddev, const char *buf, size_t len)
3632 /* If array is inactive, we can reduce the component size, but
3633 * not increase it (except from 0).
3634 * If array is active, we can try an on-line resize
3637 int err = strict_blocks_to_sectors(buf, §ors);
3642 err = update_size(mddev, sectors);
3643 md_update_sb(mddev, 1);
3645 if (mddev->dev_sectors == 0 ||
3646 mddev->dev_sectors > sectors)
3647 mddev->dev_sectors = sectors;
3651 return err ? err : len;
3654 static struct md_sysfs_entry md_size =
3655 __ATTR(component_size, S_IRUGO|S_IWUSR, size_show, size_store);
3660 * 'none' for arrays with no metadata (good luck...)
3661 * 'external' for arrays with externally managed metadata,
3662 * or N.M for internally known formats
3665 metadata_show(mddev_t *mddev, char *page)
3667 if (mddev->persistent)
3668 return sprintf(page, "%d.%d\n",
3669 mddev->major_version, mddev->minor_version);
3670 else if (mddev->external)
3671 return sprintf(page, "external:%s\n", mddev->metadata_type);
3673 return sprintf(page, "none\n");
3677 metadata_store(mddev_t *mddev, const char *buf, size_t len)
3681 /* Changing the details of 'external' metadata is
3682 * always permitted. Otherwise there must be
3683 * no devices attached to the array.
3685 if (mddev->external && strncmp(buf, "external:", 9) == 0)
3687 else if (!list_empty(&mddev->disks))
3690 if (cmd_match(buf, "none")) {
3691 mddev->persistent = 0;
3692 mddev->external = 0;
3693 mddev->major_version = 0;
3694 mddev->minor_version = 90;
3697 if (strncmp(buf, "external:", 9) == 0) {
3698 size_t namelen = len-9;
3699 if (namelen >= sizeof(mddev->metadata_type))
3700 namelen = sizeof(mddev->metadata_type)-1;
3701 strncpy(mddev->metadata_type, buf+9, namelen);
3702 mddev->metadata_type[namelen] = 0;
3703 if (namelen && mddev->metadata_type[namelen-1] == '\n')
3704 mddev->metadata_type[--namelen] = 0;
3705 mddev->persistent = 0;
3706 mddev->external = 1;
3707 mddev->major_version = 0;
3708 mddev->minor_version = 90;
3711 major = simple_strtoul(buf, &e, 10);
3712 if (e==buf || *e != '.')
3715 minor = simple_strtoul(buf, &e, 10);
3716 if (e==buf || (*e && *e != '\n') )
3718 if (major >= ARRAY_SIZE(super_types) || super_types[major].name == NULL)
3720 mddev->major_version = major;
3721 mddev->minor_version = minor;
3722 mddev->persistent = 1;
3723 mddev->external = 0;
3727 static struct md_sysfs_entry md_metadata =
3728 __ATTR(metadata_version, S_IRUGO|S_IWUSR, metadata_show, metadata_store);
3731 action_show(mddev_t *mddev, char *page)
3733 char *type = "idle";
3734 if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
3736 else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
3737 (!mddev->ro && test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))) {
3738 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
3740 else if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
3741 if (!test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
3743 else if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
3747 } else if (test_bit(MD_RECOVERY_RECOVER, &mddev->recovery))
3750 return sprintf(page, "%s\n", type);
3753 static void reap_sync_thread(mddev_t *mddev);
3756 action_store(mddev_t *mddev, const char *page, size_t len)
3758 if (!mddev->pers || !mddev->pers->sync_request)
3761 if (cmd_match(page, "frozen"))
3762 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
3764 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
3766 if (cmd_match(page, "idle") || cmd_match(page, "frozen")) {
3767 if (mddev->sync_thread) {
3768 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
3769 reap_sync_thread(mddev);
3771 } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
3772 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
3774 else if (cmd_match(page, "resync"))
3775 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3776 else if (cmd_match(page, "recover")) {
3777 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
3778 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3779 } else if (cmd_match(page, "reshape")) {
3781 if (mddev->pers->start_reshape == NULL)
3783 err = mddev->pers->start_reshape(mddev);
3786 sysfs_notify(&mddev->kobj, NULL, "degraded");
3788 if (cmd_match(page, "check"))
3789 set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
3790 else if (!cmd_match(page, "repair"))
3792 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
3793 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
3795 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3796 md_wakeup_thread(mddev->thread);
3797 sysfs_notify_dirent_safe(mddev->sysfs_action);
3802 mismatch_cnt_show(mddev_t *mddev, char *page)
3804 return sprintf(page, "%llu\n",
3805 (unsigned long long) mddev->resync_mismatches);
3808 static struct md_sysfs_entry md_scan_mode =
3809 __ATTR(sync_action, S_IRUGO|S_IWUSR, action_show, action_store);
3812 static struct md_sysfs_entry md_mismatches = __ATTR_RO(mismatch_cnt);
3815 sync_min_show(mddev_t *mddev, char *page)
3817 return sprintf(page, "%d (%s)\n", speed_min(mddev),
3818 mddev->sync_speed_min ? "local": "system");
3822 sync_min_store(mddev_t *mddev, const char *buf, size_t len)
3826 if (strncmp(buf, "system", 6)==0) {
3827 mddev->sync_speed_min = 0;
3830 min = simple_strtoul(buf, &e, 10);
3831 if (buf == e || (*e && *e != '\n') || min <= 0)
3833 mddev->sync_speed_min = min;
3837 static struct md_sysfs_entry md_sync_min =
3838 __ATTR(sync_speed_min, S_IRUGO|S_IWUSR, sync_min_show, sync_min_store);
3841 sync_max_show(mddev_t *mddev, char *page)
3843 return sprintf(page, "%d (%s)\n", speed_max(mddev),
3844 mddev->sync_speed_max ? "local": "system");
3848 sync_max_store(mddev_t *mddev, const char *buf, size_t len)
3852 if (strncmp(buf, "system", 6)==0) {
3853 mddev->sync_speed_max = 0;
3856 max = simple_strtoul(buf, &e, 10);
3857 if (buf == e || (*e && *e != '\n') || max <= 0)
3859 mddev->sync_speed_max = max;
3863 static struct md_sysfs_entry md_sync_max =
3864 __ATTR(sync_speed_max, S_IRUGO|S_IWUSR, sync_max_show, sync_max_store);
3867 degraded_show(mddev_t *mddev, char *page)
3869 return sprintf(page, "%d\n", mddev->degraded);
3871 static struct md_sysfs_entry md_degraded = __ATTR_RO(degraded);
3874 sync_force_parallel_show(mddev_t *mddev, char *page)
3876 return sprintf(page, "%d\n", mddev->parallel_resync);
3880 sync_force_parallel_store(mddev_t *mddev, const char *buf, size_t len)
3884 if (strict_strtol(buf, 10, &n))
3887 if (n != 0 && n != 1)
3890 mddev->parallel_resync = n;
3892 if (mddev->sync_thread)
3893 wake_up(&resync_wait);
3898 /* force parallel resync, even with shared block devices */
3899 static struct md_sysfs_entry md_sync_force_parallel =
3900 __ATTR(sync_force_parallel, S_IRUGO|S_IWUSR,
3901 sync_force_parallel_show, sync_force_parallel_store);
3904 sync_speed_show(mddev_t *mddev, char *page)
3906 unsigned long resync, dt, db;
3907 if (mddev->curr_resync == 0)
3908 return sprintf(page, "none\n");
3909 resync = mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active);
3910 dt = (jiffies - mddev->resync_mark) / HZ;
3912 db = resync - mddev->resync_mark_cnt;
3913 return sprintf(page, "%lu\n", db/dt/2); /* K/sec */
3916 static struct md_sysfs_entry md_sync_speed = __ATTR_RO(sync_speed);
3919 sync_completed_show(mddev_t *mddev, char *page)
3921 unsigned long max_sectors, resync;
3923 if (!test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
3924 return sprintf(page, "none\n");
3926 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
3927 max_sectors = mddev->resync_max_sectors;
3929 max_sectors = mddev->dev_sectors;
3931 resync = mddev->curr_resync_completed;
3932 return sprintf(page, "%lu / %lu\n", resync, max_sectors);
3935 static struct md_sysfs_entry md_sync_completed = __ATTR_RO(sync_completed);
3938 min_sync_show(mddev_t *mddev, char *page)
3940 return sprintf(page, "%llu\n",
3941 (unsigned long long)mddev->resync_min);
3944 min_sync_store(mddev_t *mddev, const char *buf, size_t len)
3946 unsigned long long min;
3947 if (strict_strtoull(buf, 10, &min))
3949 if (min > mddev->resync_max)
3951 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
3954 /* Must be a multiple of chunk_size */
3955 if (mddev->chunk_sectors) {
3956 sector_t temp = min;
3957 if (sector_div(temp, mddev->chunk_sectors))
3960 mddev->resync_min = min;
3965 static struct md_sysfs_entry md_min_sync =
3966 __ATTR(sync_min, S_IRUGO|S_IWUSR, min_sync_show, min_sync_store);
3969 max_sync_show(mddev_t *mddev, char *page)
3971 if (mddev->resync_max == MaxSector)
3972 return sprintf(page, "max\n");
3974 return sprintf(page, "%llu\n",
3975 (unsigned long long)mddev->resync_max);
3978 max_sync_store(mddev_t *mddev, const char *buf, size_t len)
3980 if (strncmp(buf, "max", 3) == 0)
3981 mddev->resync_max = MaxSector;
3983 unsigned long long max;
3984 if (strict_strtoull(buf, 10, &max))
3986 if (max < mddev->resync_min)
3988 if (max < mddev->resync_max &&
3990 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
3993 /* Must be a multiple of chunk_size */
3994 if (mddev->chunk_sectors) {
3995 sector_t temp = max;
3996 if (sector_div(temp, mddev->chunk_sectors))
3999 mddev->resync_max = max;
4001 wake_up(&mddev->recovery_wait);
4005 static struct md_sysfs_entry md_max_sync =
4006 __ATTR(sync_max, S_IRUGO|S_IWUSR, max_sync_show, max_sync_store);
4009 suspend_lo_show(mddev_t *mddev, char *page)
4011 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_lo);
4015 suspend_lo_store(mddev_t *mddev, const char *buf, size_t len)
4018 unsigned long long new = simple_strtoull(buf, &e, 10);
4019 unsigned long long old = mddev->suspend_lo;
4021 if (mddev->pers == NULL ||
4022 mddev->pers->quiesce == NULL)
4024 if (buf == e || (*e && *e != '\n'))
4027 mddev->suspend_lo = new;
4029 /* Shrinking suspended region */
4030 mddev->pers->quiesce(mddev, 2);
4032 /* Expanding suspended region - need to wait */
4033 mddev->pers->quiesce(mddev, 1);
4034 mddev->pers->quiesce(mddev, 0);
4038 static struct md_sysfs_entry md_suspend_lo =
4039 __ATTR(suspend_lo, S_IRUGO|S_IWUSR, suspend_lo_show, suspend_lo_store);
4043 suspend_hi_show(mddev_t *mddev, char *page)
4045 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_hi);
4049 suspend_hi_store(mddev_t *mddev, const char *buf, size_t len)
4052 unsigned long long new = simple_strtoull(buf, &e, 10);
4053 unsigned long long old = mddev->suspend_hi;
4055 if (mddev->pers == NULL ||
4056 mddev->pers->quiesce == NULL)
4058 if (buf == e || (*e && *e != '\n'))
4061 mddev->suspend_hi = new;
4063 /* Shrinking suspended region */
4064 mddev->pers->quiesce(mddev, 2);
4066 /* Expanding suspended region - need to wait */
4067 mddev->pers->quiesce(mddev, 1);
4068 mddev->pers->quiesce(mddev, 0);
4072 static struct md_sysfs_entry md_suspend_hi =
4073 __ATTR(suspend_hi, S_IRUGO|S_IWUSR, suspend_hi_show, suspend_hi_store);
4076 reshape_position_show(mddev_t *mddev, char *page)
4078 if (mddev->reshape_position != MaxSector)
4079 return sprintf(page, "%llu\n",
4080 (unsigned long long)mddev->reshape_position);
4081 strcpy(page, "none\n");
4086 reshape_position_store(mddev_t *mddev, const char *buf, size_t len)
4089 unsigned long long new = simple_strtoull(buf, &e, 10);
4092 if (buf == e || (*e && *e != '\n'))
4094 mddev->reshape_position = new;
4095 mddev->delta_disks = 0;
4096 mddev->new_level = mddev->level;
4097 mddev->new_layout = mddev->layout;
4098 mddev->new_chunk_sectors = mddev->chunk_sectors;
4102 static struct md_sysfs_entry md_reshape_position =
4103 __ATTR(reshape_position, S_IRUGO|S_IWUSR, reshape_position_show,
4104 reshape_position_store);
4107 array_size_show(mddev_t *mddev, char *page)
4109 if (mddev->external_size)
4110 return sprintf(page, "%llu\n",
4111 (unsigned long long)mddev->array_sectors/2);
4113 return sprintf(page, "default\n");
4117 array_size_store(mddev_t *mddev, const char *buf, size_t len)
4121 if (strncmp(buf, "default", 7) == 0) {
4123 sectors = mddev->pers->size(mddev, 0, 0);
4125 sectors = mddev->array_sectors;
4127 mddev->external_size = 0;
4129 if (strict_blocks_to_sectors(buf, §ors) < 0)
4131 if (mddev->pers && mddev->pers->size(mddev, 0, 0) < sectors)
4134 mddev->external_size = 1;
4137 mddev->array_sectors = sectors;
4138 set_capacity(mddev->gendisk, mddev->array_sectors);
4140 revalidate_disk(mddev->gendisk);
4145 static struct md_sysfs_entry md_array_size =
4146 __ATTR(array_size, S_IRUGO|S_IWUSR, array_size_show,
4149 static struct attribute *md_default_attrs[] = {
4152 &md_raid_disks.attr,
4153 &md_chunk_size.attr,
4155 &md_resync_start.attr,
4157 &md_new_device.attr,
4158 &md_safe_delay.attr,
4159 &md_array_state.attr,
4160 &md_reshape_position.attr,
4161 &md_array_size.attr,
4162 &max_corr_read_errors.attr,
4166 static struct attribute *md_redundancy_attrs[] = {
4168 &md_mismatches.attr,
4171 &md_sync_speed.attr,
4172 &md_sync_force_parallel.attr,
4173 &md_sync_completed.attr,
4176 &md_suspend_lo.attr,
4177 &md_suspend_hi.attr,
4182 static struct attribute_group md_redundancy_group = {
4184 .attrs = md_redundancy_attrs,
4189 md_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
4191 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
4192 mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
4197 rv = mddev_lock(mddev);
4199 rv = entry->show(mddev, page);
4200 mddev_unlock(mddev);
4206 md_attr_store(struct kobject *kobj, struct attribute *attr,
4207 const char *page, size_t length)
4209 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
4210 mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
4215 if (!capable(CAP_SYS_ADMIN))
4217 rv = mddev_lock(mddev);
4218 if (mddev->hold_active == UNTIL_IOCTL)
4219 mddev->hold_active = 0;
4221 rv = entry->store(mddev, page, length);
4222 mddev_unlock(mddev);
4227 static void md_free(struct kobject *ko)
4229 mddev_t *mddev = container_of(ko, mddev_t, kobj);
4231 if (mddev->sysfs_state)
4232 sysfs_put(mddev->sysfs_state);
4234 if (mddev->gendisk) {
4235 del_gendisk(mddev->gendisk);
4236 put_disk(mddev->gendisk);
4239 blk_cleanup_queue(mddev->queue);
4244 static const struct sysfs_ops md_sysfs_ops = {
4245 .show = md_attr_show,
4246 .store = md_attr_store,
4248 static struct kobj_type md_ktype = {
4250 .sysfs_ops = &md_sysfs_ops,
4251 .default_attrs = md_default_attrs,
4256 static void mddev_delayed_delete(struct work_struct *ws)
4258 mddev_t *mddev = container_of(ws, mddev_t, del_work);
4260 sysfs_remove_group(&mddev->kobj, &md_bitmap_group);
4261 kobject_del(&mddev->kobj);
4262 kobject_put(&mddev->kobj);
4265 static int md_alloc(dev_t dev, char *name)
4267 static DEFINE_MUTEX(disks_mutex);
4268 mddev_t *mddev = mddev_find(dev);
4269 struct gendisk *disk;
4278 partitioned = (MAJOR(mddev->unit) != MD_MAJOR);
4279 shift = partitioned ? MdpMinorShift : 0;
4280 unit = MINOR(mddev->unit) >> shift;
4282 /* wait for any previous instance of this device to be
4283 * completely removed (mddev_delayed_delete).
4285 flush_workqueue(md_misc_wq);
4287 mutex_lock(&disks_mutex);
4293 /* Need to ensure that 'name' is not a duplicate.
4296 spin_lock(&all_mddevs_lock);
4298 list_for_each_entry(mddev2, &all_mddevs, all_mddevs)
4299 if (mddev2->gendisk &&
4300 strcmp(mddev2->gendisk->disk_name, name) == 0) {
4301 spin_unlock(&all_mddevs_lock);
4304 spin_unlock(&all_mddevs_lock);
4308 mddev->queue = blk_alloc_queue(GFP_KERNEL);
4311 mddev->queue->queuedata = mddev;
4313 blk_queue_make_request(mddev->queue, md_make_request);
4315 disk = alloc_disk(1 << shift);
4317 blk_cleanup_queue(mddev->queue);
4318 mddev->queue = NULL;
4321 disk->major = MAJOR(mddev->unit);
4322 disk->first_minor = unit << shift;
4324 strcpy(disk->disk_name, name);
4325 else if (partitioned)
4326 sprintf(disk->disk_name, "md_d%d", unit);
4328 sprintf(disk->disk_name, "md%d", unit);
4329 disk->fops = &md_fops;
4330 disk->private_data = mddev;
4331 disk->queue = mddev->queue;
4332 /* Allow extended partitions. This makes the
4333 * 'mdp' device redundant, but we can't really
4336 disk->flags |= GENHD_FL_EXT_DEVT;
4338 mddev->gendisk = disk;
4339 error = kobject_init_and_add(&mddev->kobj, &md_ktype,
4340 &disk_to_dev(disk)->kobj, "%s", "md");
4342 /* This isn't possible, but as kobject_init_and_add is marked
4343 * __must_check, we must do something with the result
4345 printk(KERN_WARNING "md: cannot register %s/md - name in use\n",
4349 if (mddev->kobj.sd &&
4350 sysfs_create_group(&mddev->kobj, &md_bitmap_group))
4351 printk(KERN_DEBUG "pointless warning\n");
4353 blk_queue_flush(mddev->queue, REQ_FLUSH | REQ_FUA);
4355 mutex_unlock(&disks_mutex);
4356 if (!error && mddev->kobj.sd) {
4357 kobject_uevent(&mddev->kobj, KOBJ_ADD);
4358 mddev->sysfs_state = sysfs_get_dirent_safe(mddev->kobj.sd, "array_state");
4364 static struct kobject *md_probe(dev_t dev, int *part, void *data)
4366 md_alloc(dev, NULL);
4370 static int add_named_array(const char *val, struct kernel_param *kp)
4372 /* val must be "md_*" where * is not all digits.
4373 * We allocate an array with a large free minor number, and
4374 * set the name to val. val must not already be an active name.
4376 int len = strlen(val);
4377 char buf[DISK_NAME_LEN];
4379 while (len && val[len-1] == '\n')
4381 if (len >= DISK_NAME_LEN)
4383 strlcpy(buf, val, len+1);
4384 if (strncmp(buf, "md_", 3) != 0)
4386 return md_alloc(0, buf);
4389 static void md_safemode_timeout(unsigned long data)
4391 mddev_t *mddev = (mddev_t *) data;
4393 if (!atomic_read(&mddev->writes_pending)) {
4394 mddev->safemode = 1;
4395 if (mddev->external)
4396 sysfs_notify_dirent_safe(mddev->sysfs_state);
4398 md_wakeup_thread(mddev->thread);
4401 static int start_dirty_degraded;
4403 int md_run(mddev_t *mddev)
4407 struct mdk_personality *pers;
4409 if (list_empty(&mddev->disks))
4410 /* cannot run an array with no devices.. */
4415 /* Cannot run until previous stop completes properly */
4416 if (mddev->sysfs_active)
4420 * Analyze all RAID superblock(s)
4422 if (!mddev->raid_disks) {
4423 if (!mddev->persistent)
4428 if (mddev->level != LEVEL_NONE)
4429 request_module("md-level-%d", mddev->level);
4430 else if (mddev->clevel[0])
4431 request_module("md-%s", mddev->clevel);
4434 * Drop all container device buffers, from now on
4435 * the only valid external interface is through the md
4438 list_for_each_entry(rdev, &mddev->disks, same_set) {
4439 if (test_bit(Faulty, &rdev->flags))
4441 sync_blockdev(rdev->bdev);
4442 invalidate_bdev(rdev->bdev);
4444 /* perform some consistency tests on the device.
4445 * We don't want the data to overlap the metadata,
4446 * Internal Bitmap issues have been handled elsewhere.
4448 if (rdev->meta_bdev) {
4449 /* Nothing to check */;
4450 } else if (rdev->data_offset < rdev->sb_start) {
4451 if (mddev->dev_sectors &&
4452 rdev->data_offset + mddev->dev_sectors
4454 printk("md: %s: data overlaps metadata\n",
4459 if (rdev->sb_start + rdev->sb_size/512
4460 > rdev->data_offset) {
4461 printk("md: %s: metadata overlaps data\n",
4466 sysfs_notify_dirent_safe(rdev->sysfs_state);
4469 if (mddev->bio_set == NULL)
4470 mddev->bio_set = bioset_create(BIO_POOL_SIZE, sizeof(mddev));
4472 spin_lock(&pers_lock);
4473 pers = find_pers(mddev->level, mddev->clevel);
4474 if (!pers || !try_module_get(pers->owner)) {
4475 spin_unlock(&pers_lock);
4476 if (mddev->level != LEVEL_NONE)
4477 printk(KERN_WARNING "md: personality for level %d is not loaded!\n",
4480 printk(KERN_WARNING "md: personality for level %s is not loaded!\n",
4485 spin_unlock(&pers_lock);
4486 if (mddev->level != pers->level) {
4487 mddev->level = pers->level;
4488 mddev->new_level = pers->level;
4490 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
4492 if (mddev->reshape_position != MaxSector &&
4493 pers->start_reshape == NULL) {
4494 /* This personality cannot handle reshaping... */
4496 module_put(pers->owner);
4500 if (pers->sync_request) {
4501 /* Warn if this is a potentially silly
4504 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
4508 list_for_each_entry(rdev, &mddev->disks, same_set)
4509 list_for_each_entry(rdev2, &mddev->disks, same_set) {
4511 rdev->bdev->bd_contains ==
4512 rdev2->bdev->bd_contains) {
4514 "%s: WARNING: %s appears to be"
4515 " on the same physical disk as"
4518 bdevname(rdev->bdev,b),
4519 bdevname(rdev2->bdev,b2));
4526 "True protection against single-disk"
4527 " failure might be compromised.\n");
4530 mddev->recovery = 0;
4531 /* may be over-ridden by personality */
4532 mddev->resync_max_sectors = mddev->dev_sectors;
4534 mddev->ok_start_degraded = start_dirty_degraded;
4536 if (start_readonly && mddev->ro == 0)
4537 mddev->ro = 2; /* read-only, but switch on first write */
4539 err = mddev->pers->run(mddev);
4541 printk(KERN_ERR "md: pers->run() failed ...\n");
4542 else if (mddev->pers->size(mddev, 0, 0) < mddev->array_sectors) {
4543 WARN_ONCE(!mddev->external_size, "%s: default size too small,"
4544 " but 'external_size' not in effect?\n", __func__);
4546 "md: invalid array_size %llu > default size %llu\n",
4547 (unsigned long long)mddev->array_sectors / 2,
4548 (unsigned long long)mddev->pers->size(mddev, 0, 0) / 2);
4550 mddev->pers->stop(mddev);
4552 if (err == 0 && mddev->pers->sync_request) {
4553 err = bitmap_create(mddev);
4555 printk(KERN_ERR "%s: failed to create bitmap (%d)\n",
4556 mdname(mddev), err);
4557 mddev->pers->stop(mddev);
4561 module_put(mddev->pers->owner);
4563 bitmap_destroy(mddev);
4566 if (mddev->pers->sync_request) {
4567 if (mddev->kobj.sd &&
4568 sysfs_create_group(&mddev->kobj, &md_redundancy_group))
4570 "md: cannot register extra attributes for %s\n",
4572 mddev->sysfs_action = sysfs_get_dirent_safe(mddev->kobj.sd, "sync_action");
4573 } else if (mddev->ro == 2) /* auto-readonly not meaningful */
4576 atomic_set(&mddev->writes_pending,0);
4577 atomic_set(&mddev->max_corr_read_errors,
4578 MD_DEFAULT_MAX_CORRECTED_READ_ERRORS);
4579 mddev->safemode = 0;
4580 mddev->safemode_timer.function = md_safemode_timeout;
4581 mddev->safemode_timer.data = (unsigned long) mddev;
4582 mddev->safemode_delay = (200 * HZ)/1000 +1; /* 200 msec delay */
4586 list_for_each_entry(rdev, &mddev->disks, same_set)
4587 if (rdev->raid_disk >= 0) {
4589 sprintf(nm, "rd%d", rdev->raid_disk);
4590 if (sysfs_create_link(&mddev->kobj, &rdev->kobj, nm))
4591 /* failure here is OK */;
4594 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4597 md_update_sb(mddev, 0);
4599 md_wakeup_thread(mddev->thread);
4600 md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
4602 md_new_event(mddev);
4603 sysfs_notify_dirent_safe(mddev->sysfs_state);
4604 sysfs_notify_dirent_safe(mddev->sysfs_action);
4605 sysfs_notify(&mddev->kobj, NULL, "degraded");
4608 EXPORT_SYMBOL_GPL(md_run);
4610 static int do_md_run(mddev_t *mddev)
4614 err = md_run(mddev);
4617 err = bitmap_load(mddev);
4619 bitmap_destroy(mddev);
4622 set_capacity(mddev->gendisk, mddev->array_sectors);
4623 revalidate_disk(mddev->gendisk);
4624 kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
4629 static int restart_array(mddev_t *mddev)
4631 struct gendisk *disk = mddev->gendisk;
4633 /* Complain if it has no devices */
4634 if (list_empty(&mddev->disks))
4640 mddev->safemode = 0;
4642 set_disk_ro(disk, 0);
4643 printk(KERN_INFO "md: %s switched to read-write mode.\n",
4645 /* Kick recovery or resync if necessary */
4646 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4647 md_wakeup_thread(mddev->thread);
4648 md_wakeup_thread(mddev->sync_thread);
4649 sysfs_notify_dirent_safe(mddev->sysfs_state);
4653 /* similar to deny_write_access, but accounts for our holding a reference
4654 * to the file ourselves */
4655 static int deny_bitmap_write_access(struct file * file)
4657 struct inode *inode = file->f_mapping->host;
4659 spin_lock(&inode->i_lock);
4660 if (atomic_read(&inode->i_writecount) > 1) {
4661 spin_unlock(&inode->i_lock);
4664 atomic_set(&inode->i_writecount, -1);
4665 spin_unlock(&inode->i_lock);
4670 void restore_bitmap_write_access(struct file *file)
4672 struct inode *inode = file->f_mapping->host;
4674 spin_lock(&inode->i_lock);
4675 atomic_set(&inode->i_writecount, 1);
4676 spin_unlock(&inode->i_lock);
4679 static void md_clean(mddev_t *mddev)
4681 mddev->array_sectors = 0;
4682 mddev->external_size = 0;
4683 mddev->dev_sectors = 0;
4684 mddev->raid_disks = 0;
4685 mddev->recovery_cp = 0;
4686 mddev->resync_min = 0;
4687 mddev->resync_max = MaxSector;
4688 mddev->reshape_position = MaxSector;
4689 mddev->external = 0;
4690 mddev->persistent = 0;
4691 mddev->level = LEVEL_NONE;
4692 mddev->clevel[0] = 0;
4695 mddev->metadata_type[0] = 0;
4696 mddev->chunk_sectors = 0;
4697 mddev->ctime = mddev->utime = 0;
4699 mddev->max_disks = 0;
4701 mddev->can_decrease_events = 0;
4702 mddev->delta_disks = 0;
4703 mddev->new_level = LEVEL_NONE;
4704 mddev->new_layout = 0;
4705 mddev->new_chunk_sectors = 0;
4706 mddev->curr_resync = 0;
4707 mddev->resync_mismatches = 0;
4708 mddev->suspend_lo = mddev->suspend_hi = 0;
4709 mddev->sync_speed_min = mddev->sync_speed_max = 0;
4710 mddev->recovery = 0;
4712 mddev->degraded = 0;
4713 mddev->safemode = 0;
4714 mddev->bitmap_info.offset = 0;
4715 mddev->bitmap_info.default_offset = 0;
4716 mddev->bitmap_info.chunksize = 0;
4717 mddev->bitmap_info.daemon_sleep = 0;
4718 mddev->bitmap_info.max_write_behind = 0;
4722 static void __md_stop_writes(mddev_t *mddev)
4724 if (mddev->sync_thread) {
4725 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4726 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
4727 reap_sync_thread(mddev);
4730 del_timer_sync(&mddev->safemode_timer);
4732 bitmap_flush(mddev);
4733 md_super_wait(mddev);
4735 if (!mddev->in_sync || mddev->flags) {
4736 /* mark array as shutdown cleanly */
4738 md_update_sb(mddev, 1);
4742 void md_stop_writes(mddev_t *mddev)
4745 __md_stop_writes(mddev);
4746 mddev_unlock(mddev);
4748 EXPORT_SYMBOL_GPL(md_stop_writes);
4750 void md_stop(mddev_t *mddev)
4753 mddev->pers->stop(mddev);
4754 if (mddev->pers->sync_request && mddev->to_remove == NULL)
4755 mddev->to_remove = &md_redundancy_group;
4756 module_put(mddev->pers->owner);
4758 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4760 EXPORT_SYMBOL_GPL(md_stop);
4762 static int md_set_readonly(mddev_t *mddev, int is_open)
4765 mutex_lock(&mddev->open_mutex);
4766 if (atomic_read(&mddev->openers) > is_open) {
4767 printk("md: %s still in use.\n",mdname(mddev));
4772 __md_stop_writes(mddev);
4778 set_disk_ro(mddev->gendisk, 1);
4779 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4780 sysfs_notify_dirent_safe(mddev->sysfs_state);
4784 mutex_unlock(&mddev->open_mutex);
4789 * 0 - completely stop and dis-assemble array
4790 * 2 - stop but do not disassemble array
4792 static int do_md_stop(mddev_t * mddev, int mode, int is_open)
4794 struct gendisk *disk = mddev->gendisk;
4797 mutex_lock(&mddev->open_mutex);
4798 if (atomic_read(&mddev->openers) > is_open ||
4799 mddev->sysfs_active) {
4800 printk("md: %s still in use.\n",mdname(mddev));
4801 mutex_unlock(&mddev->open_mutex);
4807 set_disk_ro(disk, 0);
4809 __md_stop_writes(mddev);
4811 mddev->queue->merge_bvec_fn = NULL;
4812 mddev->queue->unplug_fn = NULL;
4813 mddev->queue->backing_dev_info.congested_fn = NULL;
4815 /* tell userspace to handle 'inactive' */
4816 sysfs_notify_dirent_safe(mddev->sysfs_state);
4818 list_for_each_entry(rdev, &mddev->disks, same_set)
4819 if (rdev->raid_disk >= 0) {
4821 sprintf(nm, "rd%d", rdev->raid_disk);
4822 sysfs_remove_link(&mddev->kobj, nm);
4825 set_capacity(disk, 0);
4826 mutex_unlock(&mddev->open_mutex);
4827 revalidate_disk(disk);
4832 mutex_unlock(&mddev->open_mutex);
4834 * Free resources if final stop
4837 printk(KERN_INFO "md: %s stopped.\n", mdname(mddev));
4839 bitmap_destroy(mddev);
4840 if (mddev->bitmap_info.file) {
4841 restore_bitmap_write_access(mddev->bitmap_info.file);
4842 fput(mddev->bitmap_info.file);
4843 mddev->bitmap_info.file = NULL;
4845 mddev->bitmap_info.offset = 0;
4847 export_array(mddev);
4850 kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
4851 if (mddev->hold_active == UNTIL_STOP)
4852 mddev->hold_active = 0;
4854 blk_integrity_unregister(disk);
4855 md_new_event(mddev);
4856 sysfs_notify_dirent_safe(mddev->sysfs_state);
4861 static void autorun_array(mddev_t *mddev)
4866 if (list_empty(&mddev->disks))
4869 printk(KERN_INFO "md: running: ");
4871 list_for_each_entry(rdev, &mddev->disks, same_set) {
4872 char b[BDEVNAME_SIZE];
4873 printk("<%s>", bdevname(rdev->bdev,b));
4877 err = do_md_run(mddev);
4879 printk(KERN_WARNING "md: do_md_run() returned %d\n", err);
4880 do_md_stop(mddev, 0, 0);
4885 * lets try to run arrays based on all disks that have arrived
4886 * until now. (those are in pending_raid_disks)
4888 * the method: pick the first pending disk, collect all disks with
4889 * the same UUID, remove all from the pending list and put them into
4890 * the 'same_array' list. Then order this list based on superblock
4891 * update time (freshest comes first), kick out 'old' disks and
4892 * compare superblocks. If everything's fine then run it.
4894 * If "unit" is allocated, then bump its reference count
4896 static void autorun_devices(int part)
4898 mdk_rdev_t *rdev0, *rdev, *tmp;
4900 char b[BDEVNAME_SIZE];
4902 printk(KERN_INFO "md: autorun ...\n");
4903 while (!list_empty(&pending_raid_disks)) {
4906 LIST_HEAD(candidates);
4907 rdev0 = list_entry(pending_raid_disks.next,
4908 mdk_rdev_t, same_set);
4910 printk(KERN_INFO "md: considering %s ...\n",
4911 bdevname(rdev0->bdev,b));
4912 INIT_LIST_HEAD(&candidates);
4913 rdev_for_each_list(rdev, tmp, &pending_raid_disks)
4914 if (super_90_load(rdev, rdev0, 0) >= 0) {
4915 printk(KERN_INFO "md: adding %s ...\n",
4916 bdevname(rdev->bdev,b));
4917 list_move(&rdev->same_set, &candidates);
4920 * now we have a set of devices, with all of them having
4921 * mostly sane superblocks. It's time to allocate the
4925 dev = MKDEV(mdp_major,
4926 rdev0->preferred_minor << MdpMinorShift);
4927 unit = MINOR(dev) >> MdpMinorShift;
4929 dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
4932 if (rdev0->preferred_minor != unit) {
4933 printk(KERN_INFO "md: unit number in %s is bad: %d\n",
4934 bdevname(rdev0->bdev, b), rdev0->preferred_minor);
4938 md_probe(dev, NULL, NULL);
4939 mddev = mddev_find(dev);
4940 if (!mddev || !mddev->gendisk) {
4944 "md: cannot allocate memory for md drive.\n");
4947 if (mddev_lock(mddev))
4948 printk(KERN_WARNING "md: %s locked, cannot run\n",
4950 else if (mddev->raid_disks || mddev->major_version
4951 || !list_empty(&mddev->disks)) {
4953 "md: %s already running, cannot run %s\n",
4954 mdname(mddev), bdevname(rdev0->bdev,b));
4955 mddev_unlock(mddev);
4957 printk(KERN_INFO "md: created %s\n", mdname(mddev));
4958 mddev->persistent = 1;
4959 rdev_for_each_list(rdev, tmp, &candidates) {
4960 list_del_init(&rdev->same_set);
4961 if (bind_rdev_to_array(rdev, mddev))
4964 autorun_array(mddev);
4965 mddev_unlock(mddev);
4967 /* on success, candidates will be empty, on error
4970 rdev_for_each_list(rdev, tmp, &candidates) {
4971 list_del_init(&rdev->same_set);
4976 printk(KERN_INFO "md: ... autorun DONE.\n");
4978 #endif /* !MODULE */
4980 static int get_version(void __user * arg)
4984 ver.major = MD_MAJOR_VERSION;
4985 ver.minor = MD_MINOR_VERSION;
4986 ver.patchlevel = MD_PATCHLEVEL_VERSION;
4988 if (copy_to_user(arg, &ver, sizeof(ver)))
4994 static int get_array_info(mddev_t * mddev, void __user * arg)
4996 mdu_array_info_t info;
4997 int nr,working,insync,failed,spare;
5000 nr=working=insync=failed=spare=0;
5001 list_for_each_entry(rdev, &mddev->disks, same_set) {
5003 if (test_bit(Faulty, &rdev->flags))
5007 if (test_bit(In_sync, &rdev->flags))
5014 info.major_version = mddev->major_version;
5015 info.minor_version = mddev->minor_version;
5016 info.patch_version = MD_PATCHLEVEL_VERSION;
5017 info.ctime = mddev->ctime;
5018 info.level = mddev->level;
5019 info.size = mddev->dev_sectors / 2;
5020 if (info.size != mddev->dev_sectors / 2) /* overflow */
5023 info.raid_disks = mddev->raid_disks;
5024 info.md_minor = mddev->md_minor;
5025 info.not_persistent= !mddev->persistent;
5027 info.utime = mddev->utime;
5030 info.state = (1<<MD_SB_CLEAN);
5031 if (mddev->bitmap && mddev->bitmap_info.offset)
5032 info.state = (1<<MD_SB_BITMAP_PRESENT);
5033 info.active_disks = insync;
5034 info.working_disks = working;
5035 info.failed_disks = failed;
5036 info.spare_disks = spare;
5038 info.layout = mddev->layout;
5039 info.chunk_size = mddev->chunk_sectors << 9;
5041 if (copy_to_user(arg, &info, sizeof(info)))
5047 static int get_bitmap_file(mddev_t * mddev, void __user * arg)
5049 mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */
5050 char *ptr, *buf = NULL;
5053 if (md_allow_write(mddev))
5054 file = kmalloc(sizeof(*file), GFP_NOIO);
5056 file = kmalloc(sizeof(*file), GFP_KERNEL);
5061 /* bitmap disabled, zero the first byte and copy out */
5062 if (!mddev->bitmap || !mddev->bitmap->file) {
5063 file->pathname[0] = '\0';
5067 buf = kmalloc(sizeof(file->pathname), GFP_KERNEL);
5071 ptr = d_path(&mddev->bitmap->file->f_path, buf, sizeof(file->pathname));
5075 strcpy(file->pathname, ptr);
5079 if (copy_to_user(arg, file, sizeof(*file)))
5087 static int get_disk_info(mddev_t * mddev, void __user * arg)
5089 mdu_disk_info_t info;
5092 if (copy_from_user(&info, arg, sizeof(info)))
5095 rdev = find_rdev_nr(mddev, info.number);
5097 info.major = MAJOR(rdev->bdev->bd_dev);
5098 info.minor = MINOR(rdev->bdev->bd_dev);
5099 info.raid_disk = rdev->raid_disk;
5101 if (test_bit(Faulty, &rdev->flags))
5102 info.state |= (1<<MD_DISK_FAULTY);
5103 else if (test_bit(In_sync, &rdev->flags)) {
5104 info.state |= (1<<MD_DISK_ACTIVE);
5105 info.state |= (1<<MD_DISK_SYNC);
5107 if (test_bit(WriteMostly, &rdev->flags))
5108 info.state |= (1<<MD_DISK_WRITEMOSTLY);
5110 info.major = info.minor = 0;
5111 info.raid_disk = -1;
5112 info.state = (1<<MD_DISK_REMOVED);
5115 if (copy_to_user(arg, &info, sizeof(info)))
5121 static int add_new_disk(mddev_t * mddev, mdu_disk_info_t *info)
5123 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
5125 dev_t dev = MKDEV(info->major,info->minor);
5127 if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
5130 if (!mddev->raid_disks) {
5132 /* expecting a device which has a superblock */
5133 rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
5136 "md: md_import_device returned %ld\n",
5138 return PTR_ERR(rdev);
5140 if (!list_empty(&mddev->disks)) {
5141 mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
5142 mdk_rdev_t, same_set);
5143 err = super_types[mddev->major_version]
5144 .load_super(rdev, rdev0, mddev->minor_version);
5147 "md: %s has different UUID to %s\n",
5148 bdevname(rdev->bdev,b),
5149 bdevname(rdev0->bdev,b2));
5154 err = bind_rdev_to_array(rdev, mddev);
5161 * add_new_disk can be used once the array is assembled
5162 * to add "hot spares". They must already have a superblock
5167 if (!mddev->pers->hot_add_disk) {
5169 "%s: personality does not support diskops!\n",
5173 if (mddev->persistent)
5174 rdev = md_import_device(dev, mddev->major_version,
5175 mddev->minor_version);
5177 rdev = md_import_device(dev, -1, -1);
5180 "md: md_import_device returned %ld\n",
5182 return PTR_ERR(rdev);
5184 /* set saved_raid_disk if appropriate */
5185 if (!mddev->persistent) {
5186 if (info->state & (1<<MD_DISK_SYNC) &&
5187 info->raid_disk < mddev->raid_disks) {
5188 rdev->raid_disk = info->raid_disk;
5189 set_bit(In_sync, &rdev->flags);
5191 rdev->raid_disk = -1;
5193 super_types[mddev->major_version].
5194 validate_super(mddev, rdev);
5195 if (test_bit(In_sync, &rdev->flags))
5196 rdev->saved_raid_disk = rdev->raid_disk;
5198 rdev->saved_raid_disk = -1;
5200 clear_bit(In_sync, &rdev->flags); /* just to be sure */
5201 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
5202 set_bit(WriteMostly, &rdev->flags);
5204 clear_bit(WriteMostly, &rdev->flags);
5206 rdev->raid_disk = -1;
5207 err = bind_rdev_to_array(rdev, mddev);
5208 if (!err && !mddev->pers->hot_remove_disk) {
5209 /* If there is hot_add_disk but no hot_remove_disk
5210 * then added disks for geometry changes,
5211 * and should be added immediately.
5213 super_types[mddev->major_version].
5214 validate_super(mddev, rdev);
5215 err = mddev->pers->hot_add_disk(mddev, rdev);
5217 unbind_rdev_from_array(rdev);
5222 sysfs_notify_dirent_safe(rdev->sysfs_state);
5224 md_update_sb(mddev, 1);
5225 if (mddev->degraded)
5226 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
5227 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5228 md_wakeup_thread(mddev->thread);
5232 /* otherwise, add_new_disk is only allowed
5233 * for major_version==0 superblocks
5235 if (mddev->major_version != 0) {
5236 printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n",
5241 if (!(info->state & (1<<MD_DISK_FAULTY))) {
5243 rdev = md_import_device(dev, -1, 0);
5246 "md: error, md_import_device() returned %ld\n",
5248 return PTR_ERR(rdev);
5250 rdev->desc_nr = info->number;
5251 if (info->raid_disk < mddev->raid_disks)
5252 rdev->raid_disk = info->raid_disk;
5254 rdev->raid_disk = -1;
5256 if (rdev->raid_disk < mddev->raid_disks)
5257 if (info->state & (1<<MD_DISK_SYNC))
5258 set_bit(In_sync, &rdev->flags);
5260 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
5261 set_bit(WriteMostly, &rdev->flags);
5263 if (!mddev->persistent) {
5264 printk(KERN_INFO "md: nonpersistent superblock ...\n");
5265 rdev->sb_start = i_size_read(rdev->bdev->bd_inode) / 512;
5267 rdev->sb_start = calc_dev_sboffset(rdev);
5268 rdev->sectors = rdev->sb_start;
5270 err = bind_rdev_to_array(rdev, mddev);
5280 static int hot_remove_disk(mddev_t * mddev, dev_t dev)
5282 char b[BDEVNAME_SIZE];
5285 rdev = find_rdev(mddev, dev);
5289 if (rdev->raid_disk >= 0)
5292 kick_rdev_from_array(rdev);
5293 md_update_sb(mddev, 1);
5294 md_new_event(mddev);
5298 printk(KERN_WARNING "md: cannot remove active disk %s from %s ...\n",
5299 bdevname(rdev->bdev,b), mdname(mddev));
5303 static int hot_add_disk(mddev_t * mddev, dev_t dev)
5305 char b[BDEVNAME_SIZE];
5312 if (mddev->major_version != 0) {
5313 printk(KERN_WARNING "%s: HOT_ADD may only be used with"
5314 " version-0 superblocks.\n",
5318 if (!mddev->pers->hot_add_disk) {
5320 "%s: personality does not support diskops!\n",
5325 rdev = md_import_device(dev, -1, 0);
5328 "md: error, md_import_device() returned %ld\n",
5333 if (mddev->persistent)
5334 rdev->sb_start = calc_dev_sboffset(rdev);
5336 rdev->sb_start = i_size_read(rdev->bdev->bd_inode) / 512;
5338 rdev->sectors = rdev->sb_start;
5340 if (test_bit(Faulty, &rdev->flags)) {
5342 "md: can not hot-add faulty %s disk to %s!\n",
5343 bdevname(rdev->bdev,b), mdname(mddev));
5347 clear_bit(In_sync, &rdev->flags);
5349 rdev->saved_raid_disk = -1;
5350 err = bind_rdev_to_array(rdev, mddev);
5355 * The rest should better be atomic, we can have disk failures
5356 * noticed in interrupt contexts ...
5359 rdev->raid_disk = -1;
5361 md_update_sb(mddev, 1);
5364 * Kick recovery, maybe this spare has to be added to the
5365 * array immediately.
5367 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5368 md_wakeup_thread(mddev->thread);
5369 md_new_event(mddev);
5377 static int set_bitmap_file(mddev_t *mddev, int fd)
5382 if (!mddev->pers->quiesce)
5384 if (mddev->recovery || mddev->sync_thread)
5386 /* we should be able to change the bitmap.. */
5392 return -EEXIST; /* cannot add when bitmap is present */
5393 mddev->bitmap_info.file = fget(fd);
5395 if (mddev->bitmap_info.file == NULL) {
5396 printk(KERN_ERR "%s: error: failed to get bitmap file\n",
5401 err = deny_bitmap_write_access(mddev->bitmap_info.file);
5403 printk(KERN_ERR "%s: error: bitmap file is already in use\n",
5405 fput(mddev->bitmap_info.file);
5406 mddev->bitmap_info.file = NULL;
5409 mddev->bitmap_info.offset = 0; /* file overrides offset */
5410 } else if (mddev->bitmap == NULL)
5411 return -ENOENT; /* cannot remove what isn't there */
5414 mddev->pers->quiesce(mddev, 1);
5416 err = bitmap_create(mddev);
5418 err = bitmap_load(mddev);
5420 if (fd < 0 || err) {
5421 bitmap_destroy(mddev);
5422 fd = -1; /* make sure to put the file */
5424 mddev->pers->quiesce(mddev, 0);
5427 if (mddev->bitmap_info.file) {
5428 restore_bitmap_write_access(mddev->bitmap_info.file);
5429 fput(mddev->bitmap_info.file);
5431 mddev->bitmap_info.file = NULL;
5438 * set_array_info is used two different ways
5439 * The original usage is when creating a new array.
5440 * In this usage, raid_disks is > 0 and it together with
5441 * level, size, not_persistent,layout,chunksize determine the
5442 * shape of the array.
5443 * This will always create an array with a type-0.90.0 superblock.
5444 * The newer usage is when assembling an array.
5445 * In this case raid_disks will be 0, and the major_version field is
5446 * use to determine which style super-blocks are to be found on the devices.
5447 * The minor and patch _version numbers are also kept incase the
5448 * super_block handler wishes to interpret them.
5450 static int set_array_info(mddev_t * mddev, mdu_array_info_t *info)
5453 if (info->raid_disks == 0) {
5454 /* just setting version number for superblock loading */
5455 if (info->major_version < 0 ||
5456 info->major_version >= ARRAY_SIZE(super_types) ||
5457 super_types[info->major_version].name == NULL) {
5458 /* maybe try to auto-load a module? */
5460 "md: superblock version %d not known\n",
5461 info->major_version);
5464 mddev->major_version = info->major_version;
5465 mddev->minor_version = info->minor_version;
5466 mddev->patch_version = info->patch_version;
5467 mddev->persistent = !info->not_persistent;
5468 /* ensure mddev_put doesn't delete this now that there
5469 * is some minimal configuration.
5471 mddev->ctime = get_seconds();
5474 mddev->major_version = MD_MAJOR_VERSION;
5475 mddev->minor_version = MD_MINOR_VERSION;
5476 mddev->patch_version = MD_PATCHLEVEL_VERSION;
5477 mddev->ctime = get_seconds();
5479 mddev->level = info->level;
5480 mddev->clevel[0] = 0;
5481 mddev->dev_sectors = 2 * (sector_t)info->size;
5482 mddev->raid_disks = info->raid_disks;
5483 /* don't set md_minor, it is determined by which /dev/md* was
5486 if (info->state & (1<<MD_SB_CLEAN))
5487 mddev->recovery_cp = MaxSector;
5489 mddev->recovery_cp = 0;
5490 mddev->persistent = ! info->not_persistent;
5491 mddev->external = 0;
5493 mddev->layout = info->layout;
5494 mddev->chunk_sectors = info->chunk_size >> 9;
5496 mddev->max_disks = MD_SB_DISKS;
5498 if (mddev->persistent)
5500 set_bit(MD_CHANGE_DEVS, &mddev->flags);
5502 mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9;
5503 mddev->bitmap_info.offset = 0;
5505 mddev->reshape_position = MaxSector;
5508 * Generate a 128 bit UUID
5510 get_random_bytes(mddev->uuid, 16);
5512 mddev->new_level = mddev->level;
5513 mddev->new_chunk_sectors = mddev->chunk_sectors;
5514 mddev->new_layout = mddev->layout;
5515 mddev->delta_disks = 0;
5520 void md_set_array_sectors(mddev_t *mddev, sector_t array_sectors)
5522 WARN(!mddev_is_locked(mddev), "%s: unlocked mddev!\n", __func__);
5524 if (mddev->external_size)
5527 mddev->array_sectors = array_sectors;
5529 EXPORT_SYMBOL(md_set_array_sectors);
5531 static int update_size(mddev_t *mddev, sector_t num_sectors)
5535 int fit = (num_sectors == 0);
5537 if (mddev->pers->resize == NULL)
5539 /* The "num_sectors" is the number of sectors of each device that
5540 * is used. This can only make sense for arrays with redundancy.
5541 * linear and raid0 always use whatever space is available. We can only
5542 * consider changing this number if no resync or reconstruction is
5543 * happening, and if the new size is acceptable. It must fit before the
5544 * sb_start or, if that is <data_offset, it must fit before the size
5545 * of each device. If num_sectors is zero, we find the largest size
5548 if (mddev->sync_thread)
5551 /* Sorry, cannot grow a bitmap yet, just remove it,
5555 list_for_each_entry(rdev, &mddev->disks, same_set) {
5556 sector_t avail = rdev->sectors;
5558 if (fit && (num_sectors == 0 || num_sectors > avail))
5559 num_sectors = avail;
5560 if (avail < num_sectors)
5563 rv = mddev->pers->resize(mddev, num_sectors);
5565 revalidate_disk(mddev->gendisk);
5569 static int update_raid_disks(mddev_t *mddev, int raid_disks)
5572 /* change the number of raid disks */
5573 if (mddev->pers->check_reshape == NULL)
5575 if (raid_disks <= 0 ||
5576 (mddev->max_disks && raid_disks >= mddev->max_disks))
5578 if (mddev->sync_thread || mddev->reshape_position != MaxSector)
5580 mddev->delta_disks = raid_disks - mddev->raid_disks;
5582 rv = mddev->pers->check_reshape(mddev);
5588 * update_array_info is used to change the configuration of an
5590 * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
5591 * fields in the info are checked against the array.
5592 * Any differences that cannot be handled will cause an error.
5593 * Normally, only one change can be managed at a time.
5595 static int update_array_info(mddev_t *mddev, mdu_array_info_t *info)
5601 /* calculate expected state,ignoring low bits */
5602 if (mddev->bitmap && mddev->bitmap_info.offset)
5603 state |= (1 << MD_SB_BITMAP_PRESENT);
5605 if (mddev->major_version != info->major_version ||
5606 mddev->minor_version != info->minor_version ||
5607 /* mddev->patch_version != info->patch_version || */
5608 mddev->ctime != info->ctime ||
5609 mddev->level != info->level ||
5610 /* mddev->layout != info->layout || */
5611 !mddev->persistent != info->not_persistent||
5612 mddev->chunk_sectors != info->chunk_size >> 9 ||
5613 /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
5614 ((state^info->state) & 0xfffffe00)
5617 /* Check there is only one change */
5618 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
5620 if (mddev->raid_disks != info->raid_disks)
5622 if (mddev->layout != info->layout)
5624 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT))
5631 if (mddev->layout != info->layout) {
5633 * we don't need to do anything at the md level, the
5634 * personality will take care of it all.
5636 if (mddev->pers->check_reshape == NULL)
5639 mddev->new_layout = info->layout;
5640 rv = mddev->pers->check_reshape(mddev);
5642 mddev->new_layout = mddev->layout;
5646 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
5647 rv = update_size(mddev, (sector_t)info->size * 2);
5649 if (mddev->raid_disks != info->raid_disks)
5650 rv = update_raid_disks(mddev, info->raid_disks);
5652 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) {
5653 if (mddev->pers->quiesce == NULL)
5655 if (mddev->recovery || mddev->sync_thread)
5657 if (info->state & (1<<MD_SB_BITMAP_PRESENT)) {
5658 /* add the bitmap */
5661 if (mddev->bitmap_info.default_offset == 0)
5663 mddev->bitmap_info.offset =
5664 mddev->bitmap_info.default_offset;
5665 mddev->pers->quiesce(mddev, 1);
5666 rv = bitmap_create(mddev);
5668 rv = bitmap_load(mddev);
5670 bitmap_destroy(mddev);
5671 mddev->pers->quiesce(mddev, 0);
5673 /* remove the bitmap */
5676 if (mddev->bitmap->file)
5678 mddev->pers->quiesce(mddev, 1);
5679 bitmap_destroy(mddev);
5680 mddev->pers->quiesce(mddev, 0);
5681 mddev->bitmap_info.offset = 0;
5684 md_update_sb(mddev, 1);
5688 static int set_disk_faulty(mddev_t *mddev, dev_t dev)
5692 if (mddev->pers == NULL)
5695 rdev = find_rdev(mddev, dev);
5699 md_error(mddev, rdev);
5704 * We have a problem here : there is no easy way to give a CHS
5705 * virtual geometry. We currently pretend that we have a 2 heads
5706 * 4 sectors (with a BIG number of cylinders...). This drives
5707 * dosfs just mad... ;-)
5709 static int md_getgeo(struct block_device *bdev, struct hd_geometry *geo)
5711 mddev_t *mddev = bdev->bd_disk->private_data;
5715 geo->cylinders = mddev->array_sectors / 8;
5719 static int md_ioctl(struct block_device *bdev, fmode_t mode,
5720 unsigned int cmd, unsigned long arg)
5723 void __user *argp = (void __user *)arg;
5724 mddev_t *mddev = NULL;
5727 if (!capable(CAP_SYS_ADMIN))
5731 * Commands dealing with the RAID driver but not any
5737 err = get_version(argp);
5740 case PRINT_RAID_DEBUG:
5748 autostart_arrays(arg);
5755 * Commands creating/starting a new array:
5758 mddev = bdev->bd_disk->private_data;
5765 err = mddev_lock(mddev);
5768 "md: ioctl lock interrupted, reason %d, cmd %d\n",
5775 case SET_ARRAY_INFO:
5777 mdu_array_info_t info;
5779 memset(&info, 0, sizeof(info));
5780 else if (copy_from_user(&info, argp, sizeof(info))) {
5785 err = update_array_info(mddev, &info);
5787 printk(KERN_WARNING "md: couldn't update"
5788 " array info. %d\n", err);
5793 if (!list_empty(&mddev->disks)) {
5795 "md: array %s already has disks!\n",
5800 if (mddev->raid_disks) {
5802 "md: array %s already initialised!\n",
5807 err = set_array_info(mddev, &info);
5809 printk(KERN_WARNING "md: couldn't set"
5810 " array info. %d\n", err);
5820 * Commands querying/configuring an existing array:
5822 /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
5823 * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */
5824 if ((!mddev->raid_disks && !mddev->external)
5825 && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY
5826 && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE
5827 && cmd != GET_BITMAP_FILE) {
5833 * Commands even a read-only array can execute:
5837 case GET_ARRAY_INFO:
5838 err = get_array_info(mddev, argp);
5841 case GET_BITMAP_FILE:
5842 err = get_bitmap_file(mddev, argp);
5846 err = get_disk_info(mddev, argp);
5849 case RESTART_ARRAY_RW:
5850 err = restart_array(mddev);
5854 err = do_md_stop(mddev, 0, 1);
5858 err = md_set_readonly(mddev, 1);
5862 if (get_user(ro, (int __user *)(arg))) {
5868 /* if the bdev is going readonly the value of mddev->ro
5869 * does not matter, no writes are coming
5874 /* are we are already prepared for writes? */
5878 /* transitioning to readauto need only happen for
5879 * arrays that call md_write_start
5882 err = restart_array(mddev);
5885 set_disk_ro(mddev->gendisk, 0);
5892 * The remaining ioctls are changing the state of the
5893 * superblock, so we do not allow them on read-only arrays.
5894 * However non-MD ioctls (e.g. get-size) will still come through
5895 * here and hit the 'default' below, so only disallow
5896 * 'md' ioctls, and switch to rw mode if started auto-readonly.
5898 if (_IOC_TYPE(cmd) == MD_MAJOR && mddev->ro && mddev->pers) {
5899 if (mddev->ro == 2) {
5901 sysfs_notify_dirent_safe(mddev->sysfs_state);
5902 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5903 md_wakeup_thread(mddev->thread);
5914 mdu_disk_info_t info;
5915 if (copy_from_user(&info, argp, sizeof(info)))
5918 err = add_new_disk(mddev, &info);
5922 case HOT_REMOVE_DISK:
5923 err = hot_remove_disk(mddev, new_decode_dev(arg));
5927 err = hot_add_disk(mddev, new_decode_dev(arg));
5930 case SET_DISK_FAULTY:
5931 err = set_disk_faulty(mddev, new_decode_dev(arg));
5935 err = do_md_run(mddev);
5938 case SET_BITMAP_FILE:
5939 err = set_bitmap_file(mddev, (int)arg);
5949 if (mddev->hold_active == UNTIL_IOCTL &&
5951 mddev->hold_active = 0;
5952 mddev_unlock(mddev);
5961 #ifdef CONFIG_COMPAT
5962 static int md_compat_ioctl(struct block_device *bdev, fmode_t mode,
5963 unsigned int cmd, unsigned long arg)
5966 case HOT_REMOVE_DISK:
5968 case SET_DISK_FAULTY:
5969 case SET_BITMAP_FILE:
5970 /* These take in integer arg, do not convert */
5973 arg = (unsigned long)compat_ptr(arg);
5977 return md_ioctl(bdev, mode, cmd, arg);
5979 #endif /* CONFIG_COMPAT */
5981 static int md_open(struct block_device *bdev, fmode_t mode)
5984 * Succeed if we can lock the mddev, which confirms that
5985 * it isn't being stopped right now.
5987 mddev_t *mddev = mddev_find(bdev->bd_dev);
5990 if (mddev->gendisk != bdev->bd_disk) {
5991 /* we are racing with mddev_put which is discarding this
5995 /* Wait until bdev->bd_disk is definitely gone */
5996 flush_workqueue(md_misc_wq);
5997 /* Then retry the open from the top */
5998 return -ERESTARTSYS;
6000 BUG_ON(mddev != bdev->bd_disk->private_data);
6002 if ((err = mutex_lock_interruptible(&mddev->open_mutex)))
6006 atomic_inc(&mddev->openers);
6007 mutex_unlock(&mddev->open_mutex);
6009 check_disk_size_change(mddev->gendisk, bdev);
6014 static int md_release(struct gendisk *disk, fmode_t mode)
6016 mddev_t *mddev = disk->private_data;
6019 atomic_dec(&mddev->openers);
6024 static const struct block_device_operations md_fops =
6026 .owner = THIS_MODULE,
6028 .release = md_release,
6030 #ifdef CONFIG_COMPAT
6031 .compat_ioctl = md_compat_ioctl,
6033 .getgeo = md_getgeo,
6036 static int md_thread(void * arg)
6038 mdk_thread_t *thread = arg;
6041 * md_thread is a 'system-thread', it's priority should be very
6042 * high. We avoid resource deadlocks individually in each
6043 * raid personality. (RAID5 does preallocation) We also use RR and
6044 * the very same RT priority as kswapd, thus we will never get
6045 * into a priority inversion deadlock.
6047 * we definitely have to have equal or higher priority than
6048 * bdflush, otherwise bdflush will deadlock if there are too
6049 * many dirty RAID5 blocks.
6052 allow_signal(SIGKILL);
6053 while (!kthread_should_stop()) {
6055 /* We need to wait INTERRUPTIBLE so that
6056 * we don't add to the load-average.
6057 * That means we need to be sure no signals are
6060 if (signal_pending(current))
6061 flush_signals(current);
6063 wait_event_interruptible_timeout
6065 test_bit(THREAD_WAKEUP, &thread->flags)
6066 || kthread_should_stop(),
6069 clear_bit(THREAD_WAKEUP, &thread->flags);
6070 if (!kthread_should_stop())
6071 thread->run(thread->mddev);
6077 void md_wakeup_thread(mdk_thread_t *thread)
6080 dprintk("md: waking up MD thread %s.\n", thread->tsk->comm);
6081 set_bit(THREAD_WAKEUP, &thread->flags);
6082 wake_up(&thread->wqueue);
6086 mdk_thread_t *md_register_thread(void (*run) (mddev_t *), mddev_t *mddev,
6089 mdk_thread_t *thread;
6091 thread = kzalloc(sizeof(mdk_thread_t), GFP_KERNEL);
6095 init_waitqueue_head(&thread->wqueue);
6098 thread->mddev = mddev;
6099 thread->timeout = MAX_SCHEDULE_TIMEOUT;
6100 thread->tsk = kthread_run(md_thread, thread,
6102 mdname(thread->mddev),
6103 name ?: mddev->pers->name);
6104 if (IS_ERR(thread->tsk)) {
6111 void md_unregister_thread(mdk_thread_t *thread)
6115 dprintk("interrupting MD-thread pid %d\n", task_pid_nr(thread->tsk));
6117 kthread_stop(thread->tsk);
6121 void md_error(mddev_t *mddev, mdk_rdev_t *rdev)
6128 if (!rdev || test_bit(Faulty, &rdev->flags))
6131 if (mddev->external)
6132 set_bit(Blocked, &rdev->flags);
6134 dprintk("md_error dev:%s, rdev:(%d:%d), (caller: %p,%p,%p,%p).\n",
6136 MAJOR(rdev->bdev->bd_dev), MINOR(rdev->bdev->bd_dev),
6137 __builtin_return_address(0),__builtin_return_address(1),
6138 __builtin_return_address(2),__builtin_return_address(3));
6142 if (!mddev->pers->error_handler)
6144 mddev->pers->error_handler(mddev,rdev);
6145 if (mddev->degraded)
6146 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
6147 sysfs_notify_dirent_safe(rdev->sysfs_state);
6148 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6149 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6150 md_wakeup_thread(mddev->thread);
6151 if (mddev->event_work.func)
6152 queue_work(md_misc_wq, &mddev->event_work);
6153 md_new_event_inintr(mddev);
6156 /* seq_file implementation /proc/mdstat */
6158 static void status_unused(struct seq_file *seq)
6163 seq_printf(seq, "unused devices: ");
6165 list_for_each_entry(rdev, &pending_raid_disks, same_set) {
6166 char b[BDEVNAME_SIZE];
6168 seq_printf(seq, "%s ",
6169 bdevname(rdev->bdev,b));
6172 seq_printf(seq, "<none>");
6174 seq_printf(seq, "\n");
6178 static void status_resync(struct seq_file *seq, mddev_t * mddev)
6180 sector_t max_sectors, resync, res;
6181 unsigned long dt, db;
6184 unsigned int per_milli;
6186 resync = mddev->curr_resync - atomic_read(&mddev->recovery_active);
6188 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
6189 max_sectors = mddev->resync_max_sectors;
6191 max_sectors = mddev->dev_sectors;
6194 * Should not happen.
6200 /* Pick 'scale' such that (resync>>scale)*1000 will fit
6201 * in a sector_t, and (max_sectors>>scale) will fit in a
6202 * u32, as those are the requirements for sector_div.
6203 * Thus 'scale' must be at least 10
6206 if (sizeof(sector_t) > sizeof(unsigned long)) {
6207 while ( max_sectors/2 > (1ULL<<(scale+32)))
6210 res = (resync>>scale)*1000;
6211 sector_div(res, (u32)((max_sectors>>scale)+1));
6215 int i, x = per_milli/50, y = 20-x;
6216 seq_printf(seq, "[");
6217 for (i = 0; i < x; i++)
6218 seq_printf(seq, "=");
6219 seq_printf(seq, ">");
6220 for (i = 0; i < y; i++)
6221 seq_printf(seq, ".");
6222 seq_printf(seq, "] ");
6224 seq_printf(seq, " %s =%3u.%u%% (%llu/%llu)",
6225 (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)?
6227 (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)?
6229 (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
6230 "resync" : "recovery"))),
6231 per_milli/10, per_milli % 10,
6232 (unsigned long long) resync/2,
6233 (unsigned long long) max_sectors/2);
6236 * dt: time from mark until now
6237 * db: blocks written from mark until now
6238 * rt: remaining time
6240 * rt is a sector_t, so could be 32bit or 64bit.
6241 * So we divide before multiply in case it is 32bit and close
6243 * We scale the divisor (db) by 32 to avoid loosing precision
6244 * near the end of resync when the number of remaining sectors
6246 * We then divide rt by 32 after multiplying by db to compensate.
6247 * The '+1' avoids division by zero if db is very small.
6249 dt = ((jiffies - mddev->resync_mark) / HZ);
6251 db = (mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active))
6252 - mddev->resync_mark_cnt;
6254 rt = max_sectors - resync; /* number of remaining sectors */
6255 sector_div(rt, db/32+1);
6259 seq_printf(seq, " finish=%lu.%lumin", (unsigned long)rt / 60,
6260 ((unsigned long)rt % 60)/6);
6262 seq_printf(seq, " speed=%ldK/sec", db/2/dt);
6265 static void *md_seq_start(struct seq_file *seq, loff_t *pos)
6267 struct list_head *tmp;
6277 spin_lock(&all_mddevs_lock);
6278 list_for_each(tmp,&all_mddevs)
6280 mddev = list_entry(tmp, mddev_t, all_mddevs);
6282 spin_unlock(&all_mddevs_lock);
6285 spin_unlock(&all_mddevs_lock);
6287 return (void*)2;/* tail */
6291 static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
6293 struct list_head *tmp;
6294 mddev_t *next_mddev, *mddev = v;
6300 spin_lock(&all_mddevs_lock);
6302 tmp = all_mddevs.next;
6304 tmp = mddev->all_mddevs.next;
6305 if (tmp != &all_mddevs)
6306 next_mddev = mddev_get(list_entry(tmp,mddev_t,all_mddevs));
6308 next_mddev = (void*)2;
6311 spin_unlock(&all_mddevs_lock);
6319 static void md_seq_stop(struct seq_file *seq, void *v)
6323 if (mddev && v != (void*)1 && v != (void*)2)
6327 struct mdstat_info {
6331 static int md_seq_show(struct seq_file *seq, void *v)
6336 struct mdstat_info *mi = seq->private;
6337 struct bitmap *bitmap;
6339 if (v == (void*)1) {
6340 struct mdk_personality *pers;
6341 seq_printf(seq, "Personalities : ");
6342 spin_lock(&pers_lock);
6343 list_for_each_entry(pers, &pers_list, list)
6344 seq_printf(seq, "[%s] ", pers->name);
6346 spin_unlock(&pers_lock);
6347 seq_printf(seq, "\n");
6348 mi->event = atomic_read(&md_event_count);
6351 if (v == (void*)2) {
6356 if (mddev_lock(mddev) < 0)
6359 if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
6360 seq_printf(seq, "%s : %sactive", mdname(mddev),
6361 mddev->pers ? "" : "in");
6364 seq_printf(seq, " (read-only)");
6366 seq_printf(seq, " (auto-read-only)");
6367 seq_printf(seq, " %s", mddev->pers->name);
6371 list_for_each_entry(rdev, &mddev->disks, same_set) {
6372 char b[BDEVNAME_SIZE];
6373 seq_printf(seq, " %s[%d]",
6374 bdevname(rdev->bdev,b), rdev->desc_nr);
6375 if (test_bit(WriteMostly, &rdev->flags))
6376 seq_printf(seq, "(W)");
6377 if (test_bit(Faulty, &rdev->flags)) {
6378 seq_printf(seq, "(F)");
6380 } else if (rdev->raid_disk < 0)
6381 seq_printf(seq, "(S)"); /* spare */
6382 sectors += rdev->sectors;
6385 if (!list_empty(&mddev->disks)) {
6387 seq_printf(seq, "\n %llu blocks",
6388 (unsigned long long)
6389 mddev->array_sectors / 2);
6391 seq_printf(seq, "\n %llu blocks",
6392 (unsigned long long)sectors / 2);
6394 if (mddev->persistent) {
6395 if (mddev->major_version != 0 ||
6396 mddev->minor_version != 90) {
6397 seq_printf(seq," super %d.%d",
6398 mddev->major_version,
6399 mddev->minor_version);
6401 } else if (mddev->external)
6402 seq_printf(seq, " super external:%s",
6403 mddev->metadata_type);
6405 seq_printf(seq, " super non-persistent");
6408 mddev->pers->status(seq, mddev);
6409 seq_printf(seq, "\n ");
6410 if (mddev->pers->sync_request) {
6411 if (mddev->curr_resync > 2) {
6412 status_resync(seq, mddev);
6413 seq_printf(seq, "\n ");
6414 } else if (mddev->curr_resync == 1 || mddev->curr_resync == 2)
6415 seq_printf(seq, "\tresync=DELAYED\n ");
6416 else if (mddev->recovery_cp < MaxSector)
6417 seq_printf(seq, "\tresync=PENDING\n ");
6420 seq_printf(seq, "\n ");
6422 if ((bitmap = mddev->bitmap)) {
6423 unsigned long chunk_kb;
6424 unsigned long flags;
6425 spin_lock_irqsave(&bitmap->lock, flags);
6426 chunk_kb = mddev->bitmap_info.chunksize >> 10;
6427 seq_printf(seq, "bitmap: %lu/%lu pages [%luKB], "
6429 bitmap->pages - bitmap->missing_pages,
6431 (bitmap->pages - bitmap->missing_pages)
6432 << (PAGE_SHIFT - 10),
6433 chunk_kb ? chunk_kb : mddev->bitmap_info.chunksize,
6434 chunk_kb ? "KB" : "B");
6436 seq_printf(seq, ", file: ");
6437 seq_path(seq, &bitmap->file->f_path, " \t\n");
6440 seq_printf(seq, "\n");
6441 spin_unlock_irqrestore(&bitmap->lock, flags);
6444 seq_printf(seq, "\n");
6446 mddev_unlock(mddev);
6451 static const struct seq_operations md_seq_ops = {
6452 .start = md_seq_start,
6453 .next = md_seq_next,
6454 .stop = md_seq_stop,
6455 .show = md_seq_show,
6458 static int md_seq_open(struct inode *inode, struct file *file)
6461 struct mdstat_info *mi = kmalloc(sizeof(*mi), GFP_KERNEL);
6465 error = seq_open(file, &md_seq_ops);
6469 struct seq_file *p = file->private_data;
6471 mi->event = atomic_read(&md_event_count);
6476 static unsigned int mdstat_poll(struct file *filp, poll_table *wait)
6478 struct seq_file *m = filp->private_data;
6479 struct mdstat_info *mi = m->private;
6482 poll_wait(filp, &md_event_waiters, wait);
6484 /* always allow read */
6485 mask = POLLIN | POLLRDNORM;
6487 if (mi->event != atomic_read(&md_event_count))
6488 mask |= POLLERR | POLLPRI;
6492 static const struct file_operations md_seq_fops = {
6493 .owner = THIS_MODULE,
6494 .open = md_seq_open,
6496 .llseek = seq_lseek,
6497 .release = seq_release_private,
6498 .poll = mdstat_poll,
6501 int register_md_personality(struct mdk_personality *p)
6503 spin_lock(&pers_lock);
6504 list_add_tail(&p->list, &pers_list);
6505 printk(KERN_INFO "md: %s personality registered for level %d\n", p->name, p->level);
6506 spin_unlock(&pers_lock);
6510 int unregister_md_personality(struct mdk_personality *p)
6512 printk(KERN_INFO "md: %s personality unregistered\n", p->name);
6513 spin_lock(&pers_lock);
6514 list_del_init(&p->list);
6515 spin_unlock(&pers_lock);
6519 static int is_mddev_idle(mddev_t *mddev, int init)
6527 rdev_for_each_rcu(rdev, mddev) {
6528 struct gendisk *disk = rdev->bdev->bd_contains->bd_disk;
6529 curr_events = (int)part_stat_read(&disk->part0, sectors[0]) +
6530 (int)part_stat_read(&disk->part0, sectors[1]) -
6531 atomic_read(&disk->sync_io);
6532 /* sync IO will cause sync_io to increase before the disk_stats
6533 * as sync_io is counted when a request starts, and
6534 * disk_stats is counted when it completes.
6535 * So resync activity will cause curr_events to be smaller than
6536 * when there was no such activity.
6537 * non-sync IO will cause disk_stat to increase without
6538 * increasing sync_io so curr_events will (eventually)
6539 * be larger than it was before. Once it becomes
6540 * substantially larger, the test below will cause
6541 * the array to appear non-idle, and resync will slow
6543 * If there is a lot of outstanding resync activity when
6544 * we set last_event to curr_events, then all that activity
6545 * completing might cause the array to appear non-idle
6546 * and resync will be slowed down even though there might
6547 * not have been non-resync activity. This will only
6548 * happen once though. 'last_events' will soon reflect
6549 * the state where there is little or no outstanding
6550 * resync requests, and further resync activity will
6551 * always make curr_events less than last_events.
6554 if (init || curr_events - rdev->last_events > 64) {
6555 rdev->last_events = curr_events;
6563 void md_done_sync(mddev_t *mddev, int blocks, int ok)
6565 /* another "blocks" (512byte) blocks have been synced */
6566 atomic_sub(blocks, &mddev->recovery_active);
6567 wake_up(&mddev->recovery_wait);
6569 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6570 md_wakeup_thread(mddev->thread);
6571 // stop recovery, signal do_sync ....
6576 /* md_write_start(mddev, bi)
6577 * If we need to update some array metadata (e.g. 'active' flag
6578 * in superblock) before writing, schedule a superblock update
6579 * and wait for it to complete.
6581 void md_write_start(mddev_t *mddev, struct bio *bi)
6584 if (bio_data_dir(bi) != WRITE)
6587 BUG_ON(mddev->ro == 1);
6588 if (mddev->ro == 2) {
6589 /* need to switch to read/write */
6591 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6592 md_wakeup_thread(mddev->thread);
6593 md_wakeup_thread(mddev->sync_thread);
6596 atomic_inc(&mddev->writes_pending);
6597 if (mddev->safemode == 1)
6598 mddev->safemode = 0;
6599 if (mddev->in_sync) {
6600 spin_lock_irq(&mddev->write_lock);
6601 if (mddev->in_sync) {
6603 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
6604 set_bit(MD_CHANGE_PENDING, &mddev->flags);
6605 md_wakeup_thread(mddev->thread);
6608 spin_unlock_irq(&mddev->write_lock);
6611 sysfs_notify_dirent_safe(mddev->sysfs_state);
6612 wait_event(mddev->sb_wait,
6613 !test_bit(MD_CHANGE_PENDING, &mddev->flags));
6616 void md_write_end(mddev_t *mddev)
6618 if (atomic_dec_and_test(&mddev->writes_pending)) {
6619 if (mddev->safemode == 2)
6620 md_wakeup_thread(mddev->thread);
6621 else if (mddev->safemode_delay)
6622 mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay);
6626 /* md_allow_write(mddev)
6627 * Calling this ensures that the array is marked 'active' so that writes
6628 * may proceed without blocking. It is important to call this before
6629 * attempting a GFP_KERNEL allocation while holding the mddev lock.
6630 * Must be called with mddev_lock held.
6632 * In the ->external case MD_CHANGE_CLEAN can not be cleared until mddev->lock
6633 * is dropped, so return -EAGAIN after notifying userspace.
6635 int md_allow_write(mddev_t *mddev)
6641 if (!mddev->pers->sync_request)
6644 spin_lock_irq(&mddev->write_lock);
6645 if (mddev->in_sync) {
6647 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
6648 set_bit(MD_CHANGE_PENDING, &mddev->flags);
6649 if (mddev->safemode_delay &&
6650 mddev->safemode == 0)
6651 mddev->safemode = 1;
6652 spin_unlock_irq(&mddev->write_lock);
6653 md_update_sb(mddev, 0);
6654 sysfs_notify_dirent_safe(mddev->sysfs_state);
6656 spin_unlock_irq(&mddev->write_lock);
6658 if (test_bit(MD_CHANGE_PENDING, &mddev->flags))
6663 EXPORT_SYMBOL_GPL(md_allow_write);
6665 void md_unplug(mddev_t *mddev)
6668 blk_unplug(mddev->queue);
6670 mddev->plug->unplug_fn(mddev->plug);
6673 #define SYNC_MARKS 10
6674 #define SYNC_MARK_STEP (3*HZ)
6675 void md_do_sync(mddev_t *mddev)
6678 unsigned int currspeed = 0,
6680 sector_t max_sectors,j, io_sectors;
6681 unsigned long mark[SYNC_MARKS];
6682 sector_t mark_cnt[SYNC_MARKS];
6684 struct list_head *tmp;
6685 sector_t last_check;
6690 /* just incase thread restarts... */
6691 if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
6693 if (mddev->ro) /* never try to sync a read-only array */
6696 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
6697 if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
6698 desc = "data-check";
6699 else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
6700 desc = "requested-resync";
6703 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
6708 /* we overload curr_resync somewhat here.
6709 * 0 == not engaged in resync at all
6710 * 2 == checking that there is no conflict with another sync
6711 * 1 == like 2, but have yielded to allow conflicting resync to
6713 * other == active in resync - this many blocks
6715 * Before starting a resync we must have set curr_resync to
6716 * 2, and then checked that every "conflicting" array has curr_resync
6717 * less than ours. When we find one that is the same or higher
6718 * we wait on resync_wait. To avoid deadlock, we reduce curr_resync
6719 * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
6720 * This will mean we have to start checking from the beginning again.
6725 mddev->curr_resync = 2;
6728 if (kthread_should_stop())
6729 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6731 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
6733 for_each_mddev(mddev2, tmp) {
6734 if (mddev2 == mddev)
6736 if (!mddev->parallel_resync
6737 && mddev2->curr_resync
6738 && match_mddev_units(mddev, mddev2)) {
6740 if (mddev < mddev2 && mddev->curr_resync == 2) {
6741 /* arbitrarily yield */
6742 mddev->curr_resync = 1;
6743 wake_up(&resync_wait);
6745 if (mddev > mddev2 && mddev->curr_resync == 1)
6746 /* no need to wait here, we can wait the next
6747 * time 'round when curr_resync == 2
6750 /* We need to wait 'interruptible' so as not to
6751 * contribute to the load average, and not to
6752 * be caught by 'softlockup'
6754 prepare_to_wait(&resync_wait, &wq, TASK_INTERRUPTIBLE);
6755 if (!kthread_should_stop() &&
6756 mddev2->curr_resync >= mddev->curr_resync) {
6757 printk(KERN_INFO "md: delaying %s of %s"
6758 " until %s has finished (they"
6759 " share one or more physical units)\n",
6760 desc, mdname(mddev), mdname(mddev2));
6762 if (signal_pending(current))
6763 flush_signals(current);
6765 finish_wait(&resync_wait, &wq);
6768 finish_wait(&resync_wait, &wq);
6771 } while (mddev->curr_resync < 2);
6774 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
6775 /* resync follows the size requested by the personality,
6776 * which defaults to physical size, but can be virtual size
6778 max_sectors = mddev->resync_max_sectors;
6779 mddev->resync_mismatches = 0;
6780 /* we don't use the checkpoint if there's a bitmap */
6781 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
6782 j = mddev->resync_min;
6783 else if (!mddev->bitmap)
6784 j = mddev->recovery_cp;
6786 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
6787 max_sectors = mddev->dev_sectors;
6789 /* recovery follows the physical size of devices */
6790 max_sectors = mddev->dev_sectors;
6793 list_for_each_entry_rcu(rdev, &mddev->disks, same_set)
6794 if (rdev->raid_disk >= 0 &&
6795 !test_bit(Faulty, &rdev->flags) &&
6796 !test_bit(In_sync, &rdev->flags) &&
6797 rdev->recovery_offset < j)
6798 j = rdev->recovery_offset;
6802 printk(KERN_INFO "md: %s of RAID array %s\n", desc, mdname(mddev));
6803 printk(KERN_INFO "md: minimum _guaranteed_ speed:"
6804 " %d KB/sec/disk.\n", speed_min(mddev));
6805 printk(KERN_INFO "md: using maximum available idle IO bandwidth "
6806 "(but not more than %d KB/sec) for %s.\n",
6807 speed_max(mddev), desc);
6809 is_mddev_idle(mddev, 1); /* this initializes IO event counters */
6812 for (m = 0; m < SYNC_MARKS; m++) {
6814 mark_cnt[m] = io_sectors;
6817 mddev->resync_mark = mark[last_mark];
6818 mddev->resync_mark_cnt = mark_cnt[last_mark];
6821 * Tune reconstruction:
6823 window = 32*(PAGE_SIZE/512);
6824 printk(KERN_INFO "md: using %dk window, over a total of %llu blocks.\n",
6825 window/2,(unsigned long long) max_sectors/2);
6827 atomic_set(&mddev->recovery_active, 0);
6832 "md: resuming %s of %s from checkpoint.\n",
6833 desc, mdname(mddev));
6834 mddev->curr_resync = j;
6836 mddev->curr_resync_completed = j;
6838 while (j < max_sectors) {
6843 if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
6844 ((mddev->curr_resync > mddev->curr_resync_completed &&
6845 (mddev->curr_resync - mddev->curr_resync_completed)
6846 > (max_sectors >> 4)) ||
6847 (j - mddev->curr_resync_completed)*2
6848 >= mddev->resync_max - mddev->curr_resync_completed
6850 /* time to update curr_resync_completed */
6852 wait_event(mddev->recovery_wait,
6853 atomic_read(&mddev->recovery_active) == 0);
6854 mddev->curr_resync_completed = j;
6855 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
6856 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
6859 while (j >= mddev->resync_max && !kthread_should_stop()) {
6860 /* As this condition is controlled by user-space,
6861 * we can block indefinitely, so use '_interruptible'
6862 * to avoid triggering warnings.
6864 flush_signals(current); /* just in case */
6865 wait_event_interruptible(mddev->recovery_wait,
6866 mddev->resync_max > j
6867 || kthread_should_stop());
6870 if (kthread_should_stop())
6873 sectors = mddev->pers->sync_request(mddev, j, &skipped,
6874 currspeed < speed_min(mddev));
6876 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6880 if (!skipped) { /* actual IO requested */
6881 io_sectors += sectors;
6882 atomic_add(sectors, &mddev->recovery_active);
6886 if (j>1) mddev->curr_resync = j;
6887 mddev->curr_mark_cnt = io_sectors;
6888 if (last_check == 0)
6889 /* this is the earliers that rebuilt will be
6890 * visible in /proc/mdstat
6892 md_new_event(mddev);
6894 if (last_check + window > io_sectors || j == max_sectors)
6897 last_check = io_sectors;
6899 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
6903 if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) {
6905 int next = (last_mark+1) % SYNC_MARKS;
6907 mddev->resync_mark = mark[next];
6908 mddev->resync_mark_cnt = mark_cnt[next];
6909 mark[next] = jiffies;
6910 mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active);
6915 if (kthread_should_stop())
6920 * this loop exits only if either when we are slower than
6921 * the 'hard' speed limit, or the system was IO-idle for
6923 * the system might be non-idle CPU-wise, but we only care
6924 * about not overloading the IO subsystem. (things like an
6925 * e2fsck being done on the RAID array should execute fast)
6930 currspeed = ((unsigned long)(io_sectors-mddev->resync_mark_cnt))/2
6931 /((jiffies-mddev->resync_mark)/HZ +1) +1;
6933 if (currspeed > speed_min(mddev)) {
6934 if ((currspeed > speed_max(mddev)) ||
6935 !is_mddev_idle(mddev, 0)) {
6941 printk(KERN_INFO "md: %s: %s done.\n",mdname(mddev), desc);
6943 * this also signals 'finished resyncing' to md_stop
6948 wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
6950 /* tell personality that we are finished */
6951 mddev->pers->sync_request(mddev, max_sectors, &skipped, 1);
6953 if (!test_bit(MD_RECOVERY_CHECK, &mddev->recovery) &&
6954 mddev->curr_resync > 2) {
6955 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
6956 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
6957 if (mddev->curr_resync >= mddev->recovery_cp) {
6959 "md: checkpointing %s of %s.\n",
6960 desc, mdname(mddev));
6961 mddev->recovery_cp = mddev->curr_resync;
6964 mddev->recovery_cp = MaxSector;
6966 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery))
6967 mddev->curr_resync = MaxSector;
6969 list_for_each_entry_rcu(rdev, &mddev->disks, same_set)
6970 if (rdev->raid_disk >= 0 &&
6971 mddev->delta_disks >= 0 &&
6972 !test_bit(Faulty, &rdev->flags) &&
6973 !test_bit(In_sync, &rdev->flags) &&
6974 rdev->recovery_offset < mddev->curr_resync)
6975 rdev->recovery_offset = mddev->curr_resync;
6979 set_bit(MD_CHANGE_DEVS, &mddev->flags);
6982 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
6983 /* We completed so min/max setting can be forgotten if used. */
6984 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
6985 mddev->resync_min = 0;
6986 mddev->resync_max = MaxSector;
6987 } else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
6988 mddev->resync_min = mddev->curr_resync_completed;
6989 mddev->curr_resync = 0;
6990 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery))
6991 mddev->curr_resync_completed = 0;
6992 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
6993 wake_up(&resync_wait);
6994 set_bit(MD_RECOVERY_DONE, &mddev->recovery);
6995 md_wakeup_thread(mddev->thread);
7000 * got a signal, exit.
7003 "md: md_do_sync() got signal ... exiting\n");
7004 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
7008 EXPORT_SYMBOL_GPL(md_do_sync);
7011 static int remove_and_add_spares(mddev_t *mddev)
7016 mddev->curr_resync_completed = 0;
7018 list_for_each_entry(rdev, &mddev->disks, same_set)
7019 if (rdev->raid_disk >= 0 &&
7020 !test_bit(Blocked, &rdev->flags) &&
7021 (test_bit(Faulty, &rdev->flags) ||
7022 ! test_bit(In_sync, &rdev->flags)) &&
7023 atomic_read(&rdev->nr_pending)==0) {
7024 if (mddev->pers->hot_remove_disk(
7025 mddev, rdev->raid_disk)==0) {
7027 sprintf(nm,"rd%d", rdev->raid_disk);
7028 sysfs_remove_link(&mddev->kobj, nm);
7029 rdev->raid_disk = -1;
7033 if (mddev->degraded && ! mddev->ro && !mddev->recovery_disabled) {
7034 list_for_each_entry(rdev, &mddev->disks, same_set) {
7035 if (rdev->raid_disk >= 0 &&
7036 !test_bit(In_sync, &rdev->flags) &&
7037 !test_bit(Blocked, &rdev->flags))
7039 if (rdev->raid_disk < 0
7040 && !test_bit(Faulty, &rdev->flags)) {
7041 rdev->recovery_offset = 0;
7043 hot_add_disk(mddev, rdev) == 0) {
7045 sprintf(nm, "rd%d", rdev->raid_disk);
7046 if (sysfs_create_link(&mddev->kobj,
7048 /* failure here is OK */;
7050 md_new_event(mddev);
7051 set_bit(MD_CHANGE_DEVS, &mddev->flags);
7060 static void reap_sync_thread(mddev_t *mddev)
7064 /* resync has finished, collect result */
7065 md_unregister_thread(mddev->sync_thread);
7066 mddev->sync_thread = NULL;
7067 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
7068 !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
7070 /* activate any spares */
7071 if (mddev->pers->spare_active(mddev))
7072 sysfs_notify(&mddev->kobj, NULL,
7075 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
7076 mddev->pers->finish_reshape)
7077 mddev->pers->finish_reshape(mddev);
7078 md_update_sb(mddev, 1);
7080 /* if array is no-longer degraded, then any saved_raid_disk
7081 * information must be scrapped
7083 if (!mddev->degraded)
7084 list_for_each_entry(rdev, &mddev->disks, same_set)
7085 rdev->saved_raid_disk = -1;
7087 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
7088 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
7089 clear_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
7090 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
7091 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
7092 /* flag recovery needed just to double check */
7093 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7094 sysfs_notify_dirent_safe(mddev->sysfs_action);
7095 md_new_event(mddev);
7099 * This routine is regularly called by all per-raid-array threads to
7100 * deal with generic issues like resync and super-block update.
7101 * Raid personalities that don't have a thread (linear/raid0) do not
7102 * need this as they never do any recovery or update the superblock.
7104 * It does not do any resync itself, but rather "forks" off other threads
7105 * to do that as needed.
7106 * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
7107 * "->recovery" and create a thread at ->sync_thread.
7108 * When the thread finishes it sets MD_RECOVERY_DONE
7109 * and wakeups up this thread which will reap the thread and finish up.
7110 * This thread also removes any faulty devices (with nr_pending == 0).
7112 * The overall approach is:
7113 * 1/ if the superblock needs updating, update it.
7114 * 2/ If a recovery thread is running, don't do anything else.
7115 * 3/ If recovery has finished, clean up, possibly marking spares active.
7116 * 4/ If there are any faulty devices, remove them.
7117 * 5/ If array is degraded, try to add spares devices
7118 * 6/ If array has spares or is not in-sync, start a resync thread.
7120 void md_check_recovery(mddev_t *mddev)
7123 bitmap_daemon_work(mddev);
7128 if (signal_pending(current)) {
7129 if (mddev->pers->sync_request && !mddev->external) {
7130 printk(KERN_INFO "md: %s in immediate safe mode\n",
7132 mddev->safemode = 2;
7134 flush_signals(current);
7137 if (mddev->ro && !test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
7140 (mddev->flags & ~ (1<<MD_CHANGE_PENDING)) ||
7141 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
7142 test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
7143 (mddev->external == 0 && mddev->safemode == 1) ||
7144 (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending)
7145 && !mddev->in_sync && mddev->recovery_cp == MaxSector)
7149 if (mddev_trylock(mddev)) {
7153 /* Only thing we do on a ro array is remove
7156 remove_and_add_spares(mddev);
7157 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7161 if (!mddev->external) {
7163 spin_lock_irq(&mddev->write_lock);
7164 if (mddev->safemode &&
7165 !atomic_read(&mddev->writes_pending) &&
7167 mddev->recovery_cp == MaxSector) {
7170 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
7172 if (mddev->safemode == 1)
7173 mddev->safemode = 0;
7174 spin_unlock_irq(&mddev->write_lock);
7176 sysfs_notify_dirent_safe(mddev->sysfs_state);
7180 md_update_sb(mddev, 0);
7182 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
7183 !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
7184 /* resync/recovery still happening */
7185 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7188 if (mddev->sync_thread) {
7189 reap_sync_thread(mddev);
7192 /* Set RUNNING before clearing NEEDED to avoid
7193 * any transients in the value of "sync_action".
7195 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
7196 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7197 /* Clear some bits that don't mean anything, but
7200 clear_bit(MD_RECOVERY_INTR, &mddev->recovery);
7201 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
7203 if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
7205 /* no recovery is running.
7206 * remove any failed drives, then
7207 * add spares if possible.
7208 * Spare are also removed and re-added, to allow
7209 * the personality to fail the re-add.
7212 if (mddev->reshape_position != MaxSector) {
7213 if (mddev->pers->check_reshape == NULL ||
7214 mddev->pers->check_reshape(mddev) != 0)
7215 /* Cannot proceed */
7217 set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
7218 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
7219 } else if ((spares = remove_and_add_spares(mddev))) {
7220 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
7221 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
7222 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
7223 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
7224 } else if (mddev->recovery_cp < MaxSector) {
7225 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
7226 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
7227 } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
7228 /* nothing to be done ... */
7231 if (mddev->pers->sync_request) {
7232 if (spares && mddev->bitmap && ! mddev->bitmap->file) {
7233 /* We are adding a device or devices to an array
7234 * which has the bitmap stored on all devices.
7235 * So make sure all bitmap pages get written
7237 bitmap_write_all(mddev->bitmap);
7239 mddev->sync_thread = md_register_thread(md_do_sync,
7242 if (!mddev->sync_thread) {
7243 printk(KERN_ERR "%s: could not start resync"
7246 /* leave the spares where they are, it shouldn't hurt */
7247 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
7248 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
7249 clear_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
7250 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
7251 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
7253 md_wakeup_thread(mddev->sync_thread);
7254 sysfs_notify_dirent_safe(mddev->sysfs_action);
7255 md_new_event(mddev);
7258 if (!mddev->sync_thread) {
7259 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
7260 if (test_and_clear_bit(MD_RECOVERY_RECOVER,
7262 if (mddev->sysfs_action)
7263 sysfs_notify_dirent_safe(mddev->sysfs_action);
7265 mddev_unlock(mddev);
7269 void md_wait_for_blocked_rdev(mdk_rdev_t *rdev, mddev_t *mddev)
7271 sysfs_notify_dirent_safe(rdev->sysfs_state);
7272 wait_event_timeout(rdev->blocked_wait,
7273 !test_bit(Blocked, &rdev->flags),
7274 msecs_to_jiffies(5000));
7275 rdev_dec_pending(rdev, mddev);
7277 EXPORT_SYMBOL(md_wait_for_blocked_rdev);
7279 static int md_notify_reboot(struct notifier_block *this,
7280 unsigned long code, void *x)
7282 struct list_head *tmp;
7285 if ((code == SYS_DOWN) || (code == SYS_HALT) || (code == SYS_POWER_OFF)) {
7287 printk(KERN_INFO "md: stopping all md devices.\n");
7289 for_each_mddev(mddev, tmp)
7290 if (mddev_trylock(mddev)) {
7291 /* Force a switch to readonly even array
7292 * appears to still be in use. Hence
7295 md_set_readonly(mddev, 100);
7296 mddev_unlock(mddev);
7299 * certain more exotic SCSI devices are known to be
7300 * volatile wrt too early system reboots. While the
7301 * right place to handle this issue is the given
7302 * driver, we do want to have a safe RAID driver ...
7309 static struct notifier_block md_notifier = {
7310 .notifier_call = md_notify_reboot,
7312 .priority = INT_MAX, /* before any real devices */
7315 static void md_geninit(void)
7317 dprintk("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
7319 proc_create("mdstat", S_IRUGO, NULL, &md_seq_fops);
7322 static int __init md_init(void)
7326 md_wq = alloc_workqueue("md", WQ_RESCUER, 0);
7330 md_misc_wq = alloc_workqueue("md_misc", 0, 0);
7334 if ((ret = register_blkdev(MD_MAJOR, "md")) < 0)
7337 if ((ret = register_blkdev(0, "mdp")) < 0)
7341 blk_register_region(MKDEV(MD_MAJOR, 0), 1UL<<MINORBITS, THIS_MODULE,
7342 md_probe, NULL, NULL);
7343 blk_register_region(MKDEV(mdp_major, 0), 1UL<<MINORBITS, THIS_MODULE,
7344 md_probe, NULL, NULL);
7346 register_reboot_notifier(&md_notifier);
7347 raid_table_header = register_sysctl_table(raid_root_table);
7353 unregister_blkdev(MD_MAJOR, "md");
7355 destroy_workqueue(md_misc_wq);
7357 destroy_workqueue(md_wq);
7365 * Searches all registered partitions for autorun RAID arrays
7369 static LIST_HEAD(all_detected_devices);
7370 struct detected_devices_node {
7371 struct list_head list;
7375 void md_autodetect_dev(dev_t dev)
7377 struct detected_devices_node *node_detected_dev;
7379 node_detected_dev = kzalloc(sizeof(*node_detected_dev), GFP_KERNEL);
7380 if (node_detected_dev) {
7381 node_detected_dev->dev = dev;
7382 list_add_tail(&node_detected_dev->list, &all_detected_devices);
7384 printk(KERN_CRIT "md: md_autodetect_dev: kzalloc failed"
7385 ", skipping dev(%d,%d)\n", MAJOR(dev), MINOR(dev));
7390 static void autostart_arrays(int part)
7393 struct detected_devices_node *node_detected_dev;
7395 int i_scanned, i_passed;
7400 printk(KERN_INFO "md: Autodetecting RAID arrays.\n");
7402 while (!list_empty(&all_detected_devices) && i_scanned < INT_MAX) {
7404 node_detected_dev = list_entry(all_detected_devices.next,
7405 struct detected_devices_node, list);
7406 list_del(&node_detected_dev->list);
7407 dev = node_detected_dev->dev;
7408 kfree(node_detected_dev);
7409 rdev = md_import_device(dev,0, 90);
7413 if (test_bit(Faulty, &rdev->flags)) {
7417 set_bit(AutoDetected, &rdev->flags);
7418 list_add(&rdev->same_set, &pending_raid_disks);
7422 printk(KERN_INFO "md: Scanned %d and added %d devices.\n",
7423 i_scanned, i_passed);
7425 autorun_devices(part);
7428 #endif /* !MODULE */
7430 static __exit void md_exit(void)
7433 struct list_head *tmp;
7435 blk_unregister_region(MKDEV(MD_MAJOR,0), 1U << MINORBITS);
7436 blk_unregister_region(MKDEV(mdp_major,0), 1U << MINORBITS);
7438 unregister_blkdev(MD_MAJOR,"md");
7439 unregister_blkdev(mdp_major, "mdp");
7440 unregister_reboot_notifier(&md_notifier);
7441 unregister_sysctl_table(raid_table_header);
7442 remove_proc_entry("mdstat", NULL);
7443 for_each_mddev(mddev, tmp) {
7444 export_array(mddev);
7445 mddev->hold_active = 0;
7447 destroy_workqueue(md_misc_wq);
7448 destroy_workqueue(md_wq);
7451 subsys_initcall(md_init);
7452 module_exit(md_exit)
7454 static int get_ro(char *buffer, struct kernel_param *kp)
7456 return sprintf(buffer, "%d", start_readonly);
7458 static int set_ro(const char *val, struct kernel_param *kp)
7461 int num = simple_strtoul(val, &e, 10);
7462 if (*val && (*e == '\0' || *e == '\n')) {
7463 start_readonly = num;
7469 module_param_call(start_ro, set_ro, get_ro, NULL, S_IRUSR|S_IWUSR);
7470 module_param(start_dirty_degraded, int, S_IRUGO|S_IWUSR);
7472 module_param_call(new_array, add_named_array, NULL, NULL, S_IWUSR);
7474 EXPORT_SYMBOL(register_md_personality);
7475 EXPORT_SYMBOL(unregister_md_personality);
7476 EXPORT_SYMBOL(md_error);
7477 EXPORT_SYMBOL(md_done_sync);
7478 EXPORT_SYMBOL(md_write_start);
7479 EXPORT_SYMBOL(md_write_end);
7480 EXPORT_SYMBOL(md_register_thread);
7481 EXPORT_SYMBOL(md_unregister_thread);
7482 EXPORT_SYMBOL(md_wakeup_thread);
7483 EXPORT_SYMBOL(md_check_recovery);
7484 MODULE_LICENSE("GPL");
7485 MODULE_DESCRIPTION("MD RAID framework");
7487 MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR);