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_link_disk_holder(rdev->bdev, 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 list_del_rcu(&rdev->same_set);
1916 printk(KERN_INFO "md: unbind<%s>\n", bdevname(rdev->bdev,b));
1918 sysfs_remove_link(&rdev->kobj, "block");
1919 sysfs_put(rdev->sysfs_state);
1920 rdev->sysfs_state = NULL;
1921 /* We need to delay this, otherwise we can deadlock when
1922 * writing to 'remove' to "dev/state". We also need
1923 * to delay it due to rcu usage.
1926 INIT_WORK(&rdev->del_work, md_delayed_delete);
1927 kobject_get(&rdev->kobj);
1928 queue_work(md_misc_wq, &rdev->del_work);
1932 * prevent the device from being mounted, repartitioned or
1933 * otherwise reused by a RAID array (or any other kernel
1934 * subsystem), by bd_claiming the device.
1936 static int lock_rdev(mdk_rdev_t *rdev, dev_t dev, int shared)
1939 struct block_device *bdev;
1940 char b[BDEVNAME_SIZE];
1942 bdev = blkdev_get_by_dev(dev, FMODE_READ|FMODE_WRITE|FMODE_EXCL,
1943 shared ? (mdk_rdev_t *)lock_rdev : rdev);
1945 printk(KERN_ERR "md: could not open %s.\n",
1946 __bdevname(dev, b));
1947 return PTR_ERR(bdev);
1950 set_bit(AllReserved, &rdev->flags);
1955 static void unlock_rdev(mdk_rdev_t *rdev)
1957 struct block_device *bdev = rdev->bdev;
1961 blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
1964 void md_autodetect_dev(dev_t dev);
1966 static void export_rdev(mdk_rdev_t * rdev)
1968 char b[BDEVNAME_SIZE];
1969 printk(KERN_INFO "md: export_rdev(%s)\n",
1970 bdevname(rdev->bdev,b));
1975 if (test_bit(AutoDetected, &rdev->flags))
1976 md_autodetect_dev(rdev->bdev->bd_dev);
1979 kobject_put(&rdev->kobj);
1982 static void kick_rdev_from_array(mdk_rdev_t * rdev)
1984 unbind_rdev_from_array(rdev);
1988 static void export_array(mddev_t *mddev)
1990 mdk_rdev_t *rdev, *tmp;
1992 rdev_for_each(rdev, tmp, mddev) {
1997 kick_rdev_from_array(rdev);
1999 if (!list_empty(&mddev->disks))
2001 mddev->raid_disks = 0;
2002 mddev->major_version = 0;
2005 static void print_desc(mdp_disk_t *desc)
2007 printk(" DISK<N:%d,(%d,%d),R:%d,S:%d>\n", desc->number,
2008 desc->major,desc->minor,desc->raid_disk,desc->state);
2011 static void print_sb_90(mdp_super_t *sb)
2016 "md: SB: (V:%d.%d.%d) ID:<%08x.%08x.%08x.%08x> CT:%08x\n",
2017 sb->major_version, sb->minor_version, sb->patch_version,
2018 sb->set_uuid0, sb->set_uuid1, sb->set_uuid2, sb->set_uuid3,
2020 printk(KERN_INFO "md: L%d S%08d ND:%d RD:%d md%d LO:%d CS:%d\n",
2021 sb->level, sb->size, sb->nr_disks, sb->raid_disks,
2022 sb->md_minor, sb->layout, sb->chunk_size);
2023 printk(KERN_INFO "md: UT:%08x ST:%d AD:%d WD:%d"
2024 " FD:%d SD:%d CSUM:%08x E:%08lx\n",
2025 sb->utime, sb->state, sb->active_disks, sb->working_disks,
2026 sb->failed_disks, sb->spare_disks,
2027 sb->sb_csum, (unsigned long)sb->events_lo);
2030 for (i = 0; i < MD_SB_DISKS; i++) {
2033 desc = sb->disks + i;
2034 if (desc->number || desc->major || desc->minor ||
2035 desc->raid_disk || (desc->state && (desc->state != 4))) {
2036 printk(" D %2d: ", i);
2040 printk(KERN_INFO "md: THIS: ");
2041 print_desc(&sb->this_disk);
2044 static void print_sb_1(struct mdp_superblock_1 *sb)
2048 uuid = sb->set_uuid;
2050 "md: SB: (V:%u) (F:0x%08x) Array-ID:<%pU>\n"
2051 "md: Name: \"%s\" CT:%llu\n",
2052 le32_to_cpu(sb->major_version),
2053 le32_to_cpu(sb->feature_map),
2056 (unsigned long long)le64_to_cpu(sb->ctime)
2057 & MD_SUPERBLOCK_1_TIME_SEC_MASK);
2059 uuid = sb->device_uuid;
2061 "md: L%u SZ%llu RD:%u LO:%u CS:%u DO:%llu DS:%llu SO:%llu"
2063 "md: Dev:%08x UUID: %pU\n"
2064 "md: (F:0x%08x) UT:%llu Events:%llu ResyncOffset:%llu CSUM:0x%08x\n"
2065 "md: (MaxDev:%u) \n",
2066 le32_to_cpu(sb->level),
2067 (unsigned long long)le64_to_cpu(sb->size),
2068 le32_to_cpu(sb->raid_disks),
2069 le32_to_cpu(sb->layout),
2070 le32_to_cpu(sb->chunksize),
2071 (unsigned long long)le64_to_cpu(sb->data_offset),
2072 (unsigned long long)le64_to_cpu(sb->data_size),
2073 (unsigned long long)le64_to_cpu(sb->super_offset),
2074 (unsigned long long)le64_to_cpu(sb->recovery_offset),
2075 le32_to_cpu(sb->dev_number),
2078 (unsigned long long)le64_to_cpu(sb->utime) & MD_SUPERBLOCK_1_TIME_SEC_MASK,
2079 (unsigned long long)le64_to_cpu(sb->events),
2080 (unsigned long long)le64_to_cpu(sb->resync_offset),
2081 le32_to_cpu(sb->sb_csum),
2082 le32_to_cpu(sb->max_dev)
2086 static void print_rdev(mdk_rdev_t *rdev, int major_version)
2088 char b[BDEVNAME_SIZE];
2089 printk(KERN_INFO "md: rdev %s, Sect:%08llu F:%d S:%d DN:%u\n",
2090 bdevname(rdev->bdev, b), (unsigned long long)rdev->sectors,
2091 test_bit(Faulty, &rdev->flags), test_bit(In_sync, &rdev->flags),
2093 if (rdev->sb_loaded) {
2094 printk(KERN_INFO "md: rdev superblock (MJ:%d):\n", major_version);
2095 switch (major_version) {
2097 print_sb_90((mdp_super_t*)page_address(rdev->sb_page));
2100 print_sb_1((struct mdp_superblock_1 *)page_address(rdev->sb_page));
2104 printk(KERN_INFO "md: no rdev superblock!\n");
2107 static void md_print_devices(void)
2109 struct list_head *tmp;
2112 char b[BDEVNAME_SIZE];
2115 printk("md: **********************************\n");
2116 printk("md: * <COMPLETE RAID STATE PRINTOUT> *\n");
2117 printk("md: **********************************\n");
2118 for_each_mddev(mddev, tmp) {
2121 bitmap_print_sb(mddev->bitmap);
2123 printk("%s: ", mdname(mddev));
2124 list_for_each_entry(rdev, &mddev->disks, same_set)
2125 printk("<%s>", bdevname(rdev->bdev,b));
2128 list_for_each_entry(rdev, &mddev->disks, same_set)
2129 print_rdev(rdev, mddev->major_version);
2131 printk("md: **********************************\n");
2136 static void sync_sbs(mddev_t * mddev, int nospares)
2138 /* Update each superblock (in-memory image), but
2139 * if we are allowed to, skip spares which already
2140 * have the right event counter, or have one earlier
2141 * (which would mean they aren't being marked as dirty
2142 * with the rest of the array)
2145 list_for_each_entry(rdev, &mddev->disks, same_set) {
2146 if (rdev->sb_events == mddev->events ||
2148 rdev->raid_disk < 0 &&
2149 rdev->sb_events+1 == mddev->events)) {
2150 /* Don't update this superblock */
2151 rdev->sb_loaded = 2;
2153 super_types[mddev->major_version].
2154 sync_super(mddev, rdev);
2155 rdev->sb_loaded = 1;
2160 static void md_update_sb(mddev_t * mddev, int force_change)
2167 /* First make sure individual recovery_offsets are correct */
2168 list_for_each_entry(rdev, &mddev->disks, same_set) {
2169 if (rdev->raid_disk >= 0 &&
2170 mddev->delta_disks >= 0 &&
2171 !test_bit(In_sync, &rdev->flags) &&
2172 mddev->curr_resync_completed > rdev->recovery_offset)
2173 rdev->recovery_offset = mddev->curr_resync_completed;
2176 if (!mddev->persistent) {
2177 clear_bit(MD_CHANGE_CLEAN, &mddev->flags);
2178 clear_bit(MD_CHANGE_DEVS, &mddev->flags);
2179 if (!mddev->external)
2180 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
2181 wake_up(&mddev->sb_wait);
2185 spin_lock_irq(&mddev->write_lock);
2187 mddev->utime = get_seconds();
2189 if (test_and_clear_bit(MD_CHANGE_DEVS, &mddev->flags))
2191 if (test_and_clear_bit(MD_CHANGE_CLEAN, &mddev->flags))
2192 /* just a clean<-> dirty transition, possibly leave spares alone,
2193 * though if events isn't the right even/odd, we will have to do
2199 if (mddev->degraded)
2200 /* If the array is degraded, then skipping spares is both
2201 * dangerous and fairly pointless.
2202 * Dangerous because a device that was removed from the array
2203 * might have a event_count that still looks up-to-date,
2204 * so it can be re-added without a resync.
2205 * Pointless because if there are any spares to skip,
2206 * then a recovery will happen and soon that array won't
2207 * be degraded any more and the spare can go back to sleep then.
2211 sync_req = mddev->in_sync;
2213 /* If this is just a dirty<->clean transition, and the array is clean
2214 * and 'events' is odd, we can roll back to the previous clean state */
2216 && (mddev->in_sync && mddev->recovery_cp == MaxSector)
2217 && mddev->can_decrease_events
2218 && mddev->events != 1) {
2220 mddev->can_decrease_events = 0;
2222 /* otherwise we have to go forward and ... */
2224 mddev->can_decrease_events = nospares;
2227 if (!mddev->events) {
2229 * oops, this 64-bit counter should never wrap.
2230 * Either we are in around ~1 trillion A.C., assuming
2231 * 1 reboot per second, or we have a bug:
2236 sync_sbs(mddev, nospares);
2237 spin_unlock_irq(&mddev->write_lock);
2240 "md: updating %s RAID superblock on device (in sync %d)\n",
2241 mdname(mddev),mddev->in_sync);
2243 bitmap_update_sb(mddev->bitmap);
2244 list_for_each_entry(rdev, &mddev->disks, same_set) {
2245 char b[BDEVNAME_SIZE];
2246 dprintk(KERN_INFO "md: ");
2247 if (rdev->sb_loaded != 1)
2248 continue; /* no noise on spare devices */
2249 if (test_bit(Faulty, &rdev->flags))
2250 dprintk("(skipping faulty ");
2252 dprintk("%s ", bdevname(rdev->bdev,b));
2253 if (!test_bit(Faulty, &rdev->flags)) {
2254 md_super_write(mddev,rdev,
2255 rdev->sb_start, rdev->sb_size,
2257 dprintk(KERN_INFO "(write) %s's sb offset: %llu\n",
2258 bdevname(rdev->bdev,b),
2259 (unsigned long long)rdev->sb_start);
2260 rdev->sb_events = mddev->events;
2264 if (mddev->level == LEVEL_MULTIPATH)
2265 /* only need to write one superblock... */
2268 md_super_wait(mddev);
2269 /* if there was a failure, MD_CHANGE_DEVS was set, and we re-write super */
2271 spin_lock_irq(&mddev->write_lock);
2272 if (mddev->in_sync != sync_req ||
2273 test_bit(MD_CHANGE_DEVS, &mddev->flags)) {
2274 /* have to write it out again */
2275 spin_unlock_irq(&mddev->write_lock);
2278 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
2279 spin_unlock_irq(&mddev->write_lock);
2280 wake_up(&mddev->sb_wait);
2281 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
2282 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
2286 /* words written to sysfs files may, or may not, be \n terminated.
2287 * We want to accept with case. For this we use cmd_match.
2289 static int cmd_match(const char *cmd, const char *str)
2291 /* See if cmd, written into a sysfs file, matches
2292 * str. They must either be the same, or cmd can
2293 * have a trailing newline
2295 while (*cmd && *str && *cmd == *str) {
2306 struct rdev_sysfs_entry {
2307 struct attribute attr;
2308 ssize_t (*show)(mdk_rdev_t *, char *);
2309 ssize_t (*store)(mdk_rdev_t *, const char *, size_t);
2313 state_show(mdk_rdev_t *rdev, char *page)
2318 if (test_bit(Faulty, &rdev->flags)) {
2319 len+= sprintf(page+len, "%sfaulty",sep);
2322 if (test_bit(In_sync, &rdev->flags)) {
2323 len += sprintf(page+len, "%sin_sync",sep);
2326 if (test_bit(WriteMostly, &rdev->flags)) {
2327 len += sprintf(page+len, "%swrite_mostly",sep);
2330 if (test_bit(Blocked, &rdev->flags)) {
2331 len += sprintf(page+len, "%sblocked", sep);
2334 if (!test_bit(Faulty, &rdev->flags) &&
2335 !test_bit(In_sync, &rdev->flags)) {
2336 len += sprintf(page+len, "%sspare", sep);
2339 return len+sprintf(page+len, "\n");
2343 state_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2346 * faulty - simulates and error
2347 * remove - disconnects the device
2348 * writemostly - sets write_mostly
2349 * -writemostly - clears write_mostly
2350 * blocked - sets the Blocked flag
2351 * -blocked - clears the Blocked flag
2352 * insync - sets Insync providing device isn't active
2355 if (cmd_match(buf, "faulty") && rdev->mddev->pers) {
2356 md_error(rdev->mddev, rdev);
2358 } else if (cmd_match(buf, "remove")) {
2359 if (rdev->raid_disk >= 0)
2362 mddev_t *mddev = rdev->mddev;
2363 kick_rdev_from_array(rdev);
2365 md_update_sb(mddev, 1);
2366 md_new_event(mddev);
2369 } else if (cmd_match(buf, "writemostly")) {
2370 set_bit(WriteMostly, &rdev->flags);
2372 } else if (cmd_match(buf, "-writemostly")) {
2373 clear_bit(WriteMostly, &rdev->flags);
2375 } else if (cmd_match(buf, "blocked")) {
2376 set_bit(Blocked, &rdev->flags);
2378 } else if (cmd_match(buf, "-blocked")) {
2379 clear_bit(Blocked, &rdev->flags);
2380 wake_up(&rdev->blocked_wait);
2381 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2382 md_wakeup_thread(rdev->mddev->thread);
2385 } else if (cmd_match(buf, "insync") && rdev->raid_disk == -1) {
2386 set_bit(In_sync, &rdev->flags);
2390 sysfs_notify_dirent_safe(rdev->sysfs_state);
2391 return err ? err : len;
2393 static struct rdev_sysfs_entry rdev_state =
2394 __ATTR(state, S_IRUGO|S_IWUSR, state_show, state_store);
2397 errors_show(mdk_rdev_t *rdev, char *page)
2399 return sprintf(page, "%d\n", atomic_read(&rdev->corrected_errors));
2403 errors_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2406 unsigned long n = simple_strtoul(buf, &e, 10);
2407 if (*buf && (*e == 0 || *e == '\n')) {
2408 atomic_set(&rdev->corrected_errors, n);
2413 static struct rdev_sysfs_entry rdev_errors =
2414 __ATTR(errors, S_IRUGO|S_IWUSR, errors_show, errors_store);
2417 slot_show(mdk_rdev_t *rdev, char *page)
2419 if (rdev->raid_disk < 0)
2420 return sprintf(page, "none\n");
2422 return sprintf(page, "%d\n", rdev->raid_disk);
2426 slot_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2431 int slot = simple_strtoul(buf, &e, 10);
2432 if (strncmp(buf, "none", 4)==0)
2434 else if (e==buf || (*e && *e!= '\n'))
2436 if (rdev->mddev->pers && slot == -1) {
2437 /* Setting 'slot' on an active array requires also
2438 * updating the 'rd%d' link, and communicating
2439 * with the personality with ->hot_*_disk.
2440 * For now we only support removing
2441 * failed/spare devices. This normally happens automatically,
2442 * but not when the metadata is externally managed.
2444 if (rdev->raid_disk == -1)
2446 /* personality does all needed checks */
2447 if (rdev->mddev->pers->hot_add_disk == NULL)
2449 err = rdev->mddev->pers->
2450 hot_remove_disk(rdev->mddev, rdev->raid_disk);
2453 sprintf(nm, "rd%d", rdev->raid_disk);
2454 sysfs_remove_link(&rdev->mddev->kobj, nm);
2455 rdev->raid_disk = -1;
2456 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2457 md_wakeup_thread(rdev->mddev->thread);
2458 } else if (rdev->mddev->pers) {
2460 /* Activating a spare .. or possibly reactivating
2461 * if we ever get bitmaps working here.
