Use dentry_path() to create full path to inode object
[pohmelfs.git] / drivers / md / md.c
blobce88755baf4a91a6216b628117e84a697c21d3c8
1 /*
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
7 Changes:
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)
28 any later version.
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/fs.h>
40 #include <linux/poll.h>
41 #include <linux/ctype.h>
42 #include <linux/string.h>
43 #include <linux/hdreg.h>
44 #include <linux/proc_fs.h>
45 #include <linux/random.h>
46 #include <linux/module.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>
54 #include "md.h"
55 #include "bitmap.h"
57 #ifndef MODULE
58 static void autostart_arrays(int part);
59 #endif
61 /* pers_list is a list of registered personalities protected
62 * by pers_lock.
63 * pers_lock does extra service to protect accesses to
64 * mddev->thread when the mutex cannot be held.
66 static LIST_HEAD(pers_list);
67 static DEFINE_SPINLOCK(pers_lock);
69 static void md_print_devices(void);
71 static DECLARE_WAIT_QUEUE_HEAD(resync_wait);
72 static struct workqueue_struct *md_wq;
73 static struct workqueue_struct *md_misc_wq;
75 #define MD_BUG(x...) { printk("md: bug in file %s, line %d\n", __FILE__, __LINE__); md_print_devices(); }
78 * Default number of read corrections we'll attempt on an rdev
79 * before ejecting it from the array. We divide the read error
80 * count by 2 for every hour elapsed between read errors.
82 #define MD_DEFAULT_MAX_CORRECTED_READ_ERRORS 20
84 * Current RAID-1,4,5 parallel reconstruction 'guaranteed speed limit'
85 * is 1000 KB/sec, so the extra system load does not show up that much.
86 * Increase it if you want to have more _guaranteed_ speed. Note that
87 * the RAID driver will use the maximum available bandwidth if the IO
88 * subsystem is idle. There is also an 'absolute maximum' reconstruction
89 * speed limit - in case reconstruction slows down your system despite
90 * idle IO detection.
92 * you can change it via /proc/sys/dev/raid/speed_limit_min and _max.
93 * or /sys/block/mdX/md/sync_speed_{min,max}
96 static int sysctl_speed_limit_min = 1000;
97 static int sysctl_speed_limit_max = 200000;
98 static inline int speed_min(struct mddev *mddev)
100 return mddev->sync_speed_min ?
101 mddev->sync_speed_min : sysctl_speed_limit_min;
104 static inline int speed_max(struct mddev *mddev)
106 return mddev->sync_speed_max ?
107 mddev->sync_speed_max : sysctl_speed_limit_max;
110 static struct ctl_table_header *raid_table_header;
112 static ctl_table raid_table[] = {
114 .procname = "speed_limit_min",
115 .data = &sysctl_speed_limit_min,
116 .maxlen = sizeof(int),
117 .mode = S_IRUGO|S_IWUSR,
118 .proc_handler = proc_dointvec,
121 .procname = "speed_limit_max",
122 .data = &sysctl_speed_limit_max,
123 .maxlen = sizeof(int),
124 .mode = S_IRUGO|S_IWUSR,
125 .proc_handler = proc_dointvec,
130 static ctl_table raid_dir_table[] = {
132 .procname = "raid",
133 .maxlen = 0,
134 .mode = S_IRUGO|S_IXUGO,
135 .child = raid_table,
140 static ctl_table raid_root_table[] = {
142 .procname = "dev",
143 .maxlen = 0,
144 .mode = 0555,
145 .child = raid_dir_table,
150 static const struct block_device_operations md_fops;
152 static int start_readonly;
154 /* bio_clone_mddev
155 * like bio_clone, but with a local bio set
158 static void mddev_bio_destructor(struct bio *bio)
160 struct mddev *mddev, **mddevp;
162 mddevp = (void*)bio;
163 mddev = mddevp[-1];
165 bio_free(bio, mddev->bio_set);
168 struct bio *bio_alloc_mddev(gfp_t gfp_mask, int nr_iovecs,
169 struct mddev *mddev)
171 struct bio *b;
172 struct mddev **mddevp;
174 if (!mddev || !mddev->bio_set)
175 return bio_alloc(gfp_mask, nr_iovecs);
177 b = bio_alloc_bioset(gfp_mask, nr_iovecs,
178 mddev->bio_set);
179 if (!b)
180 return NULL;
181 mddevp = (void*)b;
182 mddevp[-1] = mddev;
183 b->bi_destructor = mddev_bio_destructor;
184 return b;
186 EXPORT_SYMBOL_GPL(bio_alloc_mddev);
188 struct bio *bio_clone_mddev(struct bio *bio, gfp_t gfp_mask,
189 struct mddev *mddev)
191 struct bio *b;
192 struct mddev **mddevp;
194 if (!mddev || !mddev->bio_set)
195 return bio_clone(bio, gfp_mask);
197 b = bio_alloc_bioset(gfp_mask, bio->bi_max_vecs,
198 mddev->bio_set);
199 if (!b)
200 return NULL;
201 mddevp = (void*)b;
202 mddevp[-1] = mddev;
203 b->bi_destructor = mddev_bio_destructor;
204 __bio_clone(b, bio);
205 if (bio_integrity(bio)) {
206 int ret;
208 ret = bio_integrity_clone(b, bio, gfp_mask, mddev->bio_set);
210 if (ret < 0) {
211 bio_put(b);
212 return NULL;
216 return b;
218 EXPORT_SYMBOL_GPL(bio_clone_mddev);
220 void md_trim_bio(struct bio *bio, int offset, int size)
222 /* 'bio' is a cloned bio which we need to trim to match
223 * the given offset and size.
224 * This requires adjusting bi_sector, bi_size, and bi_io_vec
226 int i;
227 struct bio_vec *bvec;
228 int sofar = 0;
230 size <<= 9;
231 if (offset == 0 && size == bio->bi_size)
232 return;
234 bio->bi_sector += offset;
235 bio->bi_size = size;
236 offset <<= 9;
237 clear_bit(BIO_SEG_VALID, &bio->bi_flags);
239 while (bio->bi_idx < bio->bi_vcnt &&
240 bio->bi_io_vec[bio->bi_idx].bv_len <= offset) {
241 /* remove this whole bio_vec */
242 offset -= bio->bi_io_vec[bio->bi_idx].bv_len;
243 bio->bi_idx++;
245 if (bio->bi_idx < bio->bi_vcnt) {
246 bio->bi_io_vec[bio->bi_idx].bv_offset += offset;
247 bio->bi_io_vec[bio->bi_idx].bv_len -= offset;
249 /* avoid any complications with bi_idx being non-zero*/
250 if (bio->bi_idx) {
251 memmove(bio->bi_io_vec, bio->bi_io_vec+bio->bi_idx,
252 (bio->bi_vcnt - bio->bi_idx) * sizeof(struct bio_vec));
253 bio->bi_vcnt -= bio->bi_idx;
254 bio->bi_idx = 0;
256 /* Make sure vcnt and last bv are not too big */
257 bio_for_each_segment(bvec, bio, i) {
258 if (sofar + bvec->bv_len > size)
259 bvec->bv_len = size - sofar;
260 if (bvec->bv_len == 0) {
261 bio->bi_vcnt = i;
262 break;
264 sofar += bvec->bv_len;
267 EXPORT_SYMBOL_GPL(md_trim_bio);
270 * We have a system wide 'event count' that is incremented
271 * on any 'interesting' event, and readers of /proc/mdstat
272 * can use 'poll' or 'select' to find out when the event
273 * count increases.
275 * Events are:
276 * start array, stop array, error, add device, remove device,
277 * start build, activate spare
279 static DECLARE_WAIT_QUEUE_HEAD(md_event_waiters);
280 static atomic_t md_event_count;
281 void md_new_event(struct mddev *mddev)
283 atomic_inc(&md_event_count);
284 wake_up(&md_event_waiters);
286 EXPORT_SYMBOL_GPL(md_new_event);
288 /* Alternate version that can be called from interrupts
289 * when calling sysfs_notify isn't needed.
291 static void md_new_event_inintr(struct mddev *mddev)
293 atomic_inc(&md_event_count);
294 wake_up(&md_event_waiters);
298 * Enables to iterate over all existing md arrays
299 * all_mddevs_lock protects this list.
301 static LIST_HEAD(all_mddevs);
302 static DEFINE_SPINLOCK(all_mddevs_lock);
306 * iterates through all used mddevs in the system.
307 * We take care to grab the all_mddevs_lock whenever navigating
308 * the list, and to always hold a refcount when unlocked.
309 * Any code which breaks out of this loop while own
310 * a reference to the current mddev and must mddev_put it.
312 #define for_each_mddev(_mddev,_tmp) \
314 for (({ spin_lock(&all_mddevs_lock); \
315 _tmp = all_mddevs.next; \
316 _mddev = NULL;}); \
317 ({ if (_tmp != &all_mddevs) \
318 mddev_get(list_entry(_tmp, struct mddev, all_mddevs));\
319 spin_unlock(&all_mddevs_lock); \
320 if (_mddev) mddev_put(_mddev); \
321 _mddev = list_entry(_tmp, struct mddev, all_mddevs); \
322 _tmp != &all_mddevs;}); \
323 ({ spin_lock(&all_mddevs_lock); \
324 _tmp = _tmp->next;}) \
328 /* Rather than calling directly into the personality make_request function,
329 * IO requests come here first so that we can check if the device is
330 * being suspended pending a reconfiguration.
331 * We hold a refcount over the call to ->make_request. By the time that
332 * call has finished, the bio has been linked into some internal structure
333 * and so is visible to ->quiesce(), so we don't need the refcount any more.
335 static void md_make_request(struct request_queue *q, struct bio *bio)
337 const int rw = bio_data_dir(bio);
338 struct mddev *mddev = q->queuedata;
339 int cpu;
340 unsigned int sectors;
342 if (mddev == NULL || mddev->pers == NULL
343 || !mddev->ready) {
344 bio_io_error(bio);
345 return;
347 smp_rmb(); /* Ensure implications of 'active' are visible */
348 rcu_read_lock();
349 if (mddev->suspended) {
350 DEFINE_WAIT(__wait);
351 for (;;) {
352 prepare_to_wait(&mddev->sb_wait, &__wait,
353 TASK_UNINTERRUPTIBLE);
354 if (!mddev->suspended)
355 break;
356 rcu_read_unlock();
357 schedule();
358 rcu_read_lock();
360 finish_wait(&mddev->sb_wait, &__wait);
362 atomic_inc(&mddev->active_io);
363 rcu_read_unlock();
366 * save the sectors now since our bio can
367 * go away inside make_request
369 sectors = bio_sectors(bio);
370 mddev->pers->make_request(mddev, bio);
372 cpu = part_stat_lock();
373 part_stat_inc(cpu, &mddev->gendisk->part0, ios[rw]);
374 part_stat_add(cpu, &mddev->gendisk->part0, sectors[rw], sectors);
375 part_stat_unlock();
377 if (atomic_dec_and_test(&mddev->active_io) && mddev->suspended)
378 wake_up(&mddev->sb_wait);
381 /* mddev_suspend makes sure no new requests are submitted
382 * to the device, and that any requests that have been submitted
383 * are completely handled.
384 * Once ->stop is called and completes, the module will be completely
385 * unused.
387 void mddev_suspend(struct mddev *mddev)
389 BUG_ON(mddev->suspended);
390 mddev->suspended = 1;
391 synchronize_rcu();
392 wait_event(mddev->sb_wait, atomic_read(&mddev->active_io) == 0);
393 mddev->pers->quiesce(mddev, 1);
395 EXPORT_SYMBOL_GPL(mddev_suspend);
397 void mddev_resume(struct mddev *mddev)
399 mddev->suspended = 0;
400 wake_up(&mddev->sb_wait);
401 mddev->pers->quiesce(mddev, 0);
403 md_wakeup_thread(mddev->thread);
404 md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
406 EXPORT_SYMBOL_GPL(mddev_resume);
408 int mddev_congested(struct mddev *mddev, int bits)
410 return mddev->suspended;
412 EXPORT_SYMBOL(mddev_congested);
415 * Generic flush handling for md
418 static void md_end_flush(struct bio *bio, int err)
420 struct md_rdev *rdev = bio->bi_private;
421 struct mddev *mddev = rdev->mddev;
423 rdev_dec_pending(rdev, mddev);
425 if (atomic_dec_and_test(&mddev->flush_pending)) {
426 /* The pre-request flush has finished */
427 queue_work(md_wq, &mddev->flush_work);
429 bio_put(bio);
432 static void md_submit_flush_data(struct work_struct *ws);
434 static void submit_flushes(struct work_struct *ws)
436 struct mddev *mddev = container_of(ws, struct mddev, flush_work);
437 struct md_rdev *rdev;
439 INIT_WORK(&mddev->flush_work, md_submit_flush_data);
440 atomic_set(&mddev->flush_pending, 1);
441 rcu_read_lock();
442 list_for_each_entry_rcu(rdev, &mddev->disks, same_set)
443 if (rdev->raid_disk >= 0 &&
444 !test_bit(Faulty, &rdev->flags)) {
445 /* Take two references, one is dropped
446 * when request finishes, one after
447 * we reclaim rcu_read_lock
449 struct bio *bi;
450 atomic_inc(&rdev->nr_pending);
451 atomic_inc(&rdev->nr_pending);
452 rcu_read_unlock();
453 bi = bio_alloc_mddev(GFP_KERNEL, 0, mddev);
454 bi->bi_end_io = md_end_flush;
455 bi->bi_private = rdev;
456 bi->bi_bdev = rdev->bdev;
457 atomic_inc(&mddev->flush_pending);
458 submit_bio(WRITE_FLUSH, bi);
459 rcu_read_lock();
460 rdev_dec_pending(rdev, mddev);
462 rcu_read_unlock();
463 if (atomic_dec_and_test(&mddev->flush_pending))
464 queue_work(md_wq, &mddev->flush_work);
467 static void md_submit_flush_data(struct work_struct *ws)
469 struct mddev *mddev = container_of(ws, struct mddev, flush_work);
470 struct bio *bio = mddev->flush_bio;
472 if (bio->bi_size == 0)
473 /* an empty barrier - all done */
474 bio_endio(bio, 0);
475 else {
476 bio->bi_rw &= ~REQ_FLUSH;
477 mddev->pers->make_request(mddev, bio);
480 mddev->flush_bio = NULL;
481 wake_up(&mddev->sb_wait);
484 void md_flush_request(struct mddev *mddev, struct bio *bio)
486 spin_lock_irq(&mddev->write_lock);
487 wait_event_lock_irq(mddev->sb_wait,
488 !mddev->flush_bio,
489 mddev->write_lock, /*nothing*/);
490 mddev->flush_bio = bio;
491 spin_unlock_irq(&mddev->write_lock);
493 INIT_WORK(&mddev->flush_work, submit_flushes);
494 queue_work(md_wq, &mddev->flush_work);
496 EXPORT_SYMBOL(md_flush_request);
498 /* Support for plugging.
499 * This mirrors the plugging support in request_queue, but does not
500 * require having a whole queue or request structures.
501 * We allocate an md_plug_cb for each md device and each thread it gets
502 * plugged on. This links tot the private plug_handle structure in the
503 * personality data where we keep a count of the number of outstanding
504 * plugs so other code can see if a plug is active.
506 struct md_plug_cb {
507 struct blk_plug_cb cb;
508 struct mddev *mddev;
511 static void plugger_unplug(struct blk_plug_cb *cb)
513 struct md_plug_cb *mdcb = container_of(cb, struct md_plug_cb, cb);
514 if (atomic_dec_and_test(&mdcb->mddev->plug_cnt))
515 md_wakeup_thread(mdcb->mddev->thread);
516 kfree(mdcb);
519 /* Check that an unplug wakeup will come shortly.
520 * If not, wakeup the md thread immediately
522 int mddev_check_plugged(struct mddev *mddev)
524 struct blk_plug *plug = current->plug;
525 struct md_plug_cb *mdcb;
527 if (!plug)
528 return 0;
530 list_for_each_entry(mdcb, &plug->cb_list, cb.list) {
531 if (mdcb->cb.callback == plugger_unplug &&
532 mdcb->mddev == mddev) {
533 /* Already on the list, move to top */
534 if (mdcb != list_first_entry(&plug->cb_list,
535 struct md_plug_cb,
536 cb.list))
537 list_move(&mdcb->cb.list, &plug->cb_list);
538 return 1;
541 /* Not currently on the callback list */
542 mdcb = kmalloc(sizeof(*mdcb), GFP_ATOMIC);
543 if (!mdcb)
544 return 0;
546 mdcb->mddev = mddev;
547 mdcb->cb.callback = plugger_unplug;
548 atomic_inc(&mddev->plug_cnt);
549 list_add(&mdcb->cb.list, &plug->cb_list);
550 return 1;
552 EXPORT_SYMBOL_GPL(mddev_check_plugged);
554 static inline struct mddev *mddev_get(struct mddev *mddev)
556 atomic_inc(&mddev->active);
557 return mddev;
560 static void mddev_delayed_delete(struct work_struct *ws);
562 static void mddev_put(struct mddev *mddev)
564 struct bio_set *bs = NULL;
566 if (!atomic_dec_and_lock(&mddev->active, &all_mddevs_lock))
567 return;
568 if (!mddev->raid_disks && list_empty(&mddev->disks) &&
569 mddev->ctime == 0 && !mddev->hold_active) {
570 /* Array is not configured at all, and not held active,
571 * so destroy it */
572 list_del_init(&mddev->all_mddevs);
573 bs = mddev->bio_set;
574 mddev->bio_set = NULL;
575 if (mddev->gendisk) {
576 /* We did a probe so need to clean up. Call
577 * queue_work inside the spinlock so that
578 * flush_workqueue() after mddev_find will
579 * succeed in waiting for the work to be done.
581 INIT_WORK(&mddev->del_work, mddev_delayed_delete);
582 queue_work(md_misc_wq, &mddev->del_work);
583 } else
584 kfree(mddev);
586 spin_unlock(&all_mddevs_lock);
587 if (bs)
588 bioset_free(bs);
591 void mddev_init(struct mddev *mddev)
593 mutex_init(&mddev->open_mutex);
594 mutex_init(&mddev->reconfig_mutex);
595 mutex_init(&mddev->bitmap_info.mutex);
596 INIT_LIST_HEAD(&mddev->disks);
597 INIT_LIST_HEAD(&mddev->all_mddevs);
598 init_timer(&mddev->safemode_timer);
599 atomic_set(&mddev->active, 1);
600 atomic_set(&mddev->openers, 0);
601 atomic_set(&mddev->active_io, 0);
602 atomic_set(&mddev->plug_cnt, 0);
603 spin_lock_init(&mddev->write_lock);
604 atomic_set(&mddev->flush_pending, 0);
605 init_waitqueue_head(&mddev->sb_wait);
606 init_waitqueue_head(&mddev->recovery_wait);
607 mddev->reshape_position = MaxSector;
608 mddev->resync_min = 0;
609 mddev->resync_max = MaxSector;
610 mddev->level = LEVEL_NONE;
612 EXPORT_SYMBOL_GPL(mddev_init);
614 static struct mddev * mddev_find(dev_t unit)
616 struct mddev *mddev, *new = NULL;
618 if (unit && MAJOR(unit) != MD_MAJOR)
619 unit &= ~((1<<MdpMinorShift)-1);
621 retry:
622 spin_lock(&all_mddevs_lock);
624 if (unit) {
625 list_for_each_entry(mddev, &all_mddevs, all_mddevs)
626 if (mddev->unit == unit) {
627 mddev_get(mddev);
628 spin_unlock(&all_mddevs_lock);
629 kfree(new);
630 return mddev;
633 if (new) {
634 list_add(&new->all_mddevs, &all_mddevs);
635 spin_unlock(&all_mddevs_lock);
636 new->hold_active = UNTIL_IOCTL;
637 return new;
639 } else if (new) {
640 /* find an unused unit number */
641 static int next_minor = 512;
642 int start = next_minor;
643 int is_free = 0;
644 int dev = 0;
645 while (!is_free) {
646 dev = MKDEV(MD_MAJOR, next_minor);
647 next_minor++;
648 if (next_minor > MINORMASK)
649 next_minor = 0;
650 if (next_minor == start) {
651 /* Oh dear, all in use. */
652 spin_unlock(&all_mddevs_lock);
653 kfree(new);
654 return NULL;
657 is_free = 1;
658 list_for_each_entry(mddev, &all_mddevs, all_mddevs)
659 if (mddev->unit == dev) {
660 is_free = 0;
661 break;
664 new->unit = dev;
665 new->md_minor = MINOR(dev);
666 new->hold_active = UNTIL_STOP;
667 list_add(&new->all_mddevs, &all_mddevs);
668 spin_unlock(&all_mddevs_lock);
669 return new;
671 spin_unlock(&all_mddevs_lock);
673 new = kzalloc(sizeof(*new), GFP_KERNEL);
674 if (!new)
675 return NULL;
677 new->unit = unit;
678 if (MAJOR(unit) == MD_MAJOR)
679 new->md_minor = MINOR(unit);
680 else
681 new->md_minor = MINOR(unit) >> MdpMinorShift;
683 mddev_init(new);
685 goto retry;
688 static inline int mddev_lock(struct mddev * mddev)
690 return mutex_lock_interruptible(&mddev->reconfig_mutex);
693 static inline int mddev_is_locked(struct mddev *mddev)
695 return mutex_is_locked(&mddev->reconfig_mutex);
698 static inline int mddev_trylock(struct mddev * mddev)
700 return mutex_trylock(&mddev->reconfig_mutex);
703 static struct attribute_group md_redundancy_group;
705 static void mddev_unlock(struct mddev * mddev)
707 if (mddev->to_remove) {
708 /* These cannot be removed under reconfig_mutex as
709 * an access to the files will try to take reconfig_mutex
710 * while holding the file unremovable, which leads to
711 * a deadlock.
712 * So hold set sysfs_active while the remove in happeing,
713 * and anything else which might set ->to_remove or my
714 * otherwise change the sysfs namespace will fail with
715 * -EBUSY if sysfs_active is still set.
716 * We set sysfs_active under reconfig_mutex and elsewhere
717 * test it under the same mutex to ensure its correct value
718 * is seen.
720 struct attribute_group *to_remove = mddev->to_remove;
721 mddev->to_remove = NULL;
722 mddev->sysfs_active = 1;
723 mutex_unlock(&mddev->reconfig_mutex);
725 if (mddev->kobj.sd) {
726 if (to_remove != &md_redundancy_group)
727 sysfs_remove_group(&mddev->kobj, to_remove);
728 if (mddev->pers == NULL ||
729 mddev->pers->sync_request == NULL) {
730 sysfs_remove_group(&mddev->kobj, &md_redundancy_group);
731 if (mddev->sysfs_action)
732 sysfs_put(mddev->sysfs_action);
733 mddev->sysfs_action = NULL;
736 mddev->sysfs_active = 0;
737 } else
738 mutex_unlock(&mddev->reconfig_mutex);
740 /* As we've dropped the mutex we need a spinlock to
741 * make sure the thread doesn't disappear
743 spin_lock(&pers_lock);
744 md_wakeup_thread(mddev->thread);
745 spin_unlock(&pers_lock);
748 static struct md_rdev * find_rdev_nr(struct mddev *mddev, int nr)
750 struct md_rdev *rdev;
752 list_for_each_entry(rdev, &mddev->disks, same_set)
753 if (rdev->desc_nr == nr)
754 return rdev;
756 return NULL;
759 static struct md_rdev * find_rdev(struct mddev * mddev, dev_t dev)
761 struct md_rdev *rdev;
763 list_for_each_entry(rdev, &mddev->disks, same_set)
764 if (rdev->bdev->bd_dev == dev)
765 return rdev;
767 return NULL;
770 static struct md_personality *find_pers(int level, char *clevel)
772 struct md_personality *pers;
773 list_for_each_entry(pers, &pers_list, list) {
774 if (level != LEVEL_NONE && pers->level == level)
775 return pers;
776 if (strcmp(pers->name, clevel)==0)
777 return pers;
779 return NULL;
782 /* return the offset of the super block in 512byte sectors */
783 static inline sector_t calc_dev_sboffset(struct md_rdev *rdev)
785 sector_t num_sectors = i_size_read(rdev->bdev->bd_inode) / 512;
786 return MD_NEW_SIZE_SECTORS(num_sectors);
789 static int alloc_disk_sb(struct md_rdev * rdev)
791 if (rdev->sb_page)
792 MD_BUG();
794 rdev->sb_page = alloc_page(GFP_KERNEL);
795 if (!rdev->sb_page) {
796 printk(KERN_ALERT "md: out of memory.\n");
797 return -ENOMEM;
800 return 0;
803 static void free_disk_sb(struct md_rdev * rdev)
805 if (rdev->sb_page) {
806 put_page(rdev->sb_page);
807 rdev->sb_loaded = 0;
808 rdev->sb_page = NULL;
809 rdev->sb_start = 0;
810 rdev->sectors = 0;
812 if (rdev->bb_page) {
813 put_page(rdev->bb_page);
814 rdev->bb_page = NULL;
819 static void super_written(struct bio *bio, int error)
821 struct md_rdev *rdev = bio->bi_private;
822 struct mddev *mddev = rdev->mddev;
824 if (error || !test_bit(BIO_UPTODATE, &bio->bi_flags)) {
825 printk("md: super_written gets error=%d, uptodate=%d\n",
826 error, test_bit(BIO_UPTODATE, &bio->bi_flags));
827 WARN_ON(test_bit(BIO_UPTODATE, &bio->bi_flags));
828 md_error(mddev, rdev);
831 if (atomic_dec_and_test(&mddev->pending_writes))
832 wake_up(&mddev->sb_wait);
833 bio_put(bio);
836 void md_super_write(struct mddev *mddev, struct md_rdev *rdev,
837 sector_t sector, int size, struct page *page)
839 /* write first size bytes of page to sector of rdev
840 * Increment mddev->pending_writes before returning
841 * and decrement it on completion, waking up sb_wait
842 * if zero is reached.
843 * If an error occurred, call md_error
845 struct bio *bio = bio_alloc_mddev(GFP_NOIO, 1, mddev);
847 bio->bi_bdev = rdev->meta_bdev ? rdev->meta_bdev : rdev->bdev;
848 bio->bi_sector = sector;
849 bio_add_page(bio, page, size, 0);
850 bio->bi_private = rdev;
851 bio->bi_end_io = super_written;
853 atomic_inc(&mddev->pending_writes);
854 submit_bio(WRITE_FLUSH_FUA, bio);
857 void md_super_wait(struct mddev *mddev)
859 /* wait for all superblock writes that were scheduled to complete */
860 DEFINE_WAIT(wq);
861 for(;;) {
862 prepare_to_wait(&mddev->sb_wait, &wq, TASK_UNINTERRUPTIBLE);
863 if (atomic_read(&mddev->pending_writes)==0)
864 break;
865 schedule();
867 finish_wait(&mddev->sb_wait, &wq);
870 static void bi_complete(struct bio *bio, int error)
872 complete((struct completion*)bio->bi_private);
875 int sync_page_io(struct md_rdev *rdev, sector_t sector, int size,
876 struct page *page, int rw, bool metadata_op)
878 struct bio *bio = bio_alloc_mddev(GFP_NOIO, 1, rdev->mddev);
879 struct completion event;
880 int ret;
882 rw |= REQ_SYNC;
884 bio->bi_bdev = (metadata_op && rdev->meta_bdev) ?
885 rdev->meta_bdev : rdev->bdev;
886 if (metadata_op)
887 bio->bi_sector = sector + rdev->sb_start;
888 else
889 bio->bi_sector = sector + rdev->data_offset;
890 bio_add_page(bio, page, size, 0);
891 init_completion(&event);
892 bio->bi_private = &event;
893 bio->bi_end_io = bi_complete;
894 submit_bio(rw, bio);
895 wait_for_completion(&event);
897 ret = test_bit(BIO_UPTODATE, &bio->bi_flags);
898 bio_put(bio);
899 return ret;
901 EXPORT_SYMBOL_GPL(sync_page_io);
903 static int read_disk_sb(struct md_rdev * rdev, int size)
905 char b[BDEVNAME_SIZE];
906 if (!rdev->sb_page) {
907 MD_BUG();
908 return -EINVAL;
910 if (rdev->sb_loaded)
911 return 0;
914 if (!sync_page_io(rdev, 0, size, rdev->sb_page, READ, true))
915 goto fail;
916 rdev->sb_loaded = 1;
917 return 0;
919 fail:
920 printk(KERN_WARNING "md: disabled device %s, could not read superblock.\n",
921 bdevname(rdev->bdev,b));
922 return -EINVAL;
925 static int uuid_equal(mdp_super_t *sb1, mdp_super_t *sb2)
927 return sb1->set_uuid0 == sb2->set_uuid0 &&
928 sb1->set_uuid1 == sb2->set_uuid1 &&
929 sb1->set_uuid2 == sb2->set_uuid2 &&
930 sb1->set_uuid3 == sb2->set_uuid3;
933 static int sb_equal(mdp_super_t *sb1, mdp_super_t *sb2)
935 int ret;
936 mdp_super_t *tmp1, *tmp2;
938 tmp1 = kmalloc(sizeof(*tmp1),GFP_KERNEL);
939 tmp2 = kmalloc(sizeof(*tmp2),GFP_KERNEL);
941 if (!tmp1 || !tmp2) {
942 ret = 0;
943 printk(KERN_INFO "md.c sb_equal(): failed to allocate memory!\n");
944 goto abort;
947 *tmp1 = *sb1;
948 *tmp2 = *sb2;
951 * nr_disks is not constant
953 tmp1->nr_disks = 0;
954 tmp2->nr_disks = 0;
956 ret = (memcmp(tmp1, tmp2, MD_SB_GENERIC_CONSTANT_WORDS * 4) == 0);
957 abort:
958 kfree(tmp1);
959 kfree(tmp2);
960 return ret;
964 static u32 md_csum_fold(u32 csum)
966 csum = (csum & 0xffff) + (csum >> 16);
967 return (csum & 0xffff) + (csum >> 16);
970 static unsigned int calc_sb_csum(mdp_super_t * sb)
972 u64 newcsum = 0;
973 u32 *sb32 = (u32*)sb;
974 int i;
975 unsigned int disk_csum, csum;
977 disk_csum = sb->sb_csum;
978 sb->sb_csum = 0;
980 for (i = 0; i < MD_SB_BYTES/4 ; i++)
981 newcsum += sb32[i];
982 csum = (newcsum & 0xffffffff) + (newcsum>>32);
985 #ifdef CONFIG_ALPHA
986 /* This used to use csum_partial, which was wrong for several
987 * reasons including that different results are returned on
988 * different architectures. It isn't critical that we get exactly
989 * the same return value as before (we always csum_fold before
990 * testing, and that removes any differences). However as we
991 * know that csum_partial always returned a 16bit value on
992 * alphas, do a fold to maximise conformity to previous behaviour.
994 sb->sb_csum = md_csum_fold(disk_csum);
995 #else
996 sb->sb_csum = disk_csum;
997 #endif
998 return csum;
1003 * Handle superblock details.
1004 * We want to be able to handle multiple superblock formats
1005 * so we have a common interface to them all, and an array of
1006 * different handlers.
1007 * We rely on user-space to write the initial superblock, and support
1008 * reading and updating of superblocks.
1009 * Interface methods are:
1010 * int load_super(struct md_rdev *dev, struct md_rdev *refdev, int minor_version)
1011 * loads and validates a superblock on dev.
1012 * if refdev != NULL, compare superblocks on both devices
1013 * Return:
1014 * 0 - dev has a superblock that is compatible with refdev
1015 * 1 - dev has a superblock that is compatible and newer than refdev
1016 * so dev should be used as the refdev in future
1017 * -EINVAL superblock incompatible or invalid
1018 * -othererror e.g. -EIO
1020 * int validate_super(struct mddev *mddev, struct md_rdev *dev)
1021 * Verify that dev is acceptable into mddev.
1022 * The first time, mddev->raid_disks will be 0, and data from
1023 * dev should be merged in. Subsequent calls check that dev
1024 * is new enough. Return 0 or -EINVAL
1026 * void sync_super(struct mddev *mddev, struct md_rdev *dev)
1027 * Update the superblock for rdev with data in mddev
1028 * This does not write to disc.
1032 struct super_type {
1033 char *name;
1034 struct module *owner;
1035 int (*load_super)(struct md_rdev *rdev, struct md_rdev *refdev,
1036 int minor_version);
1037 int (*validate_super)(struct mddev *mddev, struct md_rdev *rdev);
1038 void (*sync_super)(struct mddev *mddev, struct md_rdev *rdev);
1039 unsigned long long (*rdev_size_change)(struct md_rdev *rdev,
1040 sector_t num_sectors);
1044 * Check that the given mddev has no bitmap.
1046 * This function is called from the run method of all personalities that do not
1047 * support bitmaps. It prints an error message and returns non-zero if mddev
1048 * has a bitmap. Otherwise, it returns 0.
1051 int md_check_no_bitmap(struct mddev *mddev)
1053 if (!mddev->bitmap_info.file && !mddev->bitmap_info.offset)
1054 return 0;
1055 printk(KERN_ERR "%s: bitmaps are not supported for %s\n",
1056 mdname(mddev), mddev->pers->name);
1057 return 1;
1059 EXPORT_SYMBOL(md_check_no_bitmap);
1062 * load_super for 0.90.0
1064 static int super_90_load(struct md_rdev *rdev, struct md_rdev *refdev, int minor_version)
1066 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
1067 mdp_super_t *sb;
1068 int ret;
1071 * Calculate the position of the superblock (512byte sectors),
1072 * it's at the end of the disk.
1074 * It also happens to be a multiple of 4Kb.
1076 rdev->sb_start = calc_dev_sboffset(rdev);
1078 ret = read_disk_sb(rdev, MD_SB_BYTES);
1079 if (ret) return ret;
1081 ret = -EINVAL;
1083 bdevname(rdev->bdev, b);
1084 sb = page_address(rdev->sb_page);
1086 if (sb->md_magic != MD_SB_MAGIC) {
1087 printk(KERN_ERR "md: invalid raid superblock magic on %s\n",
1089 goto abort;
1092 if (sb->major_version != 0 ||
1093 sb->minor_version < 90 ||
1094 sb->minor_version > 91) {
1095 printk(KERN_WARNING "Bad version number %d.%d on %s\n",
1096 sb->major_version, sb->minor_version,
1098 goto abort;
1101 if (sb->raid_disks <= 0)
1102 goto abort;
1104 if (md_csum_fold(calc_sb_csum(sb)) != md_csum_fold(sb->sb_csum)) {
1105 printk(KERN_WARNING "md: invalid superblock checksum on %s\n",
1107 goto abort;
1110 rdev->preferred_minor = sb->md_minor;
1111 rdev->data_offset = 0;
1112 rdev->sb_size = MD_SB_BYTES;
1113 rdev->badblocks.shift = -1;
1115 if (sb->level == LEVEL_MULTIPATH)
1116 rdev->desc_nr = -1;
1117 else
1118 rdev->desc_nr = sb->this_disk.number;
1120 if (!refdev) {
1121 ret = 1;
1122 } else {
1123 __u64 ev1, ev2;
1124 mdp_super_t *refsb = page_address(refdev->sb_page);
1125 if (!uuid_equal(refsb, sb)) {
1126 printk(KERN_WARNING "md: %s has different UUID to %s\n",
1127 b, bdevname(refdev->bdev,b2));
1128 goto abort;
1130 if (!sb_equal(refsb, sb)) {
1131 printk(KERN_WARNING "md: %s has same UUID"
1132 " but different superblock to %s\n",
1133 b, bdevname(refdev->bdev, b2));
1134 goto abort;
1136 ev1 = md_event(sb);
1137 ev2 = md_event(refsb);
1138 if (ev1 > ev2)
1139 ret = 1;
1140 else
1141 ret = 0;
1143 rdev->sectors = rdev->sb_start;
1144 /* Limit to 4TB as metadata cannot record more than that */
1145 if (rdev->sectors >= (2ULL << 32))
1146 rdev->sectors = (2ULL << 32) - 2;
1148 if (rdev->sectors < ((sector_t)sb->size) * 2 && sb->level >= 1)
1149 /* "this cannot possibly happen" ... */
1150 ret = -EINVAL;
1152 abort:
1153 return ret;
1157 * validate_super for 0.90.0
1159 static int super_90_validate(struct mddev *mddev, struct md_rdev *rdev)
1161 mdp_disk_t *desc;
1162 mdp_super_t *sb = page_address(rdev->sb_page);
1163 __u64 ev1 = md_event(sb);
1165 rdev->raid_disk = -1;
1166 clear_bit(Faulty, &rdev->flags);
1167 clear_bit(In_sync, &rdev->flags);
1168 clear_bit(WriteMostly, &rdev->flags);
1170 if (mddev->raid_disks == 0) {
1171 mddev->major_version = 0;
1172 mddev->minor_version = sb->minor_version;
1173 mddev->patch_version = sb->patch_version;
1174 mddev->external = 0;
1175 mddev->chunk_sectors = sb->chunk_size >> 9;
1176 mddev->ctime = sb->ctime;
1177 mddev->utime = sb->utime;
1178 mddev->level = sb->level;
1179 mddev->clevel[0] = 0;
1180 mddev->layout = sb->layout;
1181 mddev->raid_disks = sb->raid_disks;
1182 mddev->dev_sectors = ((sector_t)sb->size) * 2;
1183 mddev->events = ev1;
1184 mddev->bitmap_info.offset = 0;
1185 mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9;
1187 if (mddev->minor_version >= 91) {
1188 mddev->reshape_position = sb->reshape_position;
1189 mddev->delta_disks = sb->delta_disks;
1190 mddev->new_level = sb->new_level;
1191 mddev->new_layout = sb->new_layout;
1192 mddev->new_chunk_sectors = sb->new_chunk >> 9;
1193 } else {
1194 mddev->reshape_position = MaxSector;
1195 mddev->delta_disks = 0;
1196 mddev->new_level = mddev->level;
1197 mddev->new_layout = mddev->layout;
1198 mddev->new_chunk_sectors = mddev->chunk_sectors;
1201 if (sb->state & (1<<MD_SB_CLEAN))
1202 mddev->recovery_cp = MaxSector;
1203 else {
1204 if (sb->events_hi == sb->cp_events_hi &&
1205 sb->events_lo == sb->cp_events_lo) {
1206 mddev->recovery_cp = sb->recovery_cp;
1207 } else
1208 mddev->recovery_cp = 0;
1211 memcpy(mddev->uuid+0, &sb->set_uuid0, 4);
1212 memcpy(mddev->uuid+4, &sb->set_uuid1, 4);
1213 memcpy(mddev->uuid+8, &sb->set_uuid2, 4);
1214 memcpy(mddev->uuid+12,&sb->set_uuid3, 4);
1216 mddev->max_disks = MD_SB_DISKS;
1218 if (sb->state & (1<<MD_SB_BITMAP_PRESENT) &&
1219 mddev->bitmap_info.file == NULL)
1220 mddev->bitmap_info.offset =
1221 mddev->bitmap_info.default_offset;
1223 } else if (mddev->pers == NULL) {
1224 /* Insist on good event counter while assembling, except
1225 * for spares (which don't need an event count) */
1226 ++ev1;
1227 if (sb->disks[rdev->desc_nr].state & (
1228 (1<<MD_DISK_SYNC) | (1 << MD_DISK_ACTIVE)))
1229 if (ev1 < mddev->events)
1230 return -EINVAL;
1231 } else if (mddev->bitmap) {
1232 /* if adding to array with a bitmap, then we can accept an
1233 * older device ... but not too old.
1235 if (ev1 < mddev->bitmap->events_cleared)
1236 return 0;
1237 } else {
1238 if (ev1 < mddev->events)
1239 /* just a hot-add of a new device, leave raid_disk at -1 */
1240 return 0;
1243 if (mddev->level != LEVEL_MULTIPATH) {
1244 desc = sb->disks + rdev->desc_nr;
1246 if (desc->state & (1<<MD_DISK_FAULTY))
1247 set_bit(Faulty, &rdev->flags);
1248 else if (desc->state & (1<<MD_DISK_SYNC) /* &&
1249 desc->raid_disk < mddev->raid_disks */) {
1250 set_bit(In_sync, &rdev->flags);
1251 rdev->raid_disk = desc->raid_disk;
1252 } else if (desc->state & (1<<MD_DISK_ACTIVE)) {
1253 /* active but not in sync implies recovery up to
1254 * reshape position. We don't know exactly where
1255 * that is, so set to zero for now */
1256 if (mddev->minor_version >= 91) {
1257 rdev->recovery_offset = 0;
1258 rdev->raid_disk = desc->raid_disk;
1261 if (desc->state & (1<<MD_DISK_WRITEMOSTLY))
1262 set_bit(WriteMostly, &rdev->flags);
1263 } else /* MULTIPATH are always insync */
1264 set_bit(In_sync, &rdev->flags);
1265 return 0;
1269 * sync_super for 0.90.0
1271 static void super_90_sync(struct mddev *mddev, struct md_rdev *rdev)
1273 mdp_super_t *sb;
1274 struct md_rdev *rdev2;
1275 int next_spare = mddev->raid_disks;
1278 /* make rdev->sb match mddev data..
1280 * 1/ zero out disks
1281 * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
1282 * 3/ any empty disks < next_spare become removed
1284 * disks[0] gets initialised to REMOVED because
1285 * we cannot be sure from other fields if it has
1286 * been initialised or not.
1288 int i;
1289 int active=0, working=0,failed=0,spare=0,nr_disks=0;
1291 rdev->sb_size = MD_SB_BYTES;
1293 sb = page_address(rdev->sb_page);
1295 memset(sb, 0, sizeof(*sb));
1297 sb->md_magic = MD_SB_MAGIC;
1298 sb->major_version = mddev->major_version;
1299 sb->patch_version = mddev->patch_version;
1300 sb->gvalid_words = 0; /* ignored */
1301 memcpy(&sb->set_uuid0, mddev->uuid+0, 4);
1302 memcpy(&sb->set_uuid1, mddev->uuid+4, 4);
1303 memcpy(&sb->set_uuid2, mddev->uuid+8, 4);
1304 memcpy(&sb->set_uuid3, mddev->uuid+12,4);
1306 sb->ctime = mddev->ctime;
1307 sb->level = mddev->level;
1308 sb->size = mddev->dev_sectors / 2;
1309 sb->raid_disks = mddev->raid_disks;
1310 sb->md_minor = mddev->md_minor;
1311 sb->not_persistent = 0;
1312 sb->utime = mddev->utime;
1313 sb->state = 0;
1314 sb->events_hi = (mddev->events>>32);
1315 sb->events_lo = (u32)mddev->events;
1317 if (mddev->reshape_position == MaxSector)
1318 sb->minor_version = 90;
1319 else {
1320 sb->minor_version = 91;
1321 sb->reshape_position = mddev->reshape_position;
1322 sb->new_level = mddev->new_level;
1323 sb->delta_disks = mddev->delta_disks;
1324 sb->new_layout = mddev->new_layout;
1325 sb->new_chunk = mddev->new_chunk_sectors << 9;
1327 mddev->minor_version = sb->minor_version;
1328 if (mddev->in_sync)
1330 sb->recovery_cp = mddev->recovery_cp;
1331 sb->cp_events_hi = (mddev->events>>32);
1332 sb->cp_events_lo = (u32)mddev->events;
1333 if (mddev->recovery_cp == MaxSector)
1334 sb->state = (1<< MD_SB_CLEAN);
1335 } else
1336 sb->recovery_cp = 0;
1338 sb->layout = mddev->layout;
1339 sb->chunk_size = mddev->chunk_sectors << 9;
1341 if (mddev->bitmap && mddev->bitmap_info.file == NULL)
1342 sb->state |= (1<<MD_SB_BITMAP_PRESENT);
1344 sb->disks[0].state = (1<<MD_DISK_REMOVED);
1345 list_for_each_entry(rdev2, &mddev->disks, same_set) {
1346 mdp_disk_t *d;
1347 int desc_nr;
1348 int is_active = test_bit(In_sync, &rdev2->flags);
1350 if (rdev2->raid_disk >= 0 &&
1351 sb->minor_version >= 91)
1352 /* we have nowhere to store the recovery_offset,
1353 * but if it is not below the reshape_position,
1354 * we can piggy-back on that.
1356 is_active = 1;
1357 if (rdev2->raid_disk < 0 ||
1358 test_bit(Faulty, &rdev2->flags))
1359 is_active = 0;
1360 if (is_active)
1361 desc_nr = rdev2->raid_disk;
1362 else
1363 desc_nr = next_spare++;
1364 rdev2->desc_nr = desc_nr;
1365 d = &sb->disks[rdev2->desc_nr];
1366 nr_disks++;
1367 d->number = rdev2->desc_nr;
1368 d->major = MAJOR(rdev2->bdev->bd_dev);
1369 d->minor = MINOR(rdev2->bdev->bd_dev);
1370 if (is_active)
1371 d->raid_disk = rdev2->raid_disk;
1372 else
1373 d->raid_disk = rdev2->desc_nr; /* compatibility */
1374 if (test_bit(Faulty, &rdev2->flags))
1375 d->state = (1<<MD_DISK_FAULTY);
1376 else if (is_active) {
1377 d->state = (1<<MD_DISK_ACTIVE);
1378 if (test_bit(In_sync, &rdev2->flags))
1379 d->state |= (1<<MD_DISK_SYNC);
1380 active++;
1381 working++;
1382 } else {
1383 d->state = 0;
1384 spare++;
1385 working++;
1387 if (test_bit(WriteMostly, &rdev2->flags))
1388 d->state |= (1<<MD_DISK_WRITEMOSTLY);
1390 /* now set the "removed" and "faulty" bits on any missing devices */
1391 for (i=0 ; i < mddev->raid_disks ; i++) {
1392 mdp_disk_t *d = &sb->disks[i];
1393 if (d->state == 0 && d->number == 0) {
1394 d->number = i;
1395 d->raid_disk = i;
1396 d->state = (1<<MD_DISK_REMOVED);
1397 d->state |= (1<<MD_DISK_FAULTY);
1398 failed++;
1401 sb->nr_disks = nr_disks;
1402 sb->active_disks = active;
1403 sb->working_disks = working;
1404 sb->failed_disks = failed;
1405 sb->spare_disks = spare;
1407 sb->this_disk = sb->disks[rdev->desc_nr];
1408 sb->sb_csum = calc_sb_csum(sb);
1412 * rdev_size_change for 0.90.0
1414 static unsigned long long
1415 super_90_rdev_size_change(struct md_rdev *rdev, sector_t num_sectors)
1417 if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
1418 return 0; /* component must fit device */
1419 if (rdev->mddev->bitmap_info.offset)
1420 return 0; /* can't move bitmap */
1421 rdev->sb_start = calc_dev_sboffset(rdev);
1422 if (!num_sectors || num_sectors > rdev->sb_start)
1423 num_sectors = rdev->sb_start;
1424 /* Limit to 4TB as metadata cannot record more than that.
1425 * 4TB == 2^32 KB, or 2*2^32 sectors.
1427 if (num_sectors >= (2ULL << 32))
1428 num_sectors = (2ULL << 32) - 2;
1429 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
1430 rdev->sb_page);
1431 md_super_wait(rdev->mddev);
1432 return num_sectors;
1437 * version 1 superblock
1440 static __le32 calc_sb_1_csum(struct mdp_superblock_1 * sb)
1442 __le32 disk_csum;
1443 u32 csum;
1444 unsigned long long newcsum;
1445 int size = 256 + le32_to_cpu(sb->max_dev)*2;
1446 __le32 *isuper = (__le32*)sb;
1447 int i;
1449 disk_csum = sb->sb_csum;
1450 sb->sb_csum = 0;
1451 newcsum = 0;
1452 for (i=0; size>=4; size -= 4 )
1453 newcsum += le32_to_cpu(*isuper++);
1455 if (size == 2)
1456 newcsum += le16_to_cpu(*(__le16*) isuper);
1458 csum = (newcsum & 0xffffffff) + (newcsum >> 32);
1459 sb->sb_csum = disk_csum;
1460 return cpu_to_le32(csum);
1463 static int md_set_badblocks(struct badblocks *bb, sector_t s, int sectors,
1464 int acknowledged);
1465 static int super_1_load(struct md_rdev *rdev, struct md_rdev *refdev, int minor_version)
1467 struct mdp_superblock_1 *sb;
1468 int ret;
1469 sector_t sb_start;
1470 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
1471 int bmask;
1474 * Calculate the position of the superblock in 512byte sectors.
1475 * It is always aligned to a 4K boundary and
1476 * depeding on minor_version, it can be:
1477 * 0: At least 8K, but less than 12K, from end of device
1478 * 1: At start of device
1479 * 2: 4K from start of device.
1481 switch(minor_version) {
1482 case 0:
1483 sb_start = i_size_read(rdev->bdev->bd_inode) >> 9;
1484 sb_start -= 8*2;
1485 sb_start &= ~(sector_t)(4*2-1);
1486 break;
1487 case 1:
1488 sb_start = 0;
1489 break;
1490 case 2:
1491 sb_start = 8;
1492 break;
1493 default:
1494 return -EINVAL;
1496 rdev->sb_start = sb_start;
1498 /* superblock is rarely larger than 1K, but it can be larger,
1499 * and it is safe to read 4k, so we do that
1501 ret = read_disk_sb(rdev, 4096);
1502 if (ret) return ret;
1505 sb = page_address(rdev->sb_page);
1507 if (sb->magic != cpu_to_le32(MD_SB_MAGIC) ||
1508 sb->major_version != cpu_to_le32(1) ||
1509 le32_to_cpu(sb->max_dev) > (4096-256)/2 ||
1510 le64_to_cpu(sb->super_offset) != rdev->sb_start ||
1511 (le32_to_cpu(sb->feature_map) & ~MD_FEATURE_ALL) != 0)
1512 return -EINVAL;
1514 if (calc_sb_1_csum(sb) != sb->sb_csum) {
1515 printk("md: invalid superblock checksum on %s\n",
1516 bdevname(rdev->bdev,b));
1517 return -EINVAL;
1519 if (le64_to_cpu(sb->data_size) < 10) {
1520 printk("md: data_size too small on %s\n",
1521 bdevname(rdev->bdev,b));
1522 return -EINVAL;
1525 rdev->preferred_minor = 0xffff;
1526 rdev->data_offset = le64_to_cpu(sb->data_offset);
1527 atomic_set(&rdev->corrected_errors, le32_to_cpu(sb->cnt_corrected_read));
1529 rdev->sb_size = le32_to_cpu(sb->max_dev) * 2 + 256;
1530 bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
1531 if (rdev->sb_size & bmask)
1532 rdev->sb_size = (rdev->sb_size | bmask) + 1;
1534 if (minor_version
1535 && rdev->data_offset < sb_start + (rdev->sb_size/512))
1536 return -EINVAL;
1538 if (sb->level == cpu_to_le32(LEVEL_MULTIPATH))
1539 rdev->desc_nr = -1;
1540 else
1541 rdev->desc_nr = le32_to_cpu(sb->dev_number);
1543 if (!rdev->bb_page) {
1544 rdev->bb_page = alloc_page(GFP_KERNEL);
1545 if (!rdev->bb_page)
1546 return -ENOMEM;
1548 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BAD_BLOCKS) &&
1549 rdev->badblocks.count == 0) {
1550 /* need to load the bad block list.
1551 * Currently we limit it to one page.
1553 s32 offset;
1554 sector_t bb_sector;
1555 u64 *bbp;
1556 int i;
1557 int sectors = le16_to_cpu(sb->bblog_size);
1558 if (sectors > (PAGE_SIZE / 512))
1559 return -EINVAL;
1560 offset = le32_to_cpu(sb->bblog_offset);
1561 if (offset == 0)
1562 return -EINVAL;
1563 bb_sector = (long long)offset;
1564 if (!sync_page_io(rdev, bb_sector, sectors << 9,
1565 rdev->bb_page, READ, true))
1566 return -EIO;
1567 bbp = (u64 *)page_address(rdev->bb_page);
1568 rdev->badblocks.shift = sb->bblog_shift;
1569 for (i = 0 ; i < (sectors << (9-3)) ; i++, bbp++) {
1570 u64 bb = le64_to_cpu(*bbp);
1571 int count = bb & (0x3ff);
1572 u64 sector = bb >> 10;
1573 sector <<= sb->bblog_shift;
1574 count <<= sb->bblog_shift;
1575 if (bb + 1 == 0)
1576 break;
1577 if (md_set_badblocks(&rdev->badblocks,
1578 sector, count, 1) == 0)
1579 return -EINVAL;
1581 } else if (sb->bblog_offset == 0)
1582 rdev->badblocks.shift = -1;
1584 if (!refdev) {
1585 ret = 1;
1586 } else {
1587 __u64 ev1, ev2;
1588 struct mdp_superblock_1 *refsb = page_address(refdev->sb_page);
1590 if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 ||
1591 sb->level != refsb->level ||
1592 sb->layout != refsb->layout ||
1593 sb->chunksize != refsb->chunksize) {
1594 printk(KERN_WARNING "md: %s has strangely different"
1595 " superblock to %s\n",
1596 bdevname(rdev->bdev,b),
1597 bdevname(refdev->bdev,b2));
1598 return -EINVAL;
1600 ev1 = le64_to_cpu(sb->events);
1601 ev2 = le64_to_cpu(refsb->events);
1603 if (ev1 > ev2)
1604 ret = 1;
1605 else
1606 ret = 0;
1608 if (minor_version)
1609 rdev->sectors = (i_size_read(rdev->bdev->bd_inode) >> 9) -
1610 le64_to_cpu(sb->data_offset);
1611 else
1612 rdev->sectors = rdev->sb_start;
1613 if (rdev->sectors < le64_to_cpu(sb->data_size))
1614 return -EINVAL;
1615 rdev->sectors = le64_to_cpu(sb->data_size);
1616 if (le64_to_cpu(sb->size) > rdev->sectors)
1617 return -EINVAL;
1618 return ret;
1621 static int super_1_validate(struct mddev *mddev, struct md_rdev *rdev)
1623 struct mdp_superblock_1 *sb = page_address(rdev->sb_page);
1624 __u64 ev1 = le64_to_cpu(sb->events);
1626 rdev->raid_disk = -1;
1627 clear_bit(Faulty, &rdev->flags);
1628 clear_bit(In_sync, &rdev->flags);
1629 clear_bit(WriteMostly, &rdev->flags);
1631 if (mddev->raid_disks == 0) {
1632 mddev->major_version = 1;
1633 mddev->patch_version = 0;
1634 mddev->external = 0;
1635 mddev->chunk_sectors = le32_to_cpu(sb->chunksize);
1636 mddev->ctime = le64_to_cpu(sb->ctime) & ((1ULL << 32)-1);
1637 mddev->utime = le64_to_cpu(sb->utime) & ((1ULL << 32)-1);
1638 mddev->level = le32_to_cpu(sb->level);
1639 mddev->clevel[0] = 0;
1640 mddev->layout = le32_to_cpu(sb->layout);
1641 mddev->raid_disks = le32_to_cpu(sb->raid_disks);
1642 mddev->dev_sectors = le64_to_cpu(sb->size);
1643 mddev->events = ev1;
1644 mddev->bitmap_info.offset = 0;
1645 mddev->bitmap_info.default_offset = 1024 >> 9;
1647 mddev->recovery_cp = le64_to_cpu(sb->resync_offset);
1648 memcpy(mddev->uuid, sb->set_uuid, 16);
1650 mddev->max_disks = (4096-256)/2;
1652 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) &&
1653 mddev->bitmap_info.file == NULL )
1654 mddev->bitmap_info.offset =
1655 (__s32)le32_to_cpu(sb->bitmap_offset);
1657 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) {
1658 mddev->reshape_position = le64_to_cpu(sb->reshape_position);
1659 mddev->delta_disks = le32_to_cpu(sb->delta_disks);
1660 mddev->new_level = le32_to_cpu(sb->new_level);
1661 mddev->new_layout = le32_to_cpu(sb->new_layout);
1662 mddev->new_chunk_sectors = le32_to_cpu(sb->new_chunk);
1663 } else {
1664 mddev->reshape_position = MaxSector;
1665 mddev->delta_disks = 0;
1666 mddev->new_level = mddev->level;
1667 mddev->new_layout = mddev->layout;
1668 mddev->new_chunk_sectors = mddev->chunk_sectors;
1671 } else if (mddev->pers == NULL) {
1672 /* Insist of good event counter while assembling, except for
1673 * spares (which don't need an event count) */
1674 ++ev1;
1675 if (rdev->desc_nr >= 0 &&
1676 rdev->desc_nr < le32_to_cpu(sb->max_dev) &&
1677 le16_to_cpu(sb->dev_roles[rdev->desc_nr]) < 0xfffe)
1678 if (ev1 < mddev->events)
1679 return -EINVAL;
1680 } else if (mddev->bitmap) {
1681 /* If adding to array with a bitmap, then we can accept an
1682 * older device, but not too old.
1684 if (ev1 < mddev->bitmap->events_cleared)
1685 return 0;
1686 } else {
1687 if (ev1 < mddev->events)
1688 /* just a hot-add of a new device, leave raid_disk at -1 */
1689 return 0;
1691 if (mddev->level != LEVEL_MULTIPATH) {
1692 int role;
1693 if (rdev->desc_nr < 0 ||
1694 rdev->desc_nr >= le32_to_cpu(sb->max_dev)) {
1695 role = 0xffff;
1696 rdev->desc_nr = -1;
1697 } else
1698 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
1699 switch(role) {
1700 case 0xffff: /* spare */
1701 break;
1702 case 0xfffe: /* faulty */
1703 set_bit(Faulty, &rdev->flags);
1704 break;
1705 default:
1706 if ((le32_to_cpu(sb->feature_map) &
1707 MD_FEATURE_RECOVERY_OFFSET))
1708 rdev->recovery_offset = le64_to_cpu(sb->recovery_offset);
1709 else
1710 set_bit(In_sync, &rdev->flags);
1711 rdev->raid_disk = role;
1712 break;
1714 if (sb->devflags & WriteMostly1)
1715 set_bit(WriteMostly, &rdev->flags);
1716 if (le32_to_cpu(sb->feature_map) & MD_FEATURE_REPLACEMENT)
1717 set_bit(Replacement, &rdev->flags);
1718 } else /* MULTIPATH are always insync */
1719 set_bit(In_sync, &rdev->flags);
1721 return 0;
1724 static void super_1_sync(struct mddev *mddev, struct md_rdev *rdev)
1726 struct mdp_superblock_1 *sb;
1727 struct md_rdev *rdev2;
1728 int max_dev, i;
1729 /* make rdev->sb match mddev and rdev data. */
1731 sb = page_address(rdev->sb_page);
1733 sb->feature_map = 0;
1734 sb->pad0 = 0;
1735 sb->recovery_offset = cpu_to_le64(0);
1736 memset(sb->pad1, 0, sizeof(sb->pad1));
1737 memset(sb->pad3, 0, sizeof(sb->pad3));
1739 sb->utime = cpu_to_le64((__u64)mddev->utime);
1740 sb->events = cpu_to_le64(mddev->events);
1741 if (mddev->in_sync)
1742 sb->resync_offset = cpu_to_le64(mddev->recovery_cp);
1743 else
1744 sb->resync_offset = cpu_to_le64(0);
1746 sb->cnt_corrected_read = cpu_to_le32(atomic_read(&rdev->corrected_errors));
1748 sb->raid_disks = cpu_to_le32(mddev->raid_disks);
1749 sb->size = cpu_to_le64(mddev->dev_sectors);
1750 sb->chunksize = cpu_to_le32(mddev->chunk_sectors);
1751 sb->level = cpu_to_le32(mddev->level);
1752 sb->layout = cpu_to_le32(mddev->layout);
1754 if (test_bit(WriteMostly, &rdev->flags))
1755 sb->devflags |= WriteMostly1;
1756 else
1757 sb->devflags &= ~WriteMostly1;
1759 if (mddev->bitmap && mddev->bitmap_info.file == NULL) {
1760 sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_info.offset);
1761 sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET);
1764 if (rdev->raid_disk >= 0 &&
1765 !test_bit(In_sync, &rdev->flags)) {
1766 sb->feature_map |=
1767 cpu_to_le32(MD_FEATURE_RECOVERY_OFFSET);
1768 sb->recovery_offset =
1769 cpu_to_le64(rdev->recovery_offset);
1771 if (test_bit(Replacement, &rdev->flags))
1772 sb->feature_map |=
1773 cpu_to_le32(MD_FEATURE_REPLACEMENT);
1775 if (mddev->reshape_position != MaxSector) {
1776 sb->feature_map |= cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE);
1777 sb->reshape_position = cpu_to_le64(mddev->reshape_position);
1778 sb->new_layout = cpu_to_le32(mddev->new_layout);
1779 sb->delta_disks = cpu_to_le32(mddev->delta_disks);
1780 sb->new_level = cpu_to_le32(mddev->new_level);
1781 sb->new_chunk = cpu_to_le32(mddev->new_chunk_sectors);
1784 if (rdev->badblocks.count == 0)
1785 /* Nothing to do for bad blocks*/ ;
1786 else if (sb->bblog_offset == 0)
1787 /* Cannot record bad blocks on this device */
1788 md_error(mddev, rdev);
1789 else {
1790 struct badblocks *bb = &rdev->badblocks;
1791 u64 *bbp = (u64 *)page_address(rdev->bb_page);
1792 u64 *p = bb->page;
1793 sb->feature_map |= cpu_to_le32(MD_FEATURE_BAD_BLOCKS);
1794 if (bb->changed) {
1795 unsigned seq;
1797 retry:
1798 seq = read_seqbegin(&bb->lock);
1800 memset(bbp, 0xff, PAGE_SIZE);
1802 for (i = 0 ; i < bb->count ; i++) {
1803 u64 internal_bb = *p++;
1804 u64 store_bb = ((BB_OFFSET(internal_bb) << 10)
1805 | BB_LEN(internal_bb));
1806 *bbp++ = cpu_to_le64(store_bb);
1808 if (read_seqretry(&bb->lock, seq))
1809 goto retry;
1811 bb->sector = (rdev->sb_start +
1812 (int)le32_to_cpu(sb->bblog_offset));
1813 bb->size = le16_to_cpu(sb->bblog_size);
1814 bb->changed = 0;
1818 max_dev = 0;
1819 list_for_each_entry(rdev2, &mddev->disks, same_set)
1820 if (rdev2->desc_nr+1 > max_dev)
1821 max_dev = rdev2->desc_nr+1;
1823 if (max_dev > le32_to_cpu(sb->max_dev)) {
1824 int bmask;
1825 sb->max_dev = cpu_to_le32(max_dev);
1826 rdev->sb_size = max_dev * 2 + 256;
1827 bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
1828 if (rdev->sb_size & bmask)
1829 rdev->sb_size = (rdev->sb_size | bmask) + 1;
1830 } else
1831 max_dev = le32_to_cpu(sb->max_dev);
1833 for (i=0; i<max_dev;i++)
1834 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1836 list_for_each_entry(rdev2, &mddev->disks, same_set) {
1837 i = rdev2->desc_nr;
1838 if (test_bit(Faulty, &rdev2->flags))
1839 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1840 else if (test_bit(In_sync, &rdev2->flags))
1841 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1842 else if (rdev2->raid_disk >= 0)
1843 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1844 else
1845 sb->dev_roles[i] = cpu_to_le16(0xffff);
1848 sb->sb_csum = calc_sb_1_csum(sb);
1851 static unsigned long long
1852 super_1_rdev_size_change(struct md_rdev *rdev, sector_t num_sectors)
1854 struct mdp_superblock_1 *sb;
1855 sector_t max_sectors;
1856 if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
1857 return 0; /* component must fit device */
1858 if (rdev->sb_start < rdev->data_offset) {
1859 /* minor versions 1 and 2; superblock before data */
1860 max_sectors = i_size_read(rdev->bdev->bd_inode) >> 9;
1861 max_sectors -= rdev->data_offset;
1862 if (!num_sectors || num_sectors > max_sectors)
1863 num_sectors = max_sectors;
1864 } else if (rdev->mddev->bitmap_info.offset) {
1865 /* minor version 0 with bitmap we can't move */
1866 return 0;
1867 } else {
1868 /* minor version 0; superblock after data */
1869 sector_t sb_start;
1870 sb_start = (i_size_read(rdev->bdev->bd_inode) >> 9) - 8*2;
1871 sb_start &= ~(sector_t)(4*2 - 1);
1872 max_sectors = rdev->sectors + sb_start - rdev->sb_start;
1873 if (!num_sectors || num_sectors > max_sectors)
1874 num_sectors = max_sectors;
1875 rdev->sb_start = sb_start;
1877 sb = page_address(rdev->sb_page);
1878 sb->data_size = cpu_to_le64(num_sectors);
1879 sb->super_offset = rdev->sb_start;
1880 sb->sb_csum = calc_sb_1_csum(sb);
1881 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
1882 rdev->sb_page);
1883 md_super_wait(rdev->mddev);
1884 return num_sectors;
1887 static struct super_type super_types[] = {
1888 [0] = {
1889 .name = "0.90.0",
1890 .owner = THIS_MODULE,
1891 .load_super = super_90_load,
1892 .validate_super = super_90_validate,
1893 .sync_super = super_90_sync,
1894 .rdev_size_change = super_90_rdev_size_change,
1896 [1] = {
1897 .name = "md-1",
1898 .owner = THIS_MODULE,
1899 .load_super = super_1_load,
1900 .validate_super = super_1_validate,
1901 .sync_super = super_1_sync,
1902 .rdev_size_change = super_1_rdev_size_change,
1906 static void sync_super(struct mddev *mddev, struct md_rdev *rdev)
1908 if (mddev->sync_super) {
1909 mddev->sync_super(mddev, rdev);
1910 return;
1913 BUG_ON(mddev->major_version >= ARRAY_SIZE(super_types));
1915 super_types[mddev->major_version].sync_super(mddev, rdev);
1918 static int match_mddev_units(struct mddev *mddev1, struct mddev *mddev2)
1920 struct md_rdev *rdev, *rdev2;
1922 rcu_read_lock();
1923 rdev_for_each_rcu(rdev, mddev1)
1924 rdev_for_each_rcu(rdev2, mddev2)
1925 if (rdev->bdev->bd_contains ==
1926 rdev2->bdev->bd_contains) {
1927 rcu_read_unlock();
1928 return 1;
1930 rcu_read_unlock();
1931 return 0;
1934 static LIST_HEAD(pending_raid_disks);
1937 * Try to register data integrity profile for an mddev
1939 * This is called when an array is started and after a disk has been kicked
1940 * from the array. It only succeeds if all working and active component devices
1941 * are integrity capable with matching profiles.
1943 int md_integrity_register(struct mddev *mddev)
1945 struct md_rdev *rdev, *reference = NULL;
1947 if (list_empty(&mddev->disks))
1948 return 0; /* nothing to do */
1949 if (!mddev->gendisk || blk_get_integrity(mddev->gendisk))
1950 return 0; /* shouldn't register, or already is */
1951 list_for_each_entry(rdev, &mddev->disks, same_set) {
1952 /* skip spares and non-functional disks */
1953 if (test_bit(Faulty, &rdev->flags))
1954 continue;
1955 if (rdev->raid_disk < 0)
1956 continue;
1957 if (!reference) {
1958 /* Use the first rdev as the reference */
1959 reference = rdev;
1960 continue;
1962 /* does this rdev's profile match the reference profile? */
1963 if (blk_integrity_compare(reference->bdev->bd_disk,
1964 rdev->bdev->bd_disk) < 0)
1965 return -EINVAL;
1967 if (!reference || !bdev_get_integrity(reference->bdev))
1968 return 0;
1970 * All component devices are integrity capable and have matching
1971 * profiles, register the common profile for the md device.
1973 if (blk_integrity_register(mddev->gendisk,
1974 bdev_get_integrity(reference->bdev)) != 0) {
1975 printk(KERN_ERR "md: failed to register integrity for %s\n",
1976 mdname(mddev));
1977 return -EINVAL;
1979 printk(KERN_NOTICE "md: data integrity enabled on %s\n", mdname(mddev));
1980 if (bioset_integrity_create(mddev->bio_set, BIO_POOL_SIZE)) {
1981 printk(KERN_ERR "md: failed to create integrity pool for %s\n",
1982 mdname(mddev));
1983 return -EINVAL;
1985 return 0;
1987 EXPORT_SYMBOL(md_integrity_register);
1989 /* Disable data integrity if non-capable/non-matching disk is being added */
1990 void md_integrity_add_rdev(struct md_rdev *rdev, struct mddev *mddev)
1992 struct blk_integrity *bi_rdev = bdev_get_integrity(rdev->bdev);
1993 struct blk_integrity *bi_mddev = blk_get_integrity(mddev->gendisk);
1995 if (!bi_mddev) /* nothing to do */
1996 return;
1997 if (rdev->raid_disk < 0) /* skip spares */
1998 return;
1999 if (bi_rdev && blk_integrity_compare(mddev->gendisk,
2000 rdev->bdev->bd_disk) >= 0)
2001 return;
2002 printk(KERN_NOTICE "disabling data integrity on %s\n", mdname(mddev));
2003 blk_integrity_unregister(mddev->gendisk);
2005 EXPORT_SYMBOL(md_integrity_add_rdev);
2007 static int bind_rdev_to_array(struct md_rdev * rdev, struct mddev * mddev)
2009 char b[BDEVNAME_SIZE];
2010 struct kobject *ko;
2011 char *s;
2012 int err;
2014 if (rdev->mddev) {
2015 MD_BUG();
2016 return -EINVAL;
2019 /* prevent duplicates */
2020 if (find_rdev(mddev, rdev->bdev->bd_dev))
2021 return -EEXIST;
2023 /* make sure rdev->sectors exceeds mddev->dev_sectors */
2024 if (rdev->sectors && (mddev->dev_sectors == 0 ||
2025 rdev->sectors < mddev->dev_sectors)) {
2026 if (mddev->pers) {
2027 /* Cannot change size, so fail
2028 * If mddev->level <= 0, then we don't care
2029 * about aligning sizes (e.g. linear)
2031 if (mddev->level > 0)
2032 return -ENOSPC;
2033 } else
2034 mddev->dev_sectors = rdev->sectors;
2037 /* Verify rdev->desc_nr is unique.
2038 * If it is -1, assign a free number, else
2039 * check number is not in use
2041 if (rdev->desc_nr < 0) {
2042 int choice = 0;
2043 if (mddev->pers) choice = mddev->raid_disks;
2044 while (find_rdev_nr(mddev, choice))
2045 choice++;
2046 rdev->desc_nr = choice;
2047 } else {
2048 if (find_rdev_nr(mddev, rdev->desc_nr))
2049 return -EBUSY;
2051 if (mddev->max_disks && rdev->desc_nr >= mddev->max_disks) {
2052 printk(KERN_WARNING "md: %s: array is limited to %d devices\n",
2053 mdname(mddev), mddev->max_disks);
2054 return -EBUSY;
2056 bdevname(rdev->bdev,b);
2057 while ( (s=strchr(b, '/')) != NULL)
2058 *s = '!';
2060 rdev->mddev = mddev;
2061 printk(KERN_INFO "md: bind<%s>\n", b);
2063 if ((err = kobject_add(&rdev->kobj, &mddev->kobj, "dev-%s", b)))
2064 goto fail;
2066 ko = &part_to_dev(rdev->bdev->bd_part)->kobj;
2067 if (sysfs_create_link(&rdev->kobj, ko, "block"))
2068 /* failure here is OK */;
2069 rdev->sysfs_state = sysfs_get_dirent_safe(rdev->kobj.sd, "state");
2071 list_add_rcu(&rdev->same_set, &mddev->disks);
2072 bd_link_disk_holder(rdev->bdev, mddev->gendisk);
2074 /* May as well allow recovery to be retried once */
2075 mddev->recovery_disabled++;
2077 return 0;
2079 fail:
2080 printk(KERN_WARNING "md: failed to register dev-%s for %s\n",
2081 b, mdname(mddev));
2082 return err;
2085 static void md_delayed_delete(struct work_struct *ws)
2087 struct md_rdev *rdev = container_of(ws, struct md_rdev, del_work);
2088 kobject_del(&rdev->kobj);
2089 kobject_put(&rdev->kobj);
2092 static void unbind_rdev_from_array(struct md_rdev * rdev)
2094 char b[BDEVNAME_SIZE];
2095 if (!rdev->mddev) {
2096 MD_BUG();
2097 return;
2099 bd_unlink_disk_holder(rdev->bdev, rdev->mddev->gendisk);
2100 list_del_rcu(&rdev->same_set);
2101 printk(KERN_INFO "md: unbind<%s>\n", bdevname(rdev->bdev,b));
2102 rdev->mddev = NULL;
2103 sysfs_remove_link(&rdev->kobj, "block");
2104 sysfs_put(rdev->sysfs_state);
2105 rdev->sysfs_state = NULL;
2106 kfree(rdev->badblocks.page);
2107 rdev->badblocks.count = 0;
2108 rdev->badblocks.page = NULL;
2109 /* We need to delay this, otherwise we can deadlock when
2110 * writing to 'remove' to "dev/state". We also need
2111 * to delay it due to rcu usage.
2113 synchronize_rcu();
2114 INIT_WORK(&rdev->del_work, md_delayed_delete);
2115 kobject_get(&rdev->kobj);
2116 queue_work(md_misc_wq, &rdev->del_work);
2120 * prevent the device from being mounted, repartitioned or
2121 * otherwise reused by a RAID array (or any other kernel
2122 * subsystem), by bd_claiming the device.
2124 static int lock_rdev(struct md_rdev *rdev, dev_t dev, int shared)
2126 int err = 0;
2127 struct block_device *bdev;
2128 char b[BDEVNAME_SIZE];
2130 bdev = blkdev_get_by_dev(dev, FMODE_READ|FMODE_WRITE|FMODE_EXCL,
2131 shared ? (struct md_rdev *)lock_rdev : rdev);
2132 if (IS_ERR(bdev)) {
2133 printk(KERN_ERR "md: could not open %s.\n",
2134 __bdevname(dev, b));
2135 return PTR_ERR(bdev);
2137 rdev->bdev = bdev;
2138 return err;
2141 static void unlock_rdev(struct md_rdev *rdev)
2143 struct block_device *bdev = rdev->bdev;
2144 rdev->bdev = NULL;
2145 if (!bdev)
2146 MD_BUG();
2147 blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
2150 void md_autodetect_dev(dev_t dev);
2152 static void export_rdev(struct md_rdev * rdev)
2154 char b[BDEVNAME_SIZE];
2155 printk(KERN_INFO "md: export_rdev(%s)\n",
2156 bdevname(rdev->bdev,b));
2157 if (rdev->mddev)
2158 MD_BUG();
2159 free_disk_sb(rdev);
2160 #ifndef MODULE
2161 if (test_bit(AutoDetected, &rdev->flags))
2162 md_autodetect_dev(rdev->bdev->bd_dev);
2163 #endif
2164 unlock_rdev(rdev);
2165 kobject_put(&rdev->kobj);
2168 static void kick_rdev_from_array(struct md_rdev * rdev)
2170 unbind_rdev_from_array(rdev);
2171 export_rdev(rdev);
2174 static void export_array(struct mddev *mddev)
2176 struct md_rdev *rdev, *tmp;
2178 rdev_for_each(rdev, tmp, mddev) {
2179 if (!rdev->mddev) {
2180 MD_BUG();
2181 continue;
2183 kick_rdev_from_array(rdev);
2185 if (!list_empty(&mddev->disks))
2186 MD_BUG();
2187 mddev->raid_disks = 0;
2188 mddev->major_version = 0;
2191 static void print_desc(mdp_disk_t *desc)
2193 printk(" DISK<N:%d,(%d,%d),R:%d,S:%d>\n", desc->number,
2194 desc->major,desc->minor,desc->raid_disk,desc->state);
2197 static void print_sb_90(mdp_super_t *sb)
2199 int i;
2201 printk(KERN_INFO
2202 "md: SB: (V:%d.%d.%d) ID:<%08x.%08x.%08x.%08x> CT:%08x\n",
2203 sb->major_version, sb->minor_version, sb->patch_version,
2204 sb->set_uuid0, sb->set_uuid1, sb->set_uuid2, sb->set_uuid3,
2205 sb->ctime);
2206 printk(KERN_INFO "md: L%d S%08d ND:%d RD:%d md%d LO:%d CS:%d\n",
2207 sb->level, sb->size, sb->nr_disks, sb->raid_disks,
2208 sb->md_minor, sb->layout, sb->chunk_size);
2209 printk(KERN_INFO "md: UT:%08x ST:%d AD:%d WD:%d"
2210 " FD:%d SD:%d CSUM:%08x E:%08lx\n",
2211 sb->utime, sb->state, sb->active_disks, sb->working_disks,
2212 sb->failed_disks, sb->spare_disks,
2213 sb->sb_csum, (unsigned long)sb->events_lo);
2215 printk(KERN_INFO);
2216 for (i = 0; i < MD_SB_DISKS; i++) {
2217 mdp_disk_t *desc;
2219 desc = sb->disks + i;
2220 if (desc->number || desc->major || desc->minor ||
2221 desc->raid_disk || (desc->state && (desc->state != 4))) {
2222 printk(" D %2d: ", i);
2223 print_desc(desc);
2226 printk(KERN_INFO "md: THIS: ");
2227 print_desc(&sb->this_disk);
2230 static void print_sb_1(struct mdp_superblock_1 *sb)
2232 __u8 *uuid;
2234 uuid = sb->set_uuid;
2235 printk(KERN_INFO
2236 "md: SB: (V:%u) (F:0x%08x) Array-ID:<%pU>\n"
2237 "md: Name: \"%s\" CT:%llu\n",
2238 le32_to_cpu(sb->major_version),
2239 le32_to_cpu(sb->feature_map),
2240 uuid,
2241 sb->set_name,
2242 (unsigned long long)le64_to_cpu(sb->ctime)
2243 & MD_SUPERBLOCK_1_TIME_SEC_MASK);
2245 uuid = sb->device_uuid;
2246 printk(KERN_INFO
2247 "md: L%u SZ%llu RD:%u LO:%u CS:%u DO:%llu DS:%llu SO:%llu"
2248 " RO:%llu\n"
2249 "md: Dev:%08x UUID: %pU\n"
2250 "md: (F:0x%08x) UT:%llu Events:%llu ResyncOffset:%llu CSUM:0x%08x\n"
2251 "md: (MaxDev:%u) \n",
2252 le32_to_cpu(sb->level),
2253 (unsigned long long)le64_to_cpu(sb->size),
2254 le32_to_cpu(sb->raid_disks),
2255 le32_to_cpu(sb->layout),
2256 le32_to_cpu(sb->chunksize),
2257 (unsigned long long)le64_to_cpu(sb->data_offset),
2258 (unsigned long long)le64_to_cpu(sb->data_size),
2259 (unsigned long long)le64_to_cpu(sb->super_offset),
2260 (unsigned long long)le64_to_cpu(sb->recovery_offset),
2261 le32_to_cpu(sb->dev_number),
2262 uuid,
2263 sb->devflags,
2264 (unsigned long long)le64_to_cpu(sb->utime) & MD_SUPERBLOCK_1_TIME_SEC_MASK,
2265 (unsigned long long)le64_to_cpu(sb->events),
2266 (unsigned long long)le64_to_cpu(sb->resync_offset),
2267 le32_to_cpu(sb->sb_csum),
2268 le32_to_cpu(sb->max_dev)
2272 static void print_rdev(struct md_rdev *rdev, int major_version)
2274 char b[BDEVNAME_SIZE];
2275 printk(KERN_INFO "md: rdev %s, Sect:%08llu F:%d S:%d DN:%u\n",
2276 bdevname(rdev->bdev, b), (unsigned long long)rdev->sectors,
2277 test_bit(Faulty, &rdev->flags), test_bit(In_sync, &rdev->flags),
2278 rdev->desc_nr);
2279 if (rdev->sb_loaded) {
2280 printk(KERN_INFO "md: rdev superblock (MJ:%d):\n", major_version);
2281 switch (major_version) {
2282 case 0:
2283 print_sb_90(page_address(rdev->sb_page));
2284 break;
2285 case 1:
2286 print_sb_1(page_address(rdev->sb_page));
2287 break;
2289 } else
2290 printk(KERN_INFO "md: no rdev superblock!\n");
2293 static void md_print_devices(void)
2295 struct list_head *tmp;
2296 struct md_rdev *rdev;
2297 struct mddev *mddev;
2298 char b[BDEVNAME_SIZE];
2300 printk("\n");
2301 printk("md: **********************************\n");
2302 printk("md: * <COMPLETE RAID STATE PRINTOUT> *\n");
2303 printk("md: **********************************\n");
2304 for_each_mddev(mddev, tmp) {
2306 if (mddev->bitmap)
2307 bitmap_print_sb(mddev->bitmap);
2308 else
2309 printk("%s: ", mdname(mddev));
2310 list_for_each_entry(rdev, &mddev->disks, same_set)
2311 printk("<%s>", bdevname(rdev->bdev,b));
2312 printk("\n");
2314 list_for_each_entry(rdev, &mddev->disks, same_set)
2315 print_rdev(rdev, mddev->major_version);
2317 printk("md: **********************************\n");
2318 printk("\n");
2322 static void sync_sbs(struct mddev * mddev, int nospares)
2324 /* Update each superblock (in-memory image), but
2325 * if we are allowed to, skip spares which already
2326 * have the right event counter, or have one earlier
2327 * (which would mean they aren't being marked as dirty
2328 * with the rest of the array)
2330 struct md_rdev *rdev;
2331 list_for_each_entry(rdev, &mddev->disks, same_set) {
2332 if (rdev->sb_events == mddev->events ||
2333 (nospares &&
2334 rdev->raid_disk < 0 &&
2335 rdev->sb_events+1 == mddev->events)) {
2336 /* Don't update this superblock */
2337 rdev->sb_loaded = 2;
2338 } else {
2339 sync_super(mddev, rdev);
2340 rdev->sb_loaded = 1;
2345 static void md_update_sb(struct mddev * mddev, int force_change)
2347 struct md_rdev *rdev;
2348 int sync_req;
2349 int nospares = 0;
2350 int any_badblocks_changed = 0;
2352 repeat:
2353 /* First make sure individual recovery_offsets are correct */
2354 list_for_each_entry(rdev, &mddev->disks, same_set) {
2355 if (rdev->raid_disk >= 0 &&
2356 mddev->delta_disks >= 0 &&
2357 !test_bit(In_sync, &rdev->flags) &&
2358 mddev->curr_resync_completed > rdev->recovery_offset)
2359 rdev->recovery_offset = mddev->curr_resync_completed;
2362 if (!mddev->persistent) {
2363 clear_bit(MD_CHANGE_CLEAN, &mddev->flags);
2364 clear_bit(MD_CHANGE_DEVS, &mddev->flags);
2365 if (!mddev->external) {
2366 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
2367 list_for_each_entry(rdev, &mddev->disks, same_set) {
2368 if (rdev->badblocks.changed) {
2369 md_ack_all_badblocks(&rdev->badblocks);
2370 md_error(mddev, rdev);
2372 clear_bit(Blocked, &rdev->flags);
2373 clear_bit(BlockedBadBlocks, &rdev->flags);
2374 wake_up(&rdev->blocked_wait);
2377 wake_up(&mddev->sb_wait);
2378 return;
2381 spin_lock_irq(&mddev->write_lock);
2383 mddev->utime = get_seconds();
2385 if (test_and_clear_bit(MD_CHANGE_DEVS, &mddev->flags))
2386 force_change = 1;
2387 if (test_and_clear_bit(MD_CHANGE_CLEAN, &mddev->flags))
2388 /* just a clean<-> dirty transition, possibly leave spares alone,
2389 * though if events isn't the right even/odd, we will have to do
2390 * spares after all
2392 nospares = 1;
2393 if (force_change)
2394 nospares = 0;
2395 if (mddev->degraded)
2396 /* If the array is degraded, then skipping spares is both
2397 * dangerous and fairly pointless.
2398 * Dangerous because a device that was removed from the array
2399 * might have a event_count that still looks up-to-date,
2400 * so it can be re-added without a resync.
2401 * Pointless because if there are any spares to skip,
2402 * then a recovery will happen and soon that array won't
2403 * be degraded any more and the spare can go back to sleep then.
2405 nospares = 0;
2407 sync_req = mddev->in_sync;
2409 /* If this is just a dirty<->clean transition, and the array is clean
2410 * and 'events' is odd, we can roll back to the previous clean state */
2411 if (nospares
2412 && (mddev->in_sync && mddev->recovery_cp == MaxSector)
2413 && mddev->can_decrease_events
2414 && mddev->events != 1) {
2415 mddev->events--;
2416 mddev->can_decrease_events = 0;
2417 } else {
2418 /* otherwise we have to go forward and ... */
2419 mddev->events ++;
2420 mddev->can_decrease_events = nospares;
2423 if (!mddev->events) {
2425 * oops, this 64-bit counter should never wrap.
2426 * Either we are in around ~1 trillion A.C., assuming
2427 * 1 reboot per second, or we have a bug:
2429 MD_BUG();
2430 mddev->events --;
2433 list_for_each_entry(rdev, &mddev->disks, same_set) {
2434 if (rdev->badblocks.changed)
2435 any_badblocks_changed++;
2436 if (test_bit(Faulty, &rdev->flags))
2437 set_bit(FaultRecorded, &rdev->flags);
2440 sync_sbs(mddev, nospares);
2441 spin_unlock_irq(&mddev->write_lock);
2443 pr_debug("md: updating %s RAID superblock on device (in sync %d)\n",
2444 mdname(mddev), mddev->in_sync);
2446 bitmap_update_sb(mddev->bitmap);
2447 list_for_each_entry(rdev, &mddev->disks, same_set) {
2448 char b[BDEVNAME_SIZE];
2450 if (rdev->sb_loaded != 1)
2451 continue; /* no noise on spare devices */
2453 if (!test_bit(Faulty, &rdev->flags) &&
2454 rdev->saved_raid_disk == -1) {
2455 md_super_write(mddev,rdev,
2456 rdev->sb_start, rdev->sb_size,
2457 rdev->sb_page);
2458 pr_debug("md: (write) %s's sb offset: %llu\n",
2459 bdevname(rdev->bdev, b),
2460 (unsigned long long)rdev->sb_start);
2461 rdev->sb_events = mddev->events;
2462 if (rdev->badblocks.size) {
2463 md_super_write(mddev, rdev,
2464 rdev->badblocks.sector,
2465 rdev->badblocks.size << 9,
2466 rdev->bb_page);
2467 rdev->badblocks.size = 0;
2470 } else if (test_bit(Faulty, &rdev->flags))
2471 pr_debug("md: %s (skipping faulty)\n",
2472 bdevname(rdev->bdev, b));
2473 else
2474 pr_debug("(skipping incremental s/r ");
2476 if (mddev->level == LEVEL_MULTIPATH)
2477 /* only need to write one superblock... */
2478 break;
2480 md_super_wait(mddev);
2481 /* if there was a failure, MD_CHANGE_DEVS was set, and we re-write super */
2483 spin_lock_irq(&mddev->write_lock);
2484 if (mddev->in_sync != sync_req ||
2485 test_bit(MD_CHANGE_DEVS, &mddev->flags)) {
2486 /* have to write it out again */
2487 spin_unlock_irq(&mddev->write_lock);
2488 goto repeat;
2490 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
2491 spin_unlock_irq(&mddev->write_lock);
2492 wake_up(&mddev->sb_wait);
2493 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
2494 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
2496 list_for_each_entry(rdev, &mddev->disks, same_set) {
2497 if (test_and_clear_bit(FaultRecorded, &rdev->flags))
2498 clear_bit(Blocked, &rdev->flags);
2500 if (any_badblocks_changed)
2501 md_ack_all_badblocks(&rdev->badblocks);
2502 clear_bit(BlockedBadBlocks, &rdev->flags);
2503 wake_up(&rdev->blocked_wait);
2507 /* words written to sysfs files may, or may not, be \n terminated.
2508 * We want to accept with case. For this we use cmd_match.
2510 static int cmd_match(const char *cmd, const char *str)
2512 /* See if cmd, written into a sysfs file, matches
2513 * str. They must either be the same, or cmd can
2514 * have a trailing newline
2516 while (*cmd && *str && *cmd == *str) {
2517 cmd++;
2518 str++;
2520 if (*cmd == '\n')
2521 cmd++;
2522 if (*str || *cmd)
2523 return 0;
2524 return 1;
2527 struct rdev_sysfs_entry {
2528 struct attribute attr;
2529 ssize_t (*show)(struct md_rdev *, char *);
2530 ssize_t (*store)(struct md_rdev *, const char *, size_t);
2533 static ssize_t
2534 state_show(struct md_rdev *rdev, char *page)
2536 char *sep = "";
2537 size_t len = 0;
2539 if (test_bit(Faulty, &rdev->flags) ||
2540 rdev->badblocks.unacked_exist) {
2541 len+= sprintf(page+len, "%sfaulty",sep);
2542 sep = ",";
2544 if (test_bit(In_sync, &rdev->flags)) {
2545 len += sprintf(page+len, "%sin_sync",sep);
2546 sep = ",";
2548 if (test_bit(WriteMostly, &rdev->flags)) {
2549 len += sprintf(page+len, "%swrite_mostly",sep);
2550 sep = ",";
2552 if (test_bit(Blocked, &rdev->flags) ||
2553 (rdev->badblocks.unacked_exist
2554 && !test_bit(Faulty, &rdev->flags))) {
2555 len += sprintf(page+len, "%sblocked", sep);
2556 sep = ",";
2558 if (!test_bit(Faulty, &rdev->flags) &&
2559 !test_bit(In_sync, &rdev->flags)) {
2560 len += sprintf(page+len, "%sspare", sep);
2561 sep = ",";
2563 if (test_bit(WriteErrorSeen, &rdev->flags)) {
2564 len += sprintf(page+len, "%swrite_error", sep);
2565 sep = ",";
2567 if (test_bit(WantReplacement, &rdev->flags)) {
2568 len += sprintf(page+len, "%swant_replacement", sep);
2569 sep = ",";
2571 if (test_bit(Replacement, &rdev->flags)) {
2572 len += sprintf(page+len, "%sreplacement", sep);
2573 sep = ",";
2576 return len+sprintf(page+len, "\n");
2579 static ssize_t
2580 state_store(struct md_rdev *rdev, const char *buf, size_t len)
2582 /* can write
2583 * faulty - simulates an error
2584 * remove - disconnects the device
2585 * writemostly - sets write_mostly
2586 * -writemostly - clears write_mostly
2587 * blocked - sets the Blocked flags
2588 * -blocked - clears the Blocked and possibly simulates an error
2589 * insync - sets Insync providing device isn't active
2590 * write_error - sets WriteErrorSeen
2591 * -write_error - clears WriteErrorSeen
2593 int err = -EINVAL;
2594 if (cmd_match(buf, "faulty") && rdev->mddev->pers) {
2595 md_error(rdev->mddev, rdev);
2596 if (test_bit(Faulty, &rdev->flags))
2597 err = 0;
2598 else
2599 err = -EBUSY;
2600 } else if (cmd_match(buf, "remove")) {
2601 if (rdev->raid_disk >= 0)
2602 err = -EBUSY;
2603 else {
2604 struct mddev *mddev = rdev->mddev;
2605 kick_rdev_from_array(rdev);
2606 if (mddev->pers)
2607 md_update_sb(mddev, 1);
2608 md_new_event(mddev);
2609 err = 0;
2611 } else if (cmd_match(buf, "writemostly")) {
2612 set_bit(WriteMostly, &rdev->flags);
2613 err = 0;
2614 } else if (cmd_match(buf, "-writemostly")) {
2615 clear_bit(WriteMostly, &rdev->flags);
2616 err = 0;
2617 } else if (cmd_match(buf, "blocked")) {
2618 set_bit(Blocked, &rdev->flags);
2619 err = 0;
2620 } else if (cmd_match(buf, "-blocked")) {
2621 if (!test_bit(Faulty, &rdev->flags) &&
2622 rdev->badblocks.unacked_exist) {
2623 /* metadata handler doesn't understand badblocks,
2624 * so we need to fail the device
2626 md_error(rdev->mddev, rdev);
2628 clear_bit(Blocked, &rdev->flags);
2629 clear_bit(BlockedBadBlocks, &rdev->flags);
2630 wake_up(&rdev->blocked_wait);
2631 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2632 md_wakeup_thread(rdev->mddev->thread);
2634 err = 0;
2635 } else if (cmd_match(buf, "insync") && rdev->raid_disk == -1) {
2636 set_bit(In_sync, &rdev->flags);
2637 err = 0;
2638 } else if (cmd_match(buf, "write_error")) {
2639 set_bit(WriteErrorSeen, &rdev->flags);
2640 err = 0;
2641 } else if (cmd_match(buf, "-write_error")) {
2642 clear_bit(WriteErrorSeen, &rdev->flags);
2643 err = 0;
2644 } else if (cmd_match(buf, "want_replacement")) {
2645 /* Any non-spare device that is not a replacement can
2646 * become want_replacement at any time, but we then need to
2647 * check if recovery is needed.
2649 if (rdev->raid_disk >= 0 &&
2650 !test_bit(Replacement, &rdev->flags))
2651 set_bit(WantReplacement, &rdev->flags);
2652 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2653 md_wakeup_thread(rdev->mddev->thread);
2654 err = 0;
2655 } else if (cmd_match(buf, "-want_replacement")) {
2656 /* Clearing 'want_replacement' is always allowed.
2657 * Once replacements starts it is too late though.
2659 err = 0;
2660 clear_bit(WantReplacement, &rdev->flags);
2661 } else if (cmd_match(buf, "replacement")) {
2662 /* Can only set a device as a replacement when array has not
2663 * yet been started. Once running, replacement is automatic
2664 * from spares, or by assigning 'slot'.
2666 if (rdev->mddev->pers)
2667 err = -EBUSY;
2668 else {
2669 set_bit(Replacement, &rdev->flags);
2670 err = 0;
2672 } else if (cmd_match(buf, "-replacement")) {
2673 /* Similarly, can only clear Replacement before start */
2674 if (rdev->mddev->pers)
2675 err = -EBUSY;
2676 else {
2677 clear_bit(Replacement, &rdev->flags);
2678 err = 0;
2681 if (!err)
2682 sysfs_notify_dirent_safe(rdev->sysfs_state);
2683 return err ? err : len;
2685 static struct rdev_sysfs_entry rdev_state =
2686 __ATTR(state, S_IRUGO|S_IWUSR, state_show, state_store);
2688 static ssize_t
2689 errors_show(struct md_rdev *rdev, char *page)
2691 return sprintf(page, "%d\n", atomic_read(&rdev->corrected_errors));
2694 static ssize_t
2695 errors_store(struct md_rdev *rdev, const char *buf, size_t len)
2697 char *e;
2698 unsigned long n = simple_strtoul(buf, &e, 10);
2699 if (*buf && (*e == 0 || *e == '\n')) {
2700 atomic_set(&rdev->corrected_errors, n);
2701 return len;
2703 return -EINVAL;
2705 static struct rdev_sysfs_entry rdev_errors =
2706 __ATTR(errors, S_IRUGO|S_IWUSR, errors_show, errors_store);
2708 static ssize_t
2709 slot_show(struct md_rdev *rdev, char *page)
2711 if (rdev->raid_disk < 0)
2712 return sprintf(page, "none\n");
2713 else
2714 return sprintf(page, "%d\n", rdev->raid_disk);
2717 static ssize_t
2718 slot_store(struct md_rdev *rdev, const char *buf, size_t len)
2720 char *e;
2721 int err;
2722 int slot = simple_strtoul(buf, &e, 10);
2723 if (strncmp(buf, "none", 4)==0)
2724 slot = -1;
2725 else if (e==buf || (*e && *e!= '\n'))
2726 return -EINVAL;
2727 if (rdev->mddev->pers && slot == -1) {
2728 /* Setting 'slot' on an active array requires also
2729 * updating the 'rd%d' link, and communicating
2730 * with the personality with ->hot_*_disk.
2731 * For now we only support removing
2732 * failed/spare devices. This normally happens automatically,
2733 * but not when the metadata is externally managed.
2735 if (rdev->raid_disk == -1)
2736 return -EEXIST;
2737 /* personality does all needed checks */
2738 if (rdev->mddev->pers->hot_remove_disk == NULL)
2739 return -EINVAL;
2740 err = rdev->mddev->pers->
2741 hot_remove_disk(rdev->mddev, rdev);
2742 if (err)
2743 return err;
2744 sysfs_unlink_rdev(rdev->mddev, rdev);
2745 rdev->raid_disk = -1;
2746 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2747 md_wakeup_thread(rdev->mddev->thread);
2748 } else if (rdev->mddev->pers) {
2749 /* Activating a spare .. or possibly reactivating
2750 * if we ever get bitmaps working here.
2753 if (rdev->raid_disk != -1)
2754 return -EBUSY;
2756 if (test_bit(MD_RECOVERY_RUNNING, &rdev->mddev->recovery))
2757 return -EBUSY;
2759 if (rdev->mddev->pers->hot_add_disk == NULL)
2760 return -EINVAL;
2762 if (slot >= rdev->mddev->raid_disks &&
2763 slot >= rdev->mddev->raid_disks + rdev->mddev->delta_disks)
2764 return -ENOSPC;
2766 rdev->raid_disk = slot;
2767 if (test_bit(In_sync, &rdev->flags))
2768 rdev->saved_raid_disk = slot;
2769 else
2770 rdev->saved_raid_disk = -1;
2771 clear_bit(In_sync, &rdev->flags);
2772 err = rdev->mddev->pers->
2773 hot_add_disk(rdev->mddev, rdev);
2774 if (err) {
2775 rdev->raid_disk = -1;
2776 return err;
2777 } else
2778 sysfs_notify_dirent_safe(rdev->sysfs_state);
2779 if (sysfs_link_rdev(rdev->mddev, rdev))
2780 /* failure here is OK */;
2781 /* don't wakeup anyone, leave that to userspace. */
2782 } else {
2783 if (slot >= rdev->mddev->raid_disks &&
2784 slot >= rdev->mddev->raid_disks + rdev->mddev->delta_disks)
2785 return -ENOSPC;
2786 rdev->raid_disk = slot;
2787 /* assume it is working */
2788 clear_bit(Faulty, &rdev->flags);
2789 clear_bit(WriteMostly, &rdev->flags);
2790 set_bit(In_sync, &rdev->flags);
2791 sysfs_notify_dirent_safe(rdev->sysfs_state);
2793 return len;
2797 static struct rdev_sysfs_entry rdev_slot =
2798 __ATTR(slot, S_IRUGO|S_IWUSR, slot_show, slot_store);
2800 static ssize_t
2801 offset_show(struct md_rdev *rdev, char *page)
2803 return sprintf(page, "%llu\n", (unsigned long long)rdev->data_offset);
2806 static ssize_t
2807 offset_store(struct md_rdev *rdev, const char *buf, size_t len)
2809 char *e;
2810 unsigned long long offset = simple_strtoull(buf, &e, 10);
2811 if (e==buf || (*e && *e != '\n'))
2812 return -EINVAL;
2813 if (rdev->mddev->pers && rdev->raid_disk >= 0)
2814 return -EBUSY;
2815 if (rdev->sectors && rdev->mddev->external)
2816 /* Must set offset before size, so overlap checks
2817 * can be sane */
2818 return -EBUSY;
2819 rdev->data_offset = offset;
2820 return len;
2823 static struct rdev_sysfs_entry rdev_offset =
2824 __ATTR(offset, S_IRUGO|S_IWUSR, offset_show, offset_store);
2826 static ssize_t
2827 rdev_size_show(struct md_rdev *rdev, char *page)
2829 return sprintf(page, "%llu\n", (unsigned long long)rdev->sectors / 2);
2832 static int overlaps(sector_t s1, sector_t l1, sector_t s2, sector_t l2)
2834 /* check if two start/length pairs overlap */
2835 if (s1+l1 <= s2)
2836 return 0;
2837 if (s2+l2 <= s1)
2838 return 0;
2839 return 1;
2842 static int strict_blocks_to_sectors(const char *buf, sector_t *sectors)
2844 unsigned long long blocks;
2845 sector_t new;
2847 if (strict_strtoull(buf, 10, &blocks) < 0)
2848 return -EINVAL;
2850 if (blocks & 1ULL << (8 * sizeof(blocks) - 1))
2851 return -EINVAL; /* sector conversion overflow */
2853 new = blocks * 2;
2854 if (new != blocks * 2)
2855 return -EINVAL; /* unsigned long long to sector_t overflow */
2857 *sectors = new;
2858 return 0;
2861 static ssize_t
2862 rdev_size_store(struct md_rdev *rdev, const char *buf, size_t len)
2864 struct mddev *my_mddev = rdev->mddev;
2865 sector_t oldsectors = rdev->sectors;
2866 sector_t sectors;
2868 if (strict_blocks_to_sectors(buf, &sectors) < 0)
2869 return -EINVAL;
2870 if (my_mddev->pers && rdev->raid_disk >= 0) {
2871 if (my_mddev->persistent) {
2872 sectors = super_types[my_mddev->major_version].
2873 rdev_size_change(rdev, sectors);
2874 if (!sectors)
2875 return -EBUSY;
2876 } else if (!sectors)
2877 sectors = (i_size_read(rdev->bdev->bd_inode) >> 9) -
2878 rdev->data_offset;
2880 if (sectors < my_mddev->dev_sectors)
2881 return -EINVAL; /* component must fit device */
2883 rdev->sectors = sectors;
2884 if (sectors > oldsectors && my_mddev->external) {
2885 /* need to check that all other rdevs with the same ->bdev
2886 * do not overlap. We need to unlock the mddev to avoid
2887 * a deadlock. We have already changed rdev->sectors, and if
2888 * we have to change it back, we will have the lock again.
2890 struct mddev *mddev;
2891 int overlap = 0;
2892 struct list_head *tmp;
2894 mddev_unlock(my_mddev);
2895 for_each_mddev(mddev, tmp) {
2896 struct md_rdev *rdev2;
2898 mddev_lock(mddev);
2899 list_for_each_entry(rdev2, &mddev->disks, same_set)
2900 if (rdev->bdev == rdev2->bdev &&
2901 rdev != rdev2 &&
2902 overlaps(rdev->data_offset, rdev->sectors,
2903 rdev2->data_offset,
2904 rdev2->sectors)) {
2905 overlap = 1;
2906 break;
2908 mddev_unlock(mddev);
2909 if (overlap) {
2910 mddev_put(mddev);
2911 break;
2914 mddev_lock(my_mddev);
2915 if (overlap) {
2916 /* Someone else could have slipped in a size
2917 * change here, but doing so is just silly.
2918 * We put oldsectors back because we *know* it is
2919 * safe, and trust userspace not to race with
2920 * itself
2922 rdev->sectors = oldsectors;
2923 return -EBUSY;
2926 return len;
2929 static struct rdev_sysfs_entry rdev_size =
2930 __ATTR(size, S_IRUGO|S_IWUSR, rdev_size_show, rdev_size_store);
2933 static ssize_t recovery_start_show(struct md_rdev *rdev, char *page)
2935 unsigned long long recovery_start = rdev->recovery_offset;
2937 if (test_bit(In_sync, &rdev->flags) ||
2938 recovery_start == MaxSector)
2939 return sprintf(page, "none\n");
2941 return sprintf(page, "%llu\n", recovery_start);
2944 static ssize_t recovery_start_store(struct md_rdev *rdev, const char *buf, size_t len)
2946 unsigned long long recovery_start;
2948 if (cmd_match(buf, "none"))
2949 recovery_start = MaxSector;
2950 else if (strict_strtoull(buf, 10, &recovery_start))
2951 return -EINVAL;
2953 if (rdev->mddev->pers &&
2954 rdev->raid_disk >= 0)
2955 return -EBUSY;
2957 rdev->recovery_offset = recovery_start;
2958 if (recovery_start == MaxSector)
2959 set_bit(In_sync, &rdev->flags);
2960 else
2961 clear_bit(In_sync, &rdev->flags);
2962 return len;
2965 static struct rdev_sysfs_entry rdev_recovery_start =
2966 __ATTR(recovery_start, S_IRUGO|S_IWUSR, recovery_start_show, recovery_start_store);
2969 static ssize_t
2970 badblocks_show(struct badblocks *bb, char *page, int unack);
2971 static ssize_t
2972 badblocks_store(struct badblocks *bb, const char *page, size_t len, int unack);
2974 static ssize_t bb_show(struct md_rdev *rdev, char *page)
2976 return badblocks_show(&rdev->badblocks, page, 0);
2978 static ssize_t bb_store(struct md_rdev *rdev, const char *page, size_t len)
2980 int rv = badblocks_store(&rdev->badblocks, page, len, 0);
2981 /* Maybe that ack was all we needed */
2982 if (test_and_clear_bit(BlockedBadBlocks, &rdev->flags))
2983 wake_up(&rdev->blocked_wait);
2984 return rv;
2986 static struct rdev_sysfs_entry rdev_bad_blocks =
2987 __ATTR(bad_blocks, S_IRUGO|S_IWUSR, bb_show, bb_store);
2990 static ssize_t ubb_show(struct md_rdev *rdev, char *page)
2992 return badblocks_show(&rdev->badblocks, page, 1);
2994 static ssize_t ubb_store(struct md_rdev *rdev, const char *page, size_t len)
2996 return badblocks_store(&rdev->badblocks, page, len, 1);
2998 static struct rdev_sysfs_entry rdev_unack_bad_blocks =
2999 __ATTR(unacknowledged_bad_blocks, S_IRUGO|S_IWUSR, ubb_show, ubb_store);
3001 static struct attribute *rdev_default_attrs[] = {
3002 &rdev_state.attr,
3003 &rdev_errors.attr,
3004 &rdev_slot.attr,
3005 &rdev_offset.attr,
3006 &rdev_size.attr,
3007 &rdev_recovery_start.attr,
3008 &rdev_bad_blocks.attr,
3009 &rdev_unack_bad_blocks.attr,
3010 NULL,
3012 static ssize_t
3013 rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
3015 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
3016 struct md_rdev *rdev = container_of(kobj, struct md_rdev, kobj);
3017 struct mddev *mddev = rdev->mddev;
3018 ssize_t rv;
3020 if (!entry->show)
3021 return -EIO;
3023 rv = mddev ? mddev_lock(mddev) : -EBUSY;
3024 if (!rv) {
3025 if (rdev->mddev == NULL)
3026 rv = -EBUSY;
3027 else
3028 rv = entry->show(rdev, page);
3029 mddev_unlock(mddev);
3031 return rv;
3034 static ssize_t
3035 rdev_attr_store(struct kobject *kobj, struct attribute *attr,
3036 const char *page, size_t length)
3038 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
3039 struct md_rdev *rdev = container_of(kobj, struct md_rdev, kobj);
3040 ssize_t rv;
3041 struct mddev *mddev = rdev->mddev;
3043 if (!entry->store)
3044 return -EIO;
3045 if (!capable(CAP_SYS_ADMIN))
3046 return -EACCES;
3047 rv = mddev ? mddev_lock(mddev): -EBUSY;
3048 if (!rv) {
3049 if (rdev->mddev == NULL)
3050 rv = -EBUSY;
3051 else
3052 rv = entry->store(rdev, page, length);
3053 mddev_unlock(mddev);
3055 return rv;
3058 static void rdev_free(struct kobject *ko)
3060 struct md_rdev *rdev = container_of(ko, struct md_rdev, kobj);
3061 kfree(rdev);
3063 static const struct sysfs_ops rdev_sysfs_ops = {
3064 .show = rdev_attr_show,
3065 .store = rdev_attr_store,
3067 static struct kobj_type rdev_ktype = {
3068 .release = rdev_free,
3069 .sysfs_ops = &rdev_sysfs_ops,
3070 .default_attrs = rdev_default_attrs,
3073 int md_rdev_init(struct md_rdev *rdev)
3075 rdev->desc_nr = -1;
3076 rdev->saved_raid_disk = -1;
3077 rdev->raid_disk = -1;
3078 rdev->flags = 0;
3079 rdev->data_offset = 0;
3080 rdev->sb_events = 0;
3081 rdev->last_read_error.tv_sec = 0;
3082 rdev->last_read_error.tv_nsec = 0;
3083 rdev->sb_loaded = 0;
3084 rdev->bb_page = NULL;
3085 atomic_set(&rdev->nr_pending, 0);
3086 atomic_set(&rdev->read_errors, 0);
3087 atomic_set(&rdev->corrected_errors, 0);
3089 INIT_LIST_HEAD(&rdev->same_set);
3090 init_waitqueue_head(&rdev->blocked_wait);
3092 /* Add space to store bad block list.
3093 * This reserves the space even on arrays where it cannot
3094 * be used - I wonder if that matters
3096 rdev->badblocks.count = 0;
3097 rdev->badblocks.shift = 0;
3098 rdev->badblocks.page = kmalloc(PAGE_SIZE, GFP_KERNEL);
3099 seqlock_init(&rdev->badblocks.lock);
3100 if (rdev->badblocks.page == NULL)
3101 return -ENOMEM;
3103 return 0;
3105 EXPORT_SYMBOL_GPL(md_rdev_init);
3107 * Import a device. If 'super_format' >= 0, then sanity check the superblock
3109 * mark the device faulty if:
3111 * - the device is nonexistent (zero size)
3112 * - the device has no valid superblock
3114 * a faulty rdev _never_ has rdev->sb set.
3116 static struct md_rdev *md_import_device(dev_t newdev, int super_format, int super_minor)
3118 char b[BDEVNAME_SIZE];
3119 int err;
3120 struct md_rdev *rdev;
3121 sector_t size;
3123 rdev = kzalloc(sizeof(*rdev), GFP_KERNEL);
3124 if (!rdev) {
3125 printk(KERN_ERR "md: could not alloc mem for new device!\n");
3126 return ERR_PTR(-ENOMEM);
3129 err = md_rdev_init(rdev);
3130 if (err)
3131 goto abort_free;
3132 err = alloc_disk_sb(rdev);
3133 if (err)
3134 goto abort_free;
3136 err = lock_rdev(rdev, newdev, super_format == -2);
3137 if (err)
3138 goto abort_free;
3140 kobject_init(&rdev->kobj, &rdev_ktype);
3142 size = i_size_read(rdev->bdev->bd_inode) >> BLOCK_SIZE_BITS;
3143 if (!size) {
3144 printk(KERN_WARNING
3145 "md: %s has zero or unknown size, marking faulty!\n",
3146 bdevname(rdev->bdev,b));
3147 err = -EINVAL;
3148 goto abort_free;
3151 if (super_format >= 0) {
3152 err = super_types[super_format].
3153 load_super(rdev, NULL, super_minor);
3154 if (err == -EINVAL) {
3155 printk(KERN_WARNING
3156 "md: %s does not have a valid v%d.%d "
3157 "superblock, not importing!\n",
3158 bdevname(rdev->bdev,b),
3159 super_format, super_minor);
3160 goto abort_free;
3162 if (err < 0) {
3163 printk(KERN_WARNING
3164 "md: could not read %s's sb, not importing!\n",
3165 bdevname(rdev->bdev,b));
3166 goto abort_free;
3169 if (super_format == -1)
3170 /* hot-add for 0.90, or non-persistent: so no badblocks */
3171 rdev->badblocks.shift = -1;
3173 return rdev;
3175 abort_free:
3176 if (rdev->bdev)
3177 unlock_rdev(rdev);
3178 free_disk_sb(rdev);
3179 kfree(rdev->badblocks.page);
3180 kfree(rdev);
3181 return ERR_PTR(err);
3185 * Check a full RAID array for plausibility
3189 static void analyze_sbs(struct mddev * mddev)
3191 int i;
3192 struct md_rdev *rdev, *freshest, *tmp;
3193 char b[BDEVNAME_SIZE];
3195 freshest = NULL;
3196 rdev_for_each(rdev, tmp, mddev)
3197 switch (super_types[mddev->major_version].
3198 load_super(rdev, freshest, mddev->minor_version)) {
3199 case 1:
3200 freshest = rdev;
3201 break;
3202 case 0:
3203 break;
3204 default:
3205 printk( KERN_ERR \
3206 "md: fatal superblock inconsistency in %s"
3207 " -- removing from array\n",
3208 bdevname(rdev->bdev,b));
3209 kick_rdev_from_array(rdev);
3213 super_types[mddev->major_version].
3214 validate_super(mddev, freshest);
3216 i = 0;
3217 rdev_for_each(rdev, tmp, mddev) {
3218 if (mddev->max_disks &&
3219 (rdev->desc_nr >= mddev->max_disks ||
3220 i > mddev->max_disks)) {
3221 printk(KERN_WARNING
3222 "md: %s: %s: only %d devices permitted\n",
3223 mdname(mddev), bdevname(rdev->bdev, b),
3224 mddev->max_disks);
3225 kick_rdev_from_array(rdev);
3226 continue;
3228 if (rdev != freshest)
3229 if (super_types[mddev->major_version].
3230 validate_super(mddev, rdev)) {
3231 printk(KERN_WARNING "md: kicking non-fresh %s"
3232 " from array!\n",
3233 bdevname(rdev->bdev,b));
3234 kick_rdev_from_array(rdev);
3235 continue;
3237 if (mddev->level == LEVEL_MULTIPATH) {
3238 rdev->desc_nr = i++;
3239 rdev->raid_disk = rdev->desc_nr;
3240 set_bit(In_sync, &rdev->flags);
3241 } else if (rdev->raid_disk >= (mddev->raid_disks - min(0, mddev->delta_disks))) {
3242 rdev->raid_disk = -1;
3243 clear_bit(In_sync, &rdev->flags);
3248 /* Read a fixed-point number.
3249 * Numbers in sysfs attributes should be in "standard" units where
3250 * possible, so time should be in seconds.
3251 * However we internally use a a much smaller unit such as
3252 * milliseconds or jiffies.
3253 * This function takes a decimal number with a possible fractional
3254 * component, and produces an integer which is the result of
3255 * multiplying that number by 10^'scale'.
3256 * all without any floating-point arithmetic.
3258 int strict_strtoul_scaled(const char *cp, unsigned long *res, int scale)
3260 unsigned long result = 0;
3261 long decimals = -1;
3262 while (isdigit(*cp) || (*cp == '.' && decimals < 0)) {
3263 if (*cp == '.')
3264 decimals = 0;
3265 else if (decimals < scale) {
3266 unsigned int value;
3267 value = *cp - '0';
3268 result = result * 10 + value;
3269 if (decimals >= 0)
3270 decimals++;
3272 cp++;
3274 if (*cp == '\n')
3275 cp++;
3276 if (*cp)
3277 return -EINVAL;
3278 if (decimals < 0)
3279 decimals = 0;
3280 while (decimals < scale) {
3281 result *= 10;
3282 decimals ++;
3284 *res = result;
3285 return 0;
3289 static void md_safemode_timeout(unsigned long data);
3291 static ssize_t
3292 safe_delay_show(struct mddev *mddev, char *page)
3294 int msec = (mddev->safemode_delay*1000)/HZ;
3295 return sprintf(page, "%d.%03d\n", msec/1000, msec%1000);
3297 static ssize_t
3298 safe_delay_store(struct mddev *mddev, const char *cbuf, size_t len)
3300 unsigned long msec;
3302 if (strict_strtoul_scaled(cbuf, &msec, 3) < 0)
3303 return -EINVAL;
3304 if (msec == 0)
3305 mddev->safemode_delay = 0;
3306 else {
3307 unsigned long old_delay = mddev->safemode_delay;
3308 mddev->safemode_delay = (msec*HZ)/1000;
3309 if (mddev->safemode_delay == 0)
3310 mddev->safemode_delay = 1;
3311 if (mddev->safemode_delay < old_delay)
3312 md_safemode_timeout((unsigned long)mddev);
3314 return len;
3316 static struct md_sysfs_entry md_safe_delay =
3317 __ATTR(safe_mode_delay, S_IRUGO|S_IWUSR,safe_delay_show, safe_delay_store);
3319 static ssize_t
3320 level_show(struct mddev *mddev, char *page)
3322 struct md_personality *p = mddev->pers;
3323 if (p)
3324 return sprintf(page, "%s\n", p->name);
3325 else if (mddev->clevel[0])
3326 return sprintf(page, "%s\n", mddev->clevel);
3327 else if (mddev->level != LEVEL_NONE)
3328 return sprintf(page, "%d\n", mddev->level);
3329 else
3330 return 0;
3333 static ssize_t
3334 level_store(struct mddev *mddev, const char *buf, size_t len)
3336 char clevel[16];
3337 ssize_t rv = len;
3338 struct md_personality *pers;
3339 long level;
3340 void *priv;
3341 struct md_rdev *rdev;
3343 if (mddev->pers == NULL) {
3344 if (len == 0)
3345 return 0;
3346 if (len >= sizeof(mddev->clevel))
3347 return -ENOSPC;
3348 strncpy(mddev->clevel, buf, len);
3349 if (mddev->clevel[len-1] == '\n')
3350 len--;
3351 mddev->clevel[len] = 0;
3352 mddev->level = LEVEL_NONE;
3353 return rv;
3356 /* request to change the personality. Need to ensure:
3357 * - array is not engaged in resync/recovery/reshape
3358 * - old personality can be suspended
3359 * - new personality will access other array.
3362 if (mddev->sync_thread ||
3363 mddev->reshape_position != MaxSector ||
3364 mddev->sysfs_active)
3365 return -EBUSY;
3367 if (!mddev->pers->quiesce) {
3368 printk(KERN_WARNING "md: %s: %s does not support online personality change\n",
3369 mdname(mddev), mddev->pers->name);
3370 return -EINVAL;
3373 /* Now find the new personality */
3374 if (len == 0 || len >= sizeof(clevel))
3375 return -EINVAL;
3376 strncpy(clevel, buf, len);
3377 if (clevel[len-1] == '\n')
3378 len--;
3379 clevel[len] = 0;
3380 if (strict_strtol(clevel, 10, &level))
3381 level = LEVEL_NONE;
3383 if (request_module("md-%s", clevel) != 0)
3384 request_module("md-level-%s", clevel);
3385 spin_lock(&pers_lock);
3386 pers = find_pers(level, clevel);
3387 if (!pers || !try_module_get(pers->owner)) {
3388 spin_unlock(&pers_lock);
3389 printk(KERN_WARNING "md: personality %s not loaded\n", clevel);
3390 return -EINVAL;
3392 spin_unlock(&pers_lock);
3394 if (pers == mddev->pers) {
3395 /* Nothing to do! */
3396 module_put(pers->owner);
3397 return rv;
3399 if (!pers->takeover) {
3400 module_put(pers->owner);
3401 printk(KERN_WARNING "md: %s: %s does not support personality takeover\n",
3402 mdname(mddev), clevel);
3403 return -EINVAL;
3406 list_for_each_entry(rdev, &mddev->disks, same_set)
3407 rdev->new_raid_disk = rdev->raid_disk;
3409 /* ->takeover must set new_* and/or delta_disks
3410 * if it succeeds, and may set them when it fails.
3412 priv = pers->takeover(mddev);
3413 if (IS_ERR(priv)) {
3414 mddev->new_level = mddev->level;
3415 mddev->new_layout = mddev->layout;
3416 mddev->new_chunk_sectors = mddev->chunk_sectors;
3417 mddev->raid_disks -= mddev->delta_disks;
3418 mddev->delta_disks = 0;
3419 module_put(pers->owner);
3420 printk(KERN_WARNING "md: %s: %s would not accept array\n",
3421 mdname(mddev), clevel);
3422 return PTR_ERR(priv);
3425 /* Looks like we have a winner */
3426 mddev_suspend(mddev);
3427 mddev->pers->stop(mddev);
3429 if (mddev->pers->sync_request == NULL &&
3430 pers->sync_request != NULL) {
3431 /* need to add the md_redundancy_group */
3432 if (sysfs_create_group(&mddev->kobj, &md_redundancy_group))
3433 printk(KERN_WARNING
3434 "md: cannot register extra attributes for %s\n",
3435 mdname(mddev));
3436 mddev->sysfs_action = sysfs_get_dirent(mddev->kobj.sd, NULL, "sync_action");
3438 if (mddev->pers->sync_request != NULL &&
3439 pers->sync_request == NULL) {
3440 /* need to remove the md_redundancy_group */
3441 if (mddev->to_remove == NULL)
3442 mddev->to_remove = &md_redundancy_group;
3445 if (mddev->pers->sync_request == NULL &&
3446 mddev->external) {
3447 /* We are converting from a no-redundancy array
3448 * to a redundancy array and metadata is managed
3449 * externally so we need to be sure that writes
3450 * won't block due to a need to transition
3451 * clean->dirty
3452 * until external management is started.
3454 mddev->in_sync = 0;
3455 mddev->safemode_delay = 0;
3456 mddev->safemode = 0;
3459 list_for_each_entry(rdev, &mddev->disks, same_set) {
3460 if (rdev->raid_disk < 0)
3461 continue;
3462 if (rdev->new_raid_disk >= mddev->raid_disks)
3463 rdev->new_raid_disk = -1;
3464 if (rdev->new_raid_disk == rdev->raid_disk)
3465 continue;
3466 sysfs_unlink_rdev(mddev, rdev);
3468 list_for_each_entry(rdev, &mddev->disks, same_set) {
3469 if (rdev->raid_disk < 0)
3470 continue;
3471 if (rdev->new_raid_disk == rdev->raid_disk)
3472 continue;
3473 rdev->raid_disk = rdev->new_raid_disk;
3474 if (rdev->raid_disk < 0)
3475 clear_bit(In_sync, &rdev->flags);
3476 else {
3477 if (sysfs_link_rdev(mddev, rdev))
3478 printk(KERN_WARNING "md: cannot register rd%d"
3479 " for %s after level change\n",
3480 rdev->raid_disk, mdname(mddev));
3484 module_put(mddev->pers->owner);
3485 mddev->pers = pers;
3486 mddev->private = priv;
3487 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
3488 mddev->level = mddev->new_level;
3489 mddev->layout = mddev->new_layout;
3490 mddev->chunk_sectors = mddev->new_chunk_sectors;
3491 mddev->delta_disks = 0;
3492 mddev->degraded = 0;
3493 if (mddev->pers->sync_request == NULL) {
3494 /* this is now an array without redundancy, so
3495 * it must always be in_sync
3497 mddev->in_sync = 1;
3498 del_timer_sync(&mddev->safemode_timer);
3500 pers->run(mddev);
3501 mddev_resume(mddev);
3502 set_bit(MD_CHANGE_DEVS, &mddev->flags);
3503 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3504 md_wakeup_thread(mddev->thread);
3505 sysfs_notify(&mddev->kobj, NULL, "level");
3506 md_new_event(mddev);
3507 return rv;
3510 static struct md_sysfs_entry md_level =
3511 __ATTR(level, S_IRUGO|S_IWUSR, level_show, level_store);
3514 static ssize_t
3515 layout_show(struct mddev *mddev, char *page)
3517 /* just a number, not meaningful for all levels */
3518 if (mddev->reshape_position != MaxSector &&
3519 mddev->layout != mddev->new_layout)
3520 return sprintf(page, "%d (%d)\n",
3521 mddev->new_layout, mddev->layout);
3522 return sprintf(page, "%d\n", mddev->layout);
3525 static ssize_t
3526 layout_store(struct mddev *mddev, const char *buf, size_t len)
3528 char *e;
3529 unsigned long n = simple_strtoul(buf, &e, 10);
3531 if (!*buf || (*e && *e != '\n'))
3532 return -EINVAL;
3534 if (mddev->pers) {
3535 int err;
3536 if (mddev->pers->check_reshape == NULL)
3537 return -EBUSY;
3538 mddev->new_layout = n;
3539 err = mddev->pers->check_reshape(mddev);
3540 if (err) {
3541 mddev->new_layout = mddev->layout;
3542 return err;
3544 } else {
3545 mddev->new_layout = n;
3546 if (mddev->reshape_position == MaxSector)
3547 mddev->layout = n;
3549 return len;
3551 static struct md_sysfs_entry md_layout =
3552 __ATTR(layout, S_IRUGO|S_IWUSR, layout_show, layout_store);
3555 static ssize_t
3556 raid_disks_show(struct mddev *mddev, char *page)
3558 if (mddev->raid_disks == 0)
3559 return 0;
3560 if (mddev->reshape_position != MaxSector &&
3561 mddev->delta_disks != 0)
3562 return sprintf(page, "%d (%d)\n", mddev->raid_disks,
3563 mddev->raid_disks - mddev->delta_disks);
3564 return sprintf(page, "%d\n", mddev->raid_disks);
3567 static int update_raid_disks(struct mddev *mddev, int raid_disks);
3569 static ssize_t
3570 raid_disks_store(struct mddev *mddev, const char *buf, size_t len)
3572 char *e;
3573 int rv = 0;
3574 unsigned long n = simple_strtoul(buf, &e, 10);
3576 if (!*buf || (*e && *e != '\n'))
3577 return -EINVAL;
3579 if (mddev->pers)
3580 rv = update_raid_disks(mddev, n);
3581 else if (mddev->reshape_position != MaxSector) {
3582 int olddisks = mddev->raid_disks - mddev->delta_disks;
3583 mddev->delta_disks = n - olddisks;
3584 mddev->raid_disks = n;
3585 } else
3586 mddev->raid_disks = n;
3587 return rv ? rv : len;
3589 static struct md_sysfs_entry md_raid_disks =
3590 __ATTR(raid_disks, S_IRUGO|S_IWUSR, raid_disks_show, raid_disks_store);
3592 static ssize_t
3593 chunk_size_show(struct mddev *mddev, char *page)
3595 if (mddev->reshape_position != MaxSector &&
3596 mddev->chunk_sectors != mddev->new_chunk_sectors)
3597 return sprintf(page, "%d (%d)\n",
3598 mddev->new_chunk_sectors << 9,
3599 mddev->chunk_sectors << 9);
3600 return sprintf(page, "%d\n", mddev->chunk_sectors << 9);
3603 static ssize_t
3604 chunk_size_store(struct mddev *mddev, const char *buf, size_t len)
3606 char *e;
3607 unsigned long n = simple_strtoul(buf, &e, 10);
3609 if (!*buf || (*e && *e != '\n'))
3610 return -EINVAL;
3612 if (mddev->pers) {
3613 int err;
3614 if (mddev->pers->check_reshape == NULL)
3615 return -EBUSY;
3616 mddev->new_chunk_sectors = n >> 9;
3617 err = mddev->pers->check_reshape(mddev);
3618 if (err) {
3619 mddev->new_chunk_sectors = mddev->chunk_sectors;
3620 return err;
3622 } else {
3623 mddev->new_chunk_sectors = n >> 9;
3624 if (mddev->reshape_position == MaxSector)
3625 mddev->chunk_sectors = n >> 9;
3627 return len;
3629 static struct md_sysfs_entry md_chunk_size =
3630 __ATTR(chunk_size, S_IRUGO|S_IWUSR, chunk_size_show, chunk_size_store);
3632 static ssize_t
3633 resync_start_show(struct mddev *mddev, char *page)
3635 if (mddev->recovery_cp == MaxSector)
3636 return sprintf(page, "none\n");
3637 return sprintf(page, "%llu\n", (unsigned long long)mddev->recovery_cp);
3640 static ssize_t
3641 resync_start_store(struct mddev *mddev, const char *buf, size_t len)
3643 char *e;
3644 unsigned long long n = simple_strtoull(buf, &e, 10);
3646 if (mddev->pers && !test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
3647 return -EBUSY;
3648 if (cmd_match(buf, "none"))
3649 n = MaxSector;
3650 else if (!*buf || (*e && *e != '\n'))
3651 return -EINVAL;
3653 mddev->recovery_cp = n;
3654 return len;
3656 static struct md_sysfs_entry md_resync_start =
3657 __ATTR(resync_start, S_IRUGO|S_IWUSR, resync_start_show, resync_start_store);
3660 * The array state can be:
3662 * clear
3663 * No devices, no size, no level
3664 * Equivalent to STOP_ARRAY ioctl
3665 * inactive
3666 * May have some settings, but array is not active
3667 * all IO results in error
3668 * When written, doesn't tear down array, but just stops it
3669 * suspended (not supported yet)
3670 * All IO requests will block. The array can be reconfigured.
3671 * Writing this, if accepted, will block until array is quiescent
3672 * readonly
3673 * no resync can happen. no superblocks get written.
3674 * write requests fail
3675 * read-auto
3676 * like readonly, but behaves like 'clean' on a write request.
3678 * clean - no pending writes, but otherwise active.
3679 * When written to inactive array, starts without resync
3680 * If a write request arrives then
3681 * if metadata is known, mark 'dirty' and switch to 'active'.
3682 * if not known, block and switch to write-pending
3683 * If written to an active array that has pending writes, then fails.
3684 * active
3685 * fully active: IO and resync can be happening.
3686 * When written to inactive array, starts with resync
3688 * write-pending
3689 * clean, but writes are blocked waiting for 'active' to be written.
3691 * active-idle
3692 * like active, but no writes have been seen for a while (100msec).
3695 enum array_state { clear, inactive, suspended, readonly, read_auto, clean, active,
3696 write_pending, active_idle, bad_word};
3697 static char *array_states[] = {
3698 "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active",
3699 "write-pending", "active-idle", NULL };
3701 static int match_word(const char *word, char **list)
3703 int n;
3704 for (n=0; list[n]; n++)
3705 if (cmd_match(word, list[n]))
3706 break;
3707 return n;
3710 static ssize_t
3711 array_state_show(struct mddev *mddev, char *page)
3713 enum array_state st = inactive;
3715 if (mddev->pers)
3716 switch(mddev->ro) {
3717 case 1:
3718 st = readonly;
3719 break;
3720 case 2:
3721 st = read_auto;
3722 break;
3723 case 0:
3724 if (mddev->in_sync)
3725 st = clean;
3726 else if (test_bit(MD_CHANGE_PENDING, &mddev->flags))
3727 st = write_pending;
3728 else if (mddev->safemode)
3729 st = active_idle;
3730 else
3731 st = active;
3733 else {
3734 if (list_empty(&mddev->disks) &&
3735 mddev->raid_disks == 0 &&
3736 mddev->dev_sectors == 0)
3737 st = clear;
3738 else
3739 st = inactive;
3741 return sprintf(page, "%s\n", array_states[st]);
3744 static int do_md_stop(struct mddev * mddev, int ro, int is_open);
3745 static int md_set_readonly(struct mddev * mddev, int is_open);
3746 static int do_md_run(struct mddev * mddev);
3747 static int restart_array(struct mddev *mddev);
3749 static ssize_t
3750 array_state_store(struct mddev *mddev, const char *buf, size_t len)
3752 int err = -EINVAL;
3753 enum array_state st = match_word(buf, array_states);
3754 switch(st) {
3755 case bad_word:
3756 break;
3757 case clear:
3758 /* stopping an active array */
3759 if (atomic_read(&mddev->openers) > 0)
3760 return -EBUSY;
3761 err = do_md_stop(mddev, 0, 0);
3762 break;
3763 case inactive:
3764 /* stopping an active array */
3765 if (mddev->pers) {
3766 if (atomic_read(&mddev->openers) > 0)
3767 return -EBUSY;
3768 err = do_md_stop(mddev, 2, 0);
3769 } else
3770 err = 0; /* already inactive */
3771 break;
3772 case suspended:
3773 break; /* not supported yet */
3774 case readonly:
3775 if (mddev->pers)
3776 err = md_set_readonly(mddev, 0);
3777 else {
3778 mddev->ro = 1;
3779 set_disk_ro(mddev->gendisk, 1);
3780 err = do_md_run(mddev);
3782 break;
3783 case read_auto:
3784 if (mddev->pers) {
3785 if (mddev->ro == 0)
3786 err = md_set_readonly(mddev, 0);
3787 else if (mddev->ro == 1)
3788 err = restart_array(mddev);
3789 if (err == 0) {
3790 mddev->ro = 2;
3791 set_disk_ro(mddev->gendisk, 0);
3793 } else {
3794 mddev->ro = 2;
3795 err = do_md_run(mddev);
3797 break;
3798 case clean:
3799 if (mddev->pers) {
3800 restart_array(mddev);
3801 spin_lock_irq(&mddev->write_lock);
3802 if (atomic_read(&mddev->writes_pending) == 0) {
3803 if (mddev->in_sync == 0) {
3804 mddev->in_sync = 1;
3805 if (mddev->safemode == 1)
3806 mddev->safemode = 0;
3807 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
3809 err = 0;
3810 } else
3811 err = -EBUSY;
3812 spin_unlock_irq(&mddev->write_lock);
3813 } else
3814 err = -EINVAL;
3815 break;
3816 case active:
3817 if (mddev->pers) {
3818 restart_array(mddev);
3819 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
3820 wake_up(&mddev->sb_wait);
3821 err = 0;
3822 } else {
3823 mddev->ro = 0;
3824 set_disk_ro(mddev->gendisk, 0);
3825 err = do_md_run(mddev);
3827 break;
3828 case write_pending:
3829 case active_idle:
3830 /* these cannot be set */
3831 break;
3833 if (err)
3834 return err;
3835 else {
3836 if (mddev->hold_active == UNTIL_IOCTL)
3837 mddev->hold_active = 0;
3838 sysfs_notify_dirent_safe(mddev->sysfs_state);
3839 return len;
3842 static struct md_sysfs_entry md_array_state =
3843 __ATTR(array_state, S_IRUGO|S_IWUSR, array_state_show, array_state_store);
3845 static ssize_t
3846 max_corrected_read_errors_show(struct mddev *mddev, char *page) {
3847 return sprintf(page, "%d\n",
3848 atomic_read(&mddev->max_corr_read_errors));
3851 static ssize_t
3852 max_corrected_read_errors_store(struct mddev *mddev, const char *buf, size_t len)
3854 char *e;
3855 unsigned long n = simple_strtoul(buf, &e, 10);
3857 if (*buf && (*e == 0 || *e == '\n')) {
3858 atomic_set(&mddev->max_corr_read_errors, n);
3859 return len;
3861 return -EINVAL;
3864 static struct md_sysfs_entry max_corr_read_errors =
3865 __ATTR(max_read_errors, S_IRUGO|S_IWUSR, max_corrected_read_errors_show,
3866 max_corrected_read_errors_store);
3868 static ssize_t
3869 null_show(struct mddev *mddev, char *page)
3871 return -EINVAL;
3874 static ssize_t
3875 new_dev_store(struct mddev *mddev, const char *buf, size_t len)
3877 /* buf must be %d:%d\n? giving major and minor numbers */
3878 /* The new device is added to the array.
3879 * If the array has a persistent superblock, we read the
3880 * superblock to initialise info and check validity.
3881 * Otherwise, only checking done is that in bind_rdev_to_array,
3882 * which mainly checks size.
3884 char *e;
3885 int major = simple_strtoul(buf, &e, 10);
3886 int minor;
3887 dev_t dev;
3888 struct md_rdev *rdev;
3889 int err;
3891 if (!*buf || *e != ':' || !e[1] || e[1] == '\n')
3892 return -EINVAL;
3893 minor = simple_strtoul(e+1, &e, 10);
3894 if (*e && *e != '\n')
3895 return -EINVAL;
3896 dev = MKDEV(major, minor);
3897 if (major != MAJOR(dev) ||
3898 minor != MINOR(dev))
3899 return -EOVERFLOW;
3902 if (mddev->persistent) {
3903 rdev = md_import_device(dev, mddev->major_version,
3904 mddev->minor_version);
3905 if (!IS_ERR(rdev) && !list_empty(&mddev->disks)) {
3906 struct md_rdev *rdev0
3907 = list_entry(mddev->disks.next,
3908 struct md_rdev, same_set);
3909 err = super_types[mddev->major_version]
3910 .load_super(rdev, rdev0, mddev->minor_version);
3911 if (err < 0)
3912 goto out;
3914 } else if (mddev->external)
3915 rdev = md_import_device(dev, -2, -1);
3916 else
3917 rdev = md_import_device(dev, -1, -1);
3919 if (IS_ERR(rdev))
3920 return PTR_ERR(rdev);
3921 err = bind_rdev_to_array(rdev, mddev);
3922 out:
3923 if (err)
3924 export_rdev(rdev);
3925 return err ? err : len;
3928 static struct md_sysfs_entry md_new_device =
3929 __ATTR(new_dev, S_IWUSR, null_show, new_dev_store);
3931 static ssize_t
3932 bitmap_store(struct mddev *mddev, const char *buf, size_t len)
3934 char *end;
3935 unsigned long chunk, end_chunk;
3937 if (!mddev->bitmap)
3938 goto out;
3939 /* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */
3940 while (*buf) {
3941 chunk = end_chunk = simple_strtoul(buf, &end, 0);
3942 if (buf == end) break;
3943 if (*end == '-') { /* range */
3944 buf = end + 1;
3945 end_chunk = simple_strtoul(buf, &end, 0);
3946 if (buf == end) break;
3948 if (*end && !isspace(*end)) break;
3949 bitmap_dirty_bits(mddev->bitmap, chunk, end_chunk);
3950 buf = skip_spaces(end);
3952 bitmap_unplug(mddev->bitmap); /* flush the bits to disk */
3953 out:
3954 return len;
3957 static struct md_sysfs_entry md_bitmap =
3958 __ATTR(bitmap_set_bits, S_IWUSR, null_show, bitmap_store);
3960 static ssize_t
3961 size_show(struct mddev *mddev, char *page)
3963 return sprintf(page, "%llu\n",
3964 (unsigned long long)mddev->dev_sectors / 2);
3967 static int update_size(struct mddev *mddev, sector_t num_sectors);
3969 static ssize_t
3970 size_store(struct mddev *mddev, const char *buf, size_t len)
3972 /* If array is inactive, we can reduce the component size, but
3973 * not increase it (except from 0).
3974 * If array is active, we can try an on-line resize
3976 sector_t sectors;
3977 int err = strict_blocks_to_sectors(buf, &sectors);
3979 if (err < 0)
3980 return err;
3981 if (mddev->pers) {
3982 err = update_size(mddev, sectors);
3983 md_update_sb(mddev, 1);
3984 } else {
3985 if (mddev->dev_sectors == 0 ||
3986 mddev->dev_sectors > sectors)
3987 mddev->dev_sectors = sectors;
3988 else
3989 err = -ENOSPC;
3991 return err ? err : len;
3994 static struct md_sysfs_entry md_size =
3995 __ATTR(component_size, S_IRUGO|S_IWUSR, size_show, size_store);
3998 /* Metdata version.
3999 * This is one of
4000 * 'none' for arrays with no metadata (good luck...)
4001 * 'external' for arrays with externally managed metadata,
4002 * or N.M for internally known formats
4004 static ssize_t
4005 metadata_show(struct mddev *mddev, char *page)
4007 if (mddev->persistent)
4008 return sprintf(page, "%d.%d\n",
4009 mddev->major_version, mddev->minor_version);
4010 else if (mddev->external)
4011 return sprintf(page, "external:%s\n", mddev->metadata_type);
4012 else
4013 return sprintf(page, "none\n");
4016 static ssize_t
4017 metadata_store(struct mddev *mddev, const char *buf, size_t len)
4019 int major, minor;
4020 char *e;
4021 /* Changing the details of 'external' metadata is
4022 * always permitted. Otherwise there must be
4023 * no devices attached to the array.
4025 if (mddev->external && strncmp(buf, "external:", 9) == 0)
4027 else if (!list_empty(&mddev->disks))
4028 return -EBUSY;
4030 if (cmd_match(buf, "none")) {
4031 mddev->persistent = 0;
4032 mddev->external = 0;
4033 mddev->major_version = 0;
4034 mddev->minor_version = 90;
4035 return len;
4037 if (strncmp(buf, "external:", 9) == 0) {
4038 size_t namelen = len-9;
4039 if (namelen >= sizeof(mddev->metadata_type))
4040 namelen = sizeof(mddev->metadata_type)-1;
4041 strncpy(mddev->metadata_type, buf+9, namelen);
4042 mddev->metadata_type[namelen] = 0;
4043 if (namelen && mddev->metadata_type[namelen-1] == '\n')
4044 mddev->metadata_type[--namelen] = 0;
4045 mddev->persistent = 0;
4046 mddev->external = 1;
4047 mddev->major_version = 0;
4048 mddev->minor_version = 90;
4049 return len;
4051 major = simple_strtoul(buf, &e, 10);
4052 if (e==buf || *e != '.')
4053 return -EINVAL;
4054 buf = e+1;
4055 minor = simple_strtoul(buf, &e, 10);
4056 if (e==buf || (*e && *e != '\n') )
4057 return -EINVAL;
4058 if (major >= ARRAY_SIZE(super_types) || super_types[major].name == NULL)
4059 return -ENOENT;
4060 mddev->major_version = major;
4061 mddev->minor_version = minor;
4062 mddev->persistent = 1;
4063 mddev->external = 0;
4064 return len;
4067 static struct md_sysfs_entry md_metadata =
4068 __ATTR(metadata_version, S_IRUGO|S_IWUSR, metadata_show, metadata_store);
4070 static ssize_t
4071 action_show(struct mddev *mddev, char *page)
4073 char *type = "idle";
4074 if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
4075 type = "frozen";
4076 else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
4077 (!mddev->ro && test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))) {
4078 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
4079 type = "reshape";
4080 else if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
4081 if (!test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
4082 type = "resync";
4083 else if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
4084 type = "check";
4085 else
4086 type = "repair";
4087 } else if (test_bit(MD_RECOVERY_RECOVER, &mddev->recovery))
4088 type = "recover";
4090 return sprintf(page, "%s\n", type);
4093 static void reap_sync_thread(struct mddev *mddev);
4095 static ssize_t
4096 action_store(struct mddev *mddev, const char *page, size_t len)
4098 if (!mddev->pers || !mddev->pers->sync_request)
4099 return -EINVAL;
4101 if (cmd_match(page, "frozen"))
4102 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4103 else
4104 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4106 if (cmd_match(page, "idle") || cmd_match(page, "frozen")) {
4107 if (mddev->sync_thread) {
4108 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
4109 reap_sync_thread(mddev);
4111 } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
4112 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
4113 return -EBUSY;
4114 else if (cmd_match(page, "resync"))
4115 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4116 else if (cmd_match(page, "recover")) {
4117 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
4118 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4119 } else if (cmd_match(page, "reshape")) {
4120 int err;
4121 if (mddev->pers->start_reshape == NULL)
4122 return -EINVAL;
4123 err = mddev->pers->start_reshape(mddev);
4124 if (err)
4125 return err;
4126 sysfs_notify(&mddev->kobj, NULL, "degraded");
4127 } else {
4128 if (cmd_match(page, "check"))
4129 set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
4130 else if (!cmd_match(page, "repair"))
4131 return -EINVAL;
4132 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
4133 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
4135 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4136 md_wakeup_thread(mddev->thread);
4137 sysfs_notify_dirent_safe(mddev->sysfs_action);
4138 return len;
4141 static ssize_t
4142 mismatch_cnt_show(struct mddev *mddev, char *page)
4144 return sprintf(page, "%llu\n",
4145 (unsigned long long) mddev->resync_mismatches);
4148 static struct md_sysfs_entry md_scan_mode =
4149 __ATTR(sync_action, S_IRUGO|S_IWUSR, action_show, action_store);
4152 static struct md_sysfs_entry md_mismatches = __ATTR_RO(mismatch_cnt);
4154 static ssize_t
4155 sync_min_show(struct mddev *mddev, char *page)
4157 return sprintf(page, "%d (%s)\n", speed_min(mddev),
4158 mddev->sync_speed_min ? "local": "system");
4161 static ssize_t
4162 sync_min_store(struct mddev *mddev, const char *buf, size_t len)
4164 int min;
4165 char *e;
4166 if (strncmp(buf, "system", 6)==0) {
4167 mddev->sync_speed_min = 0;
4168 return len;
4170 min = simple_strtoul(buf, &e, 10);
4171 if (buf == e || (*e && *e != '\n') || min <= 0)
4172 return -EINVAL;
4173 mddev->sync_speed_min = min;
4174 return len;
4177 static struct md_sysfs_entry md_sync_min =
4178 __ATTR(sync_speed_min, S_IRUGO|S_IWUSR, sync_min_show, sync_min_store);
4180 static ssize_t
4181 sync_max_show(struct mddev *mddev, char *page)
4183 return sprintf(page, "%d (%s)\n", speed_max(mddev),
4184 mddev->sync_speed_max ? "local": "system");
4187 static ssize_t
4188 sync_max_store(struct mddev *mddev, const char *buf, size_t len)
4190 int max;
4191 char *e;
4192 if (strncmp(buf, "system", 6)==0) {
4193 mddev->sync_speed_max = 0;
4194 return len;
4196 max = simple_strtoul(buf, &e, 10);
4197 if (buf == e || (*e && *e != '\n') || max <= 0)
4198 return -EINVAL;
4199 mddev->sync_speed_max = max;
4200 return len;
4203 static struct md_sysfs_entry md_sync_max =
4204 __ATTR(sync_speed_max, S_IRUGO|S_IWUSR, sync_max_show, sync_max_store);
4206 static ssize_t
4207 degraded_show(struct mddev *mddev, char *page)
4209 return sprintf(page, "%d\n", mddev->degraded);
4211 static struct md_sysfs_entry md_degraded = __ATTR_RO(degraded);
4213 static ssize_t
4214 sync_force_parallel_show(struct mddev *mddev, char *page)
4216 return sprintf(page, "%d\n", mddev->parallel_resync);
4219 static ssize_t
4220 sync_force_parallel_store(struct mddev *mddev, const char *buf, size_t len)
4222 long n;
4224 if (strict_strtol(buf, 10, &n))
4225 return -EINVAL;
4227 if (n != 0 && n != 1)
4228 return -EINVAL;
4230 mddev->parallel_resync = n;
4232 if (mddev->sync_thread)
4233 wake_up(&resync_wait);
4235 return len;
4238 /* force parallel resync, even with shared block devices */
4239 static struct md_sysfs_entry md_sync_force_parallel =
4240 __ATTR(sync_force_parallel, S_IRUGO|S_IWUSR,
4241 sync_force_parallel_show, sync_force_parallel_store);
4243 static ssize_t
4244 sync_speed_show(struct mddev *mddev, char *page)
4246 unsigned long resync, dt, db;
4247 if (mddev->curr_resync == 0)
4248 return sprintf(page, "none\n");
4249 resync = mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active);
4250 dt = (jiffies - mddev->resync_mark) / HZ;
4251 if (!dt) dt++;
4252 db = resync - mddev->resync_mark_cnt;
4253 return sprintf(page, "%lu\n", db/dt/2); /* K/sec */
4256 static struct md_sysfs_entry md_sync_speed = __ATTR_RO(sync_speed);
4258 static ssize_t
4259 sync_completed_show(struct mddev *mddev, char *page)
4261 unsigned long long max_sectors, resync;
4263 if (!test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4264 return sprintf(page, "none\n");
4266 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
4267 max_sectors = mddev->resync_max_sectors;
4268 else
4269 max_sectors = mddev->dev_sectors;
4271 resync = mddev->curr_resync_completed;
4272 return sprintf(page, "%llu / %llu\n", resync, max_sectors);
4275 static struct md_sysfs_entry md_sync_completed = __ATTR_RO(sync_completed);
4277 static ssize_t
4278 min_sync_show(struct mddev *mddev, char *page)
4280 return sprintf(page, "%llu\n",
4281 (unsigned long long)mddev->resync_min);
4283 static ssize_t
4284 min_sync_store(struct mddev *mddev, const char *buf, size_t len)
4286 unsigned long long min;
4287 if (strict_strtoull(buf, 10, &min))
4288 return -EINVAL;
4289 if (min > mddev->resync_max)
4290 return -EINVAL;
4291 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4292 return -EBUSY;
4294 /* Must be a multiple of chunk_size */
4295 if (mddev->chunk_sectors) {
4296 sector_t temp = min;
4297 if (sector_div(temp, mddev->chunk_sectors))
4298 return -EINVAL;
4300 mddev->resync_min = min;
4302 return len;
4305 static struct md_sysfs_entry md_min_sync =
4306 __ATTR(sync_min, S_IRUGO|S_IWUSR, min_sync_show, min_sync_store);
4308 static ssize_t
4309 max_sync_show(struct mddev *mddev, char *page)
4311 if (mddev->resync_max == MaxSector)
4312 return sprintf(page, "max\n");
4313 else
4314 return sprintf(page, "%llu\n",
4315 (unsigned long long)mddev->resync_max);
4317 static ssize_t
4318 max_sync_store(struct mddev *mddev, const char *buf, size_t len)
4320 if (strncmp(buf, "max", 3) == 0)
4321 mddev->resync_max = MaxSector;
4322 else {
4323 unsigned long long max;
4324 if (strict_strtoull(buf, 10, &max))
4325 return -EINVAL;
4326 if (max < mddev->resync_min)
4327 return -EINVAL;
4328 if (max < mddev->resync_max &&
4329 mddev->ro == 0 &&
4330 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4331 return -EBUSY;
4333 /* Must be a multiple of chunk_size */
4334 if (mddev->chunk_sectors) {
4335 sector_t temp = max;
4336 if (sector_div(temp, mddev->chunk_sectors))
4337 return -EINVAL;
4339 mddev->resync_max = max;
4341 wake_up(&mddev->recovery_wait);
4342 return len;
4345 static struct md_sysfs_entry md_max_sync =
4346 __ATTR(sync_max, S_IRUGO|S_IWUSR, max_sync_show, max_sync_store);
4348 static ssize_t
4349 suspend_lo_show(struct mddev *mddev, char *page)
4351 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_lo);
4354 static ssize_t
4355 suspend_lo_store(struct mddev *mddev, const char *buf, size_t len)
4357 char *e;
4358 unsigned long long new = simple_strtoull(buf, &e, 10);
4359 unsigned long long old = mddev->suspend_lo;
4361 if (mddev->pers == NULL ||
4362 mddev->pers->quiesce == NULL)
4363 return -EINVAL;
4364 if (buf == e || (*e && *e != '\n'))
4365 return -EINVAL;
4367 mddev->suspend_lo = new;
4368 if (new >= old)
4369 /* Shrinking suspended region */
4370 mddev->pers->quiesce(mddev, 2);
4371 else {
4372 /* Expanding suspended region - need to wait */
4373 mddev->pers->quiesce(mddev, 1);
4374 mddev->pers->quiesce(mddev, 0);
4376 return len;
4378 static struct md_sysfs_entry md_suspend_lo =
4379 __ATTR(suspend_lo, S_IRUGO|S_IWUSR, suspend_lo_show, suspend_lo_store);
4382 static ssize_t
4383 suspend_hi_show(struct mddev *mddev, char *page)
4385 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_hi);
4388 static ssize_t
4389 suspend_hi_store(struct mddev *mddev, const char *buf, size_t len)
4391 char *e;
4392 unsigned long long new = simple_strtoull(buf, &e, 10);
4393 unsigned long long old = mddev->suspend_hi;
4395 if (mddev->pers == NULL ||
4396 mddev->pers->quiesce == NULL)
4397 return -EINVAL;
4398 if (buf == e || (*e && *e != '\n'))
4399 return -EINVAL;
4401 mddev->suspend_hi = new;
4402 if (new <= old)
4403 /* Shrinking suspended region */
4404 mddev->pers->quiesce(mddev, 2);
4405 else {
4406 /* Expanding suspended region - need to wait */
4407 mddev->pers->quiesce(mddev, 1);
4408 mddev->pers->quiesce(mddev, 0);
4410 return len;
4412 static struct md_sysfs_entry md_suspend_hi =
4413 __ATTR(suspend_hi, S_IRUGO|S_IWUSR, suspend_hi_show, suspend_hi_store);
4415 static ssize_t
4416 reshape_position_show(struct mddev *mddev, char *page)
4418 if (mddev->reshape_position != MaxSector)
4419 return sprintf(page, "%llu\n",
4420 (unsigned long long)mddev->reshape_position);
4421 strcpy(page, "none\n");
4422 return 5;
4425 static ssize_t
4426 reshape_position_store(struct mddev *mddev, const char *buf, size_t len)
4428 char *e;
4429 unsigned long long new = simple_strtoull(buf, &e, 10);
4430 if (mddev->pers)
4431 return -EBUSY;
4432 if (buf == e || (*e && *e != '\n'))
4433 return -EINVAL;
4434 mddev->reshape_position = new;
4435 mddev->delta_disks = 0;
4436 mddev->new_level = mddev->level;
4437 mddev->new_layout = mddev->layout;
4438 mddev->new_chunk_sectors = mddev->chunk_sectors;
4439 return len;
4442 static struct md_sysfs_entry md_reshape_position =
4443 __ATTR(reshape_position, S_IRUGO|S_IWUSR, reshape_position_show,
4444 reshape_position_store);
4446 static ssize_t
4447 array_size_show(struct mddev *mddev, char *page)
4449 if (mddev->external_size)
4450 return sprintf(page, "%llu\n",
4451 (unsigned long long)mddev->array_sectors/2);
4452 else
4453 return sprintf(page, "default\n");
4456 static ssize_t
4457 array_size_store(struct mddev *mddev, const char *buf, size_t len)
4459 sector_t sectors;
4461 if (strncmp(buf, "default", 7) == 0) {
4462 if (mddev->pers)
4463 sectors = mddev->pers->size(mddev, 0, 0);
4464 else
4465 sectors = mddev->array_sectors;
4467 mddev->external_size = 0;
4468 } else {
4469 if (strict_blocks_to_sectors(buf, &sectors) < 0)
4470 return -EINVAL;
4471 if (mddev->pers && mddev->pers->size(mddev, 0, 0) < sectors)
4472 return -E2BIG;
4474 mddev->external_size = 1;
4477 mddev->array_sectors = sectors;
4478 if (mddev->pers) {
4479 set_capacity(mddev->gendisk, mddev->array_sectors);
4480 revalidate_disk(mddev->gendisk);
4482 return len;
4485 static struct md_sysfs_entry md_array_size =
4486 __ATTR(array_size, S_IRUGO|S_IWUSR, array_size_show,
4487 array_size_store);
4489 static struct attribute *md_default_attrs[] = {
4490 &md_level.attr,
4491 &md_layout.attr,
4492 &md_raid_disks.attr,
4493 &md_chunk_size.attr,
4494 &md_size.attr,
4495 &md_resync_start.attr,
4496 &md_metadata.attr,
4497 &md_new_device.attr,
4498 &md_safe_delay.attr,
4499 &md_array_state.attr,
4500 &md_reshape_position.attr,
4501 &md_array_size.attr,
4502 &max_corr_read_errors.attr,
4503 NULL,
4506 static struct attribute *md_redundancy_attrs[] = {
4507 &md_scan_mode.attr,
4508 &md_mismatches.attr,
4509 &md_sync_min.attr,
4510 &md_sync_max.attr,
4511 &md_sync_speed.attr,
4512 &md_sync_force_parallel.attr,
4513 &md_sync_completed.attr,
4514 &md_min_sync.attr,
4515 &md_max_sync.attr,
4516 &md_suspend_lo.attr,
4517 &md_suspend_hi.attr,
4518 &md_bitmap.attr,
4519 &md_degraded.attr,
4520 NULL,
4522 static struct attribute_group md_redundancy_group = {
4523 .name = NULL,
4524 .attrs = md_redundancy_attrs,
4528 static ssize_t
4529 md_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
4531 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
4532 struct mddev *mddev = container_of(kobj, struct mddev, kobj);
4533 ssize_t rv;
4535 if (!entry->show)
4536 return -EIO;
4537 spin_lock(&all_mddevs_lock);
4538 if (list_empty(&mddev->all_mddevs)) {
4539 spin_unlock(&all_mddevs_lock);
4540 return -EBUSY;
4542 mddev_get(mddev);
4543 spin_unlock(&all_mddevs_lock);
4545 rv = mddev_lock(mddev);
4546 if (!rv) {
4547 rv = entry->show(mddev, page);
4548 mddev_unlock(mddev);
4550 mddev_put(mddev);
4551 return rv;
4554 static ssize_t
4555 md_attr_store(struct kobject *kobj, struct attribute *attr,
4556 const char *page, size_t length)
4558 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
4559 struct mddev *mddev = container_of(kobj, struct mddev, kobj);
4560 ssize_t rv;
4562 if (!entry->store)
4563 return -EIO;
4564 if (!capable(CAP_SYS_ADMIN))
4565 return -EACCES;
4566 spin_lock(&all_mddevs_lock);
4567 if (list_empty(&mddev->all_mddevs)) {
4568 spin_unlock(&all_mddevs_lock);
4569 return -EBUSY;
4571 mddev_get(mddev);
4572 spin_unlock(&all_mddevs_lock);
4573 rv = mddev_lock(mddev);
4574 if (!rv) {
4575 rv = entry->store(mddev, page, length);
4576 mddev_unlock(mddev);
4578 mddev_put(mddev);
4579 return rv;
4582 static void md_free(struct kobject *ko)
4584 struct mddev *mddev = container_of(ko, struct mddev, kobj);
4586 if (mddev->sysfs_state)
4587 sysfs_put(mddev->sysfs_state);
4589 if (mddev->gendisk) {
4590 del_gendisk(mddev->gendisk);
4591 put_disk(mddev->gendisk);
4593 if (mddev->queue)
4594 blk_cleanup_queue(mddev->queue);
4596 kfree(mddev);
4599 static const struct sysfs_ops md_sysfs_ops = {
4600 .show = md_attr_show,
4601 .store = md_attr_store,
4603 static struct kobj_type md_ktype = {
4604 .release = md_free,
4605 .sysfs_ops = &md_sysfs_ops,
4606 .default_attrs = md_default_attrs,
4609 int mdp_major = 0;
4611 static void mddev_delayed_delete(struct work_struct *ws)
4613 struct mddev *mddev = container_of(ws, struct mddev, del_work);
4615 sysfs_remove_group(&mddev->kobj, &md_bitmap_group);
4616 kobject_del(&mddev->kobj);
4617 kobject_put(&mddev->kobj);
4620 static int md_alloc(dev_t dev, char *name)
4622 static DEFINE_MUTEX(disks_mutex);
4623 struct mddev *mddev = mddev_find(dev);
4624 struct gendisk *disk;
4625 int partitioned;
4626 int shift;
4627 int unit;
4628 int error;
4630 if (!mddev)
4631 return -ENODEV;
4633 partitioned = (MAJOR(mddev->unit) != MD_MAJOR);
4634 shift = partitioned ? MdpMinorShift : 0;
4635 unit = MINOR(mddev->unit) >> shift;
4637 /* wait for any previous instance of this device to be
4638 * completely removed (mddev_delayed_delete).
4640 flush_workqueue(md_misc_wq);
4642 mutex_lock(&disks_mutex);
4643 error = -EEXIST;
4644 if (mddev->gendisk)
4645 goto abort;
4647 if (name) {
4648 /* Need to ensure that 'name' is not a duplicate.
4650 struct mddev *mddev2;
4651 spin_lock(&all_mddevs_lock);
4653 list_for_each_entry(mddev2, &all_mddevs, all_mddevs)
4654 if (mddev2->gendisk &&
4655 strcmp(mddev2->gendisk->disk_name, name) == 0) {
4656 spin_unlock(&all_mddevs_lock);
4657 goto abort;
4659 spin_unlock(&all_mddevs_lock);
4662 error = -ENOMEM;
4663 mddev->queue = blk_alloc_queue(GFP_KERNEL);
4664 if (!mddev->queue)
4665 goto abort;
4666 mddev->queue->queuedata = mddev;
4668 blk_queue_make_request(mddev->queue, md_make_request);
4669 blk_set_stacking_limits(&mddev->queue->limits);
4671 disk = alloc_disk(1 << shift);
4672 if (!disk) {
4673 blk_cleanup_queue(mddev->queue);
4674 mddev->queue = NULL;
4675 goto abort;
4677 disk->major = MAJOR(mddev->unit);
4678 disk->first_minor = unit << shift;
4679 if (name)
4680 strcpy(disk->disk_name, name);
4681 else if (partitioned)
4682 sprintf(disk->disk_name, "md_d%d", unit);
4683 else
4684 sprintf(disk->disk_name, "md%d", unit);
4685 disk->fops = &md_fops;
4686 disk->private_data = mddev;
4687 disk->queue = mddev->queue;
4688 blk_queue_flush(mddev->queue, REQ_FLUSH | REQ_FUA);
4689 /* Allow extended partitions. This makes the
4690 * 'mdp' device redundant, but we can't really
4691 * remove it now.
4693 disk->flags |= GENHD_FL_EXT_DEVT;
4694 mddev->gendisk = disk;
4695 /* As soon as we call add_disk(), another thread could get
4696 * through to md_open, so make sure it doesn't get too far
4698 mutex_lock(&mddev->open_mutex);
4699 add_disk(disk);
4701 error = kobject_init_and_add(&mddev->kobj, &md_ktype,
4702 &disk_to_dev(disk)->kobj, "%s", "md");
4703 if (error) {
4704 /* This isn't possible, but as kobject_init_and_add is marked
4705 * __must_check, we must do something with the result
4707 printk(KERN_WARNING "md: cannot register %s/md - name in use\n",
4708 disk->disk_name);
4709 error = 0;
4711 if (mddev->kobj.sd &&
4712 sysfs_create_group(&mddev->kobj, &md_bitmap_group))
4713 printk(KERN_DEBUG "pointless warning\n");
4714 mutex_unlock(&mddev->open_mutex);
4715 abort:
4716 mutex_unlock(&disks_mutex);
4717 if (!error && mddev->kobj.sd) {
4718 kobject_uevent(&mddev->kobj, KOBJ_ADD);
4719 mddev->sysfs_state = sysfs_get_dirent_safe(mddev->kobj.sd, "array_state");
4721 mddev_put(mddev);
4722 return error;
4725 static struct kobject *md_probe(dev_t dev, int *part, void *data)
4727 md_alloc(dev, NULL);
4728 return NULL;
4731 static int add_named_array(const char *val, struct kernel_param *kp)
4733 /* val must be "md_*" where * is not all digits.
4734 * We allocate an array with a large free minor number, and
4735 * set the name to val. val must not already be an active name.
4737 int len = strlen(val);
4738 char buf[DISK_NAME_LEN];
4740 while (len && val[len-1] == '\n')
4741 len--;
4742 if (len >= DISK_NAME_LEN)
4743 return -E2BIG;
4744 strlcpy(buf, val, len+1);
4745 if (strncmp(buf, "md_", 3) != 0)
4746 return -EINVAL;
4747 return md_alloc(0, buf);
4750 static void md_safemode_timeout(unsigned long data)
4752 struct mddev *mddev = (struct mddev *) data;
4754 if (!atomic_read(&mddev->writes_pending)) {
4755 mddev->safemode = 1;
4756 if (mddev->external)
4757 sysfs_notify_dirent_safe(mddev->sysfs_state);
4759 md_wakeup_thread(mddev->thread);
4762 static int start_dirty_degraded;
4764 int md_run(struct mddev *mddev)
4766 int err;
4767 struct md_rdev *rdev;
4768 struct md_personality *pers;
4770 if (list_empty(&mddev->disks))
4771 /* cannot run an array with no devices.. */
4772 return -EINVAL;
4774 if (mddev->pers)
4775 return -EBUSY;
4776 /* Cannot run until previous stop completes properly */
4777 if (mddev->sysfs_active)
4778 return -EBUSY;
4781 * Analyze all RAID superblock(s)
4783 if (!mddev->raid_disks) {
4784 if (!mddev->persistent)
4785 return -EINVAL;
4786 analyze_sbs(mddev);
4789 if (mddev->level != LEVEL_NONE)
4790 request_module("md-level-%d", mddev->level);
4791 else if (mddev->clevel[0])
4792 request_module("md-%s", mddev->clevel);
4795 * Drop all container device buffers, from now on
4796 * the only valid external interface is through the md
4797 * device.
4799 list_for_each_entry(rdev, &mddev->disks, same_set) {
4800 if (test_bit(Faulty, &rdev->flags))
4801 continue;
4802 sync_blockdev(rdev->bdev);
4803 invalidate_bdev(rdev->bdev);
4805 /* perform some consistency tests on the device.
4806 * We don't want the data to overlap the metadata,
4807 * Internal Bitmap issues have been handled elsewhere.
4809 if (rdev->meta_bdev) {
4810 /* Nothing to check */;
4811 } else if (rdev->data_offset < rdev->sb_start) {
4812 if (mddev->dev_sectors &&
4813 rdev->data_offset + mddev->dev_sectors
4814 > rdev->sb_start) {
4815 printk("md: %s: data overlaps metadata\n",
4816 mdname(mddev));
4817 return -EINVAL;
4819 } else {
4820 if (rdev->sb_start + rdev->sb_size/512
4821 > rdev->data_offset) {
4822 printk("md: %s: metadata overlaps data\n",
4823 mdname(mddev));
4824 return -EINVAL;
4827 sysfs_notify_dirent_safe(rdev->sysfs_state);
4830 if (mddev->bio_set == NULL)
4831 mddev->bio_set = bioset_create(BIO_POOL_SIZE,
4832 sizeof(struct mddev *));
4834 spin_lock(&pers_lock);
4835 pers = find_pers(mddev->level, mddev->clevel);
4836 if (!pers || !try_module_get(pers->owner)) {
4837 spin_unlock(&pers_lock);
4838 if (mddev->level != LEVEL_NONE)
4839 printk(KERN_WARNING "md: personality for level %d is not loaded!\n",
4840 mddev->level);
4841 else
4842 printk(KERN_WARNING "md: personality for level %s is not loaded!\n",
4843 mddev->clevel);
4844 return -EINVAL;
4846 mddev->pers = pers;
4847 spin_unlock(&pers_lock);
4848 if (mddev->level != pers->level) {
4849 mddev->level = pers->level;
4850 mddev->new_level = pers->level;
4852 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
4854 if (mddev->reshape_position != MaxSector &&
4855 pers->start_reshape == NULL) {
4856 /* This personality cannot handle reshaping... */
4857 mddev->pers = NULL;
4858 module_put(pers->owner);
4859 return -EINVAL;
4862 if (pers->sync_request) {
4863 /* Warn if this is a potentially silly
4864 * configuration.
4866 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
4867 struct md_rdev *rdev2;
4868 int warned = 0;
4870 list_for_each_entry(rdev, &mddev->disks, same_set)
4871 list_for_each_entry(rdev2, &mddev->disks, same_set) {
4872 if (rdev < rdev2 &&
4873 rdev->bdev->bd_contains ==
4874 rdev2->bdev->bd_contains) {
4875 printk(KERN_WARNING
4876 "%s: WARNING: %s appears to be"
4877 " on the same physical disk as"
4878 " %s.\n",
4879 mdname(mddev),
4880 bdevname(rdev->bdev,b),
4881 bdevname(rdev2->bdev,b2));
4882 warned = 1;
4886 if (warned)
4887 printk(KERN_WARNING
4888 "True protection against single-disk"
4889 " failure might be compromised.\n");
4892 mddev->recovery = 0;
4893 /* may be over-ridden by personality */
4894 mddev->resync_max_sectors = mddev->dev_sectors;
4896 mddev->ok_start_degraded = start_dirty_degraded;
4898 if (start_readonly && mddev->ro == 0)
4899 mddev->ro = 2; /* read-only, but switch on first write */
4901 err = mddev->pers->run(mddev);
4902 if (err)
4903 printk(KERN_ERR "md: pers->run() failed ...\n");
4904 else if (mddev->pers->size(mddev, 0, 0) < mddev->array_sectors) {
4905 WARN_ONCE(!mddev->external_size, "%s: default size too small,"
4906 " but 'external_size' not in effect?\n", __func__);
4907 printk(KERN_ERR
4908 "md: invalid array_size %llu > default size %llu\n",
4909 (unsigned long long)mddev->array_sectors / 2,
4910 (unsigned long long)mddev->pers->size(mddev, 0, 0) / 2);
4911 err = -EINVAL;
4912 mddev->pers->stop(mddev);
4914 if (err == 0 && mddev->pers->sync_request) {
4915 err = bitmap_create(mddev);
4916 if (err) {
4917 printk(KERN_ERR "%s: failed to create bitmap (%d)\n",
4918 mdname(mddev), err);
4919 mddev->pers->stop(mddev);
4922 if (err) {
4923 module_put(mddev->pers->owner);
4924 mddev->pers = NULL;
4925 bitmap_destroy(mddev);
4926 return err;
4928 if (mddev->pers->sync_request) {
4929 if (mddev->kobj.sd &&
4930 sysfs_create_group(&mddev->kobj, &md_redundancy_group))
4931 printk(KERN_WARNING
4932 "md: cannot register extra attributes for %s\n",
4933 mdname(mddev));
4934 mddev->sysfs_action = sysfs_get_dirent_safe(mddev->kobj.sd, "sync_action");
4935 } else if (mddev->ro == 2) /* auto-readonly not meaningful */
4936 mddev->ro = 0;
4938 atomic_set(&mddev->writes_pending,0);
4939 atomic_set(&mddev->max_corr_read_errors,
4940 MD_DEFAULT_MAX_CORRECTED_READ_ERRORS);
4941 mddev->safemode = 0;
4942 mddev->safemode_timer.function = md_safemode_timeout;
4943 mddev->safemode_timer.data = (unsigned long) mddev;
4944 mddev->safemode_delay = (200 * HZ)/1000 +1; /* 200 msec delay */
4945 mddev->in_sync = 1;
4946 smp_wmb();
4947 mddev->ready = 1;
4948 list_for_each_entry(rdev, &mddev->disks, same_set)
4949 if (rdev->raid_disk >= 0)
4950 if (sysfs_link_rdev(mddev, rdev))
4951 /* failure here is OK */;
4953 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4955 if (mddev->flags)
4956 md_update_sb(mddev, 0);
4958 md_new_event(mddev);
4959 sysfs_notify_dirent_safe(mddev->sysfs_state);
4960 sysfs_notify_dirent_safe(mddev->sysfs_action);
4961 sysfs_notify(&mddev->kobj, NULL, "degraded");
4962 return 0;
4964 EXPORT_SYMBOL_GPL(md_run);
4966 static int do_md_run(struct mddev *mddev)
4968 int err;
4970 err = md_run(mddev);
4971 if (err)
4972 goto out;
4973 err = bitmap_load(mddev);
4974 if (err) {
4975 bitmap_destroy(mddev);
4976 goto out;
4979 md_wakeup_thread(mddev->thread);
4980 md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
4982 set_capacity(mddev->gendisk, mddev->array_sectors);
4983 revalidate_disk(mddev->gendisk);
4984 mddev->changed = 1;
4985 kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
4986 out:
4987 return err;
4990 static int restart_array(struct mddev *mddev)
4992 struct gendisk *disk = mddev->gendisk;
4994 /* Complain if it has no devices */
4995 if (list_empty(&mddev->disks))
4996 return -ENXIO;
4997 if (!mddev->pers)
4998 return -EINVAL;
4999 if (!mddev->ro)
5000 return -EBUSY;
5001 mddev->safemode = 0;
5002 mddev->ro = 0;
5003 set_disk_ro(disk, 0);
5004 printk(KERN_INFO "md: %s switched to read-write mode.\n",
5005 mdname(mddev));
5006 /* Kick recovery or resync if necessary */
5007 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5008 md_wakeup_thread(mddev->thread);
5009 md_wakeup_thread(mddev->sync_thread);
5010 sysfs_notify_dirent_safe(mddev->sysfs_state);
5011 return 0;
5014 /* similar to deny_write_access, but accounts for our holding a reference
5015 * to the file ourselves */
5016 static int deny_bitmap_write_access(struct file * file)
5018 struct inode *inode = file->f_mapping->host;
5020 spin_lock(&inode->i_lock);
5021 if (atomic_read(&inode->i_writecount) > 1) {
5022 spin_unlock(&inode->i_lock);
5023 return -ETXTBSY;
5025 atomic_set(&inode->i_writecount, -1);
5026 spin_unlock(&inode->i_lock);
5028 return 0;
5031 void restore_bitmap_write_access(struct file *file)
5033 struct inode *inode = file->f_mapping->host;
5035 spin_lock(&inode->i_lock);
5036 atomic_set(&inode->i_writecount, 1);
5037 spin_unlock(&inode->i_lock);
5040 static void md_clean(struct mddev *mddev)
5042 mddev->array_sectors = 0;
5043 mddev->external_size = 0;
5044 mddev->dev_sectors = 0;
5045 mddev->raid_disks = 0;
5046 mddev->recovery_cp = 0;
5047 mddev->resync_min = 0;
5048 mddev->resync_max = MaxSector;
5049 mddev->reshape_position = MaxSector;
5050 mddev->external = 0;
5051 mddev->persistent = 0;
5052 mddev->level = LEVEL_NONE;
5053 mddev->clevel[0] = 0;
5054 mddev->flags = 0;
5055 mddev->ro = 0;
5056 mddev->metadata_type[0] = 0;
5057 mddev->chunk_sectors = 0;
5058 mddev->ctime = mddev->utime = 0;
5059 mddev->layout = 0;
5060 mddev->max_disks = 0;
5061 mddev->events = 0;
5062 mddev->can_decrease_events = 0;
5063 mddev->delta_disks = 0;
5064 mddev->new_level = LEVEL_NONE;
5065 mddev->new_layout = 0;
5066 mddev->new_chunk_sectors = 0;
5067 mddev->curr_resync = 0;
5068 mddev->resync_mismatches = 0;
5069 mddev->suspend_lo = mddev->suspend_hi = 0;
5070 mddev->sync_speed_min = mddev->sync_speed_max = 0;
5071 mddev->recovery = 0;
5072 mddev->in_sync = 0;
5073 mddev->changed = 0;
5074 mddev->degraded = 0;
5075 mddev->safemode = 0;
5076 mddev->bitmap_info.offset = 0;
5077 mddev->bitmap_info.default_offset = 0;
5078 mddev->bitmap_info.chunksize = 0;
5079 mddev->bitmap_info.daemon_sleep = 0;
5080 mddev->bitmap_info.max_write_behind = 0;
5083 static void __md_stop_writes(struct mddev *mddev)
5085 if (mddev->sync_thread) {
5086 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5087 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5088 reap_sync_thread(mddev);
5091 del_timer_sync(&mddev->safemode_timer);
5093 bitmap_flush(mddev);
5094 md_super_wait(mddev);
5096 if (!mddev->in_sync || mddev->flags) {
5097 /* mark array as shutdown cleanly */
5098 mddev->in_sync = 1;
5099 md_update_sb(mddev, 1);
5103 void md_stop_writes(struct mddev *mddev)
5105 mddev_lock(mddev);
5106 __md_stop_writes(mddev);
5107 mddev_unlock(mddev);
5109 EXPORT_SYMBOL_GPL(md_stop_writes);
5111 void md_stop(struct mddev *mddev)
5113 mddev->ready = 0;
5114 mddev->pers->stop(mddev);
5115 if (mddev->pers->sync_request && mddev->to_remove == NULL)
5116 mddev->to_remove = &md_redundancy_group;
5117 module_put(mddev->pers->owner);
5118 mddev->pers = NULL;
5119 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5121 EXPORT_SYMBOL_GPL(md_stop);
5123 static int md_set_readonly(struct mddev *mddev, int is_open)
5125 int err = 0;
5126 mutex_lock(&mddev->open_mutex);
5127 if (atomic_read(&mddev->openers) > is_open) {
5128 printk("md: %s still in use.\n",mdname(mddev));
5129 err = -EBUSY;
5130 goto out;
5132 if (mddev->pers) {
5133 __md_stop_writes(mddev);
5135 err = -ENXIO;
5136 if (mddev->ro==1)
5137 goto out;
5138 mddev->ro = 1;
5139 set_disk_ro(mddev->gendisk, 1);
5140 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5141 sysfs_notify_dirent_safe(mddev->sysfs_state);
5142 err = 0;
5144 out:
5145 mutex_unlock(&mddev->open_mutex);
5146 return err;
5149 /* mode:
5150 * 0 - completely stop and dis-assemble array
5151 * 2 - stop but do not disassemble array
5153 static int do_md_stop(struct mddev * mddev, int mode, int is_open)
5155 struct gendisk *disk = mddev->gendisk;
5156 struct md_rdev *rdev;
5158 mutex_lock(&mddev->open_mutex);
5159 if (atomic_read(&mddev->openers) > is_open ||
5160 mddev->sysfs_active) {
5161 printk("md: %s still in use.\n",mdname(mddev));
5162 mutex_unlock(&mddev->open_mutex);
5163 return -EBUSY;
5166 if (mddev->pers) {
5167 if (mddev->ro)
5168 set_disk_ro(disk, 0);
5170 __md_stop_writes(mddev);
5171 md_stop(mddev);
5172 mddev->queue->merge_bvec_fn = NULL;
5173 mddev->queue->backing_dev_info.congested_fn = NULL;
5175 /* tell userspace to handle 'inactive' */
5176 sysfs_notify_dirent_safe(mddev->sysfs_state);
5178 list_for_each_entry(rdev, &mddev->disks, same_set)
5179 if (rdev->raid_disk >= 0)
5180 sysfs_unlink_rdev(mddev, rdev);
5182 set_capacity(disk, 0);
5183 mutex_unlock(&mddev->open_mutex);
5184 mddev->changed = 1;
5185 revalidate_disk(disk);
5187 if (mddev->ro)
5188 mddev->ro = 0;
5189 } else
5190 mutex_unlock(&mddev->open_mutex);
5192 * Free resources if final stop
5194 if (mode == 0) {
5195 printk(KERN_INFO "md: %s stopped.\n", mdname(mddev));
5197 bitmap_destroy(mddev);
5198 if (mddev->bitmap_info.file) {
5199 restore_bitmap_write_access(mddev->bitmap_info.file);
5200 fput(mddev->bitmap_info.file);
5201 mddev->bitmap_info.file = NULL;
5203 mddev->bitmap_info.offset = 0;
5205 export_array(mddev);
5207 md_clean(mddev);
5208 kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
5209 if (mddev->hold_active == UNTIL_STOP)
5210 mddev->hold_active = 0;
5212 blk_integrity_unregister(disk);
5213 md_new_event(mddev);
5214 sysfs_notify_dirent_safe(mddev->sysfs_state);
5215 return 0;
5218 #ifndef MODULE
5219 static void autorun_array(struct mddev *mddev)
5221 struct md_rdev *rdev;
5222 int err;
5224 if (list_empty(&mddev->disks))
5225 return;
5227 printk(KERN_INFO "md: running: ");
5229 list_for_each_entry(rdev, &mddev->disks, same_set) {
5230 char b[BDEVNAME_SIZE];
5231 printk("<%s>", bdevname(rdev->bdev,b));
5233 printk("\n");
5235 err = do_md_run(mddev);
5236 if (err) {
5237 printk(KERN_WARNING "md: do_md_run() returned %d\n", err);
5238 do_md_stop(mddev, 0, 0);
5243 * lets try to run arrays based on all disks that have arrived
5244 * until now. (those are in pending_raid_disks)
5246 * the method: pick the first pending disk, collect all disks with
5247 * the same UUID, remove all from the pending list and put them into
5248 * the 'same_array' list. Then order this list based on superblock
5249 * update time (freshest comes first), kick out 'old' disks and
5250 * compare superblocks. If everything's fine then run it.
5252 * If "unit" is allocated, then bump its reference count
5254 static void autorun_devices(int part)
5256 struct md_rdev *rdev0, *rdev, *tmp;
5257 struct mddev *mddev;
5258 char b[BDEVNAME_SIZE];
5260 printk(KERN_INFO "md: autorun ...\n");
5261 while (!list_empty(&pending_raid_disks)) {
5262 int unit;
5263 dev_t dev;
5264 LIST_HEAD(candidates);
5265 rdev0 = list_entry(pending_raid_disks.next,
5266 struct md_rdev, same_set);
5268 printk(KERN_INFO "md: considering %s ...\n",
5269 bdevname(rdev0->bdev,b));
5270 INIT_LIST_HEAD(&candidates);
5271 rdev_for_each_list(rdev, tmp, &pending_raid_disks)
5272 if (super_90_load(rdev, rdev0, 0) >= 0) {
5273 printk(KERN_INFO "md: adding %s ...\n",
5274 bdevname(rdev->bdev,b));
5275 list_move(&rdev->same_set, &candidates);
5278 * now we have a set of devices, with all of them having
5279 * mostly sane superblocks. It's time to allocate the
5280 * mddev.
5282 if (part) {
5283 dev = MKDEV(mdp_major,
5284 rdev0->preferred_minor << MdpMinorShift);
5285 unit = MINOR(dev) >> MdpMinorShift;
5286 } else {
5287 dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
5288 unit = MINOR(dev);
5290 if (rdev0->preferred_minor != unit) {
5291 printk(KERN_INFO "md: unit number in %s is bad: %d\n",
5292 bdevname(rdev0->bdev, b), rdev0->preferred_minor);
5293 break;
5296 md_probe(dev, NULL, NULL);
5297 mddev = mddev_find(dev);
5298 if (!mddev || !mddev->gendisk) {
5299 if (mddev)
5300 mddev_put(mddev);
5301 printk(KERN_ERR
5302 "md: cannot allocate memory for md drive.\n");
5303 break;
5305 if (mddev_lock(mddev))
5306 printk(KERN_WARNING "md: %s locked, cannot run\n",
5307 mdname(mddev));
5308 else if (mddev->raid_disks || mddev->major_version
5309 || !list_empty(&mddev->disks)) {
5310 printk(KERN_WARNING
5311 "md: %s already running, cannot run %s\n",
5312 mdname(mddev), bdevname(rdev0->bdev,b));
5313 mddev_unlock(mddev);
5314 } else {
5315 printk(KERN_INFO "md: created %s\n", mdname(mddev));
5316 mddev->persistent = 1;
5317 rdev_for_each_list(rdev, tmp, &candidates) {
5318 list_del_init(&rdev->same_set);
5319 if (bind_rdev_to_array(rdev, mddev))
5320 export_rdev(rdev);
5322 autorun_array(mddev);
5323 mddev_unlock(mddev);
5325 /* on success, candidates will be empty, on error
5326 * it won't...
5328 rdev_for_each_list(rdev, tmp, &candidates) {
5329 list_del_init(&rdev->same_set);
5330 export_rdev(rdev);
5332 mddev_put(mddev);
5334 printk(KERN_INFO "md: ... autorun DONE.\n");
5336 #endif /* !MODULE */
5338 static int get_version(void __user * arg)
5340 mdu_version_t ver;
5342 ver.major = MD_MAJOR_VERSION;
5343 ver.minor = MD_MINOR_VERSION;
5344 ver.patchlevel = MD_PATCHLEVEL_VERSION;
5346 if (copy_to_user(arg, &ver, sizeof(ver)))
5347 return -EFAULT;
5349 return 0;
5352 static int get_array_info(struct mddev * mddev, void __user * arg)
5354 mdu_array_info_t info;
5355 int nr,working,insync,failed,spare;
5356 struct md_rdev *rdev;
5358 nr=working=insync=failed=spare=0;
5359 list_for_each_entry(rdev, &mddev->disks, same_set) {
5360 nr++;
5361 if (test_bit(Faulty, &rdev->flags))
5362 failed++;
5363 else {
5364 working++;
5365 if (test_bit(In_sync, &rdev->flags))
5366 insync++;
5367 else
5368 spare++;
5372 info.major_version = mddev->major_version;
5373 info.minor_version = mddev->minor_version;
5374 info.patch_version = MD_PATCHLEVEL_VERSION;
5375 info.ctime = mddev->ctime;
5376 info.level = mddev->level;
5377 info.size = mddev->dev_sectors / 2;
5378 if (info.size != mddev->dev_sectors / 2) /* overflow */
5379 info.size = -1;
5380 info.nr_disks = nr;
5381 info.raid_disks = mddev->raid_disks;
5382 info.md_minor = mddev->md_minor;
5383 info.not_persistent= !mddev->persistent;
5385 info.utime = mddev->utime;
5386 info.state = 0;
5387 if (mddev->in_sync)
5388 info.state = (1<<MD_SB_CLEAN);
5389 if (mddev->bitmap && mddev->bitmap_info.offset)
5390 info.state = (1<<MD_SB_BITMAP_PRESENT);
5391 info.active_disks = insync;
5392 info.working_disks = working;
5393 info.failed_disks = failed;
5394 info.spare_disks = spare;
5396 info.layout = mddev->layout;
5397 info.chunk_size = mddev->chunk_sectors << 9;
5399 if (copy_to_user(arg, &info, sizeof(info)))
5400 return -EFAULT;
5402 return 0;
5405 static int get_bitmap_file(struct mddev * mddev, void __user * arg)
5407 mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */
5408 char *ptr, *buf = NULL;
5409 int err = -ENOMEM;
5411 if (md_allow_write(mddev))
5412 file = kmalloc(sizeof(*file), GFP_NOIO);
5413 else
5414 file = kmalloc(sizeof(*file), GFP_KERNEL);
5416 if (!file)
5417 goto out;
5419 /* bitmap disabled, zero the first byte and copy out */
5420 if (!mddev->bitmap || !mddev->bitmap->file) {
5421 file->pathname[0] = '\0';
5422 goto copy_out;
5425 buf = kmalloc(sizeof(file->pathname), GFP_KERNEL);
5426 if (!buf)
5427 goto out;
5429 ptr = d_path(&mddev->bitmap->file->f_path, buf, sizeof(file->pathname));
5430 if (IS_ERR(ptr))
5431 goto out;
5433 strcpy(file->pathname, ptr);
5435 copy_out:
5436 err = 0;
5437 if (copy_to_user(arg, file, sizeof(*file)))
5438 err = -EFAULT;
5439 out:
5440 kfree(buf);
5441 kfree(file);
5442 return err;
5445 static int get_disk_info(struct mddev * mddev, void __user * arg)
5447 mdu_disk_info_t info;
5448 struct md_rdev *rdev;
5450 if (copy_from_user(&info, arg, sizeof(info)))
5451 return -EFAULT;
5453 rdev = find_rdev_nr(mddev, info.number);
5454 if (rdev) {
5455 info.major = MAJOR(rdev->bdev->bd_dev);
5456 info.minor = MINOR(rdev->bdev->bd_dev);
5457 info.raid_disk = rdev->raid_disk;
5458 info.state = 0;
5459 if (test_bit(Faulty, &rdev->flags))
5460 info.state |= (1<<MD_DISK_FAULTY);
5461 else if (test_bit(In_sync, &rdev->flags)) {
5462 info.state |= (1<<MD_DISK_ACTIVE);
5463 info.state |= (1<<MD_DISK_SYNC);
5465 if (test_bit(WriteMostly, &rdev->flags))
5466 info.state |= (1<<MD_DISK_WRITEMOSTLY);
5467 } else {
5468 info.major = info.minor = 0;
5469 info.raid_disk = -1;
5470 info.state = (1<<MD_DISK_REMOVED);
5473 if (copy_to_user(arg, &info, sizeof(info)))
5474 return -EFAULT;
5476 return 0;
5479 static int add_new_disk(struct mddev * mddev, mdu_disk_info_t *info)
5481 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
5482 struct md_rdev *rdev;
5483 dev_t dev = MKDEV(info->major,info->minor);
5485 if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
5486 return -EOVERFLOW;
5488 if (!mddev->raid_disks) {
5489 int err;
5490 /* expecting a device which has a superblock */
5491 rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
5492 if (IS_ERR(rdev)) {
5493 printk(KERN_WARNING
5494 "md: md_import_device returned %ld\n",
5495 PTR_ERR(rdev));
5496 return PTR_ERR(rdev);
5498 if (!list_empty(&mddev->disks)) {
5499 struct md_rdev *rdev0
5500 = list_entry(mddev->disks.next,
5501 struct md_rdev, same_set);
5502 err = super_types[mddev->major_version]
5503 .load_super(rdev, rdev0, mddev->minor_version);
5504 if (err < 0) {
5505 printk(KERN_WARNING
5506 "md: %s has different UUID to %s\n",
5507 bdevname(rdev->bdev,b),
5508 bdevname(rdev0->bdev,b2));
5509 export_rdev(rdev);
5510 return -EINVAL;
5513 err = bind_rdev_to_array(rdev, mddev);
5514 if (err)
5515 export_rdev(rdev);
5516 return err;
5520 * add_new_disk can be used once the array is assembled
5521 * to add "hot spares". They must already have a superblock
5522 * written
5524 if (mddev->pers) {
5525 int err;
5526 if (!mddev->pers->hot_add_disk) {
5527 printk(KERN_WARNING
5528 "%s: personality does not support diskops!\n",
5529 mdname(mddev));
5530 return -EINVAL;
5532 if (mddev->persistent)
5533 rdev = md_import_device(dev, mddev->major_version,
5534 mddev->minor_version);
5535 else
5536 rdev = md_import_device(dev, -1, -1);
5537 if (IS_ERR(rdev)) {
5538 printk(KERN_WARNING
5539 "md: md_import_device returned %ld\n",
5540 PTR_ERR(rdev));
5541 return PTR_ERR(rdev);
5543 /* set saved_raid_disk if appropriate */
5544 if (!mddev->persistent) {
5545 if (info->state & (1<<MD_DISK_SYNC) &&
5546 info->raid_disk < mddev->raid_disks) {
5547 rdev->raid_disk = info->raid_disk;
5548 set_bit(In_sync, &rdev->flags);
5549 } else
5550 rdev->raid_disk = -1;
5551 } else
5552 super_types[mddev->major_version].
5553 validate_super(mddev, rdev);
5554 if ((info->state & (1<<MD_DISK_SYNC)) &&
5555 (!test_bit(In_sync, &rdev->flags) ||
5556 rdev->raid_disk != info->raid_disk)) {
5557 /* This was a hot-add request, but events doesn't
5558 * match, so reject it.
5560 export_rdev(rdev);
5561 return -EINVAL;
5564 if (test_bit(In_sync, &rdev->flags))
5565 rdev->saved_raid_disk = rdev->raid_disk;
5566 else
5567 rdev->saved_raid_disk = -1;
5569 clear_bit(In_sync, &rdev->flags); /* just to be sure */
5570 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
5571 set_bit(WriteMostly, &rdev->flags);
5572 else
5573 clear_bit(WriteMostly, &rdev->flags);
5575 rdev->raid_disk = -1;
5576 err = bind_rdev_to_array(rdev, mddev);
5577 if (!err && !mddev->pers->hot_remove_disk) {
5578 /* If there is hot_add_disk but no hot_remove_disk
5579 * then added disks for geometry changes,
5580 * and should be added immediately.
5582 super_types[mddev->major_version].
5583 validate_super(mddev, rdev);
5584 err = mddev->pers->hot_add_disk(mddev, rdev);
5585 if (err)
5586 unbind_rdev_from_array(rdev);
5588 if (err)
5589 export_rdev(rdev);
5590 else
5591 sysfs_notify_dirent_safe(rdev->sysfs_state);
5593 md_update_sb(mddev, 1);
5594 if (mddev->degraded)
5595 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
5596 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5597 if (!err)
5598 md_new_event(mddev);
5599 md_wakeup_thread(mddev->thread);
5600 return err;
5603 /* otherwise, add_new_disk is only allowed
5604 * for major_version==0 superblocks
5606 if (mddev->major_version != 0) {
5607 printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n",
5608 mdname(mddev));
5609 return -EINVAL;
5612 if (!(info->state & (1<<MD_DISK_FAULTY))) {
5613 int err;
5614 rdev = md_import_device(dev, -1, 0);
5615 if (IS_ERR(rdev)) {
5616 printk(KERN_WARNING
5617 "md: error, md_import_device() returned %ld\n",
5618 PTR_ERR(rdev));
5619 return PTR_ERR(rdev);
5621 rdev->desc_nr = info->number;
5622 if (info->raid_disk < mddev->raid_disks)
5623 rdev->raid_disk = info->raid_disk;
5624 else
5625 rdev->raid_disk = -1;
5627 if (rdev->raid_disk < mddev->raid_disks)
5628 if (info->state & (1<<MD_DISK_SYNC))
5629 set_bit(In_sync, &rdev->flags);
5631 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
5632 set_bit(WriteMostly, &rdev->flags);
5634 if (!mddev->persistent) {
5635 printk(KERN_INFO "md: nonpersistent superblock ...\n");
5636 rdev->sb_start = i_size_read(rdev->bdev->bd_inode) / 512;
5637 } else
5638 rdev->sb_start = calc_dev_sboffset(rdev);
5639 rdev->sectors = rdev->sb_start;
5641 err = bind_rdev_to_array(rdev, mddev);
5642 if (err) {
5643 export_rdev(rdev);
5644 return err;
5648 return 0;
5651 static int hot_remove_disk(struct mddev * mddev, dev_t dev)
5653 char b[BDEVNAME_SIZE];
5654 struct md_rdev *rdev;
5656 rdev = find_rdev(mddev, dev);
5657 if (!rdev)
5658 return -ENXIO;
5660 if (rdev->raid_disk >= 0)
5661 goto busy;
5663 kick_rdev_from_array(rdev);
5664 md_update_sb(mddev, 1);
5665 md_new_event(mddev);
5667 return 0;
5668 busy:
5669 printk(KERN_WARNING "md: cannot remove active disk %s from %s ...\n",
5670 bdevname(rdev->bdev,b), mdname(mddev));
5671 return -EBUSY;
5674 static int hot_add_disk(struct mddev * mddev, dev_t dev)
5676 char b[BDEVNAME_SIZE];
5677 int err;
5678 struct md_rdev *rdev;
5680 if (!mddev->pers)
5681 return -ENODEV;
5683 if (mddev->major_version != 0) {
5684 printk(KERN_WARNING "%s: HOT_ADD may only be used with"
5685 " version-0 superblocks.\n",
5686 mdname(mddev));
5687 return -EINVAL;
5689 if (!mddev->pers->hot_add_disk) {
5690 printk(KERN_WARNING
5691 "%s: personality does not support diskops!\n",
5692 mdname(mddev));
5693 return -EINVAL;
5696 rdev = md_import_device(dev, -1, 0);
5697 if (IS_ERR(rdev)) {
5698 printk(KERN_WARNING
5699 "md: error, md_import_device() returned %ld\n",
5700 PTR_ERR(rdev));
5701 return -EINVAL;
5704 if (mddev->persistent)
5705 rdev->sb_start = calc_dev_sboffset(rdev);
5706 else
5707 rdev->sb_start = i_size_read(rdev->bdev->bd_inode) / 512;
5709 rdev->sectors = rdev->sb_start;
5711 if (test_bit(Faulty, &rdev->flags)) {
5712 printk(KERN_WARNING
5713 "md: can not hot-add faulty %s disk to %s!\n",
5714 bdevname(rdev->bdev,b), mdname(mddev));
5715 err = -EINVAL;
5716 goto abort_export;
5718 clear_bit(In_sync, &rdev->flags);
5719 rdev->desc_nr = -1;
5720 rdev->saved_raid_disk = -1;
5721 err = bind_rdev_to_array(rdev, mddev);
5722 if (err)
5723 goto abort_export;
5726 * The rest should better be atomic, we can have disk failures
5727 * noticed in interrupt contexts ...
5730 rdev->raid_disk = -1;
5732 md_update_sb(mddev, 1);
5735 * Kick recovery, maybe this spare has to be added to the
5736 * array immediately.
5738 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5739 md_wakeup_thread(mddev->thread);
5740 md_new_event(mddev);
5741 return 0;
5743 abort_export:
5744 export_rdev(rdev);
5745 return err;
5748 static int set_bitmap_file(struct mddev *mddev, int fd)
5750 int err;
5752 if (mddev->pers) {
5753 if (!mddev->pers->quiesce)
5754 return -EBUSY;
5755 if (mddev->recovery || mddev->sync_thread)
5756 return -EBUSY;
5757 /* we should be able to change the bitmap.. */
5761 if (fd >= 0) {
5762 if (mddev->bitmap)
5763 return -EEXIST; /* cannot add when bitmap is present */
5764 mddev->bitmap_info.file = fget(fd);
5766 if (mddev->bitmap_info.file == NULL) {
5767 printk(KERN_ERR "%s: error: failed to get bitmap file\n",
5768 mdname(mddev));
5769 return -EBADF;
5772 err = deny_bitmap_write_access(mddev->bitmap_info.file);
5773 if (err) {
5774 printk(KERN_ERR "%s: error: bitmap file is already in use\n",
5775 mdname(mddev));
5776 fput(mddev->bitmap_info.file);
5777 mddev->bitmap_info.file = NULL;
5778 return err;
5780 mddev->bitmap_info.offset = 0; /* file overrides offset */
5781 } else if (mddev->bitmap == NULL)
5782 return -ENOENT; /* cannot remove what isn't there */
5783 err = 0;
5784 if (mddev->pers) {
5785 mddev->pers->quiesce(mddev, 1);
5786 if (fd >= 0) {
5787 err = bitmap_create(mddev);
5788 if (!err)
5789 err = bitmap_load(mddev);
5791 if (fd < 0 || err) {
5792 bitmap_destroy(mddev);
5793 fd = -1; /* make sure to put the file */
5795 mddev->pers->quiesce(mddev, 0);
5797 if (fd < 0) {
5798 if (mddev->bitmap_info.file) {
5799 restore_bitmap_write_access(mddev->bitmap_info.file);
5800 fput(mddev->bitmap_info.file);
5802 mddev->bitmap_info.file = NULL;
5805 return err;
5809 * set_array_info is used two different ways
5810 * The original usage is when creating a new array.
5811 * In this usage, raid_disks is > 0 and it together with
5812 * level, size, not_persistent,layout,chunksize determine the
5813 * shape of the array.
5814 * This will always create an array with a type-0.90.0 superblock.
5815 * The newer usage is when assembling an array.
5816 * In this case raid_disks will be 0, and the major_version field is
5817 * use to determine which style super-blocks are to be found on the devices.
5818 * The minor and patch _version numbers are also kept incase the
5819 * super_block handler wishes to interpret them.
5821 static int set_array_info(struct mddev * mddev, mdu_array_info_t *info)
5824 if (info->raid_disks == 0) {
5825 /* just setting version number for superblock loading */
5826 if (info->major_version < 0 ||
5827 info->major_version >= ARRAY_SIZE(super_types) ||
5828 super_types[info->major_version].name == NULL) {
5829 /* maybe try to auto-load a module? */
5830 printk(KERN_INFO
5831 "md: superblock version %d not known\n",
5832 info->major_version);
5833 return -EINVAL;
5835 mddev->major_version = info->major_version;
5836 mddev->minor_version = info->minor_version;
5837 mddev->patch_version = info->patch_version;
5838 mddev->persistent = !info->not_persistent;
5839 /* ensure mddev_put doesn't delete this now that there
5840 * is some minimal configuration.
5842 mddev->ctime = get_seconds();
5843 return 0;
5845 mddev->major_version = MD_MAJOR_VERSION;
5846 mddev->minor_version = MD_MINOR_VERSION;
5847 mddev->patch_version = MD_PATCHLEVEL_VERSION;
5848 mddev->ctime = get_seconds();
5850 mddev->level = info->level;
5851 mddev->clevel[0] = 0;
5852 mddev->dev_sectors = 2 * (sector_t)info->size;
5853 mddev->raid_disks = info->raid_disks;
5854 /* don't set md_minor, it is determined by which /dev/md* was
5855 * openned
5857 if (info->state & (1<<MD_SB_CLEAN))
5858 mddev->recovery_cp = MaxSector;
5859 else
5860 mddev->recovery_cp = 0;
5861 mddev->persistent = ! info->not_persistent;
5862 mddev->external = 0;
5864 mddev->layout = info->layout;
5865 mddev->chunk_sectors = info->chunk_size >> 9;
5867 mddev->max_disks = MD_SB_DISKS;
5869 if (mddev->persistent)
5870 mddev->flags = 0;
5871 set_bit(MD_CHANGE_DEVS, &mddev->flags);
5873 mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9;
5874 mddev->bitmap_info.offset = 0;
5876 mddev->reshape_position = MaxSector;
5879 * Generate a 128 bit UUID
5881 get_random_bytes(mddev->uuid, 16);
5883 mddev->new_level = mddev->level;
5884 mddev->new_chunk_sectors = mddev->chunk_sectors;
5885 mddev->new_layout = mddev->layout;
5886 mddev->delta_disks = 0;
5888 return 0;
5891 void md_set_array_sectors(struct mddev *mddev, sector_t array_sectors)
5893 WARN(!mddev_is_locked(mddev), "%s: unlocked mddev!\n", __func__);
5895 if (mddev->external_size)
5896 return;
5898 mddev->array_sectors = array_sectors;
5900 EXPORT_SYMBOL(md_set_array_sectors);
5902 static int update_size(struct mddev *mddev, sector_t num_sectors)
5904 struct md_rdev *rdev;
5905 int rv;
5906 int fit = (num_sectors == 0);
5908 if (mddev->pers->resize == NULL)
5909 return -EINVAL;
5910 /* The "num_sectors" is the number of sectors of each device that
5911 * is used. This can only make sense for arrays with redundancy.
5912 * linear and raid0 always use whatever space is available. We can only
5913 * consider changing this number if no resync or reconstruction is
5914 * happening, and if the new size is acceptable. It must fit before the
5915 * sb_start or, if that is <data_offset, it must fit before the size
5916 * of each device. If num_sectors is zero, we find the largest size
5917 * that fits.
5919 if (mddev->sync_thread)
5920 return -EBUSY;
5921 if (mddev->bitmap)
5922 /* Sorry, cannot grow a bitmap yet, just remove it,
5923 * grow, and re-add.
5925 return -EBUSY;
5926 list_for_each_entry(rdev, &mddev->disks, same_set) {
5927 sector_t avail = rdev->sectors;
5929 if (fit && (num_sectors == 0 || num_sectors > avail))
5930 num_sectors = avail;
5931 if (avail < num_sectors)
5932 return -ENOSPC;
5934 rv = mddev->pers->resize(mddev, num_sectors);
5935 if (!rv)
5936 revalidate_disk(mddev->gendisk);
5937 return rv;
5940 static int update_raid_disks(struct mddev *mddev, int raid_disks)
5942 int rv;
5943 /* change the number of raid disks */
5944 if (mddev->pers->check_reshape == NULL)
5945 return -EINVAL;
5946 if (raid_disks <= 0 ||
5947 (mddev->max_disks && raid_disks >= mddev->max_disks))
5948 return -EINVAL;
5949 if (mddev->sync_thread || mddev->reshape_position != MaxSector)
5950 return -EBUSY;
5951 mddev->delta_disks = raid_disks - mddev->raid_disks;
5953 rv = mddev->pers->check_reshape(mddev);
5954 if (rv < 0)
5955 mddev->delta_disks = 0;
5956 return rv;
5961 * update_array_info is used to change the configuration of an
5962 * on-line array.
5963 * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
5964 * fields in the info are checked against the array.
5965 * Any differences that cannot be handled will cause an error.
5966 * Normally, only one change can be managed at a time.
5968 static int update_array_info(struct mddev *mddev, mdu_array_info_t *info)
5970 int rv = 0;
5971 int cnt = 0;
5972 int state = 0;
5974 /* calculate expected state,ignoring low bits */
5975 if (mddev->bitmap && mddev->bitmap_info.offset)
5976 state |= (1 << MD_SB_BITMAP_PRESENT);
5978 if (mddev->major_version != info->major_version ||
5979 mddev->minor_version != info->minor_version ||
5980 /* mddev->patch_version != info->patch_version || */
5981 mddev->ctime != info->ctime ||
5982 mddev->level != info->level ||
5983 /* mddev->layout != info->layout || */
5984 !mddev->persistent != info->not_persistent||
5985 mddev->chunk_sectors != info->chunk_size >> 9 ||
5986 /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
5987 ((state^info->state) & 0xfffffe00)
5989 return -EINVAL;
5990 /* Check there is only one change */
5991 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
5992 cnt++;
5993 if (mddev->raid_disks != info->raid_disks)
5994 cnt++;
5995 if (mddev->layout != info->layout)
5996 cnt++;
5997 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT))
5998 cnt++;
5999 if (cnt == 0)
6000 return 0;
6001 if (cnt > 1)
6002 return -EINVAL;
6004 if (mddev->layout != info->layout) {
6005 /* Change layout
6006 * we don't need to do anything at the md level, the
6007 * personality will take care of it all.
6009 if (mddev->pers->check_reshape == NULL)
6010 return -EINVAL;
6011 else {
6012 mddev->new_layout = info->layout;
6013 rv = mddev->pers->check_reshape(mddev);
6014 if (rv)
6015 mddev->new_layout = mddev->layout;
6016 return rv;
6019 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
6020 rv = update_size(mddev, (sector_t)info->size * 2);
6022 if (mddev->raid_disks != info->raid_disks)
6023 rv = update_raid_disks(mddev, info->raid_disks);
6025 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) {
6026 if (mddev->pers->quiesce == NULL)
6027 return -EINVAL;
6028 if (mddev->recovery || mddev->sync_thread)
6029 return -EBUSY;
6030 if (info->state & (1<<MD_SB_BITMAP_PRESENT)) {
6031 /* add the bitmap */
6032 if (mddev->bitmap)
6033 return -EEXIST;
6034 if (mddev->bitmap_info.default_offset == 0)
6035 return -EINVAL;
6036 mddev->bitmap_info.offset =
6037 mddev->bitmap_info.default_offset;
6038 mddev->pers->quiesce(mddev, 1);
6039 rv = bitmap_create(mddev);
6040 if (!rv)
6041 rv = bitmap_load(mddev);
6042 if (rv)
6043 bitmap_destroy(mddev);
6044 mddev->pers->quiesce(mddev, 0);
6045 } else {
6046 /* remove the bitmap */
6047 if (!mddev->bitmap)
6048 return -ENOENT;
6049 if (mddev->bitmap->file)
6050 return -EINVAL;
6051 mddev->pers->quiesce(mddev, 1);
6052 bitmap_destroy(mddev);
6053 mddev->pers->quiesce(mddev, 0);
6054 mddev->bitmap_info.offset = 0;
6057 md_update_sb(mddev, 1);
6058 return rv;
6061 static int set_disk_faulty(struct mddev *mddev, dev_t dev)
6063 struct md_rdev *rdev;
6065 if (mddev->pers == NULL)
6066 return -ENODEV;
6068 rdev = find_rdev(mddev, dev);
6069 if (!rdev)
6070 return -ENODEV;
6072 md_error(mddev, rdev);
6073 if (!test_bit(Faulty, &rdev->flags))
6074 return -EBUSY;
6075 return 0;
6079 * We have a problem here : there is no easy way to give a CHS
6080 * virtual geometry. We currently pretend that we have a 2 heads
6081 * 4 sectors (with a BIG number of cylinders...). This drives
6082 * dosfs just mad... ;-)
6084 static int md_getgeo(struct block_device *bdev, struct hd_geometry *geo)
6086 struct mddev *mddev = bdev->bd_disk->private_data;
6088 geo->heads = 2;
6089 geo->sectors = 4;
6090 geo->cylinders = mddev->array_sectors / 8;
6091 return 0;
6094 static int md_ioctl(struct block_device *bdev, fmode_t mode,
6095 unsigned int cmd, unsigned long arg)
6097 int err = 0;
6098 void __user *argp = (void __user *)arg;
6099 struct mddev *mddev = NULL;
6100 int ro;
6102 switch (cmd) {
6103 case RAID_VERSION:
6104 case GET_ARRAY_INFO:
6105 case GET_DISK_INFO:
6106 break;
6107 default:
6108 if (!capable(CAP_SYS_ADMIN))
6109 return -EACCES;
6113 * Commands dealing with the RAID driver but not any
6114 * particular array:
6116 switch (cmd)
6118 case RAID_VERSION:
6119 err = get_version(argp);
6120 goto done;
6122 case PRINT_RAID_DEBUG:
6123 err = 0;
6124 md_print_devices();
6125 goto done;
6127 #ifndef MODULE
6128 case RAID_AUTORUN:
6129 err = 0;
6130 autostart_arrays(arg);
6131 goto done;
6132 #endif
6133 default:;
6137 * Commands creating/starting a new array:
6140 mddev = bdev->bd_disk->private_data;
6142 if (!mddev) {
6143 BUG();
6144 goto abort;
6147 err = mddev_lock(mddev);
6148 if (err) {
6149 printk(KERN_INFO
6150 "md: ioctl lock interrupted, reason %d, cmd %d\n",
6151 err, cmd);
6152 goto abort;
6155 switch (cmd)
6157 case SET_ARRAY_INFO:
6159 mdu_array_info_t info;
6160 if (!arg)
6161 memset(&info, 0, sizeof(info));
6162 else if (copy_from_user(&info, argp, sizeof(info))) {
6163 err = -EFAULT;
6164 goto abort_unlock;
6166 if (mddev->pers) {
6167 err = update_array_info(mddev, &info);
6168 if (err) {
6169 printk(KERN_WARNING "md: couldn't update"
6170 " array info. %d\n", err);
6171 goto abort_unlock;
6173 goto done_unlock;
6175 if (!list_empty(&mddev->disks)) {
6176 printk(KERN_WARNING
6177 "md: array %s already has disks!\n",
6178 mdname(mddev));
6179 err = -EBUSY;
6180 goto abort_unlock;
6182 if (mddev->raid_disks) {
6183 printk(KERN_WARNING
6184 "md: array %s already initialised!\n",
6185 mdname(mddev));
6186 err = -EBUSY;
6187 goto abort_unlock;
6189 err = set_array_info(mddev, &info);
6190 if (err) {
6191 printk(KERN_WARNING "md: couldn't set"
6192 " array info. %d\n", err);
6193 goto abort_unlock;
6196 goto done_unlock;
6198 default:;
6202 * Commands querying/configuring an existing array:
6204 /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
6205 * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */
6206 if ((!mddev->raid_disks && !mddev->external)
6207 && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY
6208 && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE
6209 && cmd != GET_BITMAP_FILE) {
6210 err = -ENODEV;
6211 goto abort_unlock;
6215 * Commands even a read-only array can execute:
6217 switch (cmd)
6219 case GET_ARRAY_INFO:
6220 err = get_array_info(mddev, argp);
6221 goto done_unlock;
6223 case GET_BITMAP_FILE:
6224 err = get_bitmap_file(mddev, argp);
6225 goto done_unlock;
6227 case GET_DISK_INFO:
6228 err = get_disk_info(mddev, argp);
6229 goto done_unlock;
6231 case RESTART_ARRAY_RW:
6232 err = restart_array(mddev);
6233 goto done_unlock;
6235 case STOP_ARRAY:
6236 err = do_md_stop(mddev, 0, 1);
6237 goto done_unlock;
6239 case STOP_ARRAY_RO:
6240 err = md_set_readonly(mddev, 1);
6241 goto done_unlock;
6243 case BLKROSET:
6244 if (get_user(ro, (int __user *)(arg))) {
6245 err = -EFAULT;
6246 goto done_unlock;
6248 err = -EINVAL;
6250 /* if the bdev is going readonly the value of mddev->ro
6251 * does not matter, no writes are coming
6253 if (ro)
6254 goto done_unlock;
6256 /* are we are already prepared for writes? */
6257 if (mddev->ro != 1)
6258 goto done_unlock;
6260 /* transitioning to readauto need only happen for
6261 * arrays that call md_write_start
6263 if (mddev->pers) {
6264 err = restart_array(mddev);
6265 if (err == 0) {
6266 mddev->ro = 2;
6267 set_disk_ro(mddev->gendisk, 0);
6270 goto done_unlock;
6274 * The remaining ioctls are changing the state of the
6275 * superblock, so we do not allow them on read-only arrays.
6276 * However non-MD ioctls (e.g. get-size) will still come through
6277 * here and hit the 'default' below, so only disallow
6278 * 'md' ioctls, and switch to rw mode if started auto-readonly.
6280 if (_IOC_TYPE(cmd) == MD_MAJOR && mddev->ro && mddev->pers) {
6281 if (mddev->ro == 2) {
6282 mddev->ro = 0;
6283 sysfs_notify_dirent_safe(mddev->sysfs_state);
6284 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6285 md_wakeup_thread(mddev->thread);
6286 } else {
6287 err = -EROFS;
6288 goto abort_unlock;
6292 switch (cmd)
6294 case ADD_NEW_DISK:
6296 mdu_disk_info_t info;
6297 if (copy_from_user(&info, argp, sizeof(info)))
6298 err = -EFAULT;
6299 else
6300 err = add_new_disk(mddev, &info);
6301 goto done_unlock;
6304 case HOT_REMOVE_DISK:
6305 err = hot_remove_disk(mddev, new_decode_dev(arg));
6306 goto done_unlock;
6308 case HOT_ADD_DISK:
6309 err = hot_add_disk(mddev, new_decode_dev(arg));
6310 goto done_unlock;
6312 case SET_DISK_FAULTY:
6313 err = set_disk_faulty(mddev, new_decode_dev(arg));
6314 goto done_unlock;
6316 case RUN_ARRAY:
6317 err = do_md_run(mddev);
6318 goto done_unlock;
6320 case SET_BITMAP_FILE:
6321 err = set_bitmap_file(mddev, (int)arg);
6322 goto done_unlock;
6324 default:
6325 err = -EINVAL;
6326 goto abort_unlock;
6329 done_unlock:
6330 abort_unlock:
6331 if (mddev->hold_active == UNTIL_IOCTL &&
6332 err != -EINVAL)
6333 mddev->hold_active = 0;
6334 mddev_unlock(mddev);
6336 return err;
6337 done:
6338 if (err)
6339 MD_BUG();
6340 abort:
6341 return err;
6343 #ifdef CONFIG_COMPAT
6344 static int md_compat_ioctl(struct block_device *bdev, fmode_t mode,
6345 unsigned int cmd, unsigned long arg)
6347 switch (cmd) {
6348 case HOT_REMOVE_DISK:
6349 case HOT_ADD_DISK:
6350 case SET_DISK_FAULTY:
6351 case SET_BITMAP_FILE:
6352 /* These take in integer arg, do not convert */
6353 break;
6354 default:
6355 arg = (unsigned long)compat_ptr(arg);
6356 break;
6359 return md_ioctl(bdev, mode, cmd, arg);
6361 #endif /* CONFIG_COMPAT */
6363 static int md_open(struct block_device *bdev, fmode_t mode)
6366 * Succeed if we can lock the mddev, which confirms that
6367 * it isn't being stopped right now.
6369 struct mddev *mddev = mddev_find(bdev->bd_dev);
6370 int err;
6372 if (mddev->gendisk != bdev->bd_disk) {
6373 /* we are racing with mddev_put which is discarding this
6374 * bd_disk.
6376 mddev_put(mddev);
6377 /* Wait until bdev->bd_disk is definitely gone */
6378 flush_workqueue(md_misc_wq);
6379 /* Then retry the open from the top */
6380 return -ERESTARTSYS;
6382 BUG_ON(mddev != bdev->bd_disk->private_data);
6384 if ((err = mutex_lock_interruptible(&mddev->open_mutex)))
6385 goto out;
6387 err = 0;
6388 atomic_inc(&mddev->openers);
6389 mutex_unlock(&mddev->open_mutex);
6391 check_disk_change(bdev);
6392 out:
6393 return err;
6396 static int md_release(struct gendisk *disk, fmode_t mode)
6398 struct mddev *mddev = disk->private_data;
6400 BUG_ON(!mddev);
6401 atomic_dec(&mddev->openers);
6402 mddev_put(mddev);
6404 return 0;
6407 static int md_media_changed(struct gendisk *disk)
6409 struct mddev *mddev = disk->private_data;
6411 return mddev->changed;
6414 static int md_revalidate(struct gendisk *disk)
6416 struct mddev *mddev = disk->private_data;
6418 mddev->changed = 0;
6419 return 0;
6421 static const struct block_device_operations md_fops =
6423 .owner = THIS_MODULE,
6424 .open = md_open,
6425 .release = md_release,
6426 .ioctl = md_ioctl,
6427 #ifdef CONFIG_COMPAT
6428 .compat_ioctl = md_compat_ioctl,
6429 #endif
6430 .getgeo = md_getgeo,
6431 .media_changed = md_media_changed,
6432 .revalidate_disk= md_revalidate,
6435 static int md_thread(void * arg)
6437 struct md_thread *thread = arg;
6440 * md_thread is a 'system-thread', it's priority should be very
6441 * high. We avoid resource deadlocks individually in each
6442 * raid personality. (RAID5 does preallocation) We also use RR and
6443 * the very same RT priority as kswapd, thus we will never get
6444 * into a priority inversion deadlock.
6446 * we definitely have to have equal or higher priority than
6447 * bdflush, otherwise bdflush will deadlock if there are too
6448 * many dirty RAID5 blocks.
6451 allow_signal(SIGKILL);
6452 while (!kthread_should_stop()) {
6454 /* We need to wait INTERRUPTIBLE so that
6455 * we don't add to the load-average.
6456 * That means we need to be sure no signals are
6457 * pending
6459 if (signal_pending(current))
6460 flush_signals(current);
6462 wait_event_interruptible_timeout
6463 (thread->wqueue,
6464 test_bit(THREAD_WAKEUP, &thread->flags)
6465 || kthread_should_stop(),
6466 thread->timeout);
6468 clear_bit(THREAD_WAKEUP, &thread->flags);
6469 if (!kthread_should_stop())
6470 thread->run(thread->mddev);
6473 return 0;
6476 void md_wakeup_thread(struct md_thread *thread)
6478 if (thread) {
6479 pr_debug("md: waking up MD thread %s.\n", thread->tsk->comm);
6480 set_bit(THREAD_WAKEUP, &thread->flags);
6481 wake_up(&thread->wqueue);
6485 struct md_thread *md_register_thread(void (*run) (struct mddev *), struct mddev *mddev,
6486 const char *name)
6488 struct md_thread *thread;
6490 thread = kzalloc(sizeof(struct md_thread), GFP_KERNEL);
6491 if (!thread)
6492 return NULL;
6494 init_waitqueue_head(&thread->wqueue);
6496 thread->run = run;
6497 thread->mddev = mddev;
6498 thread->timeout = MAX_SCHEDULE_TIMEOUT;
6499 thread->tsk = kthread_run(md_thread, thread,
6500 "%s_%s",
6501 mdname(thread->mddev),
6502 name ?: mddev->pers->name);
6503 if (IS_ERR(thread->tsk)) {
6504 kfree(thread);
6505 return NULL;
6507 return thread;
6510 void md_unregister_thread(struct md_thread **threadp)
6512 struct md_thread *thread = *threadp;
6513 if (!thread)
6514 return;
6515 pr_debug("interrupting MD-thread pid %d\n", task_pid_nr(thread->tsk));
6516 /* Locking ensures that mddev_unlock does not wake_up a
6517 * non-existent thread
6519 spin_lock(&pers_lock);
6520 *threadp = NULL;
6521 spin_unlock(&pers_lock);
6523 kthread_stop(thread->tsk);
6524 kfree(thread);
6527 void md_error(struct mddev *mddev, struct md_rdev *rdev)
6529 if (!mddev) {
6530 MD_BUG();
6531 return;
6534 if (!rdev || test_bit(Faulty, &rdev->flags))
6535 return;
6537 if (!mddev->pers || !mddev->pers->error_handler)
6538 return;
6539 mddev->pers->error_handler(mddev,rdev);
6540 if (mddev->degraded)
6541 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
6542 sysfs_notify_dirent_safe(rdev->sysfs_state);
6543 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6544 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6545 md_wakeup_thread(mddev->thread);
6546 if (mddev->event_work.func)
6547 queue_work(md_misc_wq, &mddev->event_work);
6548 md_new_event_inintr(mddev);
6551 /* seq_file implementation /proc/mdstat */
6553 static void status_unused(struct seq_file *seq)
6555 int i = 0;
6556 struct md_rdev *rdev;
6558 seq_printf(seq, "unused devices: ");
6560 list_for_each_entry(rdev, &pending_raid_disks, same_set) {
6561 char b[BDEVNAME_SIZE];
6562 i++;
6563 seq_printf(seq, "%s ",
6564 bdevname(rdev->bdev,b));
6566 if (!i)
6567 seq_printf(seq, "<none>");
6569 seq_printf(seq, "\n");
6573 static void status_resync(struct seq_file *seq, struct mddev * mddev)
6575 sector_t max_sectors, resync, res;
6576 unsigned long dt, db;
6577 sector_t rt;
6578 int scale;
6579 unsigned int per_milli;
6581 resync = mddev->curr_resync - atomic_read(&mddev->recovery_active);
6583 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
6584 max_sectors = mddev->resync_max_sectors;
6585 else
6586 max_sectors = mddev->dev_sectors;
6589 * Should not happen.
6591 if (!max_sectors) {
6592 MD_BUG();
6593 return;
6595 /* Pick 'scale' such that (resync>>scale)*1000 will fit
6596 * in a sector_t, and (max_sectors>>scale) will fit in a
6597 * u32, as those are the requirements for sector_div.
6598 * Thus 'scale' must be at least 10
6600 scale = 10;
6601 if (sizeof(sector_t) > sizeof(unsigned long)) {
6602 while ( max_sectors/2 > (1ULL<<(scale+32)))
6603 scale++;
6605 res = (resync>>scale)*1000;
6606 sector_div(res, (u32)((max_sectors>>scale)+1));
6608 per_milli = res;
6610 int i, x = per_milli/50, y = 20-x;
6611 seq_printf(seq, "[");
6612 for (i = 0; i < x; i++)
6613 seq_printf(seq, "=");
6614 seq_printf(seq, ">");
6615 for (i = 0; i < y; i++)
6616 seq_printf(seq, ".");
6617 seq_printf(seq, "] ");
6619 seq_printf(seq, " %s =%3u.%u%% (%llu/%llu)",
6620 (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)?
6621 "reshape" :
6622 (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)?
6623 "check" :
6624 (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
6625 "resync" : "recovery"))),
6626 per_milli/10, per_milli % 10,
6627 (unsigned long long) resync/2,
6628 (unsigned long long) max_sectors/2);
6631 * dt: time from mark until now
6632 * db: blocks written from mark until now
6633 * rt: remaining time
6635 * rt is a sector_t, so could be 32bit or 64bit.
6636 * So we divide before multiply in case it is 32bit and close
6637 * to the limit.
6638 * We scale the divisor (db) by 32 to avoid losing precision
6639 * near the end of resync when the number of remaining sectors
6640 * is close to 'db'.
6641 * We then divide rt by 32 after multiplying by db to compensate.
6642 * The '+1' avoids division by zero if db is very small.
6644 dt = ((jiffies - mddev->resync_mark) / HZ);
6645 if (!dt) dt++;
6646 db = (mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active))
6647 - mddev->resync_mark_cnt;
6649 rt = max_sectors - resync; /* number of remaining sectors */
6650 sector_div(rt, db/32+1);
6651 rt *= dt;
6652 rt >>= 5;
6654 seq_printf(seq, " finish=%lu.%lumin", (unsigned long)rt / 60,
6655 ((unsigned long)rt % 60)/6);
6657 seq_printf(seq, " speed=%ldK/sec", db/2/dt);
6660 static void *md_seq_start(struct seq_file *seq, loff_t *pos)
6662 struct list_head *tmp;
6663 loff_t l = *pos;
6664 struct mddev *mddev;
6666 if (l >= 0x10000)
6667 return NULL;
6668 if (!l--)
6669 /* header */
6670 return (void*)1;
6672 spin_lock(&all_mddevs_lock);
6673 list_for_each(tmp,&all_mddevs)
6674 if (!l--) {
6675 mddev = list_entry(tmp, struct mddev, all_mddevs);
6676 mddev_get(mddev);
6677 spin_unlock(&all_mddevs_lock);
6678 return mddev;
6680 spin_unlock(&all_mddevs_lock);
6681 if (!l--)
6682 return (void*)2;/* tail */
6683 return NULL;
6686 static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
6688 struct list_head *tmp;
6689 struct mddev *next_mddev, *mddev = v;
6691 ++*pos;
6692 if (v == (void*)2)
6693 return NULL;
6695 spin_lock(&all_mddevs_lock);
6696 if (v == (void*)1)
6697 tmp = all_mddevs.next;
6698 else
6699 tmp = mddev->all_mddevs.next;
6700 if (tmp != &all_mddevs)
6701 next_mddev = mddev_get(list_entry(tmp,struct mddev,all_mddevs));
6702 else {
6703 next_mddev = (void*)2;
6704 *pos = 0x10000;
6706 spin_unlock(&all_mddevs_lock);
6708 if (v != (void*)1)
6709 mddev_put(mddev);
6710 return next_mddev;
6714 static void md_seq_stop(struct seq_file *seq, void *v)
6716 struct mddev *mddev = v;
6718 if (mddev && v != (void*)1 && v != (void*)2)
6719 mddev_put(mddev);
6722 static int md_seq_show(struct seq_file *seq, void *v)
6724 struct mddev *mddev = v;
6725 sector_t sectors;
6726 struct md_rdev *rdev;
6727 struct bitmap *bitmap;
6729 if (v == (void*)1) {
6730 struct md_personality *pers;
6731 seq_printf(seq, "Personalities : ");
6732 spin_lock(&pers_lock);
6733 list_for_each_entry(pers, &pers_list, list)
6734 seq_printf(seq, "[%s] ", pers->name);
6736 spin_unlock(&pers_lock);
6737 seq_printf(seq, "\n");
6738 seq->poll_event = atomic_read(&md_event_count);
6739 return 0;
6741 if (v == (void*)2) {
6742 status_unused(seq);
6743 return 0;
6746 if (mddev_lock(mddev) < 0)
6747 return -EINTR;
6749 if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
6750 seq_printf(seq, "%s : %sactive", mdname(mddev),
6751 mddev->pers ? "" : "in");
6752 if (mddev->pers) {
6753 if (mddev->ro==1)
6754 seq_printf(seq, " (read-only)");
6755 if (mddev->ro==2)
6756 seq_printf(seq, " (auto-read-only)");
6757 seq_printf(seq, " %s", mddev->pers->name);
6760 sectors = 0;
6761 list_for_each_entry(rdev, &mddev->disks, same_set) {
6762 char b[BDEVNAME_SIZE];
6763 seq_printf(seq, " %s[%d]",
6764 bdevname(rdev->bdev,b), rdev->desc_nr);
6765 if (test_bit(WriteMostly, &rdev->flags))
6766 seq_printf(seq, "(W)");
6767 if (test_bit(Faulty, &rdev->flags)) {
6768 seq_printf(seq, "(F)");
6769 continue;
6771 if (rdev->raid_disk < 0)
6772 seq_printf(seq, "(S)"); /* spare */
6773 if (test_bit(Replacement, &rdev->flags))
6774 seq_printf(seq, "(R)");
6775 sectors += rdev->sectors;
6778 if (!list_empty(&mddev->disks)) {
6779 if (mddev->pers)
6780 seq_printf(seq, "\n %llu blocks",
6781 (unsigned long long)
6782 mddev->array_sectors / 2);
6783 else
6784 seq_printf(seq, "\n %llu blocks",
6785 (unsigned long long)sectors / 2);
6787 if (mddev->persistent) {
6788 if (mddev->major_version != 0 ||
6789 mddev->minor_version != 90) {
6790 seq_printf(seq," super %d.%d",
6791 mddev->major_version,
6792 mddev->minor_version);
6794 } else if (mddev->external)
6795 seq_printf(seq, " super external:%s",
6796 mddev->metadata_type);
6797 else
6798 seq_printf(seq, " super non-persistent");
6800 if (mddev->pers) {
6801 mddev->pers->status(seq, mddev);
6802 seq_printf(seq, "\n ");
6803 if (mddev->pers->sync_request) {
6804 if (mddev->curr_resync > 2) {
6805 status_resync(seq, mddev);
6806 seq_printf(seq, "\n ");
6807 } else if (mddev->curr_resync == 1 || mddev->curr_resync == 2)
6808 seq_printf(seq, "\tresync=DELAYED\n ");
6809 else if (mddev->recovery_cp < MaxSector)
6810 seq_printf(seq, "\tresync=PENDING\n ");
6812 } else
6813 seq_printf(seq, "\n ");
6815 if ((bitmap = mddev->bitmap)) {
6816 unsigned long chunk_kb;
6817 unsigned long flags;
6818 spin_lock_irqsave(&bitmap->lock, flags);
6819 chunk_kb = mddev->bitmap_info.chunksize >> 10;
6820 seq_printf(seq, "bitmap: %lu/%lu pages [%luKB], "
6821 "%lu%s chunk",
6822 bitmap->pages - bitmap->missing_pages,
6823 bitmap->pages,
6824 (bitmap->pages - bitmap->missing_pages)
6825 << (PAGE_SHIFT - 10),
6826 chunk_kb ? chunk_kb : mddev->bitmap_info.chunksize,
6827 chunk_kb ? "KB" : "B");
6828 if (bitmap->file) {
6829 seq_printf(seq, ", file: ");
6830 seq_path(seq, &bitmap->file->f_path, " \t\n");
6833 seq_printf(seq, "\n");
6834 spin_unlock_irqrestore(&bitmap->lock, flags);
6837 seq_printf(seq, "\n");
6839 mddev_unlock(mddev);
6841 return 0;
6844 static const struct seq_operations md_seq_ops = {
6845 .start = md_seq_start,
6846 .next = md_seq_next,
6847 .stop = md_seq_stop,
6848 .show = md_seq_show,
6851 static int md_seq_open(struct inode *inode, struct file *file)
6853 struct seq_file *seq;
6854 int error;
6856 error = seq_open(file, &md_seq_ops);
6857 if (error)
6858 return error;
6860 seq = file->private_data;
6861 seq->poll_event = atomic_read(&md_event_count);
6862 return error;
6865 static unsigned int mdstat_poll(struct file *filp, poll_table *wait)
6867 struct seq_file *seq = filp->private_data;
6868 int mask;
6870 poll_wait(filp, &md_event_waiters, wait);
6872 /* always allow read */
6873 mask = POLLIN | POLLRDNORM;
6875 if (seq->poll_event != atomic_read(&md_event_count))
6876 mask |= POLLERR | POLLPRI;
6877 return mask;
6880 static const struct file_operations md_seq_fops = {
6881 .owner = THIS_MODULE,
6882 .open = md_seq_open,
6883 .read = seq_read,
6884 .llseek = seq_lseek,
6885 .release = seq_release_private,
6886 .poll = mdstat_poll,
6889 int register_md_personality(struct md_personality *p)
6891 spin_lock(&pers_lock);
6892 list_add_tail(&p->list, &pers_list);
6893 printk(KERN_INFO "md: %s personality registered for level %d\n", p->name, p->level);
6894 spin_unlock(&pers_lock);
6895 return 0;
6898 int unregister_md_personality(struct md_personality *p)
6900 printk(KERN_INFO "md: %s personality unregistered\n", p->name);
6901 spin_lock(&pers_lock);
6902 list_del_init(&p->list);
6903 spin_unlock(&pers_lock);
6904 return 0;
6907 static int is_mddev_idle(struct mddev *mddev, int init)
6909 struct md_rdev * rdev;
6910 int idle;
6911 int curr_events;
6913 idle = 1;
6914 rcu_read_lock();
6915 rdev_for_each_rcu(rdev, mddev) {
6916 struct gendisk *disk = rdev->bdev->bd_contains->bd_disk;
6917 curr_events = (int)part_stat_read(&disk->part0, sectors[0]) +
6918 (int)part_stat_read(&disk->part0, sectors[1]) -
6919 atomic_read(&disk->sync_io);
6920 /* sync IO will cause sync_io to increase before the disk_stats
6921 * as sync_io is counted when a request starts, and
6922 * disk_stats is counted when it completes.
6923 * So resync activity will cause curr_events to be smaller than
6924 * when there was no such activity.
6925 * non-sync IO will cause disk_stat to increase without
6926 * increasing sync_io so curr_events will (eventually)
6927 * be larger than it was before. Once it becomes
6928 * substantially larger, the test below will cause
6929 * the array to appear non-idle, and resync will slow
6930 * down.
6931 * If there is a lot of outstanding resync activity when
6932 * we set last_event to curr_events, then all that activity
6933 * completing might cause the array to appear non-idle
6934 * and resync will be slowed down even though there might
6935 * not have been non-resync activity. This will only
6936 * happen once though. 'last_events' will soon reflect
6937 * the state where there is little or no outstanding
6938 * resync requests, and further resync activity will
6939 * always make curr_events less than last_events.
6942 if (init || curr_events - rdev->last_events > 64) {
6943 rdev->last_events = curr_events;
6944 idle = 0;
6947 rcu_read_unlock();
6948 return idle;
6951 void md_done_sync(struct mddev *mddev, int blocks, int ok)
6953 /* another "blocks" (512byte) blocks have been synced */
6954 atomic_sub(blocks, &mddev->recovery_active);
6955 wake_up(&mddev->recovery_wait);
6956 if (!ok) {
6957 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6958 md_wakeup_thread(mddev->thread);
6959 // stop recovery, signal do_sync ....
6964 /* md_write_start(mddev, bi)
6965 * If we need to update some array metadata (e.g. 'active' flag
6966 * in superblock) before writing, schedule a superblock update
6967 * and wait for it to complete.
6969 void md_write_start(struct mddev *mddev, struct bio *bi)
6971 int did_change = 0;
6972 if (bio_data_dir(bi) != WRITE)
6973 return;
6975 BUG_ON(mddev->ro == 1);
6976 if (mddev->ro == 2) {
6977 /* need to switch to read/write */
6978 mddev->ro = 0;
6979 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6980 md_wakeup_thread(mddev->thread);
6981 md_wakeup_thread(mddev->sync_thread);
6982 did_change = 1;
6984 atomic_inc(&mddev->writes_pending);
6985 if (mddev->safemode == 1)
6986 mddev->safemode = 0;
6987 if (mddev->in_sync) {
6988 spin_lock_irq(&mddev->write_lock);
6989 if (mddev->in_sync) {
6990 mddev->in_sync = 0;
6991 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
6992 set_bit(MD_CHANGE_PENDING, &mddev->flags);
6993 md_wakeup_thread(mddev->thread);
6994 did_change = 1;
6996 spin_unlock_irq(&mddev->write_lock);
6998 if (did_change)
6999 sysfs_notify_dirent_safe(mddev->sysfs_state);
7000 wait_event(mddev->sb_wait,
7001 !test_bit(MD_CHANGE_PENDING, &mddev->flags));
7004 void md_write_end(struct mddev *mddev)
7006 if (atomic_dec_and_test(&mddev->writes_pending)) {
7007 if (mddev->safemode == 2)
7008 md_wakeup_thread(mddev->thread);
7009 else if (mddev->safemode_delay)
7010 mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay);
7014 /* md_allow_write(mddev)
7015 * Calling this ensures that the array is marked 'active' so that writes
7016 * may proceed without blocking. It is important to call this before
7017 * attempting a GFP_KERNEL allocation while holding the mddev lock.
7018 * Must be called with mddev_lock held.
7020 * In the ->external case MD_CHANGE_CLEAN can not be cleared until mddev->lock
7021 * is dropped, so return -EAGAIN after notifying userspace.
7023 int md_allow_write(struct mddev *mddev)
7025 if (!mddev->pers)
7026 return 0;
7027 if (mddev->ro)
7028 return 0;
7029 if (!mddev->pers->sync_request)
7030 return 0;
7032 spin_lock_irq(&mddev->write_lock);
7033 if (mddev->in_sync) {
7034 mddev->in_sync = 0;
7035 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
7036 set_bit(MD_CHANGE_PENDING, &mddev->flags);
7037 if (mddev->safemode_delay &&
7038 mddev->safemode == 0)
7039 mddev->safemode = 1;
7040 spin_unlock_irq(&mddev->write_lock);
7041 md_update_sb(mddev, 0);
7042 sysfs_notify_dirent_safe(mddev->sysfs_state);
7043 } else
7044 spin_unlock_irq(&mddev->write_lock);
7046 if (test_bit(MD_CHANGE_PENDING, &mddev->flags))
7047 return -EAGAIN;
7048 else
7049 return 0;
7051 EXPORT_SYMBOL_GPL(md_allow_write);
7053 #define SYNC_MARKS 10
7054 #define SYNC_MARK_STEP (3*HZ)
7055 void md_do_sync(struct mddev *mddev)
7057 struct mddev *mddev2;
7058 unsigned int currspeed = 0,
7059 window;
7060 sector_t max_sectors,j, io_sectors;
7061 unsigned long mark[SYNC_MARKS];
7062 sector_t mark_cnt[SYNC_MARKS];
7063 int last_mark,m;
7064 struct list_head *tmp;
7065 sector_t last_check;
7066 int skipped = 0;
7067 struct md_rdev *rdev;
7068 char *desc;
7070 /* just incase thread restarts... */
7071 if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
7072 return;
7073 if (mddev->ro) /* never try to sync a read-only array */
7074 return;
7076 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
7077 if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
7078 desc = "data-check";
7079 else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
7080 desc = "requested-resync";
7081 else
7082 desc = "resync";
7083 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
7084 desc = "reshape";
7085 else
7086 desc = "recovery";
7088 /* we overload curr_resync somewhat here.
7089 * 0 == not engaged in resync at all
7090 * 2 == checking that there is no conflict with another sync
7091 * 1 == like 2, but have yielded to allow conflicting resync to
7092 * commense
7093 * other == active in resync - this many blocks
7095 * Before starting a resync we must have set curr_resync to
7096 * 2, and then checked that every "conflicting" array has curr_resync
7097 * less than ours. When we find one that is the same or higher
7098 * we wait on resync_wait. To avoid deadlock, we reduce curr_resync
7099 * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
7100 * This will mean we have to start checking from the beginning again.
7104 do {
7105 mddev->curr_resync = 2;
7107 try_again:
7108 if (kthread_should_stop())
7109 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
7111 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
7112 goto skip;
7113 for_each_mddev(mddev2, tmp) {
7114 if (mddev2 == mddev)
7115 continue;
7116 if (!mddev->parallel_resync
7117 && mddev2->curr_resync
7118 && match_mddev_units(mddev, mddev2)) {
7119 DEFINE_WAIT(wq);
7120 if (mddev < mddev2 && mddev->curr_resync == 2) {
7121 /* arbitrarily yield */
7122 mddev->curr_resync = 1;
7123 wake_up(&resync_wait);
7125 if (mddev > mddev2 && mddev->curr_resync == 1)
7126 /* no need to wait here, we can wait the next
7127 * time 'round when curr_resync == 2
7129 continue;
7130 /* We need to wait 'interruptible' so as not to
7131 * contribute to the load average, and not to
7132 * be caught by 'softlockup'
7134 prepare_to_wait(&resync_wait, &wq, TASK_INTERRUPTIBLE);
7135 if (!kthread_should_stop() &&
7136 mddev2->curr_resync >= mddev->curr_resync) {
7137 printk(KERN_INFO "md: delaying %s of %s"
7138 " until %s has finished (they"
7139 " share one or more physical units)\n",
7140 desc, mdname(mddev), mdname(mddev2));
7141 mddev_put(mddev2);
7142 if (signal_pending(current))
7143 flush_signals(current);
7144 schedule();
7145 finish_wait(&resync_wait, &wq);
7146 goto try_again;
7148 finish_wait(&resync_wait, &wq);
7151 } while (mddev->curr_resync < 2);
7153 j = 0;
7154 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
7155 /* resync follows the size requested by the personality,
7156 * which defaults to physical size, but can be virtual size
7158 max_sectors = mddev->resync_max_sectors;
7159 mddev->resync_mismatches = 0;
7160 /* we don't use the checkpoint if there's a bitmap */
7161 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
7162 j = mddev->resync_min;
7163 else if (!mddev->bitmap)
7164 j = mddev->recovery_cp;
7166 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
7167 max_sectors = mddev->dev_sectors;
7168 else {
7169 /* recovery follows the physical size of devices */
7170 max_sectors = mddev->dev_sectors;
7171 j = MaxSector;
7172 rcu_read_lock();
7173 list_for_each_entry_rcu(rdev, &mddev->disks, same_set)
7174 if (rdev->raid_disk >= 0 &&
7175 !test_bit(Faulty, &rdev->flags) &&
7176 !test_bit(In_sync, &rdev->flags) &&
7177 rdev->recovery_offset < j)
7178 j = rdev->recovery_offset;
7179 rcu_read_unlock();
7182 printk(KERN_INFO "md: %s of RAID array %s\n", desc, mdname(mddev));
7183 printk(KERN_INFO "md: minimum _guaranteed_ speed:"
7184 " %d KB/sec/disk.\n", speed_min(mddev));
7185 printk(KERN_INFO "md: using maximum available idle IO bandwidth "
7186 "(but not more than %d KB/sec) for %s.\n",
7187 speed_max(mddev), desc);
7189 is_mddev_idle(mddev, 1); /* this initializes IO event counters */
7191 io_sectors = 0;
7192 for (m = 0; m < SYNC_MARKS; m++) {
7193 mark[m] = jiffies;
7194 mark_cnt[m] = io_sectors;
7196 last_mark = 0;
7197 mddev->resync_mark = mark[last_mark];
7198 mddev->resync_mark_cnt = mark_cnt[last_mark];
7201 * Tune reconstruction:
7203 window = 32*(PAGE_SIZE/512);
7204 printk(KERN_INFO "md: using %dk window, over a total of %lluk.\n",
7205 window/2, (unsigned long long)max_sectors/2);
7207 atomic_set(&mddev->recovery_active, 0);
7208 last_check = 0;
7210 if (j>2) {
7211 printk(KERN_INFO
7212 "md: resuming %s of %s from checkpoint.\n",
7213 desc, mdname(mddev));
7214 mddev->curr_resync = j;
7216 mddev->curr_resync_completed = j;
7218 while (j < max_sectors) {
7219 sector_t sectors;
7221 skipped = 0;
7223 if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
7224 ((mddev->curr_resync > mddev->curr_resync_completed &&
7225 (mddev->curr_resync - mddev->curr_resync_completed)
7226 > (max_sectors >> 4)) ||
7227 (j - mddev->curr_resync_completed)*2
7228 >= mddev->resync_max - mddev->curr_resync_completed
7229 )) {
7230 /* time to update curr_resync_completed */
7231 wait_event(mddev->recovery_wait,
7232 atomic_read(&mddev->recovery_active) == 0);
7233 mddev->curr_resync_completed = j;
7234 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
7235 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
7238 while (j >= mddev->resync_max && !kthread_should_stop()) {
7239 /* As this condition is controlled by user-space,
7240 * we can block indefinitely, so use '_interruptible'
7241 * to avoid triggering warnings.
7243 flush_signals(current); /* just in case */
7244 wait_event_interruptible(mddev->recovery_wait,
7245 mddev->resync_max > j
7246 || kthread_should_stop());
7249 if (kthread_should_stop())
7250 goto interrupted;
7252 sectors = mddev->pers->sync_request(mddev, j, &skipped,
7253 currspeed < speed_min(mddev));
7254 if (sectors == 0) {
7255 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
7256 goto out;
7259 if (!skipped) { /* actual IO requested */
7260 io_sectors += sectors;
7261 atomic_add(sectors, &mddev->recovery_active);
7264 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
7265 break;
7267 j += sectors;
7268 if (j>1) mddev->curr_resync = j;
7269 mddev->curr_mark_cnt = io_sectors;
7270 if (last_check == 0)
7271 /* this is the earliest that rebuild will be
7272 * visible in /proc/mdstat
7274 md_new_event(mddev);
7276 if (last_check + window > io_sectors || j == max_sectors)
7277 continue;
7279 last_check = io_sectors;
7280 repeat:
7281 if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) {
7282 /* step marks */
7283 int next = (last_mark+1) % SYNC_MARKS;
7285 mddev->resync_mark = mark[next];
7286 mddev->resync_mark_cnt = mark_cnt[next];
7287 mark[next] = jiffies;
7288 mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active);
7289 last_mark = next;
7293 if (kthread_should_stop())
7294 goto interrupted;
7298 * this loop exits only if either when we are slower than
7299 * the 'hard' speed limit, or the system was IO-idle for
7300 * a jiffy.
7301 * the system might be non-idle CPU-wise, but we only care
7302 * about not overloading the IO subsystem. (things like an
7303 * e2fsck being done on the RAID array should execute fast)
7305 cond_resched();
7307 currspeed = ((unsigned long)(io_sectors-mddev->resync_mark_cnt))/2
7308 /((jiffies-mddev->resync_mark)/HZ +1) +1;
7310 if (currspeed > speed_min(mddev)) {
7311 if ((currspeed > speed_max(mddev)) ||
7312 !is_mddev_idle(mddev, 0)) {
7313 msleep(500);
7314 goto repeat;
7318 printk(KERN_INFO "md: %s: %s done.\n",mdname(mddev), desc);
7320 * this also signals 'finished resyncing' to md_stop
7322 out:
7323 wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
7325 /* tell personality that we are finished */
7326 mddev->pers->sync_request(mddev, max_sectors, &skipped, 1);
7328 if (!test_bit(MD_RECOVERY_CHECK, &mddev->recovery) &&
7329 mddev->curr_resync > 2) {
7330 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
7331 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
7332 if (mddev->curr_resync >= mddev->recovery_cp) {
7333 printk(KERN_INFO
7334 "md: checkpointing %s of %s.\n",
7335 desc, mdname(mddev));
7336 mddev->recovery_cp =
7337 mddev->curr_resync_completed;
7339 } else
7340 mddev->recovery_cp = MaxSector;
7341 } else {
7342 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery))
7343 mddev->curr_resync = MaxSector;
7344 rcu_read_lock();
7345 list_for_each_entry_rcu(rdev, &mddev->disks, same_set)
7346 if (rdev->raid_disk >= 0 &&
7347 mddev->delta_disks >= 0 &&
7348 !test_bit(Faulty, &rdev->flags) &&
7349 !test_bit(In_sync, &rdev->flags) &&
7350 rdev->recovery_offset < mddev->curr_resync)
7351 rdev->recovery_offset = mddev->curr_resync;
7352 rcu_read_unlock();
7355 skip:
7356 set_bit(MD_CHANGE_DEVS, &mddev->flags);
7358 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
7359 /* We completed so min/max setting can be forgotten if used. */
7360 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
7361 mddev->resync_min = 0;
7362 mddev->resync_max = MaxSector;
7363 } else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
7364 mddev->resync_min = mddev->curr_resync_completed;
7365 mddev->curr_resync = 0;
7366 wake_up(&resync_wait);
7367 set_bit(MD_RECOVERY_DONE, &mddev->recovery);
7368 md_wakeup_thread(mddev->thread);
7369 return;
7371 interrupted:
7373 * got a signal, exit.
7375 printk(KERN_INFO
7376 "md: md_do_sync() got signal ... exiting\n");
7377 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
7378 goto out;
7381 EXPORT_SYMBOL_GPL(md_do_sync);
7383 static int remove_and_add_spares(struct mddev *mddev)
7385 struct md_rdev *rdev;
7386 int spares = 0;
7387 int removed = 0;
7389 mddev->curr_resync_completed = 0;
7391 list_for_each_entry(rdev, &mddev->disks, same_set)
7392 if (rdev->raid_disk >= 0 &&
7393 !test_bit(Blocked, &rdev->flags) &&
7394 (test_bit(Faulty, &rdev->flags) ||
7395 ! test_bit(In_sync, &rdev->flags)) &&
7396 atomic_read(&rdev->nr_pending)==0) {
7397 if (mddev->pers->hot_remove_disk(
7398 mddev, rdev) == 0) {
7399 sysfs_unlink_rdev(mddev, rdev);
7400 rdev->raid_disk = -1;
7401 removed++;
7404 if (removed)
7405 sysfs_notify(&mddev->kobj, NULL,
7406 "degraded");
7409 list_for_each_entry(rdev, &mddev->disks, same_set) {
7410 if (rdev->raid_disk >= 0 &&
7411 !test_bit(In_sync, &rdev->flags) &&
7412 !test_bit(Faulty, &rdev->flags))
7413 spares++;
7414 if (rdev->raid_disk < 0
7415 && !test_bit(Faulty, &rdev->flags)) {
7416 rdev->recovery_offset = 0;
7417 if (mddev->pers->
7418 hot_add_disk(mddev, rdev) == 0) {
7419 if (sysfs_link_rdev(mddev, rdev))
7420 /* failure here is OK */;
7421 spares++;
7422 md_new_event(mddev);
7423 set_bit(MD_CHANGE_DEVS, &mddev->flags);
7427 return spares;
7430 static void reap_sync_thread(struct mddev *mddev)
7432 struct md_rdev *rdev;
7434 /* resync has finished, collect result */
7435 md_unregister_thread(&mddev->sync_thread);
7436 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
7437 !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
7438 /* success...*/
7439 /* activate any spares */
7440 if (mddev->pers->spare_active(mddev))
7441 sysfs_notify(&mddev->kobj, NULL,
7442 "degraded");
7444 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
7445 mddev->pers->finish_reshape)
7446 mddev->pers->finish_reshape(mddev);
7448 /* If array is no-longer degraded, then any saved_raid_disk
7449 * information must be scrapped. Also if any device is now
7450 * In_sync we must scrape the saved_raid_disk for that device
7451 * do the superblock for an incrementally recovered device
7452 * written out.
7454 list_for_each_entry(rdev, &mddev->disks, same_set)
7455 if (!mddev->degraded ||
7456 test_bit(In_sync, &rdev->flags))
7457 rdev->saved_raid_disk = -1;
7459 md_update_sb(mddev, 1);
7460 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
7461 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
7462 clear_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
7463 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
7464 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
7465 /* flag recovery needed just to double check */
7466 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7467 sysfs_notify_dirent_safe(mddev->sysfs_action);
7468 md_new_event(mddev);
7469 if (mddev->event_work.func)
7470 queue_work(md_misc_wq, &mddev->event_work);
7474 * This routine is regularly called by all per-raid-array threads to
7475 * deal with generic issues like resync and super-block update.
7476 * Raid personalities that don't have a thread (linear/raid0) do not
7477 * need this as they never do any recovery or update the superblock.
7479 * It does not do any resync itself, but rather "forks" off other threads
7480 * to do that as needed.
7481 * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
7482 * "->recovery" and create a thread at ->sync_thread.
7483 * When the thread finishes it sets MD_RECOVERY_DONE
7484 * and wakeups up this thread which will reap the thread and finish up.
7485 * This thread also removes any faulty devices (with nr_pending == 0).
7487 * The overall approach is:
7488 * 1/ if the superblock needs updating, update it.
7489 * 2/ If a recovery thread is running, don't do anything else.
7490 * 3/ If recovery has finished, clean up, possibly marking spares active.
7491 * 4/ If there are any faulty devices, remove them.
7492 * 5/ If array is degraded, try to add spares devices
7493 * 6/ If array has spares or is not in-sync, start a resync thread.
7495 void md_check_recovery(struct mddev *mddev)
7497 if (mddev->suspended)
7498 return;
7500 if (mddev->bitmap)
7501 bitmap_daemon_work(mddev);
7503 if (signal_pending(current)) {
7504 if (mddev->pers->sync_request && !mddev->external) {
7505 printk(KERN_INFO "md: %s in immediate safe mode\n",
7506 mdname(mddev));
7507 mddev->safemode = 2;
7509 flush_signals(current);
7512 if (mddev->ro && !test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
7513 return;
7514 if ( ! (
7515 (mddev->flags & ~ (1<<MD_CHANGE_PENDING)) ||
7516 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
7517 test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
7518 (mddev->external == 0 && mddev->safemode == 1) ||
7519 (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending)
7520 && !mddev->in_sync && mddev->recovery_cp == MaxSector)
7522 return;
7524 if (mddev_trylock(mddev)) {
7525 int spares = 0;
7527 if (mddev->ro) {
7528 /* Only thing we do on a ro array is remove
7529 * failed devices.
7531 struct md_rdev *rdev;
7532 list_for_each_entry(rdev, &mddev->disks, same_set)
7533 if (rdev->raid_disk >= 0 &&
7534 !test_bit(Blocked, &rdev->flags) &&
7535 test_bit(Faulty, &rdev->flags) &&
7536 atomic_read(&rdev->nr_pending)==0) {
7537 if (mddev->pers->hot_remove_disk(
7538 mddev, rdev) == 0) {
7539 sysfs_unlink_rdev(mddev, rdev);
7540 rdev->raid_disk = -1;
7543 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7544 goto unlock;
7547 if (!mddev->external) {
7548 int did_change = 0;
7549 spin_lock_irq(&mddev->write_lock);
7550 if (mddev->safemode &&
7551 !atomic_read(&mddev->writes_pending) &&
7552 !mddev->in_sync &&
7553 mddev->recovery_cp == MaxSector) {
7554 mddev->in_sync = 1;
7555 did_change = 1;
7556 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
7558 if (mddev->safemode == 1)
7559 mddev->safemode = 0;
7560 spin_unlock_irq(&mddev->write_lock);
7561 if (did_change)
7562 sysfs_notify_dirent_safe(mddev->sysfs_state);
7565 if (mddev->flags)
7566 md_update_sb(mddev, 0);
7568 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
7569 !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
7570 /* resync/recovery still happening */
7571 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7572 goto unlock;
7574 if (mddev->sync_thread) {
7575 reap_sync_thread(mddev);
7576 goto unlock;
7578 /* Set RUNNING before clearing NEEDED to avoid
7579 * any transients in the value of "sync_action".
7581 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
7582 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7583 /* Clear some bits that don't mean anything, but
7584 * might be left set
7586 clear_bit(MD_RECOVERY_INTR, &mddev->recovery);
7587 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
7589 if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
7590 goto unlock;
7591 /* no recovery is running.
7592 * remove any failed drives, then
7593 * add spares if possible.
7594 * Spare are also removed and re-added, to allow
7595 * the personality to fail the re-add.
7598 if (mddev->reshape_position != MaxSector) {
7599 if (mddev->pers->check_reshape == NULL ||
7600 mddev->pers->check_reshape(mddev) != 0)
7601 /* Cannot proceed */
7602 goto unlock;
7603 set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
7604 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
7605 } else if ((spares = remove_and_add_spares(mddev))) {
7606 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
7607 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
7608 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
7609 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
7610 } else if (mddev->recovery_cp < MaxSector) {
7611 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
7612 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
7613 } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
7614 /* nothing to be done ... */
7615 goto unlock;
7617 if (mddev->pers->sync_request) {
7618 if (spares && mddev->bitmap && ! mddev->bitmap->file) {
7619 /* We are adding a device or devices to an array
7620 * which has the bitmap stored on all devices.
7621 * So make sure all bitmap pages get written
7623 bitmap_write_all(mddev->bitmap);
7625 mddev->sync_thread = md_register_thread(md_do_sync,
7626 mddev,
7627 "resync");
7628 if (!mddev->sync_thread) {
7629 printk(KERN_ERR "%s: could not start resync"
7630 " thread...\n",
7631 mdname(mddev));
7632 /* leave the spares where they are, it shouldn't hurt */
7633 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
7634 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
7635 clear_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
7636 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
7637 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
7638 } else
7639 md_wakeup_thread(mddev->sync_thread);
7640 sysfs_notify_dirent_safe(mddev->sysfs_action);
7641 md_new_event(mddev);
7643 unlock:
7644 if (!mddev->sync_thread) {
7645 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
7646 if (test_and_clear_bit(MD_RECOVERY_RECOVER,
7647 &mddev->recovery))
7648 if (mddev->sysfs_action)
7649 sysfs_notify_dirent_safe(mddev->sysfs_action);
7651 mddev_unlock(mddev);
7655 void md_wait_for_blocked_rdev(struct md_rdev *rdev, struct mddev *mddev)
7657 sysfs_notify_dirent_safe(rdev->sysfs_state);
7658 wait_event_timeout(rdev->blocked_wait,
7659 !test_bit(Blocked, &rdev->flags) &&
7660 !test_bit(BlockedBadBlocks, &rdev->flags),
7661 msecs_to_jiffies(5000));
7662 rdev_dec_pending(rdev, mddev);
7664 EXPORT_SYMBOL(md_wait_for_blocked_rdev);
7667 /* Bad block management.
7668 * We can record which blocks on each device are 'bad' and so just
7669 * fail those blocks, or that stripe, rather than the whole device.
7670 * Entries in the bad-block table are 64bits wide. This comprises:
7671 * Length of bad-range, in sectors: 0-511 for lengths 1-512
7672 * Start of bad-range, sector offset, 54 bits (allows 8 exbibytes)
7673 * A 'shift' can be set so that larger blocks are tracked and
7674 * consequently larger devices can be covered.
7675 * 'Acknowledged' flag - 1 bit. - the most significant bit.
7677 * Locking of the bad-block table uses a seqlock so md_is_badblock
7678 * might need to retry if it is very unlucky.
7679 * We will sometimes want to check for bad blocks in a bi_end_io function,
7680 * so we use the write_seqlock_irq variant.
7682 * When looking for a bad block we specify a range and want to
7683 * know if any block in the range is bad. So we binary-search
7684 * to the last range that starts at-or-before the given endpoint,
7685 * (or "before the sector after the target range")
7686 * then see if it ends after the given start.
7687 * We return
7688 * 0 if there are no known bad blocks in the range
7689 * 1 if there are known bad block which are all acknowledged
7690 * -1 if there are bad blocks which have not yet been acknowledged in metadata.
7691 * plus the start/length of the first bad section we overlap.
7693 int md_is_badblock(struct badblocks *bb, sector_t s, int sectors,
7694 sector_t *first_bad, int *bad_sectors)
7696 int hi;
7697 int lo = 0;
7698 u64 *p = bb->page;
7699 int rv = 0;
7700 sector_t target = s + sectors;
7701 unsigned seq;
7703 if (bb->shift > 0) {
7704 /* round the start down, and the end up */
7705 s >>= bb->shift;
7706 target += (1<<bb->shift) - 1;
7707 target >>= bb->shift;
7708 sectors = target - s;
7710 /* 'target' is now the first block after the bad range */
7712 retry:
7713 seq = read_seqbegin(&bb->lock);
7715 hi = bb->count;
7717 /* Binary search between lo and hi for 'target'
7718 * i.e. for the last range that starts before 'target'
7720 /* INVARIANT: ranges before 'lo' and at-or-after 'hi'
7721 * are known not to be the last range before target.
7722 * VARIANT: hi-lo is the number of possible
7723 * ranges, and decreases until it reaches 1
7725 while (hi - lo > 1) {
7726 int mid = (lo + hi) / 2;
7727 sector_t a = BB_OFFSET(p[mid]);
7728 if (a < target)
7729 /* This could still be the one, earlier ranges
7730 * could not. */
7731 lo = mid;
7732 else
7733 /* This and later ranges are definitely out. */
7734 hi = mid;
7736 /* 'lo' might be the last that started before target, but 'hi' isn't */
7737 if (hi > lo) {
7738 /* need to check all range that end after 's' to see if
7739 * any are unacknowledged.
7741 while (lo >= 0 &&
7742 BB_OFFSET(p[lo]) + BB_LEN(p[lo]) > s) {
7743 if (BB_OFFSET(p[lo]) < target) {
7744 /* starts before the end, and finishes after
7745 * the start, so they must overlap
7747 if (rv != -1 && BB_ACK(p[lo]))
7748 rv = 1;
7749 else
7750 rv = -1;
7751 *first_bad = BB_OFFSET(p[lo]);
7752 *bad_sectors = BB_LEN(p[lo]);
7754 lo--;
7758 if (read_seqretry(&bb->lock, seq))
7759 goto retry;
7761 return rv;
7763 EXPORT_SYMBOL_GPL(md_is_badblock);
7766 * Add a range of bad blocks to the table.
7767 * This might extend the table, or might contract it
7768 * if two adjacent ranges can be merged.
7769 * We binary-search to find the 'insertion' point, then
7770 * decide how best to handle it.
7772 static int md_set_badblocks(struct badblocks *bb, sector_t s, int sectors,
7773 int acknowledged)
7775 u64 *p;
7776 int lo, hi;
7777 int rv = 1;
7779 if (bb->shift < 0)
7780 /* badblocks are disabled */
7781 return 0;
7783 if (bb->shift) {
7784 /* round the start down, and the end up */
7785 sector_t next = s + sectors;
7786 s >>= bb->shift;
7787 next += (1<<bb->shift) - 1;
7788 next >>= bb->shift;
7789 sectors = next - s;
7792 write_seqlock_irq(&bb->lock);
7794 p = bb->page;
7795 lo = 0;
7796 hi = bb->count;
7797 /* Find the last range that starts at-or-before 's' */
7798 while (hi - lo > 1) {
7799 int mid = (lo + hi) / 2;
7800 sector_t a = BB_OFFSET(p[mid]);
7801 if (a <= s)
7802 lo = mid;
7803 else
7804 hi = mid;
7806 if (hi > lo && BB_OFFSET(p[lo]) > s)
7807 hi = lo;
7809 if (hi > lo) {
7810 /* we found a range that might merge with the start
7811 * of our new range
7813 sector_t a = BB_OFFSET(p[lo]);
7814 sector_t e = a + BB_LEN(p[lo]);
7815 int ack = BB_ACK(p[lo]);
7816 if (e >= s) {
7817 /* Yes, we can merge with a previous range */
7818 if (s == a && s + sectors >= e)
7819 /* new range covers old */
7820 ack = acknowledged;
7821 else
7822 ack = ack && acknowledged;
7824 if (e < s + sectors)
7825 e = s + sectors;
7826 if (e - a <= BB_MAX_LEN) {
7827 p[lo] = BB_MAKE(a, e-a, ack);
7828 s = e;
7829 } else {
7830 /* does not all fit in one range,
7831 * make p[lo] maximal
7833 if (BB_LEN(p[lo]) != BB_MAX_LEN)
7834 p[lo] = BB_MAKE(a, BB_MAX_LEN, ack);
7835 s = a + BB_MAX_LEN;
7837 sectors = e - s;
7840 if (sectors && hi < bb->count) {
7841 /* 'hi' points to the first range that starts after 's'.
7842 * Maybe we can merge with the start of that range */
7843 sector_t a = BB_OFFSET(p[hi]);
7844 sector_t e = a + BB_LEN(p[hi]);
7845 int ack = BB_ACK(p[hi]);
7846 if (a <= s + sectors) {
7847 /* merging is possible */
7848 if (e <= s + sectors) {
7849 /* full overlap */
7850 e = s + sectors;
7851 ack = acknowledged;
7852 } else
7853 ack = ack && acknowledged;
7855 a = s;
7856 if (e - a <= BB_MAX_LEN) {
7857 p[hi] = BB_MAKE(a, e-a, ack);
7858 s = e;
7859 } else {
7860 p[hi] = BB_MAKE(a, BB_MAX_LEN, ack);
7861 s = a + BB_MAX_LEN;
7863 sectors = e - s;
7864 lo = hi;
7865 hi++;
7868 if (sectors == 0 && hi < bb->count) {
7869 /* we might be able to combine lo and hi */
7870 /* Note: 's' is at the end of 'lo' */
7871 sector_t a = BB_OFFSET(p[hi]);
7872 int lolen = BB_LEN(p[lo]);
7873 int hilen = BB_LEN(p[hi]);
7874 int newlen = lolen + hilen - (s - a);
7875 if (s >= a && newlen < BB_MAX_LEN) {
7876 /* yes, we can combine them */
7877 int ack = BB_ACK(p[lo]) && BB_ACK(p[hi]);
7878 p[lo] = BB_MAKE(BB_OFFSET(p[lo]), newlen, ack);
7879 memmove(p + hi, p + hi + 1,
7880 (bb->count - hi - 1) * 8);
7881 bb->count--;
7884 while (sectors) {
7885 /* didn't merge (it all).
7886 * Need to add a range just before 'hi' */
7887 if (bb->count >= MD_MAX_BADBLOCKS) {
7888 /* No room for more */
7889 rv = 0;
7890 break;
7891 } else {
7892 int this_sectors = sectors;
7893 memmove(p + hi + 1, p + hi,
7894 (bb->count - hi) * 8);
7895 bb->count++;
7897 if (this_sectors > BB_MAX_LEN)
7898 this_sectors = BB_MAX_LEN;
7899 p[hi] = BB_MAKE(s, this_sectors, acknowledged);
7900 sectors -= this_sectors;
7901 s += this_sectors;
7905 bb->changed = 1;
7906 if (!acknowledged)
7907 bb->unacked_exist = 1;
7908 write_sequnlock_irq(&bb->lock);
7910 return rv;
7913 int rdev_set_badblocks(struct md_rdev *rdev, sector_t s, int sectors,
7914 int acknowledged)
7916 int rv = md_set_badblocks(&rdev->badblocks,
7917 s + rdev->data_offset, sectors, acknowledged);
7918 if (rv) {
7919 /* Make sure they get written out promptly */
7920 sysfs_notify_dirent_safe(rdev->sysfs_state);
7921 set_bit(MD_CHANGE_CLEAN, &rdev->mddev->flags);
7922 md_wakeup_thread(rdev->mddev->thread);
7924 return rv;
7926 EXPORT_SYMBOL_GPL(rdev_set_badblocks);
7929 * Remove a range of bad blocks from the table.
7930 * This may involve extending the table if we spilt a region,
7931 * but it must not fail. So if the table becomes full, we just
7932 * drop the remove request.
7934 static int md_clear_badblocks(struct badblocks *bb, sector_t s, int sectors)
7936 u64 *p;
7937 int lo, hi;
7938 sector_t target = s + sectors;
7939 int rv = 0;
7941 if (bb->shift > 0) {
7942 /* When clearing we round the start up and the end down.
7943 * This should not matter as the shift should align with
7944 * the block size and no rounding should ever be needed.
7945 * However it is better the think a block is bad when it
7946 * isn't than to think a block is not bad when it is.
7948 s += (1<<bb->shift) - 1;
7949 s >>= bb->shift;
7950 target >>= bb->shift;
7951 sectors = target - s;
7954 write_seqlock_irq(&bb->lock);
7956 p = bb->page;
7957 lo = 0;
7958 hi = bb->count;
7959 /* Find the last range that starts before 'target' */
7960 while (hi - lo > 1) {
7961 int mid = (lo + hi) / 2;
7962 sector_t a = BB_OFFSET(p[mid]);
7963 if (a < target)
7964 lo = mid;
7965 else
7966 hi = mid;
7968 if (hi > lo) {
7969 /* p[lo] is the last range that could overlap the
7970 * current range. Earlier ranges could also overlap,
7971 * but only this one can overlap the end of the range.
7973 if (BB_OFFSET(p[lo]) + BB_LEN(p[lo]) > target) {
7974 /* Partial overlap, leave the tail of this range */
7975 int ack = BB_ACK(p[lo]);
7976 sector_t a = BB_OFFSET(p[lo]);
7977 sector_t end = a + BB_LEN(p[lo]);
7979 if (a < s) {
7980 /* we need to split this range */
7981 if (bb->count >= MD_MAX_BADBLOCKS) {
7982 rv = 0;
7983 goto out;
7985 memmove(p+lo+1, p+lo, (bb->count - lo) * 8);
7986 bb->count++;
7987 p[lo] = BB_MAKE(a, s-a, ack);
7988 lo++;
7990 p[lo] = BB_MAKE(target, end - target, ack);
7991 /* there is no longer an overlap */
7992 hi = lo;
7993 lo--;
7995 while (lo >= 0 &&
7996 BB_OFFSET(p[lo]) + BB_LEN(p[lo]) > s) {
7997 /* This range does overlap */
7998 if (BB_OFFSET(p[lo]) < s) {
7999 /* Keep the early parts of this range. */
8000 int ack = BB_ACK(p[lo]);
8001 sector_t start = BB_OFFSET(p[lo]);
8002 p[lo] = BB_MAKE(start, s - start, ack);
8003 /* now low doesn't overlap, so.. */
8004 break;
8006 lo--;
8008 /* 'lo' is strictly before, 'hi' is strictly after,
8009 * anything between needs to be discarded
8011 if (hi - lo > 1) {
8012 memmove(p+lo+1, p+hi, (bb->count - hi) * 8);
8013 bb->count -= (hi - lo - 1);
8017 bb->changed = 1;
8018 out:
8019 write_sequnlock_irq(&bb->lock);
8020 return rv;
8023 int rdev_clear_badblocks(struct md_rdev *rdev, sector_t s, int sectors)
8025 return md_clear_badblocks(&rdev->badblocks,
8026 s + rdev->data_offset,
8027 sectors);
8029 EXPORT_SYMBOL_GPL(rdev_clear_badblocks);
8032 * Acknowledge all bad blocks in a list.
8033 * This only succeeds if ->changed is clear. It is used by
8034 * in-kernel metadata updates
8036 void md_ack_all_badblocks(struct badblocks *bb)
8038 if (bb->page == NULL || bb->changed)
8039 /* no point even trying */
8040 return;
8041 write_seqlock_irq(&bb->lock);
8043 if (bb->changed == 0) {
8044 u64 *p = bb->page;
8045 int i;
8046 for (i = 0; i < bb->count ; i++) {
8047 if (!BB_ACK(p[i])) {
8048 sector_t start = BB_OFFSET(p[i]);
8049 int len = BB_LEN(p[i]);
8050 p[i] = BB_MAKE(start, len, 1);
8053 bb->unacked_exist = 0;
8055 write_sequnlock_irq(&bb->lock);
8057 EXPORT_SYMBOL_GPL(md_ack_all_badblocks);
8059 /* sysfs access to bad-blocks list.
8060 * We present two files.
8061 * 'bad-blocks' lists sector numbers and lengths of ranges that
8062 * are recorded as bad. The list is truncated to fit within
8063 * the one-page limit of sysfs.
8064 * Writing "sector length" to this file adds an acknowledged
8065 * bad block list.
8066 * 'unacknowledged-bad-blocks' lists bad blocks that have not yet
8067 * been acknowledged. Writing to this file adds bad blocks
8068 * without acknowledging them. This is largely for testing.
8071 static ssize_t
8072 badblocks_show(struct badblocks *bb, char *page, int unack)
8074 size_t len;
8075 int i;
8076 u64 *p = bb->page;
8077 unsigned seq;
8079 if (bb->shift < 0)
8080 return 0;
8082 retry:
8083 seq = read_seqbegin(&bb->lock);
8085 len = 0;
8086 i = 0;
8088 while (len < PAGE_SIZE && i < bb->count) {
8089 sector_t s = BB_OFFSET(p[i]);
8090 unsigned int length = BB_LEN(p[i]);
8091 int ack = BB_ACK(p[i]);
8092 i++;
8094 if (unack && ack)
8095 continue;
8097 len += snprintf(page+len, PAGE_SIZE-len, "%llu %u\n",
8098 (unsigned long long)s << bb->shift,
8099 length << bb->shift);
8101 if (unack && len == 0)
8102 bb->unacked_exist = 0;
8104 if (read_seqretry(&bb->lock, seq))
8105 goto retry;
8107 return len;
8110 #define DO_DEBUG 1
8112 static ssize_t
8113 badblocks_store(struct badblocks *bb, const char *page, size_t len, int unack)
8115 unsigned long long sector;
8116 int length;
8117 char newline;
8118 #ifdef DO_DEBUG
8119 /* Allow clearing via sysfs *only* for testing/debugging.
8120 * Normally only a successful write may clear a badblock
8122 int clear = 0;
8123 if (page[0] == '-') {
8124 clear = 1;
8125 page++;
8127 #endif /* DO_DEBUG */
8129 switch (sscanf(page, "%llu %d%c", &sector, &length, &newline)) {
8130 case 3:
8131 if (newline != '\n')
8132 return -EINVAL;
8133 case 2:
8134 if (length <= 0)
8135 return -EINVAL;
8136 break;
8137 default:
8138 return -EINVAL;
8141 #ifdef DO_DEBUG
8142 if (clear) {
8143 md_clear_badblocks(bb, sector, length);
8144 return len;
8146 #endif /* DO_DEBUG */
8147 if (md_set_badblocks(bb, sector, length, !unack))
8148 return len;
8149 else
8150 return -ENOSPC;
8153 static int md_notify_reboot(struct notifier_block *this,
8154 unsigned long code, void *x)
8156 struct list_head *tmp;
8157 struct mddev *mddev;
8158 int need_delay = 0;
8160 if ((code == SYS_DOWN) || (code == SYS_HALT) || (code == SYS_POWER_OFF)) {
8162 printk(KERN_INFO "md: stopping all md devices.\n");
8164 for_each_mddev(mddev, tmp) {
8165 if (mddev_trylock(mddev)) {
8166 /* Force a switch to readonly even array
8167 * appears to still be in use. Hence
8168 * the '100'.
8170 md_set_readonly(mddev, 100);
8171 mddev_unlock(mddev);
8173 need_delay = 1;
8176 * certain more exotic SCSI devices are known to be
8177 * volatile wrt too early system reboots. While the
8178 * right place to handle this issue is the given
8179 * driver, we do want to have a safe RAID driver ...
8181 if (need_delay)
8182 mdelay(1000*1);
8184 return NOTIFY_DONE;
8187 static struct notifier_block md_notifier = {
8188 .notifier_call = md_notify_reboot,
8189 .next = NULL,
8190 .priority = INT_MAX, /* before any real devices */
8193 static void md_geninit(void)
8195 pr_debug("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
8197 proc_create("mdstat", S_IRUGO, NULL, &md_seq_fops);
8200 static int __init md_init(void)
8202 int ret = -ENOMEM;
8204 md_wq = alloc_workqueue("md", WQ_MEM_RECLAIM, 0);
8205 if (!md_wq)
8206 goto err_wq;
8208 md_misc_wq = alloc_workqueue("md_misc", 0, 0);
8209 if (!md_misc_wq)
8210 goto err_misc_wq;
8212 if ((ret = register_blkdev(MD_MAJOR, "md")) < 0)
8213 goto err_md;
8215 if ((ret = register_blkdev(0, "mdp")) < 0)
8216 goto err_mdp;
8217 mdp_major = ret;
8219 blk_register_region(MKDEV(MD_MAJOR, 0), 1UL<<MINORBITS, THIS_MODULE,
8220 md_probe, NULL, NULL);
8221 blk_register_region(MKDEV(mdp_major, 0), 1UL<<MINORBITS, THIS_MODULE,
8222 md_probe, NULL, NULL);
8224 register_reboot_notifier(&md_notifier);
8225 raid_table_header = register_sysctl_table(raid_root_table);
8227 md_geninit();
8228 return 0;
8230 err_mdp:
8231 unregister_blkdev(MD_MAJOR, "md");
8232 err_md:
8233 destroy_workqueue(md_misc_wq);
8234 err_misc_wq:
8235 destroy_workqueue(md_wq);
8236 err_wq:
8237 return ret;
8240 #ifndef MODULE
8243 * Searches all registered partitions for autorun RAID arrays
8244 * at boot time.
8247 static LIST_HEAD(all_detected_devices);
8248 struct detected_devices_node {
8249 struct list_head list;
8250 dev_t dev;
8253 void md_autodetect_dev(dev_t dev)
8255 struct detected_devices_node *node_detected_dev;
8257 node_detected_dev = kzalloc(sizeof(*node_detected_dev), GFP_KERNEL);
8258 if (node_detected_dev) {
8259 node_detected_dev->dev = dev;
8260 list_add_tail(&node_detected_dev->list, &all_detected_devices);
8261 } else {
8262 printk(KERN_CRIT "md: md_autodetect_dev: kzalloc failed"
8263 ", skipping dev(%d,%d)\n", MAJOR(dev), MINOR(dev));
8268 static void autostart_arrays(int part)
8270 struct md_rdev *rdev;
8271 struct detected_devices_node *node_detected_dev;
8272 dev_t dev;
8273 int i_scanned, i_passed;
8275 i_scanned = 0;
8276 i_passed = 0;
8278 printk(KERN_INFO "md: Autodetecting RAID arrays.\n");
8280 while (!list_empty(&all_detected_devices) && i_scanned < INT_MAX) {
8281 i_scanned++;
8282 node_detected_dev = list_entry(all_detected_devices.next,
8283 struct detected_devices_node, list);
8284 list_del(&node_detected_dev->list);
8285 dev = node_detected_dev->dev;
8286 kfree(node_detected_dev);
8287 rdev = md_import_device(dev,0, 90);
8288 if (IS_ERR(rdev))
8289 continue;
8291 if (test_bit(Faulty, &rdev->flags)) {
8292 MD_BUG();
8293 continue;
8295 set_bit(AutoDetected, &rdev->flags);
8296 list_add(&rdev->same_set, &pending_raid_disks);
8297 i_passed++;
8300 printk(KERN_INFO "md: Scanned %d and added %d devices.\n",
8301 i_scanned, i_passed);
8303 autorun_devices(part);
8306 #endif /* !MODULE */
8308 static __exit void md_exit(void)
8310 struct mddev *mddev;
8311 struct list_head *tmp;
8313 blk_unregister_region(MKDEV(MD_MAJOR,0), 1U << MINORBITS);
8314 blk_unregister_region(MKDEV(mdp_major,0), 1U << MINORBITS);
8316 unregister_blkdev(MD_MAJOR,"md");
8317 unregister_blkdev(mdp_major, "mdp");
8318 unregister_reboot_notifier(&md_notifier);
8319 unregister_sysctl_table(raid_table_header);
8320 remove_proc_entry("mdstat", NULL);
8321 for_each_mddev(mddev, tmp) {
8322 export_array(mddev);
8323 mddev->hold_active = 0;
8325 destroy_workqueue(md_misc_wq);
8326 destroy_workqueue(md_wq);
8329 subsys_initcall(md_init);
8330 module_exit(md_exit)
8332 static int get_ro(char *buffer, struct kernel_param *kp)
8334 return sprintf(buffer, "%d", start_readonly);
8336 static int set_ro(const char *val, struct kernel_param *kp)
8338 char *e;
8339 int num = simple_strtoul(val, &e, 10);
8340 if (*val && (*e == '\0' || *e == '\n')) {
8341 start_readonly = num;
8342 return 0;
8344 return -EINVAL;
8347 module_param_call(start_ro, set_ro, get_ro, NULL, S_IRUSR|S_IWUSR);
8348 module_param(start_dirty_degraded, int, S_IRUGO|S_IWUSR);
8350 module_param_call(new_array, add_named_array, NULL, NULL, S_IWUSR);
8352 EXPORT_SYMBOL(register_md_personality);
8353 EXPORT_SYMBOL(unregister_md_personality);
8354 EXPORT_SYMBOL(md_error);
8355 EXPORT_SYMBOL(md_done_sync);
8356 EXPORT_SYMBOL(md_write_start);
8357 EXPORT_SYMBOL(md_write_end);
8358 EXPORT_SYMBOL(md_register_thread);
8359 EXPORT_SYMBOL(md_unregister_thread);
8360 EXPORT_SYMBOL(md_wakeup_thread);
8361 EXPORT_SYMBOL(md_check_recovery);
8362 MODULE_LICENSE("GPL");
8363 MODULE_DESCRIPTION("MD RAID framework");
8364 MODULE_ALIAS("md");
8365 MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR);