xref: /openbmc/linux/drivers/md/md.c (revision d2999e1b)
1 /*
2    md.c : Multiple Devices driver for Linux
3 	  Copyright (C) 1998, 1999, 2000 Ingo Molnar
4 
5      completely rewritten, based on the MD driver code from Marc Zyngier
6 
7    Changes:
8 
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>
16 
17    - lots of fixes and improvements to the RAID1/RAID5 and generic
18      RAID code (such as request based resynchronization):
19 
20      Neil Brown <neilb@cse.unsw.edu.au>.
21 
22    - persistent bitmap code
23      Copyright (C) 2003-2004, Paul Clements, SteelEye Technology, Inc.
24 
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.
29 
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.
33 */
34 
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"
56 
57 #ifndef MODULE
58 static void autostart_arrays(int part);
59 #endif
60 
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.
65  */
66 static LIST_HEAD(pers_list);
67 static DEFINE_SPINLOCK(pers_lock);
68 
69 static void md_print_devices(void);
70 
71 static DECLARE_WAIT_QUEUE_HEAD(resync_wait);
72 static struct workqueue_struct *md_wq;
73 static struct workqueue_struct *md_misc_wq;
74 
75 static int remove_and_add_spares(struct mddev *mddev,
76 				 struct md_rdev *this);
77 
78 #define MD_BUG(x...) { printk("md: bug in file %s, line %d\n", __FILE__, __LINE__); md_print_devices(); }
79 
80 /*
81  * Default number of read corrections we'll attempt on an rdev
82  * before ejecting it from the array. We divide the read error
83  * count by 2 for every hour elapsed between read errors.
84  */
85 #define MD_DEFAULT_MAX_CORRECTED_READ_ERRORS 20
86 /*
87  * Current RAID-1,4,5 parallel reconstruction 'guaranteed speed limit'
88  * is 1000 KB/sec, so the extra system load does not show up that much.
89  * Increase it if you want to have more _guaranteed_ speed. Note that
90  * the RAID driver will use the maximum available bandwidth if the IO
91  * subsystem is idle. There is also an 'absolute maximum' reconstruction
92  * speed limit - in case reconstruction slows down your system despite
93  * idle IO detection.
94  *
95  * you can change it via /proc/sys/dev/raid/speed_limit_min and _max.
96  * or /sys/block/mdX/md/sync_speed_{min,max}
97  */
98 
99 static int sysctl_speed_limit_min = 1000;
100 static int sysctl_speed_limit_max = 200000;
101 static inline int speed_min(struct mddev *mddev)
102 {
103 	return mddev->sync_speed_min ?
104 		mddev->sync_speed_min : sysctl_speed_limit_min;
105 }
106 
107 static inline int speed_max(struct mddev *mddev)
108 {
109 	return mddev->sync_speed_max ?
110 		mddev->sync_speed_max : sysctl_speed_limit_max;
111 }
112 
113 static struct ctl_table_header *raid_table_header;
114 
115 static struct ctl_table raid_table[] = {
116 	{
117 		.procname	= "speed_limit_min",
118 		.data		= &sysctl_speed_limit_min,
119 		.maxlen		= sizeof(int),
120 		.mode		= S_IRUGO|S_IWUSR,
121 		.proc_handler	= proc_dointvec,
122 	},
123 	{
124 		.procname	= "speed_limit_max",
125 		.data		= &sysctl_speed_limit_max,
126 		.maxlen		= sizeof(int),
127 		.mode		= S_IRUGO|S_IWUSR,
128 		.proc_handler	= proc_dointvec,
129 	},
130 	{ }
131 };
132 
133 static struct ctl_table raid_dir_table[] = {
134 	{
135 		.procname	= "raid",
136 		.maxlen		= 0,
137 		.mode		= S_IRUGO|S_IXUGO,
138 		.child		= raid_table,
139 	},
140 	{ }
141 };
142 
143 static struct ctl_table raid_root_table[] = {
144 	{
145 		.procname	= "dev",
146 		.maxlen		= 0,
147 		.mode		= 0555,
148 		.child		= raid_dir_table,
149 	},
150 	{  }
151 };
152 
153 static const struct block_device_operations md_fops;
154 
155 static int start_readonly;
156 
157 /* bio_clone_mddev
158  * like bio_clone, but with a local bio set
159  */
160 
161 struct bio *bio_alloc_mddev(gfp_t gfp_mask, int nr_iovecs,
162 			    struct mddev *mddev)
163 {
164 	struct bio *b;
165 
166 	if (!mddev || !mddev->bio_set)
167 		return bio_alloc(gfp_mask, nr_iovecs);
168 
169 	b = bio_alloc_bioset(gfp_mask, nr_iovecs, mddev->bio_set);
170 	if (!b)
171 		return NULL;
172 	return b;
173 }
174 EXPORT_SYMBOL_GPL(bio_alloc_mddev);
175 
176 struct bio *bio_clone_mddev(struct bio *bio, gfp_t gfp_mask,
177 			    struct mddev *mddev)
178 {
179 	if (!mddev || !mddev->bio_set)
180 		return bio_clone(bio, gfp_mask);
181 
182 	return bio_clone_bioset(bio, gfp_mask, mddev->bio_set);
183 }
184 EXPORT_SYMBOL_GPL(bio_clone_mddev);
185 
186 /*
187  * We have a system wide 'event count' that is incremented
188  * on any 'interesting' event, and readers of /proc/mdstat
189  * can use 'poll' or 'select' to find out when the event
190  * count increases.
191  *
192  * Events are:
193  *  start array, stop array, error, add device, remove device,
194  *  start build, activate spare
195  */
196 static DECLARE_WAIT_QUEUE_HEAD(md_event_waiters);
197 static atomic_t md_event_count;
198 void md_new_event(struct mddev *mddev)
199 {
200 	atomic_inc(&md_event_count);
201 	wake_up(&md_event_waiters);
202 }
203 EXPORT_SYMBOL_GPL(md_new_event);
204 
205 /* Alternate version that can be called from interrupts
206  * when calling sysfs_notify isn't needed.
207  */
208 static void md_new_event_inintr(struct mddev *mddev)
209 {
210 	atomic_inc(&md_event_count);
211 	wake_up(&md_event_waiters);
212 }
213 
214 /*
215  * Enables to iterate over all existing md arrays
216  * all_mddevs_lock protects this list.
217  */
218 static LIST_HEAD(all_mddevs);
219 static DEFINE_SPINLOCK(all_mddevs_lock);
220 
221 
222 /*
223  * iterates through all used mddevs in the system.
224  * We take care to grab the all_mddevs_lock whenever navigating
225  * the list, and to always hold a refcount when unlocked.
226  * Any code which breaks out of this loop while own
227  * a reference to the current mddev and must mddev_put it.
228  */
229 #define for_each_mddev(_mddev,_tmp)					\
230 									\
231 	for (({ spin_lock(&all_mddevs_lock); 				\
232 		_tmp = all_mddevs.next;					\
233 		_mddev = NULL;});					\
234 	     ({ if (_tmp != &all_mddevs)				\
235 			mddev_get(list_entry(_tmp, struct mddev, all_mddevs));\
236 		spin_unlock(&all_mddevs_lock);				\
237 		if (_mddev) mddev_put(_mddev);				\
238 		_mddev = list_entry(_tmp, struct mddev, all_mddevs);	\
239 		_tmp != &all_mddevs;});					\
240 	     ({ spin_lock(&all_mddevs_lock);				\
241 		_tmp = _tmp->next;})					\
242 		)
243 
244 
245 /* Rather than calling directly into the personality make_request function,
246  * IO requests come here first so that we can check if the device is
247  * being suspended pending a reconfiguration.
248  * We hold a refcount over the call to ->make_request.  By the time that
249  * call has finished, the bio has been linked into some internal structure
250  * and so is visible to ->quiesce(), so we don't need the refcount any more.
251  */
252 static void md_make_request(struct request_queue *q, struct bio *bio)
253 {
254 	const int rw = bio_data_dir(bio);
255 	struct mddev *mddev = q->queuedata;
256 	int cpu;
257 	unsigned int sectors;
258 
259 	if (mddev == NULL || mddev->pers == NULL
260 	    || !mddev->ready) {
261 		bio_io_error(bio);
262 		return;
263 	}
264 	if (mddev->ro == 1 && unlikely(rw == WRITE)) {
265 		bio_endio(bio, bio_sectors(bio) == 0 ? 0 : -EROFS);
266 		return;
267 	}
268 	smp_rmb(); /* Ensure implications of  'active' are visible */
269 	rcu_read_lock();
270 	if (mddev->suspended) {
271 		DEFINE_WAIT(__wait);
272 		for (;;) {
273 			prepare_to_wait(&mddev->sb_wait, &__wait,
274 					TASK_UNINTERRUPTIBLE);
275 			if (!mddev->suspended)
276 				break;
277 			rcu_read_unlock();
278 			schedule();
279 			rcu_read_lock();
280 		}
281 		finish_wait(&mddev->sb_wait, &__wait);
282 	}
283 	atomic_inc(&mddev->active_io);
284 	rcu_read_unlock();
285 
286 	/*
287 	 * save the sectors now since our bio can
288 	 * go away inside make_request
289 	 */
290 	sectors = bio_sectors(bio);
291 	mddev->pers->make_request(mddev, bio);
292 
293 	cpu = part_stat_lock();
294 	part_stat_inc(cpu, &mddev->gendisk->part0, ios[rw]);
295 	part_stat_add(cpu, &mddev->gendisk->part0, sectors[rw], sectors);
296 	part_stat_unlock();
297 
298 	if (atomic_dec_and_test(&mddev->active_io) && mddev->suspended)
299 		wake_up(&mddev->sb_wait);
300 }
301 
302 /* mddev_suspend makes sure no new requests are submitted
303  * to the device, and that any requests that have been submitted
304  * are completely handled.
305  * Once ->stop is called and completes, the module will be completely
306  * unused.
307  */
308 void mddev_suspend(struct mddev *mddev)
309 {
310 	BUG_ON(mddev->suspended);
311 	mddev->suspended = 1;
312 	synchronize_rcu();
313 	wait_event(mddev->sb_wait, atomic_read(&mddev->active_io) == 0);
314 	mddev->pers->quiesce(mddev, 1);
315 
316 	del_timer_sync(&mddev->safemode_timer);
317 }
318 EXPORT_SYMBOL_GPL(mddev_suspend);
319 
320 void mddev_resume(struct mddev *mddev)
321 {
322 	mddev->suspended = 0;
323 	wake_up(&mddev->sb_wait);
324 	mddev->pers->quiesce(mddev, 0);
325 
326 	set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
327 	md_wakeup_thread(mddev->thread);
328 	md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
329 }
330 EXPORT_SYMBOL_GPL(mddev_resume);
331 
332 int mddev_congested(struct mddev *mddev, int bits)
333 {
334 	return mddev->suspended;
335 }
336 EXPORT_SYMBOL(mddev_congested);
337 
338 /*
339  * Generic flush handling for md
340  */
341 
342 static void md_end_flush(struct bio *bio, int err)
343 {
344 	struct md_rdev *rdev = bio->bi_private;
345 	struct mddev *mddev = rdev->mddev;
346 
347 	rdev_dec_pending(rdev, mddev);
348 
349 	if (atomic_dec_and_test(&mddev->flush_pending)) {
350 		/* The pre-request flush has finished */
351 		queue_work(md_wq, &mddev->flush_work);
352 	}
353 	bio_put(bio);
354 }
355 
356 static void md_submit_flush_data(struct work_struct *ws);
357 
358 static void submit_flushes(struct work_struct *ws)
359 {
360 	struct mddev *mddev = container_of(ws, struct mddev, flush_work);
361 	struct md_rdev *rdev;
362 
363 	INIT_WORK(&mddev->flush_work, md_submit_flush_data);
364 	atomic_set(&mddev->flush_pending, 1);
365 	rcu_read_lock();
366 	rdev_for_each_rcu(rdev, mddev)
367 		if (rdev->raid_disk >= 0 &&
368 		    !test_bit(Faulty, &rdev->flags)) {
369 			/* Take two references, one is dropped
370 			 * when request finishes, one after
371 			 * we reclaim rcu_read_lock
372 			 */
373 			struct bio *bi;
374 			atomic_inc(&rdev->nr_pending);
375 			atomic_inc(&rdev->nr_pending);
376 			rcu_read_unlock();
377 			bi = bio_alloc_mddev(GFP_NOIO, 0, mddev);
378 			bi->bi_end_io = md_end_flush;
379 			bi->bi_private = rdev;
380 			bi->bi_bdev = rdev->bdev;
381 			atomic_inc(&mddev->flush_pending);
382 			submit_bio(WRITE_FLUSH, bi);
383 			rcu_read_lock();
384 			rdev_dec_pending(rdev, mddev);
385 		}
386 	rcu_read_unlock();
387 	if (atomic_dec_and_test(&mddev->flush_pending))
388 		queue_work(md_wq, &mddev->flush_work);
389 }
390 
391 static void md_submit_flush_data(struct work_struct *ws)
392 {
393 	struct mddev *mddev = container_of(ws, struct mddev, flush_work);
394 	struct bio *bio = mddev->flush_bio;
395 
396 	if (bio->bi_iter.bi_size == 0)
397 		/* an empty barrier - all done */
398 		bio_endio(bio, 0);
399 	else {
400 		bio->bi_rw &= ~REQ_FLUSH;
401 		mddev->pers->make_request(mddev, bio);
402 	}
403 
404 	mddev->flush_bio = NULL;
405 	wake_up(&mddev->sb_wait);
406 }
407 
408 void md_flush_request(struct mddev *mddev, struct bio *bio)
409 {
410 	spin_lock_irq(&mddev->write_lock);
411 	wait_event_lock_irq(mddev->sb_wait,
412 			    !mddev->flush_bio,
413 			    mddev->write_lock);
414 	mddev->flush_bio = bio;
415 	spin_unlock_irq(&mddev->write_lock);
416 
417 	INIT_WORK(&mddev->flush_work, submit_flushes);
418 	queue_work(md_wq, &mddev->flush_work);
419 }
420 EXPORT_SYMBOL(md_flush_request);
421 
422 void md_unplug(struct blk_plug_cb *cb, bool from_schedule)
423 {
424 	struct mddev *mddev = cb->data;
425 	md_wakeup_thread(mddev->thread);
426 	kfree(cb);
427 }
428 EXPORT_SYMBOL(md_unplug);
429 
430 static inline struct mddev *mddev_get(struct mddev *mddev)
431 {
432 	atomic_inc(&mddev->active);
433 	return mddev;
434 }
435 
436 static void mddev_delayed_delete(struct work_struct *ws);
437 
438 static void mddev_put(struct mddev *mddev)
439 {
440 	struct bio_set *bs = NULL;
441 
442 	if (!atomic_dec_and_lock(&mddev->active, &all_mddevs_lock))
443 		return;
444 	if (!mddev->raid_disks && list_empty(&mddev->disks) &&
445 	    mddev->ctime == 0 && !mddev->hold_active) {
446 		/* Array is not configured at all, and not held active,
447 		 * so destroy it */
448 		list_del_init(&mddev->all_mddevs);
449 		bs = mddev->bio_set;
450 		mddev->bio_set = NULL;
451 		if (mddev->gendisk) {
452 			/* We did a probe so need to clean up.  Call
453 			 * queue_work inside the spinlock so that
454 			 * flush_workqueue() after mddev_find will
455 			 * succeed in waiting for the work to be done.
456 			 */
457 			INIT_WORK(&mddev->del_work, mddev_delayed_delete);
458 			queue_work(md_misc_wq, &mddev->del_work);
459 		} else
460 			kfree(mddev);
461 	}
462 	spin_unlock(&all_mddevs_lock);
463 	if (bs)
464 		bioset_free(bs);
465 }
466 
467 void mddev_init(struct mddev *mddev)
468 {
469 	mutex_init(&mddev->open_mutex);
470 	mutex_init(&mddev->reconfig_mutex);
471 	mutex_init(&mddev->bitmap_info.mutex);
472 	INIT_LIST_HEAD(&mddev->disks);
473 	INIT_LIST_HEAD(&mddev->all_mddevs);
474 	init_timer(&mddev->safemode_timer);
475 	atomic_set(&mddev->active, 1);
476 	atomic_set(&mddev->openers, 0);
477 	atomic_set(&mddev->active_io, 0);
478 	spin_lock_init(&mddev->write_lock);
479 	atomic_set(&mddev->flush_pending, 0);
480 	init_waitqueue_head(&mddev->sb_wait);
481 	init_waitqueue_head(&mddev->recovery_wait);
482 	mddev->reshape_position = MaxSector;
483 	mddev->reshape_backwards = 0;
484 	mddev->last_sync_action = "none";
485 	mddev->resync_min = 0;
486 	mddev->resync_max = MaxSector;
487 	mddev->level = LEVEL_NONE;
488 }
489 EXPORT_SYMBOL_GPL(mddev_init);
490 
491 static struct mddev * mddev_find(dev_t unit)
492 {
493 	struct mddev *mddev, *new = NULL;
494 
495 	if (unit && MAJOR(unit) != MD_MAJOR)
496 		unit &= ~((1<<MdpMinorShift)-1);
497 
498  retry:
499 	spin_lock(&all_mddevs_lock);
500 
501 	if (unit) {
502 		list_for_each_entry(mddev, &all_mddevs, all_mddevs)
503 			if (mddev->unit == unit) {
504 				mddev_get(mddev);
505 				spin_unlock(&all_mddevs_lock);
506 				kfree(new);
507 				return mddev;
508 			}
509 
510 		if (new) {
511 			list_add(&new->all_mddevs, &all_mddevs);
512 			spin_unlock(&all_mddevs_lock);
513 			new->hold_active = UNTIL_IOCTL;
514 			return new;
515 		}
516 	} else if (new) {
517 		/* find an unused unit number */
518 		static int next_minor = 512;
519 		int start = next_minor;
520 		int is_free = 0;
521 		int dev = 0;
522 		while (!is_free) {
523 			dev = MKDEV(MD_MAJOR, next_minor);
524 			next_minor++;
525 			if (next_minor > MINORMASK)
526 				next_minor = 0;
527 			if (next_minor == start) {
528 				/* Oh dear, all in use. */
529 				spin_unlock(&all_mddevs_lock);
530 				kfree(new);
531 				return NULL;
532 			}
533 
534 			is_free = 1;
535 			list_for_each_entry(mddev, &all_mddevs, all_mddevs)
536 				if (mddev->unit == dev) {
537 					is_free = 0;
538 					break;
539 				}
540 		}
541 		new->unit = dev;
542 		new->md_minor = MINOR(dev);
543 		new->hold_active = UNTIL_STOP;
544 		list_add(&new->all_mddevs, &all_mddevs);
545 		spin_unlock(&all_mddevs_lock);
546 		return new;
547 	}
548 	spin_unlock(&all_mddevs_lock);
549 
550 	new = kzalloc(sizeof(*new), GFP_KERNEL);
551 	if (!new)
552 		return NULL;
553 
554 	new->unit = unit;
555 	if (MAJOR(unit) == MD_MAJOR)
556 		new->md_minor = MINOR(unit);
557 	else
558 		new->md_minor = MINOR(unit) >> MdpMinorShift;
559 
560 	mddev_init(new);
561 
562 	goto retry;
563 }
564 
565 static inline int __must_check mddev_lock(struct mddev * mddev)
566 {
567 	return mutex_lock_interruptible(&mddev->reconfig_mutex);
568 }
569 
570 /* Sometimes we need to take the lock in a situation where
571  * failure due to interrupts is not acceptable.
572  */
573 static inline void mddev_lock_nointr(struct mddev * mddev)
574 {
575 	mutex_lock(&mddev->reconfig_mutex);
576 }
577 
578 static inline int mddev_is_locked(struct mddev *mddev)
579 {
580 	return mutex_is_locked(&mddev->reconfig_mutex);
581 }
582 
583 static inline int mddev_trylock(struct mddev * mddev)
584 {
585 	return mutex_trylock(&mddev->reconfig_mutex);
586 }
587 
588 static struct attribute_group md_redundancy_group;
589 
590 static void mddev_unlock(struct mddev * mddev)
591 {
592 	if (mddev->to_remove) {
593 		/* These cannot be removed under reconfig_mutex as
594 		 * an access to the files will try to take reconfig_mutex
595 		 * while holding the file unremovable, which leads to
596 		 * a deadlock.
597 		 * So hold set sysfs_active while the remove in happeing,
598 		 * and anything else which might set ->to_remove or my
599 		 * otherwise change the sysfs namespace will fail with
600 		 * -EBUSY if sysfs_active is still set.
601 		 * We set sysfs_active under reconfig_mutex and elsewhere
602 		 * test it under the same mutex to ensure its correct value
603 		 * is seen.
604 		 */
605 		struct attribute_group *to_remove = mddev->to_remove;
606 		mddev->to_remove = NULL;
607 		mddev->sysfs_active = 1;
608 		mutex_unlock(&mddev->reconfig_mutex);
609 
610 		if (mddev->kobj.sd) {
611 			if (to_remove != &md_redundancy_group)
612 				sysfs_remove_group(&mddev->kobj, to_remove);
613 			if (mddev->pers == NULL ||
614 			    mddev->pers->sync_request == NULL) {
615 				sysfs_remove_group(&mddev->kobj, &md_redundancy_group);
616 				if (mddev->sysfs_action)
617 					sysfs_put(mddev->sysfs_action);
618 				mddev->sysfs_action = NULL;
619 			}
620 		}
621 		mddev->sysfs_active = 0;
622 	} else
623 		mutex_unlock(&mddev->reconfig_mutex);
624 
625 	/* As we've dropped the mutex we need a spinlock to
626 	 * make sure the thread doesn't disappear
627 	 */
628 	spin_lock(&pers_lock);
629 	md_wakeup_thread(mddev->thread);
630 	spin_unlock(&pers_lock);
631 }
632 
633 static struct md_rdev * find_rdev_nr(struct mddev *mddev, int nr)
634 {
635 	struct md_rdev *rdev;
636 
637 	rdev_for_each(rdev, mddev)
638 		if (rdev->desc_nr == nr)
639 			return rdev;
640 
641 	return NULL;
642 }
643 
644 static struct md_rdev *find_rdev_nr_rcu(struct mddev *mddev, int nr)
645 {
646 	struct md_rdev *rdev;
647 
648 	rdev_for_each_rcu(rdev, mddev)
649 		if (rdev->desc_nr == nr)
650 			return rdev;
651 
652 	return NULL;
653 }
654 
655 static struct md_rdev *find_rdev(struct mddev *mddev, dev_t dev)
656 {
657 	struct md_rdev *rdev;
658 
659 	rdev_for_each(rdev, mddev)
660 		if (rdev->bdev->bd_dev == dev)
661 			return rdev;
662 
663 	return NULL;
664 }
665 
666 static struct md_rdev *find_rdev_rcu(struct mddev *mddev, dev_t dev)
667 {
668 	struct md_rdev *rdev;
669 
670 	rdev_for_each_rcu(rdev, mddev)
671 		if (rdev->bdev->bd_dev == dev)
672 			return rdev;
673 
674 	return NULL;
675 }
676 
677 static struct md_personality *find_pers(int level, char *clevel)
678 {
679 	struct md_personality *pers;
680 	list_for_each_entry(pers, &pers_list, list) {
681 		if (level != LEVEL_NONE && pers->level == level)
682 			return pers;
683 		if (strcmp(pers->name, clevel)==0)
684 			return pers;
685 	}
686 	return NULL;
687 }
688 
689 /* return the offset of the super block in 512byte sectors */
690 static inline sector_t calc_dev_sboffset(struct md_rdev *rdev)
691 {
692 	sector_t num_sectors = i_size_read(rdev->bdev->bd_inode) / 512;
693 	return MD_NEW_SIZE_SECTORS(num_sectors);
694 }
695 
696 static int alloc_disk_sb(struct md_rdev * rdev)
697 {
698 	if (rdev->sb_page)
699 		MD_BUG();
700 
701 	rdev->sb_page = alloc_page(GFP_KERNEL);
702 	if (!rdev->sb_page) {
703 		printk(KERN_ALERT "md: out of memory.\n");
704 		return -ENOMEM;
705 	}
706 
707 	return 0;
708 }
709 
710 void md_rdev_clear(struct md_rdev *rdev)
711 {
712 	if (rdev->sb_page) {
713 		put_page(rdev->sb_page);
714 		rdev->sb_loaded = 0;
715 		rdev->sb_page = NULL;
716 		rdev->sb_start = 0;
717 		rdev->sectors = 0;
718 	}
719 	if (rdev->bb_page) {
720 		put_page(rdev->bb_page);
721 		rdev->bb_page = NULL;
722 	}
723 	kfree(rdev->badblocks.page);
724 	rdev->badblocks.page = NULL;
725 }
726 EXPORT_SYMBOL_GPL(md_rdev_clear);
727 
728 static void super_written(struct bio *bio, int error)
729 {
730 	struct md_rdev *rdev = bio->bi_private;
731 	struct mddev *mddev = rdev->mddev;
732 
733 	if (error || !test_bit(BIO_UPTODATE, &bio->bi_flags)) {
734 		printk("md: super_written gets error=%d, uptodate=%d\n",
735 		       error, test_bit(BIO_UPTODATE, &bio->bi_flags));
736 		WARN_ON(test_bit(BIO_UPTODATE, &bio->bi_flags));
737 		md_error(mddev, rdev);
738 	}
739 
740 	if (atomic_dec_and_test(&mddev->pending_writes))
741 		wake_up(&mddev->sb_wait);
742 	bio_put(bio);
743 }
744 
745 void md_super_write(struct mddev *mddev, struct md_rdev *rdev,
746 		   sector_t sector, int size, struct page *page)
747 {
748 	/* write first size bytes of page to sector of rdev
749 	 * Increment mddev->pending_writes before returning
750 	 * and decrement it on completion, waking up sb_wait
751 	 * if zero is reached.
752 	 * If an error occurred, call md_error
753 	 */
754 	struct bio *bio = bio_alloc_mddev(GFP_NOIO, 1, mddev);
755 
756 	bio->bi_bdev = rdev->meta_bdev ? rdev->meta_bdev : rdev->bdev;
757 	bio->bi_iter.bi_sector = sector;
758 	bio_add_page(bio, page, size, 0);
759 	bio->bi_private = rdev;
760 	bio->bi_end_io = super_written;
761 
762 	atomic_inc(&mddev->pending_writes);
763 	submit_bio(WRITE_FLUSH_FUA, bio);
764 }
765 
766 void md_super_wait(struct mddev *mddev)
767 {
768 	/* wait for all superblock writes that were scheduled to complete */
769 	DEFINE_WAIT(wq);
770 	for(;;) {
771 		prepare_to_wait(&mddev->sb_wait, &wq, TASK_UNINTERRUPTIBLE);
772 		if (atomic_read(&mddev->pending_writes)==0)
773 			break;
774 		schedule();
775 	}
776 	finish_wait(&mddev->sb_wait, &wq);
777 }
778 
779 int sync_page_io(struct md_rdev *rdev, sector_t sector, int size,
780 		 struct page *page, int rw, bool metadata_op)
781 {
782 	struct bio *bio = bio_alloc_mddev(GFP_NOIO, 1, rdev->mddev);
783 	int ret;
784 
785 	bio->bi_bdev = (metadata_op && rdev->meta_bdev) ?
786 		rdev->meta_bdev : rdev->bdev;
787 	if (metadata_op)
788 		bio->bi_iter.bi_sector = sector + rdev->sb_start;
789 	else if (rdev->mddev->reshape_position != MaxSector &&
790 		 (rdev->mddev->reshape_backwards ==
791 		  (sector >= rdev->mddev->reshape_position)))
792 		bio->bi_iter.bi_sector = sector + rdev->new_data_offset;
793 	else
794 		bio->bi_iter.bi_sector = sector + rdev->data_offset;
795 	bio_add_page(bio, page, size, 0);
796 	submit_bio_wait(rw, bio);
797 
798 	ret = test_bit(BIO_UPTODATE, &bio->bi_flags);
799 	bio_put(bio);
800 	return ret;
801 }
802 EXPORT_SYMBOL_GPL(sync_page_io);
803 
804 static int read_disk_sb(struct md_rdev * rdev, int size)
805 {
806 	char b[BDEVNAME_SIZE];
807 	if (!rdev->sb_page) {
808 		MD_BUG();
809 		return -EINVAL;
810 	}
811 	if (rdev->sb_loaded)
812 		return 0;
813 
814 
815 	if (!sync_page_io(rdev, 0, size, rdev->sb_page, READ, true))
816 		goto fail;
817 	rdev->sb_loaded = 1;
818 	return 0;
819 
820 fail:
821 	printk(KERN_WARNING "md: disabled device %s, could not read superblock.\n",
822 		bdevname(rdev->bdev,b));
823 	return -EINVAL;
824 }
825 
826 static int uuid_equal(mdp_super_t *sb1, mdp_super_t *sb2)
827 {
828 	return 	sb1->set_uuid0 == sb2->set_uuid0 &&
829 		sb1->set_uuid1 == sb2->set_uuid1 &&
830 		sb1->set_uuid2 == sb2->set_uuid2 &&
831 		sb1->set_uuid3 == sb2->set_uuid3;
832 }
833 
834 static int sb_equal(mdp_super_t *sb1, mdp_super_t *sb2)
835 {
836 	int ret;
837 	mdp_super_t *tmp1, *tmp2;
838 
839 	tmp1 = kmalloc(sizeof(*tmp1),GFP_KERNEL);
840 	tmp2 = kmalloc(sizeof(*tmp2),GFP_KERNEL);
841 
842 	if (!tmp1 || !tmp2) {
843 		ret = 0;
844 		printk(KERN_INFO "md.c sb_equal(): failed to allocate memory!\n");
845 		goto abort;
846 	}
847 
848 	*tmp1 = *sb1;
849 	*tmp2 = *sb2;
850 
851 	/*
852 	 * nr_disks is not constant
853 	 */
854 	tmp1->nr_disks = 0;
855 	tmp2->nr_disks = 0;
856 
857 	ret = (memcmp(tmp1, tmp2, MD_SB_GENERIC_CONSTANT_WORDS * 4) == 0);
858 abort:
859 	kfree(tmp1);
860 	kfree(tmp2);
861 	return ret;
862 }
863 
864 
865 static u32 md_csum_fold(u32 csum)
866 {
867 	csum = (csum & 0xffff) + (csum >> 16);
868 	return (csum & 0xffff) + (csum >> 16);
869 }
870 
871 static unsigned int calc_sb_csum(mdp_super_t * sb)
872 {
873 	u64 newcsum = 0;
874 	u32 *sb32 = (u32*)sb;
875 	int i;
876 	unsigned int disk_csum, csum;
877 
878 	disk_csum = sb->sb_csum;
879 	sb->sb_csum = 0;
880 
881 	for (i = 0; i < MD_SB_BYTES/4 ; i++)
882 		newcsum += sb32[i];
883 	csum = (newcsum & 0xffffffff) + (newcsum>>32);
884 
885 
886 #ifdef CONFIG_ALPHA
887 	/* This used to use csum_partial, which was wrong for several
888 	 * reasons including that different results are returned on
889 	 * different architectures.  It isn't critical that we get exactly
890 	 * the same return value as before (we always csum_fold before
891 	 * testing, and that removes any differences).  However as we
892 	 * know that csum_partial always returned a 16bit value on
893 	 * alphas, do a fold to maximise conformity to previous behaviour.
894 	 */
895 	sb->sb_csum = md_csum_fold(disk_csum);
896 #else
897 	sb->sb_csum = disk_csum;
898 #endif
899 	return csum;
900 }
901 
902 
903 /*
904  * Handle superblock details.
905  * We want to be able to handle multiple superblock formats
906  * so we have a common interface to them all, and an array of
907  * different handlers.
908  * We rely on user-space to write the initial superblock, and support
909  * reading and updating of superblocks.
910  * Interface methods are:
911  *   int load_super(struct md_rdev *dev, struct md_rdev *refdev, int minor_version)
912  *      loads and validates a superblock on dev.
913  *      if refdev != NULL, compare superblocks on both devices
914  *    Return:
915  *      0 - dev has a superblock that is compatible with refdev
916  *      1 - dev has a superblock that is compatible and newer than refdev
917  *          so dev should be used as the refdev in future
918  *     -EINVAL superblock incompatible or invalid
919  *     -othererror e.g. -EIO
920  *
921  *   int validate_super(struct mddev *mddev, struct md_rdev *dev)
922  *      Verify that dev is acceptable into mddev.
923  *       The first time, mddev->raid_disks will be 0, and data from
924  *       dev should be merged in.  Subsequent calls check that dev
925  *       is new enough.  Return 0 or -EINVAL
926  *
927  *   void sync_super(struct mddev *mddev, struct md_rdev *dev)
928  *     Update the superblock for rdev with data in mddev
929  *     This does not write to disc.
930  *
931  */
932 
933 struct super_type  {
934 	char		    *name;
935 	struct module	    *owner;
936 	int		    (*load_super)(struct md_rdev *rdev,
937 					  struct md_rdev *refdev,
938 					  int minor_version);
939 	int		    (*validate_super)(struct mddev *mddev,
940 					      struct md_rdev *rdev);
941 	void		    (*sync_super)(struct mddev *mddev,
942 					  struct md_rdev *rdev);
943 	unsigned long long  (*rdev_size_change)(struct md_rdev *rdev,
944 						sector_t num_sectors);
945 	int		    (*allow_new_offset)(struct md_rdev *rdev,
946 						unsigned long long new_offset);
947 };
948 
949 /*
950  * Check that the given mddev has no bitmap.
951  *
952  * This function is called from the run method of all personalities that do not
953  * support bitmaps. It prints an error message and returns non-zero if mddev
954  * has a bitmap. Otherwise, it returns 0.
955  *
956  */
957 int md_check_no_bitmap(struct mddev *mddev)
958 {
959 	if (!mddev->bitmap_info.file && !mddev->bitmap_info.offset)
960 		return 0;
961 	printk(KERN_ERR "%s: bitmaps are not supported for %s\n",
962 		mdname(mddev), mddev->pers->name);
963 	return 1;
964 }
965 EXPORT_SYMBOL(md_check_no_bitmap);
966 
967 /*
968  * load_super for 0.90.0
969  */
970 static int super_90_load(struct md_rdev *rdev, struct md_rdev *refdev, int minor_version)
971 {
972 	char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
973 	mdp_super_t *sb;
974 	int ret;
975 
976 	/*
977 	 * Calculate the position of the superblock (512byte sectors),
978 	 * it's at the end of the disk.
979 	 *
980 	 * It also happens to be a multiple of 4Kb.
981 	 */
982 	rdev->sb_start = calc_dev_sboffset(rdev);
983 
984 	ret = read_disk_sb(rdev, MD_SB_BYTES);
985 	if (ret) return ret;
986 
987 	ret = -EINVAL;
988 
989 	bdevname(rdev->bdev, b);
990 	sb = page_address(rdev->sb_page);
991 
992 	if (sb->md_magic != MD_SB_MAGIC) {
993 		printk(KERN_ERR "md: invalid raid superblock magic on %s\n",
994 		       b);
995 		goto abort;
996 	}
997 
998 	if (sb->major_version != 0 ||
999 	    sb->minor_version < 90 ||
1000 	    sb->minor_version > 91) {
1001 		printk(KERN_WARNING "Bad version number %d.%d on %s\n",
1002 			sb->major_version, sb->minor_version,
1003 			b);
1004 		goto abort;
1005 	}
1006 
1007 	if (sb->raid_disks <= 0)
1008 		goto abort;
1009 
1010 	if (md_csum_fold(calc_sb_csum(sb)) != md_csum_fold(sb->sb_csum)) {
1011 		printk(KERN_WARNING "md: invalid superblock checksum on %s\n",
1012 			b);
1013 		goto abort;
1014 	}
1015 
1016 	rdev->preferred_minor = sb->md_minor;
1017 	rdev->data_offset = 0;
1018 	rdev->new_data_offset = 0;
1019 	rdev->sb_size = MD_SB_BYTES;
1020 	rdev->badblocks.shift = -1;
1021 
1022 	if (sb->level == LEVEL_MULTIPATH)
1023 		rdev->desc_nr = -1;
1024 	else
1025 		rdev->desc_nr = sb->this_disk.number;
1026 
1027 	if (!refdev) {
1028 		ret = 1;
1029 	} else {
1030 		__u64 ev1, ev2;
1031 		mdp_super_t *refsb = page_address(refdev->sb_page);
1032 		if (!uuid_equal(refsb, sb)) {
1033 			printk(KERN_WARNING "md: %s has different UUID to %s\n",
1034 				b, bdevname(refdev->bdev,b2));
1035 			goto abort;
1036 		}
1037 		if (!sb_equal(refsb, sb)) {
1038 			printk(KERN_WARNING "md: %s has same UUID"
1039 			       " but different superblock to %s\n",
1040 			       b, bdevname(refdev->bdev, b2));
1041 			goto abort;
1042 		}
1043 		ev1 = md_event(sb);
1044 		ev2 = md_event(refsb);
1045 		if (ev1 > ev2)
1046 			ret = 1;
1047 		else
1048 			ret = 0;
1049 	}
1050 	rdev->sectors = rdev->sb_start;
1051 	/* Limit to 4TB as metadata cannot record more than that.
1052 	 * (not needed for Linear and RAID0 as metadata doesn't
1053 	 * record this size)
1054 	 */
1055 	if (rdev->sectors >= (2ULL << 32) && sb->level >= 1)
1056 		rdev->sectors = (2ULL << 32) - 2;
1057 
1058 	if (rdev->sectors < ((sector_t)sb->size) * 2 && sb->level >= 1)
1059 		/* "this cannot possibly happen" ... */
1060 		ret = -EINVAL;
1061 
1062  abort:
1063 	return ret;
1064 }
1065 
1066 /*
1067  * validate_super for 0.90.0
1068  */
1069 static int super_90_validate(struct mddev *mddev, struct md_rdev *rdev)
1070 {
1071 	mdp_disk_t *desc;
1072 	mdp_super_t *sb = page_address(rdev->sb_page);
1073 	__u64 ev1 = md_event(sb);
1074 
1075 	rdev->raid_disk = -1;
1076 	clear_bit(Faulty, &rdev->flags);
1077 	clear_bit(In_sync, &rdev->flags);
1078 	clear_bit(Bitmap_sync, &rdev->flags);
1079 	clear_bit(WriteMostly, &rdev->flags);
1080 
1081 	if (mddev->raid_disks == 0) {
1082 		mddev->major_version = 0;
1083 		mddev->minor_version = sb->minor_version;
1084 		mddev->patch_version = sb->patch_version;
1085 		mddev->external = 0;
1086 		mddev->chunk_sectors = sb->chunk_size >> 9;
1087 		mddev->ctime = sb->ctime;
1088 		mddev->utime = sb->utime;
1089 		mddev->level = sb->level;
1090 		mddev->clevel[0] = 0;
1091 		mddev->layout = sb->layout;
1092 		mddev->raid_disks = sb->raid_disks;
1093 		mddev->dev_sectors = ((sector_t)sb->size) * 2;
1094 		mddev->events = ev1;
1095 		mddev->bitmap_info.offset = 0;
1096 		mddev->bitmap_info.space = 0;
1097 		/* bitmap can use 60 K after the 4K superblocks */
1098 		mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9;
1099 		mddev->bitmap_info.default_space = 64*2 - (MD_SB_BYTES >> 9);
1100 		mddev->reshape_backwards = 0;
1101 
1102 		if (mddev->minor_version >= 91) {
1103 			mddev->reshape_position = sb->reshape_position;
1104 			mddev->delta_disks = sb->delta_disks;
1105 			mddev->new_level = sb->new_level;
1106 			mddev->new_layout = sb->new_layout;
1107 			mddev->new_chunk_sectors = sb->new_chunk >> 9;
1108 			if (mddev->delta_disks < 0)
1109 				mddev->reshape_backwards = 1;
1110 		} else {
1111 			mddev->reshape_position = MaxSector;
1112 			mddev->delta_disks = 0;
1113 			mddev->new_level = mddev->level;
1114 			mddev->new_layout = mddev->layout;
1115 			mddev->new_chunk_sectors = mddev->chunk_sectors;
1116 		}
1117 
1118 		if (sb->state & (1<<MD_SB_CLEAN))
1119 			mddev->recovery_cp = MaxSector;
1120 		else {
1121 			if (sb->events_hi == sb->cp_events_hi &&
1122 				sb->events_lo == sb->cp_events_lo) {
1123 				mddev->recovery_cp = sb->recovery_cp;
1124 			} else
1125 				mddev->recovery_cp = 0;
1126 		}
1127 
1128 		memcpy(mddev->uuid+0, &sb->set_uuid0, 4);
1129 		memcpy(mddev->uuid+4, &sb->set_uuid1, 4);
1130 		memcpy(mddev->uuid+8, &sb->set_uuid2, 4);
1131 		memcpy(mddev->uuid+12,&sb->set_uuid3, 4);
1132 
1133 		mddev->max_disks = MD_SB_DISKS;
1134 
1135 		if (sb->state & (1<<MD_SB_BITMAP_PRESENT) &&
1136 		    mddev->bitmap_info.file == NULL) {
1137 			mddev->bitmap_info.offset =
1138 				mddev->bitmap_info.default_offset;
1139 			mddev->bitmap_info.space =
1140 				mddev->bitmap_info.default_space;
1141 		}
1142 
1143 	} else if (mddev->pers == NULL) {
1144 		/* Insist on good event counter while assembling, except
1145 		 * for spares (which don't need an event count) */
1146 		++ev1;
1147 		if (sb->disks[rdev->desc_nr].state & (
1148 			    (1<<MD_DISK_SYNC) | (1 << MD_DISK_ACTIVE)))
1149 			if (ev1 < mddev->events)
1150 				return -EINVAL;
1151 	} else if (mddev->bitmap) {
1152 		/* if adding to array with a bitmap, then we can accept an
1153 		 * older device ... but not too old.
1154 		 */
1155 		if (ev1 < mddev->bitmap->events_cleared)
1156 			return 0;
1157 		if (ev1 < mddev->events)
1158 			set_bit(Bitmap_sync, &rdev->flags);
1159 	} else {
1160 		if (ev1 < mddev->events)
1161 			/* just a hot-add of a new device, leave raid_disk at -1 */
1162 			return 0;
1163 	}
1164 
1165 	if (mddev->level != LEVEL_MULTIPATH) {
1166 		desc = sb->disks + rdev->desc_nr;
1167 
1168 		if (desc->state & (1<<MD_DISK_FAULTY))
1169 			set_bit(Faulty, &rdev->flags);
1170 		else if (desc->state & (1<<MD_DISK_SYNC) /* &&
1171 			    desc->raid_disk < mddev->raid_disks */) {
1172 			set_bit(In_sync, &rdev->flags);
1173 			rdev->raid_disk = desc->raid_disk;
1174 			rdev->saved_raid_disk = desc->raid_disk;
1175 		} else if (desc->state & (1<<MD_DISK_ACTIVE)) {
1176 			/* active but not in sync implies recovery up to
1177 			 * reshape position.  We don't know exactly where
1178 			 * that is, so set to zero for now */
1179 			if (mddev->minor_version >= 91) {
1180 				rdev->recovery_offset = 0;
1181 				rdev->raid_disk = desc->raid_disk;
1182 			}
1183 		}
1184 		if (desc->state & (1<<MD_DISK_WRITEMOSTLY))
1185 			set_bit(WriteMostly, &rdev->flags);
1186 	} else /* MULTIPATH are always insync */
1187 		set_bit(In_sync, &rdev->flags);
1188 	return 0;
1189 }
1190 
1191 /*
1192  * sync_super for 0.90.0
1193  */
1194 static void super_90_sync(struct mddev *mddev, struct md_rdev *rdev)
1195 {
1196 	mdp_super_t *sb;
1197 	struct md_rdev *rdev2;
1198 	int next_spare = mddev->raid_disks;
1199 
1200 
1201 	/* make rdev->sb match mddev data..
1202 	 *
1203 	 * 1/ zero out disks
1204 	 * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
1205 	 * 3/ any empty disks < next_spare become removed
1206 	 *
1207 	 * disks[0] gets initialised to REMOVED because
1208 	 * we cannot be sure from other fields if it has
1209 	 * been initialised or not.
1210 	 */
1211 	int i;
1212 	int active=0, working=0,failed=0,spare=0,nr_disks=0;
1213 
1214 	rdev->sb_size = MD_SB_BYTES;
1215 
1216 	sb = page_address(rdev->sb_page);
1217 
1218 	memset(sb, 0, sizeof(*sb));
1219 
1220 	sb->md_magic = MD_SB_MAGIC;
1221 	sb->major_version = mddev->major_version;
1222 	sb->patch_version = mddev->patch_version;
1223 	sb->gvalid_words  = 0; /* ignored */
1224 	memcpy(&sb->set_uuid0, mddev->uuid+0, 4);
1225 	memcpy(&sb->set_uuid1, mddev->uuid+4, 4);
1226 	memcpy(&sb->set_uuid2, mddev->uuid+8, 4);
1227 	memcpy(&sb->set_uuid3, mddev->uuid+12,4);
1228 
1229 	sb->ctime = mddev->ctime;
1230 	sb->level = mddev->level;
1231 	sb->size = mddev->dev_sectors / 2;
1232 	sb->raid_disks = mddev->raid_disks;
1233 	sb->md_minor = mddev->md_minor;
1234 	sb->not_persistent = 0;
1235 	sb->utime = mddev->utime;
1236 	sb->state = 0;
1237 	sb->events_hi = (mddev->events>>32);
1238 	sb->events_lo = (u32)mddev->events;
1239 
1240 	if (mddev->reshape_position == MaxSector)
1241 		sb->minor_version = 90;
1242 	else {
1243 		sb->minor_version = 91;
1244 		sb->reshape_position = mddev->reshape_position;
1245 		sb->new_level = mddev->new_level;
1246 		sb->delta_disks = mddev->delta_disks;
1247 		sb->new_layout = mddev->new_layout;
1248 		sb->new_chunk = mddev->new_chunk_sectors << 9;
1249 	}
1250 	mddev->minor_version = sb->minor_version;
1251 	if (mddev->in_sync)
1252 	{
1253 		sb->recovery_cp = mddev->recovery_cp;
1254 		sb->cp_events_hi = (mddev->events>>32);
1255 		sb->cp_events_lo = (u32)mddev->events;
1256 		if (mddev->recovery_cp == MaxSector)
1257 			sb->state = (1<< MD_SB_CLEAN);
1258 	} else
1259 		sb->recovery_cp = 0;
1260 
1261 	sb->layout = mddev->layout;
1262 	sb->chunk_size = mddev->chunk_sectors << 9;
1263 
1264 	if (mddev->bitmap && mddev->bitmap_info.file == NULL)
1265 		sb->state |= (1<<MD_SB_BITMAP_PRESENT);
1266 
1267 	sb->disks[0].state = (1<<MD_DISK_REMOVED);
1268 	rdev_for_each(rdev2, mddev) {
1269 		mdp_disk_t *d;
1270 		int desc_nr;
1271 		int is_active = test_bit(In_sync, &rdev2->flags);
1272 
1273 		if (rdev2->raid_disk >= 0 &&
1274 		    sb->minor_version >= 91)
1275 			/* we have nowhere to store the recovery_offset,
1276 			 * but if it is not below the reshape_position,
1277 			 * we can piggy-back on that.
1278 			 */
1279 			is_active = 1;
1280 		if (rdev2->raid_disk < 0 ||
1281 		    test_bit(Faulty, &rdev2->flags))
1282 			is_active = 0;
1283 		if (is_active)
1284 			desc_nr = rdev2->raid_disk;
1285 		else
1286 			desc_nr = next_spare++;
1287 		rdev2->desc_nr = desc_nr;
1288 		d = &sb->disks[rdev2->desc_nr];
1289 		nr_disks++;
1290 		d->number = rdev2->desc_nr;
1291 		d->major = MAJOR(rdev2->bdev->bd_dev);
1292 		d->minor = MINOR(rdev2->bdev->bd_dev);
1293 		if (is_active)
1294 			d->raid_disk = rdev2->raid_disk;
1295 		else
1296 			d->raid_disk = rdev2->desc_nr; /* compatibility */
1297 		if (test_bit(Faulty, &rdev2->flags))
1298 			d->state = (1<<MD_DISK_FAULTY);
1299 		else if (is_active) {
1300 			d->state = (1<<MD_DISK_ACTIVE);
1301 			if (test_bit(In_sync, &rdev2->flags))
1302 				d->state |= (1<<MD_DISK_SYNC);
1303 			active++;
1304 			working++;
1305 		} else {
1306 			d->state = 0;
1307 			spare++;
1308 			working++;
1309 		}
1310 		if (test_bit(WriteMostly, &rdev2->flags))
1311 			d->state |= (1<<MD_DISK_WRITEMOSTLY);
1312 	}
1313 	/* now set the "removed" and "faulty" bits on any missing devices */
1314 	for (i=0 ; i < mddev->raid_disks ; i++) {
1315 		mdp_disk_t *d = &sb->disks[i];
1316 		if (d->state == 0 && d->number == 0) {
1317 			d->number = i;
1318 			d->raid_disk = i;
1319 			d->state = (1<<MD_DISK_REMOVED);
1320 			d->state |= (1<<MD_DISK_FAULTY);
1321 			failed++;
1322 		}
1323 	}
1324 	sb->nr_disks = nr_disks;
1325 	sb->active_disks = active;
1326 	sb->working_disks = working;
1327 	sb->failed_disks = failed;
1328 	sb->spare_disks = spare;
1329 
1330 	sb->this_disk = sb->disks[rdev->desc_nr];
1331 	sb->sb_csum = calc_sb_csum(sb);
1332 }
1333 
1334 /*
1335  * rdev_size_change for 0.90.0
1336  */
1337 static unsigned long long
1338 super_90_rdev_size_change(struct md_rdev *rdev, sector_t num_sectors)
1339 {
1340 	if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
1341 		return 0; /* component must fit device */
1342 	if (rdev->mddev->bitmap_info.offset)
1343 		return 0; /* can't move bitmap */
1344 	rdev->sb_start = calc_dev_sboffset(rdev);
1345 	if (!num_sectors || num_sectors > rdev->sb_start)
1346 		num_sectors = rdev->sb_start;
1347 	/* Limit to 4TB as metadata cannot record more than that.
1348 	 * 4TB == 2^32 KB, or 2*2^32 sectors.
1349 	 */
1350 	if (num_sectors >= (2ULL << 32) && rdev->mddev->level >= 1)
1351 		num_sectors = (2ULL << 32) - 2;
1352 	md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
1353 		       rdev->sb_page);
1354 	md_super_wait(rdev->mddev);
1355 	return num_sectors;
1356 }
1357 
1358 static int
1359 super_90_allow_new_offset(struct md_rdev *rdev, unsigned long long new_offset)
1360 {
1361 	/* non-zero offset changes not possible with v0.90 */
1362 	return new_offset == 0;
1363 }
1364 
1365 /*
1366  * version 1 superblock
1367  */
1368 
1369 static __le32 calc_sb_1_csum(struct mdp_superblock_1 * sb)
1370 {
1371 	__le32 disk_csum;
1372 	u32 csum;
1373 	unsigned long long newcsum;
1374 	int size = 256 + le32_to_cpu(sb->max_dev)*2;
1375 	__le32 *isuper = (__le32*)sb;
1376 
1377 	disk_csum = sb->sb_csum;
1378 	sb->sb_csum = 0;
1379 	newcsum = 0;
1380 	for (; size >= 4; size -= 4)
1381 		newcsum += le32_to_cpu(*isuper++);
1382 
1383 	if (size == 2)
1384 		newcsum += le16_to_cpu(*(__le16*) isuper);
1385 
1386 	csum = (newcsum & 0xffffffff) + (newcsum >> 32);
1387 	sb->sb_csum = disk_csum;
1388 	return cpu_to_le32(csum);
1389 }
1390 
1391 static int md_set_badblocks(struct badblocks *bb, sector_t s, int sectors,
1392 			    int acknowledged);
1393 static int super_1_load(struct md_rdev *rdev, struct md_rdev *refdev, int minor_version)
1394 {
1395 	struct mdp_superblock_1 *sb;
1396 	int ret;
1397 	sector_t sb_start;
1398 	sector_t sectors;
1399 	char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
1400 	int bmask;
1401 
1402 	/*
1403 	 * Calculate the position of the superblock in 512byte sectors.
1404 	 * It is always aligned to a 4K boundary and
1405 	 * depeding on minor_version, it can be:
1406 	 * 0: At least 8K, but less than 12K, from end of device
1407 	 * 1: At start of device
1408 	 * 2: 4K from start of device.
1409 	 */
1410 	switch(minor_version) {
1411 	case 0:
1412 		sb_start = i_size_read(rdev->bdev->bd_inode) >> 9;
1413 		sb_start -= 8*2;
1414 		sb_start &= ~(sector_t)(4*2-1);
1415 		break;
1416 	case 1:
1417 		sb_start = 0;
1418 		break;
1419 	case 2:
1420 		sb_start = 8;
1421 		break;
1422 	default:
1423 		return -EINVAL;
1424 	}
1425 	rdev->sb_start = sb_start;
1426 
1427 	/* superblock is rarely larger than 1K, but it can be larger,
1428 	 * and it is safe to read 4k, so we do that
1429 	 */
1430 	ret = read_disk_sb(rdev, 4096);
1431 	if (ret) return ret;
1432 
1433 
1434 	sb = page_address(rdev->sb_page);
1435 
1436 	if (sb->magic != cpu_to_le32(MD_SB_MAGIC) ||
1437 	    sb->major_version != cpu_to_le32(1) ||
1438 	    le32_to_cpu(sb->max_dev) > (4096-256)/2 ||
1439 	    le64_to_cpu(sb->super_offset) != rdev->sb_start ||
1440 	    (le32_to_cpu(sb->feature_map) & ~MD_FEATURE_ALL) != 0)
1441 		return -EINVAL;
1442 
1443 	if (calc_sb_1_csum(sb) != sb->sb_csum) {
1444 		printk("md: invalid superblock checksum on %s\n",
1445 			bdevname(rdev->bdev,b));
1446 		return -EINVAL;
1447 	}
1448 	if (le64_to_cpu(sb->data_size) < 10) {
1449 		printk("md: data_size too small on %s\n",
1450 		       bdevname(rdev->bdev,b));
1451 		return -EINVAL;
1452 	}
1453 	if (sb->pad0 ||
1454 	    sb->pad3[0] ||
1455 	    memcmp(sb->pad3, sb->pad3+1, sizeof(sb->pad3) - sizeof(sb->pad3[1])))
1456 		/* Some padding is non-zero, might be a new feature */
1457 		return -EINVAL;
1458 
1459 	rdev->preferred_minor = 0xffff;
1460 	rdev->data_offset = le64_to_cpu(sb->data_offset);
1461 	rdev->new_data_offset = rdev->data_offset;
1462 	if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE) &&
1463 	    (le32_to_cpu(sb->feature_map) & MD_FEATURE_NEW_OFFSET))
1464 		rdev->new_data_offset += (s32)le32_to_cpu(sb->new_offset);
1465 	atomic_set(&rdev->corrected_errors, le32_to_cpu(sb->cnt_corrected_read));
1466 
1467 	rdev->sb_size = le32_to_cpu(sb->max_dev) * 2 + 256;
1468 	bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
1469 	if (rdev->sb_size & bmask)
1470 		rdev->sb_size = (rdev->sb_size | bmask) + 1;
1471 
1472 	if (minor_version
1473 	    && rdev->data_offset < sb_start + (rdev->sb_size/512))
1474 		return -EINVAL;
1475 	if (minor_version
1476 	    && rdev->new_data_offset < sb_start + (rdev->sb_size/512))
1477 		return -EINVAL;
1478 
1479 	if (sb->level == cpu_to_le32(LEVEL_MULTIPATH))
1480 		rdev->desc_nr = -1;
1481 	else
1482 		rdev->desc_nr = le32_to_cpu(sb->dev_number);
1483 
1484 	if (!rdev->bb_page) {
1485 		rdev->bb_page = alloc_page(GFP_KERNEL);
1486 		if (!rdev->bb_page)
1487 			return -ENOMEM;
1488 	}
1489 	if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BAD_BLOCKS) &&
1490 	    rdev->badblocks.count == 0) {
1491 		/* need to load the bad block list.
1492 		 * Currently we limit it to one page.
1493 		 */
1494 		s32 offset;
1495 		sector_t bb_sector;
1496 		u64 *bbp;
1497 		int i;
1498 		int sectors = le16_to_cpu(sb->bblog_size);
1499 		if (sectors > (PAGE_SIZE / 512))
1500 			return -EINVAL;
1501 		offset = le32_to_cpu(sb->bblog_offset);
1502 		if (offset == 0)
1503 			return -EINVAL;
1504 		bb_sector = (long long)offset;
1505 		if (!sync_page_io(rdev, bb_sector, sectors << 9,
1506 				  rdev->bb_page, READ, true))
1507 			return -EIO;
1508 		bbp = (u64 *)page_address(rdev->bb_page);
1509 		rdev->badblocks.shift = sb->bblog_shift;
1510 		for (i = 0 ; i < (sectors << (9-3)) ; i++, bbp++) {
1511 			u64 bb = le64_to_cpu(*bbp);
1512 			int count = bb & (0x3ff);
1513 			u64 sector = bb >> 10;
1514 			sector <<= sb->bblog_shift;
1515 			count <<= sb->bblog_shift;
1516 			if (bb + 1 == 0)
1517 				break;
1518 			if (md_set_badblocks(&rdev->badblocks,
1519 					     sector, count, 1) == 0)
1520 				return -EINVAL;
1521 		}
1522 	} else if (sb->bblog_offset != 0)
1523 		rdev->badblocks.shift = 0;
1524 
1525 	if (!refdev) {
1526 		ret = 1;
1527 	} else {
1528 		__u64 ev1, ev2;
1529 		struct mdp_superblock_1 *refsb = page_address(refdev->sb_page);
1530 
1531 		if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 ||
1532 		    sb->level != refsb->level ||
1533 		    sb->layout != refsb->layout ||
1534 		    sb->chunksize != refsb->chunksize) {
1535 			printk(KERN_WARNING "md: %s has strangely different"
1536 				" superblock to %s\n",
1537 				bdevname(rdev->bdev,b),
1538 				bdevname(refdev->bdev,b2));
1539 			return -EINVAL;
1540 		}
1541 		ev1 = le64_to_cpu(sb->events);
1542 		ev2 = le64_to_cpu(refsb->events);
1543 
1544 		if (ev1 > ev2)
1545 			ret = 1;
1546 		else
1547 			ret = 0;
1548 	}
1549 	if (minor_version) {
1550 		sectors = (i_size_read(rdev->bdev->bd_inode) >> 9);
1551 		sectors -= rdev->data_offset;
1552 	} else
1553 		sectors = rdev->sb_start;
1554 	if (sectors < le64_to_cpu(sb->data_size))
1555 		return -EINVAL;
1556 	rdev->sectors = le64_to_cpu(sb->data_size);
1557 	return ret;
1558 }
1559 
1560 static int super_1_validate(struct mddev *mddev, struct md_rdev *rdev)
1561 {
1562 	struct mdp_superblock_1 *sb = page_address(rdev->sb_page);
1563 	__u64 ev1 = le64_to_cpu(sb->events);
1564 
1565 	rdev->raid_disk = -1;
1566 	clear_bit(Faulty, &rdev->flags);
1567 	clear_bit(In_sync, &rdev->flags);
1568 	clear_bit(Bitmap_sync, &rdev->flags);
1569 	clear_bit(WriteMostly, &rdev->flags);
1570 
1571 	if (mddev->raid_disks == 0) {
1572 		mddev->major_version = 1;
1573 		mddev->patch_version = 0;
1574 		mddev->external = 0;
1575 		mddev->chunk_sectors = le32_to_cpu(sb->chunksize);
1576 		mddev->ctime = le64_to_cpu(sb->ctime) & ((1ULL << 32)-1);
1577 		mddev->utime = le64_to_cpu(sb->utime) & ((1ULL << 32)-1);
1578 		mddev->level = le32_to_cpu(sb->level);
1579 		mddev->clevel[0] = 0;
1580 		mddev->layout = le32_to_cpu(sb->layout);
1581 		mddev->raid_disks = le32_to_cpu(sb->raid_disks);
1582 		mddev->dev_sectors = le64_to_cpu(sb->size);
1583 		mddev->events = ev1;
1584 		mddev->bitmap_info.offset = 0;
1585 		mddev->bitmap_info.space = 0;
1586 		/* Default location for bitmap is 1K after superblock
1587 		 * using 3K - total of 4K
1588 		 */
1589 		mddev->bitmap_info.default_offset = 1024 >> 9;
1590 		mddev->bitmap_info.default_space = (4096-1024) >> 9;
1591 		mddev->reshape_backwards = 0;
1592 
1593 		mddev->recovery_cp = le64_to_cpu(sb->resync_offset);
1594 		memcpy(mddev->uuid, sb->set_uuid, 16);
1595 
1596 		mddev->max_disks =  (4096-256)/2;
1597 
1598 		if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) &&
1599 		    mddev->bitmap_info.file == NULL) {
1600 			mddev->bitmap_info.offset =
1601 				(__s32)le32_to_cpu(sb->bitmap_offset);
1602 			/* Metadata doesn't record how much space is available.
1603 			 * For 1.0, we assume we can use up to the superblock
1604 			 * if before, else to 4K beyond superblock.
1605 			 * For others, assume no change is possible.
1606 			 */
1607 			if (mddev->minor_version > 0)
1608 				mddev->bitmap_info.space = 0;
1609 			else if (mddev->bitmap_info.offset > 0)
1610 				mddev->bitmap_info.space =
1611 					8 - mddev->bitmap_info.offset;
1612 			else
1613 				mddev->bitmap_info.space =
1614 					-mddev->bitmap_info.offset;
1615 		}
1616 
1617 		if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) {
1618 			mddev->reshape_position = le64_to_cpu(sb->reshape_position);
1619 			mddev->delta_disks = le32_to_cpu(sb->delta_disks);
1620 			mddev->new_level = le32_to_cpu(sb->new_level);
1621 			mddev->new_layout = le32_to_cpu(sb->new_layout);
1622 			mddev->new_chunk_sectors = le32_to_cpu(sb->new_chunk);
1623 			if (mddev->delta_disks < 0 ||
1624 			    (mddev->delta_disks == 0 &&
1625 			     (le32_to_cpu(sb->feature_map)
1626 			      & MD_FEATURE_RESHAPE_BACKWARDS)))
1627 				mddev->reshape_backwards = 1;
1628 		} else {
1629 			mddev->reshape_position = MaxSector;
1630 			mddev->delta_disks = 0;
1631 			mddev->new_level = mddev->level;
1632 			mddev->new_layout = mddev->layout;
1633 			mddev->new_chunk_sectors = mddev->chunk_sectors;
1634 		}
1635 
1636 	} else if (mddev->pers == NULL) {
1637 		/* Insist of good event counter while assembling, except for
1638 		 * spares (which don't need an event count) */
1639 		++ev1;
1640 		if (rdev->desc_nr >= 0 &&
1641 		    rdev->desc_nr < le32_to_cpu(sb->max_dev) &&
1642 		    le16_to_cpu(sb->dev_roles[rdev->desc_nr]) < 0xfffe)
1643 			if (ev1 < mddev->events)
1644 				return -EINVAL;
1645 	} else if (mddev->bitmap) {
1646 		/* If adding to array with a bitmap, then we can accept an
1647 		 * older device, but not too old.
1648 		 */
1649 		if (ev1 < mddev->bitmap->events_cleared)
1650 			return 0;
1651 		if (ev1 < mddev->events)
1652 			set_bit(Bitmap_sync, &rdev->flags);
1653 	} else {
1654 		if (ev1 < mddev->events)
1655 			/* just a hot-add of a new device, leave raid_disk at -1 */
1656 			return 0;
1657 	}
1658 	if (mddev->level != LEVEL_MULTIPATH) {
1659 		int role;
1660 		if (rdev->desc_nr < 0 ||
1661 		    rdev->desc_nr >= le32_to_cpu(sb->max_dev)) {
1662 			role = 0xffff;
1663 			rdev->desc_nr = -1;
1664 		} else
1665 			role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
1666 		switch(role) {
1667 		case 0xffff: /* spare */
1668 			break;
1669 		case 0xfffe: /* faulty */
1670 			set_bit(Faulty, &rdev->flags);
1671 			break;
1672 		default:
1673 			rdev->saved_raid_disk = role;
1674 			if ((le32_to_cpu(sb->feature_map) &
1675 			     MD_FEATURE_RECOVERY_OFFSET)) {
1676 				rdev->recovery_offset = le64_to_cpu(sb->recovery_offset);
1677 				if (!(le32_to_cpu(sb->feature_map) &
1678 				      MD_FEATURE_RECOVERY_BITMAP))
1679 					rdev->saved_raid_disk = -1;
1680 			} else
1681 				set_bit(In_sync, &rdev->flags);
1682 			rdev->raid_disk = role;
1683 			break;
1684 		}
1685 		if (sb->devflags & WriteMostly1)
1686 			set_bit(WriteMostly, &rdev->flags);
1687 		if (le32_to_cpu(sb->feature_map) & MD_FEATURE_REPLACEMENT)
1688 			set_bit(Replacement, &rdev->flags);
1689 	} else /* MULTIPATH are always insync */
1690 		set_bit(In_sync, &rdev->flags);
1691 
1692 	return 0;
1693 }
1694 
1695 static void super_1_sync(struct mddev *mddev, struct md_rdev *rdev)
1696 {
1697 	struct mdp_superblock_1 *sb;
1698 	struct md_rdev *rdev2;
1699 	int max_dev, i;
1700 	/* make rdev->sb match mddev and rdev data. */
1701 
1702 	sb = page_address(rdev->sb_page);
1703 
1704 	sb->feature_map = 0;
1705 	sb->pad0 = 0;
1706 	sb->recovery_offset = cpu_to_le64(0);
1707 	memset(sb->pad3, 0, sizeof(sb->pad3));
1708 
1709 	sb->utime = cpu_to_le64((__u64)mddev->utime);
1710 	sb->events = cpu_to_le64(mddev->events);
1711 	if (mddev->in_sync)
1712 		sb->resync_offset = cpu_to_le64(mddev->recovery_cp);
1713 	else
1714 		sb->resync_offset = cpu_to_le64(0);
1715 
1716 	sb->cnt_corrected_read = cpu_to_le32(atomic_read(&rdev->corrected_errors));
1717 
1718 	sb->raid_disks = cpu_to_le32(mddev->raid_disks);
1719 	sb->size = cpu_to_le64(mddev->dev_sectors);
1720 	sb->chunksize = cpu_to_le32(mddev->chunk_sectors);
1721 	sb->level = cpu_to_le32(mddev->level);
1722 	sb->layout = cpu_to_le32(mddev->layout);
1723 
1724 	if (test_bit(WriteMostly, &rdev->flags))
1725 		sb->devflags |= WriteMostly1;
1726 	else
1727 		sb->devflags &= ~WriteMostly1;
1728 	sb->data_offset = cpu_to_le64(rdev->data_offset);
1729 	sb->data_size = cpu_to_le64(rdev->sectors);
1730 
1731 	if (mddev->bitmap && mddev->bitmap_info.file == NULL) {
1732 		sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_info.offset);
1733 		sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET);
1734 	}
1735 
1736 	if (rdev->raid_disk >= 0 &&
1737 	    !test_bit(In_sync, &rdev->flags)) {
1738 		sb->feature_map |=
1739 			cpu_to_le32(MD_FEATURE_RECOVERY_OFFSET);
1740 		sb->recovery_offset =
1741 			cpu_to_le64(rdev->recovery_offset);
1742 		if (rdev->saved_raid_disk >= 0 && mddev->bitmap)
1743 			sb->feature_map |=
1744 				cpu_to_le32(MD_FEATURE_RECOVERY_BITMAP);
1745 	}
1746 	if (test_bit(Replacement, &rdev->flags))
1747 		sb->feature_map |=
1748 			cpu_to_le32(MD_FEATURE_REPLACEMENT);
1749 
1750 	if (mddev->reshape_position != MaxSector) {
1751 		sb->feature_map |= cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE);
1752 		sb->reshape_position = cpu_to_le64(mddev->reshape_position);
1753 		sb->new_layout = cpu_to_le32(mddev->new_layout);
1754 		sb->delta_disks = cpu_to_le32(mddev->delta_disks);
1755 		sb->new_level = cpu_to_le32(mddev->new_level);
1756 		sb->new_chunk = cpu_to_le32(mddev->new_chunk_sectors);
1757 		if (mddev->delta_disks == 0 &&
1758 		    mddev->reshape_backwards)
1759 			sb->feature_map
1760 				|= cpu_to_le32(MD_FEATURE_RESHAPE_BACKWARDS);
1761 		if (rdev->new_data_offset != rdev->data_offset) {
1762 			sb->feature_map
1763 				|= cpu_to_le32(MD_FEATURE_NEW_OFFSET);
1764 			sb->new_offset = cpu_to_le32((__u32)(rdev->new_data_offset
1765 							     - rdev->data_offset));
1766 		}
1767 	}
1768 
1769 	if (rdev->badblocks.count == 0)
1770 		/* Nothing to do for bad blocks*/ ;
1771 	else if (sb->bblog_offset == 0)
1772 		/* Cannot record bad blocks on this device */
1773 		md_error(mddev, rdev);
1774 	else {
1775 		struct badblocks *bb = &rdev->badblocks;
1776 		u64 *bbp = (u64 *)page_address(rdev->bb_page);
1777 		u64 *p = bb->page;
1778 		sb->feature_map |= cpu_to_le32(MD_FEATURE_BAD_BLOCKS);
1779 		if (bb->changed) {
1780 			unsigned seq;
1781 
1782 retry:
1783 			seq = read_seqbegin(&bb->lock);
1784 
1785 			memset(bbp, 0xff, PAGE_SIZE);
1786 
1787 			for (i = 0 ; i < bb->count ; i++) {
1788 				u64 internal_bb = p[i];
1789 				u64 store_bb = ((BB_OFFSET(internal_bb) << 10)
1790 						| BB_LEN(internal_bb));
1791 				bbp[i] = cpu_to_le64(store_bb);
1792 			}
1793 			bb->changed = 0;
1794 			if (read_seqretry(&bb->lock, seq))
1795 				goto retry;
1796 
1797 			bb->sector = (rdev->sb_start +
1798 				      (int)le32_to_cpu(sb->bblog_offset));
1799 			bb->size = le16_to_cpu(sb->bblog_size);
1800 		}
1801 	}
1802 
1803 	max_dev = 0;
1804 	rdev_for_each(rdev2, mddev)
1805 		if (rdev2->desc_nr+1 > max_dev)
1806 			max_dev = rdev2->desc_nr+1;
1807 
1808 	if (max_dev > le32_to_cpu(sb->max_dev)) {
1809 		int bmask;
1810 		sb->max_dev = cpu_to_le32(max_dev);
1811 		rdev->sb_size = max_dev * 2 + 256;
1812 		bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
1813 		if (rdev->sb_size & bmask)
1814 			rdev->sb_size = (rdev->sb_size | bmask) + 1;
1815 	} else
1816 		max_dev = le32_to_cpu(sb->max_dev);
1817 
1818 	for (i=0; i<max_dev;i++)
1819 		sb->dev_roles[i] = cpu_to_le16(0xfffe);
1820 
1821 	rdev_for_each(rdev2, mddev) {
1822 		i = rdev2->desc_nr;
1823 		if (test_bit(Faulty, &rdev2->flags))
1824 			sb->dev_roles[i] = cpu_to_le16(0xfffe);
1825 		else if (test_bit(In_sync, &rdev2->flags))
1826 			sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1827 		else if (rdev2->raid_disk >= 0)
1828 			sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1829 		else
1830 			sb->dev_roles[i] = cpu_to_le16(0xffff);
1831 	}
1832 
1833 	sb->sb_csum = calc_sb_1_csum(sb);
1834 }
1835 
1836 static unsigned long long
1837 super_1_rdev_size_change(struct md_rdev *rdev, sector_t num_sectors)
1838 {
1839 	struct mdp_superblock_1 *sb;
1840 	sector_t max_sectors;
1841 	if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
1842 		return 0; /* component must fit device */
1843 	if (rdev->data_offset != rdev->new_data_offset)
1844 		return 0; /* too confusing */
1845 	if (rdev->sb_start < rdev->data_offset) {
1846 		/* minor versions 1 and 2; superblock before data */
1847 		max_sectors = i_size_read(rdev->bdev->bd_inode) >> 9;
1848 		max_sectors -= rdev->data_offset;
1849 		if (!num_sectors || num_sectors > max_sectors)
1850 			num_sectors = max_sectors;
1851 	} else if (rdev->mddev->bitmap_info.offset) {
1852 		/* minor version 0 with bitmap we can't move */
1853 		return 0;
1854 	} else {
1855 		/* minor version 0; superblock after data */
1856 		sector_t sb_start;
1857 		sb_start = (i_size_read(rdev->bdev->bd_inode) >> 9) - 8*2;
1858 		sb_start &= ~(sector_t)(4*2 - 1);
1859 		max_sectors = rdev->sectors + sb_start - rdev->sb_start;
1860 		if (!num_sectors || num_sectors > max_sectors)
1861 			num_sectors = max_sectors;
1862 		rdev->sb_start = sb_start;
1863 	}
1864 	sb = page_address(rdev->sb_page);
1865 	sb->data_size = cpu_to_le64(num_sectors);
1866 	sb->super_offset = rdev->sb_start;
1867 	sb->sb_csum = calc_sb_1_csum(sb);
1868 	md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
1869 		       rdev->sb_page);
1870 	md_super_wait(rdev->mddev);
1871 	return num_sectors;
1872 
1873 }
1874 
1875 static int
1876 super_1_allow_new_offset(struct md_rdev *rdev,
1877 			 unsigned long long new_offset)
1878 {
1879 	/* All necessary checks on new >= old have been done */
1880 	struct bitmap *bitmap;
1881 	if (new_offset >= rdev->data_offset)
1882 		return 1;
1883 
1884 	/* with 1.0 metadata, there is no metadata to tread on
1885 	 * so we can always move back */
1886 	if (rdev->mddev->minor_version == 0)
1887 		return 1;
1888 
1889 	/* otherwise we must be sure not to step on
1890 	 * any metadata, so stay:
1891 	 * 36K beyond start of superblock
1892 	 * beyond end of badblocks
1893 	 * beyond write-intent bitmap
1894 	 */
1895 	if (rdev->sb_start + (32+4)*2 > new_offset)
1896 		return 0;
1897 	bitmap = rdev->mddev->bitmap;
1898 	if (bitmap && !rdev->mddev->bitmap_info.file &&
1899 	    rdev->sb_start + rdev->mddev->bitmap_info.offset +
1900 	    bitmap->storage.file_pages * (PAGE_SIZE>>9) > new_offset)
1901 		return 0;
1902 	if (rdev->badblocks.sector + rdev->badblocks.size > new_offset)
1903 		return 0;
1904 
1905 	return 1;
1906 }
1907 
1908 static struct super_type super_types[] = {
1909 	[0] = {
1910 		.name	= "0.90.0",
1911 		.owner	= THIS_MODULE,
1912 		.load_super	    = super_90_load,
1913 		.validate_super	    = super_90_validate,
1914 		.sync_super	    = super_90_sync,
1915 		.rdev_size_change   = super_90_rdev_size_change,
1916 		.allow_new_offset   = super_90_allow_new_offset,
1917 	},
1918 	[1] = {
1919 		.name	= "md-1",
1920 		.owner	= THIS_MODULE,
1921 		.load_super	    = super_1_load,
1922 		.validate_super	    = super_1_validate,
1923 		.sync_super	    = super_1_sync,
1924 		.rdev_size_change   = super_1_rdev_size_change,
1925 		.allow_new_offset   = super_1_allow_new_offset,
1926 	},
1927 };
1928 
1929 static void sync_super(struct mddev *mddev, struct md_rdev *rdev)
1930 {
1931 	if (mddev->sync_super) {
1932 		mddev->sync_super(mddev, rdev);
1933 		return;
1934 	}
1935 
1936 	BUG_ON(mddev->major_version >= ARRAY_SIZE(super_types));
1937 
1938 	super_types[mddev->major_version].sync_super(mddev, rdev);
1939 }
1940 
1941 static int match_mddev_units(struct mddev *mddev1, struct mddev *mddev2)
1942 {
1943 	struct md_rdev *rdev, *rdev2;
1944 
1945 	rcu_read_lock();
1946 	rdev_for_each_rcu(rdev, mddev1)
1947 		rdev_for_each_rcu(rdev2, mddev2)
1948 			if (rdev->bdev->bd_contains ==
1949 			    rdev2->bdev->bd_contains) {
1950 				rcu_read_unlock();
1951 				return 1;
1952 			}
1953 	rcu_read_unlock();
1954 	return 0;
1955 }
1956 
1957 static LIST_HEAD(pending_raid_disks);
1958 
1959 /*
1960  * Try to register data integrity profile for an mddev
1961  *
1962  * This is called when an array is started and after a disk has been kicked
1963  * from the array. It only succeeds if all working and active component devices
1964  * are integrity capable with matching profiles.
1965  */
1966 int md_integrity_register(struct mddev *mddev)
1967 {
1968 	struct md_rdev *rdev, *reference = NULL;
1969 
1970 	if (list_empty(&mddev->disks))
1971 		return 0; /* nothing to do */
1972 	if (!mddev->gendisk || blk_get_integrity(mddev->gendisk))
1973 		return 0; /* shouldn't register, or already is */
1974 	rdev_for_each(rdev, mddev) {
1975 		/* skip spares and non-functional disks */
1976 		if (test_bit(Faulty, &rdev->flags))
1977 			continue;
1978 		if (rdev->raid_disk < 0)
1979 			continue;
1980 		if (!reference) {
1981 			/* Use the first rdev as the reference */
1982 			reference = rdev;
1983 			continue;
1984 		}
1985 		/* does this rdev's profile match the reference profile? */
1986 		if (blk_integrity_compare(reference->bdev->bd_disk,
1987 				rdev->bdev->bd_disk) < 0)
1988 			return -EINVAL;
1989 	}
1990 	if (!reference || !bdev_get_integrity(reference->bdev))
1991 		return 0;
1992 	/*
1993 	 * All component devices are integrity capable and have matching
1994 	 * profiles, register the common profile for the md device.
1995 	 */
1996 	if (blk_integrity_register(mddev->gendisk,
1997 			bdev_get_integrity(reference->bdev)) != 0) {
1998 		printk(KERN_ERR "md: failed to register integrity for %s\n",
1999 			mdname(mddev));
2000 		return -EINVAL;
2001 	}
2002 	printk(KERN_NOTICE "md: data integrity enabled on %s\n", mdname(mddev));
2003 	if (bioset_integrity_create(mddev->bio_set, BIO_POOL_SIZE)) {
2004 		printk(KERN_ERR "md: failed to create integrity pool for %s\n",
2005 		       mdname(mddev));
2006 		return -EINVAL;
2007 	}
2008 	return 0;
2009 }
2010 EXPORT_SYMBOL(md_integrity_register);
2011 
2012 /* Disable data integrity if non-capable/non-matching disk is being added */
2013 void md_integrity_add_rdev(struct md_rdev *rdev, struct mddev *mddev)
2014 {
2015 	struct blk_integrity *bi_rdev;
2016 	struct blk_integrity *bi_mddev;
2017 
2018 	if (!mddev->gendisk)
2019 		return;
2020 
2021 	bi_rdev = bdev_get_integrity(rdev->bdev);
2022 	bi_mddev = blk_get_integrity(mddev->gendisk);
2023 
2024 	if (!bi_mddev) /* nothing to do */
2025 		return;
2026 	if (rdev->raid_disk < 0) /* skip spares */
2027 		return;
2028 	if (bi_rdev && blk_integrity_compare(mddev->gendisk,
2029 					     rdev->bdev->bd_disk) >= 0)
2030 		return;
2031 	printk(KERN_NOTICE "disabling data integrity on %s\n", mdname(mddev));
2032 	blk_integrity_unregister(mddev->gendisk);
2033 }
2034 EXPORT_SYMBOL(md_integrity_add_rdev);
2035 
2036 static int bind_rdev_to_array(struct md_rdev * rdev, struct mddev * mddev)
2037 {
2038 	char b[BDEVNAME_SIZE];
2039 	struct kobject *ko;
2040 	char *s;
2041 	int err;
2042 
2043 	if (rdev->mddev) {
2044 		MD_BUG();
2045 		return -EINVAL;
2046 	}
2047 
2048 	/* prevent duplicates */
2049 	if (find_rdev(mddev, rdev->bdev->bd_dev))
2050 		return -EEXIST;
2051 
2052 	/* make sure rdev->sectors exceeds mddev->dev_sectors */
2053 	if (rdev->sectors && (mddev->dev_sectors == 0 ||
2054 			rdev->sectors < mddev->dev_sectors)) {
2055 		if (mddev->pers) {
2056 			/* Cannot change size, so fail
2057 			 * If mddev->level <= 0, then we don't care
2058 			 * about aligning sizes (e.g. linear)
2059 			 */
2060 			if (mddev->level > 0)
2061 				return -ENOSPC;
2062 		} else
2063 			mddev->dev_sectors = rdev->sectors;
2064 	}
2065 
2066 	/* Verify rdev->desc_nr is unique.
2067 	 * If it is -1, assign a free number, else
2068 	 * check number is not in use
2069 	 */
2070 	if (rdev->desc_nr < 0) {
2071 		int choice = 0;
2072 		if (mddev->pers) choice = mddev->raid_disks;
2073 		while (find_rdev_nr(mddev, choice))
2074 			choice++;
2075 		rdev->desc_nr = choice;
2076 	} else {
2077 		if (find_rdev_nr(mddev, rdev->desc_nr))
2078 			return -EBUSY;
2079 	}
2080 	if (mddev->max_disks && rdev->desc_nr >= mddev->max_disks) {
2081 		printk(KERN_WARNING "md: %s: array is limited to %d devices\n",
2082 		       mdname(mddev), mddev->max_disks);
2083 		return -EBUSY;
2084 	}
2085 	bdevname(rdev->bdev,b);
2086 	while ( (s=strchr(b, '/')) != NULL)
2087 		*s = '!';
2088 
2089 	rdev->mddev = mddev;
2090 	printk(KERN_INFO "md: bind<%s>\n", b);
2091 
2092 	if ((err = kobject_add(&rdev->kobj, &mddev->kobj, "dev-%s", b)))
2093 		goto fail;
2094 
2095 	ko = &part_to_dev(rdev->bdev->bd_part)->kobj;
2096 	if (sysfs_create_link(&rdev->kobj, ko, "block"))
2097 		/* failure here is OK */;
2098 	rdev->sysfs_state = sysfs_get_dirent_safe(rdev->kobj.sd, "state");
2099 
2100 	list_add_rcu(&rdev->same_set, &mddev->disks);
2101 	bd_link_disk_holder(rdev->bdev, mddev->gendisk);
2102 
2103 	/* May as well allow recovery to be retried once */
2104 	mddev->recovery_disabled++;
2105 
2106 	return 0;
2107 
2108  fail:
2109 	printk(KERN_WARNING "md: failed to register dev-%s for %s\n",
2110 	       b, mdname(mddev));
2111 	return err;
2112 }
2113 
2114 static void md_delayed_delete(struct work_struct *ws)
2115 {
2116 	struct md_rdev *rdev = container_of(ws, struct md_rdev, del_work);
2117 	kobject_del(&rdev->kobj);
2118 	kobject_put(&rdev->kobj);
2119 }
2120 
2121 static void unbind_rdev_from_array(struct md_rdev * rdev)
2122 {
2123 	char b[BDEVNAME_SIZE];
2124 	if (!rdev->mddev) {
2125 		MD_BUG();
2126 		return;
2127 	}
2128 	bd_unlink_disk_holder(rdev->bdev, rdev->mddev->gendisk);
2129 	list_del_rcu(&rdev->same_set);
2130 	printk(KERN_INFO "md: unbind<%s>\n", bdevname(rdev->bdev,b));
2131 	rdev->mddev = NULL;
2132 	sysfs_remove_link(&rdev->kobj, "block");
2133 	sysfs_put(rdev->sysfs_state);
2134 	rdev->sysfs_state = NULL;
2135 	rdev->badblocks.count = 0;
2136 	/* We need to delay this, otherwise we can deadlock when
2137 	 * writing to 'remove' to "dev/state".  We also need
2138 	 * to delay it due to rcu usage.
2139 	 */
2140 	synchronize_rcu();
2141 	INIT_WORK(&rdev->del_work, md_delayed_delete);
2142 	kobject_get(&rdev->kobj);
2143 	queue_work(md_misc_wq, &rdev->del_work);
2144 }
2145 
2146 /*
2147  * prevent the device from being mounted, repartitioned or
2148  * otherwise reused by a RAID array (or any other kernel
2149  * subsystem), by bd_claiming the device.
2150  */
2151 static int lock_rdev(struct md_rdev *rdev, dev_t dev, int shared)
2152 {
2153 	int err = 0;
2154 	struct block_device *bdev;
2155 	char b[BDEVNAME_SIZE];
2156 
2157 	bdev = blkdev_get_by_dev(dev, FMODE_READ|FMODE_WRITE|FMODE_EXCL,
2158 				 shared ? (struct md_rdev *)lock_rdev : rdev);
2159 	if (IS_ERR(bdev)) {
2160 		printk(KERN_ERR "md: could not open %s.\n",
2161 			__bdevname(dev, b));
2162 		return PTR_ERR(bdev);
2163 	}
2164 	rdev->bdev = bdev;
2165 	return err;
2166 }
2167 
2168 static void unlock_rdev(struct md_rdev *rdev)
2169 {
2170 	struct block_device *bdev = rdev->bdev;
2171 	rdev->bdev = NULL;
2172 	if (!bdev)
2173 		MD_BUG();
2174 	blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
2175 }
2176 
2177 void md_autodetect_dev(dev_t dev);
2178 
2179 static void export_rdev(struct md_rdev * rdev)
2180 {
2181 	char b[BDEVNAME_SIZE];
2182 	printk(KERN_INFO "md: export_rdev(%s)\n",
2183 		bdevname(rdev->bdev,b));
2184 	if (rdev->mddev)
2185 		MD_BUG();
2186 	md_rdev_clear(rdev);
2187 #ifndef MODULE
2188 	if (test_bit(AutoDetected, &rdev->flags))
2189 		md_autodetect_dev(rdev->bdev->bd_dev);
2190 #endif
2191 	unlock_rdev(rdev);
2192 	kobject_put(&rdev->kobj);
2193 }
2194 
2195 static void kick_rdev_from_array(struct md_rdev * rdev)
2196 {
2197 	unbind_rdev_from_array(rdev);
2198 	export_rdev(rdev);
2199 }
2200 
2201 static void export_array(struct mddev *mddev)
2202 {
2203 	struct md_rdev *rdev, *tmp;
2204 
2205 	rdev_for_each_safe(rdev, tmp, mddev) {
2206 		if (!rdev->mddev) {
2207 			MD_BUG();
2208 			continue;
2209 		}
2210 		kick_rdev_from_array(rdev);
2211 	}
2212 	if (!list_empty(&mddev->disks))
2213 		MD_BUG();
2214 	mddev->raid_disks = 0;
2215 	mddev->major_version = 0;
2216 }
2217 
2218 static void print_desc(mdp_disk_t *desc)
2219 {
2220 	printk(" DISK<N:%d,(%d,%d),R:%d,S:%d>\n", desc->number,
2221 		desc->major,desc->minor,desc->raid_disk,desc->state);
2222 }
2223 
2224 static void print_sb_90(mdp_super_t *sb)
2225 {
2226 	int i;
2227 
2228 	printk(KERN_INFO
2229 		"md:  SB: (V:%d.%d.%d) ID:<%08x.%08x.%08x.%08x> CT:%08x\n",
2230 		sb->major_version, sb->minor_version, sb->patch_version,
2231 		sb->set_uuid0, sb->set_uuid1, sb->set_uuid2, sb->set_uuid3,
2232 		sb->ctime);
2233 	printk(KERN_INFO "md:     L%d S%08d ND:%d RD:%d md%d LO:%d CS:%d\n",
2234 		sb->level, sb->size, sb->nr_disks, sb->raid_disks,
2235 		sb->md_minor, sb->layout, sb->chunk_size);
2236 	printk(KERN_INFO "md:     UT:%08x ST:%d AD:%d WD:%d"
2237 		" FD:%d SD:%d CSUM:%08x E:%08lx\n",
2238 		sb->utime, sb->state, sb->active_disks, sb->working_disks,
2239 		sb->failed_disks, sb->spare_disks,
2240 		sb->sb_csum, (unsigned long)sb->events_lo);
2241 
2242 	printk(KERN_INFO);
2243 	for (i = 0; i < MD_SB_DISKS; i++) {
2244 		mdp_disk_t *desc;
2245 
2246 		desc = sb->disks + i;
2247 		if (desc->number || desc->major || desc->minor ||
2248 		    desc->raid_disk || (desc->state && (desc->state != 4))) {
2249 			printk("     D %2d: ", i);
2250 			print_desc(desc);
2251 		}
2252 	}
2253 	printk(KERN_INFO "md:     THIS: ");
2254 	print_desc(&sb->this_disk);
2255 }
2256 
2257 static void print_sb_1(struct mdp_superblock_1 *sb)
2258 {
2259 	__u8 *uuid;
2260 
2261 	uuid = sb->set_uuid;
2262 	printk(KERN_INFO
2263 	       "md:  SB: (V:%u) (F:0x%08x) Array-ID:<%pU>\n"
2264 	       "md:    Name: \"%s\" CT:%llu\n",
2265 		le32_to_cpu(sb->major_version),
2266 		le32_to_cpu(sb->feature_map),
2267 		uuid,
2268 		sb->set_name,
2269 		(unsigned long long)le64_to_cpu(sb->ctime)
2270 		       & MD_SUPERBLOCK_1_TIME_SEC_MASK);
2271 
2272 	uuid = sb->device_uuid;
2273 	printk(KERN_INFO
2274 	       "md:       L%u SZ%llu RD:%u LO:%u CS:%u DO:%llu DS:%llu SO:%llu"
2275 			" RO:%llu\n"
2276 	       "md:     Dev:%08x UUID: %pU\n"
2277 	       "md:       (F:0x%08x) UT:%llu Events:%llu ResyncOffset:%llu CSUM:0x%08x\n"
2278 	       "md:         (MaxDev:%u) \n",
2279 		le32_to_cpu(sb->level),
2280 		(unsigned long long)le64_to_cpu(sb->size),
2281 		le32_to_cpu(sb->raid_disks),
2282 		le32_to_cpu(sb->layout),
2283 		le32_to_cpu(sb->chunksize),
2284 		(unsigned long long)le64_to_cpu(sb->data_offset),
2285 		(unsigned long long)le64_to_cpu(sb->data_size),
2286 		(unsigned long long)le64_to_cpu(sb->super_offset),
2287 		(unsigned long long)le64_to_cpu(sb->recovery_offset),
2288 		le32_to_cpu(sb->dev_number),
2289 		uuid,
2290 		sb->devflags,
2291 		(unsigned long long)le64_to_cpu(sb->utime) & MD_SUPERBLOCK_1_TIME_SEC_MASK,
2292 		(unsigned long long)le64_to_cpu(sb->events),
2293 		(unsigned long long)le64_to_cpu(sb->resync_offset),
2294 		le32_to_cpu(sb->sb_csum),
2295 		le32_to_cpu(sb->max_dev)
2296 		);
2297 }
2298 
2299 static void print_rdev(struct md_rdev *rdev, int major_version)
2300 {
2301 	char b[BDEVNAME_SIZE];
2302 	printk(KERN_INFO "md: rdev %s, Sect:%08llu F:%d S:%d DN:%u\n",
2303 		bdevname(rdev->bdev, b), (unsigned long long)rdev->sectors,
2304 	        test_bit(Faulty, &rdev->flags), test_bit(In_sync, &rdev->flags),
2305 	        rdev->desc_nr);
2306 	if (rdev->sb_loaded) {
2307 		printk(KERN_INFO "md: rdev superblock (MJ:%d):\n", major_version);
2308 		switch (major_version) {
2309 		case 0:
2310 			print_sb_90(page_address(rdev->sb_page));
2311 			break;
2312 		case 1:
2313 			print_sb_1(page_address(rdev->sb_page));
2314 			break;
2315 		}
2316 	} else
2317 		printk(KERN_INFO "md: no rdev superblock!\n");
2318 }
2319 
2320 static void md_print_devices(void)
2321 {
2322 	struct list_head *tmp;
2323 	struct md_rdev *rdev;
2324 	struct mddev *mddev;
2325 	char b[BDEVNAME_SIZE];
2326 
2327 	printk("\n");
2328 	printk("md:	**********************************\n");
2329 	printk("md:	* <COMPLETE RAID STATE PRINTOUT> *\n");
2330 	printk("md:	**********************************\n");
2331 	for_each_mddev(mddev, tmp) {
2332 
2333 		if (mddev->bitmap)
2334 			bitmap_print_sb(mddev->bitmap);
2335 		else
2336 			printk("%s: ", mdname(mddev));
2337 		rdev_for_each(rdev, mddev)
2338 			printk("<%s>", bdevname(rdev->bdev,b));
2339 		printk("\n");
2340 
2341 		rdev_for_each(rdev, mddev)
2342 			print_rdev(rdev, mddev->major_version);
2343 	}
2344 	printk("md:	**********************************\n");
2345 	printk("\n");
2346 }
2347 
2348 
2349 static void sync_sbs(struct mddev * mddev, int nospares)
2350 {
2351 	/* Update each superblock (in-memory image), but
2352 	 * if we are allowed to, skip spares which already
2353 	 * have the right event counter, or have one earlier
2354 	 * (which would mean they aren't being marked as dirty
2355 	 * with the rest of the array)
2356 	 */
2357 	struct md_rdev *rdev;
2358 	rdev_for_each(rdev, mddev) {
2359 		if (rdev->sb_events == mddev->events ||
2360 		    (nospares &&
2361 		     rdev->raid_disk < 0 &&
2362 		     rdev->sb_events+1 == mddev->events)) {
2363 			/* Don't update this superblock */
2364 			rdev->sb_loaded = 2;
2365 		} else {
2366 			sync_super(mddev, rdev);
2367 			rdev->sb_loaded = 1;
2368 		}
2369 	}
2370 }
2371 
2372 static void md_update_sb(struct mddev * mddev, int force_change)
2373 {
2374 	struct md_rdev *rdev;
2375 	int sync_req;
2376 	int nospares = 0;
2377 	int any_badblocks_changed = 0;
2378 
2379 	if (mddev->ro) {
2380 		if (force_change)
2381 			set_bit(MD_CHANGE_DEVS, &mddev->flags);
2382 		return;
2383 	}
2384 repeat:
2385 	/* First make sure individual recovery_offsets are correct */
2386 	rdev_for_each(rdev, mddev) {
2387 		if (rdev->raid_disk >= 0 &&
2388 		    mddev->delta_disks >= 0 &&
2389 		    !test_bit(In_sync, &rdev->flags) &&
2390 		    mddev->curr_resync_completed > rdev->recovery_offset)
2391 				rdev->recovery_offset = mddev->curr_resync_completed;
2392 
2393 	}
2394 	if (!mddev->persistent) {
2395 		clear_bit(MD_CHANGE_CLEAN, &mddev->flags);
2396 		clear_bit(MD_CHANGE_DEVS, &mddev->flags);
2397 		if (!mddev->external) {
2398 			clear_bit(MD_CHANGE_PENDING, &mddev->flags);
2399 			rdev_for_each(rdev, mddev) {
2400 				if (rdev->badblocks.changed) {
2401 					rdev->badblocks.changed = 0;
2402 					md_ack_all_badblocks(&rdev->badblocks);
2403 					md_error(mddev, rdev);
2404 				}
2405 				clear_bit(Blocked, &rdev->flags);
2406 				clear_bit(BlockedBadBlocks, &rdev->flags);
2407 				wake_up(&rdev->blocked_wait);
2408 			}
2409 		}
2410 		wake_up(&mddev->sb_wait);
2411 		return;
2412 	}
2413 
2414 	spin_lock_irq(&mddev->write_lock);
2415 
2416 	mddev->utime = get_seconds();
2417 
2418 	if (test_and_clear_bit(MD_CHANGE_DEVS, &mddev->flags))
2419 		force_change = 1;
2420 	if (test_and_clear_bit(MD_CHANGE_CLEAN, &mddev->flags))
2421 		/* just a clean<-> dirty transition, possibly leave spares alone,
2422 		 * though if events isn't the right even/odd, we will have to do
2423 		 * spares after all
2424 		 */
2425 		nospares = 1;
2426 	if (force_change)
2427 		nospares = 0;
2428 	if (mddev->degraded)
2429 		/* If the array is degraded, then skipping spares is both
2430 		 * dangerous and fairly pointless.
2431 		 * Dangerous because a device that was removed from the array
2432 		 * might have a event_count that still looks up-to-date,
2433 		 * so it can be re-added without a resync.
2434 		 * Pointless because if there are any spares to skip,
2435 		 * then a recovery will happen and soon that array won't
2436 		 * be degraded any more and the spare can go back to sleep then.
2437 		 */
2438 		nospares = 0;
2439 
2440 	sync_req = mddev->in_sync;
2441 
2442 	/* If this is just a dirty<->clean transition, and the array is clean
2443 	 * and 'events' is odd, we can roll back to the previous clean state */
2444 	if (nospares
2445 	    && (mddev->in_sync && mddev->recovery_cp == MaxSector)
2446 	    && mddev->can_decrease_events
2447 	    && mddev->events != 1) {
2448 		mddev->events--;
2449 		mddev->can_decrease_events = 0;
2450 	} else {
2451 		/* otherwise we have to go forward and ... */
2452 		mddev->events ++;
2453 		mddev->can_decrease_events = nospares;
2454 	}
2455 
2456 	if (!mddev->events) {
2457 		/*
2458 		 * oops, this 64-bit counter should never wrap.
2459 		 * Either we are in around ~1 trillion A.C., assuming
2460 		 * 1 reboot per second, or we have a bug:
2461 		 */
2462 		MD_BUG();
2463 		mddev->events --;
2464 	}
2465 
2466 	rdev_for_each(rdev, mddev) {
2467 		if (rdev->badblocks.changed)
2468 			any_badblocks_changed++;
2469 		if (test_bit(Faulty, &rdev->flags))
2470 			set_bit(FaultRecorded, &rdev->flags);
2471 	}
2472 
2473 	sync_sbs(mddev, nospares);
2474 	spin_unlock_irq(&mddev->write_lock);
2475 
2476 	pr_debug("md: updating %s RAID superblock on device (in sync %d)\n",
2477 		 mdname(mddev), mddev->in_sync);
2478 
2479 	bitmap_update_sb(mddev->bitmap);
2480 	rdev_for_each(rdev, mddev) {
2481 		char b[BDEVNAME_SIZE];
2482 
2483 		if (rdev->sb_loaded != 1)
2484 			continue; /* no noise on spare devices */
2485 
2486 		if (!test_bit(Faulty, &rdev->flags)) {
2487 			md_super_write(mddev,rdev,
2488 				       rdev->sb_start, rdev->sb_size,
2489 				       rdev->sb_page);
2490 			pr_debug("md: (write) %s's sb offset: %llu\n",
2491 				 bdevname(rdev->bdev, b),
2492 				 (unsigned long long)rdev->sb_start);
2493 			rdev->sb_events = mddev->events;
2494 			if (rdev->badblocks.size) {
2495 				md_super_write(mddev, rdev,
2496 					       rdev->badblocks.sector,
2497 					       rdev->badblocks.size << 9,
2498 					       rdev->bb_page);
2499 				rdev->badblocks.size = 0;
2500 			}
2501 
2502 		} else
2503 			pr_debug("md: %s (skipping faulty)\n",
2504 				 bdevname(rdev->bdev, b));
2505 
2506 		if (mddev->level == LEVEL_MULTIPATH)
2507 			/* only need to write one superblock... */
2508 			break;
2509 	}
2510 	md_super_wait(mddev);
2511 	/* if there was a failure, MD_CHANGE_DEVS was set, and we re-write super */
2512 
2513 	spin_lock_irq(&mddev->write_lock);
2514 	if (mddev->in_sync != sync_req ||
2515 	    test_bit(MD_CHANGE_DEVS, &mddev->flags)) {
2516 		/* have to write it out again */
2517 		spin_unlock_irq(&mddev->write_lock);
2518 		goto repeat;
2519 	}
2520 	clear_bit(MD_CHANGE_PENDING, &mddev->flags);
2521 	spin_unlock_irq(&mddev->write_lock);
2522 	wake_up(&mddev->sb_wait);
2523 	if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
2524 		sysfs_notify(&mddev->kobj, NULL, "sync_completed");
2525 
2526 	rdev_for_each(rdev, mddev) {
2527 		if (test_and_clear_bit(FaultRecorded, &rdev->flags))
2528 			clear_bit(Blocked, &rdev->flags);
2529 
2530 		if (any_badblocks_changed)
2531 			md_ack_all_badblocks(&rdev->badblocks);
2532 		clear_bit(BlockedBadBlocks, &rdev->flags);
2533 		wake_up(&rdev->blocked_wait);
2534 	}
2535 }
2536 
2537 /* words written to sysfs files may, or may not, be \n terminated.
2538  * We want to accept with case. For this we use cmd_match.
2539  */
2540 static int cmd_match(const char *cmd, const char *str)
2541 {
2542 	/* See if cmd, written into a sysfs file, matches
2543 	 * str.  They must either be the same, or cmd can
2544 	 * have a trailing newline
2545 	 */
2546 	while (*cmd && *str && *cmd == *str) {
2547 		cmd++;
2548 		str++;
2549 	}
2550 	if (*cmd == '\n')
2551 		cmd++;
2552 	if (*str || *cmd)
2553 		return 0;
2554 	return 1;
2555 }
2556 
2557 struct rdev_sysfs_entry {
2558 	struct attribute attr;
2559 	ssize_t (*show)(struct md_rdev *, char *);
2560 	ssize_t (*store)(struct md_rdev *, const char *, size_t);
2561 };
2562 
2563 static ssize_t
2564 state_show(struct md_rdev *rdev, char *page)
2565 {
2566 	char *sep = "";
2567 	size_t len = 0;
2568 
2569 	if (test_bit(Faulty, &rdev->flags) ||
2570 	    rdev->badblocks.unacked_exist) {
2571 		len+= sprintf(page+len, "%sfaulty",sep);
2572 		sep = ",";
2573 	}
2574 	if (test_bit(In_sync, &rdev->flags)) {
2575 		len += sprintf(page+len, "%sin_sync",sep);
2576 		sep = ",";
2577 	}
2578 	if (test_bit(WriteMostly, &rdev->flags)) {
2579 		len += sprintf(page+len, "%swrite_mostly",sep);
2580 		sep = ",";
2581 	}
2582 	if (test_bit(Blocked, &rdev->flags) ||
2583 	    (rdev->badblocks.unacked_exist
2584 	     && !test_bit(Faulty, &rdev->flags))) {
2585 		len += sprintf(page+len, "%sblocked", sep);
2586 		sep = ",";
2587 	}
2588 	if (!test_bit(Faulty, &rdev->flags) &&
2589 	    !test_bit(In_sync, &rdev->flags)) {
2590 		len += sprintf(page+len, "%sspare", sep);
2591 		sep = ",";
2592 	}
2593 	if (test_bit(WriteErrorSeen, &rdev->flags)) {
2594 		len += sprintf(page+len, "%swrite_error", sep);
2595 		sep = ",";
2596 	}
2597 	if (test_bit(WantReplacement, &rdev->flags)) {
2598 		len += sprintf(page+len, "%swant_replacement", sep);
2599 		sep = ",";
2600 	}
2601 	if (test_bit(Replacement, &rdev->flags)) {
2602 		len += sprintf(page+len, "%sreplacement", sep);
2603 		sep = ",";
2604 	}
2605 
2606 	return len+sprintf(page+len, "\n");
2607 }
2608 
2609 static ssize_t
2610 state_store(struct md_rdev *rdev, const char *buf, size_t len)
2611 {
2612 	/* can write
2613 	 *  faulty  - simulates an error
2614 	 *  remove  - disconnects the device
2615 	 *  writemostly - sets write_mostly
2616 	 *  -writemostly - clears write_mostly
2617 	 *  blocked - sets the Blocked flags
2618 	 *  -blocked - clears the Blocked and possibly simulates an error
2619 	 *  insync - sets Insync providing device isn't active
2620 	 *  -insync - clear Insync for a device with a slot assigned,
2621 	 *            so that it gets rebuilt based on bitmap
2622 	 *  write_error - sets WriteErrorSeen
2623 	 *  -write_error - clears WriteErrorSeen
2624 	 */
2625 	int err = -EINVAL;
2626 	if (cmd_match(buf, "faulty") && rdev->mddev->pers) {
2627 		md_error(rdev->mddev, rdev);
2628 		if (test_bit(Faulty, &rdev->flags))
2629 			err = 0;
2630 		else
2631 			err = -EBUSY;
2632 	} else if (cmd_match(buf, "remove")) {
2633 		if (rdev->raid_disk >= 0)
2634 			err = -EBUSY;
2635 		else {
2636 			struct mddev *mddev = rdev->mddev;
2637 			kick_rdev_from_array(rdev);
2638 			if (mddev->pers)
2639 				md_update_sb(mddev, 1);
2640 			md_new_event(mddev);
2641 			err = 0;
2642 		}
2643 	} else if (cmd_match(buf, "writemostly")) {
2644 		set_bit(WriteMostly, &rdev->flags);
2645 		err = 0;
2646 	} else if (cmd_match(buf, "-writemostly")) {
2647 		clear_bit(WriteMostly, &rdev->flags);
2648 		err = 0;
2649 	} else if (cmd_match(buf, "blocked")) {
2650 		set_bit(Blocked, &rdev->flags);
2651 		err = 0;
2652 	} else if (cmd_match(buf, "-blocked")) {
2653 		if (!test_bit(Faulty, &rdev->flags) &&
2654 		    rdev->badblocks.unacked_exist) {
2655 			/* metadata handler doesn't understand badblocks,
2656 			 * so we need to fail the device
2657 			 */
2658 			md_error(rdev->mddev, rdev);
2659 		}
2660 		clear_bit(Blocked, &rdev->flags);
2661 		clear_bit(BlockedBadBlocks, &rdev->flags);
2662 		wake_up(&rdev->blocked_wait);
2663 		set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2664 		md_wakeup_thread(rdev->mddev->thread);
2665 
2666 		err = 0;
2667 	} else if (cmd_match(buf, "insync") && rdev->raid_disk == -1) {
2668 		set_bit(In_sync, &rdev->flags);
2669 		err = 0;
2670 	} else if (cmd_match(buf, "-insync") && rdev->raid_disk >= 0) {
2671 		clear_bit(In_sync, &rdev->flags);
2672 		rdev->saved_raid_disk = rdev->raid_disk;
2673 		rdev->raid_disk = -1;
2674 		err = 0;
2675 	} else if (cmd_match(buf, "write_error")) {
2676 		set_bit(WriteErrorSeen, &rdev->flags);
2677 		err = 0;
2678 	} else if (cmd_match(buf, "-write_error")) {
2679 		clear_bit(WriteErrorSeen, &rdev->flags);
2680 		err = 0;
2681 	} else if (cmd_match(buf, "want_replacement")) {
2682 		/* Any non-spare device that is not a replacement can
2683 		 * become want_replacement at any time, but we then need to
2684 		 * check if recovery is needed.
2685 		 */
2686 		if (rdev->raid_disk >= 0 &&
2687 		    !test_bit(Replacement, &rdev->flags))
2688 			set_bit(WantReplacement, &rdev->flags);
2689 		set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2690 		md_wakeup_thread(rdev->mddev->thread);
2691 		err = 0;
2692 	} else if (cmd_match(buf, "-want_replacement")) {
2693 		/* Clearing 'want_replacement' is always allowed.
2694 		 * Once replacements starts it is too late though.
2695 		 */
2696 		err = 0;
2697 		clear_bit(WantReplacement, &rdev->flags);
2698 	} else if (cmd_match(buf, "replacement")) {
2699 		/* Can only set a device as a replacement when array has not
2700 		 * yet been started.  Once running, replacement is automatic
2701 		 * from spares, or by assigning 'slot'.
2702 		 */
2703 		if (rdev->mddev->pers)
2704 			err = -EBUSY;
2705 		else {
2706 			set_bit(Replacement, &rdev->flags);
2707 			err = 0;
2708 		}
2709 	} else if (cmd_match(buf, "-replacement")) {
2710 		/* Similarly, can only clear Replacement before start */
2711 		if (rdev->mddev->pers)
2712 			err = -EBUSY;
2713 		else {
2714 			clear_bit(Replacement, &rdev->flags);
2715 			err = 0;
2716 		}
2717 	}
2718 	if (!err)
2719 		sysfs_notify_dirent_safe(rdev->sysfs_state);
2720 	return err ? err : len;
2721 }
2722 static struct rdev_sysfs_entry rdev_state =
2723 __ATTR(state, S_IRUGO|S_IWUSR, state_show, state_store);
2724 
2725 static ssize_t
2726 errors_show(struct md_rdev *rdev, char *page)
2727 {
2728 	return sprintf(page, "%d\n", atomic_read(&rdev->corrected_errors));
2729 }
2730 
2731 static ssize_t
2732 errors_store(struct md_rdev *rdev, const char *buf, size_t len)
2733 {
2734 	char *e;
2735 	unsigned long n = simple_strtoul(buf, &e, 10);
2736 	if (*buf && (*e == 0 || *e == '\n')) {
2737 		atomic_set(&rdev->corrected_errors, n);
2738 		return len;
2739 	}
2740 	return -EINVAL;
2741 }
2742 static struct rdev_sysfs_entry rdev_errors =
2743 __ATTR(errors, S_IRUGO|S_IWUSR, errors_show, errors_store);
2744 
2745 static ssize_t
2746 slot_show(struct md_rdev *rdev, char *page)
2747 {
2748 	if (rdev->raid_disk < 0)
2749 		return sprintf(page, "none\n");
2750 	else
2751 		return sprintf(page, "%d\n", rdev->raid_disk);
2752 }
2753 
2754 static ssize_t
2755 slot_store(struct md_rdev *rdev, const char *buf, size_t len)
2756 {
2757 	char *e;
2758 	int err;
2759 	int slot = simple_strtoul(buf, &e, 10);
2760 	if (strncmp(buf, "none", 4)==0)
2761 		slot = -1;
2762 	else if (e==buf || (*e && *e!= '\n'))
2763 		return -EINVAL;
2764 	if (rdev->mddev->pers && slot == -1) {
2765 		/* Setting 'slot' on an active array requires also
2766 		 * updating the 'rd%d' link, and communicating
2767 		 * with the personality with ->hot_*_disk.
2768 		 * For now we only support removing
2769 		 * failed/spare devices.  This normally happens automatically,
2770 		 * but not when the metadata is externally managed.
2771 		 */
2772 		if (rdev->raid_disk == -1)
2773 			return -EEXIST;
2774 		/* personality does all needed checks */
2775 		if (rdev->mddev->pers->hot_remove_disk == NULL)
2776 			return -EINVAL;
2777 		clear_bit(Blocked, &rdev->flags);
2778 		remove_and_add_spares(rdev->mddev, rdev);
2779 		if (rdev->raid_disk >= 0)
2780 			return -EBUSY;
2781 		set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2782 		md_wakeup_thread(rdev->mddev->thread);
2783 	} else if (rdev->mddev->pers) {
2784 		/* Activating a spare .. or possibly reactivating
2785 		 * if we ever get bitmaps working here.
2786 		 */
2787 
2788 		if (rdev->raid_disk != -1)
2789 			return -EBUSY;
2790 
2791 		if (test_bit(MD_RECOVERY_RUNNING, &rdev->mddev->recovery))
2792 			return -EBUSY;
2793 
2794 		if (rdev->mddev->pers->hot_add_disk == NULL)
2795 			return -EINVAL;
2796 
2797 		if (slot >= rdev->mddev->raid_disks &&
2798 		    slot >= rdev->mddev->raid_disks + rdev->mddev->delta_disks)
2799 			return -ENOSPC;
2800 
2801 		rdev->raid_disk = slot;
2802 		if (test_bit(In_sync, &rdev->flags))
2803 			rdev->saved_raid_disk = slot;
2804 		else
2805 			rdev->saved_raid_disk = -1;
2806 		clear_bit(In_sync, &rdev->flags);
2807 		clear_bit(Bitmap_sync, &rdev->flags);
2808 		err = rdev->mddev->pers->
2809 			hot_add_disk(rdev->mddev, rdev);
2810 		if (err) {
2811 			rdev->raid_disk = -1;
2812 			return err;
2813 		} else
2814 			sysfs_notify_dirent_safe(rdev->sysfs_state);
2815 		if (sysfs_link_rdev(rdev->mddev, rdev))
2816 			/* failure here is OK */;
2817 		/* don't wakeup anyone, leave that to userspace. */
2818 	} else {
2819 		if (slot >= rdev->mddev->raid_disks &&
2820 		    slot >= rdev->mddev->raid_disks + rdev->mddev->delta_disks)
2821 			return -ENOSPC;
2822 		rdev->raid_disk = slot;
2823 		/* assume it is working */
2824 		clear_bit(Faulty, &rdev->flags);
2825 		clear_bit(WriteMostly, &rdev->flags);
2826 		set_bit(In_sync, &rdev->flags);
2827 		sysfs_notify_dirent_safe(rdev->sysfs_state);
2828 	}
2829 	return len;
2830 }
2831 
2832 
2833 static struct rdev_sysfs_entry rdev_slot =
2834 __ATTR(slot, S_IRUGO|S_IWUSR, slot_show, slot_store);
2835 
2836 static ssize_t
2837 offset_show(struct md_rdev *rdev, char *page)
2838 {
2839 	return sprintf(page, "%llu\n", (unsigned long long)rdev->data_offset);
2840 }
2841 
2842 static ssize_t
2843 offset_store(struct md_rdev *rdev, const char *buf, size_t len)
2844 {
2845 	unsigned long long offset;
2846 	if (kstrtoull(buf, 10, &offset) < 0)
2847 		return -EINVAL;
2848 	if (rdev->mddev->pers && rdev->raid_disk >= 0)
2849 		return -EBUSY;
2850 	if (rdev->sectors && rdev->mddev->external)
2851 		/* Must set offset before size, so overlap checks
2852 		 * can be sane */
2853 		return -EBUSY;
2854 	rdev->data_offset = offset;
2855 	rdev->new_data_offset = offset;
2856 	return len;
2857 }
2858 
2859 static struct rdev_sysfs_entry rdev_offset =
2860 __ATTR(offset, S_IRUGO|S_IWUSR, offset_show, offset_store);
2861 
2862 static ssize_t new_offset_show(struct md_rdev *rdev, char *page)
2863 {
2864 	return sprintf(page, "%llu\n",
2865 		       (unsigned long long)rdev->new_data_offset);
2866 }
2867 
2868 static ssize_t new_offset_store(struct md_rdev *rdev,
2869 				const char *buf, size_t len)
2870 {
2871 	unsigned long long new_offset;
2872 	struct mddev *mddev = rdev->mddev;
2873 
2874 	if (kstrtoull(buf, 10, &new_offset) < 0)
2875 		return -EINVAL;
2876 
2877 	if (mddev->sync_thread)
2878 		return -EBUSY;
2879 	if (new_offset == rdev->data_offset)
2880 		/* reset is always permitted */
2881 		;
2882 	else if (new_offset > rdev->data_offset) {
2883 		/* must not push array size beyond rdev_sectors */
2884 		if (new_offset - rdev->data_offset
2885 		    + mddev->dev_sectors > rdev->sectors)
2886 				return -E2BIG;
2887 	}
2888 	/* Metadata worries about other space details. */
2889 
2890 	/* decreasing the offset is inconsistent with a backwards
2891 	 * reshape.
2892 	 */
2893 	if (new_offset < rdev->data_offset &&
2894 	    mddev->reshape_backwards)
2895 		return -EINVAL;
2896 	/* Increasing offset is inconsistent with forwards
2897 	 * reshape.  reshape_direction should be set to
2898 	 * 'backwards' first.
2899 	 */
2900 	if (new_offset > rdev->data_offset &&
2901 	    !mddev->reshape_backwards)
2902 		return -EINVAL;
2903 
2904 	if (mddev->pers && mddev->persistent &&
2905 	    !super_types[mddev->major_version]
2906 	    .allow_new_offset(rdev, new_offset))
2907 		return -E2BIG;
2908 	rdev->new_data_offset = new_offset;
2909 	if (new_offset > rdev->data_offset)
2910 		mddev->reshape_backwards = 1;
2911 	else if (new_offset < rdev->data_offset)
2912 		mddev->reshape_backwards = 0;
2913 
2914 	return len;
2915 }
2916 static struct rdev_sysfs_entry rdev_new_offset =
2917 __ATTR(new_offset, S_IRUGO|S_IWUSR, new_offset_show, new_offset_store);
2918 
2919 static ssize_t
2920 rdev_size_show(struct md_rdev *rdev, char *page)
2921 {
2922 	return sprintf(page, "%llu\n", (unsigned long long)rdev->sectors / 2);
2923 }
2924 
2925 static int overlaps(sector_t s1, sector_t l1, sector_t s2, sector_t l2)
2926 {
2927 	/* check if two start/length pairs overlap */
2928 	if (s1+l1 <= s2)
2929 		return 0;
2930 	if (s2+l2 <= s1)
2931 		return 0;
2932 	return 1;
2933 }
2934 
2935 static int strict_blocks_to_sectors(const char *buf, sector_t *sectors)
2936 {
2937 	unsigned long long blocks;
2938 	sector_t new;
2939 
2940 	if (kstrtoull(buf, 10, &blocks) < 0)
2941 		return -EINVAL;
2942 
2943 	if (blocks & 1ULL << (8 * sizeof(blocks) - 1))
2944 		return -EINVAL; /* sector conversion overflow */
2945 
2946 	new = blocks * 2;
2947 	if (new != blocks * 2)
2948 		return -EINVAL; /* unsigned long long to sector_t overflow */
2949 
2950 	*sectors = new;
2951 	return 0;
2952 }
2953 
2954 static ssize_t
2955 rdev_size_store(struct md_rdev *rdev, const char *buf, size_t len)
2956 {
2957 	struct mddev *my_mddev = rdev->mddev;
2958 	sector_t oldsectors = rdev->sectors;
2959 	sector_t sectors;
2960 
2961 	if (strict_blocks_to_sectors(buf, &sectors) < 0)
2962 		return -EINVAL;
2963 	if (rdev->data_offset != rdev->new_data_offset)
2964 		return -EINVAL; /* too confusing */
2965 	if (my_mddev->pers && rdev->raid_disk >= 0) {
2966 		if (my_mddev->persistent) {
2967 			sectors = super_types[my_mddev->major_version].
2968 				rdev_size_change(rdev, sectors);
2969 			if (!sectors)
2970 				return -EBUSY;
2971 		} else if (!sectors)
2972 			sectors = (i_size_read(rdev->bdev->bd_inode) >> 9) -
2973 				rdev->data_offset;
2974 		if (!my_mddev->pers->resize)
2975 			/* Cannot change size for RAID0 or Linear etc */
2976 			return -EINVAL;
2977 	}
2978 	if (sectors < my_mddev->dev_sectors)
2979 		return -EINVAL; /* component must fit device */
2980 
2981 	rdev->sectors = sectors;
2982 	if (sectors > oldsectors && my_mddev->external) {
2983 		/* need to check that all other rdevs with the same ->bdev
2984 		 * do not overlap.  We need to unlock the mddev to avoid
2985 		 * a deadlock.  We have already changed rdev->sectors, and if
2986 		 * we have to change it back, we will have the lock again.
2987 		 */
2988 		struct mddev *mddev;
2989 		int overlap = 0;
2990 		struct list_head *tmp;
2991 
2992 		mddev_unlock(my_mddev);
2993 		for_each_mddev(mddev, tmp) {
2994 			struct md_rdev *rdev2;
2995 
2996 			mddev_lock_nointr(mddev);
2997 			rdev_for_each(rdev2, mddev)
2998 				if (rdev->bdev == rdev2->bdev &&
2999 				    rdev != rdev2 &&
3000 				    overlaps(rdev->data_offset, rdev->sectors,
3001 					     rdev2->data_offset,
3002 					     rdev2->sectors)) {
3003 					overlap = 1;
3004 					break;
3005 				}
3006 			mddev_unlock(mddev);
3007 			if (overlap) {
3008 				mddev_put(mddev);
3009 				break;
3010 			}
3011 		}
3012 		mddev_lock_nointr(my_mddev);
3013 		if (overlap) {
3014 			/* Someone else could have slipped in a size
3015 			 * change here, but doing so is just silly.
3016 			 * We put oldsectors back because we *know* it is
3017 			 * safe, and trust userspace not to race with
3018 			 * itself
3019 			 */
3020 			rdev->sectors = oldsectors;
3021 			return -EBUSY;
3022 		}
3023 	}
3024 	return len;
3025 }
3026 
3027 static struct rdev_sysfs_entry rdev_size =
3028 __ATTR(size, S_IRUGO|S_IWUSR, rdev_size_show, rdev_size_store);
3029 
3030 
3031 static ssize_t recovery_start_show(struct md_rdev *rdev, char *page)
3032 {
3033 	unsigned long long recovery_start = rdev->recovery_offset;
3034 
3035 	if (test_bit(In_sync, &rdev->flags) ||
3036 	    recovery_start == MaxSector)
3037 		return sprintf(page, "none\n");
3038 
3039 	return sprintf(page, "%llu\n", recovery_start);
3040 }
3041 
3042 static ssize_t recovery_start_store(struct md_rdev *rdev, const char *buf, size_t len)
3043 {
3044 	unsigned long long recovery_start;
3045 
3046 	if (cmd_match(buf, "none"))
3047 		recovery_start = MaxSector;
3048 	else if (kstrtoull(buf, 10, &recovery_start))
3049 		return -EINVAL;
3050 
3051 	if (rdev->mddev->pers &&
3052 	    rdev->raid_disk >= 0)
3053 		return -EBUSY;
3054 
3055 	rdev->recovery_offset = recovery_start;
3056 	if (recovery_start == MaxSector)
3057 		set_bit(In_sync, &rdev->flags);
3058 	else
3059 		clear_bit(In_sync, &rdev->flags);
3060 	return len;
3061 }
3062 
3063 static struct rdev_sysfs_entry rdev_recovery_start =
3064 __ATTR(recovery_start, S_IRUGO|S_IWUSR, recovery_start_show, recovery_start_store);
3065 
3066 
3067 static ssize_t
3068 badblocks_show(struct badblocks *bb, char *page, int unack);
3069 static ssize_t
3070 badblocks_store(struct badblocks *bb, const char *page, size_t len, int unack);
3071 
3072 static ssize_t bb_show(struct md_rdev *rdev, char *page)
3073 {
3074 	return badblocks_show(&rdev->badblocks, page, 0);
3075 }
3076 static ssize_t bb_store(struct md_rdev *rdev, const char *page, size_t len)
3077 {
3078 	int rv = badblocks_store(&rdev->badblocks, page, len, 0);
3079 	/* Maybe that ack was all we needed */
3080 	if (test_and_clear_bit(BlockedBadBlocks, &rdev->flags))
3081 		wake_up(&rdev->blocked_wait);
3082 	return rv;
3083 }
3084 static struct rdev_sysfs_entry rdev_bad_blocks =
3085 __ATTR(bad_blocks, S_IRUGO|S_IWUSR, bb_show, bb_store);
3086 
3087 
3088 static ssize_t ubb_show(struct md_rdev *rdev, char *page)
3089 {
3090 	return badblocks_show(&rdev->badblocks, page, 1);
3091 }
3092 static ssize_t ubb_store(struct md_rdev *rdev, const char *page, size_t len)
3093 {
3094 	return badblocks_store(&rdev->badblocks, page, len, 1);
3095 }
3096 static struct rdev_sysfs_entry rdev_unack_bad_blocks =
3097 __ATTR(unacknowledged_bad_blocks, S_IRUGO|S_IWUSR, ubb_show, ubb_store);
3098 
3099 static struct attribute *rdev_default_attrs[] = {
3100 	&rdev_state.attr,
3101 	&rdev_errors.attr,
3102 	&rdev_slot.attr,
3103 	&rdev_offset.attr,
3104 	&rdev_new_offset.attr,
3105 	&rdev_size.attr,
3106 	&rdev_recovery_start.attr,
3107 	&rdev_bad_blocks.attr,
3108 	&rdev_unack_bad_blocks.attr,
3109 	NULL,
3110 };
3111 static ssize_t
3112 rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
3113 {
3114 	struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
3115 	struct md_rdev *rdev = container_of(kobj, struct md_rdev, kobj);
3116 	struct mddev *mddev = rdev->mddev;
3117 	ssize_t rv;
3118 
3119 	if (!entry->show)
3120 		return -EIO;
3121 
3122 	rv = mddev ? mddev_lock(mddev) : -EBUSY;
3123 	if (!rv) {
3124 		if (rdev->mddev == NULL)
3125 			rv = -EBUSY;
3126 		else
3127 			rv = entry->show(rdev, page);
3128 		mddev_unlock(mddev);
3129 	}
3130 	return rv;
3131 }
3132 
3133 static ssize_t
3134 rdev_attr_store(struct kobject *kobj, struct attribute *attr,
3135 	      const char *page, size_t length)
3136 {
3137 	struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
3138 	struct md_rdev *rdev = container_of(kobj, struct md_rdev, kobj);
3139 	ssize_t rv;
3140 	struct mddev *mddev = rdev->mddev;
3141 
3142 	if (!entry->store)
3143 		return -EIO;
3144 	if (!capable(CAP_SYS_ADMIN))
3145 		return -EACCES;
3146 	rv = mddev ? mddev_lock(mddev): -EBUSY;
3147 	if (!rv) {
3148 		if (rdev->mddev == NULL)
3149 			rv = -EBUSY;
3150 		else
3151 			rv = entry->store(rdev, page, length);
3152 		mddev_unlock(mddev);
3153 	}
3154 	return rv;
3155 }
3156 
3157 static void rdev_free(struct kobject *ko)
3158 {
3159 	struct md_rdev *rdev = container_of(ko, struct md_rdev, kobj);
3160 	kfree(rdev);
3161 }
3162 static const struct sysfs_ops rdev_sysfs_ops = {
3163 	.show		= rdev_attr_show,
3164 	.store		= rdev_attr_store,
3165 };
3166 static struct kobj_type rdev_ktype = {
3167 	.release	= rdev_free,
3168 	.sysfs_ops	= &rdev_sysfs_ops,
3169 	.default_attrs	= rdev_default_attrs,
3170 };
3171 
3172 int md_rdev_init(struct md_rdev *rdev)
3173 {
3174 	rdev->desc_nr = -1;
3175 	rdev->saved_raid_disk = -1;
3176 	rdev->raid_disk = -1;
3177 	rdev->flags = 0;
3178 	rdev->data_offset = 0;
3179 	rdev->new_data_offset = 0;
3180 	rdev->sb_events = 0;
3181 	rdev->last_read_error.tv_sec  = 0;
3182 	rdev->last_read_error.tv_nsec = 0;
3183 	rdev->sb_loaded = 0;
3184 	rdev->bb_page = NULL;
3185 	atomic_set(&rdev->nr_pending, 0);
3186 	atomic_set(&rdev->read_errors, 0);
3187 	atomic_set(&rdev->corrected_errors, 0);
3188 
3189 	INIT_LIST_HEAD(&rdev->same_set);
3190 	init_waitqueue_head(&rdev->blocked_wait);
3191 
3192 	/* Add space to store bad block list.
3193 	 * This reserves the space even on arrays where it cannot
3194 	 * be used - I wonder if that matters
3195 	 */
3196 	rdev->badblocks.count = 0;
3197 	rdev->badblocks.shift = -1; /* disabled until explicitly enabled */
3198 	rdev->badblocks.page = kmalloc(PAGE_SIZE, GFP_KERNEL);
3199 	seqlock_init(&rdev->badblocks.lock);
3200 	if (rdev->badblocks.page == NULL)
3201 		return -ENOMEM;
3202 
3203 	return 0;
3204 }
3205 EXPORT_SYMBOL_GPL(md_rdev_init);
3206 /*
3207  * Import a device. If 'super_format' >= 0, then sanity check the superblock
3208  *
3209  * mark the device faulty if:
3210  *
3211  *   - the device is nonexistent (zero size)
3212  *   - the device has no valid superblock
3213  *
3214  * a faulty rdev _never_ has rdev->sb set.
3215  */
3216 static struct md_rdev *md_import_device(dev_t newdev, int super_format, int super_minor)
3217 {
3218 	char b[BDEVNAME_SIZE];
3219 	int err;
3220 	struct md_rdev *rdev;
3221 	sector_t size;
3222 
3223 	rdev = kzalloc(sizeof(*rdev), GFP_KERNEL);
3224 	if (!rdev) {
3225 		printk(KERN_ERR "md: could not alloc mem for new device!\n");
3226 		return ERR_PTR(-ENOMEM);
3227 	}
3228 
3229 	err = md_rdev_init(rdev);
3230 	if (err)
3231 		goto abort_free;
3232 	err = alloc_disk_sb(rdev);
3233 	if (err)
3234 		goto abort_free;
3235 
3236 	err = lock_rdev(rdev, newdev, super_format == -2);
3237 	if (err)
3238 		goto abort_free;
3239 
3240 	kobject_init(&rdev->kobj, &rdev_ktype);
3241 
3242 	size = i_size_read(rdev->bdev->bd_inode) >> BLOCK_SIZE_BITS;
3243 	if (!size) {
3244 		printk(KERN_WARNING
3245 			"md: %s has zero or unknown size, marking faulty!\n",
3246 			bdevname(rdev->bdev,b));
3247 		err = -EINVAL;
3248 		goto abort_free;
3249 	}
3250 
3251 	if (super_format >= 0) {
3252 		err = super_types[super_format].
3253 			load_super(rdev, NULL, super_minor);
3254 		if (err == -EINVAL) {
3255 			printk(KERN_WARNING
3256 				"md: %s does not have a valid v%d.%d "
3257 			       "superblock, not importing!\n",
3258 				bdevname(rdev->bdev,b),
3259 			       super_format, super_minor);
3260 			goto abort_free;
3261 		}
3262 		if (err < 0) {
3263 			printk(KERN_WARNING
3264 				"md: could not read %s's sb, not importing!\n",
3265 				bdevname(rdev->bdev,b));
3266 			goto abort_free;
3267 		}
3268 	}
3269 
3270 	return rdev;
3271 
3272 abort_free:
3273 	if (rdev->bdev)
3274 		unlock_rdev(rdev);
3275 	md_rdev_clear(rdev);
3276 	kfree(rdev);
3277 	return ERR_PTR(err);
3278 }
3279 
3280 /*
3281  * Check a full RAID array for plausibility
3282  */
3283 
3284 
3285 static void analyze_sbs(struct mddev * mddev)
3286 {
3287 	int i;
3288 	struct md_rdev *rdev, *freshest, *tmp;
3289 	char b[BDEVNAME_SIZE];
3290 
3291 	freshest = NULL;
3292 	rdev_for_each_safe(rdev, tmp, mddev)
3293 		switch (super_types[mddev->major_version].
3294 			load_super(rdev, freshest, mddev->minor_version)) {
3295 		case 1:
3296 			freshest = rdev;
3297 			break;
3298 		case 0:
3299 			break;
3300 		default:
3301 			printk( KERN_ERR \
3302 				"md: fatal superblock inconsistency in %s"
3303 				" -- removing from array\n",
3304 				bdevname(rdev->bdev,b));
3305 			kick_rdev_from_array(rdev);
3306 		}
3307 
3308 
3309 	super_types[mddev->major_version].
3310 		validate_super(mddev, freshest);
3311 
3312 	i = 0;
3313 	rdev_for_each_safe(rdev, tmp, mddev) {
3314 		if (mddev->max_disks &&
3315 		    (rdev->desc_nr >= mddev->max_disks ||
3316 		     i > mddev->max_disks)) {
3317 			printk(KERN_WARNING
3318 			       "md: %s: %s: only %d devices permitted\n",
3319 			       mdname(mddev), bdevname(rdev->bdev, b),
3320 			       mddev->max_disks);
3321 			kick_rdev_from_array(rdev);
3322 			continue;
3323 		}
3324 		if (rdev != freshest)
3325 			if (super_types[mddev->major_version].
3326 			    validate_super(mddev, rdev)) {
3327 				printk(KERN_WARNING "md: kicking non-fresh %s"
3328 					" from array!\n",
3329 					bdevname(rdev->bdev,b));
3330 				kick_rdev_from_array(rdev);
3331 				continue;
3332 			}
3333 		if (mddev->level == LEVEL_MULTIPATH) {
3334 			rdev->desc_nr = i++;
3335 			rdev->raid_disk = rdev->desc_nr;
3336 			set_bit(In_sync, &rdev->flags);
3337 		} else if (rdev->raid_disk >= (mddev->raid_disks - min(0, mddev->delta_disks))) {
3338 			rdev->raid_disk = -1;
3339 			clear_bit(In_sync, &rdev->flags);
3340 		}
3341 	}
3342 }
3343 
3344 /* Read a fixed-point number.
3345  * Numbers in sysfs attributes should be in "standard" units where
3346  * possible, so time should be in seconds.
3347  * However we internally use a a much smaller unit such as
3348  * milliseconds or jiffies.
3349  * This function takes a decimal number with a possible fractional
3350  * component, and produces an integer which is the result of
3351  * multiplying that number by 10^'scale'.
3352  * all without any floating-point arithmetic.
3353  */
3354 int strict_strtoul_scaled(const char *cp, unsigned long *res, int scale)
3355 {
3356 	unsigned long result = 0;
3357 	long decimals = -1;
3358 	while (isdigit(*cp) || (*cp == '.' && decimals < 0)) {
3359 		if (*cp == '.')
3360 			decimals = 0;
3361 		else if (decimals < scale) {
3362 			unsigned int value;
3363 			value = *cp - '0';
3364 			result = result * 10 + value;
3365 			if (decimals >= 0)
3366 				decimals++;
3367 		}
3368 		cp++;
3369 	}
3370 	if (*cp == '\n')
3371 		cp++;
3372 	if (*cp)
3373 		return -EINVAL;
3374 	if (decimals < 0)
3375 		decimals = 0;
3376 	while (decimals < scale) {
3377 		result *= 10;
3378 		decimals ++;
3379 	}
3380 	*res = result;
3381 	return 0;
3382 }
3383 
3384 
3385 static void md_safemode_timeout(unsigned long data);
3386 
3387 static ssize_t
3388 safe_delay_show(struct mddev *mddev, char *page)
3389 {
3390 	int msec = (mddev->safemode_delay*1000)/HZ;
3391 	return sprintf(page, "%d.%03d\n", msec/1000, msec%1000);
3392 }
3393 static ssize_t
3394 safe_delay_store(struct mddev *mddev, const char *cbuf, size_t len)
3395 {
3396 	unsigned long msec;
3397 
3398 	if (strict_strtoul_scaled(cbuf, &msec, 3) < 0)
3399 		return -EINVAL;
3400 	if (msec == 0)
3401 		mddev->safemode_delay = 0;
3402 	else {
3403 		unsigned long old_delay = mddev->safemode_delay;
3404 		mddev->safemode_delay = (msec*HZ)/1000;
3405 		if (mddev->safemode_delay == 0)
3406 			mddev->safemode_delay = 1;
3407 		if (mddev->safemode_delay < old_delay || old_delay == 0)
3408 			md_safemode_timeout((unsigned long)mddev);
3409 	}
3410 	return len;
3411 }
3412 static struct md_sysfs_entry md_safe_delay =
3413 __ATTR(safe_mode_delay, S_IRUGO|S_IWUSR,safe_delay_show, safe_delay_store);
3414 
3415 static ssize_t
3416 level_show(struct mddev *mddev, char *page)
3417 {
3418 	struct md_personality *p = mddev->pers;
3419 	if (p)
3420 		return sprintf(page, "%s\n", p->name);
3421 	else if (mddev->clevel[0])
3422 		return sprintf(page, "%s\n", mddev->clevel);
3423 	else if (mddev->level != LEVEL_NONE)
3424 		return sprintf(page, "%d\n", mddev->level);
3425 	else
3426 		return 0;
3427 }
3428 
3429 static ssize_t
3430 level_store(struct mddev *mddev, const char *buf, size_t len)
3431 {
3432 	char clevel[16];
3433 	ssize_t rv = len;
3434 	struct md_personality *pers;
3435 	long level;
3436 	void *priv;
3437 	struct md_rdev *rdev;
3438 
3439 	if (mddev->pers == NULL) {
3440 		if (len == 0)
3441 			return 0;
3442 		if (len >= sizeof(mddev->clevel))
3443 			return -ENOSPC;
3444 		strncpy(mddev->clevel, buf, len);
3445 		if (mddev->clevel[len-1] == '\n')
3446 			len--;
3447 		mddev->clevel[len] = 0;
3448 		mddev->level = LEVEL_NONE;
3449 		return rv;
3450 	}
3451 	if (mddev->ro)
3452 		return  -EROFS;
3453 
3454 	/* request to change the personality.  Need to ensure:
3455 	 *  - array is not engaged in resync/recovery/reshape
3456 	 *  - old personality can be suspended
3457 	 *  - new personality will access other array.
3458 	 */
3459 
3460 	if (mddev->sync_thread ||
3461 	    mddev->reshape_position != MaxSector ||
3462 	    mddev->sysfs_active)
3463 		return -EBUSY;
3464 
3465 	if (!mddev->pers->quiesce) {
3466 		printk(KERN_WARNING "md: %s: %s does not support online personality change\n",
3467 		       mdname(mddev), mddev->pers->name);
3468 		return -EINVAL;
3469 	}
3470 
3471 	/* Now find the new personality */
3472 	if (len == 0 || len >= sizeof(clevel))
3473 		return -EINVAL;
3474 	strncpy(clevel, buf, len);
3475 	if (clevel[len-1] == '\n')
3476 		len--;
3477 	clevel[len] = 0;
3478 	if (kstrtol(clevel, 10, &level))
3479 		level = LEVEL_NONE;
3480 
3481 	if (request_module("md-%s", clevel) != 0)
3482 		request_module("md-level-%s", clevel);
3483 	spin_lock(&pers_lock);
3484 	pers = find_pers(level, clevel);
3485 	if (!pers || !try_module_get(pers->owner)) {
3486 		spin_unlock(&pers_lock);
3487 		printk(KERN_WARNING "md: personality %s not loaded\n", clevel);
3488 		return -EINVAL;
3489 	}
3490 	spin_unlock(&pers_lock);
3491 
3492 	if (pers == mddev->pers) {
3493 		/* Nothing to do! */
3494 		module_put(pers->owner);
3495 		return rv;
3496 	}
3497 	if (!pers->takeover) {
3498 		module_put(pers->owner);
3499 		printk(KERN_WARNING "md: %s: %s does not support personality takeover\n",
3500 		       mdname(mddev), clevel);
3501 		return -EINVAL;
3502 	}
3503 
3504 	rdev_for_each(rdev, mddev)
3505 		rdev->new_raid_disk = rdev->raid_disk;
3506 
3507 	/* ->takeover must set new_* and/or delta_disks
3508 	 * if it succeeds, and may set them when it fails.
3509 	 */
3510 	priv = pers->takeover(mddev);
3511 	if (IS_ERR(priv)) {
3512 		mddev->new_level = mddev->level;
3513 		mddev->new_layout = mddev->layout;
3514 		mddev->new_chunk_sectors = mddev->chunk_sectors;
3515 		mddev->raid_disks -= mddev->delta_disks;
3516 		mddev->delta_disks = 0;
3517 		mddev->reshape_backwards = 0;
3518 		module_put(pers->owner);
3519 		printk(KERN_WARNING "md: %s: %s would not accept array\n",
3520 		       mdname(mddev), clevel);
3521 		return PTR_ERR(priv);
3522 	}
3523 
3524 	/* Looks like we have a winner */
3525 	mddev_suspend(mddev);
3526 	mddev->pers->stop(mddev);
3527 
3528 	if (mddev->pers->sync_request == NULL &&
3529 	    pers->sync_request != NULL) {
3530 		/* need to add the md_redundancy_group */
3531 		if (sysfs_create_group(&mddev->kobj, &md_redundancy_group))
3532 			printk(KERN_WARNING
3533 			       "md: cannot register extra attributes for %s\n",
3534 			       mdname(mddev));
3535 		mddev->sysfs_action = sysfs_get_dirent(mddev->kobj.sd, "sync_action");
3536 	}
3537 	if (mddev->pers->sync_request != NULL &&
3538 	    pers->sync_request == NULL) {
3539 		/* need to remove the md_redundancy_group */
3540 		if (mddev->to_remove == NULL)
3541 			mddev->to_remove = &md_redundancy_group;
3542 	}
3543 
3544 	if (mddev->pers->sync_request == NULL &&
3545 	    mddev->external) {
3546 		/* We are converting from a no-redundancy array
3547 		 * to a redundancy array and metadata is managed
3548 		 * externally so we need to be sure that writes
3549 		 * won't block due to a need to transition
3550 		 *      clean->dirty
3551 		 * until external management is started.
3552 		 */
3553 		mddev->in_sync = 0;
3554 		mddev->safemode_delay = 0;
3555 		mddev->safemode = 0;
3556 	}
3557 
3558 	rdev_for_each(rdev, mddev) {
3559 		if (rdev->raid_disk < 0)
3560 			continue;
3561 		if (rdev->new_raid_disk >= mddev->raid_disks)
3562 			rdev->new_raid_disk = -1;
3563 		if (rdev->new_raid_disk == rdev->raid_disk)
3564 			continue;
3565 		sysfs_unlink_rdev(mddev, rdev);
3566 	}
3567 	rdev_for_each(rdev, mddev) {
3568 		if (rdev->raid_disk < 0)
3569 			continue;
3570 		if (rdev->new_raid_disk == rdev->raid_disk)
3571 			continue;
3572 		rdev->raid_disk = rdev->new_raid_disk;
3573 		if (rdev->raid_disk < 0)
3574 			clear_bit(In_sync, &rdev->flags);
3575 		else {
3576 			if (sysfs_link_rdev(mddev, rdev))
3577 				printk(KERN_WARNING "md: cannot register rd%d"
3578 				       " for %s after level change\n",
3579 				       rdev->raid_disk, mdname(mddev));
3580 		}
3581 	}
3582 
3583 	module_put(mddev->pers->owner);
3584 	mddev->pers = pers;
3585 	mddev->private = priv;
3586 	strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
3587 	mddev->level = mddev->new_level;
3588 	mddev->layout = mddev->new_layout;
3589 	mddev->chunk_sectors = mddev->new_chunk_sectors;
3590 	mddev->delta_disks = 0;
3591 	mddev->reshape_backwards = 0;
3592 	mddev->degraded = 0;
3593 	if (mddev->pers->sync_request == NULL) {
3594 		/* this is now an array without redundancy, so
3595 		 * it must always be in_sync
3596 		 */
3597 		mddev->in_sync = 1;
3598 		del_timer_sync(&mddev->safemode_timer);
3599 	}
3600 	blk_set_stacking_limits(&mddev->queue->limits);
3601 	pers->run(mddev);
3602 	set_bit(MD_CHANGE_DEVS, &mddev->flags);
3603 	mddev_resume(mddev);
3604 	if (!mddev->thread)
3605 		md_update_sb(mddev, 1);
3606 	sysfs_notify(&mddev->kobj, NULL, "level");
3607 	md_new_event(mddev);
3608 	return rv;
3609 }
3610 
3611 static struct md_sysfs_entry md_level =
3612 __ATTR(level, S_IRUGO|S_IWUSR, level_show, level_store);
3613 
3614 
3615 static ssize_t
3616 layout_show(struct mddev *mddev, char *page)
3617 {
3618 	/* just a number, not meaningful for all levels */
3619 	if (mddev->reshape_position != MaxSector &&
3620 	    mddev->layout != mddev->new_layout)
3621 		return sprintf(page, "%d (%d)\n",
3622 			       mddev->new_layout, mddev->layout);
3623 	return sprintf(page, "%d\n", mddev->layout);
3624 }
3625 
3626 static ssize_t
3627 layout_store(struct mddev *mddev, const char *buf, size_t len)
3628 {
3629 	char *e;
3630 	unsigned long n = simple_strtoul(buf, &e, 10);
3631 
3632 	if (!*buf || (*e && *e != '\n'))
3633 		return -EINVAL;
3634 
3635 	if (mddev->pers) {
3636 		int err;
3637 		if (mddev->pers->check_reshape == NULL)
3638 			return -EBUSY;
3639 		if (mddev->ro)
3640 			return -EROFS;
3641 		mddev->new_layout = n;
3642 		err = mddev->pers->check_reshape(mddev);
3643 		if (err) {
3644 			mddev->new_layout = mddev->layout;
3645 			return err;
3646 		}
3647 	} else {
3648 		mddev->new_layout = n;
3649 		if (mddev->reshape_position == MaxSector)
3650 			mddev->layout = n;
3651 	}
3652 	return len;
3653 }
3654 static struct md_sysfs_entry md_layout =
3655 __ATTR(layout, S_IRUGO|S_IWUSR, layout_show, layout_store);
3656 
3657 
3658 static ssize_t
3659 raid_disks_show(struct mddev *mddev, char *page)
3660 {
3661 	if (mddev->raid_disks == 0)
3662 		return 0;
3663 	if (mddev->reshape_position != MaxSector &&
3664 	    mddev->delta_disks != 0)
3665 		return sprintf(page, "%d (%d)\n", mddev->raid_disks,
3666 			       mddev->raid_disks - mddev->delta_disks);
3667 	return sprintf(page, "%d\n", mddev->raid_disks);
3668 }
3669 
3670 static int update_raid_disks(struct mddev *mddev, int raid_disks);
3671 
3672 static ssize_t
3673 raid_disks_store(struct mddev *mddev, const char *buf, size_t len)
3674 {
3675 	char *e;
3676 	int rv = 0;
3677 	unsigned long n = simple_strtoul(buf, &e, 10);
3678 
3679 	if (!*buf || (*e && *e != '\n'))
3680 		return -EINVAL;
3681 
3682 	if (mddev->pers)
3683 		rv = update_raid_disks(mddev, n);
3684 	else if (mddev->reshape_position != MaxSector) {
3685 		struct md_rdev *rdev;
3686 		int olddisks = mddev->raid_disks - mddev->delta_disks;
3687 
3688 		rdev_for_each(rdev, mddev) {
3689 			if (olddisks < n &&
3690 			    rdev->data_offset < rdev->new_data_offset)
3691 				return -EINVAL;
3692 			if (olddisks > n &&
3693 			    rdev->data_offset > rdev->new_data_offset)
3694 				return -EINVAL;
3695 		}
3696 		mddev->delta_disks = n - olddisks;
3697 		mddev->raid_disks = n;
3698 		mddev->reshape_backwards = (mddev->delta_disks < 0);
3699 	} else
3700 		mddev->raid_disks = n;
3701 	return rv ? rv : len;
3702 }
3703 static struct md_sysfs_entry md_raid_disks =
3704 __ATTR(raid_disks, S_IRUGO|S_IWUSR, raid_disks_show, raid_disks_store);
3705 
3706 static ssize_t
3707 chunk_size_show(struct mddev *mddev, char *page)
3708 {
3709 	if (mddev->reshape_position != MaxSector &&
3710 	    mddev->chunk_sectors != mddev->new_chunk_sectors)
3711 		return sprintf(page, "%d (%d)\n",
3712 			       mddev->new_chunk_sectors << 9,
3713 			       mddev->chunk_sectors << 9);
3714 	return sprintf(page, "%d\n", mddev->chunk_sectors << 9);
3715 }
3716 
3717 static ssize_t
3718 chunk_size_store(struct mddev *mddev, const char *buf, size_t len)
3719 {
3720 	char *e;
3721 	unsigned long n = simple_strtoul(buf, &e, 10);
3722 
3723 	if (!*buf || (*e && *e != '\n'))
3724 		return -EINVAL;
3725 
3726 	if (mddev->pers) {
3727 		int err;
3728 		if (mddev->pers->check_reshape == NULL)
3729 			return -EBUSY;
3730 		if (mddev->ro)
3731 			return -EROFS;
3732 		mddev->new_chunk_sectors = n >> 9;
3733 		err = mddev->pers->check_reshape(mddev);
3734 		if (err) {
3735 			mddev->new_chunk_sectors = mddev->chunk_sectors;
3736 			return err;
3737 		}
3738 	} else {
3739 		mddev->new_chunk_sectors = n >> 9;
3740 		if (mddev->reshape_position == MaxSector)
3741 			mddev->chunk_sectors = n >> 9;
3742 	}
3743 	return len;
3744 }
3745 static struct md_sysfs_entry md_chunk_size =
3746 __ATTR(chunk_size, S_IRUGO|S_IWUSR, chunk_size_show, chunk_size_store);
3747 
3748 static ssize_t
3749 resync_start_show(struct mddev *mddev, char *page)
3750 {
3751 	if (mddev->recovery_cp == MaxSector)
3752 		return sprintf(page, "none\n");
3753 	return sprintf(page, "%llu\n", (unsigned long long)mddev->recovery_cp);
3754 }
3755 
3756 static ssize_t
3757 resync_start_store(struct mddev *mddev, const char *buf, size_t len)
3758 {
3759 	char *e;
3760 	unsigned long long n = simple_strtoull(buf, &e, 10);
3761 
3762 	if (mddev->pers && !test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
3763 		return -EBUSY;
3764 	if (cmd_match(buf, "none"))
3765 		n = MaxSector;
3766 	else if (!*buf || (*e && *e != '\n'))
3767 		return -EINVAL;
3768 
3769 	mddev->recovery_cp = n;
3770 	if (mddev->pers)
3771 		set_bit(MD_CHANGE_CLEAN, &mddev->flags);
3772 	return len;
3773 }
3774 static struct md_sysfs_entry md_resync_start =
3775 __ATTR(resync_start, S_IRUGO|S_IWUSR, resync_start_show, resync_start_store);
3776 
3777 /*
3778  * The array state can be:
3779  *
3780  * clear
3781  *     No devices, no size, no level
3782  *     Equivalent to STOP_ARRAY ioctl
3783  * inactive
3784  *     May have some settings, but array is not active
3785  *        all IO results in error
3786  *     When written, doesn't tear down array, but just stops it
3787  * suspended (not supported yet)
3788  *     All IO requests will block. The array can be reconfigured.
3789  *     Writing this, if accepted, will block until array is quiescent
3790  * readonly
3791  *     no resync can happen.  no superblocks get written.
3792  *     write requests fail
3793  * read-auto
3794  *     like readonly, but behaves like 'clean' on a write request.
3795  *
3796  * clean - no pending writes, but otherwise active.
3797  *     When written to inactive array, starts without resync
3798  *     If a write request arrives then
3799  *       if metadata is known, mark 'dirty' and switch to 'active'.
3800  *       if not known, block and switch to write-pending
3801  *     If written to an active array that has pending writes, then fails.
3802  * active
3803  *     fully active: IO and resync can be happening.
3804  *     When written to inactive array, starts with resync
3805  *
3806  * write-pending
3807  *     clean, but writes are blocked waiting for 'active' to be written.
3808  *
3809  * active-idle
3810  *     like active, but no writes have been seen for a while (100msec).
3811  *
3812  */
3813 enum array_state { clear, inactive, suspended, readonly, read_auto, clean, active,
3814 		   write_pending, active_idle, bad_word};
3815 static char *array_states[] = {
3816 	"clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active",
3817 	"write-pending", "active-idle", NULL };
3818 
3819 static int match_word(const char *word, char **list)
3820 {
3821 	int n;
3822 	for (n=0; list[n]; n++)
3823 		if (cmd_match(word, list[n]))
3824 			break;
3825 	return n;
3826 }
3827 
3828 static ssize_t
3829 array_state_show(struct mddev *mddev, char *page)
3830 {
3831 	enum array_state st = inactive;
3832 
3833 	if (mddev->pers)
3834 		switch(mddev->ro) {
3835 		case 1:
3836 			st = readonly;
3837 			break;
3838 		case 2:
3839 			st = read_auto;
3840 			break;
3841 		case 0:
3842 			if (mddev->in_sync)
3843 				st = clean;
3844 			else if (test_bit(MD_CHANGE_PENDING, &mddev->flags))
3845 				st = write_pending;
3846 			else if (mddev->safemode)
3847 				st = active_idle;
3848 			else
3849 				st = active;
3850 		}
3851 	else {
3852 		if (list_empty(&mddev->disks) &&
3853 		    mddev->raid_disks == 0 &&
3854 		    mddev->dev_sectors == 0)
3855 			st = clear;
3856 		else
3857 			st = inactive;
3858 	}
3859 	return sprintf(page, "%s\n", array_states[st]);
3860 }
3861 
3862 static int do_md_stop(struct mddev * mddev, int ro, struct block_device *bdev);
3863 static int md_set_readonly(struct mddev * mddev, struct block_device *bdev);
3864 static int do_md_run(struct mddev * mddev);
3865 static int restart_array(struct mddev *mddev);
3866 
3867 static ssize_t
3868 array_state_store(struct mddev *mddev, const char *buf, size_t len)
3869 {
3870 	int err = -EINVAL;
3871 	enum array_state st = match_word(buf, array_states);
3872 	switch(st) {
3873 	case bad_word:
3874 		break;
3875 	case clear:
3876 		/* stopping an active array */
3877 		err = do_md_stop(mddev, 0, NULL);
3878 		break;
3879 	case inactive:
3880 		/* stopping an active array */
3881 		if (mddev->pers)
3882 			err = do_md_stop(mddev, 2, NULL);
3883 		else
3884 			err = 0; /* already inactive */
3885 		break;
3886 	case suspended:
3887 		break; /* not supported yet */
3888 	case readonly:
3889 		if (mddev->pers)
3890 			err = md_set_readonly(mddev, NULL);
3891 		else {
3892 			mddev->ro = 1;
3893 			set_disk_ro(mddev->gendisk, 1);
3894 			err = do_md_run(mddev);
3895 		}
3896 		break;
3897 	case read_auto:
3898 		if (mddev->pers) {
3899 			if (mddev->ro == 0)
3900 				err = md_set_readonly(mddev, NULL);
3901 			else if (mddev->ro == 1)
3902 				err = restart_array(mddev);
3903 			if (err == 0) {
3904 				mddev->ro = 2;
3905 				set_disk_ro(mddev->gendisk, 0);
3906 			}
3907 		} else {
3908 			mddev->ro = 2;
3909 			err = do_md_run(mddev);
3910 		}
3911 		break;
3912 	case clean:
3913 		if (mddev->pers) {
3914 			restart_array(mddev);
3915 			spin_lock_irq(&mddev->write_lock);
3916 			if (atomic_read(&mddev->writes_pending) == 0) {
3917 				if (mddev->in_sync == 0) {
3918 					mddev->in_sync = 1;
3919 					if (mddev->safemode == 1)
3920 						mddev->safemode = 0;
3921 					set_bit(MD_CHANGE_CLEAN, &mddev->flags);
3922 				}
3923 				err = 0;
3924 			} else
3925 				err = -EBUSY;
3926 			spin_unlock_irq(&mddev->write_lock);
3927 		} else
3928 			err = -EINVAL;
3929 		break;
3930 	case active:
3931 		if (mddev->pers) {
3932 			restart_array(mddev);
3933 			clear_bit(MD_CHANGE_PENDING, &mddev->flags);
3934 			wake_up(&mddev->sb_wait);
3935 			err = 0;
3936 		} else {
3937 			mddev->ro = 0;
3938 			set_disk_ro(mddev->gendisk, 0);
3939 			err = do_md_run(mddev);
3940 		}
3941 		break;
3942 	case write_pending:
3943 	case active_idle:
3944 		/* these cannot be set */
3945 		break;
3946 	}
3947 	if (err)
3948 		return err;
3949 	else {
3950 		if (mddev->hold_active == UNTIL_IOCTL)
3951 			mddev->hold_active = 0;
3952 		sysfs_notify_dirent_safe(mddev->sysfs_state);
3953 		return len;
3954 	}
3955 }
3956 static struct md_sysfs_entry md_array_state =
3957 __ATTR(array_state, S_IRUGO|S_IWUSR, array_state_show, array_state_store);
3958 
3959 static ssize_t
3960 max_corrected_read_errors_show(struct mddev *mddev, char *page) {
3961 	return sprintf(page, "%d\n",
3962 		       atomic_read(&mddev->max_corr_read_errors));
3963 }
3964 
3965 static ssize_t
3966 max_corrected_read_errors_store(struct mddev *mddev, const char *buf, size_t len)
3967 {
3968 	char *e;
3969 	unsigned long n = simple_strtoul(buf, &e, 10);
3970 
3971 	if (*buf && (*e == 0 || *e == '\n')) {
3972 		atomic_set(&mddev->max_corr_read_errors, n);
3973 		return len;
3974 	}
3975 	return -EINVAL;
3976 }
3977 
3978 static struct md_sysfs_entry max_corr_read_errors =
3979 __ATTR(max_read_errors, S_IRUGO|S_IWUSR, max_corrected_read_errors_show,
3980 	max_corrected_read_errors_store);
3981 
3982 static ssize_t
3983 null_show(struct mddev *mddev, char *page)
3984 {
3985 	return -EINVAL;
3986 }
3987 
3988 static ssize_t
3989 new_dev_store(struct mddev *mddev, const char *buf, size_t len)
3990 {
3991 	/* buf must be %d:%d\n? giving major and minor numbers */
3992 	/* The new device is added to the array.
3993 	 * If the array has a persistent superblock, we read the
3994 	 * superblock to initialise info and check validity.
3995 	 * Otherwise, only checking done is that in bind_rdev_to_array,
3996 	 * which mainly checks size.
3997 	 */
3998 	char *e;
3999 	int major = simple_strtoul(buf, &e, 10);
4000 	int minor;
4001 	dev_t dev;
4002 	struct md_rdev *rdev;
4003 	int err;
4004 
4005 	if (!*buf || *e != ':' || !e[1] || e[1] == '\n')
4006 		return -EINVAL;
4007 	minor = simple_strtoul(e+1, &e, 10);
4008 	if (*e && *e != '\n')
4009 		return -EINVAL;
4010 	dev = MKDEV(major, minor);
4011 	if (major != MAJOR(dev) ||
4012 	    minor != MINOR(dev))
4013 		return -EOVERFLOW;
4014 
4015 
4016 	if (mddev->persistent) {
4017 		rdev = md_import_device(dev, mddev->major_version,
4018 					mddev->minor_version);
4019 		if (!IS_ERR(rdev) && !list_empty(&mddev->disks)) {
4020 			struct md_rdev *rdev0
4021 				= list_entry(mddev->disks.next,
4022 					     struct md_rdev, same_set);
4023 			err = super_types[mddev->major_version]
4024 				.load_super(rdev, rdev0, mddev->minor_version);
4025 			if (err < 0)
4026 				goto out;
4027 		}
4028 	} else if (mddev->external)
4029 		rdev = md_import_device(dev, -2, -1);
4030 	else
4031 		rdev = md_import_device(dev, -1, -1);
4032 
4033 	if (IS_ERR(rdev))
4034 		return PTR_ERR(rdev);
4035 	err = bind_rdev_to_array(rdev, mddev);
4036  out:
4037 	if (err)
4038 		export_rdev(rdev);
4039 	return err ? err : len;
4040 }
4041 
4042 static struct md_sysfs_entry md_new_device =
4043 __ATTR(new_dev, S_IWUSR, null_show, new_dev_store);
4044 
4045 static ssize_t
4046 bitmap_store(struct mddev *mddev, const char *buf, size_t len)
4047 {
4048 	char *end;
4049 	unsigned long chunk, end_chunk;
4050 
4051 	if (!mddev->bitmap)
4052 		goto out;
4053 	/* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */
4054 	while (*buf) {
4055 		chunk = end_chunk = simple_strtoul(buf, &end, 0);
4056 		if (buf == end) break;
4057 		if (*end == '-') { /* range */
4058 			buf = end + 1;
4059 			end_chunk = simple_strtoul(buf, &end, 0);
4060 			if (buf == end) break;
4061 		}
4062 		if (*end && !isspace(*end)) break;
4063 		bitmap_dirty_bits(mddev->bitmap, chunk, end_chunk);
4064 		buf = skip_spaces(end);
4065 	}
4066 	bitmap_unplug(mddev->bitmap); /* flush the bits to disk */
4067 out:
4068 	return len;
4069 }
4070 
4071 static struct md_sysfs_entry md_bitmap =
4072 __ATTR(bitmap_set_bits, S_IWUSR, null_show, bitmap_store);
4073 
4074 static ssize_t
4075 size_show(struct mddev *mddev, char *page)
4076 {
4077 	return sprintf(page, "%llu\n",
4078 		(unsigned long long)mddev->dev_sectors / 2);
4079 }
4080 
4081 static int update_size(struct mddev *mddev, sector_t num_sectors);
4082 
4083 static ssize_t
4084 size_store(struct mddev *mddev, const char *buf, size_t len)
4085 {
4086 	/* If array is inactive, we can reduce the component size, but
4087 	 * not increase it (except from 0).
4088 	 * If array is active, we can try an on-line resize
4089 	 */
4090 	sector_t sectors;
4091 	int err = strict_blocks_to_sectors(buf, &sectors);
4092 
4093 	if (err < 0)
4094 		return err;
4095 	if (mddev->pers) {
4096 		err = update_size(mddev, sectors);
4097 		md_update_sb(mddev, 1);
4098 	} else {
4099 		if (mddev->dev_sectors == 0 ||
4100 		    mddev->dev_sectors > sectors)
4101 			mddev->dev_sectors = sectors;
4102 		else
4103 			err = -ENOSPC;
4104 	}
4105 	return err ? err : len;
4106 }
4107 
4108 static struct md_sysfs_entry md_size =
4109 __ATTR(component_size, S_IRUGO|S_IWUSR, size_show, size_store);
4110 
4111 
4112 /* Metadata version.
4113  * This is one of
4114  *   'none' for arrays with no metadata (good luck...)
4115  *   'external' for arrays with externally managed metadata,
4116  * or N.M for internally known formats
4117  */
4118 static ssize_t
4119 metadata_show(struct mddev *mddev, char *page)
4120 {
4121 	if (mddev->persistent)
4122 		return sprintf(page, "%d.%d\n",
4123 			       mddev->major_version, mddev->minor_version);
4124 	else if (mddev->external)
4125 		return sprintf(page, "external:%s\n", mddev->metadata_type);
4126 	else
4127 		return sprintf(page, "none\n");
4128 }
4129 
4130 static ssize_t
4131 metadata_store(struct mddev *mddev, const char *buf, size_t len)
4132 {
4133 	int major, minor;
4134 	char *e;
4135 	/* Changing the details of 'external' metadata is
4136 	 * always permitted.  Otherwise there must be
4137 	 * no devices attached to the array.
4138 	 */
4139 	if (mddev->external && strncmp(buf, "external:", 9) == 0)
4140 		;
4141 	else if (!list_empty(&mddev->disks))
4142 		return -EBUSY;
4143 
4144 	if (cmd_match(buf, "none")) {
4145 		mddev->persistent = 0;
4146 		mddev->external = 0;
4147 		mddev->major_version = 0;
4148 		mddev->minor_version = 90;
4149 		return len;
4150 	}
4151 	if (strncmp(buf, "external:", 9) == 0) {
4152 		size_t namelen = len-9;
4153 		if (namelen >= sizeof(mddev->metadata_type))
4154 			namelen = sizeof(mddev->metadata_type)-1;
4155 		strncpy(mddev->metadata_type, buf+9, namelen);
4156 		mddev->metadata_type[namelen] = 0;
4157 		if (namelen && mddev->metadata_type[namelen-1] == '\n')
4158 			mddev->metadata_type[--namelen] = 0;
4159 		mddev->persistent = 0;
4160 		mddev->external = 1;
4161 		mddev->major_version = 0;
4162 		mddev->minor_version = 90;
4163 		return len;
4164 	}
4165 	major = simple_strtoul(buf, &e, 10);
4166 	if (e==buf || *e != '.')
4167 		return -EINVAL;
4168 	buf = e+1;
4169 	minor = simple_strtoul(buf, &e, 10);
4170 	if (e==buf || (*e && *e != '\n') )
4171 		return -EINVAL;
4172 	if (major >= ARRAY_SIZE(super_types) || super_types[major].name == NULL)
4173 		return -ENOENT;
4174 	mddev->major_version = major;
4175 	mddev->minor_version = minor;
4176 	mddev->persistent = 1;
4177 	mddev->external = 0;
4178 	return len;
4179 }
4180 
4181 static struct md_sysfs_entry md_metadata =
4182 __ATTR(metadata_version, S_IRUGO|S_IWUSR, metadata_show, metadata_store);
4183 
4184 static ssize_t
4185 action_show(struct mddev *mddev, char *page)
4186 {
4187 	char *type = "idle";
4188 	if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
4189 		type = "frozen";
4190 	else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
4191 	    (!mddev->ro && test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))) {
4192 		if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
4193 			type = "reshape";
4194 		else if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
4195 			if (!test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
4196 				type = "resync";
4197 			else if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
4198 				type = "check";
4199 			else
4200 				type = "repair";
4201 		} else if (test_bit(MD_RECOVERY_RECOVER, &mddev->recovery))
4202 			type = "recover";
4203 	}
4204 	return sprintf(page, "%s\n", type);
4205 }
4206 
4207 static ssize_t
4208 action_store(struct mddev *mddev, const char *page, size_t len)
4209 {
4210 	if (!mddev->pers || !mddev->pers->sync_request)
4211 		return -EINVAL;
4212 
4213 	if (cmd_match(page, "frozen"))
4214 		set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4215 	else
4216 		clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4217 
4218 	if (cmd_match(page, "idle") || cmd_match(page, "frozen")) {
4219 		if (mddev->sync_thread) {
4220 			set_bit(MD_RECOVERY_INTR, &mddev->recovery);
4221 			md_reap_sync_thread(mddev);
4222 		}
4223 	} else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
4224 		   test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
4225 		return -EBUSY;
4226 	else if (cmd_match(page, "resync"))
4227 		set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4228 	else if (cmd_match(page, "recover")) {
4229 		set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
4230 		set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4231 	} else if (cmd_match(page, "reshape")) {
4232 		int err;
4233 		if (mddev->pers->start_reshape == NULL)
4234 			return -EINVAL;
4235 		err = mddev->pers->start_reshape(mddev);
4236 		if (err)
4237 			return err;
4238 		sysfs_notify(&mddev->kobj, NULL, "degraded");
4239 	} else {
4240 		if (cmd_match(page, "check"))
4241 			set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
4242 		else if (!cmd_match(page, "repair"))
4243 			return -EINVAL;
4244 		set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
4245 		set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
4246 	}
4247 	if (mddev->ro == 2) {
4248 		/* A write to sync_action is enough to justify
4249 		 * canceling read-auto mode
4250 		 */
4251 		mddev->ro = 0;
4252 		md_wakeup_thread(mddev->sync_thread);
4253 	}
4254 	set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4255 	md_wakeup_thread(mddev->thread);
4256 	sysfs_notify_dirent_safe(mddev->sysfs_action);
4257 	return len;
4258 }
4259 
4260 static struct md_sysfs_entry md_scan_mode =
4261 __ATTR(sync_action, S_IRUGO|S_IWUSR, action_show, action_store);
4262 
4263 static ssize_t
4264 last_sync_action_show(struct mddev *mddev, char *page)
4265 {
4266 	return sprintf(page, "%s\n", mddev->last_sync_action);
4267 }
4268 
4269 static struct md_sysfs_entry md_last_scan_mode = __ATTR_RO(last_sync_action);
4270 
4271 static ssize_t
4272 mismatch_cnt_show(struct mddev *mddev, char *page)
4273 {
4274 	return sprintf(page, "%llu\n",
4275 		       (unsigned long long)
4276 		       atomic64_read(&mddev->resync_mismatches));
4277 }
4278 
4279 static struct md_sysfs_entry md_mismatches = __ATTR_RO(mismatch_cnt);
4280 
4281 static ssize_t
4282 sync_min_show(struct mddev *mddev, char *page)
4283 {
4284 	return sprintf(page, "%d (%s)\n", speed_min(mddev),
4285 		       mddev->sync_speed_min ? "local": "system");
4286 }
4287 
4288 static ssize_t
4289 sync_min_store(struct mddev *mddev, const char *buf, size_t len)
4290 {
4291 	int min;
4292 	char *e;
4293 	if (strncmp(buf, "system", 6)==0) {
4294 		mddev->sync_speed_min = 0;
4295 		return len;
4296 	}
4297 	min = simple_strtoul(buf, &e, 10);
4298 	if (buf == e || (*e && *e != '\n') || min <= 0)
4299 		return -EINVAL;
4300 	mddev->sync_speed_min = min;
4301 	return len;
4302 }
4303 
4304 static struct md_sysfs_entry md_sync_min =
4305 __ATTR(sync_speed_min, S_IRUGO|S_IWUSR, sync_min_show, sync_min_store);
4306 
4307 static ssize_t
4308 sync_max_show(struct mddev *mddev, char *page)
4309 {
4310 	return sprintf(page, "%d (%s)\n", speed_max(mddev),
4311 		       mddev->sync_speed_max ? "local": "system");
4312 }
4313 
4314 static ssize_t
4315 sync_max_store(struct mddev *mddev, const char *buf, size_t len)
4316 {
4317 	int max;
4318 	char *e;
4319 	if (strncmp(buf, "system", 6)==0) {
4320 		mddev->sync_speed_max = 0;
4321 		return len;
4322 	}
4323 	max = simple_strtoul(buf, &e, 10);
4324 	if (buf == e || (*e && *e != '\n') || max <= 0)
4325 		return -EINVAL;
4326 	mddev->sync_speed_max = max;
4327 	return len;
4328 }
4329 
4330 static struct md_sysfs_entry md_sync_max =
4331 __ATTR(sync_speed_max, S_IRUGO|S_IWUSR, sync_max_show, sync_max_store);
4332 
4333 static ssize_t
4334 degraded_show(struct mddev *mddev, char *page)
4335 {
4336 	return sprintf(page, "%d\n", mddev->degraded);
4337 }
4338 static struct md_sysfs_entry md_degraded = __ATTR_RO(degraded);
4339 
4340 static ssize_t
4341 sync_force_parallel_show(struct mddev *mddev, char *page)
4342 {
4343 	return sprintf(page, "%d\n", mddev->parallel_resync);
4344 }
4345 
4346 static ssize_t
4347 sync_force_parallel_store(struct mddev *mddev, const char *buf, size_t len)
4348 {
4349 	long n;
4350 
4351 	if (kstrtol(buf, 10, &n))
4352 		return -EINVAL;
4353 
4354 	if (n != 0 && n != 1)
4355 		return -EINVAL;
4356 
4357 	mddev->parallel_resync = n;
4358 
4359 	if (mddev->sync_thread)
4360 		wake_up(&resync_wait);
4361 
4362 	return len;
4363 }
4364 
4365 /* force parallel resync, even with shared block devices */
4366 static struct md_sysfs_entry md_sync_force_parallel =
4367 __ATTR(sync_force_parallel, S_IRUGO|S_IWUSR,
4368        sync_force_parallel_show, sync_force_parallel_store);
4369 
4370 static ssize_t
4371 sync_speed_show(struct mddev *mddev, char *page)
4372 {
4373 	unsigned long resync, dt, db;
4374 	if (mddev->curr_resync == 0)
4375 		return sprintf(page, "none\n");
4376 	resync = mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active);
4377 	dt = (jiffies - mddev->resync_mark) / HZ;
4378 	if (!dt) dt++;
4379 	db = resync - mddev->resync_mark_cnt;
4380 	return sprintf(page, "%lu\n", db/dt/2); /* K/sec */
4381 }
4382 
4383 static struct md_sysfs_entry md_sync_speed = __ATTR_RO(sync_speed);
4384 
4385 static ssize_t
4386 sync_completed_show(struct mddev *mddev, char *page)
4387 {
4388 	unsigned long long max_sectors, resync;
4389 
4390 	if (!test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4391 		return sprintf(page, "none\n");
4392 
4393 	if (mddev->curr_resync == 1 ||
4394 	    mddev->curr_resync == 2)
4395 		return sprintf(page, "delayed\n");
4396 
4397 	if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ||
4398 	    test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
4399 		max_sectors = mddev->resync_max_sectors;
4400 	else
4401 		max_sectors = mddev->dev_sectors;
4402 
4403 	resync = mddev->curr_resync_completed;
4404 	return sprintf(page, "%llu / %llu\n", resync, max_sectors);
4405 }
4406 
4407 static struct md_sysfs_entry md_sync_completed = __ATTR_RO(sync_completed);
4408 
4409 static ssize_t
4410 min_sync_show(struct mddev *mddev, char *page)
4411 {
4412 	return sprintf(page, "%llu\n",
4413 		       (unsigned long long)mddev->resync_min);
4414 }
4415 static ssize_t
4416 min_sync_store(struct mddev *mddev, const char *buf, size_t len)
4417 {
4418 	unsigned long long min;
4419 	if (kstrtoull(buf, 10, &min))
4420 		return -EINVAL;
4421 	if (min > mddev->resync_max)
4422 		return -EINVAL;
4423 	if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4424 		return -EBUSY;
4425 
4426 	/* Must be a multiple of chunk_size */
4427 	if (mddev->chunk_sectors) {
4428 		sector_t temp = min;
4429 		if (sector_div(temp, mddev->chunk_sectors))
4430 			return -EINVAL;
4431 	}
4432 	mddev->resync_min = min;
4433 
4434 	return len;
4435 }
4436 
4437 static struct md_sysfs_entry md_min_sync =
4438 __ATTR(sync_min, S_IRUGO|S_IWUSR, min_sync_show, min_sync_store);
4439 
4440 static ssize_t
4441 max_sync_show(struct mddev *mddev, char *page)
4442 {
4443 	if (mddev->resync_max == MaxSector)
4444 		return sprintf(page, "max\n");
4445 	else
4446 		return sprintf(page, "%llu\n",
4447 			       (unsigned long long)mddev->resync_max);
4448 }
4449 static ssize_t
4450 max_sync_store(struct mddev *mddev, const char *buf, size_t len)
4451 {
4452 	if (strncmp(buf, "max", 3) == 0)
4453 		mddev->resync_max = MaxSector;
4454 	else {
4455 		unsigned long long max;
4456 		if (kstrtoull(buf, 10, &max))
4457 			return -EINVAL;
4458 		if (max < mddev->resync_min)
4459 			return -EINVAL;
4460 		if (max < mddev->resync_max &&
4461 		    mddev->ro == 0 &&
4462 		    test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4463 			return -EBUSY;
4464 
4465 		/* Must be a multiple of chunk_size */
4466 		if (mddev->chunk_sectors) {
4467 			sector_t temp = max;
4468 			if (sector_div(temp, mddev->chunk_sectors))
4469 				return -EINVAL;
4470 		}
4471 		mddev->resync_max = max;
4472 	}
4473 	wake_up(&mddev->recovery_wait);
4474 	return len;
4475 }
4476 
4477 static struct md_sysfs_entry md_max_sync =
4478 __ATTR(sync_max, S_IRUGO|S_IWUSR, max_sync_show, max_sync_store);
4479 
4480 static ssize_t
4481 suspend_lo_show(struct mddev *mddev, char *page)
4482 {
4483 	return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_lo);
4484 }
4485 
4486 static ssize_t
4487 suspend_lo_store(struct mddev *mddev, const char *buf, size_t len)
4488 {
4489 	char *e;
4490 	unsigned long long new = simple_strtoull(buf, &e, 10);
4491 	unsigned long long old = mddev->suspend_lo;
4492 
4493 	if (mddev->pers == NULL ||
4494 	    mddev->pers->quiesce == NULL)
4495 		return -EINVAL;
4496 	if (buf == e || (*e && *e != '\n'))
4497 		return -EINVAL;
4498 
4499 	mddev->suspend_lo = new;
4500 	if (new >= old)
4501 		/* Shrinking suspended region */
4502 		mddev->pers->quiesce(mddev, 2);
4503 	else {
4504 		/* Expanding suspended region - need to wait */
4505 		mddev->pers->quiesce(mddev, 1);
4506 		mddev->pers->quiesce(mddev, 0);
4507 	}
4508 	return len;
4509 }
4510 static struct md_sysfs_entry md_suspend_lo =
4511 __ATTR(suspend_lo, S_IRUGO|S_IWUSR, suspend_lo_show, suspend_lo_store);
4512 
4513 
4514 static ssize_t
4515 suspend_hi_show(struct mddev *mddev, char *page)
4516 {
4517 	return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_hi);
4518 }
4519 
4520 static ssize_t
4521 suspend_hi_store(struct mddev *mddev, const char *buf, size_t len)
4522 {
4523 	char *e;
4524 	unsigned long long new = simple_strtoull(buf, &e, 10);
4525 	unsigned long long old = mddev->suspend_hi;
4526 
4527 	if (mddev->pers == NULL ||
4528 	    mddev->pers->quiesce == NULL)
4529 		return -EINVAL;
4530 	if (buf == e || (*e && *e != '\n'))
4531 		return -EINVAL;
4532 
4533 	mddev->suspend_hi = new;
4534 	if (new <= old)
4535 		/* Shrinking suspended region */
4536 		mddev->pers->quiesce(mddev, 2);
4537 	else {
4538 		/* Expanding suspended region - need to wait */
4539 		mddev->pers->quiesce(mddev, 1);
4540 		mddev->pers->quiesce(mddev, 0);
4541 	}
4542 	return len;
4543 }
4544 static struct md_sysfs_entry md_suspend_hi =
4545 __ATTR(suspend_hi, S_IRUGO|S_IWUSR, suspend_hi_show, suspend_hi_store);
4546 
4547 static ssize_t
4548 reshape_position_show(struct mddev *mddev, char *page)
4549 {
4550 	if (mddev->reshape_position != MaxSector)
4551 		return sprintf(page, "%llu\n",
4552 			       (unsigned long long)mddev->reshape_position);
4553 	strcpy(page, "none\n");
4554 	return 5;
4555 }
4556 
4557 static ssize_t
4558 reshape_position_store(struct mddev *mddev, const char *buf, size_t len)
4559 {
4560 	struct md_rdev *rdev;
4561 	char *e;
4562 	unsigned long long new = simple_strtoull(buf, &e, 10);
4563 	if (mddev->pers)
4564 		return -EBUSY;
4565 	if (buf == e || (*e && *e != '\n'))
4566 		return -EINVAL;
4567 	mddev->reshape_position = new;
4568 	mddev->delta_disks = 0;
4569 	mddev->reshape_backwards = 0;
4570 	mddev->new_level = mddev->level;
4571 	mddev->new_layout = mddev->layout;
4572 	mddev->new_chunk_sectors = mddev->chunk_sectors;
4573 	rdev_for_each(rdev, mddev)
4574 		rdev->new_data_offset = rdev->data_offset;
4575 	return len;
4576 }
4577 
4578 static struct md_sysfs_entry md_reshape_position =
4579 __ATTR(reshape_position, S_IRUGO|S_IWUSR, reshape_position_show,
4580        reshape_position_store);
4581 
4582 static ssize_t
4583 reshape_direction_show(struct mddev *mddev, char *page)
4584 {
4585 	return sprintf(page, "%s\n",
4586 		       mddev->reshape_backwards ? "backwards" : "forwards");
4587 }
4588 
4589 static ssize_t
4590 reshape_direction_store(struct mddev *mddev, const char *buf, size_t len)
4591 {
4592 	int backwards = 0;
4593 	if (cmd_match(buf, "forwards"))
4594 		backwards = 0;
4595 	else if (cmd_match(buf, "backwards"))
4596 		backwards = 1;
4597 	else
4598 		return -EINVAL;
4599 	if (mddev->reshape_backwards == backwards)
4600 		return len;
4601 
4602 	/* check if we are allowed to change */
4603 	if (mddev->delta_disks)
4604 		return -EBUSY;
4605 
4606 	if (mddev->persistent &&
4607 	    mddev->major_version == 0)
4608 		return -EINVAL;
4609 
4610 	mddev->reshape_backwards = backwards;
4611 	return len;
4612 }
4613 
4614 static struct md_sysfs_entry md_reshape_direction =
4615 __ATTR(reshape_direction, S_IRUGO|S_IWUSR, reshape_direction_show,
4616        reshape_direction_store);
4617 
4618 static ssize_t
4619 array_size_show(struct mddev *mddev, char *page)
4620 {
4621 	if (mddev->external_size)
4622 		return sprintf(page, "%llu\n",
4623 			       (unsigned long long)mddev->array_sectors/2);
4624 	else
4625 		return sprintf(page, "default\n");
4626 }
4627 
4628 static ssize_t
4629 array_size_store(struct mddev *mddev, const char *buf, size_t len)
4630 {
4631 	sector_t sectors;
4632 
4633 	if (strncmp(buf, "default", 7) == 0) {
4634 		if (mddev->pers)
4635 			sectors = mddev->pers->size(mddev, 0, 0);
4636 		else
4637 			sectors = mddev->array_sectors;
4638 
4639 		mddev->external_size = 0;
4640 	} else {
4641 		if (strict_blocks_to_sectors(buf, &sectors) < 0)
4642 			return -EINVAL;
4643 		if (mddev->pers && mddev->pers->size(mddev, 0, 0) < sectors)
4644 			return -E2BIG;
4645 
4646 		mddev->external_size = 1;
4647 	}
4648 
4649 	mddev->array_sectors = sectors;
4650 	if (mddev->pers) {
4651 		set_capacity(mddev->gendisk, mddev->array_sectors);
4652 		revalidate_disk(mddev->gendisk);
4653 	}
4654 	return len;
4655 }
4656 
4657 static struct md_sysfs_entry md_array_size =
4658 __ATTR(array_size, S_IRUGO|S_IWUSR, array_size_show,
4659        array_size_store);
4660 
4661 static struct attribute *md_default_attrs[] = {
4662 	&md_level.attr,
4663 	&md_layout.attr,
4664 	&md_raid_disks.attr,
4665 	&md_chunk_size.attr,
4666 	&md_size.attr,
4667 	&md_resync_start.attr,
4668 	&md_metadata.attr,
4669 	&md_new_device.attr,
4670 	&md_safe_delay.attr,
4671 	&md_array_state.attr,
4672 	&md_reshape_position.attr,
4673 	&md_reshape_direction.attr,
4674 	&md_array_size.attr,
4675 	&max_corr_read_errors.attr,
4676 	NULL,
4677 };
4678 
4679 static struct attribute *md_redundancy_attrs[] = {
4680 	&md_scan_mode.attr,
4681 	&md_last_scan_mode.attr,
4682 	&md_mismatches.attr,
4683 	&md_sync_min.attr,
4684 	&md_sync_max.attr,
4685 	&md_sync_speed.attr,
4686 	&md_sync_force_parallel.attr,
4687 	&md_sync_completed.attr,
4688 	&md_min_sync.attr,
4689 	&md_max_sync.attr,
4690 	&md_suspend_lo.attr,
4691 	&md_suspend_hi.attr,
4692 	&md_bitmap.attr,
4693 	&md_degraded.attr,
4694 	NULL,
4695 };
4696 static struct attribute_group md_redundancy_group = {
4697 	.name = NULL,
4698 	.attrs = md_redundancy_attrs,
4699 };
4700 
4701 
4702 static ssize_t
4703 md_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
4704 {
4705 	struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
4706 	struct mddev *mddev = container_of(kobj, struct mddev, kobj);
4707 	ssize_t rv;
4708 
4709 	if (!entry->show)
4710 		return -EIO;
4711 	spin_lock(&all_mddevs_lock);
4712 	if (list_empty(&mddev->all_mddevs)) {
4713 		spin_unlock(&all_mddevs_lock);
4714 		return -EBUSY;
4715 	}
4716 	mddev_get(mddev);
4717 	spin_unlock(&all_mddevs_lock);
4718 
4719 	rv = mddev_lock(mddev);
4720 	if (!rv) {
4721 		rv = entry->show(mddev, page);
4722 		mddev_unlock(mddev);
4723 	}
4724 	mddev_put(mddev);
4725 	return rv;
4726 }
4727 
4728 static ssize_t
4729 md_attr_store(struct kobject *kobj, struct attribute *attr,
4730 	      const char *page, size_t length)
4731 {
4732 	struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
4733 	struct mddev *mddev = container_of(kobj, struct mddev, kobj);
4734 	ssize_t rv;
4735 
4736 	if (!entry->store)
4737 		return -EIO;
4738 	if (!capable(CAP_SYS_ADMIN))
4739 		return -EACCES;
4740 	spin_lock(&all_mddevs_lock);
4741 	if (list_empty(&mddev->all_mddevs)) {
4742 		spin_unlock(&all_mddevs_lock);
4743 		return -EBUSY;
4744 	}
4745 	mddev_get(mddev);
4746 	spin_unlock(&all_mddevs_lock);
4747 	if (entry->store == new_dev_store)
4748 		flush_workqueue(md_misc_wq);
4749 	rv = mddev_lock(mddev);
4750 	if (!rv) {
4751 		rv = entry->store(mddev, page, length);
4752 		mddev_unlock(mddev);
4753 	}
4754 	mddev_put(mddev);
4755 	return rv;
4756 }
4757 
4758 static void md_free(struct kobject *ko)
4759 {
4760 	struct mddev *mddev = container_of(ko, struct mddev, kobj);
4761 
4762 	if (mddev->sysfs_state)
4763 		sysfs_put(mddev->sysfs_state);
4764 
4765 	if (mddev->gendisk) {
4766 		del_gendisk(mddev->gendisk);
4767 		put_disk(mddev->gendisk);
4768 	}
4769 	if (mddev->queue)
4770 		blk_cleanup_queue(mddev->queue);
4771 
4772 	kfree(mddev);
4773 }
4774 
4775 static const struct sysfs_ops md_sysfs_ops = {
4776 	.show	= md_attr_show,
4777 	.store	= md_attr_store,
4778 };
4779 static struct kobj_type md_ktype = {
4780 	.release	= md_free,
4781 	.sysfs_ops	= &md_sysfs_ops,
4782 	.default_attrs	= md_default_attrs,
4783 };
4784 
4785 int mdp_major = 0;
4786 
4787 static void mddev_delayed_delete(struct work_struct *ws)
4788 {
4789 	struct mddev *mddev = container_of(ws, struct mddev, del_work);
4790 
4791 	sysfs_remove_group(&mddev->kobj, &md_bitmap_group);
4792 	kobject_del(&mddev->kobj);
4793 	kobject_put(&mddev->kobj);
4794 }
4795 
4796 static int md_alloc(dev_t dev, char *name)
4797 {
4798 	static DEFINE_MUTEX(disks_mutex);
4799 	struct mddev *mddev = mddev_find(dev);
4800 	struct gendisk *disk;
4801 	int partitioned;
4802 	int shift;
4803 	int unit;
4804 	int error;
4805 
4806 	if (!mddev)
4807 		return -ENODEV;
4808 
4809 	partitioned = (MAJOR(mddev->unit) != MD_MAJOR);
4810 	shift = partitioned ? MdpMinorShift : 0;
4811 	unit = MINOR(mddev->unit) >> shift;
4812 
4813 	/* wait for any previous instance of this device to be
4814 	 * completely removed (mddev_delayed_delete).
4815 	 */
4816 	flush_workqueue(md_misc_wq);
4817 
4818 	mutex_lock(&disks_mutex);
4819 	error = -EEXIST;
4820 	if (mddev->gendisk)
4821 		goto abort;
4822 
4823 	if (name) {
4824 		/* Need to ensure that 'name' is not a duplicate.
4825 		 */
4826 		struct mddev *mddev2;
4827 		spin_lock(&all_mddevs_lock);
4828 
4829 		list_for_each_entry(mddev2, &all_mddevs, all_mddevs)
4830 			if (mddev2->gendisk &&
4831 			    strcmp(mddev2->gendisk->disk_name, name) == 0) {
4832 				spin_unlock(&all_mddevs_lock);
4833 				goto abort;
4834 			}
4835 		spin_unlock(&all_mddevs_lock);
4836 	}
4837 
4838 	error = -ENOMEM;
4839 	mddev->queue = blk_alloc_queue(GFP_KERNEL);
4840 	if (!mddev->queue)
4841 		goto abort;
4842 	mddev->queue->queuedata = mddev;
4843 
4844 	blk_queue_make_request(mddev->queue, md_make_request);
4845 	blk_set_stacking_limits(&mddev->queue->limits);
4846 
4847 	disk = alloc_disk(1 << shift);
4848 	if (!disk) {
4849 		blk_cleanup_queue(mddev->queue);
4850 		mddev->queue = NULL;
4851 		goto abort;
4852 	}
4853 	disk->major = MAJOR(mddev->unit);
4854 	disk->first_minor = unit << shift;
4855 	if (name)
4856 		strcpy(disk->disk_name, name);
4857 	else if (partitioned)
4858 		sprintf(disk->disk_name, "md_d%d", unit);
4859 	else
4860 		sprintf(disk->disk_name, "md%d", unit);
4861 	disk->fops = &md_fops;
4862 	disk->private_data = mddev;
4863 	disk->queue = mddev->queue;
4864 	blk_queue_flush(mddev->queue, REQ_FLUSH | REQ_FUA);
4865 	/* Allow extended partitions.  This makes the
4866 	 * 'mdp' device redundant, but we can't really
4867 	 * remove it now.
4868 	 */
4869 	disk->flags |= GENHD_FL_EXT_DEVT;
4870 	mddev->gendisk = disk;
4871 	/* As soon as we call add_disk(), another thread could get
4872 	 * through to md_open, so make sure it doesn't get too far
4873 	 */
4874 	mutex_lock(&mddev->open_mutex);
4875 	add_disk(disk);
4876 
4877 	error = kobject_init_and_add(&mddev->kobj, &md_ktype,
4878 				     &disk_to_dev(disk)->kobj, "%s", "md");
4879 	if (error) {
4880 		/* This isn't possible, but as kobject_init_and_add is marked
4881 		 * __must_check, we must do something with the result
4882 		 */
4883 		printk(KERN_WARNING "md: cannot register %s/md - name in use\n",
4884 		       disk->disk_name);
4885 		error = 0;
4886 	}
4887 	if (mddev->kobj.sd &&
4888 	    sysfs_create_group(&mddev->kobj, &md_bitmap_group))
4889 		printk(KERN_DEBUG "pointless warning\n");
4890 	mutex_unlock(&mddev->open_mutex);
4891  abort:
4892 	mutex_unlock(&disks_mutex);
4893 	if (!error && mddev->kobj.sd) {
4894 		kobject_uevent(&mddev->kobj, KOBJ_ADD);
4895 		mddev->sysfs_state = sysfs_get_dirent_safe(mddev->kobj.sd, "array_state");
4896 	}
4897 	mddev_put(mddev);
4898 	return error;
4899 }
4900 
4901 static struct kobject *md_probe(dev_t dev, int *part, void *data)
4902 {
4903 	md_alloc(dev, NULL);
4904 	return NULL;
4905 }
4906 
4907 static int add_named_array(const char *val, struct kernel_param *kp)
4908 {
4909 	/* val must be "md_*" where * is not all digits.
4910 	 * We allocate an array with a large free minor number, and
4911 	 * set the name to val.  val must not already be an active name.
4912 	 */
4913 	int len = strlen(val);
4914 	char buf[DISK_NAME_LEN];
4915 
4916 	while (len && val[len-1] == '\n')
4917 		len--;
4918 	if (len >= DISK_NAME_LEN)
4919 		return -E2BIG;
4920 	strlcpy(buf, val, len+1);
4921 	if (strncmp(buf, "md_", 3) != 0)
4922 		return -EINVAL;
4923 	return md_alloc(0, buf);
4924 }
4925 
4926 static void md_safemode_timeout(unsigned long data)
4927 {
4928 	struct mddev *mddev = (struct mddev *) data;
4929 
4930 	if (!atomic_read(&mddev->writes_pending)) {
4931 		mddev->safemode = 1;
4932 		if (mddev->external)
4933 			sysfs_notify_dirent_safe(mddev->sysfs_state);
4934 	}
4935 	md_wakeup_thread(mddev->thread);
4936 }
4937 
4938 static int start_dirty_degraded;
4939 
4940 int md_run(struct mddev *mddev)
4941 {
4942 	int err;
4943 	struct md_rdev *rdev;
4944 	struct md_personality *pers;
4945 
4946 	if (list_empty(&mddev->disks))
4947 		/* cannot run an array with no devices.. */
4948 		return -EINVAL;
4949 
4950 	if (mddev->pers)
4951 		return -EBUSY;
4952 	/* Cannot run until previous stop completes properly */
4953 	if (mddev->sysfs_active)
4954 		return -EBUSY;
4955 
4956 	/*
4957 	 * Analyze all RAID superblock(s)
4958 	 */
4959 	if (!mddev->raid_disks) {
4960 		if (!mddev->persistent)
4961 			return -EINVAL;
4962 		analyze_sbs(mddev);
4963 	}
4964 
4965 	if (mddev->level != LEVEL_NONE)
4966 		request_module("md-level-%d", mddev->level);
4967 	else if (mddev->clevel[0])
4968 		request_module("md-%s", mddev->clevel);
4969 
4970 	/*
4971 	 * Drop all container device buffers, from now on
4972 	 * the only valid external interface is through the md
4973 	 * device.
4974 	 */
4975 	rdev_for_each(rdev, mddev) {
4976 		if (test_bit(Faulty, &rdev->flags))
4977 			continue;
4978 		sync_blockdev(rdev->bdev);
4979 		invalidate_bdev(rdev->bdev);
4980 
4981 		/* perform some consistency tests on the device.
4982 		 * We don't want the data to overlap the metadata,
4983 		 * Internal Bitmap issues have been handled elsewhere.
4984 		 */
4985 		if (rdev->meta_bdev) {
4986 			/* Nothing to check */;
4987 		} else if (rdev->data_offset < rdev->sb_start) {
4988 			if (mddev->dev_sectors &&
4989 			    rdev->data_offset + mddev->dev_sectors
4990 			    > rdev->sb_start) {
4991 				printk("md: %s: data overlaps metadata\n",
4992 				       mdname(mddev));
4993 				return -EINVAL;
4994 			}
4995 		} else {
4996 			if (rdev->sb_start + rdev->sb_size/512
4997 			    > rdev->data_offset) {
4998 				printk("md: %s: metadata overlaps data\n",
4999 				       mdname(mddev));
5000 				return -EINVAL;
5001 			}
5002 		}
5003 		sysfs_notify_dirent_safe(rdev->sysfs_state);
5004 	}
5005 
5006 	if (mddev->bio_set == NULL)
5007 		mddev->bio_set = bioset_create(BIO_POOL_SIZE, 0);
5008 
5009 	spin_lock(&pers_lock);
5010 	pers = find_pers(mddev->level, mddev->clevel);
5011 	if (!pers || !try_module_get(pers->owner)) {
5012 		spin_unlock(&pers_lock);
5013 		if (mddev->level != LEVEL_NONE)
5014 			printk(KERN_WARNING "md: personality for level %d is not loaded!\n",
5015 			       mddev->level);
5016 		else
5017 			printk(KERN_WARNING "md: personality for level %s is not loaded!\n",
5018 			       mddev->clevel);
5019 		return -EINVAL;
5020 	}
5021 	mddev->pers = pers;
5022 	spin_unlock(&pers_lock);
5023 	if (mddev->level != pers->level) {
5024 		mddev->level = pers->level;
5025 		mddev->new_level = pers->level;
5026 	}
5027 	strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
5028 
5029 	if (mddev->reshape_position != MaxSector &&
5030 	    pers->start_reshape == NULL) {
5031 		/* This personality cannot handle reshaping... */
5032 		mddev->pers = NULL;
5033 		module_put(pers->owner);
5034 		return -EINVAL;
5035 	}
5036 
5037 	if (pers->sync_request) {
5038 		/* Warn if this is a potentially silly
5039 		 * configuration.
5040 		 */
5041 		char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
5042 		struct md_rdev *rdev2;
5043 		int warned = 0;
5044 
5045 		rdev_for_each(rdev, mddev)
5046 			rdev_for_each(rdev2, mddev) {
5047 				if (rdev < rdev2 &&
5048 				    rdev->bdev->bd_contains ==
5049 				    rdev2->bdev->bd_contains) {
5050 					printk(KERN_WARNING
5051 					       "%s: WARNING: %s appears to be"
5052 					       " on the same physical disk as"
5053 					       " %s.\n",
5054 					       mdname(mddev),
5055 					       bdevname(rdev->bdev,b),
5056 					       bdevname(rdev2->bdev,b2));
5057 					warned = 1;
5058 				}
5059 			}
5060 
5061 		if (warned)
5062 			printk(KERN_WARNING
5063 			       "True protection against single-disk"
5064 			       " failure might be compromised.\n");
5065 	}
5066 
5067 	mddev->recovery = 0;
5068 	/* may be over-ridden by personality */
5069 	mddev->resync_max_sectors = mddev->dev_sectors;
5070 
5071 	mddev->ok_start_degraded = start_dirty_degraded;
5072 
5073 	if (start_readonly && mddev->ro == 0)
5074 		mddev->ro = 2; /* read-only, but switch on first write */
5075 
5076 	err = mddev->pers->run(mddev);
5077 	if (err)
5078 		printk(KERN_ERR "md: pers->run() failed ...\n");
5079 	else if (mddev->pers->size(mddev, 0, 0) < mddev->array_sectors) {
5080 		WARN_ONCE(!mddev->external_size, "%s: default size too small,"
5081 			  " but 'external_size' not in effect?\n", __func__);
5082 		printk(KERN_ERR
5083 		       "md: invalid array_size %llu > default size %llu\n",
5084 		       (unsigned long long)mddev->array_sectors / 2,
5085 		       (unsigned long long)mddev->pers->size(mddev, 0, 0) / 2);
5086 		err = -EINVAL;
5087 		mddev->pers->stop(mddev);
5088 	}
5089 	if (err == 0 && mddev->pers->sync_request &&
5090 	    (mddev->bitmap_info.file || mddev->bitmap_info.offset)) {
5091 		err = bitmap_create(mddev);
5092 		if (err) {
5093 			printk(KERN_ERR "%s: failed to create bitmap (%d)\n",
5094 			       mdname(mddev), err);
5095 			mddev->pers->stop(mddev);
5096 		}
5097 	}
5098 	if (err) {
5099 		module_put(mddev->pers->owner);
5100 		mddev->pers = NULL;
5101 		bitmap_destroy(mddev);
5102 		return err;
5103 	}
5104 	if (mddev->pers->sync_request) {
5105 		if (mddev->kobj.sd &&
5106 		    sysfs_create_group(&mddev->kobj, &md_redundancy_group))
5107 			printk(KERN_WARNING
5108 			       "md: cannot register extra attributes for %s\n",
5109 			       mdname(mddev));
5110 		mddev->sysfs_action = sysfs_get_dirent_safe(mddev->kobj.sd, "sync_action");
5111 	} else if (mddev->ro == 2) /* auto-readonly not meaningful */
5112 		mddev->ro = 0;
5113 
5114  	atomic_set(&mddev->writes_pending,0);
5115 	atomic_set(&mddev->max_corr_read_errors,
5116 		   MD_DEFAULT_MAX_CORRECTED_READ_ERRORS);
5117 	mddev->safemode = 0;
5118 	mddev->safemode_timer.function = md_safemode_timeout;
5119 	mddev->safemode_timer.data = (unsigned long) mddev;
5120 	mddev->safemode_delay = (200 * HZ)/1000 +1; /* 200 msec delay */
5121 	mddev->in_sync = 1;
5122 	smp_wmb();
5123 	mddev->ready = 1;
5124 	rdev_for_each(rdev, mddev)
5125 		if (rdev->raid_disk >= 0)
5126 			if (sysfs_link_rdev(mddev, rdev))
5127 				/* failure here is OK */;
5128 
5129 	set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5130 
5131 	if (mddev->flags & MD_UPDATE_SB_FLAGS)
5132 		md_update_sb(mddev, 0);
5133 
5134 	md_new_event(mddev);
5135 	sysfs_notify_dirent_safe(mddev->sysfs_state);
5136 	sysfs_notify_dirent_safe(mddev->sysfs_action);
5137 	sysfs_notify(&mddev->kobj, NULL, "degraded");
5138 	return 0;
5139 }
5140 EXPORT_SYMBOL_GPL(md_run);
5141 
5142 static int do_md_run(struct mddev *mddev)
5143 {
5144 	int err;
5145 
5146 	err = md_run(mddev);
5147 	if (err)
5148 		goto out;
5149 	err = bitmap_load(mddev);
5150 	if (err) {
5151 		bitmap_destroy(mddev);
5152 		goto out;
5153 	}
5154 
5155 	md_wakeup_thread(mddev->thread);
5156 	md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
5157 
5158 	set_capacity(mddev->gendisk, mddev->array_sectors);
5159 	revalidate_disk(mddev->gendisk);
5160 	mddev->changed = 1;
5161 	kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
5162 out:
5163 	return err;
5164 }
5165 
5166 static int restart_array(struct mddev *mddev)
5167 {
5168 	struct gendisk *disk = mddev->gendisk;
5169 
5170 	/* Complain if it has no devices */
5171 	if (list_empty(&mddev->disks))
5172 		return -ENXIO;
5173 	if (!mddev->pers)
5174 		return -EINVAL;
5175 	if (!mddev->ro)
5176 		return -EBUSY;
5177 	mddev->safemode = 0;
5178 	mddev->ro = 0;
5179 	set_disk_ro(disk, 0);
5180 	printk(KERN_INFO "md: %s switched to read-write mode.\n",
5181 		mdname(mddev));
5182 	/* Kick recovery or resync if necessary */
5183 	set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5184 	md_wakeup_thread(mddev->thread);
5185 	md_wakeup_thread(mddev->sync_thread);
5186 	sysfs_notify_dirent_safe(mddev->sysfs_state);
5187 	return 0;
5188 }
5189 
5190 static void md_clean(struct mddev *mddev)
5191 {
5192 	mddev->array_sectors = 0;
5193 	mddev->external_size = 0;
5194 	mddev->dev_sectors = 0;
5195 	mddev->raid_disks = 0;
5196 	mddev->recovery_cp = 0;
5197 	mddev->resync_min = 0;
5198 	mddev->resync_max = MaxSector;
5199 	mddev->reshape_position = MaxSector;
5200 	mddev->external = 0;
5201 	mddev->persistent = 0;
5202 	mddev->level = LEVEL_NONE;
5203 	mddev->clevel[0] = 0;
5204 	mddev->flags = 0;
5205 	mddev->ro = 0;
5206 	mddev->metadata_type[0] = 0;
5207 	mddev->chunk_sectors = 0;
5208 	mddev->ctime = mddev->utime = 0;
5209 	mddev->layout = 0;
5210 	mddev->max_disks = 0;
5211 	mddev->events = 0;
5212 	mddev->can_decrease_events = 0;
5213 	mddev->delta_disks = 0;
5214 	mddev->reshape_backwards = 0;
5215 	mddev->new_level = LEVEL_NONE;
5216 	mddev->new_layout = 0;
5217 	mddev->new_chunk_sectors = 0;
5218 	mddev->curr_resync = 0;
5219 	atomic64_set(&mddev->resync_mismatches, 0);
5220 	mddev->suspend_lo = mddev->suspend_hi = 0;
5221 	mddev->sync_speed_min = mddev->sync_speed_max = 0;
5222 	mddev->recovery = 0;
5223 	mddev->in_sync = 0;
5224 	mddev->changed = 0;
5225 	mddev->degraded = 0;
5226 	mddev->safemode = 0;
5227 	mddev->merge_check_needed = 0;
5228 	mddev->bitmap_info.offset = 0;
5229 	mddev->bitmap_info.default_offset = 0;
5230 	mddev->bitmap_info.default_space = 0;
5231 	mddev->bitmap_info.chunksize = 0;
5232 	mddev->bitmap_info.daemon_sleep = 0;
5233 	mddev->bitmap_info.max_write_behind = 0;
5234 }
5235 
5236 static void __md_stop_writes(struct mddev *mddev)
5237 {
5238 	set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5239 	if (mddev->sync_thread) {
5240 		set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5241 		md_reap_sync_thread(mddev);
5242 	}
5243 
5244 	del_timer_sync(&mddev->safemode_timer);
5245 
5246 	bitmap_flush(mddev);
5247 	md_super_wait(mddev);
5248 
5249 	if (mddev->ro == 0 &&
5250 	    (!mddev->in_sync || (mddev->flags & MD_UPDATE_SB_FLAGS))) {
5251 		/* mark array as shutdown cleanly */
5252 		mddev->in_sync = 1;
5253 		md_update_sb(mddev, 1);
5254 	}
5255 }
5256 
5257 void md_stop_writes(struct mddev *mddev)
5258 {
5259 	mddev_lock_nointr(mddev);
5260 	__md_stop_writes(mddev);
5261 	mddev_unlock(mddev);
5262 }
5263 EXPORT_SYMBOL_GPL(md_stop_writes);
5264 
5265 static void __md_stop(struct mddev *mddev)
5266 {
5267 	mddev->ready = 0;
5268 	mddev->pers->stop(mddev);
5269 	if (mddev->pers->sync_request && mddev->to_remove == NULL)
5270 		mddev->to_remove = &md_redundancy_group;
5271 	module_put(mddev->pers->owner);
5272 	mddev->pers = NULL;
5273 	clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5274 }
5275 
5276 void md_stop(struct mddev *mddev)
5277 {
5278 	/* stop the array and free an attached data structures.
5279 	 * This is called from dm-raid
5280 	 */
5281 	__md_stop(mddev);
5282 	bitmap_destroy(mddev);
5283 	if (mddev->bio_set)
5284 		bioset_free(mddev->bio_set);
5285 }
5286 
5287 EXPORT_SYMBOL_GPL(md_stop);
5288 
5289 static int md_set_readonly(struct mddev *mddev, struct block_device *bdev)
5290 {
5291 	int err = 0;
5292 	int did_freeze = 0;
5293 
5294 	if (!test_bit(MD_RECOVERY_FROZEN, &mddev->recovery)) {
5295 		did_freeze = 1;
5296 		set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5297 		md_wakeup_thread(mddev->thread);
5298 	}
5299 	if (mddev->sync_thread) {
5300 		set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5301 		/* Thread might be blocked waiting for metadata update
5302 		 * which will now never happen */
5303 		wake_up_process(mddev->sync_thread->tsk);
5304 	}
5305 	mddev_unlock(mddev);
5306 	wait_event(resync_wait, mddev->sync_thread == NULL);
5307 	mddev_lock_nointr(mddev);
5308 
5309 	mutex_lock(&mddev->open_mutex);
5310 	if (atomic_read(&mddev->openers) > !!bdev ||
5311 	    mddev->sync_thread ||
5312 	    (bdev && !test_bit(MD_STILL_CLOSED, &mddev->flags))) {
5313 		printk("md: %s still in use.\n",mdname(mddev));
5314 		if (did_freeze) {
5315 			clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5316 			md_wakeup_thread(mddev->thread);
5317 		}
5318 		err = -EBUSY;
5319 		goto out;
5320 	}
5321 	if (mddev->pers) {
5322 		__md_stop_writes(mddev);
5323 
5324 		err  = -ENXIO;
5325 		if (mddev->ro==1)
5326 			goto out;
5327 		mddev->ro = 1;
5328 		set_disk_ro(mddev->gendisk, 1);
5329 		clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5330 		sysfs_notify_dirent_safe(mddev->sysfs_state);
5331 		err = 0;
5332 	}
5333 out:
5334 	mutex_unlock(&mddev->open_mutex);
5335 	return err;
5336 }
5337 
5338 /* mode:
5339  *   0 - completely stop and dis-assemble array
5340  *   2 - stop but do not disassemble array
5341  */
5342 static int do_md_stop(struct mddev * mddev, int mode,
5343 		      struct block_device *bdev)
5344 {
5345 	struct gendisk *disk = mddev->gendisk;
5346 	struct md_rdev *rdev;
5347 	int did_freeze = 0;
5348 
5349 	if (!test_bit(MD_RECOVERY_FROZEN, &mddev->recovery)) {
5350 		did_freeze = 1;
5351 		set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5352 		md_wakeup_thread(mddev->thread);
5353 	}
5354 	if (mddev->sync_thread) {
5355 		set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5356 		/* Thread might be blocked waiting for metadata update
5357 		 * which will now never happen */
5358 		wake_up_process(mddev->sync_thread->tsk);
5359 	}
5360 	mddev_unlock(mddev);
5361 	wait_event(resync_wait, mddev->sync_thread == NULL);
5362 	mddev_lock_nointr(mddev);
5363 
5364 	mutex_lock(&mddev->open_mutex);
5365 	if (atomic_read(&mddev->openers) > !!bdev ||
5366 	    mddev->sysfs_active ||
5367 	    mddev->sync_thread ||
5368 	    (bdev && !test_bit(MD_STILL_CLOSED, &mddev->flags))) {
5369 		printk("md: %s still in use.\n",mdname(mddev));
5370 		mutex_unlock(&mddev->open_mutex);
5371 		if (did_freeze) {
5372 			clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5373 			md_wakeup_thread(mddev->thread);
5374 		}
5375 		return -EBUSY;
5376 	}
5377 	if (mddev->pers) {
5378 		if (mddev->ro)
5379 			set_disk_ro(disk, 0);
5380 
5381 		__md_stop_writes(mddev);
5382 		__md_stop(mddev);
5383 		mddev->queue->merge_bvec_fn = NULL;
5384 		mddev->queue->backing_dev_info.congested_fn = NULL;
5385 
5386 		/* tell userspace to handle 'inactive' */
5387 		sysfs_notify_dirent_safe(mddev->sysfs_state);
5388 
5389 		rdev_for_each(rdev, mddev)
5390 			if (rdev->raid_disk >= 0)
5391 				sysfs_unlink_rdev(mddev, rdev);
5392 
5393 		set_capacity(disk, 0);
5394 		mutex_unlock(&mddev->open_mutex);
5395 		mddev->changed = 1;
5396 		revalidate_disk(disk);
5397 
5398 		if (mddev->ro)
5399 			mddev->ro = 0;
5400 	} else
5401 		mutex_unlock(&mddev->open_mutex);
5402 	/*
5403 	 * Free resources if final stop
5404 	 */
5405 	if (mode == 0) {
5406 		printk(KERN_INFO "md: %s stopped.\n", mdname(mddev));
5407 
5408 		bitmap_destroy(mddev);
5409 		if (mddev->bitmap_info.file) {
5410 			fput(mddev->bitmap_info.file);
5411 			mddev->bitmap_info.file = NULL;
5412 		}
5413 		mddev->bitmap_info.offset = 0;
5414 
5415 		export_array(mddev);
5416 
5417 		md_clean(mddev);
5418 		kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
5419 		if (mddev->hold_active == UNTIL_STOP)
5420 			mddev->hold_active = 0;
5421 	}
5422 	blk_integrity_unregister(disk);
5423 	md_new_event(mddev);
5424 	sysfs_notify_dirent_safe(mddev->sysfs_state);
5425 	return 0;
5426 }
5427 
5428 #ifndef MODULE
5429 static void autorun_array(struct mddev *mddev)
5430 {
5431 	struct md_rdev *rdev;
5432 	int err;
5433 
5434 	if (list_empty(&mddev->disks))
5435 		return;
5436 
5437 	printk(KERN_INFO "md: running: ");
5438 
5439 	rdev_for_each(rdev, mddev) {
5440 		char b[BDEVNAME_SIZE];
5441 		printk("<%s>", bdevname(rdev->bdev,b));
5442 	}
5443 	printk("\n");
5444 
5445 	err = do_md_run(mddev);
5446 	if (err) {
5447 		printk(KERN_WARNING "md: do_md_run() returned %d\n", err);
5448 		do_md_stop(mddev, 0, NULL);
5449 	}
5450 }
5451 
5452 /*
5453  * lets try to run arrays based on all disks that have arrived
5454  * until now. (those are in pending_raid_disks)
5455  *
5456  * the method: pick the first pending disk, collect all disks with
5457  * the same UUID, remove all from the pending list and put them into
5458  * the 'same_array' list. Then order this list based on superblock
5459  * update time (freshest comes first), kick out 'old' disks and
5460  * compare superblocks. If everything's fine then run it.
5461  *
5462  * If "unit" is allocated, then bump its reference count
5463  */
5464 static void autorun_devices(int part)
5465 {
5466 	struct md_rdev *rdev0, *rdev, *tmp;
5467 	struct mddev *mddev;
5468 	char b[BDEVNAME_SIZE];
5469 
5470 	printk(KERN_INFO "md: autorun ...\n");
5471 	while (!list_empty(&pending_raid_disks)) {
5472 		int unit;
5473 		dev_t dev;
5474 		LIST_HEAD(candidates);
5475 		rdev0 = list_entry(pending_raid_disks.next,
5476 					 struct md_rdev, same_set);
5477 
5478 		printk(KERN_INFO "md: considering %s ...\n",
5479 			bdevname(rdev0->bdev,b));
5480 		INIT_LIST_HEAD(&candidates);
5481 		rdev_for_each_list(rdev, tmp, &pending_raid_disks)
5482 			if (super_90_load(rdev, rdev0, 0) >= 0) {
5483 				printk(KERN_INFO "md:  adding %s ...\n",
5484 					bdevname(rdev->bdev,b));
5485 				list_move(&rdev->same_set, &candidates);
5486 			}
5487 		/*
5488 		 * now we have a set of devices, with all of them having
5489 		 * mostly sane superblocks. It's time to allocate the
5490 		 * mddev.
5491 		 */
5492 		if (part) {
5493 			dev = MKDEV(mdp_major,
5494 				    rdev0->preferred_minor << MdpMinorShift);
5495 			unit = MINOR(dev) >> MdpMinorShift;
5496 		} else {
5497 			dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
5498 			unit = MINOR(dev);
5499 		}
5500 		if (rdev0->preferred_minor != unit) {
5501 			printk(KERN_INFO "md: unit number in %s is bad: %d\n",
5502 			       bdevname(rdev0->bdev, b), rdev0->preferred_minor);
5503 			break;
5504 		}
5505 
5506 		md_probe(dev, NULL, NULL);
5507 		mddev = mddev_find(dev);
5508 		if (!mddev || !mddev->gendisk) {
5509 			if (mddev)
5510 				mddev_put(mddev);
5511 			printk(KERN_ERR
5512 				"md: cannot allocate memory for md drive.\n");
5513 			break;
5514 		}
5515 		if (mddev_lock(mddev))
5516 			printk(KERN_WARNING "md: %s locked, cannot run\n",
5517 			       mdname(mddev));
5518 		else if (mddev->raid_disks || mddev->major_version
5519 			 || !list_empty(&mddev->disks)) {
5520 			printk(KERN_WARNING
5521 				"md: %s already running, cannot run %s\n",
5522 				mdname(mddev), bdevname(rdev0->bdev,b));
5523 			mddev_unlock(mddev);
5524 		} else {
5525 			printk(KERN_INFO "md: created %s\n", mdname(mddev));
5526 			mddev->persistent = 1;
5527 			rdev_for_each_list(rdev, tmp, &candidates) {
5528 				list_del_init(&rdev->same_set);
5529 				if (bind_rdev_to_array(rdev, mddev))
5530 					export_rdev(rdev);
5531 			}
5532 			autorun_array(mddev);
5533 			mddev_unlock(mddev);
5534 		}
5535 		/* on success, candidates will be empty, on error
5536 		 * it won't...
5537 		 */
5538 		rdev_for_each_list(rdev, tmp, &candidates) {
5539 			list_del_init(&rdev->same_set);
5540 			export_rdev(rdev);
5541 		}
5542 		mddev_put(mddev);
5543 	}
5544 	printk(KERN_INFO "md: ... autorun DONE.\n");
5545 }
5546 #endif /* !MODULE */
5547 
5548 static int get_version(void __user * arg)
5549 {
5550 	mdu_version_t ver;
5551 
5552 	ver.major = MD_MAJOR_VERSION;
5553 	ver.minor = MD_MINOR_VERSION;
5554 	ver.patchlevel = MD_PATCHLEVEL_VERSION;
5555 
5556 	if (copy_to_user(arg, &ver, sizeof(ver)))
5557 		return -EFAULT;
5558 
5559 	return 0;
5560 }
5561 
5562 static int get_array_info(struct mddev * mddev, void __user * arg)
5563 {
5564 	mdu_array_info_t info;
5565 	int nr,working,insync,failed,spare;
5566 	struct md_rdev *rdev;
5567 
5568 	nr = working = insync = failed = spare = 0;
5569 	rcu_read_lock();
5570 	rdev_for_each_rcu(rdev, mddev) {
5571 		nr++;
5572 		if (test_bit(Faulty, &rdev->flags))
5573 			failed++;
5574 		else {
5575 			working++;
5576 			if (test_bit(In_sync, &rdev->flags))
5577 				insync++;
5578 			else
5579 				spare++;
5580 		}
5581 	}
5582 	rcu_read_unlock();
5583 
5584 	info.major_version = mddev->major_version;
5585 	info.minor_version = mddev->minor_version;
5586 	info.patch_version = MD_PATCHLEVEL_VERSION;
5587 	info.ctime         = mddev->ctime;
5588 	info.level         = mddev->level;
5589 	info.size          = mddev->dev_sectors / 2;
5590 	if (info.size != mddev->dev_sectors / 2) /* overflow */
5591 		info.size = -1;
5592 	info.nr_disks      = nr;
5593 	info.raid_disks    = mddev->raid_disks;
5594 	info.md_minor      = mddev->md_minor;
5595 	info.not_persistent= !mddev->persistent;
5596 
5597 	info.utime         = mddev->utime;
5598 	info.state         = 0;
5599 	if (mddev->in_sync)
5600 		info.state = (1<<MD_SB_CLEAN);
5601 	if (mddev->bitmap && mddev->bitmap_info.offset)
5602 		info.state = (1<<MD_SB_BITMAP_PRESENT);
5603 	info.active_disks  = insync;
5604 	info.working_disks = working;
5605 	info.failed_disks  = failed;
5606 	info.spare_disks   = spare;
5607 
5608 	info.layout        = mddev->layout;
5609 	info.chunk_size    = mddev->chunk_sectors << 9;
5610 
5611 	if (copy_to_user(arg, &info, sizeof(info)))
5612 		return -EFAULT;
5613 
5614 	return 0;
5615 }
5616 
5617 static int get_bitmap_file(struct mddev * mddev, void __user * arg)
5618 {
5619 	mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */
5620 	char *ptr, *buf = NULL;
5621 	int err = -ENOMEM;
5622 
5623 	file = kmalloc(sizeof(*file), GFP_NOIO);
5624 
5625 	if (!file)
5626 		goto out;
5627 
5628 	/* bitmap disabled, zero the first byte and copy out */
5629 	if (!mddev->bitmap || !mddev->bitmap->storage.file) {
5630 		file->pathname[0] = '\0';
5631 		goto copy_out;
5632 	}
5633 
5634 	buf = kmalloc(sizeof(file->pathname), GFP_KERNEL);
5635 	if (!buf)
5636 		goto out;
5637 
5638 	ptr = d_path(&mddev->bitmap->storage.file->f_path,
5639 		     buf, sizeof(file->pathname));
5640 	if (IS_ERR(ptr))
5641 		goto out;
5642 
5643 	strcpy(file->pathname, ptr);
5644 
5645 copy_out:
5646 	err = 0;
5647 	if (copy_to_user(arg, file, sizeof(*file)))
5648 		err = -EFAULT;
5649 out:
5650 	kfree(buf);
5651 	kfree(file);
5652 	return err;
5653 }
5654 
5655 static int get_disk_info(struct mddev * mddev, void __user * arg)
5656 {
5657 	mdu_disk_info_t info;
5658 	struct md_rdev *rdev;
5659 
5660 	if (copy_from_user(&info, arg, sizeof(info)))
5661 		return -EFAULT;
5662 
5663 	rcu_read_lock();
5664 	rdev = find_rdev_nr_rcu(mddev, info.number);
5665 	if (rdev) {
5666 		info.major = MAJOR(rdev->bdev->bd_dev);
5667 		info.minor = MINOR(rdev->bdev->bd_dev);
5668 		info.raid_disk = rdev->raid_disk;
5669 		info.state = 0;
5670 		if (test_bit(Faulty, &rdev->flags))
5671 			info.state |= (1<<MD_DISK_FAULTY);
5672 		else if (test_bit(In_sync, &rdev->flags)) {
5673 			info.state |= (1<<MD_DISK_ACTIVE);
5674 			info.state |= (1<<MD_DISK_SYNC);
5675 		}
5676 		if (test_bit(WriteMostly, &rdev->flags))
5677 			info.state |= (1<<MD_DISK_WRITEMOSTLY);
5678 	} else {
5679 		info.major = info.minor = 0;
5680 		info.raid_disk = -1;
5681 		info.state = (1<<MD_DISK_REMOVED);
5682 	}
5683 	rcu_read_unlock();
5684 
5685 	if (copy_to_user(arg, &info, sizeof(info)))
5686 		return -EFAULT;
5687 
5688 	return 0;
5689 }
5690 
5691 static int add_new_disk(struct mddev * mddev, mdu_disk_info_t *info)
5692 {
5693 	char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
5694 	struct md_rdev *rdev;
5695 	dev_t dev = MKDEV(info->major,info->minor);
5696 
5697 	if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
5698 		return -EOVERFLOW;
5699 
5700 	if (!mddev->raid_disks) {
5701 		int err;
5702 		/* expecting a device which has a superblock */
5703 		rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
5704 		if (IS_ERR(rdev)) {
5705 			printk(KERN_WARNING
5706 				"md: md_import_device returned %ld\n",
5707 				PTR_ERR(rdev));
5708 			return PTR_ERR(rdev);
5709 		}
5710 		if (!list_empty(&mddev->disks)) {
5711 			struct md_rdev *rdev0
5712 				= list_entry(mddev->disks.next,
5713 					     struct md_rdev, same_set);
5714 			err = super_types[mddev->major_version]
5715 				.load_super(rdev, rdev0, mddev->minor_version);
5716 			if (err < 0) {
5717 				printk(KERN_WARNING
5718 					"md: %s has different UUID to %s\n",
5719 					bdevname(rdev->bdev,b),
5720 					bdevname(rdev0->bdev,b2));
5721 				export_rdev(rdev);
5722 				return -EINVAL;
5723 			}
5724 		}
5725 		err = bind_rdev_to_array(rdev, mddev);
5726 		if (err)
5727 			export_rdev(rdev);
5728 		return err;
5729 	}
5730 
5731 	/*
5732 	 * add_new_disk can be used once the array is assembled
5733 	 * to add "hot spares".  They must already have a superblock
5734 	 * written
5735 	 */
5736 	if (mddev->pers) {
5737 		int err;
5738 		if (!mddev->pers->hot_add_disk) {
5739 			printk(KERN_WARNING
5740 				"%s: personality does not support diskops!\n",
5741 			       mdname(mddev));
5742 			return -EINVAL;
5743 		}
5744 		if (mddev->persistent)
5745 			rdev = md_import_device(dev, mddev->major_version,
5746 						mddev->minor_version);
5747 		else
5748 			rdev = md_import_device(dev, -1, -1);
5749 		if (IS_ERR(rdev)) {
5750 			printk(KERN_WARNING
5751 				"md: md_import_device returned %ld\n",
5752 				PTR_ERR(rdev));
5753 			return PTR_ERR(rdev);
5754 		}
5755 		/* set saved_raid_disk if appropriate */
5756 		if (!mddev->persistent) {
5757 			if (info->state & (1<<MD_DISK_SYNC)  &&
5758 			    info->raid_disk < mddev->raid_disks) {
5759 				rdev->raid_disk = info->raid_disk;
5760 				set_bit(In_sync, &rdev->flags);
5761 				clear_bit(Bitmap_sync, &rdev->flags);
5762 			} else
5763 				rdev->raid_disk = -1;
5764 			rdev->saved_raid_disk = rdev->raid_disk;
5765 		} else
5766 			super_types[mddev->major_version].
5767 				validate_super(mddev, rdev);
5768 		if ((info->state & (1<<MD_DISK_SYNC)) &&
5769 		     rdev->raid_disk != info->raid_disk) {
5770 			/* This was a hot-add request, but events doesn't
5771 			 * match, so reject it.
5772 			 */
5773 			export_rdev(rdev);
5774 			return -EINVAL;
5775 		}
5776 
5777 		clear_bit(In_sync, &rdev->flags); /* just to be sure */
5778 		if (info->state & (1<<MD_DISK_WRITEMOSTLY))
5779 			set_bit(WriteMostly, &rdev->flags);
5780 		else
5781 			clear_bit(WriteMostly, &rdev->flags);
5782 
5783 		rdev->raid_disk = -1;
5784 		err = bind_rdev_to_array(rdev, mddev);
5785 		if (!err && !mddev->pers->hot_remove_disk) {
5786 			/* If there is hot_add_disk but no hot_remove_disk
5787 			 * then added disks for geometry changes,
5788 			 * and should be added immediately.
5789 			 */
5790 			super_types[mddev->major_version].
5791 				validate_super(mddev, rdev);
5792 			err = mddev->pers->hot_add_disk(mddev, rdev);
5793 			if (err)
5794 				unbind_rdev_from_array(rdev);
5795 		}
5796 		if (err)
5797 			export_rdev(rdev);
5798 		else
5799 			sysfs_notify_dirent_safe(rdev->sysfs_state);
5800 
5801 		set_bit(MD_CHANGE_DEVS, &mddev->flags);
5802 		if (mddev->degraded)
5803 			set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
5804 		set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5805 		if (!err)
5806 			md_new_event(mddev);
5807 		md_wakeup_thread(mddev->thread);
5808 		return err;
5809 	}
5810 
5811 	/* otherwise, add_new_disk is only allowed
5812 	 * for major_version==0 superblocks
5813 	 */
5814 	if (mddev->major_version != 0) {
5815 		printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n",
5816 		       mdname(mddev));
5817 		return -EINVAL;
5818 	}
5819 
5820 	if (!(info->state & (1<<MD_DISK_FAULTY))) {
5821 		int err;
5822 		rdev = md_import_device(dev, -1, 0);
5823 		if (IS_ERR(rdev)) {
5824 			printk(KERN_WARNING
5825 				"md: error, md_import_device() returned %ld\n",
5826 				PTR_ERR(rdev));
5827 			return PTR_ERR(rdev);
5828 		}
5829 		rdev->desc_nr = info->number;
5830 		if (info->raid_disk < mddev->raid_disks)
5831 			rdev->raid_disk = info->raid_disk;
5832 		else
5833 			rdev->raid_disk = -1;
5834 
5835 		if (rdev->raid_disk < mddev->raid_disks)
5836 			if (info->state & (1<<MD_DISK_SYNC))
5837 				set_bit(In_sync, &rdev->flags);
5838 
5839 		if (info->state & (1<<MD_DISK_WRITEMOSTLY))
5840 			set_bit(WriteMostly, &rdev->flags);
5841 
5842 		if (!mddev->persistent) {
5843 			printk(KERN_INFO "md: nonpersistent superblock ...\n");
5844 			rdev->sb_start = i_size_read(rdev->bdev->bd_inode) / 512;
5845 		} else
5846 			rdev->sb_start = calc_dev_sboffset(rdev);
5847 		rdev->sectors = rdev->sb_start;
5848 
5849 		err = bind_rdev_to_array(rdev, mddev);
5850 		if (err) {
5851 			export_rdev(rdev);
5852 			return err;
5853 		}
5854 	}
5855 
5856 	return 0;
5857 }
5858 
5859 static int hot_remove_disk(struct mddev * mddev, dev_t dev)
5860 {
5861 	char b[BDEVNAME_SIZE];
5862 	struct md_rdev *rdev;
5863 
5864 	rdev = find_rdev(mddev, dev);
5865 	if (!rdev)
5866 		return -ENXIO;
5867 
5868 	clear_bit(Blocked, &rdev->flags);
5869 	remove_and_add_spares(mddev, rdev);
5870 
5871 	if (rdev->raid_disk >= 0)
5872 		goto busy;
5873 
5874 	kick_rdev_from_array(rdev);
5875 	md_update_sb(mddev, 1);
5876 	md_new_event(mddev);
5877 
5878 	return 0;
5879 busy:
5880 	printk(KERN_WARNING "md: cannot remove active disk %s from %s ...\n",
5881 		bdevname(rdev->bdev,b), mdname(mddev));
5882 	return -EBUSY;
5883 }
5884 
5885 static int hot_add_disk(struct mddev * mddev, dev_t dev)
5886 {
5887 	char b[BDEVNAME_SIZE];
5888 	int err;
5889 	struct md_rdev *rdev;
5890 
5891 	if (!mddev->pers)
5892 		return -ENODEV;
5893 
5894 	if (mddev->major_version != 0) {
5895 		printk(KERN_WARNING "%s: HOT_ADD may only be used with"
5896 			" version-0 superblocks.\n",
5897 			mdname(mddev));
5898 		return -EINVAL;
5899 	}
5900 	if (!mddev->pers->hot_add_disk) {
5901 		printk(KERN_WARNING
5902 			"%s: personality does not support diskops!\n",
5903 			mdname(mddev));
5904 		return -EINVAL;
5905 	}
5906 
5907 	rdev = md_import_device(dev, -1, 0);
5908 	if (IS_ERR(rdev)) {
5909 		printk(KERN_WARNING
5910 			"md: error, md_import_device() returned %ld\n",
5911 			PTR_ERR(rdev));
5912 		return -EINVAL;
5913 	}
5914 
5915 	if (mddev->persistent)
5916 		rdev->sb_start = calc_dev_sboffset(rdev);
5917 	else
5918 		rdev->sb_start = i_size_read(rdev->bdev->bd_inode) / 512;
5919 
5920 	rdev->sectors = rdev->sb_start;
5921 
5922 	if (test_bit(Faulty, &rdev->flags)) {
5923 		printk(KERN_WARNING
5924 			"md: can not hot-add faulty %s disk to %s!\n",
5925 			bdevname(rdev->bdev,b), mdname(mddev));
5926 		err = -EINVAL;
5927 		goto abort_export;
5928 	}
5929 	clear_bit(In_sync, &rdev->flags);
5930 	rdev->desc_nr = -1;
5931 	rdev->saved_raid_disk = -1;
5932 	err = bind_rdev_to_array(rdev, mddev);
5933 	if (err)
5934 		goto abort_export;
5935 
5936 	/*
5937 	 * The rest should better be atomic, we can have disk failures
5938 	 * noticed in interrupt contexts ...
5939 	 */
5940 
5941 	rdev->raid_disk = -1;
5942 
5943 	md_update_sb(mddev, 1);
5944 
5945 	/*
5946 	 * Kick recovery, maybe this spare has to be added to the
5947 	 * array immediately.
5948 	 */
5949 	set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5950 	md_wakeup_thread(mddev->thread);
5951 	md_new_event(mddev);
5952 	return 0;
5953 
5954 abort_export:
5955 	export_rdev(rdev);
5956 	return err;
5957 }
5958 
5959 static int set_bitmap_file(struct mddev *mddev, int fd)
5960 {
5961 	int err = 0;
5962 
5963 	if (mddev->pers) {
5964 		if (!mddev->pers->quiesce)
5965 			return -EBUSY;
5966 		if (mddev->recovery || mddev->sync_thread)
5967 			return -EBUSY;
5968 		/* we should be able to change the bitmap.. */
5969 	}
5970 
5971 
5972 	if (fd >= 0) {
5973 		struct inode *inode;
5974 		if (mddev->bitmap)
5975 			return -EEXIST; /* cannot add when bitmap is present */
5976 		mddev->bitmap_info.file = fget(fd);
5977 
5978 		if (mddev->bitmap_info.file == NULL) {
5979 			printk(KERN_ERR "%s: error: failed to get bitmap file\n",
5980 			       mdname(mddev));
5981 			return -EBADF;
5982 		}
5983 
5984 		inode = mddev->bitmap_info.file->f_mapping->host;
5985 		if (!S_ISREG(inode->i_mode)) {
5986 			printk(KERN_ERR "%s: error: bitmap file must be a regular file\n",
5987 			       mdname(mddev));
5988 			err = -EBADF;
5989 		} else if (!(mddev->bitmap_info.file->f_mode & FMODE_WRITE)) {
5990 			printk(KERN_ERR "%s: error: bitmap file must open for write\n",
5991 			       mdname(mddev));
5992 			err = -EBADF;
5993 		} else if (atomic_read(&inode->i_writecount) != 1) {
5994 			printk(KERN_ERR "%s: error: bitmap file is already in use\n",
5995 			       mdname(mddev));
5996 			err = -EBUSY;
5997 		}
5998 		if (err) {
5999 			fput(mddev->bitmap_info.file);
6000 			mddev->bitmap_info.file = NULL;
6001 			return err;
6002 		}
6003 		mddev->bitmap_info.offset = 0; /* file overrides offset */
6004 	} else if (mddev->bitmap == NULL)
6005 		return -ENOENT; /* cannot remove what isn't there */
6006 	err = 0;
6007 	if (mddev->pers) {
6008 		mddev->pers->quiesce(mddev, 1);
6009 		if (fd >= 0) {
6010 			err = bitmap_create(mddev);
6011 			if (!err)
6012 				err = bitmap_load(mddev);
6013 		}
6014 		if (fd < 0 || err) {
6015 			bitmap_destroy(mddev);
6016 			fd = -1; /* make sure to put the file */
6017 		}
6018 		mddev->pers->quiesce(mddev, 0);
6019 	}
6020 	if (fd < 0) {
6021 		if (mddev->bitmap_info.file)
6022 			fput(mddev->bitmap_info.file);
6023 		mddev->bitmap_info.file = NULL;
6024 	}
6025 
6026 	return err;
6027 }
6028 
6029 /*
6030  * set_array_info is used two different ways
6031  * The original usage is when creating a new array.
6032  * In this usage, raid_disks is > 0 and it together with
6033  *  level, size, not_persistent,layout,chunksize determine the
6034  *  shape of the array.
6035  *  This will always create an array with a type-0.90.0 superblock.
6036  * The newer usage is when assembling an array.
6037  *  In this case raid_disks will be 0, and the major_version field is
6038  *  use to determine which style super-blocks are to be found on the devices.
6039  *  The minor and patch _version numbers are also kept incase the
6040  *  super_block handler wishes to interpret them.
6041  */
6042 static int set_array_info(struct mddev * mddev, mdu_array_info_t *info)
6043 {
6044 
6045 	if (info->raid_disks == 0) {
6046 		/* just setting version number for superblock loading */
6047 		if (info->major_version < 0 ||
6048 		    info->major_version >= ARRAY_SIZE(super_types) ||
6049 		    super_types[info->major_version].name == NULL) {
6050 			/* maybe try to auto-load a module? */
6051 			printk(KERN_INFO
6052 				"md: superblock version %d not known\n",
6053 				info->major_version);
6054 			return -EINVAL;
6055 		}
6056 		mddev->major_version = info->major_version;
6057 		mddev->minor_version = info->minor_version;
6058 		mddev->patch_version = info->patch_version;
6059 		mddev->persistent = !info->not_persistent;
6060 		/* ensure mddev_put doesn't delete this now that there
6061 		 * is some minimal configuration.
6062 		 */
6063 		mddev->ctime         = get_seconds();
6064 		return 0;
6065 	}
6066 	mddev->major_version = MD_MAJOR_VERSION;
6067 	mddev->minor_version = MD_MINOR_VERSION;
6068 	mddev->patch_version = MD_PATCHLEVEL_VERSION;
6069 	mddev->ctime         = get_seconds();
6070 
6071 	mddev->level         = info->level;
6072 	mddev->clevel[0]     = 0;
6073 	mddev->dev_sectors   = 2 * (sector_t)info->size;
6074 	mddev->raid_disks    = info->raid_disks;
6075 	/* don't set md_minor, it is determined by which /dev/md* was
6076 	 * openned
6077 	 */
6078 	if (info->state & (1<<MD_SB_CLEAN))
6079 		mddev->recovery_cp = MaxSector;
6080 	else
6081 		mddev->recovery_cp = 0;
6082 	mddev->persistent    = ! info->not_persistent;
6083 	mddev->external	     = 0;
6084 
6085 	mddev->layout        = info->layout;
6086 	mddev->chunk_sectors = info->chunk_size >> 9;
6087 
6088 	mddev->max_disks     = MD_SB_DISKS;
6089 
6090 	if (mddev->persistent)
6091 		mddev->flags         = 0;
6092 	set_bit(MD_CHANGE_DEVS, &mddev->flags);
6093 
6094 	mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9;
6095 	mddev->bitmap_info.default_space = 64*2 - (MD_SB_BYTES >> 9);
6096 	mddev->bitmap_info.offset = 0;
6097 
6098 	mddev->reshape_position = MaxSector;
6099 
6100 	/*
6101 	 * Generate a 128 bit UUID
6102 	 */
6103 	get_random_bytes(mddev->uuid, 16);
6104 
6105 	mddev->new_level = mddev->level;
6106 	mddev->new_chunk_sectors = mddev->chunk_sectors;
6107 	mddev->new_layout = mddev->layout;
6108 	mddev->delta_disks = 0;
6109 	mddev->reshape_backwards = 0;
6110 
6111 	return 0;
6112 }
6113 
6114 void md_set_array_sectors(struct mddev *mddev, sector_t array_sectors)
6115 {
6116 	WARN(!mddev_is_locked(mddev), "%s: unlocked mddev!\n", __func__);
6117 
6118 	if (mddev->external_size)
6119 		return;
6120 
6121 	mddev->array_sectors = array_sectors;
6122 }
6123 EXPORT_SYMBOL(md_set_array_sectors);
6124 
6125 static int update_size(struct mddev *mddev, sector_t num_sectors)
6126 {
6127 	struct md_rdev *rdev;
6128 	int rv;
6129 	int fit = (num_sectors == 0);
6130 
6131 	if (mddev->pers->resize == NULL)
6132 		return -EINVAL;
6133 	/* The "num_sectors" is the number of sectors of each device that
6134 	 * is used.  This can only make sense for arrays with redundancy.
6135 	 * linear and raid0 always use whatever space is available. We can only
6136 	 * consider changing this number if no resync or reconstruction is
6137 	 * happening, and if the new size is acceptable. It must fit before the
6138 	 * sb_start or, if that is <data_offset, it must fit before the size
6139 	 * of each device.  If num_sectors is zero, we find the largest size
6140 	 * that fits.
6141 	 */
6142 	if (mddev->sync_thread)
6143 		return -EBUSY;
6144 	if (mddev->ro)
6145 		return -EROFS;
6146 
6147 	rdev_for_each(rdev, mddev) {
6148 		sector_t avail = rdev->sectors;
6149 
6150 		if (fit && (num_sectors == 0 || num_sectors > avail))
6151 			num_sectors = avail;
6152 		if (avail < num_sectors)
6153 			return -ENOSPC;
6154 	}
6155 	rv = mddev->pers->resize(mddev, num_sectors);
6156 	if (!rv)
6157 		revalidate_disk(mddev->gendisk);
6158 	return rv;
6159 }
6160 
6161 static int update_raid_disks(struct mddev *mddev, int raid_disks)
6162 {
6163 	int rv;
6164 	struct md_rdev *rdev;
6165 	/* change the number of raid disks */
6166 	if (mddev->pers->check_reshape == NULL)
6167 		return -EINVAL;
6168 	if (mddev->ro)
6169 		return -EROFS;
6170 	if (raid_disks <= 0 ||
6171 	    (mddev->max_disks && raid_disks >= mddev->max_disks))
6172 		return -EINVAL;
6173 	if (mddev->sync_thread || mddev->reshape_position != MaxSector)
6174 		return -EBUSY;
6175 
6176 	rdev_for_each(rdev, mddev) {
6177 		if (mddev->raid_disks < raid_disks &&
6178 		    rdev->data_offset < rdev->new_data_offset)
6179 			return -EINVAL;
6180 		if (mddev->raid_disks > raid_disks &&
6181 		    rdev->data_offset > rdev->new_data_offset)
6182 			return -EINVAL;
6183 	}
6184 
6185 	mddev->delta_disks = raid_disks - mddev->raid_disks;
6186 	if (mddev->delta_disks < 0)
6187 		mddev->reshape_backwards = 1;
6188 	else if (mddev->delta_disks > 0)
6189 		mddev->reshape_backwards = 0;
6190 
6191 	rv = mddev->pers->check_reshape(mddev);
6192 	if (rv < 0) {
6193 		mddev->delta_disks = 0;
6194 		mddev->reshape_backwards = 0;
6195 	}
6196 	return rv;
6197 }
6198 
6199 
6200 /*
6201  * update_array_info is used to change the configuration of an
6202  * on-line array.
6203  * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
6204  * fields in the info are checked against the array.
6205  * Any differences that cannot be handled will cause an error.
6206  * Normally, only one change can be managed at a time.
6207  */
6208 static int update_array_info(struct mddev *mddev, mdu_array_info_t *info)
6209 {
6210 	int rv = 0;
6211 	int cnt = 0;
6212 	int state = 0;
6213 
6214 	/* calculate expected state,ignoring low bits */
6215 	if (mddev->bitmap && mddev->bitmap_info.offset)
6216 		state |= (1 << MD_SB_BITMAP_PRESENT);
6217 
6218 	if (mddev->major_version != info->major_version ||
6219 	    mddev->minor_version != info->minor_version ||
6220 /*	    mddev->patch_version != info->patch_version || */
6221 	    mddev->ctime         != info->ctime         ||
6222 	    mddev->level         != info->level         ||
6223 /*	    mddev->layout        != info->layout        || */
6224 	    !mddev->persistent	 != info->not_persistent||
6225 	    mddev->chunk_sectors != info->chunk_size >> 9 ||
6226 	    /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
6227 	    ((state^info->state) & 0xfffffe00)
6228 		)
6229 		return -EINVAL;
6230 	/* Check there is only one change */
6231 	if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
6232 		cnt++;
6233 	if (mddev->raid_disks != info->raid_disks)
6234 		cnt++;
6235 	if (mddev->layout != info->layout)
6236 		cnt++;
6237 	if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT))
6238 		cnt++;
6239 	if (cnt == 0)
6240 		return 0;
6241 	if (cnt > 1)
6242 		return -EINVAL;
6243 
6244 	if (mddev->layout != info->layout) {
6245 		/* Change layout
6246 		 * we don't need to do anything at the md level, the
6247 		 * personality will take care of it all.
6248 		 */
6249 		if (mddev->pers->check_reshape == NULL)
6250 			return -EINVAL;
6251 		else {
6252 			mddev->new_layout = info->layout;
6253 			rv = mddev->pers->check_reshape(mddev);
6254 			if (rv)
6255 				mddev->new_layout = mddev->layout;
6256 			return rv;
6257 		}
6258 	}
6259 	if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
6260 		rv = update_size(mddev, (sector_t)info->size * 2);
6261 
6262 	if (mddev->raid_disks    != info->raid_disks)
6263 		rv = update_raid_disks(mddev, info->raid_disks);
6264 
6265 	if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) {
6266 		if (mddev->pers->quiesce == NULL)
6267 			return -EINVAL;
6268 		if (mddev->recovery || mddev->sync_thread)
6269 			return -EBUSY;
6270 		if (info->state & (1<<MD_SB_BITMAP_PRESENT)) {
6271 			/* add the bitmap */
6272 			if (mddev->bitmap)
6273 				return -EEXIST;
6274 			if (mddev->bitmap_info.default_offset == 0)
6275 				return -EINVAL;
6276 			mddev->bitmap_info.offset =
6277 				mddev->bitmap_info.default_offset;
6278 			mddev->bitmap_info.space =
6279 				mddev->bitmap_info.default_space;
6280 			mddev->pers->quiesce(mddev, 1);
6281 			rv = bitmap_create(mddev);
6282 			if (!rv)
6283 				rv = bitmap_load(mddev);
6284 			if (rv)
6285 				bitmap_destroy(mddev);
6286 			mddev->pers->quiesce(mddev, 0);
6287 		} else {
6288 			/* remove the bitmap */
6289 			if (!mddev->bitmap)
6290 				return -ENOENT;
6291 			if (mddev->bitmap->storage.file)
6292 				return -EINVAL;
6293 			mddev->pers->quiesce(mddev, 1);
6294 			bitmap_destroy(mddev);
6295 			mddev->pers->quiesce(mddev, 0);
6296 			mddev->bitmap_info.offset = 0;
6297 		}
6298 	}
6299 	md_update_sb(mddev, 1);
6300 	return rv;
6301 }
6302 
6303 static int set_disk_faulty(struct mddev *mddev, dev_t dev)
6304 {
6305 	struct md_rdev *rdev;
6306 	int err = 0;
6307 
6308 	if (mddev->pers == NULL)
6309 		return -ENODEV;
6310 
6311 	rcu_read_lock();
6312 	rdev = find_rdev_rcu(mddev, dev);
6313 	if (!rdev)
6314 		err =  -ENODEV;
6315 	else {
6316 		md_error(mddev, rdev);
6317 		if (!test_bit(Faulty, &rdev->flags))
6318 			err = -EBUSY;
6319 	}
6320 	rcu_read_unlock();
6321 	return err;
6322 }
6323 
6324 /*
6325  * We have a problem here : there is no easy way to give a CHS
6326  * virtual geometry. We currently pretend that we have a 2 heads
6327  * 4 sectors (with a BIG number of cylinders...). This drives
6328  * dosfs just mad... ;-)
6329  */
6330 static int md_getgeo(struct block_device *bdev, struct hd_geometry *geo)
6331 {
6332 	struct mddev *mddev = bdev->bd_disk->private_data;
6333 
6334 	geo->heads = 2;
6335 	geo->sectors = 4;
6336 	geo->cylinders = mddev->array_sectors / 8;
6337 	return 0;
6338 }
6339 
6340 static inline bool md_ioctl_valid(unsigned int cmd)
6341 {
6342 	switch (cmd) {
6343 	case ADD_NEW_DISK:
6344 	case BLKROSET:
6345 	case GET_ARRAY_INFO:
6346 	case GET_BITMAP_FILE:
6347 	case GET_DISK_INFO:
6348 	case HOT_ADD_DISK:
6349 	case HOT_REMOVE_DISK:
6350 	case PRINT_RAID_DEBUG:
6351 	case RAID_AUTORUN:
6352 	case RAID_VERSION:
6353 	case RESTART_ARRAY_RW:
6354 	case RUN_ARRAY:
6355 	case SET_ARRAY_INFO:
6356 	case SET_BITMAP_FILE:
6357 	case SET_DISK_FAULTY:
6358 	case STOP_ARRAY:
6359 	case STOP_ARRAY_RO:
6360 		return true;
6361 	default:
6362 		return false;
6363 	}
6364 }
6365 
6366 static int md_ioctl(struct block_device *bdev, fmode_t mode,
6367 			unsigned int cmd, unsigned long arg)
6368 {
6369 	int err = 0;
6370 	void __user *argp = (void __user *)arg;
6371 	struct mddev *mddev = NULL;
6372 	int ro;
6373 
6374 	if (!md_ioctl_valid(cmd))
6375 		return -ENOTTY;
6376 
6377 	switch (cmd) {
6378 	case RAID_VERSION:
6379 	case GET_ARRAY_INFO:
6380 	case GET_DISK_INFO:
6381 		break;
6382 	default:
6383 		if (!capable(CAP_SYS_ADMIN))
6384 			return -EACCES;
6385 	}
6386 
6387 	/*
6388 	 * Commands dealing with the RAID driver but not any
6389 	 * particular array:
6390 	 */
6391 	switch (cmd) {
6392 	case RAID_VERSION:
6393 		err = get_version(argp);
6394 		goto done;
6395 
6396 	case PRINT_RAID_DEBUG:
6397 		err = 0;
6398 		md_print_devices();
6399 		goto done;
6400 
6401 #ifndef MODULE
6402 	case RAID_AUTORUN:
6403 		err = 0;
6404 		autostart_arrays(arg);
6405 		goto done;
6406 #endif
6407 	default:;
6408 	}
6409 
6410 	/*
6411 	 * Commands creating/starting a new array:
6412 	 */
6413 
6414 	mddev = bdev->bd_disk->private_data;
6415 
6416 	if (!mddev) {
6417 		BUG();
6418 		goto abort;
6419 	}
6420 
6421 	/* Some actions do not requires the mutex */
6422 	switch (cmd) {
6423 	case GET_ARRAY_INFO:
6424 		if (!mddev->raid_disks && !mddev->external)
6425 			err = -ENODEV;
6426 		else
6427 			err = get_array_info(mddev, argp);
6428 		goto abort;
6429 
6430 	case GET_DISK_INFO:
6431 		if (!mddev->raid_disks && !mddev->external)
6432 			err = -ENODEV;
6433 		else
6434 			err = get_disk_info(mddev, argp);
6435 		goto abort;
6436 
6437 	case SET_DISK_FAULTY:
6438 		err = set_disk_faulty(mddev, new_decode_dev(arg));
6439 		goto abort;
6440 	}
6441 
6442 	if (cmd == ADD_NEW_DISK)
6443 		/* need to ensure md_delayed_delete() has completed */
6444 		flush_workqueue(md_misc_wq);
6445 
6446 	if (cmd == HOT_REMOVE_DISK)
6447 		/* need to ensure recovery thread has run */
6448 		wait_event_interruptible_timeout(mddev->sb_wait,
6449 						 !test_bit(MD_RECOVERY_NEEDED,
6450 							   &mddev->flags),
6451 						 msecs_to_jiffies(5000));
6452 	if (cmd == STOP_ARRAY || cmd == STOP_ARRAY_RO) {
6453 		/* Need to flush page cache, and ensure no-one else opens
6454 		 * and writes
6455 		 */
6456 		mutex_lock(&mddev->open_mutex);
6457 		if (atomic_read(&mddev->openers) > 1) {
6458 			mutex_unlock(&mddev->open_mutex);
6459 			err = -EBUSY;
6460 			goto abort;
6461 		}
6462 		set_bit(MD_STILL_CLOSED, &mddev->flags);
6463 		mutex_unlock(&mddev->open_mutex);
6464 		sync_blockdev(bdev);
6465 	}
6466 	err = mddev_lock(mddev);
6467 	if (err) {
6468 		printk(KERN_INFO
6469 			"md: ioctl lock interrupted, reason %d, cmd %d\n",
6470 			err, cmd);
6471 		goto abort;
6472 	}
6473 
6474 	if (cmd == SET_ARRAY_INFO) {
6475 		mdu_array_info_t info;
6476 		if (!arg)
6477 			memset(&info, 0, sizeof(info));
6478 		else if (copy_from_user(&info, argp, sizeof(info))) {
6479 			err = -EFAULT;
6480 			goto abort_unlock;
6481 		}
6482 		if (mddev->pers) {
6483 			err = update_array_info(mddev, &info);
6484 			if (err) {
6485 				printk(KERN_WARNING "md: couldn't update"
6486 				       " array info. %d\n", err);
6487 				goto abort_unlock;
6488 			}
6489 			goto done_unlock;
6490 		}
6491 		if (!list_empty(&mddev->disks)) {
6492 			printk(KERN_WARNING
6493 			       "md: array %s already has disks!\n",
6494 			       mdname(mddev));
6495 			err = -EBUSY;
6496 			goto abort_unlock;
6497 		}
6498 		if (mddev->raid_disks) {
6499 			printk(KERN_WARNING
6500 			       "md: array %s already initialised!\n",
6501 			       mdname(mddev));
6502 			err = -EBUSY;
6503 			goto abort_unlock;
6504 		}
6505 		err = set_array_info(mddev, &info);
6506 		if (err) {
6507 			printk(KERN_WARNING "md: couldn't set"
6508 			       " array info. %d\n", err);
6509 			goto abort_unlock;
6510 		}
6511 		goto done_unlock;
6512 	}
6513 
6514 	/*
6515 	 * Commands querying/configuring an existing array:
6516 	 */
6517 	/* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
6518 	 * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */
6519 	if ((!mddev->raid_disks && !mddev->external)
6520 	    && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY
6521 	    && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE
6522 	    && cmd != GET_BITMAP_FILE) {
6523 		err = -ENODEV;
6524 		goto abort_unlock;
6525 	}
6526 
6527 	/*
6528 	 * Commands even a read-only array can execute:
6529 	 */
6530 	switch (cmd) {
6531 	case GET_BITMAP_FILE:
6532 		err = get_bitmap_file(mddev, argp);
6533 		goto done_unlock;
6534 
6535 	case RESTART_ARRAY_RW:
6536 		err = restart_array(mddev);
6537 		goto done_unlock;
6538 
6539 	case STOP_ARRAY:
6540 		err = do_md_stop(mddev, 0, bdev);
6541 		goto done_unlock;
6542 
6543 	case STOP_ARRAY_RO:
6544 		err = md_set_readonly(mddev, bdev);
6545 		goto done_unlock;
6546 
6547 	case HOT_REMOVE_DISK:
6548 		err = hot_remove_disk(mddev, new_decode_dev(arg));
6549 		goto done_unlock;
6550 
6551 	case ADD_NEW_DISK:
6552 		/* We can support ADD_NEW_DISK on read-only arrays
6553 		 * on if we are re-adding a preexisting device.
6554 		 * So require mddev->pers and MD_DISK_SYNC.
6555 		 */
6556 		if (mddev->pers) {
6557 			mdu_disk_info_t info;
6558 			if (copy_from_user(&info, argp, sizeof(info)))
6559 				err = -EFAULT;
6560 			else if (!(info.state & (1<<MD_DISK_SYNC)))
6561 				/* Need to clear read-only for this */
6562 				break;
6563 			else
6564 				err = add_new_disk(mddev, &info);
6565 			goto done_unlock;
6566 		}
6567 		break;
6568 
6569 	case BLKROSET:
6570 		if (get_user(ro, (int __user *)(arg))) {
6571 			err = -EFAULT;
6572 			goto done_unlock;
6573 		}
6574 		err = -EINVAL;
6575 
6576 		/* if the bdev is going readonly the value of mddev->ro
6577 		 * does not matter, no writes are coming
6578 		 */
6579 		if (ro)
6580 			goto done_unlock;
6581 
6582 		/* are we are already prepared for writes? */
6583 		if (mddev->ro != 1)
6584 			goto done_unlock;
6585 
6586 		/* transitioning to readauto need only happen for
6587 		 * arrays that call md_write_start
6588 		 */
6589 		if (mddev->pers) {
6590 			err = restart_array(mddev);
6591 			if (err == 0) {
6592 				mddev->ro = 2;
6593 				set_disk_ro(mddev->gendisk, 0);
6594 			}
6595 		}
6596 		goto done_unlock;
6597 	}
6598 
6599 	/*
6600 	 * The remaining ioctls are changing the state of the
6601 	 * superblock, so we do not allow them on read-only arrays.
6602 	 * However non-MD ioctls (e.g. get-size) will still come through
6603 	 * here and hit the 'default' below, so only disallow
6604 	 * 'md' ioctls, and switch to rw mode if started auto-readonly.
6605 	 */
6606 	if (_IOC_TYPE(cmd) == MD_MAJOR && mddev->ro && mddev->pers) {
6607 		if (mddev->ro == 2) {
6608 			mddev->ro = 0;
6609 			sysfs_notify_dirent_safe(mddev->sysfs_state);
6610 			set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6611 			/* mddev_unlock will wake thread */
6612 			/* If a device failed while we were read-only, we
6613 			 * need to make sure the metadata is updated now.
6614 			 */
6615 			if (test_bit(MD_CHANGE_DEVS, &mddev->flags)) {
6616 				mddev_unlock(mddev);
6617 				wait_event(mddev->sb_wait,
6618 					   !test_bit(MD_CHANGE_DEVS, &mddev->flags) &&
6619 					   !test_bit(MD_CHANGE_PENDING, &mddev->flags));
6620 				mddev_lock_nointr(mddev);
6621 			}
6622 		} else {
6623 			err = -EROFS;
6624 			goto abort_unlock;
6625 		}
6626 	}
6627 
6628 	switch (cmd) {
6629 	case ADD_NEW_DISK:
6630 	{
6631 		mdu_disk_info_t info;
6632 		if (copy_from_user(&info, argp, sizeof(info)))
6633 			err = -EFAULT;
6634 		else
6635 			err = add_new_disk(mddev, &info);
6636 		goto done_unlock;
6637 	}
6638 
6639 	case HOT_ADD_DISK:
6640 		err = hot_add_disk(mddev, new_decode_dev(arg));
6641 		goto done_unlock;
6642 
6643 	case RUN_ARRAY:
6644 		err = do_md_run(mddev);
6645 		goto done_unlock;
6646 
6647 	case SET_BITMAP_FILE:
6648 		err = set_bitmap_file(mddev, (int)arg);
6649 		goto done_unlock;
6650 
6651 	default:
6652 		err = -EINVAL;
6653 		goto abort_unlock;
6654 	}
6655 
6656 done_unlock:
6657 abort_unlock:
6658 	if (mddev->hold_active == UNTIL_IOCTL &&
6659 	    err != -EINVAL)
6660 		mddev->hold_active = 0;
6661 	mddev_unlock(mddev);
6662 
6663 	return err;
6664 done:
6665 	if (err)
6666 		MD_BUG();
6667 abort:
6668 	return err;
6669 }
6670 #ifdef CONFIG_COMPAT
6671 static int md_compat_ioctl(struct block_device *bdev, fmode_t mode,
6672 		    unsigned int cmd, unsigned long arg)
6673 {
6674 	switch (cmd) {
6675 	case HOT_REMOVE_DISK:
6676 	case HOT_ADD_DISK:
6677 	case SET_DISK_FAULTY:
6678 	case SET_BITMAP_FILE:
6679 		/* These take in integer arg, do not convert */
6680 		break;
6681 	default:
6682 		arg = (unsigned long)compat_ptr(arg);
6683 		break;
6684 	}
6685 
6686 	return md_ioctl(bdev, mode, cmd, arg);
6687 }
6688 #endif /* CONFIG_COMPAT */
6689 
6690 static int md_open(struct block_device *bdev, fmode_t mode)
6691 {
6692 	/*
6693 	 * Succeed if we can lock the mddev, which confirms that
6694 	 * it isn't being stopped right now.
6695 	 */
6696 	struct mddev *mddev = mddev_find(bdev->bd_dev);
6697 	int err;
6698 
6699 	if (!mddev)
6700 		return -ENODEV;
6701 
6702 	if (mddev->gendisk != bdev->bd_disk) {
6703 		/* we are racing with mddev_put which is discarding this
6704 		 * bd_disk.
6705 		 */
6706 		mddev_put(mddev);
6707 		/* Wait until bdev->bd_disk is definitely gone */
6708 		flush_workqueue(md_misc_wq);
6709 		/* Then retry the open from the top */
6710 		return -ERESTARTSYS;
6711 	}
6712 	BUG_ON(mddev != bdev->bd_disk->private_data);
6713 
6714 	if ((err = mutex_lock_interruptible(&mddev->open_mutex)))
6715 		goto out;
6716 
6717 	err = 0;
6718 	atomic_inc(&mddev->openers);
6719 	clear_bit(MD_STILL_CLOSED, &mddev->flags);
6720 	mutex_unlock(&mddev->open_mutex);
6721 
6722 	check_disk_change(bdev);
6723  out:
6724 	return err;
6725 }
6726 
6727 static void md_release(struct gendisk *disk, fmode_t mode)
6728 {
6729  	struct mddev *mddev = disk->private_data;
6730 
6731 	BUG_ON(!mddev);
6732 	atomic_dec(&mddev->openers);
6733 	mddev_put(mddev);
6734 }
6735 
6736 static int md_media_changed(struct gendisk *disk)
6737 {
6738 	struct mddev *mddev = disk->private_data;
6739 
6740 	return mddev->changed;
6741 }
6742 
6743 static int md_revalidate(struct gendisk *disk)
6744 {
6745 	struct mddev *mddev = disk->private_data;
6746 
6747 	mddev->changed = 0;
6748 	return 0;
6749 }
6750 static const struct block_device_operations md_fops =
6751 {
6752 	.owner		= THIS_MODULE,
6753 	.open		= md_open,
6754 	.release	= md_release,
6755 	.ioctl		= md_ioctl,
6756 #ifdef CONFIG_COMPAT
6757 	.compat_ioctl	= md_compat_ioctl,
6758 #endif
6759 	.getgeo		= md_getgeo,
6760 	.media_changed  = md_media_changed,
6761 	.revalidate_disk= md_revalidate,
6762 };
6763 
6764 static int md_thread(void * arg)
6765 {
6766 	struct md_thread *thread = arg;
6767 
6768 	/*
6769 	 * md_thread is a 'system-thread', it's priority should be very
6770 	 * high. We avoid resource deadlocks individually in each
6771 	 * raid personality. (RAID5 does preallocation) We also use RR and
6772 	 * the very same RT priority as kswapd, thus we will never get
6773 	 * into a priority inversion deadlock.
6774 	 *
6775 	 * we definitely have to have equal or higher priority than
6776 	 * bdflush, otherwise bdflush will deadlock if there are too
6777 	 * many dirty RAID5 blocks.
6778 	 */
6779 
6780 	allow_signal(SIGKILL);
6781 	while (!kthread_should_stop()) {
6782 
6783 		/* We need to wait INTERRUPTIBLE so that
6784 		 * we don't add to the load-average.
6785 		 * That means we need to be sure no signals are
6786 		 * pending
6787 		 */
6788 		if (signal_pending(current))
6789 			flush_signals(current);
6790 
6791 		wait_event_interruptible_timeout
6792 			(thread->wqueue,
6793 			 test_bit(THREAD_WAKEUP, &thread->flags)
6794 			 || kthread_should_stop(),
6795 			 thread->timeout);
6796 
6797 		clear_bit(THREAD_WAKEUP, &thread->flags);
6798 		if (!kthread_should_stop())
6799 			thread->run(thread);
6800 	}
6801 
6802 	return 0;
6803 }
6804 
6805 void md_wakeup_thread(struct md_thread *thread)
6806 {
6807 	if (thread) {
6808 		pr_debug("md: waking up MD thread %s.\n", thread->tsk->comm);
6809 		set_bit(THREAD_WAKEUP, &thread->flags);
6810 		wake_up(&thread->wqueue);
6811 	}
6812 }
6813 
6814 struct md_thread *md_register_thread(void (*run) (struct md_thread *),
6815 		struct mddev *mddev, const char *name)
6816 {
6817 	struct md_thread *thread;
6818 
6819 	thread = kzalloc(sizeof(struct md_thread), GFP_KERNEL);
6820 	if (!thread)
6821 		return NULL;
6822 
6823 	init_waitqueue_head(&thread->wqueue);
6824 
6825 	thread->run = run;
6826 	thread->mddev = mddev;
6827 	thread->timeout = MAX_SCHEDULE_TIMEOUT;
6828 	thread->tsk = kthread_run(md_thread, thread,
6829 				  "%s_%s",
6830 				  mdname(thread->mddev),
6831 				  name);
6832 	if (IS_ERR(thread->tsk)) {
6833 		kfree(thread);
6834 		return NULL;
6835 	}
6836 	return thread;
6837 }
6838 
6839 void md_unregister_thread(struct md_thread **threadp)
6840 {
6841 	struct md_thread *thread = *threadp;
6842 	if (!thread)
6843 		return;
6844 	pr_debug("interrupting MD-thread pid %d\n", task_pid_nr(thread->tsk));
6845 	/* Locking ensures that mddev_unlock does not wake_up a
6846 	 * non-existent thread
6847 	 */
6848 	spin_lock(&pers_lock);
6849 	*threadp = NULL;
6850 	spin_unlock(&pers_lock);
6851 
6852 	kthread_stop(thread->tsk);
6853 	kfree(thread);
6854 }
6855 
6856 void md_error(struct mddev *mddev, struct md_rdev *rdev)
6857 {
6858 	if (!mddev) {
6859 		MD_BUG();
6860 		return;
6861 	}
6862 
6863 	if (!rdev || test_bit(Faulty, &rdev->flags))
6864 		return;
6865 
6866 	if (!mddev->pers || !mddev->pers->error_handler)
6867 		return;
6868 	mddev->pers->error_handler(mddev,rdev);
6869 	if (mddev->degraded)
6870 		set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
6871 	sysfs_notify_dirent_safe(rdev->sysfs_state);
6872 	set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6873 	set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6874 	md_wakeup_thread(mddev->thread);
6875 	if (mddev->event_work.func)
6876 		queue_work(md_misc_wq, &mddev->event_work);
6877 	md_new_event_inintr(mddev);
6878 }
6879 
6880 /* seq_file implementation /proc/mdstat */
6881 
6882 static void status_unused(struct seq_file *seq)
6883 {
6884 	int i = 0;
6885 	struct md_rdev *rdev;
6886 
6887 	seq_printf(seq, "unused devices: ");
6888 
6889 	list_for_each_entry(rdev, &pending_raid_disks, same_set) {
6890 		char b[BDEVNAME_SIZE];
6891 		i++;
6892 		seq_printf(seq, "%s ",
6893 			      bdevname(rdev->bdev,b));
6894 	}
6895 	if (!i)
6896 		seq_printf(seq, "<none>");
6897 
6898 	seq_printf(seq, "\n");
6899 }
6900 
6901 
6902 static void status_resync(struct seq_file *seq, struct mddev * mddev)
6903 {
6904 	sector_t max_sectors, resync, res;
6905 	unsigned long dt, db;
6906 	sector_t rt;
6907 	int scale;
6908 	unsigned int per_milli;
6909 
6910 	if (mddev->curr_resync <= 3)
6911 		resync = 0;
6912 	else
6913 		resync = mddev->curr_resync
6914 			- atomic_read(&mddev->recovery_active);
6915 
6916 	if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ||
6917 	    test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
6918 		max_sectors = mddev->resync_max_sectors;
6919 	else
6920 		max_sectors = mddev->dev_sectors;
6921 
6922 	/*
6923 	 * Should not happen.
6924 	 */
6925 	if (!max_sectors) {
6926 		MD_BUG();
6927 		return;
6928 	}
6929 	/* Pick 'scale' such that (resync>>scale)*1000 will fit
6930 	 * in a sector_t, and (max_sectors>>scale) will fit in a
6931 	 * u32, as those are the requirements for sector_div.
6932 	 * Thus 'scale' must be at least 10
6933 	 */
6934 	scale = 10;
6935 	if (sizeof(sector_t) > sizeof(unsigned long)) {
6936 		while ( max_sectors/2 > (1ULL<<(scale+32)))
6937 			scale++;
6938 	}
6939 	res = (resync>>scale)*1000;
6940 	sector_div(res, (u32)((max_sectors>>scale)+1));
6941 
6942 	per_milli = res;
6943 	{
6944 		int i, x = per_milli/50, y = 20-x;
6945 		seq_printf(seq, "[");
6946 		for (i = 0; i < x; i++)
6947 			seq_printf(seq, "=");
6948 		seq_printf(seq, ">");
6949 		for (i = 0; i < y; i++)
6950 			seq_printf(seq, ".");
6951 		seq_printf(seq, "] ");
6952 	}
6953 	seq_printf(seq, " %s =%3u.%u%% (%llu/%llu)",
6954 		   (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)?
6955 		    "reshape" :
6956 		    (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)?
6957 		     "check" :
6958 		     (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
6959 		      "resync" : "recovery"))),
6960 		   per_milli/10, per_milli % 10,
6961 		   (unsigned long long) resync/2,
6962 		   (unsigned long long) max_sectors/2);
6963 
6964 	/*
6965 	 * dt: time from mark until now
6966 	 * db: blocks written from mark until now
6967 	 * rt: remaining time
6968 	 *
6969 	 * rt is a sector_t, so could be 32bit or 64bit.
6970 	 * So we divide before multiply in case it is 32bit and close
6971 	 * to the limit.
6972 	 * We scale the divisor (db) by 32 to avoid losing precision
6973 	 * near the end of resync when the number of remaining sectors
6974 	 * is close to 'db'.
6975 	 * We then divide rt by 32 after multiplying by db to compensate.
6976 	 * The '+1' avoids division by zero if db is very small.
6977 	 */
6978 	dt = ((jiffies - mddev->resync_mark) / HZ);
6979 	if (!dt) dt++;
6980 	db = (mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active))
6981 		- mddev->resync_mark_cnt;
6982 
6983 	rt = max_sectors - resync;    /* number of remaining sectors */
6984 	sector_div(rt, db/32+1);
6985 	rt *= dt;
6986 	rt >>= 5;
6987 
6988 	seq_printf(seq, " finish=%lu.%lumin", (unsigned long)rt / 60,
6989 		   ((unsigned long)rt % 60)/6);
6990 
6991 	seq_printf(seq, " speed=%ldK/sec", db/2/dt);
6992 }
6993 
6994 static void *md_seq_start(struct seq_file *seq, loff_t *pos)
6995 {
6996 	struct list_head *tmp;
6997 	loff_t l = *pos;
6998 	struct mddev *mddev;
6999 
7000 	if (l >= 0x10000)
7001 		return NULL;
7002 	if (!l--)
7003 		/* header */
7004 		return (void*)1;
7005 
7006 	spin_lock(&all_mddevs_lock);
7007 	list_for_each(tmp,&all_mddevs)
7008 		if (!l--) {
7009 			mddev = list_entry(tmp, struct mddev, all_mddevs);
7010 			mddev_get(mddev);
7011 			spin_unlock(&all_mddevs_lock);
7012 			return mddev;
7013 		}
7014 	spin_unlock(&all_mddevs_lock);
7015 	if (!l--)
7016 		return (void*)2;/* tail */
7017 	return NULL;
7018 }
7019 
7020 static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
7021 {
7022 	struct list_head *tmp;
7023 	struct mddev *next_mddev, *mddev = v;
7024 
7025 	++*pos;
7026 	if (v == (void*)2)
7027 		return NULL;
7028 
7029 	spin_lock(&all_mddevs_lock);
7030 	if (v == (void*)1)
7031 		tmp = all_mddevs.next;
7032 	else
7033 		tmp = mddev->all_mddevs.next;
7034 	if (tmp != &all_mddevs)
7035 		next_mddev = mddev_get(list_entry(tmp,struct mddev,all_mddevs));
7036 	else {
7037 		next_mddev = (void*)2;
7038 		*pos = 0x10000;
7039 	}
7040 	spin_unlock(&all_mddevs_lock);
7041 
7042 	if (v != (void*)1)
7043 		mddev_put(mddev);
7044 	return next_mddev;
7045 
7046 }
7047 
7048 static void md_seq_stop(struct seq_file *seq, void *v)
7049 {
7050 	struct mddev *mddev = v;
7051 
7052 	if (mddev && v != (void*)1 && v != (void*)2)
7053 		mddev_put(mddev);
7054 }
7055 
7056 static int md_seq_show(struct seq_file *seq, void *v)
7057 {
7058 	struct mddev *mddev = v;
7059 	sector_t sectors;
7060 	struct md_rdev *rdev;
7061 
7062 	if (v == (void*)1) {
7063 		struct md_personality *pers;
7064 		seq_printf(seq, "Personalities : ");
7065 		spin_lock(&pers_lock);
7066 		list_for_each_entry(pers, &pers_list, list)
7067 			seq_printf(seq, "[%s] ", pers->name);
7068 
7069 		spin_unlock(&pers_lock);
7070 		seq_printf(seq, "\n");
7071 		seq->poll_event = atomic_read(&md_event_count);
7072 		return 0;
7073 	}
7074 	if (v == (void*)2) {
7075 		status_unused(seq);
7076 		return 0;
7077 	}
7078 
7079 	if (mddev_lock(mddev) < 0)
7080 		return -EINTR;
7081 
7082 	if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
7083 		seq_printf(seq, "%s : %sactive", mdname(mddev),
7084 						mddev->pers ? "" : "in");
7085 		if (mddev->pers) {
7086 			if (mddev->ro==1)
7087 				seq_printf(seq, " (read-only)");
7088 			if (mddev->ro==2)
7089 				seq_printf(seq, " (auto-read-only)");
7090 			seq_printf(seq, " %s", mddev->pers->name);
7091 		}
7092 
7093 		sectors = 0;
7094 		rdev_for_each(rdev, mddev) {
7095 			char b[BDEVNAME_SIZE];
7096 			seq_printf(seq, " %s[%d]",
7097 				bdevname(rdev->bdev,b), rdev->desc_nr);
7098 			if (test_bit(WriteMostly, &rdev->flags))
7099 				seq_printf(seq, "(W)");
7100 			if (test_bit(Faulty, &rdev->flags)) {
7101 				seq_printf(seq, "(F)");
7102 				continue;
7103 			}
7104 			if (rdev->raid_disk < 0)
7105 				seq_printf(seq, "(S)"); /* spare */
7106 			if (test_bit(Replacement, &rdev->flags))
7107 				seq_printf(seq, "(R)");
7108 			sectors += rdev->sectors;
7109 		}
7110 
7111 		if (!list_empty(&mddev->disks)) {
7112 			if (mddev->pers)
7113 				seq_printf(seq, "\n      %llu blocks",
7114 					   (unsigned long long)
7115 					   mddev->array_sectors / 2);
7116 			else
7117 				seq_printf(seq, "\n      %llu blocks",
7118 					   (unsigned long long)sectors / 2);
7119 		}
7120 		if (mddev->persistent) {
7121 			if (mddev->major_version != 0 ||
7122 			    mddev->minor_version != 90) {
7123 				seq_printf(seq," super %d.%d",
7124 					   mddev->major_version,
7125 					   mddev->minor_version);
7126 			}
7127 		} else if (mddev->external)
7128 			seq_printf(seq, " super external:%s",
7129 				   mddev->metadata_type);
7130 		else
7131 			seq_printf(seq, " super non-persistent");
7132 
7133 		if (mddev->pers) {
7134 			mddev->pers->status(seq, mddev);
7135 	 		seq_printf(seq, "\n      ");
7136 			if (mddev->pers->sync_request) {
7137 				if (mddev->curr_resync > 2) {
7138 					status_resync(seq, mddev);
7139 					seq_printf(seq, "\n      ");
7140 				} else if (mddev->curr_resync >= 1)
7141 					seq_printf(seq, "\tresync=DELAYED\n      ");
7142 				else if (mddev->recovery_cp < MaxSector)
7143 					seq_printf(seq, "\tresync=PENDING\n      ");
7144 			}
7145 		} else
7146 			seq_printf(seq, "\n       ");
7147 
7148 		bitmap_status(seq, mddev->bitmap);
7149 
7150 		seq_printf(seq, "\n");
7151 	}
7152 	mddev_unlock(mddev);
7153 
7154 	return 0;
7155 }
7156 
7157 static const struct seq_operations md_seq_ops = {
7158 	.start  = md_seq_start,
7159 	.next   = md_seq_next,
7160 	.stop   = md_seq_stop,
7161 	.show   = md_seq_show,
7162 };
7163 
7164 static int md_seq_open(struct inode *inode, struct file *file)
7165 {
7166 	struct seq_file *seq;
7167 	int error;
7168 
7169 	error = seq_open(file, &md_seq_ops);
7170 	if (error)
7171 		return error;
7172 
7173 	seq = file->private_data;
7174 	seq->poll_event = atomic_read(&md_event_count);
7175 	return error;
7176 }
7177 
7178 static int md_unloading;
7179 static unsigned int mdstat_poll(struct file *filp, poll_table *wait)
7180 {
7181 	struct seq_file *seq = filp->private_data;
7182 	int mask;
7183 
7184 	if (md_unloading)
7185 		return POLLIN|POLLRDNORM|POLLERR|POLLPRI;;
7186 	poll_wait(filp, &md_event_waiters, wait);
7187 
7188 	/* always allow read */
7189 	mask = POLLIN | POLLRDNORM;
7190 
7191 	if (seq->poll_event != atomic_read(&md_event_count))
7192 		mask |= POLLERR | POLLPRI;
7193 	return mask;
7194 }
7195 
7196 static const struct file_operations md_seq_fops = {
7197 	.owner		= THIS_MODULE,
7198 	.open           = md_seq_open,
7199 	.read           = seq_read,
7200 	.llseek         = seq_lseek,
7201 	.release	= seq_release_private,
7202 	.poll		= mdstat_poll,
7203 };
7204 
7205 int register_md_personality(struct md_personality *p)
7206 {
7207 	spin_lock(&pers_lock);
7208 	list_add_tail(&p->list, &pers_list);
7209 	printk(KERN_INFO "md: %s personality registered for level %d\n", p->name, p->level);
7210 	spin_unlock(&pers_lock);
7211 	return 0;
7212 }
7213 
7214 int unregister_md_personality(struct md_personality *p)
7215 {
7216 	printk(KERN_INFO "md: %s personality unregistered\n", p->name);
7217 	spin_lock(&pers_lock);
7218 	list_del_init(&p->list);
7219 	spin_unlock(&pers_lock);
7220 	return 0;
7221 }
7222 
7223 static int is_mddev_idle(struct mddev *mddev, int init)
7224 {
7225 	struct md_rdev * rdev;
7226 	int idle;
7227 	int curr_events;
7228 
7229 	idle = 1;
7230 	rcu_read_lock();
7231 	rdev_for_each_rcu(rdev, mddev) {
7232 		struct gendisk *disk = rdev->bdev->bd_contains->bd_disk;
7233 		curr_events = (int)part_stat_read(&disk->part0, sectors[0]) +
7234 			      (int)part_stat_read(&disk->part0, sectors[1]) -
7235 			      atomic_read(&disk->sync_io);
7236 		/* sync IO will cause sync_io to increase before the disk_stats
7237 		 * as sync_io is counted when a request starts, and
7238 		 * disk_stats is counted when it completes.
7239 		 * So resync activity will cause curr_events to be smaller than
7240 		 * when there was no such activity.
7241 		 * non-sync IO will cause disk_stat to increase without
7242 		 * increasing sync_io so curr_events will (eventually)
7243 		 * be larger than it was before.  Once it becomes
7244 		 * substantially larger, the test below will cause
7245 		 * the array to appear non-idle, and resync will slow
7246 		 * down.
7247 		 * If there is a lot of outstanding resync activity when
7248 		 * we set last_event to curr_events, then all that activity
7249 		 * completing might cause the array to appear non-idle
7250 		 * and resync will be slowed down even though there might
7251 		 * not have been non-resync activity.  This will only
7252 		 * happen once though.  'last_events' will soon reflect
7253 		 * the state where there is little or no outstanding
7254 		 * resync requests, and further resync activity will
7255 		 * always make curr_events less than last_events.
7256 		 *
7257 		 */
7258 		if (init || curr_events - rdev->last_events > 64) {
7259 			rdev->last_events = curr_events;
7260 			idle = 0;
7261 		}
7262 	}
7263 	rcu_read_unlock();
7264 	return idle;
7265 }
7266 
7267 void md_done_sync(struct mddev *mddev, int blocks, int ok)
7268 {
7269 	/* another "blocks" (512byte) blocks have been synced */
7270 	atomic_sub(blocks, &mddev->recovery_active);
7271 	wake_up(&mddev->recovery_wait);
7272 	if (!ok) {
7273 		set_bit(MD_RECOVERY_INTR, &mddev->recovery);
7274 		set_bit(MD_RECOVERY_ERROR, &mddev->recovery);
7275 		md_wakeup_thread(mddev->thread);
7276 		// stop recovery, signal do_sync ....
7277 	}
7278 }
7279 
7280 
7281 /* md_write_start(mddev, bi)
7282  * If we need to update some array metadata (e.g. 'active' flag
7283  * in superblock) before writing, schedule a superblock update
7284  * and wait for it to complete.
7285  */
7286 void md_write_start(struct mddev *mddev, struct bio *bi)
7287 {
7288 	int did_change = 0;
7289 	if (bio_data_dir(bi) != WRITE)
7290 		return;
7291 
7292 	BUG_ON(mddev->ro == 1);
7293 	if (mddev->ro == 2) {
7294 		/* need to switch to read/write */
7295 		mddev->ro = 0;
7296 		set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7297 		md_wakeup_thread(mddev->thread);
7298 		md_wakeup_thread(mddev->sync_thread);
7299 		did_change = 1;
7300 	}
7301 	atomic_inc(&mddev->writes_pending);
7302 	if (mddev->safemode == 1)
7303 		mddev->safemode = 0;
7304 	if (mddev->in_sync) {
7305 		spin_lock_irq(&mddev->write_lock);
7306 		if (mddev->in_sync) {
7307 			mddev->in_sync = 0;
7308 			set_bit(MD_CHANGE_CLEAN, &mddev->flags);
7309 			set_bit(MD_CHANGE_PENDING, &mddev->flags);
7310 			md_wakeup_thread(mddev->thread);
7311 			did_change = 1;
7312 		}
7313 		spin_unlock_irq(&mddev->write_lock);
7314 	}
7315 	if (did_change)
7316 		sysfs_notify_dirent_safe(mddev->sysfs_state);
7317 	wait_event(mddev->sb_wait,
7318 		   !test_bit(MD_CHANGE_PENDING, &mddev->flags));
7319 }
7320 
7321 void md_write_end(struct mddev *mddev)
7322 {
7323 	if (atomic_dec_and_test(&mddev->writes_pending)) {
7324 		if (mddev->safemode == 2)
7325 			md_wakeup_thread(mddev->thread);
7326 		else if (mddev->safemode_delay)
7327 			mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay);
7328 	}
7329 }
7330 
7331 /* md_allow_write(mddev)
7332  * Calling this ensures that the array is marked 'active' so that writes
7333  * may proceed without blocking.  It is important to call this before
7334  * attempting a GFP_KERNEL allocation while holding the mddev lock.
7335  * Must be called with mddev_lock held.
7336  *
7337  * In the ->external case MD_CHANGE_CLEAN can not be cleared until mddev->lock
7338  * is dropped, so return -EAGAIN after notifying userspace.
7339  */
7340 int md_allow_write(struct mddev *mddev)
7341 {
7342 	if (!mddev->pers)
7343 		return 0;
7344 	if (mddev->ro)
7345 		return 0;
7346 	if (!mddev->pers->sync_request)
7347 		return 0;
7348 
7349 	spin_lock_irq(&mddev->write_lock);
7350 	if (mddev->in_sync) {
7351 		mddev->in_sync = 0;
7352 		set_bit(MD_CHANGE_CLEAN, &mddev->flags);
7353 		set_bit(MD_CHANGE_PENDING, &mddev->flags);
7354 		if (mddev->safemode_delay &&
7355 		    mddev->safemode == 0)
7356 			mddev->safemode = 1;
7357 		spin_unlock_irq(&mddev->write_lock);
7358 		md_update_sb(mddev, 0);
7359 		sysfs_notify_dirent_safe(mddev->sysfs_state);
7360 	} else
7361 		spin_unlock_irq(&mddev->write_lock);
7362 
7363 	if (test_bit(MD_CHANGE_PENDING, &mddev->flags))
7364 		return -EAGAIN;
7365 	else
7366 		return 0;
7367 }
7368 EXPORT_SYMBOL_GPL(md_allow_write);
7369 
7370 #define SYNC_MARKS	10
7371 #define	SYNC_MARK_STEP	(3*HZ)
7372 #define UPDATE_FREQUENCY (5*60*HZ)
7373 void md_do_sync(struct md_thread *thread)
7374 {
7375 	struct mddev *mddev = thread->mddev;
7376 	struct mddev *mddev2;
7377 	unsigned int currspeed = 0,
7378 		 window;
7379 	sector_t max_sectors,j, io_sectors;
7380 	unsigned long mark[SYNC_MARKS];
7381 	unsigned long update_time;
7382 	sector_t mark_cnt[SYNC_MARKS];
7383 	int last_mark,m;
7384 	struct list_head *tmp;
7385 	sector_t last_check;
7386 	int skipped = 0;
7387 	struct md_rdev *rdev;
7388 	char *desc, *action = NULL;
7389 	struct blk_plug plug;
7390 
7391 	/* just incase thread restarts... */
7392 	if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
7393 		return;
7394 	if (mddev->ro) {/* never try to sync a read-only array */
7395 		set_bit(MD_RECOVERY_INTR, &mddev->recovery);
7396 		return;
7397 	}
7398 
7399 	if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
7400 		if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)) {
7401 			desc = "data-check";
7402 			action = "check";
7403 		} else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
7404 			desc = "requested-resync";
7405 			action = "repair";
7406 		} else
7407 			desc = "resync";
7408 	} else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
7409 		desc = "reshape";
7410 	else
7411 		desc = "recovery";
7412 
7413 	mddev->last_sync_action = action ?: desc;
7414 
7415 	/* we overload curr_resync somewhat here.
7416 	 * 0 == not engaged in resync at all
7417 	 * 2 == checking that there is no conflict with another sync
7418 	 * 1 == like 2, but have yielded to allow conflicting resync to
7419 	 *		commense
7420 	 * other == active in resync - this many blocks
7421 	 *
7422 	 * Before starting a resync we must have set curr_resync to
7423 	 * 2, and then checked that every "conflicting" array has curr_resync
7424 	 * less than ours.  When we find one that is the same or higher
7425 	 * we wait on resync_wait.  To avoid deadlock, we reduce curr_resync
7426 	 * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
7427 	 * This will mean we have to start checking from the beginning again.
7428 	 *
7429 	 */
7430 
7431 	do {
7432 		mddev->curr_resync = 2;
7433 
7434 	try_again:
7435 		if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
7436 			goto skip;
7437 		for_each_mddev(mddev2, tmp) {
7438 			if (mddev2 == mddev)
7439 				continue;
7440 			if (!mddev->parallel_resync
7441 			&&  mddev2->curr_resync
7442 			&&  match_mddev_units(mddev, mddev2)) {
7443 				DEFINE_WAIT(wq);
7444 				if (mddev < mddev2 && mddev->curr_resync == 2) {
7445 					/* arbitrarily yield */
7446 					mddev->curr_resync = 1;
7447 					wake_up(&resync_wait);
7448 				}
7449 				if (mddev > mddev2 && mddev->curr_resync == 1)
7450 					/* no need to wait here, we can wait the next
7451 					 * time 'round when curr_resync == 2
7452 					 */
7453 					continue;
7454 				/* We need to wait 'interruptible' so as not to
7455 				 * contribute to the load average, and not to
7456 				 * be caught by 'softlockup'
7457 				 */
7458 				prepare_to_wait(&resync_wait, &wq, TASK_INTERRUPTIBLE);
7459 				if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
7460 				    mddev2->curr_resync >= mddev->curr_resync) {
7461 					printk(KERN_INFO "md: delaying %s of %s"
7462 					       " until %s has finished (they"
7463 					       " share one or more physical units)\n",
7464 					       desc, mdname(mddev), mdname(mddev2));
7465 					mddev_put(mddev2);
7466 					if (signal_pending(current))
7467 						flush_signals(current);
7468 					schedule();
7469 					finish_wait(&resync_wait, &wq);
7470 					goto try_again;
7471 				}
7472 				finish_wait(&resync_wait, &wq);
7473 			}
7474 		}
7475 	} while (mddev->curr_resync < 2);
7476 
7477 	j = 0;
7478 	if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
7479 		/* resync follows the size requested by the personality,
7480 		 * which defaults to physical size, but can be virtual size
7481 		 */
7482 		max_sectors = mddev->resync_max_sectors;
7483 		atomic64_set(&mddev->resync_mismatches, 0);
7484 		/* we don't use the checkpoint if there's a bitmap */
7485 		if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
7486 			j = mddev->resync_min;
7487 		else if (!mddev->bitmap)
7488 			j = mddev->recovery_cp;
7489 
7490 	} else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
7491 		max_sectors = mddev->resync_max_sectors;
7492 	else {
7493 		/* recovery follows the physical size of devices */
7494 		max_sectors = mddev->dev_sectors;
7495 		j = MaxSector;
7496 		rcu_read_lock();
7497 		rdev_for_each_rcu(rdev, mddev)
7498 			if (rdev->raid_disk >= 0 &&
7499 			    !test_bit(Faulty, &rdev->flags) &&
7500 			    !test_bit(In_sync, &rdev->flags) &&
7501 			    rdev->recovery_offset < j)
7502 				j = rdev->recovery_offset;
7503 		rcu_read_unlock();
7504 	}
7505 
7506 	printk(KERN_INFO "md: %s of RAID array %s\n", desc, mdname(mddev));
7507 	printk(KERN_INFO "md: minimum _guaranteed_  speed:"
7508 		" %d KB/sec/disk.\n", speed_min(mddev));
7509 	printk(KERN_INFO "md: using maximum available idle IO bandwidth "
7510 	       "(but not more than %d KB/sec) for %s.\n",
7511 	       speed_max(mddev), desc);
7512 
7513 	is_mddev_idle(mddev, 1); /* this initializes IO event counters */
7514 
7515 	io_sectors = 0;
7516 	for (m = 0; m < SYNC_MARKS; m++) {
7517 		mark[m] = jiffies;
7518 		mark_cnt[m] = io_sectors;
7519 	}
7520 	last_mark = 0;
7521 	mddev->resync_mark = mark[last_mark];
7522 	mddev->resync_mark_cnt = mark_cnt[last_mark];
7523 
7524 	/*
7525 	 * Tune reconstruction:
7526 	 */
7527 	window = 32*(PAGE_SIZE/512);
7528 	printk(KERN_INFO "md: using %dk window, over a total of %lluk.\n",
7529 		window/2, (unsigned long long)max_sectors/2);
7530 
7531 	atomic_set(&mddev->recovery_active, 0);
7532 	last_check = 0;
7533 
7534 	if (j>2) {
7535 		printk(KERN_INFO
7536 		       "md: resuming %s of %s from checkpoint.\n",
7537 		       desc, mdname(mddev));
7538 		mddev->curr_resync = j;
7539 	} else
7540 		mddev->curr_resync = 3; /* no longer delayed */
7541 	mddev->curr_resync_completed = j;
7542 	sysfs_notify(&mddev->kobj, NULL, "sync_completed");
7543 	md_new_event(mddev);
7544 	update_time = jiffies;
7545 
7546 	blk_start_plug(&plug);
7547 	while (j < max_sectors) {
7548 		sector_t sectors;
7549 
7550 		skipped = 0;
7551 
7552 		if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
7553 		    ((mddev->curr_resync > mddev->curr_resync_completed &&
7554 		      (mddev->curr_resync - mddev->curr_resync_completed)
7555 		      > (max_sectors >> 4)) ||
7556 		     time_after_eq(jiffies, update_time + UPDATE_FREQUENCY) ||
7557 		     (j - mddev->curr_resync_completed)*2
7558 		     >= mddev->resync_max - mddev->curr_resync_completed
7559 			    )) {
7560 			/* time to update curr_resync_completed */
7561 			wait_event(mddev->recovery_wait,
7562 				   atomic_read(&mddev->recovery_active) == 0);
7563 			mddev->curr_resync_completed = j;
7564 			if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) &&
7565 			    j > mddev->recovery_cp)
7566 				mddev->recovery_cp = j;
7567 			update_time = jiffies;
7568 			set_bit(MD_CHANGE_CLEAN, &mddev->flags);
7569 			sysfs_notify(&mddev->kobj, NULL, "sync_completed");
7570 		}
7571 
7572 		while (j >= mddev->resync_max &&
7573 		       !test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
7574 			/* As this condition is controlled by user-space,
7575 			 * we can block indefinitely, so use '_interruptible'
7576 			 * to avoid triggering warnings.
7577 			 */
7578 			flush_signals(current); /* just in case */
7579 			wait_event_interruptible(mddev->recovery_wait,
7580 						 mddev->resync_max > j
7581 						 || test_bit(MD_RECOVERY_INTR,
7582 							     &mddev->recovery));
7583 		}
7584 
7585 		if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
7586 			break;
7587 
7588 		sectors = mddev->pers->sync_request(mddev, j, &skipped,
7589 						  currspeed < speed_min(mddev));
7590 		if (sectors == 0) {
7591 			set_bit(MD_RECOVERY_INTR, &mddev->recovery);
7592 			break;
7593 		}
7594 
7595 		if (!skipped) { /* actual IO requested */
7596 			io_sectors += sectors;
7597 			atomic_add(sectors, &mddev->recovery_active);
7598 		}
7599 
7600 		if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
7601 			break;
7602 
7603 		j += sectors;
7604 		if (j > 2)
7605 			mddev->curr_resync = j;
7606 		mddev->curr_mark_cnt = io_sectors;
7607 		if (last_check == 0)
7608 			/* this is the earliest that rebuild will be
7609 			 * visible in /proc/mdstat
7610 			 */
7611 			md_new_event(mddev);
7612 
7613 		if (last_check + window > io_sectors || j == max_sectors)
7614 			continue;
7615 
7616 		last_check = io_sectors;
7617 	repeat:
7618 		if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) {
7619 			/* step marks */
7620 			int next = (last_mark+1) % SYNC_MARKS;
7621 
7622 			mddev->resync_mark = mark[next];
7623 			mddev->resync_mark_cnt = mark_cnt[next];
7624 			mark[next] = jiffies;
7625 			mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active);
7626 			last_mark = next;
7627 		}
7628 
7629 		if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
7630 			break;
7631 
7632 		/*
7633 		 * this loop exits only if either when we are slower than
7634 		 * the 'hard' speed limit, or the system was IO-idle for
7635 		 * a jiffy.
7636 		 * the system might be non-idle CPU-wise, but we only care
7637 		 * about not overloading the IO subsystem. (things like an
7638 		 * e2fsck being done on the RAID array should execute fast)
7639 		 */
7640 		cond_resched();
7641 
7642 		currspeed = ((unsigned long)(io_sectors-mddev->resync_mark_cnt))/2
7643 			/((jiffies-mddev->resync_mark)/HZ +1) +1;
7644 
7645 		if (currspeed > speed_min(mddev)) {
7646 			if ((currspeed > speed_max(mddev)) ||
7647 					!is_mddev_idle(mddev, 0)) {
7648 				msleep(500);
7649 				goto repeat;
7650 			}
7651 		}
7652 	}
7653 	printk(KERN_INFO "md: %s: %s %s.\n",mdname(mddev), desc,
7654 	       test_bit(MD_RECOVERY_INTR, &mddev->recovery)
7655 	       ? "interrupted" : "done");
7656 	/*
7657 	 * this also signals 'finished resyncing' to md_stop
7658 	 */
7659 	blk_finish_plug(&plug);
7660 	wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
7661 
7662 	/* tell personality that we are finished */
7663 	mddev->pers->sync_request(mddev, max_sectors, &skipped, 1);
7664 
7665 	if (!test_bit(MD_RECOVERY_CHECK, &mddev->recovery) &&
7666 	    mddev->curr_resync > 2) {
7667 		if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
7668 			if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
7669 				if (mddev->curr_resync >= mddev->recovery_cp) {
7670 					printk(KERN_INFO
7671 					       "md: checkpointing %s of %s.\n",
7672 					       desc, mdname(mddev));
7673 					if (test_bit(MD_RECOVERY_ERROR,
7674 						&mddev->recovery))
7675 						mddev->recovery_cp =
7676 							mddev->curr_resync_completed;
7677 					else
7678 						mddev->recovery_cp =
7679 							mddev->curr_resync;
7680 				}
7681 			} else
7682 				mddev->recovery_cp = MaxSector;
7683 		} else {
7684 			if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery))
7685 				mddev->curr_resync = MaxSector;
7686 			rcu_read_lock();
7687 			rdev_for_each_rcu(rdev, mddev)
7688 				if (rdev->raid_disk >= 0 &&
7689 				    mddev->delta_disks >= 0 &&
7690 				    !test_bit(Faulty, &rdev->flags) &&
7691 				    !test_bit(In_sync, &rdev->flags) &&
7692 				    rdev->recovery_offset < mddev->curr_resync)
7693 					rdev->recovery_offset = mddev->curr_resync;
7694 			rcu_read_unlock();
7695 		}
7696 	}
7697  skip:
7698 	set_bit(MD_CHANGE_DEVS, &mddev->flags);
7699 
7700 	if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
7701 		/* We completed so min/max setting can be forgotten if used. */
7702 		if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
7703 			mddev->resync_min = 0;
7704 		mddev->resync_max = MaxSector;
7705 	} else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
7706 		mddev->resync_min = mddev->curr_resync_completed;
7707 	mddev->curr_resync = 0;
7708 	wake_up(&resync_wait);
7709 	set_bit(MD_RECOVERY_DONE, &mddev->recovery);
7710 	md_wakeup_thread(mddev->thread);
7711 	return;
7712 }
7713 EXPORT_SYMBOL_GPL(md_do_sync);
7714 
7715 static int remove_and_add_spares(struct mddev *mddev,
7716 				 struct md_rdev *this)
7717 {
7718 	struct md_rdev *rdev;
7719 	int spares = 0;
7720 	int removed = 0;
7721 
7722 	rdev_for_each(rdev, mddev)
7723 		if ((this == NULL || rdev == this) &&
7724 		    rdev->raid_disk >= 0 &&
7725 		    !test_bit(Blocked, &rdev->flags) &&
7726 		    (test_bit(Faulty, &rdev->flags) ||
7727 		     ! test_bit(In_sync, &rdev->flags)) &&
7728 		    atomic_read(&rdev->nr_pending)==0) {
7729 			if (mddev->pers->hot_remove_disk(
7730 				    mddev, rdev) == 0) {
7731 				sysfs_unlink_rdev(mddev, rdev);
7732 				rdev->raid_disk = -1;
7733 				removed++;
7734 			}
7735 		}
7736 	if (removed && mddev->kobj.sd)
7737 		sysfs_notify(&mddev->kobj, NULL, "degraded");
7738 
7739 	if (this)
7740 		goto no_add;
7741 
7742 	rdev_for_each(rdev, mddev) {
7743 		if (rdev->raid_disk >= 0 &&
7744 		    !test_bit(In_sync, &rdev->flags) &&
7745 		    !test_bit(Faulty, &rdev->flags))
7746 			spares++;
7747 		if (rdev->raid_disk >= 0)
7748 			continue;
7749 		if (test_bit(Faulty, &rdev->flags))
7750 			continue;
7751 		if (mddev->ro &&
7752 		    ! (rdev->saved_raid_disk >= 0 &&
7753 		       !test_bit(Bitmap_sync, &rdev->flags)))
7754 			continue;
7755 
7756 		if (rdev->saved_raid_disk < 0)
7757 			rdev->recovery_offset = 0;
7758 		if (mddev->pers->
7759 		    hot_add_disk(mddev, rdev) == 0) {
7760 			if (sysfs_link_rdev(mddev, rdev))
7761 				/* failure here is OK */;
7762 			spares++;
7763 			md_new_event(mddev);
7764 			set_bit(MD_CHANGE_DEVS, &mddev->flags);
7765 		}
7766 	}
7767 no_add:
7768 	if (removed)
7769 		set_bit(MD_CHANGE_DEVS, &mddev->flags);
7770 	return spares;
7771 }
7772 
7773 /*
7774  * This routine is regularly called by all per-raid-array threads to
7775  * deal with generic issues like resync and super-block update.
7776  * Raid personalities that don't have a thread (linear/raid0) do not
7777  * need this as they never do any recovery or update the superblock.
7778  *
7779  * It does not do any resync itself, but rather "forks" off other threads
7780  * to do that as needed.
7781  * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
7782  * "->recovery" and create a thread at ->sync_thread.
7783  * When the thread finishes it sets MD_RECOVERY_DONE
7784  * and wakeups up this thread which will reap the thread and finish up.
7785  * This thread also removes any faulty devices (with nr_pending == 0).
7786  *
7787  * The overall approach is:
7788  *  1/ if the superblock needs updating, update it.
7789  *  2/ If a recovery thread is running, don't do anything else.
7790  *  3/ If recovery has finished, clean up, possibly marking spares active.
7791  *  4/ If there are any faulty devices, remove them.
7792  *  5/ If array is degraded, try to add spares devices
7793  *  6/ If array has spares or is not in-sync, start a resync thread.
7794  */
7795 void md_check_recovery(struct mddev *mddev)
7796 {
7797 	if (mddev->suspended)
7798 		return;
7799 
7800 	if (mddev->bitmap)
7801 		bitmap_daemon_work(mddev);
7802 
7803 	if (signal_pending(current)) {
7804 		if (mddev->pers->sync_request && !mddev->external) {
7805 			printk(KERN_INFO "md: %s in immediate safe mode\n",
7806 			       mdname(mddev));
7807 			mddev->safemode = 2;
7808 		}
7809 		flush_signals(current);
7810 	}
7811 
7812 	if (mddev->ro && !test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
7813 		return;
7814 	if ( ! (
7815 		(mddev->flags & MD_UPDATE_SB_FLAGS & ~ (1<<MD_CHANGE_PENDING)) ||
7816 		test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
7817 		test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
7818 		(mddev->external == 0 && mddev->safemode == 1) ||
7819 		(mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending)
7820 		 && !mddev->in_sync && mddev->recovery_cp == MaxSector)
7821 		))
7822 		return;
7823 
7824 	if (mddev_trylock(mddev)) {
7825 		int spares = 0;
7826 
7827 		if (mddev->ro) {
7828 			/* On a read-only array we can:
7829 			 * - remove failed devices
7830 			 * - add already-in_sync devices if the array itself
7831 			 *   is in-sync.
7832 			 * As we only add devices that are already in-sync,
7833 			 * we can activate the spares immediately.
7834 			 */
7835 			remove_and_add_spares(mddev, NULL);
7836 			/* There is no thread, but we need to call
7837 			 * ->spare_active and clear saved_raid_disk
7838 			 */
7839 			set_bit(MD_RECOVERY_INTR, &mddev->recovery);
7840 			md_reap_sync_thread(mddev);
7841 			clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7842 			goto unlock;
7843 		}
7844 
7845 		if (!mddev->external) {
7846 			int did_change = 0;
7847 			spin_lock_irq(&mddev->write_lock);
7848 			if (mddev->safemode &&
7849 			    !atomic_read(&mddev->writes_pending) &&
7850 			    !mddev->in_sync &&
7851 			    mddev->recovery_cp == MaxSector) {
7852 				mddev->in_sync = 1;
7853 				did_change = 1;
7854 				set_bit(MD_CHANGE_CLEAN, &mddev->flags);
7855 			}
7856 			if (mddev->safemode == 1)
7857 				mddev->safemode = 0;
7858 			spin_unlock_irq(&mddev->write_lock);
7859 			if (did_change)
7860 				sysfs_notify_dirent_safe(mddev->sysfs_state);
7861 		}
7862 
7863 		if (mddev->flags & MD_UPDATE_SB_FLAGS)
7864 			md_update_sb(mddev, 0);
7865 
7866 		if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
7867 		    !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
7868 			/* resync/recovery still happening */
7869 			clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7870 			goto unlock;
7871 		}
7872 		if (mddev->sync_thread) {
7873 			md_reap_sync_thread(mddev);
7874 			goto unlock;
7875 		}
7876 		/* Set RUNNING before clearing NEEDED to avoid
7877 		 * any transients in the value of "sync_action".
7878 		 */
7879 		mddev->curr_resync_completed = 0;
7880 		set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
7881 		/* Clear some bits that don't mean anything, but
7882 		 * might be left set
7883 		 */
7884 		clear_bit(MD_RECOVERY_INTR, &mddev->recovery);
7885 		clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
7886 
7887 		if (!test_and_clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
7888 		    test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
7889 			goto unlock;
7890 		/* no recovery is running.
7891 		 * remove any failed drives, then
7892 		 * add spares if possible.
7893 		 * Spares are also removed and re-added, to allow
7894 		 * the personality to fail the re-add.
7895 		 */
7896 
7897 		if (mddev->reshape_position != MaxSector) {
7898 			if (mddev->pers->check_reshape == NULL ||
7899 			    mddev->pers->check_reshape(mddev) != 0)
7900 				/* Cannot proceed */
7901 				goto unlock;
7902 			set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
7903 			clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
7904 		} else if ((spares = remove_and_add_spares(mddev, NULL))) {
7905 			clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
7906 			clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
7907 			clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
7908 			set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
7909 		} else if (mddev->recovery_cp < MaxSector) {
7910 			set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
7911 			clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
7912 		} else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
7913 			/* nothing to be done ... */
7914 			goto unlock;
7915 
7916 		if (mddev->pers->sync_request) {
7917 			if (spares) {
7918 				/* We are adding a device or devices to an array
7919 				 * which has the bitmap stored on all devices.
7920 				 * So make sure all bitmap pages get written
7921 				 */
7922 				bitmap_write_all(mddev->bitmap);
7923 			}
7924 			mddev->sync_thread = md_register_thread(md_do_sync,
7925 								mddev,
7926 								"resync");
7927 			if (!mddev->sync_thread) {
7928 				printk(KERN_ERR "%s: could not start resync"
7929 					" thread...\n",
7930 					mdname(mddev));
7931 				/* leave the spares where they are, it shouldn't hurt */
7932 				clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
7933 				clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
7934 				clear_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
7935 				clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
7936 				clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
7937 			} else
7938 				md_wakeup_thread(mddev->sync_thread);
7939 			sysfs_notify_dirent_safe(mddev->sysfs_action);
7940 			md_new_event(mddev);
7941 		}
7942 	unlock:
7943 		wake_up(&mddev->sb_wait);
7944 
7945 		if (!mddev->sync_thread) {
7946 			clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
7947 			if (test_and_clear_bit(MD_RECOVERY_RECOVER,
7948 					       &mddev->recovery))
7949 				if (mddev->sysfs_action)
7950 					sysfs_notify_dirent_safe(mddev->sysfs_action);
7951 		}
7952 		mddev_unlock(mddev);
7953 	}
7954 }
7955 
7956 void md_reap_sync_thread(struct mddev *mddev)
7957 {
7958 	struct md_rdev *rdev;
7959 
7960 	/* resync has finished, collect result */
7961 	md_unregister_thread(&mddev->sync_thread);
7962 	wake_up(&resync_wait);
7963 	if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
7964 	    !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
7965 		/* success...*/
7966 		/* activate any spares */
7967 		if (mddev->pers->spare_active(mddev)) {
7968 			sysfs_notify(&mddev->kobj, NULL,
7969 				     "degraded");
7970 			set_bit(MD_CHANGE_DEVS, &mddev->flags);
7971 		}
7972 	}
7973 	if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
7974 	    mddev->pers->finish_reshape)
7975 		mddev->pers->finish_reshape(mddev);
7976 
7977 	/* If array is no-longer degraded, then any saved_raid_disk
7978 	 * information must be scrapped.
7979 	 */
7980 	if (!mddev->degraded)
7981 		rdev_for_each(rdev, mddev)
7982 			rdev->saved_raid_disk = -1;
7983 
7984 	md_update_sb(mddev, 1);
7985 	clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
7986 	clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
7987 	clear_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
7988 	clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
7989 	clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
7990 	/* flag recovery needed just to double check */
7991 	set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7992 	sysfs_notify_dirent_safe(mddev->sysfs_action);
7993 	md_new_event(mddev);
7994 	if (mddev->event_work.func)
7995 		queue_work(md_misc_wq, &mddev->event_work);
7996 }
7997 
7998 void md_wait_for_blocked_rdev(struct md_rdev *rdev, struct mddev *mddev)
7999 {
8000 	sysfs_notify_dirent_safe(rdev->sysfs_state);
8001 	wait_event_timeout(rdev->blocked_wait,
8002 			   !test_bit(Blocked, &rdev->flags) &&
8003 			   !test_bit(BlockedBadBlocks, &rdev->flags),
8004 			   msecs_to_jiffies(5000));
8005 	rdev_dec_pending(rdev, mddev);
8006 }
8007 EXPORT_SYMBOL(md_wait_for_blocked_rdev);
8008 
8009 void md_finish_reshape(struct mddev *mddev)
8010 {
8011 	/* called be personality module when reshape completes. */
8012 	struct md_rdev *rdev;
8013 
8014 	rdev_for_each(rdev, mddev) {
8015 		if (rdev->data_offset > rdev->new_data_offset)
8016 			rdev->sectors += rdev->data_offset - rdev->new_data_offset;
8017 		else
8018 			rdev->sectors -= rdev->new_data_offset - rdev->data_offset;
8019 		rdev->data_offset = rdev->new_data_offset;
8020 	}
8021 }
8022 EXPORT_SYMBOL(md_finish_reshape);
8023 
8024 /* Bad block management.
8025  * We can record which blocks on each device are 'bad' and so just
8026  * fail those blocks, or that stripe, rather than the whole device.
8027  * Entries in the bad-block table are 64bits wide.  This comprises:
8028  * Length of bad-range, in sectors: 0-511 for lengths 1-512
8029  * Start of bad-range, sector offset, 54 bits (allows 8 exbibytes)
8030  *  A 'shift' can be set so that larger blocks are tracked and
8031  *  consequently larger devices can be covered.
8032  * 'Acknowledged' flag - 1 bit. - the most significant bit.
8033  *
8034  * Locking of the bad-block table uses a seqlock so md_is_badblock
8035  * might need to retry if it is very unlucky.
8036  * We will sometimes want to check for bad blocks in a bi_end_io function,
8037  * so we use the write_seqlock_irq variant.
8038  *
8039  * When looking for a bad block we specify a range and want to
8040  * know if any block in the range is bad.  So we binary-search
8041  * to the last range that starts at-or-before the given endpoint,
8042  * (or "before the sector after the target range")
8043  * then see if it ends after the given start.
8044  * We return
8045  *  0 if there are no known bad blocks in the range
8046  *  1 if there are known bad block which are all acknowledged
8047  * -1 if there are bad blocks which have not yet been acknowledged in metadata.
8048  * plus the start/length of the first bad section we overlap.
8049  */
8050 int md_is_badblock(struct badblocks *bb, sector_t s, int sectors,
8051 		   sector_t *first_bad, int *bad_sectors)
8052 {
8053 	int hi;
8054 	int lo;
8055 	u64 *p = bb->page;
8056 	int rv;
8057 	sector_t target = s + sectors;
8058 	unsigned seq;
8059 
8060 	if (bb->shift > 0) {
8061 		/* round the start down, and the end up */
8062 		s >>= bb->shift;
8063 		target += (1<<bb->shift) - 1;
8064 		target >>= bb->shift;
8065 		sectors = target - s;
8066 	}
8067 	/* 'target' is now the first block after the bad range */
8068 
8069 retry:
8070 	seq = read_seqbegin(&bb->lock);
8071 	lo = 0;
8072 	rv = 0;
8073 	hi = bb->count;
8074 
8075 	/* Binary search between lo and hi for 'target'
8076 	 * i.e. for the last range that starts before 'target'
8077 	 */
8078 	/* INVARIANT: ranges before 'lo' and at-or-after 'hi'
8079 	 * are known not to be the last range before target.
8080 	 * VARIANT: hi-lo is the number of possible
8081 	 * ranges, and decreases until it reaches 1
8082 	 */
8083 	while (hi - lo > 1) {
8084 		int mid = (lo + hi) / 2;
8085 		sector_t a = BB_OFFSET(p[mid]);
8086 		if (a < target)
8087 			/* This could still be the one, earlier ranges
8088 			 * could not. */
8089 			lo = mid;
8090 		else
8091 			/* This and later ranges are definitely out. */
8092 			hi = mid;
8093 	}
8094 	/* 'lo' might be the last that started before target, but 'hi' isn't */
8095 	if (hi > lo) {
8096 		/* need to check all range that end after 's' to see if
8097 		 * any are unacknowledged.
8098 		 */
8099 		while (lo >= 0 &&
8100 		       BB_OFFSET(p[lo]) + BB_LEN(p[lo]) > s) {
8101 			if (BB_OFFSET(p[lo]) < target) {
8102 				/* starts before the end, and finishes after
8103 				 * the start, so they must overlap
8104 				 */
8105 				if (rv != -1 && BB_ACK(p[lo]))
8106 					rv = 1;
8107 				else
8108 					rv = -1;
8109 				*first_bad = BB_OFFSET(p[lo]);
8110 				*bad_sectors = BB_LEN(p[lo]);
8111 			}
8112 			lo--;
8113 		}
8114 	}
8115 
8116 	if (read_seqretry(&bb->lock, seq))
8117 		goto retry;
8118 
8119 	return rv;
8120 }
8121 EXPORT_SYMBOL_GPL(md_is_badblock);
8122 
8123 /*
8124  * Add a range of bad blocks to the table.
8125  * This might extend the table, or might contract it
8126  * if two adjacent ranges can be merged.
8127  * We binary-search to find the 'insertion' point, then
8128  * decide how best to handle it.
8129  */
8130 static int md_set_badblocks(struct badblocks *bb, sector_t s, int sectors,
8131 			    int acknowledged)
8132 {
8133 	u64 *p;
8134 	int lo, hi;
8135 	int rv = 1;
8136 	unsigned long flags;
8137 
8138 	if (bb->shift < 0)
8139 		/* badblocks are disabled */
8140 		return 0;
8141 
8142 	if (bb->shift) {
8143 		/* round the start down, and the end up */
8144 		sector_t next = s + sectors;
8145 		s >>= bb->shift;
8146 		next += (1<<bb->shift) - 1;
8147 		next >>= bb->shift;
8148 		sectors = next - s;
8149 	}
8150 
8151 	write_seqlock_irqsave(&bb->lock, flags);
8152 
8153 	p = bb->page;
8154 	lo = 0;
8155 	hi = bb->count;
8156 	/* Find the last range that starts at-or-before 's' */
8157 	while (hi - lo > 1) {
8158 		int mid = (lo + hi) / 2;
8159 		sector_t a = BB_OFFSET(p[mid]);
8160 		if (a <= s)
8161 			lo = mid;
8162 		else
8163 			hi = mid;
8164 	}
8165 	if (hi > lo && BB_OFFSET(p[lo]) > s)
8166 		hi = lo;
8167 
8168 	if (hi > lo) {
8169 		/* we found a range that might merge with the start
8170 		 * of our new range
8171 		 */
8172 		sector_t a = BB_OFFSET(p[lo]);
8173 		sector_t e = a + BB_LEN(p[lo]);
8174 		int ack = BB_ACK(p[lo]);
8175 		if (e >= s) {
8176 			/* Yes, we can merge with a previous range */
8177 			if (s == a && s + sectors >= e)
8178 				/* new range covers old */
8179 				ack = acknowledged;
8180 			else
8181 				ack = ack && acknowledged;
8182 
8183 			if (e < s + sectors)
8184 				e = s + sectors;
8185 			if (e - a <= BB_MAX_LEN) {
8186 				p[lo] = BB_MAKE(a, e-a, ack);
8187 				s = e;
8188 			} else {
8189 				/* does not all fit in one range,
8190 				 * make p[lo] maximal
8191 				 */
8192 				if (BB_LEN(p[lo]) != BB_MAX_LEN)
8193 					p[lo] = BB_MAKE(a, BB_MAX_LEN, ack);
8194 				s = a + BB_MAX_LEN;
8195 			}
8196 			sectors = e - s;
8197 		}
8198 	}
8199 	if (sectors && hi < bb->count) {
8200 		/* 'hi' points to the first range that starts after 's'.
8201 		 * Maybe we can merge with the start of that range */
8202 		sector_t a = BB_OFFSET(p[hi]);
8203 		sector_t e = a + BB_LEN(p[hi]);
8204 		int ack = BB_ACK(p[hi]);
8205 		if (a <= s + sectors) {
8206 			/* merging is possible */
8207 			if (e <= s + sectors) {
8208 				/* full overlap */
8209 				e = s + sectors;
8210 				ack = acknowledged;
8211 			} else
8212 				ack = ack && acknowledged;
8213 
8214 			a = s;
8215 			if (e - a <= BB_MAX_LEN) {
8216 				p[hi] = BB_MAKE(a, e-a, ack);
8217 				s = e;
8218 			} else {
8219 				p[hi] = BB_MAKE(a, BB_MAX_LEN, ack);
8220 				s = a + BB_MAX_LEN;
8221 			}
8222 			sectors = e - s;
8223 			lo = hi;
8224 			hi++;
8225 		}
8226 	}
8227 	if (sectors == 0 && hi < bb->count) {
8228 		/* we might be able to combine lo and hi */
8229 		/* Note: 's' is at the end of 'lo' */
8230 		sector_t a = BB_OFFSET(p[hi]);
8231 		int lolen = BB_LEN(p[lo]);
8232 		int hilen = BB_LEN(p[hi]);
8233 		int newlen = lolen + hilen - (s - a);
8234 		if (s >= a && newlen < BB_MAX_LEN) {
8235 			/* yes, we can combine them */
8236 			int ack = BB_ACK(p[lo]) && BB_ACK(p[hi]);
8237 			p[lo] = BB_MAKE(BB_OFFSET(p[lo]), newlen, ack);
8238 			memmove(p + hi, p + hi + 1,
8239 				(bb->count - hi - 1) * 8);
8240 			bb->count--;
8241 		}
8242 	}
8243 	while (sectors) {
8244 		/* didn't merge (it all).
8245 		 * Need to add a range just before 'hi' */
8246 		if (bb->count >= MD_MAX_BADBLOCKS) {
8247 			/* No room for more */
8248 			rv = 0;
8249 			break;
8250 		} else {
8251 			int this_sectors = sectors;
8252 			memmove(p + hi + 1, p + hi,
8253 				(bb->count - hi) * 8);
8254 			bb->count++;
8255 
8256 			if (this_sectors > BB_MAX_LEN)
8257 				this_sectors = BB_MAX_LEN;
8258 			p[hi] = BB_MAKE(s, this_sectors, acknowledged);
8259 			sectors -= this_sectors;
8260 			s += this_sectors;
8261 		}
8262 	}
8263 
8264 	bb->changed = 1;
8265 	if (!acknowledged)
8266 		bb->unacked_exist = 1;
8267 	write_sequnlock_irqrestore(&bb->lock, flags);
8268 
8269 	return rv;
8270 }
8271 
8272 int rdev_set_badblocks(struct md_rdev *rdev, sector_t s, int sectors,
8273 		       int is_new)
8274 {
8275 	int rv;
8276 	if (is_new)
8277 		s += rdev->new_data_offset;
8278 	else
8279 		s += rdev->data_offset;
8280 	rv = md_set_badblocks(&rdev->badblocks,
8281 			      s, sectors, 0);
8282 	if (rv) {
8283 		/* Make sure they get written out promptly */
8284 		sysfs_notify_dirent_safe(rdev->sysfs_state);
8285 		set_bit(MD_CHANGE_CLEAN, &rdev->mddev->flags);
8286 		md_wakeup_thread(rdev->mddev->thread);
8287 	}
8288 	return rv;
8289 }
8290 EXPORT_SYMBOL_GPL(rdev_set_badblocks);
8291 
8292 /*
8293  * Remove a range of bad blocks from the table.
8294  * This may involve extending the table if we spilt a region,
8295  * but it must not fail.  So if the table becomes full, we just
8296  * drop the remove request.
8297  */
8298 static int md_clear_badblocks(struct badblocks *bb, sector_t s, int sectors)
8299 {
8300 	u64 *p;
8301 	int lo, hi;
8302 	sector_t target = s + sectors;
8303 	int rv = 0;
8304 
8305 	if (bb->shift > 0) {
8306 		/* When clearing we round the start up and the end down.
8307 		 * This should not matter as the shift should align with
8308 		 * the block size and no rounding should ever be needed.
8309 		 * However it is better the think a block is bad when it
8310 		 * isn't than to think a block is not bad when it is.
8311 		 */
8312 		s += (1<<bb->shift) - 1;
8313 		s >>= bb->shift;
8314 		target >>= bb->shift;
8315 		sectors = target - s;
8316 	}
8317 
8318 	write_seqlock_irq(&bb->lock);
8319 
8320 	p = bb->page;
8321 	lo = 0;
8322 	hi = bb->count;
8323 	/* Find the last range that starts before 'target' */
8324 	while (hi - lo > 1) {
8325 		int mid = (lo + hi) / 2;
8326 		sector_t a = BB_OFFSET(p[mid]);
8327 		if (a < target)
8328 			lo = mid;
8329 		else
8330 			hi = mid;
8331 	}
8332 	if (hi > lo) {
8333 		/* p[lo] is the last range that could overlap the
8334 		 * current range.  Earlier ranges could also overlap,
8335 		 * but only this one can overlap the end of the range.
8336 		 */
8337 		if (BB_OFFSET(p[lo]) + BB_LEN(p[lo]) > target) {
8338 			/* Partial overlap, leave the tail of this range */
8339 			int ack = BB_ACK(p[lo]);
8340 			sector_t a = BB_OFFSET(p[lo]);
8341 			sector_t end = a + BB_LEN(p[lo]);
8342 
8343 			if (a < s) {
8344 				/* we need to split this range */
8345 				if (bb->count >= MD_MAX_BADBLOCKS) {
8346 					rv = -ENOSPC;
8347 					goto out;
8348 				}
8349 				memmove(p+lo+1, p+lo, (bb->count - lo) * 8);
8350 				bb->count++;
8351 				p[lo] = BB_MAKE(a, s-a, ack);
8352 				lo++;
8353 			}
8354 			p[lo] = BB_MAKE(target, end - target, ack);
8355 			/* there is no longer an overlap */
8356 			hi = lo;
8357 			lo--;
8358 		}
8359 		while (lo >= 0 &&
8360 		       BB_OFFSET(p[lo]) + BB_LEN(p[lo]) > s) {
8361 			/* This range does overlap */
8362 			if (BB_OFFSET(p[lo]) < s) {
8363 				/* Keep the early parts of this range. */
8364 				int ack = BB_ACK(p[lo]);
8365 				sector_t start = BB_OFFSET(p[lo]);
8366 				p[lo] = BB_MAKE(start, s - start, ack);
8367 				/* now low doesn't overlap, so.. */
8368 				break;
8369 			}
8370 			lo--;
8371 		}
8372 		/* 'lo' is strictly before, 'hi' is strictly after,
8373 		 * anything between needs to be discarded
8374 		 */
8375 		if (hi - lo > 1) {
8376 			memmove(p+lo+1, p+hi, (bb->count - hi) * 8);
8377 			bb->count -= (hi - lo - 1);
8378 		}
8379 	}
8380 
8381 	bb->changed = 1;
8382 out:
8383 	write_sequnlock_irq(&bb->lock);
8384 	return rv;
8385 }
8386 
8387 int rdev_clear_badblocks(struct md_rdev *rdev, sector_t s, int sectors,
8388 			 int is_new)
8389 {
8390 	if (is_new)
8391 		s += rdev->new_data_offset;
8392 	else
8393 		s += rdev->data_offset;
8394 	return md_clear_badblocks(&rdev->badblocks,
8395 				  s, sectors);
8396 }
8397 EXPORT_SYMBOL_GPL(rdev_clear_badblocks);
8398 
8399 /*
8400  * Acknowledge all bad blocks in a list.
8401  * This only succeeds if ->changed is clear.  It is used by
8402  * in-kernel metadata updates
8403  */
8404 void md_ack_all_badblocks(struct badblocks *bb)
8405 {
8406 	if (bb->page == NULL || bb->changed)
8407 		/* no point even trying */
8408 		return;
8409 	write_seqlock_irq(&bb->lock);
8410 
8411 	if (bb->changed == 0 && bb->unacked_exist) {
8412 		u64 *p = bb->page;
8413 		int i;
8414 		for (i = 0; i < bb->count ; i++) {
8415 			if (!BB_ACK(p[i])) {
8416 				sector_t start = BB_OFFSET(p[i]);
8417 				int len = BB_LEN(p[i]);
8418 				p[i] = BB_MAKE(start, len, 1);
8419 			}
8420 		}
8421 		bb->unacked_exist = 0;
8422 	}
8423 	write_sequnlock_irq(&bb->lock);
8424 }
8425 EXPORT_SYMBOL_GPL(md_ack_all_badblocks);
8426 
8427 /* sysfs access to bad-blocks list.
8428  * We present two files.
8429  * 'bad-blocks' lists sector numbers and lengths of ranges that
8430  *    are recorded as bad.  The list is truncated to fit within
8431  *    the one-page limit of sysfs.
8432  *    Writing "sector length" to this file adds an acknowledged
8433  *    bad block list.
8434  * 'unacknowledged-bad-blocks' lists bad blocks that have not yet
8435  *    been acknowledged.  Writing to this file adds bad blocks
8436  *    without acknowledging them.  This is largely for testing.
8437  */
8438 
8439 static ssize_t
8440 badblocks_show(struct badblocks *bb, char *page, int unack)
8441 {
8442 	size_t len;
8443 	int i;
8444 	u64 *p = bb->page;
8445 	unsigned seq;
8446 
8447 	if (bb->shift < 0)
8448 		return 0;
8449 
8450 retry:
8451 	seq = read_seqbegin(&bb->lock);
8452 
8453 	len = 0;
8454 	i = 0;
8455 
8456 	while (len < PAGE_SIZE && i < bb->count) {
8457 		sector_t s = BB_OFFSET(p[i]);
8458 		unsigned int length = BB_LEN(p[i]);
8459 		int ack = BB_ACK(p[i]);
8460 		i++;
8461 
8462 		if (unack && ack)
8463 			continue;
8464 
8465 		len += snprintf(page+len, PAGE_SIZE-len, "%llu %u\n",
8466 				(unsigned long long)s << bb->shift,
8467 				length << bb->shift);
8468 	}
8469 	if (unack && len == 0)
8470 		bb->unacked_exist = 0;
8471 
8472 	if (read_seqretry(&bb->lock, seq))
8473 		goto retry;
8474 
8475 	return len;
8476 }
8477 
8478 #define DO_DEBUG 1
8479 
8480 static ssize_t
8481 badblocks_store(struct badblocks *bb, const char *page, size_t len, int unack)
8482 {
8483 	unsigned long long sector;
8484 	int length;
8485 	char newline;
8486 #ifdef DO_DEBUG
8487 	/* Allow clearing via sysfs *only* for testing/debugging.
8488 	 * Normally only a successful write may clear a badblock
8489 	 */
8490 	int clear = 0;
8491 	if (page[0] == '-') {
8492 		clear = 1;
8493 		page++;
8494 	}
8495 #endif /* DO_DEBUG */
8496 
8497 	switch (sscanf(page, "%llu %d%c", &sector, &length, &newline)) {
8498 	case 3:
8499 		if (newline != '\n')
8500 			return -EINVAL;
8501 	case 2:
8502 		if (length <= 0)
8503 			return -EINVAL;
8504 		break;
8505 	default:
8506 		return -EINVAL;
8507 	}
8508 
8509 #ifdef DO_DEBUG
8510 	if (clear) {
8511 		md_clear_badblocks(bb, sector, length);
8512 		return len;
8513 	}
8514 #endif /* DO_DEBUG */
8515 	if (md_set_badblocks(bb, sector, length, !unack))
8516 		return len;
8517 	else
8518 		return -ENOSPC;
8519 }
8520 
8521 static int md_notify_reboot(struct notifier_block *this,
8522 			    unsigned long code, void *x)
8523 {
8524 	struct list_head *tmp;
8525 	struct mddev *mddev;
8526 	int need_delay = 0;
8527 
8528 	for_each_mddev(mddev, tmp) {
8529 		if (mddev_trylock(mddev)) {
8530 			if (mddev->pers)
8531 				__md_stop_writes(mddev);
8532 			if (mddev->persistent)
8533 				mddev->safemode = 2;
8534 			mddev_unlock(mddev);
8535 		}
8536 		need_delay = 1;
8537 	}
8538 	/*
8539 	 * certain more exotic SCSI devices are known to be
8540 	 * volatile wrt too early system reboots. While the
8541 	 * right place to handle this issue is the given
8542 	 * driver, we do want to have a safe RAID driver ...
8543 	 */
8544 	if (need_delay)
8545 		mdelay(1000*1);
8546 
8547 	return NOTIFY_DONE;
8548 }
8549 
8550 static struct notifier_block md_notifier = {
8551 	.notifier_call	= md_notify_reboot,
8552 	.next		= NULL,
8553 	.priority	= INT_MAX, /* before any real devices */
8554 };
8555 
8556 static void md_geninit(void)
8557 {
8558 	pr_debug("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
8559 
8560 	proc_create("mdstat", S_IRUGO, NULL, &md_seq_fops);
8561 }
8562 
8563 static int __init md_init(void)
8564 {
8565 	int ret = -ENOMEM;
8566 
8567 	md_wq = alloc_workqueue("md", WQ_MEM_RECLAIM, 0);
8568 	if (!md_wq)
8569 		goto err_wq;
8570 
8571 	md_misc_wq = alloc_workqueue("md_misc", 0, 0);
8572 	if (!md_misc_wq)
8573 		goto err_misc_wq;
8574 
8575 	if ((ret = register_blkdev(MD_MAJOR, "md")) < 0)
8576 		goto err_md;
8577 
8578 	if ((ret = register_blkdev(0, "mdp")) < 0)
8579 		goto err_mdp;
8580 	mdp_major = ret;
8581 
8582 	blk_register_region(MKDEV(MD_MAJOR, 0), 1UL<<MINORBITS, THIS_MODULE,
8583 			    md_probe, NULL, NULL);
8584 	blk_register_region(MKDEV(mdp_major, 0), 1UL<<MINORBITS, THIS_MODULE,
8585 			    md_probe, NULL, NULL);
8586 
8587 	register_reboot_notifier(&md_notifier);
8588 	raid_table_header = register_sysctl_table(raid_root_table);
8589 
8590 	md_geninit();
8591 	return 0;
8592 
8593 err_mdp:
8594 	unregister_blkdev(MD_MAJOR, "md");
8595 err_md:
8596 	destroy_workqueue(md_misc_wq);
8597 err_misc_wq:
8598 	destroy_workqueue(md_wq);
8599 err_wq:
8600 	return ret;
8601 }
8602 
8603 #ifndef MODULE
8604 
8605 /*
8606  * Searches all registered partitions for autorun RAID arrays
8607  * at boot time.
8608  */
8609 
8610 static LIST_HEAD(all_detected_devices);
8611 struct detected_devices_node {
8612 	struct list_head list;
8613 	dev_t dev;
8614 };
8615 
8616 void md_autodetect_dev(dev_t dev)
8617 {
8618 	struct detected_devices_node *node_detected_dev;
8619 
8620 	node_detected_dev = kzalloc(sizeof(*node_detected_dev), GFP_KERNEL);
8621 	if (node_detected_dev) {
8622 		node_detected_dev->dev = dev;
8623 		list_add_tail(&node_detected_dev->list, &all_detected_devices);
8624 	} else {
8625 		printk(KERN_CRIT "md: md_autodetect_dev: kzalloc failed"
8626 			", skipping dev(%d,%d)\n", MAJOR(dev), MINOR(dev));
8627 	}
8628 }
8629 
8630 
8631 static void autostart_arrays(int part)
8632 {
8633 	struct md_rdev *rdev;
8634 	struct detected_devices_node *node_detected_dev;
8635 	dev_t dev;
8636 	int i_scanned, i_passed;
8637 
8638 	i_scanned = 0;
8639 	i_passed = 0;
8640 
8641 	printk(KERN_INFO "md: Autodetecting RAID arrays.\n");
8642 
8643 	while (!list_empty(&all_detected_devices) && i_scanned < INT_MAX) {
8644 		i_scanned++;
8645 		node_detected_dev = list_entry(all_detected_devices.next,
8646 					struct detected_devices_node, list);
8647 		list_del(&node_detected_dev->list);
8648 		dev = node_detected_dev->dev;
8649 		kfree(node_detected_dev);
8650 		rdev = md_import_device(dev,0, 90);
8651 		if (IS_ERR(rdev))
8652 			continue;
8653 
8654 		if (test_bit(Faulty, &rdev->flags)) {
8655 			MD_BUG();
8656 			continue;
8657 		}
8658 		set_bit(AutoDetected, &rdev->flags);
8659 		list_add(&rdev->same_set, &pending_raid_disks);
8660 		i_passed++;
8661 	}
8662 
8663 	printk(KERN_INFO "md: Scanned %d and added %d devices.\n",
8664 						i_scanned, i_passed);
8665 
8666 	autorun_devices(part);
8667 }
8668 
8669 #endif /* !MODULE */
8670 
8671 static __exit void md_exit(void)
8672 {
8673 	struct mddev *mddev;
8674 	struct list_head *tmp;
8675 	int delay = 1;
8676 
8677 	blk_unregister_region(MKDEV(MD_MAJOR,0), 1U << MINORBITS);
8678 	blk_unregister_region(MKDEV(mdp_major,0), 1U << MINORBITS);
8679 
8680 	unregister_blkdev(MD_MAJOR,"md");
8681 	unregister_blkdev(mdp_major, "mdp");
8682 	unregister_reboot_notifier(&md_notifier);
8683 	unregister_sysctl_table(raid_table_header);
8684 
8685 	/* We cannot unload the modules while some process is
8686 	 * waiting for us in select() or poll() - wake them up
8687 	 */
8688 	md_unloading = 1;
8689 	while (waitqueue_active(&md_event_waiters)) {
8690 		/* not safe to leave yet */
8691 		wake_up(&md_event_waiters);
8692 		msleep(delay);
8693 		delay += delay;
8694 	}
8695 	remove_proc_entry("mdstat", NULL);
8696 
8697 	for_each_mddev(mddev, tmp) {
8698 		export_array(mddev);
8699 		mddev->hold_active = 0;
8700 	}
8701 	destroy_workqueue(md_misc_wq);
8702 	destroy_workqueue(md_wq);
8703 }
8704 
8705 subsys_initcall(md_init);
8706 module_exit(md_exit)
8707 
8708 static int get_ro(char *buffer, struct kernel_param *kp)
8709 {
8710 	return sprintf(buffer, "%d", start_readonly);
8711 }
8712 static int set_ro(const char *val, struct kernel_param *kp)
8713 {
8714 	char *e;
8715 	int num = simple_strtoul(val, &e, 10);
8716 	if (*val && (*e == '\0' || *e == '\n')) {
8717 		start_readonly = num;
8718 		return 0;
8719 	}
8720 	return -EINVAL;
8721 }
8722 
8723 module_param_call(start_ro, set_ro, get_ro, NULL, S_IRUSR|S_IWUSR);
8724 module_param(start_dirty_degraded, int, S_IRUGO|S_IWUSR);
8725 
8726 module_param_call(new_array, add_named_array, NULL, NULL, S_IWUSR);
8727 
8728 EXPORT_SYMBOL(register_md_personality);
8729 EXPORT_SYMBOL(unregister_md_personality);
8730 EXPORT_SYMBOL(md_error);
8731 EXPORT_SYMBOL(md_done_sync);
8732 EXPORT_SYMBOL(md_write_start);
8733 EXPORT_SYMBOL(md_write_end);
8734 EXPORT_SYMBOL(md_register_thread);
8735 EXPORT_SYMBOL(md_unregister_thread);
8736 EXPORT_SYMBOL(md_wakeup_thread);
8737 EXPORT_SYMBOL(md_check_recovery);
8738 EXPORT_SYMBOL(md_reap_sync_thread);
8739 MODULE_LICENSE("GPL");
8740 MODULE_DESCRIPTION("MD RAID framework");
8741 MODULE_ALIAS("md");
8742 MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR);
8743