2464 if (rdev->raid_disk != -1)
2467 if (rdev->mddev->pers->hot_add_disk == NULL)
2470 list_for_each_entry(rdev2, &rdev->mddev->disks, same_set)
2471 if (rdev2->raid_disk == slot)
2474 if (slot >= rdev->mddev->raid_disks &&
2475 slot >= rdev->mddev->raid_disks + rdev->mddev->delta_disks)
2478 rdev->raid_disk = slot;
2479 if (test_bit(In_sync, &rdev->flags))
2480 rdev->saved_raid_disk = slot;
2482 rdev->saved_raid_disk = -1;
2483 err = rdev->mddev->pers->
2484 hot_add_disk(rdev->mddev, rdev);
2486 rdev->raid_disk = -1;
2489 sysfs_notify_dirent_safe(rdev->sysfs_state);
2490 sprintf(nm, "rd%d", rdev->raid_disk);
2491 if (sysfs_create_link(&rdev->mddev->kobj, &rdev->kobj, nm))
2492 /* failure here is OK */;
2493 /* don't wakeup anyone, leave that to userspace. */
2495 if (slot >= rdev->mddev->raid_disks &&
2496 slot >= rdev->mddev->raid_disks + rdev->mddev->delta_disks)
2498 rdev->raid_disk = slot;
2499 /* assume it is working */
2500 clear_bit(Faulty, &rdev->flags);
2501 clear_bit(WriteMostly, &rdev->flags);
2502 set_bit(In_sync, &rdev->flags);
2503 sysfs_notify_dirent_safe(rdev->sysfs_state);
2509 static struct rdev_sysfs_entry rdev_slot =
2510 __ATTR(slot, S_IRUGO|S_IWUSR, slot_show, slot_store);
2513 offset_show(mdk_rdev_t *rdev, char *page)
2515 return sprintf(page, "%llu\n", (unsigned long long)rdev->data_offset);
2519 offset_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2522 unsigned long long offset = simple_strtoull(buf, &e, 10);
2523 if (e==buf || (*e && *e != '\n'))
2525 if (rdev->mddev->pers && rdev->raid_disk >= 0)
2527 if (rdev->sectors && rdev->mddev->external)
2528 /* Must set offset before size, so overlap checks
2531 rdev->data_offset = offset;
2535 static struct rdev_sysfs_entry rdev_offset =
2536 __ATTR(offset, S_IRUGO|S_IWUSR, offset_show, offset_store);
2539 rdev_size_show(mdk_rdev_t *rdev, char *page)
2541 return sprintf(page, "%llu\n", (unsigned long long)rdev->sectors / 2);
2544 static int overlaps(sector_t s1, sector_t l1, sector_t s2, sector_t l2)
2546 /* check if two start/length pairs overlap */
2554 static int strict_blocks_to_sectors(const char *buf, sector_t *sectors)
2556 unsigned long long blocks;
2559 if (strict_strtoull(buf, 10, &blocks) < 0)
2562 if (blocks & 1ULL << (8 * sizeof(blocks) - 1))
2563 return -EINVAL; /* sector conversion overflow */
2566 if (new != blocks * 2)
2567 return -EINVAL; /* unsigned long long to sector_t overflow */
2574 rdev_size_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2576 mddev_t *my_mddev = rdev->mddev;
2577 sector_t oldsectors = rdev->sectors;
2580 if (strict_blocks_to_sectors(buf, §ors) < 0)
2582 if (my_mddev->pers && rdev->raid_disk >= 0) {
2583 if (my_mddev->persistent) {
2584 sectors = super_types[my_mddev->major_version].
2585 rdev_size_change(rdev, sectors);
2588 } else if (!sectors)
2589 sectors = (i_size_read(rdev->bdev->bd_inode) >> 9) -
2592 if (sectors < my_mddev->dev_sectors)
2593 return -EINVAL; /* component must fit device */
2595 rdev->sectors = sectors;
2596 if (sectors > oldsectors && my_mddev->external) {
2597 /* need to check that all other rdevs with the same ->bdev
2598 * do not overlap. We need to unlock the mddev to avoid
2599 * a deadlock. We have already changed rdev->sectors, and if
2600 * we have to change it back, we will have the lock again.
2604 struct list_head *tmp;
2606 mddev_unlock(my_mddev);
2607 for_each_mddev(mddev, tmp) {
2611 list_for_each_entry(rdev2, &mddev->disks, same_set)
2612 if (test_bit(AllReserved, &rdev2->flags) ||
2613 (rdev->bdev == rdev2->bdev &&
2615 overlaps(rdev->data_offset, rdev->sectors,
2621 mddev_unlock(mddev);
2627 mddev_lock(my_mddev);
2629 /* Someone else could have slipped in a size
2630 * change here, but doing so is just silly.
2631 * We put oldsectors back because we *know* it is
2632 * safe, and trust userspace not to race with
2635 rdev->sectors = oldsectors;
2642 static struct rdev_sysfs_entry rdev_size =
2643 __ATTR(size, S_IRUGO|S_IWUSR, rdev_size_show, rdev_size_store);
2646 static ssize_t recovery_start_show(mdk_rdev_t *rdev, char *page)
2648 unsigned long long recovery_start = rdev->recovery_offset;
2650 if (test_bit(In_sync, &rdev->flags) ||
2651 recovery_start == MaxSector)
2652 return sprintf(page, "none\n");
2654 return sprintf(page, "%llu\n", recovery_start);
2657 static ssize_t recovery_start_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2659 unsigned long long recovery_start;
2661 if (cmd_match(buf, "none"))
2662 recovery_start = MaxSector;
2663 else if (strict_strtoull(buf, 10, &recovery_start))
2666 if (rdev->mddev->pers &&
2667 rdev->raid_disk >= 0)
2670 rdev->recovery_offset = recovery_start;
2671 if (recovery_start == MaxSector)
2672 set_bit(In_sync, &rdev->flags);
2674 clear_bit(In_sync, &rdev->flags);
2678 static struct rdev_sysfs_entry rdev_recovery_start =
2679 __ATTR(recovery_start, S_IRUGO|S_IWUSR, recovery_start_show, recovery_start_store);
2681 static struct attribute *rdev_default_attrs[] = {
2687 &rdev_recovery_start.attr,
2691 rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
2693 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
2694 mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
2695 mddev_t *mddev = rdev->mddev;
2701 rv = mddev ? mddev_lock(mddev) : -EBUSY;
2703 if (rdev->mddev == NULL)
2706 rv = entry->show(rdev, page);
2707 mddev_unlock(mddev);
2713 rdev_attr_store(struct kobject *kobj, struct attribute *attr,
2714 const char *page, size_t length)
2716 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
2717 mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
2719 mddev_t *mddev = rdev->mddev;
2723 if (!capable(CAP_SYS_ADMIN))
2725 rv = mddev ? mddev_lock(mddev): -EBUSY;
2727 if (rdev->mddev == NULL)
2730 rv = entry->store(rdev, page, length);
2731 mddev_unlock(mddev);
2736 static void rdev_free(struct kobject *ko)
2738 mdk_rdev_t *rdev = container_of(ko, mdk_rdev_t, kobj);
2741 static const struct sysfs_ops rdev_sysfs_ops = {
2742 .show = rdev_attr_show,
2743 .store = rdev_attr_store,
2745 static struct kobj_type rdev_ktype = {
2746 .release = rdev_free,
2747 .sysfs_ops = &rdev_sysfs_ops,
2748 .default_attrs = rdev_default_attrs,
2751 void md_rdev_init(mdk_rdev_t *rdev)
2754 rdev->saved_raid_disk = -1;
2755 rdev->raid_disk = -1;
2757 rdev->data_offset = 0;
2758 rdev->sb_events = 0;
2759 rdev->last_read_error.tv_sec = 0;
2760 rdev->last_read_error.tv_nsec = 0;
2761 atomic_set(&rdev->nr_pending, 0);
2762 atomic_set(&rdev->read_errors, 0);
2763 atomic_set(&rdev->corrected_errors, 0);
2765 INIT_LIST_HEAD(&rdev->same_set);
2766 init_waitqueue_head(&rdev->blocked_wait);
2768 EXPORT_SYMBOL_GPL(md_rdev_init);
2770 * Import a device. If 'super_format' >= 0, then sanity check the superblock
2772 * mark the device faulty if:
2774 * - the device is nonexistent (zero size)
2775 * - the device has no valid superblock
2777 * a faulty rdev _never_ has rdev->sb set.
2779 static mdk_rdev_t *md_import_device(dev_t newdev, int super_format, int super_minor)
2781 char b[BDEVNAME_SIZE];
2786 rdev = kzalloc(sizeof(*rdev), GFP_KERNEL);
2788 printk(KERN_ERR "md: could not alloc mem for new device!\n");
2789 return ERR_PTR(-ENOMEM);
2793 if ((err = alloc_disk_sb(rdev)))
2796 err = lock_rdev(rdev, newdev, super_format == -2);
2800 kobject_init(&rdev->kobj, &rdev_ktype);
2802 size = i_size_read(rdev->bdev->bd_inode) >> BLOCK_SIZE_BITS;
2805 "md: %s has zero or unknown size, marking faulty!\n",
2806 bdevname(rdev->bdev,b));
2811 if (super_format >= 0) {
2812 err = super_types[super_format].
2813 load_super(rdev, NULL, super_minor);
2814 if (err == -EINVAL) {
2816 "md: %s does not have a valid v%d.%d "
2817 "superblock, not importing!\n",
2818 bdevname(rdev->bdev,b),
2819 super_format, super_minor);
2824 "md: could not read %s's sb, not importing!\n",
2825 bdevname(rdev->bdev,b));
2833 if (rdev->sb_page) {
2839 return ERR_PTR(err);
2843 * Check a full RAID array for plausibility
2847 static void analyze_sbs(mddev_t * mddev)
2850 mdk_rdev_t *rdev, *freshest, *tmp;
2851 char b[BDEVNAME_SIZE];
2854 rdev_for_each(rdev, tmp, mddev)
2855 switch (super_types[mddev->major_version].
2856 load_super(rdev, freshest, mddev->minor_version)) {
2864 "md: fatal superblock inconsistency in %s"
2865 " -- removing from array\n",
2866 bdevname(rdev->bdev,b));
2867 kick_rdev_from_array(rdev);
2871 super_types[mddev->major_version].
2872 validate_super(mddev, freshest);
2875 rdev_for_each(rdev, tmp, mddev) {
2876 if (mddev->max_disks &&
2877 (rdev->desc_nr >= mddev->max_disks ||
2878 i > mddev->max_disks)) {
2880 "md: %s: %s: only %d devices permitted\n",
2881 mdname(mddev), bdevname(rdev->bdev, b),
2883 kick_rdev_from_array(rdev);
2886 if (rdev != freshest)
2887 if (super_types[mddev->major_version].
2888 validate_super(mddev, rdev)) {
2889 printk(KERN_WARNING "md: kicking non-fresh %s"
2891 bdevname(rdev->bdev,b));
2892 kick_rdev_from_array(rdev);
2895 if (mddev->level == LEVEL_MULTIPATH) {
2896 rdev->desc_nr = i++;
2897 rdev->raid_disk = rdev->desc_nr;
2898 set_bit(In_sync, &rdev->flags);
2899 } else if (rdev->raid_disk >= (mddev->raid_disks - min(0, mddev->delta_disks))) {
2900 rdev->raid_disk = -1;
2901 clear_bit(In_sync, &rdev->flags);
2906 /* Read a fixed-point number.
2907 * Numbers in sysfs attributes should be in "standard" units where
2908 * possible, so time should be in seconds.
2909 * However we internally use a a much smaller unit such as
2910 * milliseconds or jiffies.
2911 * This function takes a decimal number with a possible fractional
2912 * component, and produces an integer which is the result of
2913 * multiplying that number by 10^'scale'.
2914 * all without any floating-point arithmetic.
2916 int strict_strtoul_scaled(const char *cp, unsigned long *res, int scale)
2918 unsigned long result = 0;
2920 while (isdigit(*cp) || (*cp == '.' && decimals < 0)) {
2923 else if (decimals < scale) {
2926 result = result * 10 + value;
2938 while (decimals < scale) {
2947 static void md_safemode_timeout(unsigned long data);
2950 safe_delay_show(mddev_t *mddev, char *page)
2952 int msec = (mddev->safemode_delay*1000)/HZ;
2953 return sprintf(page, "%d.%03d\n", msec/1000, msec%1000);
2956 safe_delay_store(mddev_t *mddev, const char *cbuf, size_t len)
2960 if (strict_strtoul_scaled(cbuf, &msec, 3) < 0)
2963 mddev->safemode_delay = 0;
2965 unsigned long old_delay = mddev->safemode_delay;
2966 mddev->safemode_delay = (msec*HZ)/1000;
2967 if (mddev->safemode_delay == 0)
2968 mddev->safemode_delay = 1;
2969 if (mddev->safemode_delay < old_delay)
2970 md_safemode_timeout((unsigned long)mddev);
2974 static struct md_sysfs_entry md_safe_delay =
2975 __ATTR(safe_mode_delay, S_IRUGO|S_IWUSR,safe_delay_show, safe_delay_store);
2978 level_show(mddev_t *mddev, char *page)
2980 struct mdk_personality *p = mddev->pers;
2982 return sprintf(page, "%s\n", p->name);
2983 else if (mddev->clevel[0])
2984 return sprintf(page, "%s\n", mddev->clevel);
2985 else if (mddev->level != LEVEL_NONE)
2986 return sprintf(page, "%d\n", mddev->level);
2992 level_store(mddev_t *mddev, const char *buf, size_t len)
2996 struct mdk_personality *pers;
3001 if (mddev->pers == NULL) {
3004 if (len >= sizeof(mddev->clevel))
3006 strncpy(mddev->clevel, buf, len);
3007 if (mddev->clevel[len-1] == '\n')
3009 mddev->clevel[len] = 0;
3010 mddev->level = LEVEL_NONE;
3014 /* request to change the personality. Need to ensure:
3015 * - array is not engaged in resync/recovery/reshape
3016 * - old personality can be suspended
3017 * - new personality will access other array.
3020 if (mddev->sync_thread ||
3021 mddev->reshape_position != MaxSector ||
3022 mddev->sysfs_active)
3025 if (!mddev->pers->quiesce) {
3026 printk(KERN_WARNING "md: %s: %s does not support online personality change\n",
3027 mdname(mddev), mddev->pers->name);
3031 /* Now find the new personality */
3032 if (len == 0 || len >= sizeof(clevel))
3034 strncpy(clevel, buf, len);
3035 if (clevel[len-1] == '\n')
3038 if (strict_strtol(clevel, 10, &level))
3041 if (request_module("md-%s", clevel) != 0)
3042 request_module("md-level-%s", clevel);
3043 spin_lock(&pers_lock);
3044 pers = find_pers(level, clevel);
3045 if (!pers || !try_module_get(pers->owner)) {
3046 spin_unlock(&pers_lock);
3047 printk(KERN_WARNING "md: personality %s not loaded\n", clevel);
3050 spin_unlock(&pers_lock);
3052 if (pers == mddev->pers) {
3053 /* Nothing to do! */
3054 module_put(pers->owner);
3057 if (!pers->takeover) {
3058 module_put(pers->owner);
3059 printk(KERN_WARNING "md: %s: %s does not support personality takeover\n",
3060 mdname(mddev), clevel);
3064 list_for_each_entry(rdev, &mddev->disks, same_set)
3065 rdev->new_raid_disk = rdev->raid_disk;
3067 /* ->takeover must set new_* and/or delta_disks
3068 * if it succeeds, and may set them when it fails.
3070 priv = pers->takeover(mddev);
3072 mddev->new_level = mddev->level;
3073 mddev->new_layout = mddev->layout;
3074 mddev->new_chunk_sectors = mddev->chunk_sectors;
3075 mddev->raid_disks -= mddev->delta_disks;
3076 mddev->delta_disks = 0;
3077 module_put(pers->owner);
3078 printk(KERN_WARNING "md: %s: %s would not accept array\n",
3079 mdname(mddev), clevel);
3080 return PTR_ERR(priv);
3083 /* Looks like we have a winner */
3084 mddev_suspend(mddev);
3085 mddev->pers->stop(mddev);
3087 if (mddev->pers->sync_request == NULL &&
3088 pers->sync_request != NULL) {
3089 /* need to add the md_redundancy_group */
3090 if (sysfs_create_group(&mddev->kobj, &md_redundancy_group))
3092 "md: cannot register extra attributes for %s\n",
3094 mddev->sysfs_action = sysfs_get_dirent(mddev->kobj.sd, NULL, "sync_action");
3096 if (mddev->pers->sync_request != NULL &&
3097 pers->sync_request == NULL) {
3098 /* need to remove the md_redundancy_group */
3099 if (mddev->to_remove == NULL)
3100 mddev->to_remove = &md_redundancy_group;
3103 if (mddev->pers->sync_request == NULL &&
3105 /* We are converting from a no-redundancy array
3106 * to a redundancy array and metadata is managed
3107 * externally so we need to be sure that writes
3108 * won't block due to a need to transition
3110 * until external management is started.
3113 mddev->safemode_delay = 0;
3114 mddev->safemode = 0;
3117 list_for_each_entry(rdev, &mddev->disks, same_set) {
3119 if (rdev->raid_disk < 0)
3121 if (rdev->new_raid_disk >= mddev->raid_disks)
3122 rdev->new_raid_disk = -1;
3123 if (rdev->new_raid_disk == rdev->raid_disk)
3125 sprintf(nm, "rd%d", rdev->raid_disk);
3126 sysfs_remove_link(&mddev->kobj, nm);
3128 list_for_each_entry(rdev, &mddev->disks, same_set) {
3129 if (rdev->raid_disk < 0)
3131 if (rdev->new_raid_disk == rdev->raid_disk)
3133 rdev->raid_disk = rdev->new_raid_disk;
3134 if (rdev->raid_disk < 0)
3135 clear_bit(In_sync, &rdev->flags);
3138 sprintf(nm, "rd%d", rdev->raid_disk);
3139 if(sysfs_create_link(&mddev->kobj, &rdev->kobj, nm))
3140 printk("md: cannot register %s for %s after level change\n",
3145 module_put(mddev->pers->owner);
3147 mddev->private = priv;
3148 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
3149 mddev->level = mddev->new_level;
3150 mddev->layout = mddev->new_layout;
3151 mddev->chunk_sectors = mddev->new_chunk_sectors;
3152 mddev->delta_disks = 0;
3153 if (mddev->pers->sync_request == NULL) {
3154 /* this is now an array without redundancy, so
3155 * it must always be in_sync
3158 del_timer_sync(&mddev->safemode_timer);
3161 mddev_resume(mddev);
3162 set_bit(MD_CHANGE_DEVS, &mddev->flags);
3163 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3164 md_wakeup_thread(mddev->thread);
3165 sysfs_notify(&mddev->kobj, NULL, "level");
3166 md_new_event(mddev);
3170 static struct md_sysfs_entry md_level =
3171 __ATTR(level, S_IRUGO|S_IWUSR, level_show, level_store);
3175 layout_show(mddev_t *mddev, char *page)
3177 /* just a number, not meaningful for all levels */
3178 if (mddev->reshape_position != MaxSector &&
3179 mddev->layout != mddev->new_layout)
3180 return sprintf(page, "%d (%d)\n",
3181 mddev->new_layout, mddev->layout);
3182 return sprintf(page, "%d\n", mddev->layout);
3186 layout_store(mddev_t *mddev, const char *buf, size_t len)
3189 unsigned long n = simple_strtoul(buf, &e, 10);
3191 if (!*buf || (*e && *e != '\n'))
3196 if (mddev->pers->check_reshape == NULL)
3198 mddev->new_layout = n;
3199 err = mddev->pers->check_reshape(mddev);
3201 mddev->new_layout = mddev->layout;
3205 mddev->new_layout = n;
3206 if (mddev->reshape_position == MaxSector)
3211 static struct md_sysfs_entry md_layout =
3212 __ATTR(layout, S_IRUGO|S_IWUSR, layout_show, layout_store);
3216 raid_disks_show(mddev_t *mddev, char *page)
3218 if (mddev->raid_disks == 0)
3220 if (mddev->reshape_position != MaxSector &&
3221 mddev->delta_disks != 0)
3222 return sprintf(page, "%d (%d)\n", mddev->raid_disks,
3223 mddev->raid_disks - mddev->delta_disks);
3224 return sprintf(page, "%d\n", mddev->raid_disks);
3227 static int update_raid_disks(mddev_t *mddev, int raid_disks);
3230 raid_disks_store(mddev_t *mddev, const char *buf, size_t len)
3234 unsigned long n = simple_strtoul(buf, &e, 10);
3236 if (!*buf || (*e && *e != '\n'))
3240 rv = update_raid_disks(mddev, n);
3241 else if (mddev->reshape_position != MaxSector) {
3242 int olddisks = mddev->raid_disks - mddev->delta_disks;
3243 mddev->delta_disks = n - olddisks;
3244 mddev->raid_disks = n;
3246 mddev->raid_disks = n;
3247 return rv ? rv : len;
3249 static struct md_sysfs_entry md_raid_disks =
3250 __ATTR(raid_disks, S_IRUGO|S_IWUSR, raid_disks_show, raid_disks_store);
3253 chunk_size_show(mddev_t *mddev, char *page)
3255 if (mddev->reshape_position != MaxSector &&
3256 mddev->chunk_sectors != mddev->new_chunk_sectors)
3257 return sprintf(page, "%d (%d)\n",
3258 mddev->new_chunk_sectors << 9,
3259 mddev->chunk_sectors << 9);
3260 return sprintf(page, "%d\n", mddev->chunk_sectors << 9);
3264 chunk_size_store(mddev_t *mddev, const char *buf, size_t len)
3267 unsigned long n = simple_strtoul(buf, &e, 10);
3269 if (!*buf || (*e && *e != '\n'))
3274 if (mddev->pers->check_reshape == NULL)
3276 mddev->new_chunk_sectors = n >> 9;
3277 err = mddev->pers->check_reshape(mddev);
3279 mddev->new_chunk_sectors = mddev->chunk_sectors;
3283 mddev->new_chunk_sectors = n >> 9;
3284 if (mddev->reshape_position == MaxSector)
3285 mddev->chunk_sectors = n >> 9;
3289 static struct md_sysfs_entry md_chunk_size =
3290 __ATTR(chunk_size, S_IRUGO|S_IWUSR, chunk_size_show, chunk_size_store);
3293 resync_start_show(mddev_t *mddev, char *page)
3295 if (mddev->recovery_cp == MaxSector)
3296 return sprintf(page, "none\n");
3297 return sprintf(page, "%llu\n", (unsigned long long)mddev->recovery_cp);
3301 resync_start_store(mddev_t *mddev, const char *buf, size_t len)
3304 unsigned long long n = simple_strtoull(buf, &e, 10);
3308 if (cmd_match(buf, "none"))
3310 else if (!*buf || (*e && *e != '\n'))
3313 mddev->recovery_cp = n;
3316 static struct md_sysfs_entry md_resync_start =
3317 __ATTR(resync_start, S_IRUGO|S_IWUSR, resync_start_show, resync_start_store);
3320 * The array state can be:
3323 * No devices, no size, no level
3324 * Equivalent to STOP_ARRAY ioctl
3326 * May have some settings, but array is not active
3327 * all IO results in error
3328 * When written, doesn't tear down array, but just stops it
3329 * suspended (not supported yet)
3330 * All IO requests will block. The array can be reconfigured.
3331 * Writing this, if accepted, will block until array is quiescent
3333 * no resync can happen. no superblocks get written.
3334 * write requests fail
3336 * like readonly, but behaves like 'clean' on a write request.
3338 * clean - no pending writes, but otherwise active.
3339 * When written to inactive array, starts without resync
3340 * If a write request arrives then
3341 * if metadata is known, mark 'dirty' and switch to 'active'.
3342 * if not known, block and switch to write-pending
3343 * If written to an active array that has pending writes, then fails.
3345 * fully active: IO and resync can be happening.
3346 * When written to inactive array, starts with resync
3349 * clean, but writes are blocked waiting for 'active' to be written.
3352 * like active, but no writes have been seen for a while (100msec).
3355 enum array_state { clear, inactive, suspended, readonly, read_auto, clean, active,
3356 write_pending, active_idle, bad_word};
3357 static char *array_states[] = {
3358 "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active",
3359 "write-pending", "active-idle", NULL };
3361 static int match_word(const char *word, char **list)
3364 for (n=0; list[n]; n++)
3365 if (cmd_match(word, list[n]))
3371 array_state_show(mddev_t *mddev, char *page)
3373 enum array_state st = inactive;
3386 else if (test_bit(MD_CHANGE_PENDING, &mddev->flags))
3388 else if (mddev->safemode)
3394 if (list_empty(&mddev->disks) &&
3395 mddev->raid_disks == 0 &&
3396 mddev->dev_sectors == 0)
3401 return sprintf(page, "%s\n", array_states[st]);
3404 static int do_md_stop(mddev_t * mddev, int ro, int is_open);
3405 static int md_set_readonly(mddev_t * mddev, int is_open);
3406 static int do_md_run(mddev_t * mddev);
3407 static int restart_array(mddev_t *mddev);
3410 array_state_store(mddev_t *mddev, const char *buf, size_t len)
3413 enum array_state st = match_word(buf, array_states);
3418 /* stopping an active array */
3419 if (atomic_read(&mddev->openers) > 0)
3421 err = do_md_stop(mddev, 0, 0);
3424 /* stopping an active array */
3426 if (atomic_read(&mddev->openers) > 0)
3428 err = do_md_stop(mddev, 2, 0);
3430 err = 0; /* already inactive */
3433 break; /* not supported yet */
3436 err = md_set_readonly(mddev, 0);
3439 set_disk_ro(mddev->gendisk, 1);
3440 err = do_md_run(mddev);
3446 err = md_set_readonly(mddev, 0);
3447 else if (mddev->ro == 1)
3448 err = restart_array(mddev);
3451 set_disk_ro(mddev->gendisk, 0);
3455 err = do_md_run(mddev);
3460 restart_array(mddev);
3461 spin_lock_irq(&mddev->write_lock);
3462 if (atomic_read(&mddev->writes_pending) == 0) {
3463 if (mddev->in_sync == 0) {
3465 if (mddev->safemode == 1)
3466 mddev->safemode = 0;
3467 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
3472 spin_unlock_irq(&mddev->write_lock);
3478 restart_array(mddev);
3479 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
3480 wake_up(&mddev->sb_wait);
3484 set_disk_ro(mddev->gendisk, 0);
3485 err = do_md_run(mddev);
3490 /* these cannot be set */
3496 sysfs_notify_dirent_safe(mddev->sysfs_state);
3500 static struct md_sysfs_entry md_array_state =
3501 __ATTR(array_state, S_IRUGO|S_IWUSR, array_state_show, array_state_store);
3504 max_corrected_read_errors_show(mddev_t *mddev, char *page) {
3505 return sprintf(page, "%d\n",
3506 atomic_read(&mddev->max_corr_read_errors));
3510 max_corrected_read_errors_store(mddev_t *mddev, const char *buf, size_t len)
3513 unsigned long n = simple_strtoul(buf, &e, 10);
3515 if (*buf && (*e == 0 || *e == '\n')) {
3516 atomic_set(&mddev->max_corr_read_errors, n);
3522 static struct md_sysfs_entry max_corr_read_errors =
3523 __ATTR(max_read_errors, S_IRUGO|S_IWUSR, max_corrected_read_errors_show,
3524 max_corrected_read_errors_store);
3527 null_show(mddev_t *mddev, char *page)
3533 new_dev_store(mddev_t *mddev, const char *buf, size_t len)
3535 /* buf must be %d:%d\n? giving major and minor numbers */
3536 /* The new device is added to the array.
3537 * If the array has a persistent superblock, we read the
3538 * superblock to initialise info and check validity.
3539 * Otherwise, only checking done is that in bind_rdev_to_array,
3540 * which mainly checks size.
3543 int major = simple_strtoul(buf, &e, 10);
3549 if (!*buf || *e != ':' || !e[1] || e[1] == '\n')
3551 minor = simple_strtoul(e+1, &e, 10);
3552 if (*e && *e != '\n')
3554 dev = MKDEV(major, minor);
3555 if (major != MAJOR(dev) ||
3556 minor != MINOR(dev))
3560 if (mddev->persistent) {
3561 rdev = md_import_device(dev, mddev->major_version,
3562 mddev->minor_version);
3563 if (!IS_ERR(rdev) && !list_empty(&mddev->disks)) {
3564 mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
3565 mdk_rdev_t, same_set);
3566 err = super_types[mddev->major_version]
3567 .load_super(rdev, rdev0, mddev->minor_version);
3571 } else if (mddev->external)
3572 rdev = md_import_device(dev, -2, -1);
3574 rdev = md_import_device(dev, -1, -1);
3577 return PTR_ERR(rdev);
3578 err = bind_rdev_to_array(rdev, mddev);
3582 return err ? err : len;
3585 static struct md_sysfs_entry md_new_device =
3586 __ATTR(new_dev, S_IWUSR, null_show, new_dev_store);
3589 bitmap_store(mddev_t *mddev, const char *buf, size_t len)
3592 unsigned long chunk, end_chunk;
3596 /* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */
3598 chunk = end_chunk = simple_strtoul(buf, &end, 0);
3599 if (buf == end) break;
3600 if (*end == '-') { /* range */
3602 end_chunk = simple_strtoul(buf, &end, 0);
3603 if (buf == end) break;
3605 if (*end && !isspace(*end)) break;
3606 bitmap_dirty_bits(mddev->bitmap, chunk, end_chunk);
3607 buf = skip_spaces(end);
3609 bitmap_unplug(mddev->bitmap); /* flush the bits to disk */
3614 static struct md_sysfs_entry md_bitmap =
3615 __ATTR(bitmap_set_bits, S_IWUSR, null_show, bitmap_store);
3618 size_show(mddev_t *mddev, char *page)
3620 return sprintf(page, "%llu\n",
3621 (unsigned long long)mddev->dev_sectors / 2);
3624 static int update_size(mddev_t *mddev, sector_t num_sectors);
3627 size_store(mddev_t *mddev, const char *buf, size_t len)
3629 /* If array is inactive, we can reduce the component size, but
3630 * not increase it (except from 0).
3631 * If array is active, we can try an on-line resize
3634 int err = strict_blocks_to_sectors(buf, §ors);
3639 err = update_size(mddev, sectors);
3640 md_update_sb(mddev, 1);
3642 if (mddev->dev_sectors == 0 ||
3643 mddev->dev_sectors > sectors)
3644 mddev->dev_sectors = sectors;
3648 return err ? err : len;
3651 static struct md_sysfs_entry md_size =
3652 __ATTR(component_size, S_IRUGO|S_IWUSR, size_show, size_store);
3657 * 'none' for arrays with no metadata (good luck...)
3658 * 'external' for arrays with externally managed metadata,
3659 * or N.M for internally known formats
3662 metadata_show(mddev_t *mddev, char *page)
3664 if (mddev->persistent)
3665 return sprintf(page, "%d.%d\n",
3666 mddev->major_version, mddev->minor_version);
3667 else if (mddev->external)
3668 return sprintf(page, "external:%s\n", mddev->metadata_type);
3670 return sprintf(page, "none\n");
3674 metadata_store(mddev_t *mddev, const char *buf, size_t len)
3678 /* Changing the details of 'external' metadata is
3679 * always permitted. Otherwise there must be
3680 * no devices attached to the array.
3682 if (mddev->external && strncmp(buf, "external:", 9) == 0)
3684 else if (!list_empty(&mddev->disks))
3687 if (cmd_match(buf, "none")) {
3688 mddev->persistent = 0;
3689 mddev->external = 0;
3690 mddev->major_version = 0;
3691 mddev->minor_version = 90;
3694 if (strncmp(buf, "external:", 9) == 0) {
3695 size_t namelen = len-9;
3696 if (namelen >= sizeof(mddev->metadata_type))
3697 namelen = sizeof(mddev->metadata_type)-1;
3698 strncpy(mddev->metadata_type, buf+9, namelen);
3699 mddev->metadata_type[namelen] = 0;
3700 if (namelen && mddev->metadata_type[namelen-1] == '\n')
3701 mddev->metadata_type[--namelen] = 0;
3702 mddev->persistent = 0;
3703 mddev->external = 1;
3704 mddev->major_version = 0;
3705 mddev->minor_version = 90;
3708 major = simple_strtoul(buf, &e, 10);
3709 if (e==buf || *e != '.')
3712 minor = simple_strtoul(buf, &e, 10);
3713 if (e==buf || (*e && *e != '\n') )
3715 if (major >= ARRAY_SIZE(super_types) || super_types[major].name == NULL)
3717 mddev->major_version = major;
3718 mddev->minor_version = minor;
3719 mddev->persistent = 1;
3720 mddev->external = 0;
3724 static struct md_sysfs_entry md_metadata =
3725 __ATTR(metadata_version, S_IRUGO|S_IWUSR, metadata_show, metadata_store);
3728 action_show(mddev_t *mddev, char *page)
3730 char *type = "idle";
3731 if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
3733 else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
3734 (!mddev->ro && test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))) {
3735 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
3737 else if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
3738 if (!test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
3740 else if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
3744 } else if (test_bit(MD_RECOVERY_RECOVER, &mddev->recovery))
3747 return sprintf(page, "%s\n", type);
3750 static void reap_sync_thread(mddev_t *mddev);
3753 action_store(mddev_t *mddev, const char *page, size_t len)
3755 if (!mddev->pers || !mddev->pers->sync_request)
3758 if (cmd_match(page, "frozen"))
3759 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
3761 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
3763 if (cmd_match(page, "idle") || cmd_match(page, "frozen")) {
3764 if (mddev->sync_thread) {
3765 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
3766 reap_sync_thread(mddev);
3768 } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
3769 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
3771 else if (cmd_match(page, "resync"))
3772 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3773 else if (cmd_match(page, "recover")) {
3774 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
3775 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3776 } else if (cmd_match(page, "reshape")) {
3778 if (mddev->pers->start_reshape == NULL)
3780 err = mddev->pers->start_reshape(mddev);
3783 sysfs_notify(&mddev->kobj, NULL, "degraded");
3785 if (cmd_match(page, "check"))
3786 set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
3787 else if (!cmd_match(page, "repair"))
3789 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
3790 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
3792 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3793 md_wakeup_thread(mddev->thread);
3794 sysfs_notify_dirent_safe(mddev->sysfs_action);
3799 mismatch_cnt_show(mddev_t *mddev, char *page)
3801 return sprintf(page, "%llu\n",
3802 (unsigned long long) mddev->resync_mismatches);
3805 static struct md_sysfs_entry md_scan_mode =
3806 __ATTR(sync_action, S_IRUGO|S_IWUSR, action_show, action_store);
3809 static struct md_sysfs_entry md_mismatches = __ATTR_RO(mismatch_cnt);
3812 sync_min_show(mddev_t *mddev, char *page)
3814 return sprintf(page, "%d (%s)\n", speed_min(mddev),
3815 mddev->sync_speed_min ? "local": "system");
3819 sync_min_store(mddev_t *mddev, const char *buf, size_t len)
3823 if (strncmp(buf, "system", 6)==0) {
3824 mddev->sync_speed_min = 0;
3827 min = simple_strtoul(buf, &e, 10);
3828 if (buf == e || (*e && *e != '\n') || min <= 0)
3830 mddev->sync_speed_min = min;
3834 static struct md_sysfs_entry md_sync_min =
3835 __ATTR(sync_speed_min, S_IRUGO|S_IWUSR, sync_min_show, sync_min_store);
3838 sync_max_show(mddev_t *mddev, char *page)
3840 return sprintf(page, "%d (%s)\n", speed_max(mddev),
3841 mddev->sync_speed_max ? "local": "system");
3845 sync_max_store(mddev_t *mddev, const char *buf, size_t len)
3849 if (strncmp(buf, "system", 6)==0) {
3850 mddev->sync_speed_max = 0;
3853 max = simple_strtoul(buf, &e, 10);
3854 if (buf == e || (*e && *e != '\n') || max <= 0)
3856 mddev->sync_speed_max = max;
3860 static struct md_sysfs_entry md_sync_max =
3861 __ATTR(sync_speed_max, S_IRUGO|S_IWUSR, sync_max_show, sync_max_store);
3864 degraded_show(mddev_t *mddev, char *page)
3866 return sprintf(page, "%d\n", mddev->degraded);
3868 static struct md_sysfs_entry md_degraded = __ATTR_RO(degraded);
3871 sync_force_parallel_show(mddev_t *mddev, char *page)
3873 return sprintf(page, "%d\n", mddev->parallel_resync);
3877 sync_force_parallel_store(mddev_t *mddev, const char *buf, size_t len)
3881 if (strict_strtol(buf, 10, &n))
3884 if (n != 0 && n != 1)
3887 mddev->parallel_resync = n;
3889 if (mddev->sync_thread)
3890 wake_up(&resync_wait);
3895 /* force parallel resync, even with shared block devices */
3896 static struct md_sysfs_entry md_sync_force_parallel =
3897 __ATTR(sync_force_parallel, S_IRUGO|S_IWUSR,
3898 sync_force_parallel_show, sync_force_parallel_store);
3901 sync_speed_show(mddev_t *mddev, char *page)
3903 unsigned long resync, dt, db;
3904 if (mddev->curr_resync == 0)
3905 return sprintf(page, "none\n");
3906 resync = mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active);
3907 dt = (jiffies - mddev->resync_mark) / HZ;
3909 db = resync - mddev->resync_mark_cnt;
3910 return sprintf(page, "%lu\n", db/dt/2); /* K/sec */
3913 static struct md_sysfs_entry md_sync_speed = __ATTR_RO(sync_speed);
3916 sync_completed_show(mddev_t *mddev, char *page)
3918 unsigned long long max_sectors, resync;
3920 if (!test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
3921 return sprintf(page, "none\n");
3923 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
3924 max_sectors = mddev->resync_max_sectors;
3926 max_sectors = mddev->dev_sectors;
3928 resync = mddev->curr_resync_completed;
3929 return sprintf(page, "%llu / %llu\n", resync, max_sectors);
3932 static struct md_sysfs_entry md_sync_completed = __ATTR_RO(sync_completed);
3935 min_sync_show(mddev_t *mddev, char *page)
3937 return sprintf(page, "%llu\n",
3938 (unsigned long long)mddev->resync_min);
3941 min_sync_store(mddev_t *mddev, const char *buf, size_t len)
3943 unsigned long long min;
3944 if (strict_strtoull(buf, 10, &min))
3946 if (min > mddev->resync_max)
3948 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
3951 /* Must be a multiple of chunk_size */
3952 if (mddev->chunk_sectors) {
3953 sector_t temp = min;
3954 if (sector_div(temp, mddev->chunk_sectors))
3957 mddev->resync_min = min;
3962 static struct md_sysfs_entry md_min_sync =
3963 __ATTR(sync_min, S_IRUGO|S_IWUSR, min_sync_show, min_sync_store);
3966 max_sync_show(mddev_t *mddev, char *page)
3968 if (mddev->resync_max == MaxSector)
3969 return sprintf(page, "max\n");
3971 return sprintf(page, "%llu\n",
3972 (unsigned long long)mddev->resync_max);
3975 max_sync_store(mddev_t *mddev, const char *buf, size_t len)
3977 if (strncmp(buf, "max", 3) == 0)
3978 mddev->resync_max = MaxSector;
3980 unsigned long long max;
3981 if (strict_strtoull(buf, 10, &max))
3983 if (max < mddev->resync_min)
3985 if (max < mddev->resync_max &&
3987 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
3990 /* Must be a multiple of chunk_size */
3991 if (mddev->chunk_sectors) {
3992 sector_t temp = max;
3993 if (sector_div(temp, mddev->chunk_sectors))
3996 mddev->resync_max = max;
3998 wake_up(&mddev->recovery_wait);
4002 static struct md_sysfs_entry md_max_sync =
4003 __ATTR(sync_max, S_IRUGO|S_IWUSR, max_sync_show, max_sync_store);
4006 suspend_lo_show(mddev_t *mddev, char *page)
4008 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_lo);
4012 suspend_lo_store(mddev_t *mddev, const char *buf, size_t len)
4015 unsigned long long new = simple_strtoull(buf, &e, 10);
4016 unsigned long long old = mddev->suspend_lo;
4018 if (mddev->pers == NULL ||
4019 mddev->pers->quiesce == NULL)
4021 if (buf == e || (*e && *e != '\n'))
4024 mddev->suspend_lo = new;
4026 /* Shrinking suspended region */
4027 mddev->pers->quiesce(mddev, 2);
4029 /* Expanding suspended region - need to wait */
4030 mddev->pers->quiesce(mddev, 1);
4031 mddev->pers->quiesce(mddev, 0);
4035 static struct md_sysfs_entry md_suspend_lo =
4036 __ATTR(suspend_lo, S_IRUGO|S_IWUSR, suspend_lo_show, suspend_lo_store);
4040 suspend_hi_show(mddev_t *mddev, char *page)
4042 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_hi);
4046 suspend_hi_store(mddev_t *mddev, const char *buf, size_t len)
4049 unsigned long long new = simple_strtoull(buf, &e, 10);
4050 unsigned long long old = mddev->suspend_hi;
4052 if (mddev->pers == NULL ||
4053 mddev->pers->quiesce == NULL)
4055 if (buf == e || (*e && *e != '\n'))
4058 mddev->suspend_hi = new;
4060 /* Shrinking suspended region */
4061 mddev->pers->quiesce(mddev, 2);
4063 /* Expanding suspended region - need to wait */
4064 mddev->pers->quiesce(mddev, 1);
4065 mddev->pers->quiesce(mddev, 0);
4069 static struct md_sysfs_entry md_suspend_hi =
4070 __ATTR(suspend_hi, S_IRUGO|S_IWUSR, suspend_hi_show, suspend_hi_store);
4073 reshape_position_show(mddev_t *mddev, char *page)
4075 if (mddev->reshape_position != MaxSector)
4076 return sprintf(page, "%llu\n",
4077 (unsigned long long)mddev->reshape_position);
4078 strcpy(page, "none\n");
4083 reshape_position_store(mddev_t *mddev, const char *buf, size_t len)
4086 unsigned long long new = simple_strtoull(buf, &e, 10);
4089 if (buf == e || (*e && *e != '\n'))
4091 mddev->reshape_position = new;
4092 mddev->delta_disks = 0;
4093 mddev->new_level = mddev->level;
4094 mddev->new_layout = mddev->layout;
4095 mddev->new_chunk_sectors = mddev->chunk_sectors;
4099 static struct md_sysfs_entry md_reshape_position =
4100 __ATTR(reshape_position, S_IRUGO|S_IWUSR, reshape_position_show,
4101 reshape_position_store);
4104 array_size_show(mddev_t *mddev, char *page)
4106 if (mddev->external_size)
4107 return sprintf(page, "%llu\n",
4108 (unsigned long long)mddev->array_sectors/2);
4110 return sprintf(page, "default\n");
4114 array_size_store(mddev_t *mddev, const char *buf, size_t len)
4118 if (strncmp(buf, "default", 7) == 0) {
4120 sectors = mddev->pers->size(mddev, 0, 0);
4122 sectors = mddev->array_sectors;
4124 mddev->external_size = 0;
4126 if (strict_blocks_to_sectors(buf, §ors) < 0)
4128 if (mddev->pers && mddev->pers->size(mddev, 0, 0) < sectors)
4131 mddev->external_size = 1;
4134 mddev->array_sectors = sectors;
4135 set_capacity(mddev->gendisk, mddev->array_sectors);
4137 revalidate_disk(mddev->gendisk);
4142 static struct md_sysfs_entry md_array_size =
4143 __ATTR(array_size, S_IRUGO|S_IWUSR, array_size_show,
4146 static struct attribute *md_default_attrs[] = {
4149 &md_raid_disks.attr,
4150 &md_chunk_size.attr,
4152 &md_resync_start.attr,
4154 &md_new_device.attr,
4155 &md_safe_delay.attr,
4156 &md_array_state.attr,
4157 &md_reshape_position.attr,
4158 &md_array_size.attr,
4159 &max_corr_read_errors.attr,
4163 static struct attribute *md_redundancy_attrs[] = {
4165 &md_mismatches.attr,
4168 &md_sync_speed.attr,
4169 &md_sync_force_parallel.attr,
4170 &md_sync_completed.attr,
4173 &md_suspend_lo.attr,
4174 &md_suspend_hi.attr,
4179 static struct attribute_group md_redundancy_group = {
4181 .attrs = md_redundancy_attrs,
4186 md_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
4188 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
4189 mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
4194 rv = mddev_lock(mddev);
4196 rv = entry->show(mddev, page);
4197 mddev_unlock(mddev);
4203 md_attr_store(struct kobject *kobj, struct attribute *attr,
4204 const char *page, size_t length)
4206 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
4207 mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
4212 if (!capable(CAP_SYS_ADMIN))
4214 rv = mddev_lock(mddev);
4215 if (mddev->hold_active == UNTIL_IOCTL)
4216 mddev->hold_active = 0;
4218 rv = entry->store(mddev, page, length);
4219 mddev_unlock(mddev);
4224 static void md_free(struct kobject *ko)
4226 mddev_t *mddev = container_of(ko, mddev_t, kobj);
4228 if (mddev->sysfs_state)
4229 sysfs_put(mddev->sysfs_state);
4231 if (mddev->gendisk) {
4232 del_gendisk(mddev->gendisk);
4233 put_disk(mddev->gendisk);
4236 blk_cleanup_queue(mddev->queue);
4241 static const struct sysfs_ops md_sysfs_ops = {
4242 .show = md_attr_show,
4243 .store = md_attr_store,
4245 static struct kobj_type md_ktype = {
4247 .sysfs_ops = &md_sysfs_ops,
4248 .default_attrs = md_default_attrs,
4253 static void mddev_delayed_delete(struct work_struct *ws)
4255 mddev_t *mddev = container_of(ws, mddev_t, del_work);
4257 sysfs_remove_group(&mddev->kobj, &md_bitmap_group);
4258 kobject_del(&mddev->kobj);
4259 kobject_put(&mddev->kobj);
4262 static int md_alloc(dev_t dev, char *name)
4264 static DEFINE_MUTEX(disks_mutex);
4265 mddev_t *mddev = mddev_find(dev);
4266 struct gendisk *disk;
4275 partitioned = (MAJOR(mddev->unit) != MD_MAJOR);
4276 shift = partitioned ? MdpMinorShift : 0;
4277 unit = MINOR(mddev->unit) >> shift;
4279 /* wait for any previous instance of this device to be
4280 * completely removed (mddev_delayed_delete).
4282 flush_workqueue(md_misc_wq);
4284 mutex_lock(&disks_mutex);
4290 /* Need to ensure that 'name' is not a duplicate.
4293 spin_lock(&all_mddevs_lock);
4295 list_for_each_entry(mddev2, &all_mddevs, all_mddevs)
4296 if (mddev2->gendisk &&
4297 strcmp(mddev2->gendisk->disk_name, name) == 0) {
4298 spin_unlock(&all_mddevs_lock);
4301 spin_unlock(&all_mddevs_lock);
4305 mddev->queue = blk_alloc_queue(GFP_KERNEL);
4308 mddev->queue->queuedata = mddev;
4310 blk_queue_make_request(mddev->queue, md_make_request);
4312 disk = alloc_disk(1 << shift);
4314 blk_cleanup_queue(mddev->queue);
4315 mddev->queue = NULL;
4318 disk->major = MAJOR(mddev->unit);
4319 disk->first_minor = unit << shift;
4321 strcpy(disk->disk_name, name);
4322 else if (partitioned)
4323 sprintf(disk->disk_name, "md_d%d", unit);
4325 sprintf(disk->disk_name, "md%d", unit);
4326 disk->fops = &md_fops;
4327 disk->private_data = mddev;
4328 disk->queue = mddev->queue;
4329 /* Allow extended partitions. This makes the
4330 * 'mdp' device redundant, but we can't really
4333 disk->flags |= GENHD_FL_EXT_DEVT;
4335 mddev->gendisk = disk;
4336 error = kobject_init_and_add(&mddev->kobj, &md_ktype,
4337 &disk_to_dev(disk)->kobj, "%s", "md");
4339 /* This isn't possible, but as kobject_init_and_add is marked
4340 * __must_check, we must do something with the result
4342 printk(KERN_WARNING "md: cannot register %s/md - name in use\n",
4346 if (mddev->kobj.sd &&
4347 sysfs_create_group(&mddev->kobj, &md_bitmap_group))
4348 printk(KERN_DEBUG "pointless warning\n");
4350 blk_queue_flush(mddev->queue, REQ_FLUSH | REQ_FUA);
4352 mutex_unlock(&disks_mutex);
4353 if (!error && mddev->kobj.sd) {
4354 kobject_uevent(&mddev->kobj, KOBJ_ADD);
4355 mddev->sysfs_state = sysfs_get_dirent_safe(mddev->kobj.sd, "array_state");
4361 static struct kobject *md_probe(dev_t dev, int *part, void *data)
4363 md_alloc(dev, NULL);
4367 static int add_named_array(const char *val, struct kernel_param *kp)
4369 /* val must be "md_*" where * is not all digits.
4370 * We allocate an array with a large free minor number, and
4371 * set the name to val. val must not already be an active name.
4373 int len = strlen(val);
4374 char buf[DISK_NAME_LEN];
4376 while (len && val[len-1] == '\n')
4378 if (len >= DISK_NAME_LEN)
4380 strlcpy(buf, val, len+1);
4381 if (strncmp(buf, "md_", 3) != 0)
4383 return md_alloc(0, buf);
4386 static void md_safemode_timeout(unsigned long data)
4388 mddev_t *mddev = (mddev_t *) data;
4390 if (!atomic_read(&mddev->writes_pending)) {
4391 mddev->safemode = 1;
4392 if (mddev->external)
4393 sysfs_notify_dirent_safe(mddev->sysfs_state);
4395 md_wakeup_thread(mddev->thread);
4398 static int start_dirty_degraded;
4400 int md_run(mddev_t *mddev)
4404 struct mdk_personality *pers;
4406 if (list_empty(&mddev->disks))
4407 /* cannot run an array with no devices.. */
4412 /* Cannot run until previous stop completes properly */
4413 if (mddev->sysfs_active)
4417 * Analyze all RAID superblock(s)
4419 if (!mddev->raid_disks) {
4420 if (!mddev->persistent)
4425 if (mddev->level != LEVEL_NONE)
4426 request_module("md-level-%d", mddev->level);
4427 else if (mddev->clevel[0])
4428 request_module("md-%s", mddev->clevel);
4431 * Drop all container device buffers, from now on
4432 * the only valid external interface is through the md
4435 list_for_each_entry(rdev, &mddev->disks, same_set) {
4436 if (test_bit(Faulty, &rdev->flags))
4438 sync_blockdev(rdev->bdev);
4439 invalidate_bdev(rdev->bdev);
4441 /* perform some consistency tests on the device.
4442 * We don't want the data to overlap the metadata,
4443 * Internal Bitmap issues have been handled elsewhere.
4445 if (rdev->meta_bdev) {
4446 /* Nothing to check */;
4447 } else if (rdev->data_offset < rdev->sb_start) {
4448 if (mddev->dev_sectors &&
4449 rdev->data_offset + mddev->dev_sectors
4451 printk("md: %s: data overlaps metadata\n",
4456 if (rdev->sb_start + rdev->sb_size/512
4457 > rdev->data_offset) {
4458 printk("md: %s: metadata overlaps data\n",
4463 sysfs_notify_dirent_safe(rdev->sysfs_state);
4466 if (mddev->bio_set == NULL)
4467 mddev->bio_set = bioset_create(BIO_POOL_SIZE, sizeof(mddev));
4469 spin_lock(&pers_lock);
4470 pers = find_pers(mddev->level, mddev->clevel);
4471 if (!pers || !try_module_get(pers->owner)) {
4472 spin_unlock(&pers_lock);
4473 if (mddev->level != LEVEL_NONE)
4474 printk(KERN_WARNING "md: personality for level %d is not loaded!\n",
4477 printk(KERN_WARNING "md: personality for level %s is not loaded!\n",
4482 spin_unlock(&pers_lock);
4483 if (mddev->level != pers->level) {
4484 mddev->level = pers->level;
4485 mddev->new_level = pers->level;
4487 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
4489 if (mddev->reshape_position != MaxSector &&
4490 pers->start_reshape == NULL) {
4491 /* This personality cannot handle reshaping... */
4493 module_put(pers->owner);
4497 if (pers->sync_request) {
4498 /* Warn if this is a potentially silly
4501 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
4505 list_for_each_entry(rdev, &mddev->disks, same_set)
4506 list_for_each_entry(rdev2, &mddev->disks, same_set) {
4508 rdev->bdev->bd_contains ==
4509 rdev2->bdev->bd_contains) {
4511 "%s: WARNING: %s appears to be"
4512 " on the same physical disk as"
4515 bdevname(rdev->bdev,b),
4516 bdevname(rdev2->bdev,b2));
4523 "True protection against single-disk"
4524 " failure might be compromised.\n");
4527 mddev->recovery = 0;
4528 /* may be over-ridden by personality */
4529 mddev->resync_max_sectors = mddev->dev_sectors;
4531 mddev->ok_start_degraded = start_dirty_degraded;
4533 if (start_readonly && mddev->ro == 0)
4534 mddev->ro = 2; /* read-only, but switch on first write */
4536 err = mddev->pers->run(mddev);
4538 printk(KERN_ERR "md: pers->run() failed ...\n");
4539 else if (mddev->pers->size(mddev, 0, 0) < mddev->array_sectors) {
4540 WARN_ONCE(!mddev->external_size, "%s: default size too small,"
4541 " but 'external_size' not in effect?\n", __func__);
4543 "md: invalid array_size %llu > default size %llu\n",
4544 (unsigned long long)mddev->array_sectors / 2,
4545 (unsigned long long)mddev->pers->size(mddev, 0, 0) / 2);
4547 mddev->pers->stop(mddev);
4549 if (err == 0 && mddev->pers->sync_request) {
4550 err = bitmap_create(mddev);
4552 printk(KERN_ERR "%s: failed to create bitmap (%d)\n",
4553 mdname(mddev), err);
4554 mddev->pers->stop(mddev);
4558 module_put(mddev->pers->owner);
4560 bitmap_destroy(mddev);
4563 if (mddev->pers->sync_request) {
4564 if (mddev->kobj.sd &&
4565 sysfs_create_group(&mddev->kobj, &md_redundancy_group))
4567 "md: cannot register extra attributes for %s\n",
4569 mddev->sysfs_action = sysfs_get_dirent_safe(mddev->kobj.sd, "sync_action");
4570 } else if (mddev->ro == 2) /* auto-readonly not meaningful */
4573 atomic_set(&mddev->writes_pending,0);
4574 atomic_set(&mddev->max_corr_read_errors,
4575 MD_DEFAULT_MAX_CORRECTED_READ_ERRORS);
4576 mddev->safemode = 0;
4577 mddev->safemode_timer.function = md_safemode_timeout;
4578 mddev->safemode_timer.data = (unsigned long) mddev;
4579 mddev->safemode_delay = (200 * HZ)/1000 +1; /* 200 msec delay */
4583 list_for_each_entry(rdev, &mddev->disks, same_set)
4584 if (rdev->raid_disk >= 0) {
4586 sprintf(nm, "rd%d", rdev->raid_disk);
4587 if (sysfs_create_link(&mddev->kobj, &rdev->kobj, nm))
4588 /* failure here is OK */;
4591 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4594 md_update_sb(mddev, 0);
4596 md_wakeup_thread(mddev->thread);
4597 md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
4599 md_new_event(mddev);
4600 sysfs_notify_dirent_safe(mddev->sysfs_state);
4601 sysfs_notify_dirent_safe(mddev->sysfs_action);
4602 sysfs_notify(&mddev->kobj, NULL, "degraded");
4605 EXPORT_SYMBOL_GPL(md_run);
4607 static int do_md_run(mddev_t *mddev)
4611 err = md_run(mddev);
4614 err = bitmap_load(mddev);
4616 bitmap_destroy(mddev);
4619 set_capacity(mddev->gendisk, mddev->array_sectors);
4620 revalidate_disk(mddev->gendisk);
4621 kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
4626 static int restart_array(mddev_t *mddev)
4628 struct gendisk *disk = mddev->gendisk;
4630 /* Complain if it has no devices */
4631 if (list_empty(&mddev->disks))
4637 mddev->safemode = 0;
4639 set_disk_ro(disk, 0);
4640 printk(KERN_INFO "md: %s switched to read-write mode.\n",
4642 /* Kick recovery or resync if necessary */
4643 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4644 md_wakeup_thread(mddev->thread);
4645 md_wakeup_thread(mddev->sync_thread);
4646 sysfs_notify_dirent_safe(mddev->sysfs_state);
4650 /* similar to deny_write_access, but accounts for our holding a reference
4651 * to the file ourselves */
4652 static int deny_bitmap_write_access(struct file * file)
4654 struct inode *inode = file->f_mapping->host;
4656 spin_lock(&inode->i_lock);
4657 if (atomic_read(&inode->i_writecount) > 1) {
4658 spin_unlock(&inode->i_lock);
4661 atomic_set(&inode->i_writecount, -1);
4662 spin_unlock(&inode->i_lock);
4667 void restore_bitmap_write_access(struct file *file)
4669 struct inode *inode = file->f_mapping->host;
4671 spin_lock(&inode->i_lock);
4672 atomic_set(&inode->i_writecount, 1);
4673 spin_unlock(&inode->i_lock);
4676 static void md_clean(mddev_t *mddev)
4678 mddev->array_sectors = 0;
4679 mddev->external_size = 0;
4680 mddev->dev_sectors = 0;
4681 mddev->raid_disks = 0;
4682 mddev->recovery_cp = 0;
4683 mddev->resync_min = 0;
4684 mddev->resync_max = MaxSector;
4685 mddev->reshape_position = MaxSector;
4686 mddev->external = 0;
4687 mddev->persistent = 0;
4688 mddev->level = LEVEL_NONE;
4689 mddev->clevel[0] = 0;
4692 mddev->metadata_type[0] = 0;
4693 mddev->chunk_sectors = 0;
4694 mddev->ctime = mddev->utime = 0;
4696 mddev->max_disks = 0;
4698 mddev->can_decrease_events = 0;
4699 mddev->delta_disks = 0;
4700 mddev->new_level = LEVEL_NONE;
4701 mddev->new_layout = 0;
4702 mddev->new_chunk_sectors = 0;
4703 mddev->curr_resync = 0;
4704 mddev->resync_mismatches = 0;
4705 mddev->suspend_lo = mddev->suspend_hi = 0;
4706 mddev->sync_speed_min = mddev->sync_speed_max = 0;
4707 mddev->recovery = 0;
4709 mddev->degraded = 0;
4710 mddev->safemode = 0;
4711 mddev->bitmap_info.offset = 0;
4712 mddev->bitmap_info.default_offset = 0;
4713 mddev->bitmap_info.chunksize = 0;
4714 mddev->bitmap_info.daemon_sleep = 0;
4715 mddev->bitmap_info.max_write_behind = 0;
4719 static void __md_stop_writes(mddev_t *mddev)
4721 if (mddev->sync_thread) {
4722 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4723 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
4724 reap_sync_thread(mddev);
4727 del_timer_sync(&mddev->safemode_timer);
4729 bitmap_flush(mddev);
4730 md_super_wait(mddev);
4732 if (!mddev->in_sync || mddev->flags) {
4733 /* mark array as shutdown cleanly */
4735 md_update_sb(mddev, 1);
4739 void md_stop_writes(mddev_t *mddev)
4742 __md_stop_writes(mddev);
4743 mddev_unlock(mddev);
4745 EXPORT_SYMBOL_GPL(md_stop_writes);
4747 void md_stop(mddev_t *mddev)
4750 mddev->pers->stop(mddev);
4751 if (mddev->pers->sync_request && mddev->to_remove == NULL)
4752 mddev->to_remove = &md_redundancy_group;
4753 module_put(mddev->pers->owner);
4755 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4757 EXPORT_SYMBOL_GPL(md_stop);
4759 static int md_set_readonly(mddev_t *mddev, int is_open)
4762 mutex_lock(&mddev->open_mutex);
4763 if (atomic_read(&mddev->openers) > is_open) {
4764 printk("md: %s still in use.\n",mdname(mddev));
4769 __md_stop_writes(mddev);
4775 set_disk_ro(mddev->gendisk, 1);
4776 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4777 sysfs_notify_dirent_safe(mddev->sysfs_state);
4781 mutex_unlock(&mddev->open_mutex);
4786 * 0 - completely stop and dis-assemble array
4787 * 2 - stop but do not disassemble array
4789 static int do_md_stop(mddev_t * mddev, int mode, int is_open)
4791 struct gendisk *disk = mddev->gendisk;
4794 mutex_lock(&mddev->open_mutex);
4795 if (atomic_read(&mddev->openers) > is_open ||
4796 mddev->sysfs_active) {
4797 printk("md: %s still in use.\n",mdname(mddev));
4798 mutex_unlock(&mddev->open_mutex);
4804 set_disk_ro(disk, 0);
4806 __md_stop_writes(mddev);
4808 mddev->queue->merge_bvec_fn = NULL;
4809 mddev->queue->unplug_fn = NULL;
4810 mddev->queue->backing_dev_info.congested_fn = NULL;
4812 /* tell userspace to handle 'inactive' */
4813 sysfs_notify_dirent_safe(mddev->sysfs_state);
4815 list_for_each_entry(rdev, &mddev->disks, same_set)
4816 if (rdev->raid_disk >= 0) {
4818 sprintf(nm, "rd%d", rdev->raid_disk);
4819 sysfs_remove_link(&mddev->kobj, nm);
4822 set_capacity(disk, 0);
4823 mutex_unlock(&mddev->open_mutex);
4824 revalidate_disk(disk);
4829 mutex_unlock(&mddev->open_mutex);
4831 * Free resources if final stop
4834 printk(KERN_INFO "md: %s stopped.\n", mdname(mddev));
4836 bitmap_destroy(mddev);
4837 if (mddev->bitmap_info.file) {
4838 restore_bitmap_write_access(mddev->bitmap_info.file);
4839 fput(mddev->bitmap_info.file);
4840 mddev->bitmap_info.file = NULL;
4842 mddev->bitmap_info.offset = 0;
4844 export_array(mddev);
4847 kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
4848 if (mddev->hold_active == UNTIL_STOP)
4849 mddev->hold_active = 0;
4851 blk_integrity_unregister(disk);
4852 md_new_event(mddev);
4853 sysfs_notify_dirent_safe(mddev->sysfs_state);
4858 static void autorun_array(mddev_t *mddev)
4863 if (list_empty(&mddev->disks))
4866 printk(KERN_INFO "md: running: ");
4868 list_for_each_entry(rdev, &mddev->disks, same_set) {
4869 char b[BDEVNAME_SIZE];
4870 printk("<%s>", bdevname(rdev->bdev,b));
4874 err = do_md_run(mddev);
4876 printk(KERN_WARNING "md: do_md_run() returned %d\n", err);
4877 do_md_stop(mddev, 0, 0);
4882 * lets try to run arrays based on all disks that have arrived
4883 * until now. (those are in pending_raid_disks)
4885 * the method: pick the first pending disk, collect all disks with
4886 * the same UUID, remove all from the pending list and put them into
4887 * the 'same_array' list. Then order this list based on superblock
4888 * update time (freshest comes first), kick out 'old' disks and
4889 * compare superblocks. If everything's fine then run it.
4891 * If "unit" is allocated, then bump its reference count
4893 static void autorun_devices(int part)
4895 mdk_rdev_t *rdev0, *rdev, *tmp;
4897 char b[BDEVNAME_SIZE];
4899 printk(KERN_INFO "md: autorun ...\n");
4900 while (!list_empty(&pending_raid_disks)) {
4903 LIST_HEAD(candidates);
4904 rdev0 = list_entry(pending_raid_disks.next,
4905 mdk_rdev_t, same_set);
4907 printk(KERN_INFO "md: considering %s ...\n",
4908 bdevname(rdev0->bdev,b));
4909 INIT_LIST_HEAD(&candidates);
4910 rdev_for_each_list(rdev, tmp, &pending_raid_disks)
4911 if (super_90_load(rdev, rdev0, 0) >= 0) {
4912 printk(KERN_INFO "md: adding %s ...\n",
4913 bdevname(rdev->bdev,b));
4914 list_move(&rdev->same_set, &candidates);
4917 * now we have a set of devices, with all of them having
4918 * mostly sane superblocks. It's time to allocate the
4922 dev = MKDEV(mdp_major,
4923 rdev0->preferred_minor << MdpMinorShift);
4924 unit = MINOR(dev) >> MdpMinorShift;
4926 dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
4929 if (rdev0->preferred_minor != unit) {
4930 printk(KERN_INFO "md: unit number in %s is bad: %d\n",
4931 bdevname(rdev0->bdev, b), rdev0->preferred_minor);
4935 md_probe(dev, NULL, NULL);
4936 mddev = mddev_find(dev);
4937 if (!mddev || !mddev->gendisk) {
4941 "md: cannot allocate memory for md drive.\n");
4944 if (mddev_lock(mddev))
4945 printk(KERN_WARNING "md: %s locked, cannot run\n",
4947 else if (mddev->raid_disks || mddev->major_version
4948 || !list_empty(&mddev->disks)) {
4950 "md: %s already running, cannot run %s\n",
4951 mdname(mddev), bdevname(rdev0->bdev,b));
4952 mddev_unlock(mddev);
4954 printk(KERN_INFO "md: created %s\n", mdname(mddev));
4955 mddev->persistent = 1;
4956 rdev_for_each_list(rdev, tmp, &candidates) {
4957 list_del_init(&rdev->same_set);
4958 if (bind_rdev_to_array(rdev, mddev))
4961 autorun_array(mddev);
4962 mddev_unlock(mddev);
4964 /* on success, candidates will be empty, on error
4967 rdev_for_each_list(rdev, tmp, &candidates) {
4968 list_del_init(&rdev->same_set);
4973 printk(KERN_INFO "md: ... autorun DONE.\n");
4975 #endif /* !MODULE */
4977 static int get_version(void __user * arg)
4981 ver.major = MD_MAJOR_VERSION;
4982 ver.minor = MD_MINOR_VERSION;
4983 ver.patchlevel = MD_PATCHLEVEL_VERSION;
4985 if (copy_to_user(arg, &ver, sizeof(ver)))
4991 static int get_array_info(mddev_t * mddev, void __user * arg)
4993 mdu_array_info_t info;
4994 int nr,working,insync,failed,spare;
4997 nr=working=insync=failed=spare=0;
4998 list_for_each_entry(rdev, &mddev->disks, same_set) {
5000 if (test_bit(Faulty, &rdev->flags))
5004 if (test_bit(In_sync, &rdev->flags))
5011 info.major_version = mddev->major_version;
5012 info.minor_version = mddev->minor_version;
5013 info.patch_version = MD_PATCHLEVEL_VERSION;
5014 info.ctime = mddev->ctime;
5015 info.level = mddev->level;
5016 info.size = mddev->dev_sectors / 2;
5017 if (info.size != mddev->dev_sectors / 2) /* overflow */
5020 info.raid_disks = mddev->raid_disks;
5021 info.md_minor = mddev->md_minor;
5022 info.not_persistent= !mddev->persistent;
5024 info.utime = mddev->utime;
5027 info.state = (1<<MD_SB_CLEAN);
5028 if (mddev->bitmap && mddev->bitmap_info.offset)
5029 info.state = (1<<MD_SB_BITMAP_PRESENT);
5030 info.active_disks = insync;
5031 info.working_disks = working;
5032 info.failed_disks = failed;
5033 info.spare_disks = spare;
5035 info.layout = mddev->layout;
5036 info.chunk_size = mddev->chunk_sectors << 9;
5038 if (copy_to_user(arg, &info, sizeof(info)))
5044 static int get_bitmap_file(mddev_t * mddev, void __user * arg)
5046 mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */
5047 char *ptr, *buf = NULL;
5050 if (md_allow_write(mddev))
5051 file = kmalloc(sizeof(*file), GFP_NOIO);
5053 file = kmalloc(sizeof(*file), GFP_KERNEL);
5058 /* bitmap disabled, zero the first byte and copy out */
5059 if (!mddev->bitmap || !mddev->bitmap->file) {
5060 file->pathname[0] = '\0';
5064 buf = kmalloc(sizeof(file->pathname), GFP_KERNEL);
5068 ptr = d_path(&mddev->bitmap->file->f_path, buf, sizeof(file->pathname));
5072 strcpy(file->pathname, ptr);
5076 if (copy_to_user(arg, file, sizeof(*file)))
5084 static int get_disk_info(mddev_t * mddev, void __user * arg)
5086 mdu_disk_info_t info;
5089 if (copy_from_user(&info, arg, sizeof(info)))
5092 rdev = find_rdev_nr(mddev, info.number);
5094 info.major = MAJOR(rdev->bdev->bd_dev);
5095 info.minor = MINOR(rdev->bdev->bd_dev);
5096 info.raid_disk = rdev->raid_disk;
5098 if (test_bit(Faulty, &rdev->flags))
5099 info.state |= (1<<MD_DISK_FAULTY);
5100 else if (test_bit(In_sync, &rdev->flags)) {
5101 info.state |= (1<<MD_DISK_ACTIVE);
5102 info.state |= (1<<MD_DISK_SYNC);
5104 if (test_bit(WriteMostly, &rdev->flags))
5105 info.state |= (1<<MD_DISK_WRITEMOSTLY);
5107 info.major = info.minor = 0;
5108 info.raid_disk = -1;
5109 info.state = (1<<MD_DISK_REMOVED);
5112 if (copy_to_user(arg, &info, sizeof(info)))
5118 static int add_new_disk(mddev_t * mddev, mdu_disk_info_t *info)
5120 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
5122 dev_t dev = MKDEV(info->major,info->minor);
5124 if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
5127 if (!mddev->raid_disks) {
5129 /* expecting a device which has a superblock */
5130 rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
5133 "md: md_import_device returned %ld\n",
5135 return PTR_ERR(rdev);
5137 if (!list_empty(&mddev->disks)) {
5138 mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
5139 mdk_rdev_t, same_set);
5140 err = super_types[mddev->major_version]
5141 .load_super(rdev, rdev0, mddev->minor_version);
5144 "md: %s has different UUID to %s\n",
5145 bdevname(rdev->bdev,b),
5146 bdevname(rdev0->bdev,b2));
5151 err = bind_rdev_to_array(rdev, mddev);
5158 * add_new_disk can be used once the array is assembled
5159 * to add "hot spares". They must already have a superblock
5164 if (!mddev->pers->hot_add_disk) {
5166 "%s: personality does not support diskops!\n",
5170 if (mddev->persistent)
5171 rdev = md_import_device(dev, mddev->major_version,
5172 mddev->minor_version);
5174 rdev = md_import_device(dev, -1, -1);
5177 "md: md_import_device returned %ld\n",
5179 return PTR_ERR(rdev);
5181 /* set saved_raid_disk if appropriate */
5182 if (!mddev->persistent) {
5183 if (info->state & (1<<MD_DISK_SYNC) &&
5184 info->raid_disk < mddev->raid_disks) {
5185 rdev->raid_disk = info->raid_disk;
5186 set_bit(In_sync, &rdev->flags);
5188 rdev->raid_disk = -1;
5190 super_types[mddev->major_version].
5191 validate_super(mddev, rdev);
5192 if (test_bit(In_sync, &rdev->flags))
5193 rdev->saved_raid_disk = rdev->raid_disk;
5195 rdev->saved_raid_disk = -1;
5197 clear_bit(In_sync, &rdev->flags); /* just to be sure */
5198 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
5199 set_bit(WriteMostly, &rdev->flags);
5201 clear_bit(WriteMostly, &rdev->flags);
5203 rdev->raid_disk = -1;
5204 err = bind_rdev_to_array(rdev, mddev);
5205 if (!err && !mddev->pers->hot_remove_disk) {
5206 /* If there is hot_add_disk but no hot_remove_disk
5207 * then added disks for geometry changes,
5208 * and should be added immediately.
5210 super_types[mddev->major_version].
5211 validate_super(mddev, rdev);
5212 err = mddev->pers->hot_add_disk(mddev, rdev);
5214 unbind_rdev_from_array(rdev);
5219 sysfs_notify_dirent_safe(rdev->sysfs_state);
5221 md_update_sb(mddev, 1);
5222 if (mddev->degraded)
5223 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
5224 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5225 md_wakeup_thread(mddev->thread);
5229 /* otherwise, add_new_disk is only allowed
5230 * for major_version==0 superblocks
5232 if (mddev->major_version != 0) {
5233 printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n",
5238 if (!(info->state & (1<<MD_DISK_FAULTY))) {
5240 rdev = md_import_device(dev, -1, 0);
5243 "md: error, md_import_device() returned %ld\n",
5245 return PTR_ERR(rdev);
5247 rdev->desc_nr = info->number;
5248 if (info->raid_disk < mddev->raid_disks)
5249 rdev->raid_disk = info->raid_disk;
5251 rdev->raid_disk = -1;
5253 if (rdev->raid_disk < mddev->raid_disks)
5254 if (info->state & (1<<MD_DISK_SYNC))
5255 set_bit(In_sync, &rdev->flags);
5257 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
5258 set_bit(WriteMostly, &rdev->flags);
5260 if (!mddev->persistent) {
5261 printk(KERN_INFO "md: nonpersistent superblock ...\n");
5262 rdev->sb_start = i_size_read(rdev->bdev->bd_inode) / 512;
5264 rdev->sb_start = calc_dev_sboffset(rdev);
5265 rdev->sectors = rdev->sb_start;
5267 err = bind_rdev_to_array(rdev, mddev);
5277 static int hot_remove_disk(mddev_t * mddev, dev_t dev)
5279 char b[BDEVNAME_SIZE];
5282 rdev = find_rdev(mddev, dev);
5286 if (rdev->raid_disk >= 0)
5289 kick_rdev_from_array(rdev);
5290 md_update_sb(mddev, 1);
5291 md_new_event(mddev);
5295 printk(KERN_WARNING "md: cannot remove active disk %s from %s ...\n",
5296 bdevname(rdev->bdev,b), mdname(mddev));
5300 static int hot_add_disk(mddev_t * mddev, dev_t dev)
5302 char b[BDEVNAME_SIZE];
5309 if (mddev->major_version != 0) {
5310 printk(KERN_WARNING "%s: HOT_ADD may only be used with"
5311 " version-0 superblocks.\n",
5315 if (!mddev->pers->hot_add_disk) {
5317 "%s: personality does not support diskops!\n",
5322 rdev = md_import_device(dev, -1, 0);
5325 "md: error, md_import_device() returned %ld\n",
5330 if (mddev->persistent)
5331 rdev->sb_start = calc_dev_sboffset(rdev);
5333 rdev->sb_start = i_size_read(rdev->bdev->bd_inode) / 512;
5335 rdev->sectors = rdev->sb_start;
5337 if (test_bit(Faulty, &rdev->flags)) {
5339 "md: can not hot-add faulty %s disk to %s!\n",
5340 bdevname(rdev->bdev,b), mdname(mddev));
5344 clear_bit(In_sync, &rdev->flags);
5346 rdev->saved_raid_disk = -1;
5347 err = bind_rdev_to_array(rdev, mddev);
5352 * The rest should better be atomic, we can have disk failures
5353 * noticed in interrupt contexts ...
5356 rdev->raid_disk = -1;
5358 md_update_sb(mddev, 1);
5361 * Kick recovery, maybe this spare has to be added to the
5362 * array immediately.
5364 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5365 md_wakeup_thread(mddev->thread);
5366 md_new_event(mddev);
5374 static int set_bitmap_file(mddev_t *mddev, int fd)
5379 if (!mddev->pers->quiesce)
5381 if (mddev->recovery || mddev->sync_thread)
5383 /* we should be able to change the bitmap.. */
5389 return -EEXIST; /* cannot add when bitmap is present */
5390 mddev->bitmap_info.file = fget(fd);
5392 if (mddev->bitmap_info.file == NULL) {
5393 printk(KERN_ERR "%s: error: failed to get bitmap file\n",
5398 err = deny_bitmap_write_access(mddev->bitmap_info.file);
5400 printk(KERN_ERR "%s: error: bitmap file is already in use\n",
5402 fput(mddev->bitmap_info.file);
5403 mddev->bitmap_info.file = NULL;
5406 mddev->bitmap_info.offset = 0; /* file overrides offset */
5407 } else if (mddev->bitmap == NULL)
5408 return -ENOENT; /* cannot remove what isn't there */
5411 mddev->pers->quiesce(mddev, 1);
5413 err = bitmap_create(mddev);
5415 err = bitmap_load(mddev);
5417 if (fd < 0 || err) {
5418 bitmap_destroy(mddev);
5419 fd = -1; /* make sure to put the file */
5421 mddev->pers->quiesce(mddev, 0);
5424 if (mddev->bitmap_info.file) {
5425 restore_bitmap_write_access(mddev->bitmap_info.file);
5426 fput(mddev->bitmap_info.file);
5428 mddev->bitmap_info.file = NULL;
5435 * set_array_info is used two different ways
5436 * The original usage is when creating a new array.
5437 * In this usage, raid_disks is > 0 and it together with
5438 * level, size, not_persistent,layout,chunksize determine the
5439 * shape of the array.
5440 * This will always create an array with a type-0.90.0 superblock.
5441 * The newer usage is when assembling an array.
5442 * In this case raid_disks will be 0, and the major_version field is
5443 * use to determine which style super-blocks are to be found on the devices.
5444 * The minor and patch _version numbers are also kept incase the
5445 * super_block handler wishes to interpret them.
5447 static int set_array_info(mddev_t * mddev, mdu_array_info_t *info)
5450 if (info->raid_disks == 0) {
5451 /* just setting version number for superblock loading */
5452 if (info->major_version < 0 ||
5453 info->major_version >= ARRAY_SIZE(super_types) ||
5454 super_types[info->major_version].name == NULL) {
5455 /* maybe try to auto-load a module? */
5457 "md: superblock version %d not known\n",
5458 info->major_version);
5461 mddev->major_version = info->major_version;
5462 mddev->minor_version = info->minor_version;
5463 mddev->patch_version = info->patch_version;
5464 mddev->persistent = !info->not_persistent;
5465 /* ensure mddev_put doesn't delete this now that there
5466 * is some minimal configuration.
5468 mddev->ctime = get_seconds();
5471 mddev->major_version = MD_MAJOR_VERSION;
5472 mddev->minor_version = MD_MINOR_VERSION;
5473 mddev->patch_version = MD_PATCHLEVEL_VERSION;
5474 mddev->ctime = get_seconds();
5476 mddev->level = info->level;
5477 mddev->clevel[0] = 0;
5478 mddev->dev_sectors = 2 * (sector_t)info->size;
5479 mddev->raid_disks = info->raid_disks;
5480 /* don't set md_minor, it is determined by which /dev/md* was
5483 if (info->state & (1<<MD_SB_CLEAN))
5484 mddev->recovery_cp = MaxSector;
5486 mddev->recovery_cp = 0;
5487 mddev->persistent = ! info->not_persistent;
5488 mddev->external = 0;
5490 mddev->layout = info->layout;
5491 mddev->chunk_sectors = info->chunk_size >> 9;
5493 mddev->max_disks = MD_SB_DISKS;
5495 if (mddev->persistent)
5497 set_bit(MD_CHANGE_DEVS, &mddev->flags);
5499 mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9;
5500 mddev->bitmap_info.offset = 0;
5502 mddev->reshape_position = MaxSector;
5505 * Generate a 128 bit UUID
5507 get_random_bytes(mddev->uuid, 16);
5509 mddev->new_level = mddev->level;
5510 mddev->new_chunk_sectors = mddev->chunk_sectors;
5511 mddev->new_layout = mddev->layout;
5512 mddev->delta_disks = 0;
5517 void md_set_array_sectors(mddev_t *mddev, sector_t array_sectors)
5519 WARN(!mddev_is_locked(mddev), "%s: unlocked mddev!\n", __func__);
5521 if (mddev->external_size)
5524 mddev->array_sectors = array_sectors;
5526 EXPORT_SYMBOL(md_set_array_sectors);
5528 static int update_size(mddev_t *mddev, sector_t num_sectors)
5532 int fit = (num_sectors == 0);
5534 if (mddev->pers->resize == NULL)
5536 /* The "num_sectors" is the number of sectors of each device that
5537 * is used. This can only make sense for arrays with redundancy.
5538 * linear and raid0 always use whatever space is available. We can only
5539 * consider changing this number if no resync or reconstruction is
5540 * happening, and if the new size is acceptable. It must fit before the
5541 * sb_start or, if that is <data_offset, it must fit before the size
5542 * of each device. If num_sectors is zero, we find the largest size
5545 if (mddev->sync_thread)
5548 /* Sorry, cannot grow a bitmap yet, just remove it,
5552 list_for_each_entry(rdev, &mddev->disks, same_set) {
5553 sector_t avail = rdev->sectors;
5555 if (fit && (num_sectors == 0 || num_sectors > avail))
5556 num_sectors = avail;
5557 if (avail < num_sectors)
5560 rv = mddev->pers->resize(mddev, num_sectors);
5562 revalidate_disk(mddev->gendisk);
5566 static int update_raid_disks(mddev_t *mddev, int raid_disks)
5569 /* change the number of raid disks */
5570 if (mddev->pers->check_reshape == NULL)
5572 if (raid_disks <= 0 ||
5573 (mddev->max_disks && raid_disks >= mddev->max_disks))
5575 if (mddev->sync_thread || mddev->reshape_position != MaxSector)
5577 mddev->delta_disks = raid_disks - mddev->raid_disks;
5579 rv = mddev->pers->check_reshape(mddev);
5585 * update_array_info is used to change the configuration of an
5587 * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
5588 * fields in the info are checked against the array.
5589 * Any differences that cannot be handled will cause an error.
5590 * Normally, only one change can be managed at a time.
5592 static int update_array_info(mddev_t *mddev, mdu_array_info_t *info)
5598 /* calculate expected state,ignoring low bits */
5599 if (mddev->bitmap && mddev->bitmap_info.offset)
5600 state |= (1 << MD_SB_BITMAP_PRESENT);
5602 if (mddev->major_version != info->major_version ||
5603 mddev->minor_version != info->minor_version ||
5604 /* mddev->patch_version != info->patch_version || */
5605 mddev->ctime != info->ctime ||
5606 mddev->level != info->level ||
5607 /* mddev->layout != info->layout || */
5608 !mddev->persistent != info->not_persistent||
5609 mddev->chunk_sectors != info->chunk_size >> 9 ||
5610 /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
5611 ((state^info->state) & 0xfffffe00)
5614 /* Check there is only one change */
5615 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
5617 if (mddev->raid_disks != info->raid_disks)
5619 if (mddev->layout != info->layout)
5621 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT))
5628 if (mddev->layout != info->layout) {
5630 * we don't need to do anything at the md level, the
5631 * personality will take care of it all.
5633 if (mddev->pers->check_reshape == NULL)
5636 mddev->new_layout = info->layout;
5637 rv = mddev->pers->check_reshape(mddev);
5639 mddev->new_layout = mddev->layout;
5643 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
5644 rv = update_size(mddev, (sector_t)info->size * 2);
5646 if (mddev->raid_disks != info->raid_disks)
5647 rv = update_raid_disks(mddev, info->raid_disks);
5649 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) {
5650 if (mddev->pers->quiesce == NULL)
5652 if (mddev->recovery || mddev->sync_thread)
5654 if (info->state & (1<<MD_SB_BITMAP_PRESENT)) {
5655 /* add the bitmap */
5658 if (mddev->bitmap_info.default_offset == 0)
5660 mddev->bitmap_info.offset =
5661 mddev->bitmap_info.default_offset;
5662 mddev->pers->quiesce(mddev, 1);
5663 rv = bitmap_create(mddev);
5665 rv = bitmap_load(mddev);
5667 bitmap_destroy(mddev);
5668 mddev->pers->quiesce(mddev, 0);
5670 /* remove the bitmap */
5673 if (mddev->bitmap->file)
5675 mddev->pers->quiesce(mddev, 1);
5676 bitmap_destroy(mddev);
5677 mddev->pers->quiesce(mddev, 0);
5678 mddev->bitmap_info.offset = 0;
5681 md_update_sb(mddev, 1);
5685 static int set_disk_faulty(mddev_t *mddev, dev_t dev)
5689 if (mddev->pers == NULL)
5692 rdev = find_rdev(mddev, dev);
5696 md_error(mddev, rdev);
5701 * We have a problem here : there is no easy way to give a CHS
5702 * virtual geometry. We currently pretend that we have a 2 heads
5703 * 4 sectors (with a BIG number of cylinders...). This drives
5704 * dosfs just mad... ;-)
5706 static int md_getgeo(struct block_device *bdev, struct hd_geometry *geo)
5708 mddev_t *mddev = bdev->bd_disk->private_data;
5712 geo->cylinders = mddev->array_sectors / 8;
5716 static int md_ioctl(struct block_device *bdev, fmode_t mode,
5717 unsigned int cmd, unsigned long arg)
5720 void __user *argp = (void __user *)arg;
5721 mddev_t *mddev = NULL;
5724 if (!capable(CAP_SYS_ADMIN))
5728 * Commands dealing with the RAID driver but not any
5734 err = get_version(argp);
5737 case PRINT_RAID_DEBUG:
5745 autostart_arrays(arg);
5752 * Commands creating/starting a new array:
5755 mddev = bdev->bd_disk->private_data;
5762 err = mddev_lock(mddev);
5765 "md: ioctl lock interrupted, reason %d, cmd %d\n",
5772 case SET_ARRAY_INFO:
5774 mdu_array_info_t info;
5776 memset(&info, 0, sizeof(info));
5777 else if (copy_from_user(&info, argp, sizeof(info))) {
5782 err = update_array_info(mddev, &info);
5784 printk(KERN_WARNING "md: couldn't update"
5785 " array info. %d\n", err);
5790 if (!list_empty(&mddev->disks)) {
5792 "md: array %s already has disks!\n",
5797 if (mddev->raid_disks) {
5799 "md: array %s already initialised!\n",
5804 err = set_array_info(mddev, &info);
5806 printk(KERN_WARNING "md: couldn't set"
5807 " array info. %d\n", err);
5817 * Commands querying/configuring an existing array:
5819 /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
5820 * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */
5821 if ((!mddev->raid_disks && !mddev->external)
5822 && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY
5823 && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE
5824 && cmd != GET_BITMAP_FILE) {
5830 * Commands even a read-only array can execute:
5834 case GET_ARRAY_INFO:
5835 err = get_array_info(mddev, argp);
5838 case GET_BITMAP_FILE:
5839 err = get_bitmap_file(mddev, argp);
5843 err = get_disk_info(mddev, argp);
5846 case RESTART_ARRAY_RW:
5847 err = restart_array(mddev);
5851 err = do_md_stop(mddev, 0, 1);
5855 err = md_set_readonly(mddev, 1);
5859 if (get_user(ro, (int __user *)(arg))) {
5865 /* if the bdev is going readonly the value of mddev->ro
5866 * does not matter, no writes are coming
5871 /* are we are already prepared for writes? */
5875 /* transitioning to readauto need only happen for
5876 * arrays that call md_write_start
5879 err = restart_array(mddev);
5882 set_disk_ro(mddev->gendisk, 0);
5889 * The remaining ioctls are changing the state of the
5890 * superblock, so we do not allow them on read-only arrays.
5891 * However non-MD ioctls (e.g. get-size) will still come through
5892 * here and hit the 'default' below, so only disallow
5893 * 'md' ioctls, and switch to rw mode if started auto-readonly.
5895 if (_IOC_TYPE(cmd) == MD_MAJOR && mddev->ro && mddev->pers) {
5896 if (mddev->ro == 2) {
5898 sysfs_notify_dirent_safe(mddev->sysfs_state);
5899 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5900 md_wakeup_thread(mddev->thread);
5911 mdu_disk_info_t info;
5912 if (copy_from_user(&info, argp, sizeof(info)))
5915 err = add_new_disk(mddev, &info);
5919 case HOT_REMOVE_DISK:
5920 err = hot_remove_disk(mddev, new_decode_dev(arg));
5924 err = hot_add_disk(mddev, new_decode_dev(arg));
5927 case SET_DISK_FAULTY:
5928 err = set_disk_faulty(mddev, new_decode_dev(arg));
5932 err = do_md_run(mddev);
5935 case SET_BITMAP_FILE:
5936 err = set_bitmap_file(mddev, (int)arg);
5946 if (mddev->hold_active == UNTIL_IOCTL &&
5948 mddev->hold_active = 0;
5949 mddev_unlock(mddev);
5958 #ifdef CONFIG_COMPAT
5959 static int md_compat_ioctl(struct block_device *bdev, fmode_t mode,
5960 unsigned int cmd, unsigned long arg)
5963 case HOT_REMOVE_DISK:
5965 case SET_DISK_FAULTY:
5966 case SET_BITMAP_FILE:
5967 /* These take in integer arg, do not convert */
5970 arg = (unsigned long)compat_ptr(arg);
5974 return md_ioctl(bdev, mode, cmd, arg);
5976 #endif /* CONFIG_COMPAT */
5978 static int md_open(struct block_device *bdev, fmode_t mode)
5981 * Succeed if we can lock the mddev, which confirms that
5982 * it isn't being stopped right now.
5984 mddev_t *mddev = mddev_find(bdev->bd_dev);
5987 if (mddev->gendisk != bdev->bd_disk) {
5988 /* we are racing with mddev_put which is discarding this
5992 /* Wait until bdev->bd_disk is definitely gone */
5993 flush_workqueue(md_misc_wq);
5994 /* Then retry the open from the top */
5995 return -ERESTARTSYS;
5997 BUG_ON(mddev != bdev->bd_disk->private_data);
5999 if ((err = mutex_lock_interruptible(&mddev->open_mutex)))
6003 atomic_inc(&mddev->openers);
6004 mutex_unlock(&mddev->open_mutex);
6006 check_disk_size_change(mddev->gendisk, bdev);
6011 static int md_release(struct gendisk *disk, fmode_t mode)
6013 mddev_t *mddev = disk->private_data;
6016 atomic_dec(&mddev->openers);
6021 static const struct block_device_operations md_fops =
6023 .owner = THIS_MODULE,
6025 .release = md_release,
6027 #ifdef CONFIG_COMPAT
6028 .compat_ioctl = md_compat_ioctl,
6030 .getgeo = md_getgeo,
6033 static int md_thread(void * arg)
6035 mdk_thread_t *thread = arg;
6038 * md_thread is a 'system-thread', it's priority should be very
6039 * high. We avoid resource deadlocks individually in each
6040 * raid personality. (RAID5 does preallocation) We also use RR and
6041 * the very same RT priority as kswapd, thus we will never get
6042 * into a priority inversion deadlock.
6044 * we definitely have to have equal or higher priority than
6045 * bdflush, otherwise bdflush will deadlock if there are too
6046 * many dirty RAID5 blocks.
6049 allow_signal(SIGKILL);
6050 while (!kthread_should_stop()) {
6052 /* We need to wait INTERRUPTIBLE so that
6053 * we don't add to the load-average.
6054 * That means we need to be sure no signals are
6057 if (signal_pending(current))
6058 flush_signals(current);
6060 wait_event_interruptible_timeout
6062 test_bit(THREAD_WAKEUP, &thread->flags)
6063 || kthread_should_stop(),
6066 clear_bit(THREAD_WAKEUP, &thread->flags);
6067 if (!kthread_should_stop())
6068 thread->run(thread->mddev);
6074 void md_wakeup_thread(mdk_thread_t *thread)
6077 dprintk("md: waking up MD thread %s.\n", thread->tsk->comm);
6078 set_bit(THREAD_WAKEUP, &thread->flags);
6079 wake_up(&thread->wqueue);
6083 mdk_thread_t *md_register_thread(void (*run) (mddev_t *), mddev_t *mddev,
6086 mdk_thread_t *thread;
6088 thread = kzalloc(sizeof(mdk_thread_t), GFP_KERNEL);
6092 init_waitqueue_head(&thread->wqueue);
6095 thread->mddev = mddev;
6096 thread->timeout = MAX_SCHEDULE_TIMEOUT;
6097 thread->tsk = kthread_run(md_thread, thread,
6099 mdname(thread->mddev),
6100 name ?: mddev->pers->name);
6101 if (IS_ERR(thread->tsk)) {
6108 void md_unregister_thread(mdk_thread_t *thread)
6112 dprintk("interrupting MD-thread pid %d\n", task_pid_nr(thread->tsk));
6114 kthread_stop(thread->tsk);
6118 void md_error(mddev_t *mddev, mdk_rdev_t *rdev)
6125 if (!rdev || test_bit(Faulty, &rdev->flags))
6128 if (mddev->external)
6129 set_bit(Blocked, &rdev->flags);
6131 dprintk("md_error dev:%s, rdev:(%d:%d), (caller: %p,%p,%p,%p).\n",
6133 MAJOR(rdev->bdev->bd_dev), MINOR(rdev->bdev->bd_dev),
6134 __builtin_return_address(0),__builtin_return_address(1),
6135 __builtin_return_address(2),__builtin_return_address(3));
6139 if (!mddev->pers->error_handler)
6141 mddev->pers->error_handler(mddev,rdev);
6142 if (mddev->degraded)
6143 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
6144 sysfs_notify_dirent_safe(rdev->sysfs_state);
6145 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6146 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6147 md_wakeup_thread(mddev->thread);
6148 if (mddev->event_work.func)
6149 queue_work(md_misc_wq, &mddev->event_work);
6150 md_new_event_inintr(mddev);
6153 /* seq_file implementation /proc/mdstat */
6155 static void status_unused(struct seq_file *seq)
6160 seq_printf(seq, "unused devices: ");
6162 list_for_each_entry(rdev, &pending_raid_disks, same_set) {
6163 char b[BDEVNAME_SIZE];
6165 seq_printf(seq, "%s ",
6166 bdevname(rdev->bdev,b));
6169 seq_printf(seq, "<none>");
6171 seq_printf(seq, "\n");
6175 static void status_resync(struct seq_file *seq, mddev_t * mddev)
6177 sector_t max_sectors, resync, res;
6178 unsigned long dt, db;
6181 unsigned int per_milli;
6183 resync = mddev->curr_resync - atomic_read(&mddev->recovery_active);
6185 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
6186 max_sectors = mddev->resync_max_sectors;
6188 max_sectors = mddev->dev_sectors;
6191 * Should not happen.
6197 /* Pick 'scale' such that (resync>>scale)*1000 will fit
6198 * in a sector_t, and (max_sectors>>scale) will fit in a
6199 * u32, as those are the requirements for sector_div.
6200 * Thus 'scale' must be at least 10
6203 if (sizeof(sector_t) > sizeof(unsigned long)) {
6204 while ( max_sectors/2 > (1ULL<<(scale+32)))
6207 res = (resync>>scale)*1000;
6208 sector_div(res, (u32)((max_sectors>>scale)+1));
6212 int i, x = per_milli/50, y = 20-x;
6213 seq_printf(seq, "[");
6214 for (i = 0; i < x; i++)
6215 seq_printf(seq, "=");
6216 seq_printf(seq, ">");
6217 for (i = 0; i < y; i++)
6218 seq_printf(seq, ".");
6219 seq_printf(seq, "] ");
6221 seq_printf(seq, " %s =%3u.%u%% (%llu/%llu)",
6222 (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)?
6224 (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)?
6226 (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
6227 "resync" : "recovery"))),
6228 per_milli/10, per_milli % 10,
6229 (unsigned long long) resync/2,
6230 (unsigned long long) max_sectors/2);
6233 * dt: time from mark until now
6234 * db: blocks written from mark until now
6235 * rt: remaining time
6237 * rt is a sector_t, so could be 32bit or 64bit.
6238 * So we divide before multiply in case it is 32bit and close
6240 * We scale the divisor (db) by 32 to avoid loosing precision
6241 * near the end of resync when the number of remaining sectors
6243 * We then divide rt by 32 after multiplying by db to compensate.
6244 * The '+1' avoids division by zero if db is very small.
6246 dt = ((jiffies - mddev->resync_mark) / HZ);
6248 db = (mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active))
6249 - mddev->resync_mark_cnt;
6251 rt = max_sectors - resync; /* number of remaining sectors */
6252 sector_div(rt, db/32+1);
6256 seq_printf(seq, " finish=%lu.%lumin", (unsigned long)rt / 60,
6257 ((unsigned long)rt % 60)/6);
6259 seq_printf(seq, " speed=%ldK/sec", db/2/dt);
6262 static void *md_seq_start(struct seq_file *seq, loff_t *pos)
6264 struct list_head *tmp;
6274 spin_lock(&all_mddevs_lock);
6275 list_for_each(tmp,&all_mddevs)
6277 mddev = list_entry(tmp, mddev_t, all_mddevs);
6279 spin_unlock(&all_mddevs_lock);
6282 spin_unlock(&all_mddevs_lock);
6284 return (void*)2;/* tail */
6288 static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
6290 struct list_head *tmp;
6291 mddev_t *next_mddev, *mddev = v;
6297 spin_lock(&all_mddevs_lock);
6299 tmp = all_mddevs.next;
6301 tmp = mddev->all_mddevs.next;
6302 if (tmp != &all_mddevs)
6303 next_mddev = mddev_get(list_entry(tmp,mddev_t,all_mddevs));
6305 next_mddev = (void*)2;
6308 spin_unlock(&all_mddevs_lock);
6316 static void md_seq_stop(struct seq_file *seq, void *v)
6320 if (mddev && v != (void*)1 && v != (void*)2)
6324 struct mdstat_info {
6328 static int md_seq_show(struct seq_file *seq, void *v)
6333 struct mdstat_info *mi = seq->private;
6334 struct bitmap *bitmap;
6336 if (v == (void*)1) {
6337 struct mdk_personality *pers;
6338 seq_printf(seq, "Personalities : ");
6339 spin_lock(&pers_lock);
6340 list_for_each_entry(pers, &pers_list, list)
6341 seq_printf(seq, "[%s] ", pers->name);
6343 spin_unlock(&pers_lock);
6344 seq_printf(seq, "\n");
6345 mi->event = atomic_read(&md_event_count);
6348 if (v == (void*)2) {
6353 if (mddev_lock(mddev) < 0)
6356 if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
6357 seq_printf(seq, "%s : %sactive", mdname(mddev),
6358 mddev->pers ? "" : "in");
6361 seq_printf(seq, " (read-only)");
6363 seq_printf(seq, " (auto-read-only)");
6364 seq_printf(seq, " %s", mddev->pers->name);
6368 list_for_each_entry(rdev, &mddev->disks, same_set) {
6369 char b[BDEVNAME_SIZE];
6370 seq_printf(seq, " %s[%d]",
6371 bdevname(rdev->bdev,b), rdev->desc_nr);
6372 if (test_bit(WriteMostly, &rdev->flags))
6373 seq_printf(seq, "(W)");
6374 if (test_bit(Faulty, &rdev->flags)) {
6375 seq_printf(seq, "(F)");
6377 } else if (rdev->raid_disk < 0)
6378 seq_printf(seq, "(S)"); /* spare */
6379 sectors += rdev->sectors;
6382 if (!list_empty(&mddev->disks)) {
6384 seq_printf(seq, "\n %llu blocks",
6385 (unsigned long long)
6386 mddev->array_sectors / 2);
6388 seq_printf(seq, "\n %llu blocks",
6389 (unsigned long long)sectors / 2);
6391 if (mddev->persistent) {
6392 if (mddev->major_version != 0 ||
6393 mddev->minor_version != 90) {
6394 seq_printf(seq," super %d.%d",
6395 mddev->major_version,
6396 mddev->minor_version);
6398 } else if (mddev->external)
6399 seq_printf(seq, " super external:%s",
6400 mddev->metadata_type);
6402 seq_printf(seq, " super non-persistent");
6405 mddev->pers->status(seq, mddev);
6406 seq_printf(seq, "\n ");
6407 if (mddev->pers->sync_request) {
6408 if (mddev->curr_resync > 2) {
6409 status_resync(seq, mddev);
6410 seq_printf(seq, "\n ");
6411 } else if (mddev->curr_resync == 1 || mddev->curr_resync == 2)
6412 seq_printf(seq, "\tresync=DELAYED\n ");
6413 else if (mddev->recovery_cp < MaxSector)
6414 seq_printf(seq, "\tresync=PENDING\n ");
6417 seq_printf(seq, "\n ");
6419 if ((bitmap = mddev->bitmap)) {
6420 unsigned long chunk_kb;
6421 unsigned long flags;
6422 spin_lock_irqsave(&bitmap->lock, flags);
6423 chunk_kb = mddev->bitmap_info.chunksize >> 10;
6424 seq_printf(seq, "bitmap: %lu/%lu pages [%luKB], "
6426 bitmap->pages - bitmap->missing_pages,
6428 (bitmap->pages - bitmap->missing_pages)
6429 << (PAGE_SHIFT - 10),
6430 chunk_kb ? chunk_kb : mddev->bitmap_info.chunksize,
6431 chunk_kb ? "KB" : "B");
6433 seq_printf(seq, ", file: ");
6434 seq_path(seq, &bitmap->file->f_path, " \t\n");
6437 seq_printf(seq, "\n");
6438 spin_unlock_irqrestore(&bitmap->lock, flags);
6441 seq_printf(seq, "\n");
6443 mddev_unlock(mddev);
6448 static const struct seq_operations md_seq_ops = {
6449 .start = md_seq_start,
6450 .next = md_seq_next,
6451 .stop = md_seq_stop,
6452 .show = md_seq_show,
6455 static int md_seq_open(struct inode *inode, struct file *file)
6458 struct mdstat_info *mi = kmalloc(sizeof(*mi), GFP_KERNEL);
6462 error = seq_open(file, &md_seq_ops);
6466 struct seq_file *p = file->private_data;
6468 mi->event = atomic_read(&md_event_count);
6473 static unsigned int mdstat_poll(struct file *filp, poll_table *wait)
6475 struct seq_file *m = filp->private_data;
6476 struct mdstat_info *mi = m->private;
6479 poll_wait(filp, &md_event_waiters, wait);
6481 /* always allow read */
6482 mask = POLLIN | POLLRDNORM;
6484 if (mi->event != atomic_read(&md_event_count))
6485 mask |= POLLERR | POLLPRI;
6489 static const struct file_operations md_seq_fops = {
6490 .owner = THIS_MODULE,
6491 .open = md_seq_open,
6493 .llseek = seq_lseek,
6494 .release = seq_release_private,
6495 .poll = mdstat_poll,
6498 int register_md_personality(struct mdk_personality *p)
6500 spin_lock(&pers_lock);
6501 list_add_tail(&p->list, &pers_list);
6502 printk(KERN_INFO "md: %s personality registered for level %d\n", p->name, p->level);
6503 spin_unlock(&pers_lock);
6507 int unregister_md_personality(struct mdk_personality *p)
6509 printk(KERN_INFO "md: %s personality unregistered\n", p->name);
6510 spin_lock(&pers_lock);
6511 list_del_init(&p->list);
6512 spin_unlock(&pers_lock);
6516 static int is_mddev_idle(mddev_t *mddev, int init)
6524 rdev_for_each_rcu(rdev, mddev) {
6525 struct gendisk *disk = rdev->bdev->bd_contains->bd_disk;
6526 curr_events = (int)part_stat_read(&disk->part0, sectors[0]) +
6527 (int)part_stat_read(&disk->part0, sectors[1]) -
6528 atomic_read(&disk->sync_io);
6529 /* sync IO will cause sync_io to increase before the disk_stats
6530 * as sync_io is counted when a request starts, and
6531 * disk_stats is counted when it completes.
6532 * So resync activity will cause curr_events to be smaller than
6533 * when there was no such activity.
6534 * non-sync IO will cause disk_stat to increase without
6535 * increasing sync_io so curr_events will (eventually)
6536 * be larger than it was before. Once it becomes
6537 * substantially larger, the test below will cause
6538 * the array to appear non-idle, and resync will slow
6540 * If there is a lot of outstanding resync activity when
6541 * we set last_event to curr_events, then all that activity
6542 * completing might cause the array to appear non-idle
6543 * and resync will be slowed down even though there might
6544 * not have been non-resync activity. This will only
6545 * happen once though. 'last_events' will soon reflect
6546 * the state where there is little or no outstanding
6547 * resync requests, and further resync activity will
6548 * always make curr_events less than last_events.
6551 if (init || curr_events - rdev->last_events > 64) {
6552 rdev->last_events = curr_events;
6560 void md_done_sync(mddev_t *mddev, int blocks, int ok)
6562 /* another "blocks" (512byte) blocks have been synced */
6563 atomic_sub(blocks, &mddev->recovery_active);
6564 wake_up(&mddev->recovery_wait);
6566 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6567 md_wakeup_thread(mddev->thread);
6568 // stop recovery, signal do_sync ....
6573 /* md_write_start(mddev, bi)
6574 * If we need to update some array metadata (e.g. 'active' flag
6575 * in superblock) before writing, schedule a superblock update
6576 * and wait for it to complete.
6578 void md_write_start(mddev_t *mddev, struct bio *bi)
6581 if (bio_data_dir(bi) != WRITE)
6584 BUG_ON(mddev->ro == 1);
6585 if (mddev->ro == 2) {
6586 /* need to switch to read/write */
6588 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6589 md_wakeup_thread(mddev->thread);
6590 md_wakeup_thread(mddev->sync_thread);
6593 atomic_inc(&mddev->writes_pending);
6594 if (mddev->safemode == 1)
6595 mddev->safemode = 0;
6596 if (mddev->in_sync) {
6597 spin_lock_irq(&mddev->write_lock);
6598 if (mddev->in_sync) {
6600 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
6601 set_bit(MD_CHANGE_PENDING, &mddev->flags);
6602 md_wakeup_thread(mddev->thread);
6605 spin_unlock_irq(&mddev->write_lock);
6608 sysfs_notify_dirent_safe(mddev->sysfs_state);
6609 wait_event(mddev->sb_wait,
6610 !test_bit(MD_CHANGE_PENDING, &mddev->flags));
6613 void md_write_end(mddev_t *mddev)
6615 if (atomic_dec_and_test(&mddev->writes_pending)) {
6616 if (mddev->safemode == 2)
6617 md_wakeup_thread(mddev->thread);
6618 else if (mddev->safemode_delay)
6619 mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay);
6623 /* md_allow_write(mddev)
6624 * Calling this ensures that the array is marked 'active' so that writes
6625 * may proceed without blocking. It is important to call this before
6626 * attempting a GFP_KERNEL allocation while holding the mddev lock.
6627 * Must be called with mddev_lock held.
6629 * In the ->external case MD_CHANGE_CLEAN can not be cleared until mddev->lock
6630 * is dropped, so return -EAGAIN after notifying userspace.
6632 int md_allow_write(mddev_t *mddev)
6638 if (!mddev->pers->sync_request)
6641 spin_lock_irq(&mddev->write_lock);
6642 if (mddev->in_sync) {
6644 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
6645 set_bit(MD_CHANGE_PENDING, &mddev->flags);
6646 if (mddev->safemode_delay &&
6647 mddev->safemode == 0)
6648 mddev->safemode = 1;
6649 spin_unlock_irq(&mddev->write_lock);
6650 md_update_sb(mddev, 0);
6651 sysfs_notify_dirent_safe(mddev->sysfs_state);
6653 spin_unlock_irq(&mddev->write_lock);
6655 if (test_bit(MD_CHANGE_PENDING, &mddev->flags))
6660 EXPORT_SYMBOL_GPL(md_allow_write);
6662 void md_unplug(mddev_t *mddev)
6665 blk_unplug(mddev->queue);
6667 mddev->plug->unplug_fn(mddev->plug);
6670 #define SYNC_MARKS 10
6671 #define SYNC_MARK_STEP (3*HZ)
6672 void md_do_sync(mddev_t *mddev)
6675 unsigned int currspeed = 0,
6677 sector_t max_sectors,j, io_sectors;
6678 unsigned long mark[SYNC_MARKS];
6679 sector_t mark_cnt[SYNC_MARKS];
6681 struct list_head *tmp;
6682 sector_t last_check;
6687 /* just incase thread restarts... */
6688 if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
6690 if (mddev->ro) /* never try to sync a read-only array */
6693 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
6694 if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
6695 desc = "data-check";
6696 else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
6697 desc = "requested-resync";
6700 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
6705 /* we overload curr_resync somewhat here.
6706 * 0 == not engaged in resync at all
6707 * 2 == checking that there is no conflict with another sync
6708 * 1 == like 2, but have yielded to allow conflicting resync to
6710 * other == active in resync - this many blocks
6712 * Before starting a resync we must have set curr_resync to
6713 * 2, and then checked that every "conflicting" array has curr_resync
6714 * less than ours. When we find one that is the same or higher
6715 * we wait on resync_wait. To avoid deadlock, we reduce curr_resync
6716 * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
6717 * This will mean we have to start checking from the beginning again.
6722 mddev->curr_resync = 2;
6725 if (kthread_should_stop())
6726 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6728 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
6730 for_each_mddev(mddev2, tmp) {
6731 if (mddev2 == mddev)
6733 if (!mddev->parallel_resync
6734 && mddev2->curr_resync
6735 && match_mddev_units(mddev, mddev2)) {
6737 if (mddev < mddev2 && mddev->curr_resync == 2) {
6738 /* arbitrarily yield */
6739 mddev->curr_resync = 1;
6740 wake_up(&resync_wait);
6742 if (mddev > mddev2 && mddev->curr_resync == 1)
6743 /* no need to wait here, we can wait the next
6744 * time 'round when curr_resync == 2
6747 /* We need to wait 'interruptible' so as not to
6748 * contribute to the load average, and not to
6749 * be caught by 'softlockup'
6751 prepare_to_wait(&resync_wait, &wq, TASK_INTERRUPTIBLE);
6752 if (!kthread_should_stop() &&
6753 mddev2->curr_resync >= mddev->curr_resync) {
6754 printk(KERN_INFO "md: delaying %s of %s"
6755 " until %s has finished (they"
6756 " share one or more physical units)\n",
6757 desc, mdname(mddev), mdname(mddev2));
6759 if (signal_pending(current))
6760 flush_signals(current);
6762 finish_wait(&resync_wait, &wq);
6765 finish_wait(&resync_wait, &wq);
6768 } while (mddev->curr_resync < 2);
6771 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
6772 /* resync follows the size requested by the personality,
6773 * which defaults to physical size, but can be virtual size
6775 max_sectors = mddev->resync_max_sectors;
6776 mddev->resync_mismatches = 0;
6777 /* we don't use the checkpoint if there's a bitmap */
6778 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
6779 j = mddev->resync_min;
6780 else if (!mddev->bitmap)
6781 j = mddev->recovery_cp;
6783 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
6784 max_sectors = mddev->dev_sectors;
6786 /* recovery follows the physical size of devices */
6787 max_sectors = mddev->dev_sectors;
6790 list_for_each_entry_rcu(rdev, &mddev->disks, same_set)
6791 if (rdev->raid_disk >= 0 &&
6792 !test_bit(Faulty, &rdev->flags) &&
6793 !test_bit(In_sync, &rdev->flags) &&
6794 rdev->recovery_offset < j)
6795 j = rdev->recovery_offset;
6799 printk(KERN_INFO "md: %s of RAID array %s\n", desc, mdname(mddev));
6800 printk(KERN_INFO "md: minimum _guaranteed_ speed:"
6801 " %d KB/sec/disk.\n", speed_min(mddev));
6802 printk(KERN_INFO "md: using maximum available idle IO bandwidth "
6803 "(but not more than %d KB/sec) for %s.\n",
6804 speed_max(mddev), desc);
6806 is_mddev_idle(mddev, 1); /* this initializes IO event counters */
6809 for (m = 0; m < SYNC_MARKS; m++) {
6811 mark_cnt[m] = io_sectors;
6814 mddev->resync_mark = mark[last_mark];
6815 mddev->resync_mark_cnt = mark_cnt[last_mark];
6818 * Tune reconstruction:
6820 window = 32*(PAGE_SIZE/512);
6821 printk(KERN_INFO "md: using %dk window, over a total of %llu blocks.\n",
6822 window/2,(unsigned long long) max_sectors/2);
6824 atomic_set(&mddev->recovery_active, 0);
6829 "md: resuming %s of %s from checkpoint.\n",
6830 desc, mdname(mddev));
6831 mddev->curr_resync = j;
6833 mddev->curr_resync_completed = j;
6835 while (j < max_sectors) {
6840 if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
6841 ((mddev->curr_resync > mddev->curr_resync_completed &&
6842 (mddev->curr_resync - mddev->curr_resync_completed)
6843 > (max_sectors >> 4)) ||
6844 (j - mddev->curr_resync_completed)*2
6845 >= mddev->resync_max - mddev->curr_resync_completed
6847 /* time to update curr_resync_completed */
6849 wait_event(mddev->recovery_wait,
6850 atomic_read(&mddev->recovery_active) == 0);
6851 mddev->curr_resync_completed = j;
6852 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
6853 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
6856 while (j >= mddev->resync_max && !kthread_should_stop()) {
6857 /* As this condition is controlled by user-space,
6858 * we can block indefinitely, so use '_interruptible'
6859 * to avoid triggering warnings.
6861 flush_signals(current); /* just in case */
6862 wait_event_interruptible(mddev->recovery_wait,
6863 mddev->resync_max > j
6864 || kthread_should_stop());
6867 if (kthread_should_stop())
6870 sectors = mddev->pers->sync_request(mddev, j, &skipped,
6871 currspeed < speed_min(mddev));
6873 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6877 if (!skipped) { /* actual IO requested */
6878 io_sectors += sectors;
6879 atomic_add(sectors, &mddev->recovery_active);
6883 if (j>1) mddev->curr_resync = j;
6884 mddev->curr_mark_cnt = io_sectors;
6885 if (last_check == 0)
6886 /* this is the earliers that rebuilt will be
6887 * visible in /proc/mdstat
6889 md_new_event(mddev);
6891 if (last_check + window > io_sectors || j == max_sectors)
6894 last_check = io_sectors;
6896 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
6900 if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) {
6902 int next = (last_mark+1) % SYNC_MARKS;
6904 mddev->resync_mark = mark[next];
6905 mddev->resync_mark_cnt = mark_cnt[next];
6906 mark[next] = jiffies;
6907 mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active);
6912 if (kthread_should_stop())
6917 * this loop exits only if either when we are slower than
6918 * the 'hard' speed limit, or the system was IO-idle for
6920 * the system might be non-idle CPU-wise, but we only care
6921 * about not overloading the IO subsystem. (things like an
6922 * e2fsck being done on the RAID array should execute fast)
6927 currspeed = ((unsigned long)(io_sectors-mddev->resync_mark_cnt))/2
6928 /((jiffies-mddev->resync_mark)/HZ +1) +1;
6930 if (currspeed > speed_min(mddev)) {
6931 if ((currspeed > speed_max(mddev)) ||
6932 !is_mddev_idle(mddev, 0)) {
6938 printk(KERN_INFO "md: %s: %s done.\n",mdname(mddev), desc);
6940 * this also signals 'finished resyncing' to md_stop
6945 wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
6947 /* tell personality that we are finished */
6948 mddev->pers->sync_request(mddev, max_sectors, &skipped, 1);
6950 if (!test_bit(MD_RECOVERY_CHECK, &mddev->recovery) &&
6951 mddev->curr_resync > 2) {
6952 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
6953 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
6954 if (mddev->curr_resync >= mddev->recovery_cp) {
6956 "md: checkpointing %s of %s.\n",
6957 desc, mdname(mddev));
6958 mddev->recovery_cp = mddev->curr_resync;
6961 mddev->recovery_cp = MaxSector;
6963 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery))
6964 mddev->curr_resync = MaxSector;
6966 list_for_each_entry_rcu(rdev, &mddev->disks, same_set)
6967 if (rdev->raid_disk >= 0 &&
6968 mddev->delta_disks >= 0 &&
6969 !test_bit(Faulty, &rdev->flags) &&
6970 !test_bit(In_sync, &rdev->flags) &&
6971 rdev->recovery_offset < mddev->curr_resync)
6972 rdev->recovery_offset = mddev->curr_resync;
6976 set_bit(MD_CHANGE_DEVS, &mddev->flags);
6979 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
6980 /* We completed so min/max setting can be forgotten if used. */
6981 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
6982 mddev->resync_min = 0;
6983 mddev->resync_max = MaxSector;
6984 } else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
6985 mddev->resync_min = mddev->curr_resync_completed;
6986 mddev->curr_resync = 0;
6987 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery))
6988 mddev->curr_resync_completed = 0;
6989 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
6990 wake_up(&resync_wait);
6991 set_bit(MD_RECOVERY_DONE, &mddev->recovery);
6992 md_wakeup_thread(mddev->thread);
6997 * got a signal, exit.
7000 "md: md_do_sync() got signal ... exiting\n");
7001 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
7005 EXPORT_SYMBOL_GPL(md_do_sync);
7008 static int remove_and_add_spares(mddev_t *mddev)
7013 mddev->curr_resync_completed = 0;
7015 list_for_each_entry(rdev, &mddev->disks, same_set)
7016 if (rdev->raid_disk >= 0 &&
7017 !test_bit(Blocked, &rdev->flags) &&
7018 (test_bit(Faulty, &rdev->flags) ||
7019 ! test_bit(In_sync, &rdev->flags)) &&
7020 atomic_read(&rdev->nr_pending)==0) {
7021 if (mddev->pers->hot_remove_disk(
7022 mddev, rdev->raid_disk)==0) {
7024 sprintf(nm,"rd%d", rdev->raid_disk);
7025 sysfs_remove_link(&mddev->kobj, nm);
7026 rdev->raid_disk = -1;
7030 if (mddev->degraded && ! mddev->ro && !mddev->recovery_disabled) {
7031 list_for_each_entry(rdev, &mddev->disks, same_set) {
7032 if (rdev->raid_disk >= 0 &&
7033 !test_bit(In_sync, &rdev->flags) &&
7034 !test_bit(Blocked, &rdev->flags))
7036 if (rdev->raid_disk < 0
7037 && !test_bit(Faulty, &rdev->flags)) {
7038 rdev->recovery_offset = 0;
7040 hot_add_disk(mddev, rdev) == 0) {
7042 sprintf(nm, "rd%d", rdev->raid_disk);
7043 if (sysfs_create_link(&mddev->kobj,
7045 /* failure here is OK */;
7047 md_new_event(mddev);
7048 set_bit(MD_CHANGE_DEVS, &mddev->flags);
7057 static void reap_sync_thread(mddev_t *mddev)
7061 /* resync has finished, collect result */
7062 md_unregister_thread(mddev->sync_thread);
7063 mddev->sync_thread = NULL;
7064 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
7065 !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
7067 /* activate any spares */
7068 if (mddev->pers->spare_active(mddev))
7069 sysfs_notify(&mddev->kobj, NULL,
7072 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
7073 mddev->pers->finish_reshape)
7074 mddev->pers->finish_reshape(mddev);
7075 md_update_sb(mddev, 1);
7077 /* if array is no-longer degraded, then any saved_raid_disk
7078 * information must be scrapped
7080 if (!mddev->degraded)
7081 list_for_each_entry(rdev, &mddev->disks, same_set)
7082 rdev->saved_raid_disk = -1;
7084 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
7085 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
7086 clear_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
7087 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
7088 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
7089 /* flag recovery needed just to double check */
7090 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7091 sysfs_notify_dirent_safe(mddev->sysfs_action);
7092 md_new_event(mddev);
7096 * This routine is regularly called by all per-raid-array threads to
7097 * deal with generic issues like resync and super-block update.
7098 * Raid personalities that don't have a thread (linear/raid0) do not
7099 * need this as they never do any recovery or update the superblock.
7101 * It does not do any resync itself, but rather "forks" off other threads
7102 * to do that as needed.
7103 * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
7104 * "->recovery" and create a thread at ->sync_thread.
7105 * When the thread finishes it sets MD_RECOVERY_DONE
7106 * and wakeups up this thread which will reap the thread and finish up.
7107 * This thread also removes any faulty devices (with nr_pending == 0).
7109 * The overall approach is:
7110 * 1/ if the superblock needs updating, update it.
7111 * 2/ If a recovery thread is running, don't do anything else.
7112 * 3/ If recovery has finished, clean up, possibly marking spares active.
7113 * 4/ If there are any faulty devices, remove them.
7114 * 5/ If array is degraded, try to add spares devices
7115 * 6/ If array has spares or is not in-sync, start a resync thread.
7117 void md_check_recovery(mddev_t *mddev)
7120 bitmap_daemon_work(mddev);
7125 if (signal_pending(current)) {
7126 if (mddev->pers->sync_request && !mddev->external) {
7127 printk(KERN_INFO "md: %s in immediate safe mode\n",
7129 mddev->safemode = 2;
7131 flush_signals(current);
7134 if (mddev->ro && !test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
7137 (mddev->flags & ~ (1<<MD_CHANGE_PENDING)) ||
7138 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
7139 test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
7140 (mddev->external == 0 && mddev->safemode == 1) ||
7141 (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending)
7142 && !mddev->in_sync && mddev->recovery_cp == MaxSector)
7146 if (mddev_trylock(mddev)) {
7150 /* Only thing we do on a ro array is remove
7153 remove_and_add_spares(mddev);
7154 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7158 if (!mddev->external) {
7160 spin_lock_irq(&mddev->write_lock);
7161 if (mddev->safemode &&
7162 !atomic_read(&mddev->writes_pending) &&
7164 mddev->recovery_cp == MaxSector) {
7167 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
7169 if (mddev->safemode == 1)
7170 mddev->safemode = 0;
7171 spin_unlock_irq(&mddev->write_lock);
7173 sysfs_notify_dirent_safe(mddev->sysfs_state);
7177 md_update_sb(mddev, 0);
7179 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
7180 !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
7181 /* resync/recovery still happening */
7182 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7185 if (mddev->sync_thread) {
7186 reap_sync_thread(mddev);
7189 /* Set RUNNING before clearing NEEDED to avoid
7190 * any transients in the value of "sync_action".
7192 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
7193 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7194 /* Clear some bits that don't mean anything, but
7197 clear_bit(MD_RECOVERY_INTR, &mddev->recovery);
7198 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
7200 if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
7202 /* no recovery is running.
7203 * remove any failed drives, then
7204 * add spares if possible.
7205 * Spare are also removed and re-added, to allow
7206 * the personality to fail the re-add.
7209 if (mddev->reshape_position != MaxSector) {
7210 if (mddev->pers->check_reshape == NULL ||
7211 mddev->pers->check_reshape(mddev) != 0)
7212 /* Cannot proceed */
7214 set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
7215 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
7216 } else if ((spares = remove_and_add_spares(mddev))) {
7217 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
7218 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
7219 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
7220 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
7221 } else if (mddev->recovery_cp < MaxSector) {
7222 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
7223 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
7224 } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
7225 /* nothing to be done ... */
7228 if (mddev->pers->sync_request) {
7229 if (spares && mddev->bitmap && ! mddev->bitmap->file) {
7230 /* We are adding a device or devices to an array
7231 * which has the bitmap stored on all devices.
7232 * So make sure all bitmap pages get written
7234 bitmap_write_all(mddev->bitmap);
7236 mddev->sync_thread = md_register_thread(md_do_sync,
7239 if (!mddev->sync_thread) {
7240 printk(KERN_ERR "%s: could not start resync"
7243 /* leave the spares where they are, it shouldn't hurt */
7244 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
7245 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
7246 clear_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
7247 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
7248 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
7250 md_wakeup_thread(mddev->sync_thread);
7251 sysfs_notify_dirent_safe(mddev->sysfs_action);
7252 md_new_event(mddev);
7255 if (!mddev->sync_thread) {
7256 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
7257 if (test_and_clear_bit(MD_RECOVERY_RECOVER,
7259 if (mddev->sysfs_action)
7260 sysfs_notify_dirent_safe(mddev->sysfs_action);
7262 mddev_unlock(mddev);
7266 void md_wait_for_blocked_rdev(mdk_rdev_t *rdev, mddev_t *mddev)
7268 sysfs_notify_dirent_safe(rdev->sysfs_state);
7269 wait_event_timeout(rdev->blocked_wait,
7270 !test_bit(Blocked, &rdev->flags),
7271 msecs_to_jiffies(5000));
7272 rdev_dec_pending(rdev, mddev);
7274 EXPORT_SYMBOL(md_wait_for_blocked_rdev);
7276 static int md_notify_reboot(struct notifier_block *this,
7277 unsigned long code, void *x)
7279 struct list_head *tmp;
7282 if ((code == SYS_DOWN) || (code == SYS_HALT) || (code == SYS_POWER_OFF)) {
7284 printk(KERN_INFO "md: stopping all md devices.\n");
7286 for_each_mddev(mddev, tmp)
7287 if (mddev_trylock(mddev)) {
7288 /* Force a switch to readonly even array
7289 * appears to still be in use. Hence
7292 md_set_readonly(mddev, 100);
7293 mddev_unlock(mddev);
7296 * certain more exotic SCSI devices are known to be
7297 * volatile wrt too early system reboots. While the
7298 * right place to handle this issue is the given
7299 * driver, we do want to have a safe RAID driver ...
7306 static struct notifier_block md_notifier = {
7307 .notifier_call = md_notify_reboot,
7309 .priority = INT_MAX, /* before any real devices */
7312 static void md_geninit(void)
7314 dprintk("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
7316 proc_create("mdstat", S_IRUGO, NULL, &md_seq_fops);
7319 static int __init md_init(void)
7323 md_wq = alloc_workqueue("md", WQ_RESCUER, 0);
7327 md_misc_wq = alloc_workqueue("md_misc", 0, 0);
7331 if ((ret = register_blkdev(MD_MAJOR, "md")) < 0)
7334 if ((ret = register_blkdev(0, "mdp")) < 0)
7338 blk_register_region(MKDEV(MD_MAJOR, 0), 1UL<<MINORBITS, THIS_MODULE,
7339 md_probe, NULL, NULL);
7340 blk_register_region(MKDEV(mdp_major, 0), 1UL<<MINORBITS, THIS_MODULE,
7341 md_probe, NULL, NULL);
7343 register_reboot_notifier(&md_notifier);
7344 raid_table_header = register_sysctl_table(raid_root_table);
7350 unregister_blkdev(MD_MAJOR, "md");
7352 destroy_workqueue(md_misc_wq);
7354 destroy_workqueue(md_wq);
7362 * Searches all registered partitions for autorun RAID arrays
7366 static LIST_HEAD(all_detected_devices);
7367 struct detected_devices_node {
7368 struct list_head list;
7372 void md_autodetect_dev(dev_t dev)
7374 struct detected_devices_node *node_detected_dev;
7376 node_detected_dev = kzalloc(sizeof(*node_detected_dev), GFP_KERNEL);
7377 if (node_detected_dev) {
7378 node_detected_dev->dev = dev;
7379 list_add_tail(&node_detected_dev->list, &all_detected_devices);
7381 printk(KERN_CRIT "md: md_autodetect_dev: kzalloc failed"
7382 ", skipping dev(%d,%d)\n", MAJOR(dev), MINOR(dev));
7387 static void autostart_arrays(int part)
7390 struct detected_devices_node *node_detected_dev;
7392 int i_scanned, i_passed;
7397 printk(KERN_INFO "md: Autodetecting RAID arrays.\n");
7399 while (!list_empty(&all_detected_devices) && i_scanned < INT_MAX) {
7401 node_detected_dev = list_entry(all_detected_devices.next,
7402 struct detected_devices_node, list);
7403 list_del(&node_detected_dev->list);
7404 dev = node_detected_dev->dev;
7405 kfree(node_detected_dev);
7406 rdev = md_import_device(dev,0, 90);
7410 if (test_bit(Faulty, &rdev->flags)) {
7414 set_bit(AutoDetected, &rdev->flags);
7415 list_add(&rdev->same_set, &pending_raid_disks);
7419 printk(KERN_INFO "md: Scanned %d and added %d devices.\n",
7420 i_scanned, i_passed);
7422 autorun_devices(part);
7425 #endif /* !MODULE */
7427 static __exit void md_exit(void)
7430 struct list_head *tmp;
7432 blk_unregister_region(MKDEV(MD_MAJOR,0), 1U << MINORBITS);
7433 blk_unregister_region(MKDEV(mdp_major,0), 1U << MINORBITS);
7435 unregister_blkdev(MD_MAJOR,"md");
7436 unregister_blkdev(mdp_major, "mdp");
7437 unregister_reboot_notifier(&md_notifier);
7438 unregister_sysctl_table(raid_table_header);
7439 remove_proc_entry("mdstat", NULL);
7440 for_each_mddev(mddev, tmp) {
7441 export_array(mddev);
7442 mddev->hold_active = 0;
7444 destroy_workqueue(md_misc_wq);
7445 destroy_workqueue(md_wq);
7448 subsys_initcall(md_init);
7449 module_exit(md_exit)
7451 static int get_ro(char *buffer, struct kernel_param *kp)
7453 return sprintf(buffer, "%d", start_readonly);
7455 static int set_ro(const char *val, struct kernel_param *kp)
7458 int num = simple_strtoul(val, &e, 10);
7459 if (*val && (*e == '\0' || *e == '\n')) {
7460 start_readonly = num;
7466 module_param_call(start_ro, set_ro, get_ro, NULL, S_IRUSR|S_IWUSR);
7467 module_param(start_dirty_degraded, int, S_IRUGO|S_IWUSR);
7469 module_param_call(new_array, add_named_array, NULL, NULL, S_IWUSR);
7471 EXPORT_SYMBOL(register_md_personality);
7472 EXPORT_SYMBOL(unregister_md_personality);
7473 EXPORT_SYMBOL(md_error);
7474 EXPORT_SYMBOL(md_done_sync);
7475 EXPORT_SYMBOL(md_write_start);
7476 EXPORT_SYMBOL(md_write_end);
7477 EXPORT_SYMBOL(md_register_thread);
7478 EXPORT_SYMBOL(md_unregister_thread);
7479 EXPORT_SYMBOL(md_wakeup_thread);
7480 EXPORT_SYMBOL(md_check_recovery);
7481 MODULE_LICENSE("GPL");
7482 MODULE_DESCRIPTION("MD RAID framework");
7484 MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR);