xref: /linux/drivers/md/md.c (revision 507e190946297c34a27d9366b0661d5e506fdd03)
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    Errors, Warnings, etc.
35    Please use:
36      pr_crit() for error conditions that risk data loss
37      pr_err() for error conditions that are unexpected, like an IO error
38          or internal inconsistency
39      pr_warn() for error conditions that could have been predicated, like
40          adding a device to an array when it has incompatible metadata
41      pr_info() for every interesting, very rare events, like an array starting
42          or stopping, or resync starting or stopping
43      pr_debug() for everything else.
44 
45 */
46 
47 #include <linux/sched/signal.h>
48 #include <linux/kthread.h>
49 #include <linux/blkdev.h>
50 #include <linux/badblocks.h>
51 #include <linux/sysctl.h>
52 #include <linux/seq_file.h>
53 #include <linux/fs.h>
54 #include <linux/poll.h>
55 #include <linux/ctype.h>
56 #include <linux/string.h>
57 #include <linux/hdreg.h>
58 #include <linux/proc_fs.h>
59 #include <linux/random.h>
60 #include <linux/module.h>
61 #include <linux/reboot.h>
62 #include <linux/file.h>
63 #include <linux/compat.h>
64 #include <linux/delay.h>
65 #include <linux/raid/md_p.h>
66 #include <linux/raid/md_u.h>
67 #include <linux/slab.h>
68 #include <linux/percpu-refcount.h>
69 
70 #include <trace/events/block.h>
71 #include "md.h"
72 #include "bitmap.h"
73 #include "md-cluster.h"
74 
75 #ifndef MODULE
76 static void autostart_arrays(int part);
77 #endif
78 
79 /* pers_list is a list of registered personalities protected
80  * by pers_lock.
81  * pers_lock does extra service to protect accesses to
82  * mddev->thread when the mutex cannot be held.
83  */
84 static LIST_HEAD(pers_list);
85 static DEFINE_SPINLOCK(pers_lock);
86 
87 struct md_cluster_operations *md_cluster_ops;
88 EXPORT_SYMBOL(md_cluster_ops);
89 struct module *md_cluster_mod;
90 EXPORT_SYMBOL(md_cluster_mod);
91 
92 static DECLARE_WAIT_QUEUE_HEAD(resync_wait);
93 static struct workqueue_struct *md_wq;
94 static struct workqueue_struct *md_misc_wq;
95 
96 static int remove_and_add_spares(struct mddev *mddev,
97 				 struct md_rdev *this);
98 static void mddev_detach(struct mddev *mddev);
99 
100 /*
101  * Default number of read corrections we'll attempt on an rdev
102  * before ejecting it from the array. We divide the read error
103  * count by 2 for every hour elapsed between read errors.
104  */
105 #define MD_DEFAULT_MAX_CORRECTED_READ_ERRORS 20
106 /*
107  * Current RAID-1,4,5 parallel reconstruction 'guaranteed speed limit'
108  * is 1000 KB/sec, so the extra system load does not show up that much.
109  * Increase it if you want to have more _guaranteed_ speed. Note that
110  * the RAID driver will use the maximum available bandwidth if the IO
111  * subsystem is idle. There is also an 'absolute maximum' reconstruction
112  * speed limit - in case reconstruction slows down your system despite
113  * idle IO detection.
114  *
115  * you can change it via /proc/sys/dev/raid/speed_limit_min and _max.
116  * or /sys/block/mdX/md/sync_speed_{min,max}
117  */
118 
119 static int sysctl_speed_limit_min = 1000;
120 static int sysctl_speed_limit_max = 200000;
121 static inline int speed_min(struct mddev *mddev)
122 {
123 	return mddev->sync_speed_min ?
124 		mddev->sync_speed_min : sysctl_speed_limit_min;
125 }
126 
127 static inline int speed_max(struct mddev *mddev)
128 {
129 	return mddev->sync_speed_max ?
130 		mddev->sync_speed_max : sysctl_speed_limit_max;
131 }
132 
133 static struct ctl_table_header *raid_table_header;
134 
135 static struct ctl_table raid_table[] = {
136 	{
137 		.procname	= "speed_limit_min",
138 		.data		= &sysctl_speed_limit_min,
139 		.maxlen		= sizeof(int),
140 		.mode		= S_IRUGO|S_IWUSR,
141 		.proc_handler	= proc_dointvec,
142 	},
143 	{
144 		.procname	= "speed_limit_max",
145 		.data		= &sysctl_speed_limit_max,
146 		.maxlen		= sizeof(int),
147 		.mode		= S_IRUGO|S_IWUSR,
148 		.proc_handler	= proc_dointvec,
149 	},
150 	{ }
151 };
152 
153 static struct ctl_table raid_dir_table[] = {
154 	{
155 		.procname	= "raid",
156 		.maxlen		= 0,
157 		.mode		= S_IRUGO|S_IXUGO,
158 		.child		= raid_table,
159 	},
160 	{ }
161 };
162 
163 static struct ctl_table raid_root_table[] = {
164 	{
165 		.procname	= "dev",
166 		.maxlen		= 0,
167 		.mode		= 0555,
168 		.child		= raid_dir_table,
169 	},
170 	{  }
171 };
172 
173 static const struct block_device_operations md_fops;
174 
175 static int start_readonly;
176 
177 /*
178  * The original mechanism for creating an md device is to create
179  * a device node in /dev and to open it.  This causes races with device-close.
180  * The preferred method is to write to the "new_array" module parameter.
181  * This can avoid races.
182  * Setting create_on_open to false disables the original mechanism
183  * so all the races disappear.
184  */
185 static bool create_on_open = true;
186 
187 /* bio_clone_mddev
188  * like bio_clone_bioset, but with a local bio set
189  */
190 
191 struct bio *bio_alloc_mddev(gfp_t gfp_mask, int nr_iovecs,
192 			    struct mddev *mddev)
193 {
194 	struct bio *b;
195 
196 	if (!mddev || !mddev->bio_set)
197 		return bio_alloc(gfp_mask, nr_iovecs);
198 
199 	b = bio_alloc_bioset(gfp_mask, nr_iovecs, mddev->bio_set);
200 	if (!b)
201 		return NULL;
202 	return b;
203 }
204 EXPORT_SYMBOL_GPL(bio_alloc_mddev);
205 
206 static struct bio *md_bio_alloc_sync(struct mddev *mddev)
207 {
208 	if (!mddev || !mddev->sync_set)
209 		return bio_alloc(GFP_NOIO, 1);
210 
211 	return bio_alloc_bioset(GFP_NOIO, 1, mddev->sync_set);
212 }
213 
214 /*
215  * We have a system wide 'event count' that is incremented
216  * on any 'interesting' event, and readers of /proc/mdstat
217  * can use 'poll' or 'select' to find out when the event
218  * count increases.
219  *
220  * Events are:
221  *  start array, stop array, error, add device, remove device,
222  *  start build, activate spare
223  */
224 static DECLARE_WAIT_QUEUE_HEAD(md_event_waiters);
225 static atomic_t md_event_count;
226 void md_new_event(struct mddev *mddev)
227 {
228 	atomic_inc(&md_event_count);
229 	wake_up(&md_event_waiters);
230 }
231 EXPORT_SYMBOL_GPL(md_new_event);
232 
233 /*
234  * Enables to iterate over all existing md arrays
235  * all_mddevs_lock protects this list.
236  */
237 static LIST_HEAD(all_mddevs);
238 static DEFINE_SPINLOCK(all_mddevs_lock);
239 
240 /*
241  * iterates through all used mddevs in the system.
242  * We take care to grab the all_mddevs_lock whenever navigating
243  * the list, and to always hold a refcount when unlocked.
244  * Any code which breaks out of this loop while own
245  * a reference to the current mddev and must mddev_put it.
246  */
247 #define for_each_mddev(_mddev,_tmp)					\
248 									\
249 	for (({ spin_lock(&all_mddevs_lock);				\
250 		_tmp = all_mddevs.next;					\
251 		_mddev = NULL;});					\
252 	     ({ if (_tmp != &all_mddevs)				\
253 			mddev_get(list_entry(_tmp, struct mddev, all_mddevs));\
254 		spin_unlock(&all_mddevs_lock);				\
255 		if (_mddev) mddev_put(_mddev);				\
256 		_mddev = list_entry(_tmp, struct mddev, all_mddevs);	\
257 		_tmp != &all_mddevs;});					\
258 	     ({ spin_lock(&all_mddevs_lock);				\
259 		_tmp = _tmp->next;})					\
260 		)
261 
262 /* Rather than calling directly into the personality make_request function,
263  * IO requests come here first so that we can check if the device is
264  * being suspended pending a reconfiguration.
265  * We hold a refcount over the call to ->make_request.  By the time that
266  * call has finished, the bio has been linked into some internal structure
267  * and so is visible to ->quiesce(), so we don't need the refcount any more.
268  */
269 static blk_qc_t md_make_request(struct request_queue *q, struct bio *bio)
270 {
271 	const int rw = bio_data_dir(bio);
272 	struct mddev *mddev = q->queuedata;
273 	unsigned int sectors;
274 	int cpu;
275 
276 	blk_queue_split(q, &bio);
277 
278 	if (mddev == NULL || mddev->pers == NULL) {
279 		bio_io_error(bio);
280 		return BLK_QC_T_NONE;
281 	}
282 	if (mddev->ro == 1 && unlikely(rw == WRITE)) {
283 		if (bio_sectors(bio) != 0)
284 			bio->bi_status = BLK_STS_IOERR;
285 		bio_endio(bio);
286 		return BLK_QC_T_NONE;
287 	}
288 check_suspended:
289 	rcu_read_lock();
290 	if (mddev->suspended) {
291 		DEFINE_WAIT(__wait);
292 		for (;;) {
293 			prepare_to_wait(&mddev->sb_wait, &__wait,
294 					TASK_UNINTERRUPTIBLE);
295 			if (!mddev->suspended)
296 				break;
297 			rcu_read_unlock();
298 			schedule();
299 			rcu_read_lock();
300 		}
301 		finish_wait(&mddev->sb_wait, &__wait);
302 	}
303 	atomic_inc(&mddev->active_io);
304 	rcu_read_unlock();
305 
306 	/*
307 	 * save the sectors now since our bio can
308 	 * go away inside make_request
309 	 */
310 	sectors = bio_sectors(bio);
311 	/* bio could be mergeable after passing to underlayer */
312 	bio->bi_opf &= ~REQ_NOMERGE;
313 	if (!mddev->pers->make_request(mddev, bio)) {
314 		atomic_dec(&mddev->active_io);
315 		wake_up(&mddev->sb_wait);
316 		goto check_suspended;
317 	}
318 
319 	cpu = part_stat_lock();
320 	part_stat_inc(cpu, &mddev->gendisk->part0, ios[rw]);
321 	part_stat_add(cpu, &mddev->gendisk->part0, sectors[rw], sectors);
322 	part_stat_unlock();
323 
324 	if (atomic_dec_and_test(&mddev->active_io) && mddev->suspended)
325 		wake_up(&mddev->sb_wait);
326 
327 	return BLK_QC_T_NONE;
328 }
329 
330 /* mddev_suspend makes sure no new requests are submitted
331  * to the device, and that any requests that have been submitted
332  * are completely handled.
333  * Once mddev_detach() is called and completes, the module will be
334  * completely unused.
335  */
336 void mddev_suspend(struct mddev *mddev)
337 {
338 	WARN_ON_ONCE(mddev->thread && current == mddev->thread->tsk);
339 	if (mddev->suspended++)
340 		return;
341 	synchronize_rcu();
342 	wake_up(&mddev->sb_wait);
343 	wait_event(mddev->sb_wait, atomic_read(&mddev->active_io) == 0);
344 	mddev->pers->quiesce(mddev, 1);
345 
346 	del_timer_sync(&mddev->safemode_timer);
347 }
348 EXPORT_SYMBOL_GPL(mddev_suspend);
349 
350 void mddev_resume(struct mddev *mddev)
351 {
352 	if (--mddev->suspended)
353 		return;
354 	wake_up(&mddev->sb_wait);
355 	mddev->pers->quiesce(mddev, 0);
356 
357 	set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
358 	md_wakeup_thread(mddev->thread);
359 	md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
360 }
361 EXPORT_SYMBOL_GPL(mddev_resume);
362 
363 int mddev_congested(struct mddev *mddev, int bits)
364 {
365 	struct md_personality *pers = mddev->pers;
366 	int ret = 0;
367 
368 	rcu_read_lock();
369 	if (mddev->suspended)
370 		ret = 1;
371 	else if (pers && pers->congested)
372 		ret = pers->congested(mddev, bits);
373 	rcu_read_unlock();
374 	return ret;
375 }
376 EXPORT_SYMBOL_GPL(mddev_congested);
377 static int md_congested(void *data, int bits)
378 {
379 	struct mddev *mddev = data;
380 	return mddev_congested(mddev, bits);
381 }
382 
383 /*
384  * Generic flush handling for md
385  */
386 
387 static void md_end_flush(struct bio *bio)
388 {
389 	struct md_rdev *rdev = bio->bi_private;
390 	struct mddev *mddev = rdev->mddev;
391 
392 	rdev_dec_pending(rdev, mddev);
393 
394 	if (atomic_dec_and_test(&mddev->flush_pending)) {
395 		/* The pre-request flush has finished */
396 		queue_work(md_wq, &mddev->flush_work);
397 	}
398 	bio_put(bio);
399 }
400 
401 static void md_submit_flush_data(struct work_struct *ws);
402 
403 static void submit_flushes(struct work_struct *ws)
404 {
405 	struct mddev *mddev = container_of(ws, struct mddev, flush_work);
406 	struct md_rdev *rdev;
407 
408 	INIT_WORK(&mddev->flush_work, md_submit_flush_data);
409 	atomic_set(&mddev->flush_pending, 1);
410 	rcu_read_lock();
411 	rdev_for_each_rcu(rdev, mddev)
412 		if (rdev->raid_disk >= 0 &&
413 		    !test_bit(Faulty, &rdev->flags)) {
414 			/* Take two references, one is dropped
415 			 * when request finishes, one after
416 			 * we reclaim rcu_read_lock
417 			 */
418 			struct bio *bi;
419 			atomic_inc(&rdev->nr_pending);
420 			atomic_inc(&rdev->nr_pending);
421 			rcu_read_unlock();
422 			bi = bio_alloc_mddev(GFP_NOIO, 0, mddev);
423 			bi->bi_end_io = md_end_flush;
424 			bi->bi_private = rdev;
425 			bi->bi_bdev = rdev->bdev;
426 			bi->bi_opf = REQ_OP_WRITE | REQ_PREFLUSH;
427 			atomic_inc(&mddev->flush_pending);
428 			submit_bio(bi);
429 			rcu_read_lock();
430 			rdev_dec_pending(rdev, mddev);
431 		}
432 	rcu_read_unlock();
433 	if (atomic_dec_and_test(&mddev->flush_pending))
434 		queue_work(md_wq, &mddev->flush_work);
435 }
436 
437 static void md_submit_flush_data(struct work_struct *ws)
438 {
439 	struct mddev *mddev = container_of(ws, struct mddev, flush_work);
440 	struct bio *bio = mddev->flush_bio;
441 
442 	if (bio->bi_iter.bi_size == 0)
443 		/* an empty barrier - all done */
444 		bio_endio(bio);
445 	else {
446 		bio->bi_opf &= ~REQ_PREFLUSH;
447 		mddev->pers->make_request(mddev, bio);
448 	}
449 
450 	mddev->flush_bio = NULL;
451 	wake_up(&mddev->sb_wait);
452 }
453 
454 void md_flush_request(struct mddev *mddev, struct bio *bio)
455 {
456 	spin_lock_irq(&mddev->lock);
457 	wait_event_lock_irq(mddev->sb_wait,
458 			    !mddev->flush_bio,
459 			    mddev->lock);
460 	mddev->flush_bio = bio;
461 	spin_unlock_irq(&mddev->lock);
462 
463 	INIT_WORK(&mddev->flush_work, submit_flushes);
464 	queue_work(md_wq, &mddev->flush_work);
465 }
466 EXPORT_SYMBOL(md_flush_request);
467 
468 static inline struct mddev *mddev_get(struct mddev *mddev)
469 {
470 	atomic_inc(&mddev->active);
471 	return mddev;
472 }
473 
474 static void mddev_delayed_delete(struct work_struct *ws);
475 
476 static void mddev_put(struct mddev *mddev)
477 {
478 	struct bio_set *bs = NULL, *sync_bs = NULL;
479 
480 	if (!atomic_dec_and_lock(&mddev->active, &all_mddevs_lock))
481 		return;
482 	if (!mddev->raid_disks && list_empty(&mddev->disks) &&
483 	    mddev->ctime == 0 && !mddev->hold_active) {
484 		/* Array is not configured at all, and not held active,
485 		 * so destroy it */
486 		list_del_init(&mddev->all_mddevs);
487 		bs = mddev->bio_set;
488 		sync_bs = mddev->sync_set;
489 		mddev->bio_set = NULL;
490 		mddev->sync_set = NULL;
491 		if (mddev->gendisk) {
492 			/* We did a probe so need to clean up.  Call
493 			 * queue_work inside the spinlock so that
494 			 * flush_workqueue() after mddev_find will
495 			 * succeed in waiting for the work to be done.
496 			 */
497 			INIT_WORK(&mddev->del_work, mddev_delayed_delete);
498 			queue_work(md_misc_wq, &mddev->del_work);
499 		} else
500 			kfree(mddev);
501 	}
502 	spin_unlock(&all_mddevs_lock);
503 	if (bs)
504 		bioset_free(bs);
505 	if (sync_bs)
506 		bioset_free(sync_bs);
507 }
508 
509 static void md_safemode_timeout(unsigned long data);
510 
511 void mddev_init(struct mddev *mddev)
512 {
513 	mutex_init(&mddev->open_mutex);
514 	mutex_init(&mddev->reconfig_mutex);
515 	mutex_init(&mddev->bitmap_info.mutex);
516 	INIT_LIST_HEAD(&mddev->disks);
517 	INIT_LIST_HEAD(&mddev->all_mddevs);
518 	setup_timer(&mddev->safemode_timer, md_safemode_timeout,
519 		    (unsigned long) mddev);
520 	atomic_set(&mddev->active, 1);
521 	atomic_set(&mddev->openers, 0);
522 	atomic_set(&mddev->active_io, 0);
523 	spin_lock_init(&mddev->lock);
524 	atomic_set(&mddev->flush_pending, 0);
525 	init_waitqueue_head(&mddev->sb_wait);
526 	init_waitqueue_head(&mddev->recovery_wait);
527 	mddev->reshape_position = MaxSector;
528 	mddev->reshape_backwards = 0;
529 	mddev->last_sync_action = "none";
530 	mddev->resync_min = 0;
531 	mddev->resync_max = MaxSector;
532 	mddev->level = LEVEL_NONE;
533 }
534 EXPORT_SYMBOL_GPL(mddev_init);
535 
536 static struct mddev *mddev_find(dev_t unit)
537 {
538 	struct mddev *mddev, *new = NULL;
539 
540 	if (unit && MAJOR(unit) != MD_MAJOR)
541 		unit &= ~((1<<MdpMinorShift)-1);
542 
543  retry:
544 	spin_lock(&all_mddevs_lock);
545 
546 	if (unit) {
547 		list_for_each_entry(mddev, &all_mddevs, all_mddevs)
548 			if (mddev->unit == unit) {
549 				mddev_get(mddev);
550 				spin_unlock(&all_mddevs_lock);
551 				kfree(new);
552 				return mddev;
553 			}
554 
555 		if (new) {
556 			list_add(&new->all_mddevs, &all_mddevs);
557 			spin_unlock(&all_mddevs_lock);
558 			new->hold_active = UNTIL_IOCTL;
559 			return new;
560 		}
561 	} else if (new) {
562 		/* find an unused unit number */
563 		static int next_minor = 512;
564 		int start = next_minor;
565 		int is_free = 0;
566 		int dev = 0;
567 		while (!is_free) {
568 			dev = MKDEV(MD_MAJOR, next_minor);
569 			next_minor++;
570 			if (next_minor > MINORMASK)
571 				next_minor = 0;
572 			if (next_minor == start) {
573 				/* Oh dear, all in use. */
574 				spin_unlock(&all_mddevs_lock);
575 				kfree(new);
576 				return NULL;
577 			}
578 
579 			is_free = 1;
580 			list_for_each_entry(mddev, &all_mddevs, all_mddevs)
581 				if (mddev->unit == dev) {
582 					is_free = 0;
583 					break;
584 				}
585 		}
586 		new->unit = dev;
587 		new->md_minor = MINOR(dev);
588 		new->hold_active = UNTIL_STOP;
589 		list_add(&new->all_mddevs, &all_mddevs);
590 		spin_unlock(&all_mddevs_lock);
591 		return new;
592 	}
593 	spin_unlock(&all_mddevs_lock);
594 
595 	new = kzalloc(sizeof(*new), GFP_KERNEL);
596 	if (!new)
597 		return NULL;
598 
599 	new->unit = unit;
600 	if (MAJOR(unit) == MD_MAJOR)
601 		new->md_minor = MINOR(unit);
602 	else
603 		new->md_minor = MINOR(unit) >> MdpMinorShift;
604 
605 	mddev_init(new);
606 
607 	goto retry;
608 }
609 
610 static struct attribute_group md_redundancy_group;
611 
612 void mddev_unlock(struct mddev *mddev)
613 {
614 	if (mddev->to_remove) {
615 		/* These cannot be removed under reconfig_mutex as
616 		 * an access to the files will try to take reconfig_mutex
617 		 * while holding the file unremovable, which leads to
618 		 * a deadlock.
619 		 * So hold set sysfs_active while the remove in happeing,
620 		 * and anything else which might set ->to_remove or my
621 		 * otherwise change the sysfs namespace will fail with
622 		 * -EBUSY if sysfs_active is still set.
623 		 * We set sysfs_active under reconfig_mutex and elsewhere
624 		 * test it under the same mutex to ensure its correct value
625 		 * is seen.
626 		 */
627 		struct attribute_group *to_remove = mddev->to_remove;
628 		mddev->to_remove = NULL;
629 		mddev->sysfs_active = 1;
630 		mutex_unlock(&mddev->reconfig_mutex);
631 
632 		if (mddev->kobj.sd) {
633 			if (to_remove != &md_redundancy_group)
634 				sysfs_remove_group(&mddev->kobj, to_remove);
635 			if (mddev->pers == NULL ||
636 			    mddev->pers->sync_request == NULL) {
637 				sysfs_remove_group(&mddev->kobj, &md_redundancy_group);
638 				if (mddev->sysfs_action)
639 					sysfs_put(mddev->sysfs_action);
640 				mddev->sysfs_action = NULL;
641 			}
642 		}
643 		mddev->sysfs_active = 0;
644 	} else
645 		mutex_unlock(&mddev->reconfig_mutex);
646 
647 	/* As we've dropped the mutex we need a spinlock to
648 	 * make sure the thread doesn't disappear
649 	 */
650 	spin_lock(&pers_lock);
651 	md_wakeup_thread(mddev->thread);
652 	spin_unlock(&pers_lock);
653 }
654 EXPORT_SYMBOL_GPL(mddev_unlock);
655 
656 struct md_rdev *md_find_rdev_nr_rcu(struct mddev *mddev, int nr)
657 {
658 	struct md_rdev *rdev;
659 
660 	rdev_for_each_rcu(rdev, mddev)
661 		if (rdev->desc_nr == nr)
662 			return rdev;
663 
664 	return NULL;
665 }
666 EXPORT_SYMBOL_GPL(md_find_rdev_nr_rcu);
667 
668 static struct md_rdev *find_rdev(struct mddev *mddev, dev_t dev)
669 {
670 	struct md_rdev *rdev;
671 
672 	rdev_for_each(rdev, mddev)
673 		if (rdev->bdev->bd_dev == dev)
674 			return rdev;
675 
676 	return NULL;
677 }
678 
679 static struct md_rdev *find_rdev_rcu(struct mddev *mddev, dev_t dev)
680 {
681 	struct md_rdev *rdev;
682 
683 	rdev_for_each_rcu(rdev, mddev)
684 		if (rdev->bdev->bd_dev == dev)
685 			return rdev;
686 
687 	return NULL;
688 }
689 
690 static struct md_personality *find_pers(int level, char *clevel)
691 {
692 	struct md_personality *pers;
693 	list_for_each_entry(pers, &pers_list, list) {
694 		if (level != LEVEL_NONE && pers->level == level)
695 			return pers;
696 		if (strcmp(pers->name, clevel)==0)
697 			return pers;
698 	}
699 	return NULL;
700 }
701 
702 /* return the offset of the super block in 512byte sectors */
703 static inline sector_t calc_dev_sboffset(struct md_rdev *rdev)
704 {
705 	sector_t num_sectors = i_size_read(rdev->bdev->bd_inode) / 512;
706 	return MD_NEW_SIZE_SECTORS(num_sectors);
707 }
708 
709 static int alloc_disk_sb(struct md_rdev *rdev)
710 {
711 	rdev->sb_page = alloc_page(GFP_KERNEL);
712 	if (!rdev->sb_page)
713 		return -ENOMEM;
714 	return 0;
715 }
716 
717 void md_rdev_clear(struct md_rdev *rdev)
718 {
719 	if (rdev->sb_page) {
720 		put_page(rdev->sb_page);
721 		rdev->sb_loaded = 0;
722 		rdev->sb_page = NULL;
723 		rdev->sb_start = 0;
724 		rdev->sectors = 0;
725 	}
726 	if (rdev->bb_page) {
727 		put_page(rdev->bb_page);
728 		rdev->bb_page = NULL;
729 	}
730 	badblocks_exit(&rdev->badblocks);
731 }
732 EXPORT_SYMBOL_GPL(md_rdev_clear);
733 
734 static void super_written(struct bio *bio)
735 {
736 	struct md_rdev *rdev = bio->bi_private;
737 	struct mddev *mddev = rdev->mddev;
738 
739 	if (bio->bi_status) {
740 		pr_err("md: super_written gets error=%d\n", bio->bi_status);
741 		md_error(mddev, rdev);
742 		if (!test_bit(Faulty, &rdev->flags)
743 		    && (bio->bi_opf & MD_FAILFAST)) {
744 			set_bit(MD_SB_NEED_REWRITE, &mddev->sb_flags);
745 			set_bit(LastDev, &rdev->flags);
746 		}
747 	} else
748 		clear_bit(LastDev, &rdev->flags);
749 
750 	if (atomic_dec_and_test(&mddev->pending_writes))
751 		wake_up(&mddev->sb_wait);
752 	rdev_dec_pending(rdev, mddev);
753 	bio_put(bio);
754 }
755 
756 void md_super_write(struct mddev *mddev, struct md_rdev *rdev,
757 		   sector_t sector, int size, struct page *page)
758 {
759 	/* write first size bytes of page to sector of rdev
760 	 * Increment mddev->pending_writes before returning
761 	 * and decrement it on completion, waking up sb_wait
762 	 * if zero is reached.
763 	 * If an error occurred, call md_error
764 	 */
765 	struct bio *bio;
766 	int ff = 0;
767 
768 	if (test_bit(Faulty, &rdev->flags))
769 		return;
770 
771 	bio = md_bio_alloc_sync(mddev);
772 
773 	atomic_inc(&rdev->nr_pending);
774 
775 	bio->bi_bdev = rdev->meta_bdev ? rdev->meta_bdev : rdev->bdev;
776 	bio->bi_iter.bi_sector = sector;
777 	bio_add_page(bio, page, size, 0);
778 	bio->bi_private = rdev;
779 	bio->bi_end_io = super_written;
780 
781 	if (test_bit(MD_FAILFAST_SUPPORTED, &mddev->flags) &&
782 	    test_bit(FailFast, &rdev->flags) &&
783 	    !test_bit(LastDev, &rdev->flags))
784 		ff = MD_FAILFAST;
785 	bio->bi_opf = REQ_OP_WRITE | REQ_SYNC | REQ_PREFLUSH | REQ_FUA | ff;
786 
787 	atomic_inc(&mddev->pending_writes);
788 	submit_bio(bio);
789 }
790 
791 int md_super_wait(struct mddev *mddev)
792 {
793 	/* wait for all superblock writes that were scheduled to complete */
794 	wait_event(mddev->sb_wait, atomic_read(&mddev->pending_writes)==0);
795 	if (test_and_clear_bit(MD_SB_NEED_REWRITE, &mddev->sb_flags))
796 		return -EAGAIN;
797 	return 0;
798 }
799 
800 int sync_page_io(struct md_rdev *rdev, sector_t sector, int size,
801 		 struct page *page, int op, int op_flags, bool metadata_op)
802 {
803 	struct bio *bio = md_bio_alloc_sync(rdev->mddev);
804 	int ret;
805 
806 	bio->bi_bdev = (metadata_op && rdev->meta_bdev) ?
807 		rdev->meta_bdev : rdev->bdev;
808 	bio_set_op_attrs(bio, op, op_flags);
809 	if (metadata_op)
810 		bio->bi_iter.bi_sector = sector + rdev->sb_start;
811 	else if (rdev->mddev->reshape_position != MaxSector &&
812 		 (rdev->mddev->reshape_backwards ==
813 		  (sector >= rdev->mddev->reshape_position)))
814 		bio->bi_iter.bi_sector = sector + rdev->new_data_offset;
815 	else
816 		bio->bi_iter.bi_sector = sector + rdev->data_offset;
817 	bio_add_page(bio, page, size, 0);
818 
819 	submit_bio_wait(bio);
820 
821 	ret = !bio->bi_status;
822 	bio_put(bio);
823 	return ret;
824 }
825 EXPORT_SYMBOL_GPL(sync_page_io);
826 
827 static int read_disk_sb(struct md_rdev *rdev, int size)
828 {
829 	char b[BDEVNAME_SIZE];
830 
831 	if (rdev->sb_loaded)
832 		return 0;
833 
834 	if (!sync_page_io(rdev, 0, size, rdev->sb_page, REQ_OP_READ, 0, true))
835 		goto fail;
836 	rdev->sb_loaded = 1;
837 	return 0;
838 
839 fail:
840 	pr_err("md: disabled device %s, could not read superblock.\n",
841 	       bdevname(rdev->bdev,b));
842 	return -EINVAL;
843 }
844 
845 static int md_uuid_equal(mdp_super_t *sb1, mdp_super_t *sb2)
846 {
847 	return	sb1->set_uuid0 == sb2->set_uuid0 &&
848 		sb1->set_uuid1 == sb2->set_uuid1 &&
849 		sb1->set_uuid2 == sb2->set_uuid2 &&
850 		sb1->set_uuid3 == sb2->set_uuid3;
851 }
852 
853 static int md_sb_equal(mdp_super_t *sb1, mdp_super_t *sb2)
854 {
855 	int ret;
856 	mdp_super_t *tmp1, *tmp2;
857 
858 	tmp1 = kmalloc(sizeof(*tmp1),GFP_KERNEL);
859 	tmp2 = kmalloc(sizeof(*tmp2),GFP_KERNEL);
860 
861 	if (!tmp1 || !tmp2) {
862 		ret = 0;
863 		goto abort;
864 	}
865 
866 	*tmp1 = *sb1;
867 	*tmp2 = *sb2;
868 
869 	/*
870 	 * nr_disks is not constant
871 	 */
872 	tmp1->nr_disks = 0;
873 	tmp2->nr_disks = 0;
874 
875 	ret = (memcmp(tmp1, tmp2, MD_SB_GENERIC_CONSTANT_WORDS * 4) == 0);
876 abort:
877 	kfree(tmp1);
878 	kfree(tmp2);
879 	return ret;
880 }
881 
882 static u32 md_csum_fold(u32 csum)
883 {
884 	csum = (csum & 0xffff) + (csum >> 16);
885 	return (csum & 0xffff) + (csum >> 16);
886 }
887 
888 static unsigned int calc_sb_csum(mdp_super_t *sb)
889 {
890 	u64 newcsum = 0;
891 	u32 *sb32 = (u32*)sb;
892 	int i;
893 	unsigned int disk_csum, csum;
894 
895 	disk_csum = sb->sb_csum;
896 	sb->sb_csum = 0;
897 
898 	for (i = 0; i < MD_SB_BYTES/4 ; i++)
899 		newcsum += sb32[i];
900 	csum = (newcsum & 0xffffffff) + (newcsum>>32);
901 
902 #ifdef CONFIG_ALPHA
903 	/* This used to use csum_partial, which was wrong for several
904 	 * reasons including that different results are returned on
905 	 * different architectures.  It isn't critical that we get exactly
906 	 * the same return value as before (we always csum_fold before
907 	 * testing, and that removes any differences).  However as we
908 	 * know that csum_partial always returned a 16bit value on
909 	 * alphas, do a fold to maximise conformity to previous behaviour.
910 	 */
911 	sb->sb_csum = md_csum_fold(disk_csum);
912 #else
913 	sb->sb_csum = disk_csum;
914 #endif
915 	return csum;
916 }
917 
918 /*
919  * Handle superblock details.
920  * We want to be able to handle multiple superblock formats
921  * so we have a common interface to them all, and an array of
922  * different handlers.
923  * We rely on user-space to write the initial superblock, and support
924  * reading and updating of superblocks.
925  * Interface methods are:
926  *   int load_super(struct md_rdev *dev, struct md_rdev *refdev, int minor_version)
927  *      loads and validates a superblock on dev.
928  *      if refdev != NULL, compare superblocks on both devices
929  *    Return:
930  *      0 - dev has a superblock that is compatible with refdev
931  *      1 - dev has a superblock that is compatible and newer than refdev
932  *          so dev should be used as the refdev in future
933  *     -EINVAL superblock incompatible or invalid
934  *     -othererror e.g. -EIO
935  *
936  *   int validate_super(struct mddev *mddev, struct md_rdev *dev)
937  *      Verify that dev is acceptable into mddev.
938  *       The first time, mddev->raid_disks will be 0, and data from
939  *       dev should be merged in.  Subsequent calls check that dev
940  *       is new enough.  Return 0 or -EINVAL
941  *
942  *   void sync_super(struct mddev *mddev, struct md_rdev *dev)
943  *     Update the superblock for rdev with data in mddev
944  *     This does not write to disc.
945  *
946  */
947 
948 struct super_type  {
949 	char		    *name;
950 	struct module	    *owner;
951 	int		    (*load_super)(struct md_rdev *rdev,
952 					  struct md_rdev *refdev,
953 					  int minor_version);
954 	int		    (*validate_super)(struct mddev *mddev,
955 					      struct md_rdev *rdev);
956 	void		    (*sync_super)(struct mddev *mddev,
957 					  struct md_rdev *rdev);
958 	unsigned long long  (*rdev_size_change)(struct md_rdev *rdev,
959 						sector_t num_sectors);
960 	int		    (*allow_new_offset)(struct md_rdev *rdev,
961 						unsigned long long new_offset);
962 };
963 
964 /*
965  * Check that the given mddev has no bitmap.
966  *
967  * This function is called from the run method of all personalities that do not
968  * support bitmaps. It prints an error message and returns non-zero if mddev
969  * has a bitmap. Otherwise, it returns 0.
970  *
971  */
972 int md_check_no_bitmap(struct mddev *mddev)
973 {
974 	if (!mddev->bitmap_info.file && !mddev->bitmap_info.offset)
975 		return 0;
976 	pr_warn("%s: bitmaps are not supported for %s\n",
977 		mdname(mddev), mddev->pers->name);
978 	return 1;
979 }
980 EXPORT_SYMBOL(md_check_no_bitmap);
981 
982 /*
983  * load_super for 0.90.0
984  */
985 static int super_90_load(struct md_rdev *rdev, struct md_rdev *refdev, int minor_version)
986 {
987 	char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
988 	mdp_super_t *sb;
989 	int ret;
990 
991 	/*
992 	 * Calculate the position of the superblock (512byte sectors),
993 	 * it's at the end of the disk.
994 	 *
995 	 * It also happens to be a multiple of 4Kb.
996 	 */
997 	rdev->sb_start = calc_dev_sboffset(rdev);
998 
999 	ret = read_disk_sb(rdev, MD_SB_BYTES);
1000 	if (ret)
1001 		return ret;
1002 
1003 	ret = -EINVAL;
1004 
1005 	bdevname(rdev->bdev, b);
1006 	sb = page_address(rdev->sb_page);
1007 
1008 	if (sb->md_magic != MD_SB_MAGIC) {
1009 		pr_warn("md: invalid raid superblock magic on %s\n", b);
1010 		goto abort;
1011 	}
1012 
1013 	if (sb->major_version != 0 ||
1014 	    sb->minor_version < 90 ||
1015 	    sb->minor_version > 91) {
1016 		pr_warn("Bad version number %d.%d on %s\n",
1017 			sb->major_version, sb->minor_version, b);
1018 		goto abort;
1019 	}
1020 
1021 	if (sb->raid_disks <= 0)
1022 		goto abort;
1023 
1024 	if (md_csum_fold(calc_sb_csum(sb)) != md_csum_fold(sb->sb_csum)) {
1025 		pr_warn("md: invalid superblock checksum on %s\n", b);
1026 		goto abort;
1027 	}
1028 
1029 	rdev->preferred_minor = sb->md_minor;
1030 	rdev->data_offset = 0;
1031 	rdev->new_data_offset = 0;
1032 	rdev->sb_size = MD_SB_BYTES;
1033 	rdev->badblocks.shift = -1;
1034 
1035 	if (sb->level == LEVEL_MULTIPATH)
1036 		rdev->desc_nr = -1;
1037 	else
1038 		rdev->desc_nr = sb->this_disk.number;
1039 
1040 	if (!refdev) {
1041 		ret = 1;
1042 	} else {
1043 		__u64 ev1, ev2;
1044 		mdp_super_t *refsb = page_address(refdev->sb_page);
1045 		if (!md_uuid_equal(refsb, sb)) {
1046 			pr_warn("md: %s has different UUID to %s\n",
1047 				b, bdevname(refdev->bdev,b2));
1048 			goto abort;
1049 		}
1050 		if (!md_sb_equal(refsb, sb)) {
1051 			pr_warn("md: %s has same UUID but different superblock to %s\n",
1052 				b, bdevname(refdev->bdev, b2));
1053 			goto abort;
1054 		}
1055 		ev1 = md_event(sb);
1056 		ev2 = md_event(refsb);
1057 		if (ev1 > ev2)
1058 			ret = 1;
1059 		else
1060 			ret = 0;
1061 	}
1062 	rdev->sectors = rdev->sb_start;
1063 	/* Limit to 4TB as metadata cannot record more than that.
1064 	 * (not needed for Linear and RAID0 as metadata doesn't
1065 	 * record this size)
1066 	 */
1067 	if (IS_ENABLED(CONFIG_LBDAF) && (u64)rdev->sectors >= (2ULL << 32) &&
1068 	    sb->level >= 1)
1069 		rdev->sectors = (sector_t)(2ULL << 32) - 2;
1070 
1071 	if (rdev->sectors < ((sector_t)sb->size) * 2 && sb->level >= 1)
1072 		/* "this cannot possibly happen" ... */
1073 		ret = -EINVAL;
1074 
1075  abort:
1076 	return ret;
1077 }
1078 
1079 /*
1080  * validate_super for 0.90.0
1081  */
1082 static int super_90_validate(struct mddev *mddev, struct md_rdev *rdev)
1083 {
1084 	mdp_disk_t *desc;
1085 	mdp_super_t *sb = page_address(rdev->sb_page);
1086 	__u64 ev1 = md_event(sb);
1087 
1088 	rdev->raid_disk = -1;
1089 	clear_bit(Faulty, &rdev->flags);
1090 	clear_bit(In_sync, &rdev->flags);
1091 	clear_bit(Bitmap_sync, &rdev->flags);
1092 	clear_bit(WriteMostly, &rdev->flags);
1093 
1094 	if (mddev->raid_disks == 0) {
1095 		mddev->major_version = 0;
1096 		mddev->minor_version = sb->minor_version;
1097 		mddev->patch_version = sb->patch_version;
1098 		mddev->external = 0;
1099 		mddev->chunk_sectors = sb->chunk_size >> 9;
1100 		mddev->ctime = sb->ctime;
1101 		mddev->utime = sb->utime;
1102 		mddev->level = sb->level;
1103 		mddev->clevel[0] = 0;
1104 		mddev->layout = sb->layout;
1105 		mddev->raid_disks = sb->raid_disks;
1106 		mddev->dev_sectors = ((sector_t)sb->size) * 2;
1107 		mddev->events = ev1;
1108 		mddev->bitmap_info.offset = 0;
1109 		mddev->bitmap_info.space = 0;
1110 		/* bitmap can use 60 K after the 4K superblocks */
1111 		mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9;
1112 		mddev->bitmap_info.default_space = 64*2 - (MD_SB_BYTES >> 9);
1113 		mddev->reshape_backwards = 0;
1114 
1115 		if (mddev->minor_version >= 91) {
1116 			mddev->reshape_position = sb->reshape_position;
1117 			mddev->delta_disks = sb->delta_disks;
1118 			mddev->new_level = sb->new_level;
1119 			mddev->new_layout = sb->new_layout;
1120 			mddev->new_chunk_sectors = sb->new_chunk >> 9;
1121 			if (mddev->delta_disks < 0)
1122 				mddev->reshape_backwards = 1;
1123 		} else {
1124 			mddev->reshape_position = MaxSector;
1125 			mddev->delta_disks = 0;
1126 			mddev->new_level = mddev->level;
1127 			mddev->new_layout = mddev->layout;
1128 			mddev->new_chunk_sectors = mddev->chunk_sectors;
1129 		}
1130 
1131 		if (sb->state & (1<<MD_SB_CLEAN))
1132 			mddev->recovery_cp = MaxSector;
1133 		else {
1134 			if (sb->events_hi == sb->cp_events_hi &&
1135 				sb->events_lo == sb->cp_events_lo) {
1136 				mddev->recovery_cp = sb->recovery_cp;
1137 			} else
1138 				mddev->recovery_cp = 0;
1139 		}
1140 
1141 		memcpy(mddev->uuid+0, &sb->set_uuid0, 4);
1142 		memcpy(mddev->uuid+4, &sb->set_uuid1, 4);
1143 		memcpy(mddev->uuid+8, &sb->set_uuid2, 4);
1144 		memcpy(mddev->uuid+12,&sb->set_uuid3, 4);
1145 
1146 		mddev->max_disks = MD_SB_DISKS;
1147 
1148 		if (sb->state & (1<<MD_SB_BITMAP_PRESENT) &&
1149 		    mddev->bitmap_info.file == NULL) {
1150 			mddev->bitmap_info.offset =
1151 				mddev->bitmap_info.default_offset;
1152 			mddev->bitmap_info.space =
1153 				mddev->bitmap_info.default_space;
1154 		}
1155 
1156 	} else if (mddev->pers == NULL) {
1157 		/* Insist on good event counter while assembling, except
1158 		 * for spares (which don't need an event count) */
1159 		++ev1;
1160 		if (sb->disks[rdev->desc_nr].state & (
1161 			    (1<<MD_DISK_SYNC) | (1 << MD_DISK_ACTIVE)))
1162 			if (ev1 < mddev->events)
1163 				return -EINVAL;
1164 	} else if (mddev->bitmap) {
1165 		/* if adding to array with a bitmap, then we can accept an
1166 		 * older device ... but not too old.
1167 		 */
1168 		if (ev1 < mddev->bitmap->events_cleared)
1169 			return 0;
1170 		if (ev1 < mddev->events)
1171 			set_bit(Bitmap_sync, &rdev->flags);
1172 	} else {
1173 		if (ev1 < mddev->events)
1174 			/* just a hot-add of a new device, leave raid_disk at -1 */
1175 			return 0;
1176 	}
1177 
1178 	if (mddev->level != LEVEL_MULTIPATH) {
1179 		desc = sb->disks + rdev->desc_nr;
1180 
1181 		if (desc->state & (1<<MD_DISK_FAULTY))
1182 			set_bit(Faulty, &rdev->flags);
1183 		else if (desc->state & (1<<MD_DISK_SYNC) /* &&
1184 			    desc->raid_disk < mddev->raid_disks */) {
1185 			set_bit(In_sync, &rdev->flags);
1186 			rdev->raid_disk = desc->raid_disk;
1187 			rdev->saved_raid_disk = desc->raid_disk;
1188 		} else if (desc->state & (1<<MD_DISK_ACTIVE)) {
1189 			/* active but not in sync implies recovery up to
1190 			 * reshape position.  We don't know exactly where
1191 			 * that is, so set to zero for now */
1192 			if (mddev->minor_version >= 91) {
1193 				rdev->recovery_offset = 0;
1194 				rdev->raid_disk = desc->raid_disk;
1195 			}
1196 		}
1197 		if (desc->state & (1<<MD_DISK_WRITEMOSTLY))
1198 			set_bit(WriteMostly, &rdev->flags);
1199 		if (desc->state & (1<<MD_DISK_FAILFAST))
1200 			set_bit(FailFast, &rdev->flags);
1201 	} else /* MULTIPATH are always insync */
1202 		set_bit(In_sync, &rdev->flags);
1203 	return 0;
1204 }
1205 
1206 /*
1207  * sync_super for 0.90.0
1208  */
1209 static void super_90_sync(struct mddev *mddev, struct md_rdev *rdev)
1210 {
1211 	mdp_super_t *sb;
1212 	struct md_rdev *rdev2;
1213 	int next_spare = mddev->raid_disks;
1214 
1215 	/* make rdev->sb match mddev data..
1216 	 *
1217 	 * 1/ zero out disks
1218 	 * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
1219 	 * 3/ any empty disks < next_spare become removed
1220 	 *
1221 	 * disks[0] gets initialised to REMOVED because
1222 	 * we cannot be sure from other fields if it has
1223 	 * been initialised or not.
1224 	 */
1225 	int i;
1226 	int active=0, working=0,failed=0,spare=0,nr_disks=0;
1227 
1228 	rdev->sb_size = MD_SB_BYTES;
1229 
1230 	sb = page_address(rdev->sb_page);
1231 
1232 	memset(sb, 0, sizeof(*sb));
1233 
1234 	sb->md_magic = MD_SB_MAGIC;
1235 	sb->major_version = mddev->major_version;
1236 	sb->patch_version = mddev->patch_version;
1237 	sb->gvalid_words  = 0; /* ignored */
1238 	memcpy(&sb->set_uuid0, mddev->uuid+0, 4);
1239 	memcpy(&sb->set_uuid1, mddev->uuid+4, 4);
1240 	memcpy(&sb->set_uuid2, mddev->uuid+8, 4);
1241 	memcpy(&sb->set_uuid3, mddev->uuid+12,4);
1242 
1243 	sb->ctime = clamp_t(time64_t, mddev->ctime, 0, U32_MAX);
1244 	sb->level = mddev->level;
1245 	sb->size = mddev->dev_sectors / 2;
1246 	sb->raid_disks = mddev->raid_disks;
1247 	sb->md_minor = mddev->md_minor;
1248 	sb->not_persistent = 0;
1249 	sb->utime = clamp_t(time64_t, mddev->utime, 0, U32_MAX);
1250 	sb->state = 0;
1251 	sb->events_hi = (mddev->events>>32);
1252 	sb->events_lo = (u32)mddev->events;
1253 
1254 	if (mddev->reshape_position == MaxSector)
1255 		sb->minor_version = 90;
1256 	else {
1257 		sb->minor_version = 91;
1258 		sb->reshape_position = mddev->reshape_position;
1259 		sb->new_level = mddev->new_level;
1260 		sb->delta_disks = mddev->delta_disks;
1261 		sb->new_layout = mddev->new_layout;
1262 		sb->new_chunk = mddev->new_chunk_sectors << 9;
1263 	}
1264 	mddev->minor_version = sb->minor_version;
1265 	if (mddev->in_sync)
1266 	{
1267 		sb->recovery_cp = mddev->recovery_cp;
1268 		sb->cp_events_hi = (mddev->events>>32);
1269 		sb->cp_events_lo = (u32)mddev->events;
1270 		if (mddev->recovery_cp == MaxSector)
1271 			sb->state = (1<< MD_SB_CLEAN);
1272 	} else
1273 		sb->recovery_cp = 0;
1274 
1275 	sb->layout = mddev->layout;
1276 	sb->chunk_size = mddev->chunk_sectors << 9;
1277 
1278 	if (mddev->bitmap && mddev->bitmap_info.file == NULL)
1279 		sb->state |= (1<<MD_SB_BITMAP_PRESENT);
1280 
1281 	sb->disks[0].state = (1<<MD_DISK_REMOVED);
1282 	rdev_for_each(rdev2, mddev) {
1283 		mdp_disk_t *d;
1284 		int desc_nr;
1285 		int is_active = test_bit(In_sync, &rdev2->flags);
1286 
1287 		if (rdev2->raid_disk >= 0 &&
1288 		    sb->minor_version >= 91)
1289 			/* we have nowhere to store the recovery_offset,
1290 			 * but if it is not below the reshape_position,
1291 			 * we can piggy-back on that.
1292 			 */
1293 			is_active = 1;
1294 		if (rdev2->raid_disk < 0 ||
1295 		    test_bit(Faulty, &rdev2->flags))
1296 			is_active = 0;
1297 		if (is_active)
1298 			desc_nr = rdev2->raid_disk;
1299 		else
1300 			desc_nr = next_spare++;
1301 		rdev2->desc_nr = desc_nr;
1302 		d = &sb->disks[rdev2->desc_nr];
1303 		nr_disks++;
1304 		d->number = rdev2->desc_nr;
1305 		d->major = MAJOR(rdev2->bdev->bd_dev);
1306 		d->minor = MINOR(rdev2->bdev->bd_dev);
1307 		if (is_active)
1308 			d->raid_disk = rdev2->raid_disk;
1309 		else
1310 			d->raid_disk = rdev2->desc_nr; /* compatibility */
1311 		if (test_bit(Faulty, &rdev2->flags))
1312 			d->state = (1<<MD_DISK_FAULTY);
1313 		else if (is_active) {
1314 			d->state = (1<<MD_DISK_ACTIVE);
1315 			if (test_bit(In_sync, &rdev2->flags))
1316 				d->state |= (1<<MD_DISK_SYNC);
1317 			active++;
1318 			working++;
1319 		} else {
1320 			d->state = 0;
1321 			spare++;
1322 			working++;
1323 		}
1324 		if (test_bit(WriteMostly, &rdev2->flags))
1325 			d->state |= (1<<MD_DISK_WRITEMOSTLY);
1326 		if (test_bit(FailFast, &rdev2->flags))
1327 			d->state |= (1<<MD_DISK_FAILFAST);
1328 	}
1329 	/* now set the "removed" and "faulty" bits on any missing devices */
1330 	for (i=0 ; i < mddev->raid_disks ; i++) {
1331 		mdp_disk_t *d = &sb->disks[i];
1332 		if (d->state == 0 && d->number == 0) {
1333 			d->number = i;
1334 			d->raid_disk = i;
1335 			d->state = (1<<MD_DISK_REMOVED);
1336 			d->state |= (1<<MD_DISK_FAULTY);
1337 			failed++;
1338 		}
1339 	}
1340 	sb->nr_disks = nr_disks;
1341 	sb->active_disks = active;
1342 	sb->working_disks = working;
1343 	sb->failed_disks = failed;
1344 	sb->spare_disks = spare;
1345 
1346 	sb->this_disk = sb->disks[rdev->desc_nr];
1347 	sb->sb_csum = calc_sb_csum(sb);
1348 }
1349 
1350 /*
1351  * rdev_size_change for 0.90.0
1352  */
1353 static unsigned long long
1354 super_90_rdev_size_change(struct md_rdev *rdev, sector_t num_sectors)
1355 {
1356 	if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
1357 		return 0; /* component must fit device */
1358 	if (rdev->mddev->bitmap_info.offset)
1359 		return 0; /* can't move bitmap */
1360 	rdev->sb_start = calc_dev_sboffset(rdev);
1361 	if (!num_sectors || num_sectors > rdev->sb_start)
1362 		num_sectors = rdev->sb_start;
1363 	/* Limit to 4TB as metadata cannot record more than that.
1364 	 * 4TB == 2^32 KB, or 2*2^32 sectors.
1365 	 */
1366 	if (IS_ENABLED(CONFIG_LBDAF) && (u64)num_sectors >= (2ULL << 32) &&
1367 	    rdev->mddev->level >= 1)
1368 		num_sectors = (sector_t)(2ULL << 32) - 2;
1369 	do {
1370 		md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
1371 		       rdev->sb_page);
1372 	} while (md_super_wait(rdev->mddev) < 0);
1373 	return num_sectors;
1374 }
1375 
1376 static int
1377 super_90_allow_new_offset(struct md_rdev *rdev, unsigned long long new_offset)
1378 {
1379 	/* non-zero offset changes not possible with v0.90 */
1380 	return new_offset == 0;
1381 }
1382 
1383 /*
1384  * version 1 superblock
1385  */
1386 
1387 static __le32 calc_sb_1_csum(struct mdp_superblock_1 *sb)
1388 {
1389 	__le32 disk_csum;
1390 	u32 csum;
1391 	unsigned long long newcsum;
1392 	int size = 256 + le32_to_cpu(sb->max_dev)*2;
1393 	__le32 *isuper = (__le32*)sb;
1394 
1395 	disk_csum = sb->sb_csum;
1396 	sb->sb_csum = 0;
1397 	newcsum = 0;
1398 	for (; size >= 4; size -= 4)
1399 		newcsum += le32_to_cpu(*isuper++);
1400 
1401 	if (size == 2)
1402 		newcsum += le16_to_cpu(*(__le16*) isuper);
1403 
1404 	csum = (newcsum & 0xffffffff) + (newcsum >> 32);
1405 	sb->sb_csum = disk_csum;
1406 	return cpu_to_le32(csum);
1407 }
1408 
1409 static int super_1_load(struct md_rdev *rdev, struct md_rdev *refdev, int minor_version)
1410 {
1411 	struct mdp_superblock_1 *sb;
1412 	int ret;
1413 	sector_t sb_start;
1414 	sector_t sectors;
1415 	char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
1416 	int bmask;
1417 
1418 	/*
1419 	 * Calculate the position of the superblock in 512byte sectors.
1420 	 * It is always aligned to a 4K boundary and
1421 	 * depeding on minor_version, it can be:
1422 	 * 0: At least 8K, but less than 12K, from end of device
1423 	 * 1: At start of device
1424 	 * 2: 4K from start of device.
1425 	 */
1426 	switch(minor_version) {
1427 	case 0:
1428 		sb_start = i_size_read(rdev->bdev->bd_inode) >> 9;
1429 		sb_start -= 8*2;
1430 		sb_start &= ~(sector_t)(4*2-1);
1431 		break;
1432 	case 1:
1433 		sb_start = 0;
1434 		break;
1435 	case 2:
1436 		sb_start = 8;
1437 		break;
1438 	default:
1439 		return -EINVAL;
1440 	}
1441 	rdev->sb_start = sb_start;
1442 
1443 	/* superblock is rarely larger than 1K, but it can be larger,
1444 	 * and it is safe to read 4k, so we do that
1445 	 */
1446 	ret = read_disk_sb(rdev, 4096);
1447 	if (ret) return ret;
1448 
1449 	sb = page_address(rdev->sb_page);
1450 
1451 	if (sb->magic != cpu_to_le32(MD_SB_MAGIC) ||
1452 	    sb->major_version != cpu_to_le32(1) ||
1453 	    le32_to_cpu(sb->max_dev) > (4096-256)/2 ||
1454 	    le64_to_cpu(sb->super_offset) != rdev->sb_start ||
1455 	    (le32_to_cpu(sb->feature_map) & ~MD_FEATURE_ALL) != 0)
1456 		return -EINVAL;
1457 
1458 	if (calc_sb_1_csum(sb) != sb->sb_csum) {
1459 		pr_warn("md: invalid superblock checksum on %s\n",
1460 			bdevname(rdev->bdev,b));
1461 		return -EINVAL;
1462 	}
1463 	if (le64_to_cpu(sb->data_size) < 10) {
1464 		pr_warn("md: data_size too small on %s\n",
1465 			bdevname(rdev->bdev,b));
1466 		return -EINVAL;
1467 	}
1468 	if (sb->pad0 ||
1469 	    sb->pad3[0] ||
1470 	    memcmp(sb->pad3, sb->pad3+1, sizeof(sb->pad3) - sizeof(sb->pad3[1])))
1471 		/* Some padding is non-zero, might be a new feature */
1472 		return -EINVAL;
1473 
1474 	rdev->preferred_minor = 0xffff;
1475 	rdev->data_offset = le64_to_cpu(sb->data_offset);
1476 	rdev->new_data_offset = rdev->data_offset;
1477 	if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE) &&
1478 	    (le32_to_cpu(sb->feature_map) & MD_FEATURE_NEW_OFFSET))
1479 		rdev->new_data_offset += (s32)le32_to_cpu(sb->new_offset);
1480 	atomic_set(&rdev->corrected_errors, le32_to_cpu(sb->cnt_corrected_read));
1481 
1482 	rdev->sb_size = le32_to_cpu(sb->max_dev) * 2 + 256;
1483 	bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
1484 	if (rdev->sb_size & bmask)
1485 		rdev->sb_size = (rdev->sb_size | bmask) + 1;
1486 
1487 	if (minor_version
1488 	    && rdev->data_offset < sb_start + (rdev->sb_size/512))
1489 		return -EINVAL;
1490 	if (minor_version
1491 	    && rdev->new_data_offset < sb_start + (rdev->sb_size/512))
1492 		return -EINVAL;
1493 
1494 	if (sb->level == cpu_to_le32(LEVEL_MULTIPATH))
1495 		rdev->desc_nr = -1;
1496 	else
1497 		rdev->desc_nr = le32_to_cpu(sb->dev_number);
1498 
1499 	if (!rdev->bb_page) {
1500 		rdev->bb_page = alloc_page(GFP_KERNEL);
1501 		if (!rdev->bb_page)
1502 			return -ENOMEM;
1503 	}
1504 	if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BAD_BLOCKS) &&
1505 	    rdev->badblocks.count == 0) {
1506 		/* need to load the bad block list.
1507 		 * Currently we limit it to one page.
1508 		 */
1509 		s32 offset;
1510 		sector_t bb_sector;
1511 		u64 *bbp;
1512 		int i;
1513 		int sectors = le16_to_cpu(sb->bblog_size);
1514 		if (sectors > (PAGE_SIZE / 512))
1515 			return -EINVAL;
1516 		offset = le32_to_cpu(sb->bblog_offset);
1517 		if (offset == 0)
1518 			return -EINVAL;
1519 		bb_sector = (long long)offset;
1520 		if (!sync_page_io(rdev, bb_sector, sectors << 9,
1521 				  rdev->bb_page, REQ_OP_READ, 0, true))
1522 			return -EIO;
1523 		bbp = (u64 *)page_address(rdev->bb_page);
1524 		rdev->badblocks.shift = sb->bblog_shift;
1525 		for (i = 0 ; i < (sectors << (9-3)) ; i++, bbp++) {
1526 			u64 bb = le64_to_cpu(*bbp);
1527 			int count = bb & (0x3ff);
1528 			u64 sector = bb >> 10;
1529 			sector <<= sb->bblog_shift;
1530 			count <<= sb->bblog_shift;
1531 			if (bb + 1 == 0)
1532 				break;
1533 			if (badblocks_set(&rdev->badblocks, sector, count, 1))
1534 				return -EINVAL;
1535 		}
1536 	} else if (sb->bblog_offset != 0)
1537 		rdev->badblocks.shift = 0;
1538 
1539 	if (le32_to_cpu(sb->feature_map) & MD_FEATURE_PPL) {
1540 		rdev->ppl.offset = (__s16)le16_to_cpu(sb->ppl.offset);
1541 		rdev->ppl.size = le16_to_cpu(sb->ppl.size);
1542 		rdev->ppl.sector = rdev->sb_start + rdev->ppl.offset;
1543 	}
1544 
1545 	if (!refdev) {
1546 		ret = 1;
1547 	} else {
1548 		__u64 ev1, ev2;
1549 		struct mdp_superblock_1 *refsb = page_address(refdev->sb_page);
1550 
1551 		if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 ||
1552 		    sb->level != refsb->level ||
1553 		    sb->layout != refsb->layout ||
1554 		    sb->chunksize != refsb->chunksize) {
1555 			pr_warn("md: %s has strangely different superblock to %s\n",
1556 				bdevname(rdev->bdev,b),
1557 				bdevname(refdev->bdev,b2));
1558 			return -EINVAL;
1559 		}
1560 		ev1 = le64_to_cpu(sb->events);
1561 		ev2 = le64_to_cpu(refsb->events);
1562 
1563 		if (ev1 > ev2)
1564 			ret = 1;
1565 		else
1566 			ret = 0;
1567 	}
1568 	if (minor_version) {
1569 		sectors = (i_size_read(rdev->bdev->bd_inode) >> 9);
1570 		sectors -= rdev->data_offset;
1571 	} else
1572 		sectors = rdev->sb_start;
1573 	if (sectors < le64_to_cpu(sb->data_size))
1574 		return -EINVAL;
1575 	rdev->sectors = le64_to_cpu(sb->data_size);
1576 	return ret;
1577 }
1578 
1579 static int super_1_validate(struct mddev *mddev, struct md_rdev *rdev)
1580 {
1581 	struct mdp_superblock_1 *sb = page_address(rdev->sb_page);
1582 	__u64 ev1 = le64_to_cpu(sb->events);
1583 
1584 	rdev->raid_disk = -1;
1585 	clear_bit(Faulty, &rdev->flags);
1586 	clear_bit(In_sync, &rdev->flags);
1587 	clear_bit(Bitmap_sync, &rdev->flags);
1588 	clear_bit(WriteMostly, &rdev->flags);
1589 
1590 	if (mddev->raid_disks == 0) {
1591 		mddev->major_version = 1;
1592 		mddev->patch_version = 0;
1593 		mddev->external = 0;
1594 		mddev->chunk_sectors = le32_to_cpu(sb->chunksize);
1595 		mddev->ctime = le64_to_cpu(sb->ctime);
1596 		mddev->utime = le64_to_cpu(sb->utime);
1597 		mddev->level = le32_to_cpu(sb->level);
1598 		mddev->clevel[0] = 0;
1599 		mddev->layout = le32_to_cpu(sb->layout);
1600 		mddev->raid_disks = le32_to_cpu(sb->raid_disks);
1601 		mddev->dev_sectors = le64_to_cpu(sb->size);
1602 		mddev->events = ev1;
1603 		mddev->bitmap_info.offset = 0;
1604 		mddev->bitmap_info.space = 0;
1605 		/* Default location for bitmap is 1K after superblock
1606 		 * using 3K - total of 4K
1607 		 */
1608 		mddev->bitmap_info.default_offset = 1024 >> 9;
1609 		mddev->bitmap_info.default_space = (4096-1024) >> 9;
1610 		mddev->reshape_backwards = 0;
1611 
1612 		mddev->recovery_cp = le64_to_cpu(sb->resync_offset);
1613 		memcpy(mddev->uuid, sb->set_uuid, 16);
1614 
1615 		mddev->max_disks =  (4096-256)/2;
1616 
1617 		if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) &&
1618 		    mddev->bitmap_info.file == NULL) {
1619 			mddev->bitmap_info.offset =
1620 				(__s32)le32_to_cpu(sb->bitmap_offset);
1621 			/* Metadata doesn't record how much space is available.
1622 			 * For 1.0, we assume we can use up to the superblock
1623 			 * if before, else to 4K beyond superblock.
1624 			 * For others, assume no change is possible.
1625 			 */
1626 			if (mddev->minor_version > 0)
1627 				mddev->bitmap_info.space = 0;
1628 			else if (mddev->bitmap_info.offset > 0)
1629 				mddev->bitmap_info.space =
1630 					8 - mddev->bitmap_info.offset;
1631 			else
1632 				mddev->bitmap_info.space =
1633 					-mddev->bitmap_info.offset;
1634 		}
1635 
1636 		if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) {
1637 			mddev->reshape_position = le64_to_cpu(sb->reshape_position);
1638 			mddev->delta_disks = le32_to_cpu(sb->delta_disks);
1639 			mddev->new_level = le32_to_cpu(sb->new_level);
1640 			mddev->new_layout = le32_to_cpu(sb->new_layout);
1641 			mddev->new_chunk_sectors = le32_to_cpu(sb->new_chunk);
1642 			if (mddev->delta_disks < 0 ||
1643 			    (mddev->delta_disks == 0 &&
1644 			     (le32_to_cpu(sb->feature_map)
1645 			      & MD_FEATURE_RESHAPE_BACKWARDS)))
1646 				mddev->reshape_backwards = 1;
1647 		} else {
1648 			mddev->reshape_position = MaxSector;
1649 			mddev->delta_disks = 0;
1650 			mddev->new_level = mddev->level;
1651 			mddev->new_layout = mddev->layout;
1652 			mddev->new_chunk_sectors = mddev->chunk_sectors;
1653 		}
1654 
1655 		if (le32_to_cpu(sb->feature_map) & MD_FEATURE_JOURNAL)
1656 			set_bit(MD_HAS_JOURNAL, &mddev->flags);
1657 
1658 		if (le32_to_cpu(sb->feature_map) & MD_FEATURE_PPL) {
1659 			if (le32_to_cpu(sb->feature_map) &
1660 			    (MD_FEATURE_BITMAP_OFFSET | MD_FEATURE_JOURNAL))
1661 				return -EINVAL;
1662 			set_bit(MD_HAS_PPL, &mddev->flags);
1663 		}
1664 	} else if (mddev->pers == NULL) {
1665 		/* Insist of good event counter while assembling, except for
1666 		 * spares (which don't need an event count) */
1667 		++ev1;
1668 		if (rdev->desc_nr >= 0 &&
1669 		    rdev->desc_nr < le32_to_cpu(sb->max_dev) &&
1670 		    (le16_to_cpu(sb->dev_roles[rdev->desc_nr]) < MD_DISK_ROLE_MAX ||
1671 		     le16_to_cpu(sb->dev_roles[rdev->desc_nr]) == MD_DISK_ROLE_JOURNAL))
1672 			if (ev1 < mddev->events)
1673 				return -EINVAL;
1674 	} else if (mddev->bitmap) {
1675 		/* If adding to array with a bitmap, then we can accept an
1676 		 * older device, but not too old.
1677 		 */
1678 		if (ev1 < mddev->bitmap->events_cleared)
1679 			return 0;
1680 		if (ev1 < mddev->events)
1681 			set_bit(Bitmap_sync, &rdev->flags);
1682 	} else {
1683 		if (ev1 < mddev->events)
1684 			/* just a hot-add of a new device, leave raid_disk at -1 */
1685 			return 0;
1686 	}
1687 	if (mddev->level != LEVEL_MULTIPATH) {
1688 		int role;
1689 		if (rdev->desc_nr < 0 ||
1690 		    rdev->desc_nr >= le32_to_cpu(sb->max_dev)) {
1691 			role = MD_DISK_ROLE_SPARE;
1692 			rdev->desc_nr = -1;
1693 		} else
1694 			role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
1695 		switch(role) {
1696 		case MD_DISK_ROLE_SPARE: /* spare */
1697 			break;
1698 		case MD_DISK_ROLE_FAULTY: /* faulty */
1699 			set_bit(Faulty, &rdev->flags);
1700 			break;
1701 		case MD_DISK_ROLE_JOURNAL: /* journal device */
1702 			if (!(le32_to_cpu(sb->feature_map) & MD_FEATURE_JOURNAL)) {
1703 				/* journal device without journal feature */
1704 				pr_warn("md: journal device provided without journal feature, ignoring the device\n");
1705 				return -EINVAL;
1706 			}
1707 			set_bit(Journal, &rdev->flags);
1708 			rdev->journal_tail = le64_to_cpu(sb->journal_tail);
1709 			rdev->raid_disk = 0;
1710 			break;
1711 		default:
1712 			rdev->saved_raid_disk = role;
1713 			if ((le32_to_cpu(sb->feature_map) &
1714 			     MD_FEATURE_RECOVERY_OFFSET)) {
1715 				rdev->recovery_offset = le64_to_cpu(sb->recovery_offset);
1716 				if (!(le32_to_cpu(sb->feature_map) &
1717 				      MD_FEATURE_RECOVERY_BITMAP))
1718 					rdev->saved_raid_disk = -1;
1719 			} else
1720 				set_bit(In_sync, &rdev->flags);
1721 			rdev->raid_disk = role;
1722 			break;
1723 		}
1724 		if (sb->devflags & WriteMostly1)
1725 			set_bit(WriteMostly, &rdev->flags);
1726 		if (sb->devflags & FailFast1)
1727 			set_bit(FailFast, &rdev->flags);
1728 		if (le32_to_cpu(sb->feature_map) & MD_FEATURE_REPLACEMENT)
1729 			set_bit(Replacement, &rdev->flags);
1730 	} else /* MULTIPATH are always insync */
1731 		set_bit(In_sync, &rdev->flags);
1732 
1733 	return 0;
1734 }
1735 
1736 static void super_1_sync(struct mddev *mddev, struct md_rdev *rdev)
1737 {
1738 	struct mdp_superblock_1 *sb;
1739 	struct md_rdev *rdev2;
1740 	int max_dev, i;
1741 	/* make rdev->sb match mddev and rdev data. */
1742 
1743 	sb = page_address(rdev->sb_page);
1744 
1745 	sb->feature_map = 0;
1746 	sb->pad0 = 0;
1747 	sb->recovery_offset = cpu_to_le64(0);
1748 	memset(sb->pad3, 0, sizeof(sb->pad3));
1749 
1750 	sb->utime = cpu_to_le64((__u64)mddev->utime);
1751 	sb->events = cpu_to_le64(mddev->events);
1752 	if (mddev->in_sync)
1753 		sb->resync_offset = cpu_to_le64(mddev->recovery_cp);
1754 	else if (test_bit(MD_JOURNAL_CLEAN, &mddev->flags))
1755 		sb->resync_offset = cpu_to_le64(MaxSector);
1756 	else
1757 		sb->resync_offset = cpu_to_le64(0);
1758 
1759 	sb->cnt_corrected_read = cpu_to_le32(atomic_read(&rdev->corrected_errors));
1760 
1761 	sb->raid_disks = cpu_to_le32(mddev->raid_disks);
1762 	sb->size = cpu_to_le64(mddev->dev_sectors);
1763 	sb->chunksize = cpu_to_le32(mddev->chunk_sectors);
1764 	sb->level = cpu_to_le32(mddev->level);
1765 	sb->layout = cpu_to_le32(mddev->layout);
1766 	if (test_bit(FailFast, &rdev->flags))
1767 		sb->devflags |= FailFast1;
1768 	else
1769 		sb->devflags &= ~FailFast1;
1770 
1771 	if (test_bit(WriteMostly, &rdev->flags))
1772 		sb->devflags |= WriteMostly1;
1773 	else
1774 		sb->devflags &= ~WriteMostly1;
1775 	sb->data_offset = cpu_to_le64(rdev->data_offset);
1776 	sb->data_size = cpu_to_le64(rdev->sectors);
1777 
1778 	if (mddev->bitmap && mddev->bitmap_info.file == NULL) {
1779 		sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_info.offset);
1780 		sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET);
1781 	}
1782 
1783 	if (rdev->raid_disk >= 0 && !test_bit(Journal, &rdev->flags) &&
1784 	    !test_bit(In_sync, &rdev->flags)) {
1785 		sb->feature_map |=
1786 			cpu_to_le32(MD_FEATURE_RECOVERY_OFFSET);
1787 		sb->recovery_offset =
1788 			cpu_to_le64(rdev->recovery_offset);
1789 		if (rdev->saved_raid_disk >= 0 && mddev->bitmap)
1790 			sb->feature_map |=
1791 				cpu_to_le32(MD_FEATURE_RECOVERY_BITMAP);
1792 	}
1793 	/* Note: recovery_offset and journal_tail share space  */
1794 	if (test_bit(Journal, &rdev->flags))
1795 		sb->journal_tail = cpu_to_le64(rdev->journal_tail);
1796 	if (test_bit(Replacement, &rdev->flags))
1797 		sb->feature_map |=
1798 			cpu_to_le32(MD_FEATURE_REPLACEMENT);
1799 
1800 	if (mddev->reshape_position != MaxSector) {
1801 		sb->feature_map |= cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE);
1802 		sb->reshape_position = cpu_to_le64(mddev->reshape_position);
1803 		sb->new_layout = cpu_to_le32(mddev->new_layout);
1804 		sb->delta_disks = cpu_to_le32(mddev->delta_disks);
1805 		sb->new_level = cpu_to_le32(mddev->new_level);
1806 		sb->new_chunk = cpu_to_le32(mddev->new_chunk_sectors);
1807 		if (mddev->delta_disks == 0 &&
1808 		    mddev->reshape_backwards)
1809 			sb->feature_map
1810 				|= cpu_to_le32(MD_FEATURE_RESHAPE_BACKWARDS);
1811 		if (rdev->new_data_offset != rdev->data_offset) {
1812 			sb->feature_map
1813 				|= cpu_to_le32(MD_FEATURE_NEW_OFFSET);
1814 			sb->new_offset = cpu_to_le32((__u32)(rdev->new_data_offset
1815 							     - rdev->data_offset));
1816 		}
1817 	}
1818 
1819 	if (mddev_is_clustered(mddev))
1820 		sb->feature_map |= cpu_to_le32(MD_FEATURE_CLUSTERED);
1821 
1822 	if (rdev->badblocks.count == 0)
1823 		/* Nothing to do for bad blocks*/ ;
1824 	else if (sb->bblog_offset == 0)
1825 		/* Cannot record bad blocks on this device */
1826 		md_error(mddev, rdev);
1827 	else {
1828 		struct badblocks *bb = &rdev->badblocks;
1829 		u64 *bbp = (u64 *)page_address(rdev->bb_page);
1830 		u64 *p = bb->page;
1831 		sb->feature_map |= cpu_to_le32(MD_FEATURE_BAD_BLOCKS);
1832 		if (bb->changed) {
1833 			unsigned seq;
1834 
1835 retry:
1836 			seq = read_seqbegin(&bb->lock);
1837 
1838 			memset(bbp, 0xff, PAGE_SIZE);
1839 
1840 			for (i = 0 ; i < bb->count ; i++) {
1841 				u64 internal_bb = p[i];
1842 				u64 store_bb = ((BB_OFFSET(internal_bb) << 10)
1843 						| BB_LEN(internal_bb));
1844 				bbp[i] = cpu_to_le64(store_bb);
1845 			}
1846 			bb->changed = 0;
1847 			if (read_seqretry(&bb->lock, seq))
1848 				goto retry;
1849 
1850 			bb->sector = (rdev->sb_start +
1851 				      (int)le32_to_cpu(sb->bblog_offset));
1852 			bb->size = le16_to_cpu(sb->bblog_size);
1853 		}
1854 	}
1855 
1856 	max_dev = 0;
1857 	rdev_for_each(rdev2, mddev)
1858 		if (rdev2->desc_nr+1 > max_dev)
1859 			max_dev = rdev2->desc_nr+1;
1860 
1861 	if (max_dev > le32_to_cpu(sb->max_dev)) {
1862 		int bmask;
1863 		sb->max_dev = cpu_to_le32(max_dev);
1864 		rdev->sb_size = max_dev * 2 + 256;
1865 		bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
1866 		if (rdev->sb_size & bmask)
1867 			rdev->sb_size = (rdev->sb_size | bmask) + 1;
1868 	} else
1869 		max_dev = le32_to_cpu(sb->max_dev);
1870 
1871 	for (i=0; i<max_dev;i++)
1872 		sb->dev_roles[i] = cpu_to_le16(MD_DISK_ROLE_SPARE);
1873 
1874 	if (test_bit(MD_HAS_JOURNAL, &mddev->flags))
1875 		sb->feature_map |= cpu_to_le32(MD_FEATURE_JOURNAL);
1876 
1877 	if (test_bit(MD_HAS_PPL, &mddev->flags)) {
1878 		sb->feature_map |= cpu_to_le32(MD_FEATURE_PPL);
1879 		sb->ppl.offset = cpu_to_le16(rdev->ppl.offset);
1880 		sb->ppl.size = cpu_to_le16(rdev->ppl.size);
1881 	}
1882 
1883 	rdev_for_each(rdev2, mddev) {
1884 		i = rdev2->desc_nr;
1885 		if (test_bit(Faulty, &rdev2->flags))
1886 			sb->dev_roles[i] = cpu_to_le16(MD_DISK_ROLE_FAULTY);
1887 		else if (test_bit(In_sync, &rdev2->flags))
1888 			sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1889 		else if (test_bit(Journal, &rdev2->flags))
1890 			sb->dev_roles[i] = cpu_to_le16(MD_DISK_ROLE_JOURNAL);
1891 		else if (rdev2->raid_disk >= 0)
1892 			sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1893 		else
1894 			sb->dev_roles[i] = cpu_to_le16(MD_DISK_ROLE_SPARE);
1895 	}
1896 
1897 	sb->sb_csum = calc_sb_1_csum(sb);
1898 }
1899 
1900 static unsigned long long
1901 super_1_rdev_size_change(struct md_rdev *rdev, sector_t num_sectors)
1902 {
1903 	struct mdp_superblock_1 *sb;
1904 	sector_t max_sectors;
1905 	if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
1906 		return 0; /* component must fit device */
1907 	if (rdev->data_offset != rdev->new_data_offset)
1908 		return 0; /* too confusing */
1909 	if (rdev->sb_start < rdev->data_offset) {
1910 		/* minor versions 1 and 2; superblock before data */
1911 		max_sectors = i_size_read(rdev->bdev->bd_inode) >> 9;
1912 		max_sectors -= rdev->data_offset;
1913 		if (!num_sectors || num_sectors > max_sectors)
1914 			num_sectors = max_sectors;
1915 	} else if (rdev->mddev->bitmap_info.offset) {
1916 		/* minor version 0 with bitmap we can't move */
1917 		return 0;
1918 	} else {
1919 		/* minor version 0; superblock after data */
1920 		sector_t sb_start;
1921 		sb_start = (i_size_read(rdev->bdev->bd_inode) >> 9) - 8*2;
1922 		sb_start &= ~(sector_t)(4*2 - 1);
1923 		max_sectors = rdev->sectors + sb_start - rdev->sb_start;
1924 		if (!num_sectors || num_sectors > max_sectors)
1925 			num_sectors = max_sectors;
1926 		rdev->sb_start = sb_start;
1927 	}
1928 	sb = page_address(rdev->sb_page);
1929 	sb->data_size = cpu_to_le64(num_sectors);
1930 	sb->super_offset = cpu_to_le64(rdev->sb_start);
1931 	sb->sb_csum = calc_sb_1_csum(sb);
1932 	do {
1933 		md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
1934 			       rdev->sb_page);
1935 	} while (md_super_wait(rdev->mddev) < 0);
1936 	return num_sectors;
1937 
1938 }
1939 
1940 static int
1941 super_1_allow_new_offset(struct md_rdev *rdev,
1942 			 unsigned long long new_offset)
1943 {
1944 	/* All necessary checks on new >= old have been done */
1945 	struct bitmap *bitmap;
1946 	if (new_offset >= rdev->data_offset)
1947 		return 1;
1948 
1949 	/* with 1.0 metadata, there is no metadata to tread on
1950 	 * so we can always move back */
1951 	if (rdev->mddev->minor_version == 0)
1952 		return 1;
1953 
1954 	/* otherwise we must be sure not to step on
1955 	 * any metadata, so stay:
1956 	 * 36K beyond start of superblock
1957 	 * beyond end of badblocks
1958 	 * beyond write-intent bitmap
1959 	 */
1960 	if (rdev->sb_start + (32+4)*2 > new_offset)
1961 		return 0;
1962 	bitmap = rdev->mddev->bitmap;
1963 	if (bitmap && !rdev->mddev->bitmap_info.file &&
1964 	    rdev->sb_start + rdev->mddev->bitmap_info.offset +
1965 	    bitmap->storage.file_pages * (PAGE_SIZE>>9) > new_offset)
1966 		return 0;
1967 	if (rdev->badblocks.sector + rdev->badblocks.size > new_offset)
1968 		return 0;
1969 
1970 	return 1;
1971 }
1972 
1973 static struct super_type super_types[] = {
1974 	[0] = {
1975 		.name	= "0.90.0",
1976 		.owner	= THIS_MODULE,
1977 		.load_super	    = super_90_load,
1978 		.validate_super	    = super_90_validate,
1979 		.sync_super	    = super_90_sync,
1980 		.rdev_size_change   = super_90_rdev_size_change,
1981 		.allow_new_offset   = super_90_allow_new_offset,
1982 	},
1983 	[1] = {
1984 		.name	= "md-1",
1985 		.owner	= THIS_MODULE,
1986 		.load_super	    = super_1_load,
1987 		.validate_super	    = super_1_validate,
1988 		.sync_super	    = super_1_sync,
1989 		.rdev_size_change   = super_1_rdev_size_change,
1990 		.allow_new_offset   = super_1_allow_new_offset,
1991 	},
1992 };
1993 
1994 static void sync_super(struct mddev *mddev, struct md_rdev *rdev)
1995 {
1996 	if (mddev->sync_super) {
1997 		mddev->sync_super(mddev, rdev);
1998 		return;
1999 	}
2000 
2001 	BUG_ON(mddev->major_version >= ARRAY_SIZE(super_types));
2002 
2003 	super_types[mddev->major_version].sync_super(mddev, rdev);
2004 }
2005 
2006 static int match_mddev_units(struct mddev *mddev1, struct mddev *mddev2)
2007 {
2008 	struct md_rdev *rdev, *rdev2;
2009 
2010 	rcu_read_lock();
2011 	rdev_for_each_rcu(rdev, mddev1) {
2012 		if (test_bit(Faulty, &rdev->flags) ||
2013 		    test_bit(Journal, &rdev->flags) ||
2014 		    rdev->raid_disk == -1)
2015 			continue;
2016 		rdev_for_each_rcu(rdev2, mddev2) {
2017 			if (test_bit(Faulty, &rdev2->flags) ||
2018 			    test_bit(Journal, &rdev2->flags) ||
2019 			    rdev2->raid_disk == -1)
2020 				continue;
2021 			if (rdev->bdev->bd_contains ==
2022 			    rdev2->bdev->bd_contains) {
2023 				rcu_read_unlock();
2024 				return 1;
2025 			}
2026 		}
2027 	}
2028 	rcu_read_unlock();
2029 	return 0;
2030 }
2031 
2032 static LIST_HEAD(pending_raid_disks);
2033 
2034 /*
2035  * Try to register data integrity profile for an mddev
2036  *
2037  * This is called when an array is started and after a disk has been kicked
2038  * from the array. It only succeeds if all working and active component devices
2039  * are integrity capable with matching profiles.
2040  */
2041 int md_integrity_register(struct mddev *mddev)
2042 {
2043 	struct md_rdev *rdev, *reference = NULL;
2044 
2045 	if (list_empty(&mddev->disks))
2046 		return 0; /* nothing to do */
2047 	if (!mddev->gendisk || blk_get_integrity(mddev->gendisk))
2048 		return 0; /* shouldn't register, or already is */
2049 	rdev_for_each(rdev, mddev) {
2050 		/* skip spares and non-functional disks */
2051 		if (test_bit(Faulty, &rdev->flags))
2052 			continue;
2053 		if (rdev->raid_disk < 0)
2054 			continue;
2055 		if (!reference) {
2056 			/* Use the first rdev as the reference */
2057 			reference = rdev;
2058 			continue;
2059 		}
2060 		/* does this rdev's profile match the reference profile? */
2061 		if (blk_integrity_compare(reference->bdev->bd_disk,
2062 				rdev->bdev->bd_disk) < 0)
2063 			return -EINVAL;
2064 	}
2065 	if (!reference || !bdev_get_integrity(reference->bdev))
2066 		return 0;
2067 	/*
2068 	 * All component devices are integrity capable and have matching
2069 	 * profiles, register the common profile for the md device.
2070 	 */
2071 	blk_integrity_register(mddev->gendisk,
2072 			       bdev_get_integrity(reference->bdev));
2073 
2074 	pr_debug("md: data integrity enabled on %s\n", mdname(mddev));
2075 	if (bioset_integrity_create(mddev->bio_set, BIO_POOL_SIZE)) {
2076 		pr_err("md: failed to create integrity pool for %s\n",
2077 		       mdname(mddev));
2078 		return -EINVAL;
2079 	}
2080 	return 0;
2081 }
2082 EXPORT_SYMBOL(md_integrity_register);
2083 
2084 /*
2085  * Attempt to add an rdev, but only if it is consistent with the current
2086  * integrity profile
2087  */
2088 int md_integrity_add_rdev(struct md_rdev *rdev, struct mddev *mddev)
2089 {
2090 	struct blk_integrity *bi_rdev;
2091 	struct blk_integrity *bi_mddev;
2092 	char name[BDEVNAME_SIZE];
2093 
2094 	if (!mddev->gendisk)
2095 		return 0;
2096 
2097 	bi_rdev = bdev_get_integrity(rdev->bdev);
2098 	bi_mddev = blk_get_integrity(mddev->gendisk);
2099 
2100 	if (!bi_mddev) /* nothing to do */
2101 		return 0;
2102 
2103 	if (blk_integrity_compare(mddev->gendisk, rdev->bdev->bd_disk) != 0) {
2104 		pr_err("%s: incompatible integrity profile for %s\n",
2105 		       mdname(mddev), bdevname(rdev->bdev, name));
2106 		return -ENXIO;
2107 	}
2108 
2109 	return 0;
2110 }
2111 EXPORT_SYMBOL(md_integrity_add_rdev);
2112 
2113 static int bind_rdev_to_array(struct md_rdev *rdev, struct mddev *mddev)
2114 {
2115 	char b[BDEVNAME_SIZE];
2116 	struct kobject *ko;
2117 	int err;
2118 
2119 	/* prevent duplicates */
2120 	if (find_rdev(mddev, rdev->bdev->bd_dev))
2121 		return -EEXIST;
2122 
2123 	if ((bdev_read_only(rdev->bdev) || bdev_read_only(rdev->meta_bdev)) &&
2124 	    mddev->pers)
2125 		return -EROFS;
2126 
2127 	/* make sure rdev->sectors exceeds mddev->dev_sectors */
2128 	if (!test_bit(Journal, &rdev->flags) &&
2129 	    rdev->sectors &&
2130 	    (mddev->dev_sectors == 0 || rdev->sectors < mddev->dev_sectors)) {
2131 		if (mddev->pers) {
2132 			/* Cannot change size, so fail
2133 			 * If mddev->level <= 0, then we don't care
2134 			 * about aligning sizes (e.g. linear)
2135 			 */
2136 			if (mddev->level > 0)
2137 				return -ENOSPC;
2138 		} else
2139 			mddev->dev_sectors = rdev->sectors;
2140 	}
2141 
2142 	/* Verify rdev->desc_nr is unique.
2143 	 * If it is -1, assign a free number, else
2144 	 * check number is not in use
2145 	 */
2146 	rcu_read_lock();
2147 	if (rdev->desc_nr < 0) {
2148 		int choice = 0;
2149 		if (mddev->pers)
2150 			choice = mddev->raid_disks;
2151 		while (md_find_rdev_nr_rcu(mddev, choice))
2152 			choice++;
2153 		rdev->desc_nr = choice;
2154 	} else {
2155 		if (md_find_rdev_nr_rcu(mddev, rdev->desc_nr)) {
2156 			rcu_read_unlock();
2157 			return -EBUSY;
2158 		}
2159 	}
2160 	rcu_read_unlock();
2161 	if (!test_bit(Journal, &rdev->flags) &&
2162 	    mddev->max_disks && rdev->desc_nr >= mddev->max_disks) {
2163 		pr_warn("md: %s: array is limited to %d devices\n",
2164 			mdname(mddev), mddev->max_disks);
2165 		return -EBUSY;
2166 	}
2167 	bdevname(rdev->bdev,b);
2168 	strreplace(b, '/', '!');
2169 
2170 	rdev->mddev = mddev;
2171 	pr_debug("md: bind<%s>\n", b);
2172 
2173 	if ((err = kobject_add(&rdev->kobj, &mddev->kobj, "dev-%s", b)))
2174 		goto fail;
2175 
2176 	ko = &part_to_dev(rdev->bdev->bd_part)->kobj;
2177 	if (sysfs_create_link(&rdev->kobj, ko, "block"))
2178 		/* failure here is OK */;
2179 	rdev->sysfs_state = sysfs_get_dirent_safe(rdev->kobj.sd, "state");
2180 
2181 	list_add_rcu(&rdev->same_set, &mddev->disks);
2182 	bd_link_disk_holder(rdev->bdev, mddev->gendisk);
2183 
2184 	/* May as well allow recovery to be retried once */
2185 	mddev->recovery_disabled++;
2186 
2187 	return 0;
2188 
2189  fail:
2190 	pr_warn("md: failed to register dev-%s for %s\n",
2191 		b, mdname(mddev));
2192 	return err;
2193 }
2194 
2195 static void md_delayed_delete(struct work_struct *ws)
2196 {
2197 	struct md_rdev *rdev = container_of(ws, struct md_rdev, del_work);
2198 	kobject_del(&rdev->kobj);
2199 	kobject_put(&rdev->kobj);
2200 }
2201 
2202 static void unbind_rdev_from_array(struct md_rdev *rdev)
2203 {
2204 	char b[BDEVNAME_SIZE];
2205 
2206 	bd_unlink_disk_holder(rdev->bdev, rdev->mddev->gendisk);
2207 	list_del_rcu(&rdev->same_set);
2208 	pr_debug("md: unbind<%s>\n", bdevname(rdev->bdev,b));
2209 	rdev->mddev = NULL;
2210 	sysfs_remove_link(&rdev->kobj, "block");
2211 	sysfs_put(rdev->sysfs_state);
2212 	rdev->sysfs_state = NULL;
2213 	rdev->badblocks.count = 0;
2214 	/* We need to delay this, otherwise we can deadlock when
2215 	 * writing to 'remove' to "dev/state".  We also need
2216 	 * to delay it due to rcu usage.
2217 	 */
2218 	synchronize_rcu();
2219 	INIT_WORK(&rdev->del_work, md_delayed_delete);
2220 	kobject_get(&rdev->kobj);
2221 	queue_work(md_misc_wq, &rdev->del_work);
2222 }
2223 
2224 /*
2225  * prevent the device from being mounted, repartitioned or
2226  * otherwise reused by a RAID array (or any other kernel
2227  * subsystem), by bd_claiming the device.
2228  */
2229 static int lock_rdev(struct md_rdev *rdev, dev_t dev, int shared)
2230 {
2231 	int err = 0;
2232 	struct block_device *bdev;
2233 	char b[BDEVNAME_SIZE];
2234 
2235 	bdev = blkdev_get_by_dev(dev, FMODE_READ|FMODE_WRITE|FMODE_EXCL,
2236 				 shared ? (struct md_rdev *)lock_rdev : rdev);
2237 	if (IS_ERR(bdev)) {
2238 		pr_warn("md: could not open %s.\n", __bdevname(dev, b));
2239 		return PTR_ERR(bdev);
2240 	}
2241 	rdev->bdev = bdev;
2242 	return err;
2243 }
2244 
2245 static void unlock_rdev(struct md_rdev *rdev)
2246 {
2247 	struct block_device *bdev = rdev->bdev;
2248 	rdev->bdev = NULL;
2249 	blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
2250 }
2251 
2252 void md_autodetect_dev(dev_t dev);
2253 
2254 static void export_rdev(struct md_rdev *rdev)
2255 {
2256 	char b[BDEVNAME_SIZE];
2257 
2258 	pr_debug("md: export_rdev(%s)\n", bdevname(rdev->bdev,b));
2259 	md_rdev_clear(rdev);
2260 #ifndef MODULE
2261 	if (test_bit(AutoDetected, &rdev->flags))
2262 		md_autodetect_dev(rdev->bdev->bd_dev);
2263 #endif
2264 	unlock_rdev(rdev);
2265 	kobject_put(&rdev->kobj);
2266 }
2267 
2268 void md_kick_rdev_from_array(struct md_rdev *rdev)
2269 {
2270 	unbind_rdev_from_array(rdev);
2271 	export_rdev(rdev);
2272 }
2273 EXPORT_SYMBOL_GPL(md_kick_rdev_from_array);
2274 
2275 static void export_array(struct mddev *mddev)
2276 {
2277 	struct md_rdev *rdev;
2278 
2279 	while (!list_empty(&mddev->disks)) {
2280 		rdev = list_first_entry(&mddev->disks, struct md_rdev,
2281 					same_set);
2282 		md_kick_rdev_from_array(rdev);
2283 	}
2284 	mddev->raid_disks = 0;
2285 	mddev->major_version = 0;
2286 }
2287 
2288 static bool set_in_sync(struct mddev *mddev)
2289 {
2290 	WARN_ON_ONCE(NR_CPUS != 1 && !spin_is_locked(&mddev->lock));
2291 	if (!mddev->in_sync) {
2292 		mddev->sync_checkers++;
2293 		spin_unlock(&mddev->lock);
2294 		percpu_ref_switch_to_atomic_sync(&mddev->writes_pending);
2295 		spin_lock(&mddev->lock);
2296 		if (!mddev->in_sync &&
2297 		    percpu_ref_is_zero(&mddev->writes_pending)) {
2298 			mddev->in_sync = 1;
2299 			/*
2300 			 * Ensure ->in_sync is visible before we clear
2301 			 * ->sync_checkers.
2302 			 */
2303 			smp_mb();
2304 			set_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags);
2305 			sysfs_notify_dirent_safe(mddev->sysfs_state);
2306 		}
2307 		if (--mddev->sync_checkers == 0)
2308 			percpu_ref_switch_to_percpu(&mddev->writes_pending);
2309 	}
2310 	if (mddev->safemode == 1)
2311 		mddev->safemode = 0;
2312 	return mddev->in_sync;
2313 }
2314 
2315 static void sync_sbs(struct mddev *mddev, int nospares)
2316 {
2317 	/* Update each superblock (in-memory image), but
2318 	 * if we are allowed to, skip spares which already
2319 	 * have the right event counter, or have one earlier
2320 	 * (which would mean they aren't being marked as dirty
2321 	 * with the rest of the array)
2322 	 */
2323 	struct md_rdev *rdev;
2324 	rdev_for_each(rdev, mddev) {
2325 		if (rdev->sb_events == mddev->events ||
2326 		    (nospares &&
2327 		     rdev->raid_disk < 0 &&
2328 		     rdev->sb_events+1 == mddev->events)) {
2329 			/* Don't update this superblock */
2330 			rdev->sb_loaded = 2;
2331 		} else {
2332 			sync_super(mddev, rdev);
2333 			rdev->sb_loaded = 1;
2334 		}
2335 	}
2336 }
2337 
2338 static bool does_sb_need_changing(struct mddev *mddev)
2339 {
2340 	struct md_rdev *rdev;
2341 	struct mdp_superblock_1 *sb;
2342 	int role;
2343 
2344 	/* Find a good rdev */
2345 	rdev_for_each(rdev, mddev)
2346 		if ((rdev->raid_disk >= 0) && !test_bit(Faulty, &rdev->flags))
2347 			break;
2348 
2349 	/* No good device found. */
2350 	if (!rdev)
2351 		return false;
2352 
2353 	sb = page_address(rdev->sb_page);
2354 	/* Check if a device has become faulty or a spare become active */
2355 	rdev_for_each(rdev, mddev) {
2356 		role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
2357 		/* Device activated? */
2358 		if (role == 0xffff && rdev->raid_disk >=0 &&
2359 		    !test_bit(Faulty, &rdev->flags))
2360 			return true;
2361 		/* Device turned faulty? */
2362 		if (test_bit(Faulty, &rdev->flags) && (role < 0xfffd))
2363 			return true;
2364 	}
2365 
2366 	/* Check if any mddev parameters have changed */
2367 	if ((mddev->dev_sectors != le64_to_cpu(sb->size)) ||
2368 	    (mddev->reshape_position != le64_to_cpu(sb->reshape_position)) ||
2369 	    (mddev->layout != le32_to_cpu(sb->layout)) ||
2370 	    (mddev->raid_disks != le32_to_cpu(sb->raid_disks)) ||
2371 	    (mddev->chunk_sectors != le32_to_cpu(sb->chunksize)))
2372 		return true;
2373 
2374 	return false;
2375 }
2376 
2377 void md_update_sb(struct mddev *mddev, int force_change)
2378 {
2379 	struct md_rdev *rdev;
2380 	int sync_req;
2381 	int nospares = 0;
2382 	int any_badblocks_changed = 0;
2383 	int ret = -1;
2384 
2385 	if (mddev->ro) {
2386 		if (force_change)
2387 			set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
2388 		return;
2389 	}
2390 
2391 repeat:
2392 	if (mddev_is_clustered(mddev)) {
2393 		if (test_and_clear_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags))
2394 			force_change = 1;
2395 		if (test_and_clear_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags))
2396 			nospares = 1;
2397 		ret = md_cluster_ops->metadata_update_start(mddev);
2398 		/* Has someone else has updated the sb */
2399 		if (!does_sb_need_changing(mddev)) {
2400 			if (ret == 0)
2401 				md_cluster_ops->metadata_update_cancel(mddev);
2402 			bit_clear_unless(&mddev->sb_flags, BIT(MD_SB_CHANGE_PENDING),
2403 							 BIT(MD_SB_CHANGE_DEVS) |
2404 							 BIT(MD_SB_CHANGE_CLEAN));
2405 			return;
2406 		}
2407 	}
2408 
2409 	/* First make sure individual recovery_offsets are correct */
2410 	rdev_for_each(rdev, mddev) {
2411 		if (rdev->raid_disk >= 0 &&
2412 		    mddev->delta_disks >= 0 &&
2413 		    !test_bit(Journal, &rdev->flags) &&
2414 		    !test_bit(In_sync, &rdev->flags) &&
2415 		    mddev->curr_resync_completed > rdev->recovery_offset)
2416 				rdev->recovery_offset = mddev->curr_resync_completed;
2417 
2418 	}
2419 	if (!mddev->persistent) {
2420 		clear_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags);
2421 		clear_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
2422 		if (!mddev->external) {
2423 			clear_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags);
2424 			rdev_for_each(rdev, mddev) {
2425 				if (rdev->badblocks.changed) {
2426 					rdev->badblocks.changed = 0;
2427 					ack_all_badblocks(&rdev->badblocks);
2428 					md_error(mddev, rdev);
2429 				}
2430 				clear_bit(Blocked, &rdev->flags);
2431 				clear_bit(BlockedBadBlocks, &rdev->flags);
2432 				wake_up(&rdev->blocked_wait);
2433 			}
2434 		}
2435 		wake_up(&mddev->sb_wait);
2436 		return;
2437 	}
2438 
2439 	spin_lock(&mddev->lock);
2440 
2441 	mddev->utime = ktime_get_real_seconds();
2442 
2443 	if (test_and_clear_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags))
2444 		force_change = 1;
2445 	if (test_and_clear_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags))
2446 		/* just a clean<-> dirty transition, possibly leave spares alone,
2447 		 * though if events isn't the right even/odd, we will have to do
2448 		 * spares after all
2449 		 */
2450 		nospares = 1;
2451 	if (force_change)
2452 		nospares = 0;
2453 	if (mddev->degraded)
2454 		/* If the array is degraded, then skipping spares is both
2455 		 * dangerous and fairly pointless.
2456 		 * Dangerous because a device that was removed from the array
2457 		 * might have a event_count that still looks up-to-date,
2458 		 * so it can be re-added without a resync.
2459 		 * Pointless because if there are any spares to skip,
2460 		 * then a recovery will happen and soon that array won't
2461 		 * be degraded any more and the spare can go back to sleep then.
2462 		 */
2463 		nospares = 0;
2464 
2465 	sync_req = mddev->in_sync;
2466 
2467 	/* If this is just a dirty<->clean transition, and the array is clean
2468 	 * and 'events' is odd, we can roll back to the previous clean state */
2469 	if (nospares
2470 	    && (mddev->in_sync && mddev->recovery_cp == MaxSector)
2471 	    && mddev->can_decrease_events
2472 	    && mddev->events != 1) {
2473 		mddev->events--;
2474 		mddev->can_decrease_events = 0;
2475 	} else {
2476 		/* otherwise we have to go forward and ... */
2477 		mddev->events ++;
2478 		mddev->can_decrease_events = nospares;
2479 	}
2480 
2481 	/*
2482 	 * This 64-bit counter should never wrap.
2483 	 * Either we are in around ~1 trillion A.C., assuming
2484 	 * 1 reboot per second, or we have a bug...
2485 	 */
2486 	WARN_ON(mddev->events == 0);
2487 
2488 	rdev_for_each(rdev, mddev) {
2489 		if (rdev->badblocks.changed)
2490 			any_badblocks_changed++;
2491 		if (test_bit(Faulty, &rdev->flags))
2492 			set_bit(FaultRecorded, &rdev->flags);
2493 	}
2494 
2495 	sync_sbs(mddev, nospares);
2496 	spin_unlock(&mddev->lock);
2497 
2498 	pr_debug("md: updating %s RAID superblock on device (in sync %d)\n",
2499 		 mdname(mddev), mddev->in_sync);
2500 
2501 	if (mddev->queue)
2502 		blk_add_trace_msg(mddev->queue, "md md_update_sb");
2503 rewrite:
2504 	bitmap_update_sb(mddev->bitmap);
2505 	rdev_for_each(rdev, mddev) {
2506 		char b[BDEVNAME_SIZE];
2507 
2508 		if (rdev->sb_loaded != 1)
2509 			continue; /* no noise on spare devices */
2510 
2511 		if (!test_bit(Faulty, &rdev->flags)) {
2512 			md_super_write(mddev,rdev,
2513 				       rdev->sb_start, rdev->sb_size,
2514 				       rdev->sb_page);
2515 			pr_debug("md: (write) %s's sb offset: %llu\n",
2516 				 bdevname(rdev->bdev, b),
2517 				 (unsigned long long)rdev->sb_start);
2518 			rdev->sb_events = mddev->events;
2519 			if (rdev->badblocks.size) {
2520 				md_super_write(mddev, rdev,
2521 					       rdev->badblocks.sector,
2522 					       rdev->badblocks.size << 9,
2523 					       rdev->bb_page);
2524 				rdev->badblocks.size = 0;
2525 			}
2526 
2527 		} else
2528 			pr_debug("md: %s (skipping faulty)\n",
2529 				 bdevname(rdev->bdev, b));
2530 
2531 		if (mddev->level == LEVEL_MULTIPATH)
2532 			/* only need to write one superblock... */
2533 			break;
2534 	}
2535 	if (md_super_wait(mddev) < 0)
2536 		goto rewrite;
2537 	/* if there was a failure, MD_SB_CHANGE_DEVS was set, and we re-write super */
2538 
2539 	if (mddev_is_clustered(mddev) && ret == 0)
2540 		md_cluster_ops->metadata_update_finish(mddev);
2541 
2542 	if (mddev->in_sync != sync_req ||
2543 	    !bit_clear_unless(&mddev->sb_flags, BIT(MD_SB_CHANGE_PENDING),
2544 			       BIT(MD_SB_CHANGE_DEVS) | BIT(MD_SB_CHANGE_CLEAN)))
2545 		/* have to write it out again */
2546 		goto repeat;
2547 	wake_up(&mddev->sb_wait);
2548 	if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
2549 		sysfs_notify(&mddev->kobj, NULL, "sync_completed");
2550 
2551 	rdev_for_each(rdev, mddev) {
2552 		if (test_and_clear_bit(FaultRecorded, &rdev->flags))
2553 			clear_bit(Blocked, &rdev->flags);
2554 
2555 		if (any_badblocks_changed)
2556 			ack_all_badblocks(&rdev->badblocks);
2557 		clear_bit(BlockedBadBlocks, &rdev->flags);
2558 		wake_up(&rdev->blocked_wait);
2559 	}
2560 }
2561 EXPORT_SYMBOL(md_update_sb);
2562 
2563 static int add_bound_rdev(struct md_rdev *rdev)
2564 {
2565 	struct mddev *mddev = rdev->mddev;
2566 	int err = 0;
2567 	bool add_journal = test_bit(Journal, &rdev->flags);
2568 
2569 	if (!mddev->pers->hot_remove_disk || add_journal) {
2570 		/* If there is hot_add_disk but no hot_remove_disk
2571 		 * then added disks for geometry changes,
2572 		 * and should be added immediately.
2573 		 */
2574 		super_types[mddev->major_version].
2575 			validate_super(mddev, rdev);
2576 		if (add_journal)
2577 			mddev_suspend(mddev);
2578 		err = mddev->pers->hot_add_disk(mddev, rdev);
2579 		if (add_journal)
2580 			mddev_resume(mddev);
2581 		if (err) {
2582 			md_kick_rdev_from_array(rdev);
2583 			return err;
2584 		}
2585 	}
2586 	sysfs_notify_dirent_safe(rdev->sysfs_state);
2587 
2588 	set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
2589 	if (mddev->degraded)
2590 		set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
2591 	set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2592 	md_new_event(mddev);
2593 	md_wakeup_thread(mddev->thread);
2594 	return 0;
2595 }
2596 
2597 /* words written to sysfs files may, or may not, be \n terminated.
2598  * We want to accept with case. For this we use cmd_match.
2599  */
2600 static int cmd_match(const char *cmd, const char *str)
2601 {
2602 	/* See if cmd, written into a sysfs file, matches
2603 	 * str.  They must either be the same, or cmd can
2604 	 * have a trailing newline
2605 	 */
2606 	while (*cmd && *str && *cmd == *str) {
2607 		cmd++;
2608 		str++;
2609 	}
2610 	if (*cmd == '\n')
2611 		cmd++;
2612 	if (*str || *cmd)
2613 		return 0;
2614 	return 1;
2615 }
2616 
2617 struct rdev_sysfs_entry {
2618 	struct attribute attr;
2619 	ssize_t (*show)(struct md_rdev *, char *);
2620 	ssize_t (*store)(struct md_rdev *, const char *, size_t);
2621 };
2622 
2623 static ssize_t
2624 state_show(struct md_rdev *rdev, char *page)
2625 {
2626 	char *sep = ",";
2627 	size_t len = 0;
2628 	unsigned long flags = ACCESS_ONCE(rdev->flags);
2629 
2630 	if (test_bit(Faulty, &flags) ||
2631 	    (!test_bit(ExternalBbl, &flags) &&
2632 	    rdev->badblocks.unacked_exist))
2633 		len += sprintf(page+len, "faulty%s", sep);
2634 	if (test_bit(In_sync, &flags))
2635 		len += sprintf(page+len, "in_sync%s", sep);
2636 	if (test_bit(Journal, &flags))
2637 		len += sprintf(page+len, "journal%s", sep);
2638 	if (test_bit(WriteMostly, &flags))
2639 		len += sprintf(page+len, "write_mostly%s", sep);
2640 	if (test_bit(Blocked, &flags) ||
2641 	    (rdev->badblocks.unacked_exist
2642 	     && !test_bit(Faulty, &flags)))
2643 		len += sprintf(page+len, "blocked%s", sep);
2644 	if (!test_bit(Faulty, &flags) &&
2645 	    !test_bit(Journal, &flags) &&
2646 	    !test_bit(In_sync, &flags))
2647 		len += sprintf(page+len, "spare%s", sep);
2648 	if (test_bit(WriteErrorSeen, &flags))
2649 		len += sprintf(page+len, "write_error%s", sep);
2650 	if (test_bit(WantReplacement, &flags))
2651 		len += sprintf(page+len, "want_replacement%s", sep);
2652 	if (test_bit(Replacement, &flags))
2653 		len += sprintf(page+len, "replacement%s", sep);
2654 	if (test_bit(ExternalBbl, &flags))
2655 		len += sprintf(page+len, "external_bbl%s", sep);
2656 	if (test_bit(FailFast, &flags))
2657 		len += sprintf(page+len, "failfast%s", sep);
2658 
2659 	if (len)
2660 		len -= strlen(sep);
2661 
2662 	return len+sprintf(page+len, "\n");
2663 }
2664 
2665 static ssize_t
2666 state_store(struct md_rdev *rdev, const char *buf, size_t len)
2667 {
2668 	/* can write
2669 	 *  faulty  - simulates an error
2670 	 *  remove  - disconnects the device
2671 	 *  writemostly - sets write_mostly
2672 	 *  -writemostly - clears write_mostly
2673 	 *  blocked - sets the Blocked flags
2674 	 *  -blocked - clears the Blocked and possibly simulates an error
2675 	 *  insync - sets Insync providing device isn't active
2676 	 *  -insync - clear Insync for a device with a slot assigned,
2677 	 *            so that it gets rebuilt based on bitmap
2678 	 *  write_error - sets WriteErrorSeen
2679 	 *  -write_error - clears WriteErrorSeen
2680 	 *  {,-}failfast - set/clear FailFast
2681 	 */
2682 	int err = -EINVAL;
2683 	if (cmd_match(buf, "faulty") && rdev->mddev->pers) {
2684 		md_error(rdev->mddev, rdev);
2685 		if (test_bit(Faulty, &rdev->flags))
2686 			err = 0;
2687 		else
2688 			err = -EBUSY;
2689 	} else if (cmd_match(buf, "remove")) {
2690 		if (rdev->mddev->pers) {
2691 			clear_bit(Blocked, &rdev->flags);
2692 			remove_and_add_spares(rdev->mddev, rdev);
2693 		}
2694 		if (rdev->raid_disk >= 0)
2695 			err = -EBUSY;
2696 		else {
2697 			struct mddev *mddev = rdev->mddev;
2698 			err = 0;
2699 			if (mddev_is_clustered(mddev))
2700 				err = md_cluster_ops->remove_disk(mddev, rdev);
2701 
2702 			if (err == 0) {
2703 				md_kick_rdev_from_array(rdev);
2704 				if (mddev->pers) {
2705 					set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
2706 					md_wakeup_thread(mddev->thread);
2707 				}
2708 				md_new_event(mddev);
2709 			}
2710 		}
2711 	} else if (cmd_match(buf, "writemostly")) {
2712 		set_bit(WriteMostly, &rdev->flags);
2713 		err = 0;
2714 	} else if (cmd_match(buf, "-writemostly")) {
2715 		clear_bit(WriteMostly, &rdev->flags);
2716 		err = 0;
2717 	} else if (cmd_match(buf, "blocked")) {
2718 		set_bit(Blocked, &rdev->flags);
2719 		err = 0;
2720 	} else if (cmd_match(buf, "-blocked")) {
2721 		if (!test_bit(Faulty, &rdev->flags) &&
2722 		    !test_bit(ExternalBbl, &rdev->flags) &&
2723 		    rdev->badblocks.unacked_exist) {
2724 			/* metadata handler doesn't understand badblocks,
2725 			 * so we need to fail the device
2726 			 */
2727 			md_error(rdev->mddev, rdev);
2728 		}
2729 		clear_bit(Blocked, &rdev->flags);
2730 		clear_bit(BlockedBadBlocks, &rdev->flags);
2731 		wake_up(&rdev->blocked_wait);
2732 		set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2733 		md_wakeup_thread(rdev->mddev->thread);
2734 
2735 		err = 0;
2736 	} else if (cmd_match(buf, "insync") && rdev->raid_disk == -1) {
2737 		set_bit(In_sync, &rdev->flags);
2738 		err = 0;
2739 	} else if (cmd_match(buf, "failfast")) {
2740 		set_bit(FailFast, &rdev->flags);
2741 		err = 0;
2742 	} else if (cmd_match(buf, "-failfast")) {
2743 		clear_bit(FailFast, &rdev->flags);
2744 		err = 0;
2745 	} else if (cmd_match(buf, "-insync") && rdev->raid_disk >= 0 &&
2746 		   !test_bit(Journal, &rdev->flags)) {
2747 		if (rdev->mddev->pers == NULL) {
2748 			clear_bit(In_sync, &rdev->flags);
2749 			rdev->saved_raid_disk = rdev->raid_disk;
2750 			rdev->raid_disk = -1;
2751 			err = 0;
2752 		}
2753 	} else if (cmd_match(buf, "write_error")) {
2754 		set_bit(WriteErrorSeen, &rdev->flags);
2755 		err = 0;
2756 	} else if (cmd_match(buf, "-write_error")) {
2757 		clear_bit(WriteErrorSeen, &rdev->flags);
2758 		err = 0;
2759 	} else if (cmd_match(buf, "want_replacement")) {
2760 		/* Any non-spare device that is not a replacement can
2761 		 * become want_replacement at any time, but we then need to
2762 		 * check if recovery is needed.
2763 		 */
2764 		if (rdev->raid_disk >= 0 &&
2765 		    !test_bit(Journal, &rdev->flags) &&
2766 		    !test_bit(Replacement, &rdev->flags))
2767 			set_bit(WantReplacement, &rdev->flags);
2768 		set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2769 		md_wakeup_thread(rdev->mddev->thread);
2770 		err = 0;
2771 	} else if (cmd_match(buf, "-want_replacement")) {
2772 		/* Clearing 'want_replacement' is always allowed.
2773 		 * Once replacements starts it is too late though.
2774 		 */
2775 		err = 0;
2776 		clear_bit(WantReplacement, &rdev->flags);
2777 	} else if (cmd_match(buf, "replacement")) {
2778 		/* Can only set a device as a replacement when array has not
2779 		 * yet been started.  Once running, replacement is automatic
2780 		 * from spares, or by assigning 'slot'.
2781 		 */
2782 		if (rdev->mddev->pers)
2783 			err = -EBUSY;
2784 		else {
2785 			set_bit(Replacement, &rdev->flags);
2786 			err = 0;
2787 		}
2788 	} else if (cmd_match(buf, "-replacement")) {
2789 		/* Similarly, can only clear Replacement before start */
2790 		if (rdev->mddev->pers)
2791 			err = -EBUSY;
2792 		else {
2793 			clear_bit(Replacement, &rdev->flags);
2794 			err = 0;
2795 		}
2796 	} else if (cmd_match(buf, "re-add")) {
2797 		if (test_bit(Faulty, &rdev->flags) && (rdev->raid_disk == -1)) {
2798 			/* clear_bit is performed _after_ all the devices
2799 			 * have their local Faulty bit cleared. If any writes
2800 			 * happen in the meantime in the local node, they
2801 			 * will land in the local bitmap, which will be synced
2802 			 * by this node eventually
2803 			 */
2804 			if (!mddev_is_clustered(rdev->mddev) ||
2805 			    (err = md_cluster_ops->gather_bitmaps(rdev)) == 0) {
2806 				clear_bit(Faulty, &rdev->flags);
2807 				err = add_bound_rdev(rdev);
2808 			}
2809 		} else
2810 			err = -EBUSY;
2811 	} else if (cmd_match(buf, "external_bbl") && (rdev->mddev->external)) {
2812 		set_bit(ExternalBbl, &rdev->flags);
2813 		rdev->badblocks.shift = 0;
2814 		err = 0;
2815 	} else if (cmd_match(buf, "-external_bbl") && (rdev->mddev->external)) {
2816 		clear_bit(ExternalBbl, &rdev->flags);
2817 		err = 0;
2818 	}
2819 	if (!err)
2820 		sysfs_notify_dirent_safe(rdev->sysfs_state);
2821 	return err ? err : len;
2822 }
2823 static struct rdev_sysfs_entry rdev_state =
2824 __ATTR_PREALLOC(state, S_IRUGO|S_IWUSR, state_show, state_store);
2825 
2826 static ssize_t
2827 errors_show(struct md_rdev *rdev, char *page)
2828 {
2829 	return sprintf(page, "%d\n", atomic_read(&rdev->corrected_errors));
2830 }
2831 
2832 static ssize_t
2833 errors_store(struct md_rdev *rdev, const char *buf, size_t len)
2834 {
2835 	unsigned int n;
2836 	int rv;
2837 
2838 	rv = kstrtouint(buf, 10, &n);
2839 	if (rv < 0)
2840 		return rv;
2841 	atomic_set(&rdev->corrected_errors, n);
2842 	return len;
2843 }
2844 static struct rdev_sysfs_entry rdev_errors =
2845 __ATTR(errors, S_IRUGO|S_IWUSR, errors_show, errors_store);
2846 
2847 static ssize_t
2848 slot_show(struct md_rdev *rdev, char *page)
2849 {
2850 	if (test_bit(Journal, &rdev->flags))
2851 		return sprintf(page, "journal\n");
2852 	else if (rdev->raid_disk < 0)
2853 		return sprintf(page, "none\n");
2854 	else
2855 		return sprintf(page, "%d\n", rdev->raid_disk);
2856 }
2857 
2858 static ssize_t
2859 slot_store(struct md_rdev *rdev, const char *buf, size_t len)
2860 {
2861 	int slot;
2862 	int err;
2863 
2864 	if (test_bit(Journal, &rdev->flags))
2865 		return -EBUSY;
2866 	if (strncmp(buf, "none", 4)==0)
2867 		slot = -1;
2868 	else {
2869 		err = kstrtouint(buf, 10, (unsigned int *)&slot);
2870 		if (err < 0)
2871 			return err;
2872 	}
2873 	if (rdev->mddev->pers && slot == -1) {
2874 		/* Setting 'slot' on an active array requires also
2875 		 * updating the 'rd%d' link, and communicating
2876 		 * with the personality with ->hot_*_disk.
2877 		 * For now we only support removing
2878 		 * failed/spare devices.  This normally happens automatically,
2879 		 * but not when the metadata is externally managed.
2880 		 */
2881 		if (rdev->raid_disk == -1)
2882 			return -EEXIST;
2883 		/* personality does all needed checks */
2884 		if (rdev->mddev->pers->hot_remove_disk == NULL)
2885 			return -EINVAL;
2886 		clear_bit(Blocked, &rdev->flags);
2887 		remove_and_add_spares(rdev->mddev, rdev);
2888 		if (rdev->raid_disk >= 0)
2889 			return -EBUSY;
2890 		set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2891 		md_wakeup_thread(rdev->mddev->thread);
2892 	} else if (rdev->mddev->pers) {
2893 		/* Activating a spare .. or possibly reactivating
2894 		 * if we ever get bitmaps working here.
2895 		 */
2896 		int err;
2897 
2898 		if (rdev->raid_disk != -1)
2899 			return -EBUSY;
2900 
2901 		if (test_bit(MD_RECOVERY_RUNNING, &rdev->mddev->recovery))
2902 			return -EBUSY;
2903 
2904 		if (rdev->mddev->pers->hot_add_disk == NULL)
2905 			return -EINVAL;
2906 
2907 		if (slot >= rdev->mddev->raid_disks &&
2908 		    slot >= rdev->mddev->raid_disks + rdev->mddev->delta_disks)
2909 			return -ENOSPC;
2910 
2911 		rdev->raid_disk = slot;
2912 		if (test_bit(In_sync, &rdev->flags))
2913 			rdev->saved_raid_disk = slot;
2914 		else
2915 			rdev->saved_raid_disk = -1;
2916 		clear_bit(In_sync, &rdev->flags);
2917 		clear_bit(Bitmap_sync, &rdev->flags);
2918 		err = rdev->mddev->pers->
2919 			hot_add_disk(rdev->mddev, rdev);
2920 		if (err) {
2921 			rdev->raid_disk = -1;
2922 			return err;
2923 		} else
2924 			sysfs_notify_dirent_safe(rdev->sysfs_state);
2925 		if (sysfs_link_rdev(rdev->mddev, rdev))
2926 			/* failure here is OK */;
2927 		/* don't wakeup anyone, leave that to userspace. */
2928 	} else {
2929 		if (slot >= rdev->mddev->raid_disks &&
2930 		    slot >= rdev->mddev->raid_disks + rdev->mddev->delta_disks)
2931 			return -ENOSPC;
2932 		rdev->raid_disk = slot;
2933 		/* assume it is working */
2934 		clear_bit(Faulty, &rdev->flags);
2935 		clear_bit(WriteMostly, &rdev->flags);
2936 		set_bit(In_sync, &rdev->flags);
2937 		sysfs_notify_dirent_safe(rdev->sysfs_state);
2938 	}
2939 	return len;
2940 }
2941 
2942 static struct rdev_sysfs_entry rdev_slot =
2943 __ATTR(slot, S_IRUGO|S_IWUSR, slot_show, slot_store);
2944 
2945 static ssize_t
2946 offset_show(struct md_rdev *rdev, char *page)
2947 {
2948 	return sprintf(page, "%llu\n", (unsigned long long)rdev->data_offset);
2949 }
2950 
2951 static ssize_t
2952 offset_store(struct md_rdev *rdev, const char *buf, size_t len)
2953 {
2954 	unsigned long long offset;
2955 	if (kstrtoull(buf, 10, &offset) < 0)
2956 		return -EINVAL;
2957 	if (rdev->mddev->pers && rdev->raid_disk >= 0)
2958 		return -EBUSY;
2959 	if (rdev->sectors && rdev->mddev->external)
2960 		/* Must set offset before size, so overlap checks
2961 		 * can be sane */
2962 		return -EBUSY;
2963 	rdev->data_offset = offset;
2964 	rdev->new_data_offset = offset;
2965 	return len;
2966 }
2967 
2968 static struct rdev_sysfs_entry rdev_offset =
2969 __ATTR(offset, S_IRUGO|S_IWUSR, offset_show, offset_store);
2970 
2971 static ssize_t new_offset_show(struct md_rdev *rdev, char *page)
2972 {
2973 	return sprintf(page, "%llu\n",
2974 		       (unsigned long long)rdev->new_data_offset);
2975 }
2976 
2977 static ssize_t new_offset_store(struct md_rdev *rdev,
2978 				const char *buf, size_t len)
2979 {
2980 	unsigned long long new_offset;
2981 	struct mddev *mddev = rdev->mddev;
2982 
2983 	if (kstrtoull(buf, 10, &new_offset) < 0)
2984 		return -EINVAL;
2985 
2986 	if (mddev->sync_thread ||
2987 	    test_bit(MD_RECOVERY_RUNNING,&mddev->recovery))
2988 		return -EBUSY;
2989 	if (new_offset == rdev->data_offset)
2990 		/* reset is always permitted */
2991 		;
2992 	else if (new_offset > rdev->data_offset) {
2993 		/* must not push array size beyond rdev_sectors */
2994 		if (new_offset - rdev->data_offset
2995 		    + mddev->dev_sectors > rdev->sectors)
2996 				return -E2BIG;
2997 	}
2998 	/* Metadata worries about other space details. */
2999 
3000 	/* decreasing the offset is inconsistent with a backwards
3001 	 * reshape.
3002 	 */
3003 	if (new_offset < rdev->data_offset &&
3004 	    mddev->reshape_backwards)
3005 		return -EINVAL;
3006 	/* Increasing offset is inconsistent with forwards
3007 	 * reshape.  reshape_direction should be set to
3008 	 * 'backwards' first.
3009 	 */
3010 	if (new_offset > rdev->data_offset &&
3011 	    !mddev->reshape_backwards)
3012 		return -EINVAL;
3013 
3014 	if (mddev->pers && mddev->persistent &&
3015 	    !super_types[mddev->major_version]
3016 	    .allow_new_offset(rdev, new_offset))
3017 		return -E2BIG;
3018 	rdev->new_data_offset = new_offset;
3019 	if (new_offset > rdev->data_offset)
3020 		mddev->reshape_backwards = 1;
3021 	else if (new_offset < rdev->data_offset)
3022 		mddev->reshape_backwards = 0;
3023 
3024 	return len;
3025 }
3026 static struct rdev_sysfs_entry rdev_new_offset =
3027 __ATTR(new_offset, S_IRUGO|S_IWUSR, new_offset_show, new_offset_store);
3028 
3029 static ssize_t
3030 rdev_size_show(struct md_rdev *rdev, char *page)
3031 {
3032 	return sprintf(page, "%llu\n", (unsigned long long)rdev->sectors / 2);
3033 }
3034 
3035 static int overlaps(sector_t s1, sector_t l1, sector_t s2, sector_t l2)
3036 {
3037 	/* check if two start/length pairs overlap */
3038 	if (s1+l1 <= s2)
3039 		return 0;
3040 	if (s2+l2 <= s1)
3041 		return 0;
3042 	return 1;
3043 }
3044 
3045 static int strict_blocks_to_sectors(const char *buf, sector_t *sectors)
3046 {
3047 	unsigned long long blocks;
3048 	sector_t new;
3049 
3050 	if (kstrtoull(buf, 10, &blocks) < 0)
3051 		return -EINVAL;
3052 
3053 	if (blocks & 1ULL << (8 * sizeof(blocks) - 1))
3054 		return -EINVAL; /* sector conversion overflow */
3055 
3056 	new = blocks * 2;
3057 	if (new != blocks * 2)
3058 		return -EINVAL; /* unsigned long long to sector_t overflow */
3059 
3060 	*sectors = new;
3061 	return 0;
3062 }
3063 
3064 static ssize_t
3065 rdev_size_store(struct md_rdev *rdev, const char *buf, size_t len)
3066 {
3067 	struct mddev *my_mddev = rdev->mddev;
3068 	sector_t oldsectors = rdev->sectors;
3069 	sector_t sectors;
3070 
3071 	if (test_bit(Journal, &rdev->flags))
3072 		return -EBUSY;
3073 	if (strict_blocks_to_sectors(buf, &sectors) < 0)
3074 		return -EINVAL;
3075 	if (rdev->data_offset != rdev->new_data_offset)
3076 		return -EINVAL; /* too confusing */
3077 	if (my_mddev->pers && rdev->raid_disk >= 0) {
3078 		if (my_mddev->persistent) {
3079 			sectors = super_types[my_mddev->major_version].
3080 				rdev_size_change(rdev, sectors);
3081 			if (!sectors)
3082 				return -EBUSY;
3083 		} else if (!sectors)
3084 			sectors = (i_size_read(rdev->bdev->bd_inode) >> 9) -
3085 				rdev->data_offset;
3086 		if (!my_mddev->pers->resize)
3087 			/* Cannot change size for RAID0 or Linear etc */
3088 			return -EINVAL;
3089 	}
3090 	if (sectors < my_mddev->dev_sectors)
3091 		return -EINVAL; /* component must fit device */
3092 
3093 	rdev->sectors = sectors;
3094 	if (sectors > oldsectors && my_mddev->external) {
3095 		/* Need to check that all other rdevs with the same
3096 		 * ->bdev do not overlap.  'rcu' is sufficient to walk
3097 		 * the rdev lists safely.
3098 		 * This check does not provide a hard guarantee, it
3099 		 * just helps avoid dangerous mistakes.
3100 		 */
3101 		struct mddev *mddev;
3102 		int overlap = 0;
3103 		struct list_head *tmp;
3104 
3105 		rcu_read_lock();
3106 		for_each_mddev(mddev, tmp) {
3107 			struct md_rdev *rdev2;
3108 
3109 			rdev_for_each(rdev2, mddev)
3110 				if (rdev->bdev == rdev2->bdev &&
3111 				    rdev != rdev2 &&
3112 				    overlaps(rdev->data_offset, rdev->sectors,
3113 					     rdev2->data_offset,
3114 					     rdev2->sectors)) {
3115 					overlap = 1;
3116 					break;
3117 				}
3118 			if (overlap) {
3119 				mddev_put(mddev);
3120 				break;
3121 			}
3122 		}
3123 		rcu_read_unlock();
3124 		if (overlap) {
3125 			/* Someone else could have slipped in a size
3126 			 * change here, but doing so is just silly.
3127 			 * We put oldsectors back because we *know* it is
3128 			 * safe, and trust userspace not to race with
3129 			 * itself
3130 			 */
3131 			rdev->sectors = oldsectors;
3132 			return -EBUSY;
3133 		}
3134 	}
3135 	return len;
3136 }
3137 
3138 static struct rdev_sysfs_entry rdev_size =
3139 __ATTR(size, S_IRUGO|S_IWUSR, rdev_size_show, rdev_size_store);
3140 
3141 static ssize_t recovery_start_show(struct md_rdev *rdev, char *page)
3142 {
3143 	unsigned long long recovery_start = rdev->recovery_offset;
3144 
3145 	if (test_bit(In_sync, &rdev->flags) ||
3146 	    recovery_start == MaxSector)
3147 		return sprintf(page, "none\n");
3148 
3149 	return sprintf(page, "%llu\n", recovery_start);
3150 }
3151 
3152 static ssize_t recovery_start_store(struct md_rdev *rdev, const char *buf, size_t len)
3153 {
3154 	unsigned long long recovery_start;
3155 
3156 	if (cmd_match(buf, "none"))
3157 		recovery_start = MaxSector;
3158 	else if (kstrtoull(buf, 10, &recovery_start))
3159 		return -EINVAL;
3160 
3161 	if (rdev->mddev->pers &&
3162 	    rdev->raid_disk >= 0)
3163 		return -EBUSY;
3164 
3165 	rdev->recovery_offset = recovery_start;
3166 	if (recovery_start == MaxSector)
3167 		set_bit(In_sync, &rdev->flags);
3168 	else
3169 		clear_bit(In_sync, &rdev->flags);
3170 	return len;
3171 }
3172 
3173 static struct rdev_sysfs_entry rdev_recovery_start =
3174 __ATTR(recovery_start, S_IRUGO|S_IWUSR, recovery_start_show, recovery_start_store);
3175 
3176 /* sysfs access to bad-blocks list.
3177  * We present two files.
3178  * 'bad-blocks' lists sector numbers and lengths of ranges that
3179  *    are recorded as bad.  The list is truncated to fit within
3180  *    the one-page limit of sysfs.
3181  *    Writing "sector length" to this file adds an acknowledged
3182  *    bad block list.
3183  * 'unacknowledged-bad-blocks' lists bad blocks that have not yet
3184  *    been acknowledged.  Writing to this file adds bad blocks
3185  *    without acknowledging them.  This is largely for testing.
3186  */
3187 static ssize_t bb_show(struct md_rdev *rdev, char *page)
3188 {
3189 	return badblocks_show(&rdev->badblocks, page, 0);
3190 }
3191 static ssize_t bb_store(struct md_rdev *rdev, const char *page, size_t len)
3192 {
3193 	int rv = badblocks_store(&rdev->badblocks, page, len, 0);
3194 	/* Maybe that ack was all we needed */
3195 	if (test_and_clear_bit(BlockedBadBlocks, &rdev->flags))
3196 		wake_up(&rdev->blocked_wait);
3197 	return rv;
3198 }
3199 static struct rdev_sysfs_entry rdev_bad_blocks =
3200 __ATTR(bad_blocks, S_IRUGO|S_IWUSR, bb_show, bb_store);
3201 
3202 static ssize_t ubb_show(struct md_rdev *rdev, char *page)
3203 {
3204 	return badblocks_show(&rdev->badblocks, page, 1);
3205 }
3206 static ssize_t ubb_store(struct md_rdev *rdev, const char *page, size_t len)
3207 {
3208 	return badblocks_store(&rdev->badblocks, page, len, 1);
3209 }
3210 static struct rdev_sysfs_entry rdev_unack_bad_blocks =
3211 __ATTR(unacknowledged_bad_blocks, S_IRUGO|S_IWUSR, ubb_show, ubb_store);
3212 
3213 static ssize_t
3214 ppl_sector_show(struct md_rdev *rdev, char *page)
3215 {
3216 	return sprintf(page, "%llu\n", (unsigned long long)rdev->ppl.sector);
3217 }
3218 
3219 static ssize_t
3220 ppl_sector_store(struct md_rdev *rdev, const char *buf, size_t len)
3221 {
3222 	unsigned long long sector;
3223 
3224 	if (kstrtoull(buf, 10, &sector) < 0)
3225 		return -EINVAL;
3226 	if (sector != (sector_t)sector)
3227 		return -EINVAL;
3228 
3229 	if (rdev->mddev->pers && test_bit(MD_HAS_PPL, &rdev->mddev->flags) &&
3230 	    rdev->raid_disk >= 0)
3231 		return -EBUSY;
3232 
3233 	if (rdev->mddev->persistent) {
3234 		if (rdev->mddev->major_version == 0)
3235 			return -EINVAL;
3236 		if ((sector > rdev->sb_start &&
3237 		     sector - rdev->sb_start > S16_MAX) ||
3238 		    (sector < rdev->sb_start &&
3239 		     rdev->sb_start - sector > -S16_MIN))
3240 			return -EINVAL;
3241 		rdev->ppl.offset = sector - rdev->sb_start;
3242 	} else if (!rdev->mddev->external) {
3243 		return -EBUSY;
3244 	}
3245 	rdev->ppl.sector = sector;
3246 	return len;
3247 }
3248 
3249 static struct rdev_sysfs_entry rdev_ppl_sector =
3250 __ATTR(ppl_sector, S_IRUGO|S_IWUSR, ppl_sector_show, ppl_sector_store);
3251 
3252 static ssize_t
3253 ppl_size_show(struct md_rdev *rdev, char *page)
3254 {
3255 	return sprintf(page, "%u\n", rdev->ppl.size);
3256 }
3257 
3258 static ssize_t
3259 ppl_size_store(struct md_rdev *rdev, const char *buf, size_t len)
3260 {
3261 	unsigned int size;
3262 
3263 	if (kstrtouint(buf, 10, &size) < 0)
3264 		return -EINVAL;
3265 
3266 	if (rdev->mddev->pers && test_bit(MD_HAS_PPL, &rdev->mddev->flags) &&
3267 	    rdev->raid_disk >= 0)
3268 		return -EBUSY;
3269 
3270 	if (rdev->mddev->persistent) {
3271 		if (rdev->mddev->major_version == 0)
3272 			return -EINVAL;
3273 		if (size > U16_MAX)
3274 			return -EINVAL;
3275 	} else if (!rdev->mddev->external) {
3276 		return -EBUSY;
3277 	}
3278 	rdev->ppl.size = size;
3279 	return len;
3280 }
3281 
3282 static struct rdev_sysfs_entry rdev_ppl_size =
3283 __ATTR(ppl_size, S_IRUGO|S_IWUSR, ppl_size_show, ppl_size_store);
3284 
3285 static struct attribute *rdev_default_attrs[] = {
3286 	&rdev_state.attr,
3287 	&rdev_errors.attr,
3288 	&rdev_slot.attr,
3289 	&rdev_offset.attr,
3290 	&rdev_new_offset.attr,
3291 	&rdev_size.attr,
3292 	&rdev_recovery_start.attr,
3293 	&rdev_bad_blocks.attr,
3294 	&rdev_unack_bad_blocks.attr,
3295 	&rdev_ppl_sector.attr,
3296 	&rdev_ppl_size.attr,
3297 	NULL,
3298 };
3299 static ssize_t
3300 rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
3301 {
3302 	struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
3303 	struct md_rdev *rdev = container_of(kobj, struct md_rdev, kobj);
3304 
3305 	if (!entry->show)
3306 		return -EIO;
3307 	if (!rdev->mddev)
3308 		return -EBUSY;
3309 	return entry->show(rdev, page);
3310 }
3311 
3312 static ssize_t
3313 rdev_attr_store(struct kobject *kobj, struct attribute *attr,
3314 	      const char *page, size_t length)
3315 {
3316 	struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
3317 	struct md_rdev *rdev = container_of(kobj, struct md_rdev, kobj);
3318 	ssize_t rv;
3319 	struct mddev *mddev = rdev->mddev;
3320 
3321 	if (!entry->store)
3322 		return -EIO;
3323 	if (!capable(CAP_SYS_ADMIN))
3324 		return -EACCES;
3325 	rv = mddev ? mddev_lock(mddev): -EBUSY;
3326 	if (!rv) {
3327 		if (rdev->mddev == NULL)
3328 			rv = -EBUSY;
3329 		else
3330 			rv = entry->store(rdev, page, length);
3331 		mddev_unlock(mddev);
3332 	}
3333 	return rv;
3334 }
3335 
3336 static void rdev_free(struct kobject *ko)
3337 {
3338 	struct md_rdev *rdev = container_of(ko, struct md_rdev, kobj);
3339 	kfree(rdev);
3340 }
3341 static const struct sysfs_ops rdev_sysfs_ops = {
3342 	.show		= rdev_attr_show,
3343 	.store		= rdev_attr_store,
3344 };
3345 static struct kobj_type rdev_ktype = {
3346 	.release	= rdev_free,
3347 	.sysfs_ops	= &rdev_sysfs_ops,
3348 	.default_attrs	= rdev_default_attrs,
3349 };
3350 
3351 int md_rdev_init(struct md_rdev *rdev)
3352 {
3353 	rdev->desc_nr = -1;
3354 	rdev->saved_raid_disk = -1;
3355 	rdev->raid_disk = -1;
3356 	rdev->flags = 0;
3357 	rdev->data_offset = 0;
3358 	rdev->new_data_offset = 0;
3359 	rdev->sb_events = 0;
3360 	rdev->last_read_error = 0;
3361 	rdev->sb_loaded = 0;
3362 	rdev->bb_page = NULL;
3363 	atomic_set(&rdev->nr_pending, 0);
3364 	atomic_set(&rdev->read_errors, 0);
3365 	atomic_set(&rdev->corrected_errors, 0);
3366 
3367 	INIT_LIST_HEAD(&rdev->same_set);
3368 	init_waitqueue_head(&rdev->blocked_wait);
3369 
3370 	/* Add space to store bad block list.
3371 	 * This reserves the space even on arrays where it cannot
3372 	 * be used - I wonder if that matters
3373 	 */
3374 	return badblocks_init(&rdev->badblocks, 0);
3375 }
3376 EXPORT_SYMBOL_GPL(md_rdev_init);
3377 /*
3378  * Import a device. If 'super_format' >= 0, then sanity check the superblock
3379  *
3380  * mark the device faulty if:
3381  *
3382  *   - the device is nonexistent (zero size)
3383  *   - the device has no valid superblock
3384  *
3385  * a faulty rdev _never_ has rdev->sb set.
3386  */
3387 static struct md_rdev *md_import_device(dev_t newdev, int super_format, int super_minor)
3388 {
3389 	char b[BDEVNAME_SIZE];
3390 	int err;
3391 	struct md_rdev *rdev;
3392 	sector_t size;
3393 
3394 	rdev = kzalloc(sizeof(*rdev), GFP_KERNEL);
3395 	if (!rdev)
3396 		return ERR_PTR(-ENOMEM);
3397 
3398 	err = md_rdev_init(rdev);
3399 	if (err)
3400 		goto abort_free;
3401 	err = alloc_disk_sb(rdev);
3402 	if (err)
3403 		goto abort_free;
3404 
3405 	err = lock_rdev(rdev, newdev, super_format == -2);
3406 	if (err)
3407 		goto abort_free;
3408 
3409 	kobject_init(&rdev->kobj, &rdev_ktype);
3410 
3411 	size = i_size_read(rdev->bdev->bd_inode) >> BLOCK_SIZE_BITS;
3412 	if (!size) {
3413 		pr_warn("md: %s has zero or unknown size, marking faulty!\n",
3414 			bdevname(rdev->bdev,b));
3415 		err = -EINVAL;
3416 		goto abort_free;
3417 	}
3418 
3419 	if (super_format >= 0) {
3420 		err = super_types[super_format].
3421 			load_super(rdev, NULL, super_minor);
3422 		if (err == -EINVAL) {
3423 			pr_warn("md: %s does not have a valid v%d.%d superblock, not importing!\n",
3424 				bdevname(rdev->bdev,b),
3425 				super_format, super_minor);
3426 			goto abort_free;
3427 		}
3428 		if (err < 0) {
3429 			pr_warn("md: could not read %s's sb, not importing!\n",
3430 				bdevname(rdev->bdev,b));
3431 			goto abort_free;
3432 		}
3433 	}
3434 
3435 	return rdev;
3436 
3437 abort_free:
3438 	if (rdev->bdev)
3439 		unlock_rdev(rdev);
3440 	md_rdev_clear(rdev);
3441 	kfree(rdev);
3442 	return ERR_PTR(err);
3443 }
3444 
3445 /*
3446  * Check a full RAID array for plausibility
3447  */
3448 
3449 static void analyze_sbs(struct mddev *mddev)
3450 {
3451 	int i;
3452 	struct md_rdev *rdev, *freshest, *tmp;
3453 	char b[BDEVNAME_SIZE];
3454 
3455 	freshest = NULL;
3456 	rdev_for_each_safe(rdev, tmp, mddev)
3457 		switch (super_types[mddev->major_version].
3458 			load_super(rdev, freshest, mddev->minor_version)) {
3459 		case 1:
3460 			freshest = rdev;
3461 			break;
3462 		case 0:
3463 			break;
3464 		default:
3465 			pr_warn("md: fatal superblock inconsistency in %s -- removing from array\n",
3466 				bdevname(rdev->bdev,b));
3467 			md_kick_rdev_from_array(rdev);
3468 		}
3469 
3470 	super_types[mddev->major_version].
3471 		validate_super(mddev, freshest);
3472 
3473 	i = 0;
3474 	rdev_for_each_safe(rdev, tmp, mddev) {
3475 		if (mddev->max_disks &&
3476 		    (rdev->desc_nr >= mddev->max_disks ||
3477 		     i > mddev->max_disks)) {
3478 			pr_warn("md: %s: %s: only %d devices permitted\n",
3479 				mdname(mddev), bdevname(rdev->bdev, b),
3480 				mddev->max_disks);
3481 			md_kick_rdev_from_array(rdev);
3482 			continue;
3483 		}
3484 		if (rdev != freshest) {
3485 			if (super_types[mddev->major_version].
3486 			    validate_super(mddev, rdev)) {
3487 				pr_warn("md: kicking non-fresh %s from array!\n",
3488 					bdevname(rdev->bdev,b));
3489 				md_kick_rdev_from_array(rdev);
3490 				continue;
3491 			}
3492 		}
3493 		if (mddev->level == LEVEL_MULTIPATH) {
3494 			rdev->desc_nr = i++;
3495 			rdev->raid_disk = rdev->desc_nr;
3496 			set_bit(In_sync, &rdev->flags);
3497 		} else if (rdev->raid_disk >=
3498 			    (mddev->raid_disks - min(0, mddev->delta_disks)) &&
3499 			   !test_bit(Journal, &rdev->flags)) {
3500 			rdev->raid_disk = -1;
3501 			clear_bit(In_sync, &rdev->flags);
3502 		}
3503 	}
3504 }
3505 
3506 /* Read a fixed-point number.
3507  * Numbers in sysfs attributes should be in "standard" units where
3508  * possible, so time should be in seconds.
3509  * However we internally use a a much smaller unit such as
3510  * milliseconds or jiffies.
3511  * This function takes a decimal number with a possible fractional
3512  * component, and produces an integer which is the result of
3513  * multiplying that number by 10^'scale'.
3514  * all without any floating-point arithmetic.
3515  */
3516 int strict_strtoul_scaled(const char *cp, unsigned long *res, int scale)
3517 {
3518 	unsigned long result = 0;
3519 	long decimals = -1;
3520 	while (isdigit(*cp) || (*cp == '.' && decimals < 0)) {
3521 		if (*cp == '.')
3522 			decimals = 0;
3523 		else if (decimals < scale) {
3524 			unsigned int value;
3525 			value = *cp - '0';
3526 			result = result * 10 + value;
3527 			if (decimals >= 0)
3528 				decimals++;
3529 		}
3530 		cp++;
3531 	}
3532 	if (*cp == '\n')
3533 		cp++;
3534 	if (*cp)
3535 		return -EINVAL;
3536 	if (decimals < 0)
3537 		decimals = 0;
3538 	while (decimals < scale) {
3539 		result *= 10;
3540 		decimals ++;
3541 	}
3542 	*res = result;
3543 	return 0;
3544 }
3545 
3546 static ssize_t
3547 safe_delay_show(struct mddev *mddev, char *page)
3548 {
3549 	int msec = (mddev->safemode_delay*1000)/HZ;
3550 	return sprintf(page, "%d.%03d\n", msec/1000, msec%1000);
3551 }
3552 static ssize_t
3553 safe_delay_store(struct mddev *mddev, const char *cbuf, size_t len)
3554 {
3555 	unsigned long msec;
3556 
3557 	if (mddev_is_clustered(mddev)) {
3558 		pr_warn("md: Safemode is disabled for clustered mode\n");
3559 		return -EINVAL;
3560 	}
3561 
3562 	if (strict_strtoul_scaled(cbuf, &msec, 3) < 0)
3563 		return -EINVAL;
3564 	if (msec == 0)
3565 		mddev->safemode_delay = 0;
3566 	else {
3567 		unsigned long old_delay = mddev->safemode_delay;
3568 		unsigned long new_delay = (msec*HZ)/1000;
3569 
3570 		if (new_delay == 0)
3571 			new_delay = 1;
3572 		mddev->safemode_delay = new_delay;
3573 		if (new_delay < old_delay || old_delay == 0)
3574 			mod_timer(&mddev->safemode_timer, jiffies+1);
3575 	}
3576 	return len;
3577 }
3578 static struct md_sysfs_entry md_safe_delay =
3579 __ATTR(safe_mode_delay, S_IRUGO|S_IWUSR,safe_delay_show, safe_delay_store);
3580 
3581 static ssize_t
3582 level_show(struct mddev *mddev, char *page)
3583 {
3584 	struct md_personality *p;
3585 	int ret;
3586 	spin_lock(&mddev->lock);
3587 	p = mddev->pers;
3588 	if (p)
3589 		ret = sprintf(page, "%s\n", p->name);
3590 	else if (mddev->clevel[0])
3591 		ret = sprintf(page, "%s\n", mddev->clevel);
3592 	else if (mddev->level != LEVEL_NONE)
3593 		ret = sprintf(page, "%d\n", mddev->level);
3594 	else
3595 		ret = 0;
3596 	spin_unlock(&mddev->lock);
3597 	return ret;
3598 }
3599 
3600 static ssize_t
3601 level_store(struct mddev *mddev, const char *buf, size_t len)
3602 {
3603 	char clevel[16];
3604 	ssize_t rv;
3605 	size_t slen = len;
3606 	struct md_personality *pers, *oldpers;
3607 	long level;
3608 	void *priv, *oldpriv;
3609 	struct md_rdev *rdev;
3610 
3611 	if (slen == 0 || slen >= sizeof(clevel))
3612 		return -EINVAL;
3613 
3614 	rv = mddev_lock(mddev);
3615 	if (rv)
3616 		return rv;
3617 
3618 	if (mddev->pers == NULL) {
3619 		strncpy(mddev->clevel, buf, slen);
3620 		if (mddev->clevel[slen-1] == '\n')
3621 			slen--;
3622 		mddev->clevel[slen] = 0;
3623 		mddev->level = LEVEL_NONE;
3624 		rv = len;
3625 		goto out_unlock;
3626 	}
3627 	rv = -EROFS;
3628 	if (mddev->ro)
3629 		goto out_unlock;
3630 
3631 	/* request to change the personality.  Need to ensure:
3632 	 *  - array is not engaged in resync/recovery/reshape
3633 	 *  - old personality can be suspended
3634 	 *  - new personality will access other array.
3635 	 */
3636 
3637 	rv = -EBUSY;
3638 	if (mddev->sync_thread ||
3639 	    test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
3640 	    mddev->reshape_position != MaxSector ||
3641 	    mddev->sysfs_active)
3642 		goto out_unlock;
3643 
3644 	rv = -EINVAL;
3645 	if (!mddev->pers->quiesce) {
3646 		pr_warn("md: %s: %s does not support online personality change\n",
3647 			mdname(mddev), mddev->pers->name);
3648 		goto out_unlock;
3649 	}
3650 
3651 	/* Now find the new personality */
3652 	strncpy(clevel, buf, slen);
3653 	if (clevel[slen-1] == '\n')
3654 		slen--;
3655 	clevel[slen] = 0;
3656 	if (kstrtol(clevel, 10, &level))
3657 		level = LEVEL_NONE;
3658 
3659 	if (request_module("md-%s", clevel) != 0)
3660 		request_module("md-level-%s", clevel);
3661 	spin_lock(&pers_lock);
3662 	pers = find_pers(level, clevel);
3663 	if (!pers || !try_module_get(pers->owner)) {
3664 		spin_unlock(&pers_lock);
3665 		pr_warn("md: personality %s not loaded\n", clevel);
3666 		rv = -EINVAL;
3667 		goto out_unlock;
3668 	}
3669 	spin_unlock(&pers_lock);
3670 
3671 	if (pers == mddev->pers) {
3672 		/* Nothing to do! */
3673 		module_put(pers->owner);
3674 		rv = len;
3675 		goto out_unlock;
3676 	}
3677 	if (!pers->takeover) {
3678 		module_put(pers->owner);
3679 		pr_warn("md: %s: %s does not support personality takeover\n",
3680 			mdname(mddev), clevel);
3681 		rv = -EINVAL;
3682 		goto out_unlock;
3683 	}
3684 
3685 	rdev_for_each(rdev, mddev)
3686 		rdev->new_raid_disk = rdev->raid_disk;
3687 
3688 	/* ->takeover must set new_* and/or delta_disks
3689 	 * if it succeeds, and may set them when it fails.
3690 	 */
3691 	priv = pers->takeover(mddev);
3692 	if (IS_ERR(priv)) {
3693 		mddev->new_level = mddev->level;
3694 		mddev->new_layout = mddev->layout;
3695 		mddev->new_chunk_sectors = mddev->chunk_sectors;
3696 		mddev->raid_disks -= mddev->delta_disks;
3697 		mddev->delta_disks = 0;
3698 		mddev->reshape_backwards = 0;
3699 		module_put(pers->owner);
3700 		pr_warn("md: %s: %s would not accept array\n",
3701 			mdname(mddev), clevel);
3702 		rv = PTR_ERR(priv);
3703 		goto out_unlock;
3704 	}
3705 
3706 	/* Looks like we have a winner */
3707 	mddev_suspend(mddev);
3708 	mddev_detach(mddev);
3709 
3710 	spin_lock(&mddev->lock);
3711 	oldpers = mddev->pers;
3712 	oldpriv = mddev->private;
3713 	mddev->pers = pers;
3714 	mddev->private = priv;
3715 	strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
3716 	mddev->level = mddev->new_level;
3717 	mddev->layout = mddev->new_layout;
3718 	mddev->chunk_sectors = mddev->new_chunk_sectors;
3719 	mddev->delta_disks = 0;
3720 	mddev->reshape_backwards = 0;
3721 	mddev->degraded = 0;
3722 	spin_unlock(&mddev->lock);
3723 
3724 	if (oldpers->sync_request == NULL &&
3725 	    mddev->external) {
3726 		/* We are converting from a no-redundancy array
3727 		 * to a redundancy array and metadata is managed
3728 		 * externally so we need to be sure that writes
3729 		 * won't block due to a need to transition
3730 		 *      clean->dirty
3731 		 * until external management is started.
3732 		 */
3733 		mddev->in_sync = 0;
3734 		mddev->safemode_delay = 0;
3735 		mddev->safemode = 0;
3736 	}
3737 
3738 	oldpers->free(mddev, oldpriv);
3739 
3740 	if (oldpers->sync_request == NULL &&
3741 	    pers->sync_request != NULL) {
3742 		/* need to add the md_redundancy_group */
3743 		if (sysfs_create_group(&mddev->kobj, &md_redundancy_group))
3744 			pr_warn("md: cannot register extra attributes for %s\n",
3745 				mdname(mddev));
3746 		mddev->sysfs_action = sysfs_get_dirent(mddev->kobj.sd, "sync_action");
3747 	}
3748 	if (oldpers->sync_request != NULL &&
3749 	    pers->sync_request == NULL) {
3750 		/* need to remove the md_redundancy_group */
3751 		if (mddev->to_remove == NULL)
3752 			mddev->to_remove = &md_redundancy_group;
3753 	}
3754 
3755 	module_put(oldpers->owner);
3756 
3757 	rdev_for_each(rdev, mddev) {
3758 		if (rdev->raid_disk < 0)
3759 			continue;
3760 		if (rdev->new_raid_disk >= mddev->raid_disks)
3761 			rdev->new_raid_disk = -1;
3762 		if (rdev->new_raid_disk == rdev->raid_disk)
3763 			continue;
3764 		sysfs_unlink_rdev(mddev, rdev);
3765 	}
3766 	rdev_for_each(rdev, mddev) {
3767 		if (rdev->raid_disk < 0)
3768 			continue;
3769 		if (rdev->new_raid_disk == rdev->raid_disk)
3770 			continue;
3771 		rdev->raid_disk = rdev->new_raid_disk;
3772 		if (rdev->raid_disk < 0)
3773 			clear_bit(In_sync, &rdev->flags);
3774 		else {
3775 			if (sysfs_link_rdev(mddev, rdev))
3776 				pr_warn("md: cannot register rd%d for %s after level change\n",
3777 					rdev->raid_disk, mdname(mddev));
3778 		}
3779 	}
3780 
3781 	if (pers->sync_request == NULL) {
3782 		/* this is now an array without redundancy, so
3783 		 * it must always be in_sync
3784 		 */
3785 		mddev->in_sync = 1;
3786 		del_timer_sync(&mddev->safemode_timer);
3787 	}
3788 	blk_set_stacking_limits(&mddev->queue->limits);
3789 	pers->run(mddev);
3790 	set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
3791 	mddev_resume(mddev);
3792 	if (!mddev->thread)
3793 		md_update_sb(mddev, 1);
3794 	sysfs_notify(&mddev->kobj, NULL, "level");
3795 	md_new_event(mddev);
3796 	rv = len;
3797 out_unlock:
3798 	mddev_unlock(mddev);
3799 	return rv;
3800 }
3801 
3802 static struct md_sysfs_entry md_level =
3803 __ATTR(level, S_IRUGO|S_IWUSR, level_show, level_store);
3804 
3805 static ssize_t
3806 layout_show(struct mddev *mddev, char *page)
3807 {
3808 	/* just a number, not meaningful for all levels */
3809 	if (mddev->reshape_position != MaxSector &&
3810 	    mddev->layout != mddev->new_layout)
3811 		return sprintf(page, "%d (%d)\n",
3812 			       mddev->new_layout, mddev->layout);
3813 	return sprintf(page, "%d\n", mddev->layout);
3814 }
3815 
3816 static ssize_t
3817 layout_store(struct mddev *mddev, const char *buf, size_t len)
3818 {
3819 	unsigned int n;
3820 	int err;
3821 
3822 	err = kstrtouint(buf, 10, &n);
3823 	if (err < 0)
3824 		return err;
3825 	err = mddev_lock(mddev);
3826 	if (err)
3827 		return err;
3828 
3829 	if (mddev->pers) {
3830 		if (mddev->pers->check_reshape == NULL)
3831 			err = -EBUSY;
3832 		else if (mddev->ro)
3833 			err = -EROFS;
3834 		else {
3835 			mddev->new_layout = n;
3836 			err = mddev->pers->check_reshape(mddev);
3837 			if (err)
3838 				mddev->new_layout = mddev->layout;
3839 		}
3840 	} else {
3841 		mddev->new_layout = n;
3842 		if (mddev->reshape_position == MaxSector)
3843 			mddev->layout = n;
3844 	}
3845 	mddev_unlock(mddev);
3846 	return err ?: len;
3847 }
3848 static struct md_sysfs_entry md_layout =
3849 __ATTR(layout, S_IRUGO|S_IWUSR, layout_show, layout_store);
3850 
3851 static ssize_t
3852 raid_disks_show(struct mddev *mddev, char *page)
3853 {
3854 	if (mddev->raid_disks == 0)
3855 		return 0;
3856 	if (mddev->reshape_position != MaxSector &&
3857 	    mddev->delta_disks != 0)
3858 		return sprintf(page, "%d (%d)\n", mddev->raid_disks,
3859 			       mddev->raid_disks - mddev->delta_disks);
3860 	return sprintf(page, "%d\n", mddev->raid_disks);
3861 }
3862 
3863 static int update_raid_disks(struct mddev *mddev, int raid_disks);
3864 
3865 static ssize_t
3866 raid_disks_store(struct mddev *mddev, const char *buf, size_t len)
3867 {
3868 	unsigned int n;
3869 	int err;
3870 
3871 	err = kstrtouint(buf, 10, &n);
3872 	if (err < 0)
3873 		return err;
3874 
3875 	err = mddev_lock(mddev);
3876 	if (err)
3877 		return err;
3878 	if (mddev->pers)
3879 		err = update_raid_disks(mddev, n);
3880 	else if (mddev->reshape_position != MaxSector) {
3881 		struct md_rdev *rdev;
3882 		int olddisks = mddev->raid_disks - mddev->delta_disks;
3883 
3884 		err = -EINVAL;
3885 		rdev_for_each(rdev, mddev) {
3886 			if (olddisks < n &&
3887 			    rdev->data_offset < rdev->new_data_offset)
3888 				goto out_unlock;
3889 			if (olddisks > n &&
3890 			    rdev->data_offset > rdev->new_data_offset)
3891 				goto out_unlock;
3892 		}
3893 		err = 0;
3894 		mddev->delta_disks = n - olddisks;
3895 		mddev->raid_disks = n;
3896 		mddev->reshape_backwards = (mddev->delta_disks < 0);
3897 	} else
3898 		mddev->raid_disks = n;
3899 out_unlock:
3900 	mddev_unlock(mddev);
3901 	return err ? err : len;
3902 }
3903 static struct md_sysfs_entry md_raid_disks =
3904 __ATTR(raid_disks, S_IRUGO|S_IWUSR, raid_disks_show, raid_disks_store);
3905 
3906 static ssize_t
3907 chunk_size_show(struct mddev *mddev, char *page)
3908 {
3909 	if (mddev->reshape_position != MaxSector &&
3910 	    mddev->chunk_sectors != mddev->new_chunk_sectors)
3911 		return sprintf(page, "%d (%d)\n",
3912 			       mddev->new_chunk_sectors << 9,
3913 			       mddev->chunk_sectors << 9);
3914 	return sprintf(page, "%d\n", mddev->chunk_sectors << 9);
3915 }
3916 
3917 static ssize_t
3918 chunk_size_store(struct mddev *mddev, const char *buf, size_t len)
3919 {
3920 	unsigned long n;
3921 	int err;
3922 
3923 	err = kstrtoul(buf, 10, &n);
3924 	if (err < 0)
3925 		return err;
3926 
3927 	err = mddev_lock(mddev);
3928 	if (err)
3929 		return err;
3930 	if (mddev->pers) {
3931 		if (mddev->pers->check_reshape == NULL)
3932 			err = -EBUSY;
3933 		else if (mddev->ro)
3934 			err = -EROFS;
3935 		else {
3936 			mddev->new_chunk_sectors = n >> 9;
3937 			err = mddev->pers->check_reshape(mddev);
3938 			if (err)
3939 				mddev->new_chunk_sectors = mddev->chunk_sectors;
3940 		}
3941 	} else {
3942 		mddev->new_chunk_sectors = n >> 9;
3943 		if (mddev->reshape_position == MaxSector)
3944 			mddev->chunk_sectors = n >> 9;
3945 	}
3946 	mddev_unlock(mddev);
3947 	return err ?: len;
3948 }
3949 static struct md_sysfs_entry md_chunk_size =
3950 __ATTR(chunk_size, S_IRUGO|S_IWUSR, chunk_size_show, chunk_size_store);
3951 
3952 static ssize_t
3953 resync_start_show(struct mddev *mddev, char *page)
3954 {
3955 	if (mddev->recovery_cp == MaxSector)
3956 		return sprintf(page, "none\n");
3957 	return sprintf(page, "%llu\n", (unsigned long long)mddev->recovery_cp);
3958 }
3959 
3960 static ssize_t
3961 resync_start_store(struct mddev *mddev, const char *buf, size_t len)
3962 {
3963 	unsigned long long n;
3964 	int err;
3965 
3966 	if (cmd_match(buf, "none"))
3967 		n = MaxSector;
3968 	else {
3969 		err = kstrtoull(buf, 10, &n);
3970 		if (err < 0)
3971 			return err;
3972 		if (n != (sector_t)n)
3973 			return -EINVAL;
3974 	}
3975 
3976 	err = mddev_lock(mddev);
3977 	if (err)
3978 		return err;
3979 	if (mddev->pers && !test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
3980 		err = -EBUSY;
3981 
3982 	if (!err) {
3983 		mddev->recovery_cp = n;
3984 		if (mddev->pers)
3985 			set_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags);
3986 	}
3987 	mddev_unlock(mddev);
3988 	return err ?: len;
3989 }
3990 static struct md_sysfs_entry md_resync_start =
3991 __ATTR_PREALLOC(resync_start, S_IRUGO|S_IWUSR,
3992 		resync_start_show, resync_start_store);
3993 
3994 /*
3995  * The array state can be:
3996  *
3997  * clear
3998  *     No devices, no size, no level
3999  *     Equivalent to STOP_ARRAY ioctl
4000  * inactive
4001  *     May have some settings, but array is not active
4002  *        all IO results in error
4003  *     When written, doesn't tear down array, but just stops it
4004  * suspended (not supported yet)
4005  *     All IO requests will block. The array can be reconfigured.
4006  *     Writing this, if accepted, will block until array is quiescent
4007  * readonly
4008  *     no resync can happen.  no superblocks get written.
4009  *     write requests fail
4010  * read-auto
4011  *     like readonly, but behaves like 'clean' on a write request.
4012  *
4013  * clean - no pending writes, but otherwise active.
4014  *     When written to inactive array, starts without resync
4015  *     If a write request arrives then
4016  *       if metadata is known, mark 'dirty' and switch to 'active'.
4017  *       if not known, block and switch to write-pending
4018  *     If written to an active array that has pending writes, then fails.
4019  * active
4020  *     fully active: IO and resync can be happening.
4021  *     When written to inactive array, starts with resync
4022  *
4023  * write-pending
4024  *     clean, but writes are blocked waiting for 'active' to be written.
4025  *
4026  * active-idle
4027  *     like active, but no writes have been seen for a while (100msec).
4028  *
4029  */
4030 enum array_state { clear, inactive, suspended, readonly, read_auto, clean, active,
4031 		   write_pending, active_idle, bad_word};
4032 static char *array_states[] = {
4033 	"clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active",
4034 	"write-pending", "active-idle", NULL };
4035 
4036 static int match_word(const char *word, char **list)
4037 {
4038 	int n;
4039 	for (n=0; list[n]; n++)
4040 		if (cmd_match(word, list[n]))
4041 			break;
4042 	return n;
4043 }
4044 
4045 static ssize_t
4046 array_state_show(struct mddev *mddev, char *page)
4047 {
4048 	enum array_state st = inactive;
4049 
4050 	if (mddev->pers)
4051 		switch(mddev->ro) {
4052 		case 1:
4053 			st = readonly;
4054 			break;
4055 		case 2:
4056 			st = read_auto;
4057 			break;
4058 		case 0:
4059 			spin_lock(&mddev->lock);
4060 			if (test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags))
4061 				st = write_pending;
4062 			else if (mddev->in_sync)
4063 				st = clean;
4064 			else if (mddev->safemode)
4065 				st = active_idle;
4066 			else
4067 				st = active;
4068 			spin_unlock(&mddev->lock);
4069 		}
4070 	else {
4071 		if (list_empty(&mddev->disks) &&
4072 		    mddev->raid_disks == 0 &&
4073 		    mddev->dev_sectors == 0)
4074 			st = clear;
4075 		else
4076 			st = inactive;
4077 	}
4078 	return sprintf(page, "%s\n", array_states[st]);
4079 }
4080 
4081 static int do_md_stop(struct mddev *mddev, int ro, struct block_device *bdev);
4082 static int md_set_readonly(struct mddev *mddev, struct block_device *bdev);
4083 static int do_md_run(struct mddev *mddev);
4084 static int restart_array(struct mddev *mddev);
4085 
4086 static ssize_t
4087 array_state_store(struct mddev *mddev, const char *buf, size_t len)
4088 {
4089 	int err = 0;
4090 	enum array_state st = match_word(buf, array_states);
4091 
4092 	if (mddev->pers && (st == active || st == clean) && mddev->ro != 1) {
4093 		/* don't take reconfig_mutex when toggling between
4094 		 * clean and active
4095 		 */
4096 		spin_lock(&mddev->lock);
4097 		if (st == active) {
4098 			restart_array(mddev);
4099 			clear_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags);
4100 			md_wakeup_thread(mddev->thread);
4101 			wake_up(&mddev->sb_wait);
4102 		} else /* st == clean */ {
4103 			restart_array(mddev);
4104 			if (!set_in_sync(mddev))
4105 				err = -EBUSY;
4106 		}
4107 		if (!err)
4108 			sysfs_notify_dirent_safe(mddev->sysfs_state);
4109 		spin_unlock(&mddev->lock);
4110 		return err ?: len;
4111 	}
4112 	err = mddev_lock(mddev);
4113 	if (err)
4114 		return err;
4115 	err = -EINVAL;
4116 	switch(st) {
4117 	case bad_word:
4118 		break;
4119 	case clear:
4120 		/* stopping an active array */
4121 		err = do_md_stop(mddev, 0, NULL);
4122 		break;
4123 	case inactive:
4124 		/* stopping an active array */
4125 		if (mddev->pers)
4126 			err = do_md_stop(mddev, 2, NULL);
4127 		else
4128 			err = 0; /* already inactive */
4129 		break;
4130 	case suspended:
4131 		break; /* not supported yet */
4132 	case readonly:
4133 		if (mddev->pers)
4134 			err = md_set_readonly(mddev, NULL);
4135 		else {
4136 			mddev->ro = 1;
4137 			set_disk_ro(mddev->gendisk, 1);
4138 			err = do_md_run(mddev);
4139 		}
4140 		break;
4141 	case read_auto:
4142 		if (mddev->pers) {
4143 			if (mddev->ro == 0)
4144 				err = md_set_readonly(mddev, NULL);
4145 			else if (mddev->ro == 1)
4146 				err = restart_array(mddev);
4147 			if (err == 0) {
4148 				mddev->ro = 2;
4149 				set_disk_ro(mddev->gendisk, 0);
4150 			}
4151 		} else {
4152 			mddev->ro = 2;
4153 			err = do_md_run(mddev);
4154 		}
4155 		break;
4156 	case clean:
4157 		if (mddev->pers) {
4158 			err = restart_array(mddev);
4159 			if (err)
4160 				break;
4161 			spin_lock(&mddev->lock);
4162 			if (!set_in_sync(mddev))
4163 				err = -EBUSY;
4164 			spin_unlock(&mddev->lock);
4165 		} else
4166 			err = -EINVAL;
4167 		break;
4168 	case active:
4169 		if (mddev->pers) {
4170 			err = restart_array(mddev);
4171 			if (err)
4172 				break;
4173 			clear_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags);
4174 			wake_up(&mddev->sb_wait);
4175 			err = 0;
4176 		} else {
4177 			mddev->ro = 0;
4178 			set_disk_ro(mddev->gendisk, 0);
4179 			err = do_md_run(mddev);
4180 		}
4181 		break;
4182 	case write_pending:
4183 	case active_idle:
4184 		/* these cannot be set */
4185 		break;
4186 	}
4187 
4188 	if (!err) {
4189 		if (mddev->hold_active == UNTIL_IOCTL)
4190 			mddev->hold_active = 0;
4191 		sysfs_notify_dirent_safe(mddev->sysfs_state);
4192 	}
4193 	mddev_unlock(mddev);
4194 	return err ?: len;
4195 }
4196 static struct md_sysfs_entry md_array_state =
4197 __ATTR_PREALLOC(array_state, S_IRUGO|S_IWUSR, array_state_show, array_state_store);
4198 
4199 static ssize_t
4200 max_corrected_read_errors_show(struct mddev *mddev, char *page) {
4201 	return sprintf(page, "%d\n",
4202 		       atomic_read(&mddev->max_corr_read_errors));
4203 }
4204 
4205 static ssize_t
4206 max_corrected_read_errors_store(struct mddev *mddev, const char *buf, size_t len)
4207 {
4208 	unsigned int n;
4209 	int rv;
4210 
4211 	rv = kstrtouint(buf, 10, &n);
4212 	if (rv < 0)
4213 		return rv;
4214 	atomic_set(&mddev->max_corr_read_errors, n);
4215 	return len;
4216 }
4217 
4218 static struct md_sysfs_entry max_corr_read_errors =
4219 __ATTR(max_read_errors, S_IRUGO|S_IWUSR, max_corrected_read_errors_show,
4220 	max_corrected_read_errors_store);
4221 
4222 static ssize_t
4223 null_show(struct mddev *mddev, char *page)
4224 {
4225 	return -EINVAL;
4226 }
4227 
4228 static ssize_t
4229 new_dev_store(struct mddev *mddev, const char *buf, size_t len)
4230 {
4231 	/* buf must be %d:%d\n? giving major and minor numbers */
4232 	/* The new device is added to the array.
4233 	 * If the array has a persistent superblock, we read the
4234 	 * superblock to initialise info and check validity.
4235 	 * Otherwise, only checking done is that in bind_rdev_to_array,
4236 	 * which mainly checks size.
4237 	 */
4238 	char *e;
4239 	int major = simple_strtoul(buf, &e, 10);
4240 	int minor;
4241 	dev_t dev;
4242 	struct md_rdev *rdev;
4243 	int err;
4244 
4245 	if (!*buf || *e != ':' || !e[1] || e[1] == '\n')
4246 		return -EINVAL;
4247 	minor = simple_strtoul(e+1, &e, 10);
4248 	if (*e && *e != '\n')
4249 		return -EINVAL;
4250 	dev = MKDEV(major, minor);
4251 	if (major != MAJOR(dev) ||
4252 	    minor != MINOR(dev))
4253 		return -EOVERFLOW;
4254 
4255 	flush_workqueue(md_misc_wq);
4256 
4257 	err = mddev_lock(mddev);
4258 	if (err)
4259 		return err;
4260 	if (mddev->persistent) {
4261 		rdev = md_import_device(dev, mddev->major_version,
4262 					mddev->minor_version);
4263 		if (!IS_ERR(rdev) && !list_empty(&mddev->disks)) {
4264 			struct md_rdev *rdev0
4265 				= list_entry(mddev->disks.next,
4266 					     struct md_rdev, same_set);
4267 			err = super_types[mddev->major_version]
4268 				.load_super(rdev, rdev0, mddev->minor_version);
4269 			if (err < 0)
4270 				goto out;
4271 		}
4272 	} else if (mddev->external)
4273 		rdev = md_import_device(dev, -2, -1);
4274 	else
4275 		rdev = md_import_device(dev, -1, -1);
4276 
4277 	if (IS_ERR(rdev)) {
4278 		mddev_unlock(mddev);
4279 		return PTR_ERR(rdev);
4280 	}
4281 	err = bind_rdev_to_array(rdev, mddev);
4282  out:
4283 	if (err)
4284 		export_rdev(rdev);
4285 	mddev_unlock(mddev);
4286 	return err ? err : len;
4287 }
4288 
4289 static struct md_sysfs_entry md_new_device =
4290 __ATTR(new_dev, S_IWUSR, null_show, new_dev_store);
4291 
4292 static ssize_t
4293 bitmap_store(struct mddev *mddev, const char *buf, size_t len)
4294 {
4295 	char *end;
4296 	unsigned long chunk, end_chunk;
4297 	int err;
4298 
4299 	err = mddev_lock(mddev);
4300 	if (err)
4301 		return err;
4302 	if (!mddev->bitmap)
4303 		goto out;
4304 	/* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */
4305 	while (*buf) {
4306 		chunk = end_chunk = simple_strtoul(buf, &end, 0);
4307 		if (buf == end) break;
4308 		if (*end == '-') { /* range */
4309 			buf = end + 1;
4310 			end_chunk = simple_strtoul(buf, &end, 0);
4311 			if (buf == end) break;
4312 		}
4313 		if (*end && !isspace(*end)) break;
4314 		bitmap_dirty_bits(mddev->bitmap, chunk, end_chunk);
4315 		buf = skip_spaces(end);
4316 	}
4317 	bitmap_unplug(mddev->bitmap); /* flush the bits to disk */
4318 out:
4319 	mddev_unlock(mddev);
4320 	return len;
4321 }
4322 
4323 static struct md_sysfs_entry md_bitmap =
4324 __ATTR(bitmap_set_bits, S_IWUSR, null_show, bitmap_store);
4325 
4326 static ssize_t
4327 size_show(struct mddev *mddev, char *page)
4328 {
4329 	return sprintf(page, "%llu\n",
4330 		(unsigned long long)mddev->dev_sectors / 2);
4331 }
4332 
4333 static int update_size(struct mddev *mddev, sector_t num_sectors);
4334 
4335 static ssize_t
4336 size_store(struct mddev *mddev, const char *buf, size_t len)
4337 {
4338 	/* If array is inactive, we can reduce the component size, but
4339 	 * not increase it (except from 0).
4340 	 * If array is active, we can try an on-line resize
4341 	 */
4342 	sector_t sectors;
4343 	int err = strict_blocks_to_sectors(buf, &sectors);
4344 
4345 	if (err < 0)
4346 		return err;
4347 	err = mddev_lock(mddev);
4348 	if (err)
4349 		return err;
4350 	if (mddev->pers) {
4351 		err = update_size(mddev, sectors);
4352 		if (err == 0)
4353 			md_update_sb(mddev, 1);
4354 	} else {
4355 		if (mddev->dev_sectors == 0 ||
4356 		    mddev->dev_sectors > sectors)
4357 			mddev->dev_sectors = sectors;
4358 		else
4359 			err = -ENOSPC;
4360 	}
4361 	mddev_unlock(mddev);
4362 	return err ? err : len;
4363 }
4364 
4365 static struct md_sysfs_entry md_size =
4366 __ATTR(component_size, S_IRUGO|S_IWUSR, size_show, size_store);
4367 
4368 /* Metadata version.
4369  * This is one of
4370  *   'none' for arrays with no metadata (good luck...)
4371  *   'external' for arrays with externally managed metadata,
4372  * or N.M for internally known formats
4373  */
4374 static ssize_t
4375 metadata_show(struct mddev *mddev, char *page)
4376 {
4377 	if (mddev->persistent)
4378 		return sprintf(page, "%d.%d\n",
4379 			       mddev->major_version, mddev->minor_version);
4380 	else if (mddev->external)
4381 		return sprintf(page, "external:%s\n", mddev->metadata_type);
4382 	else
4383 		return sprintf(page, "none\n");
4384 }
4385 
4386 static ssize_t
4387 metadata_store(struct mddev *mddev, const char *buf, size_t len)
4388 {
4389 	int major, minor;
4390 	char *e;
4391 	int err;
4392 	/* Changing the details of 'external' metadata is
4393 	 * always permitted.  Otherwise there must be
4394 	 * no devices attached to the array.
4395 	 */
4396 
4397 	err = mddev_lock(mddev);
4398 	if (err)
4399 		return err;
4400 	err = -EBUSY;
4401 	if (mddev->external && strncmp(buf, "external:", 9) == 0)
4402 		;
4403 	else if (!list_empty(&mddev->disks))
4404 		goto out_unlock;
4405 
4406 	err = 0;
4407 	if (cmd_match(buf, "none")) {
4408 		mddev->persistent = 0;
4409 		mddev->external = 0;
4410 		mddev->major_version = 0;
4411 		mddev->minor_version = 90;
4412 		goto out_unlock;
4413 	}
4414 	if (strncmp(buf, "external:", 9) == 0) {
4415 		size_t namelen = len-9;
4416 		if (namelen >= sizeof(mddev->metadata_type))
4417 			namelen = sizeof(mddev->metadata_type)-1;
4418 		strncpy(mddev->metadata_type, buf+9, namelen);
4419 		mddev->metadata_type[namelen] = 0;
4420 		if (namelen && mddev->metadata_type[namelen-1] == '\n')
4421 			mddev->metadata_type[--namelen] = 0;
4422 		mddev->persistent = 0;
4423 		mddev->external = 1;
4424 		mddev->major_version = 0;
4425 		mddev->minor_version = 90;
4426 		goto out_unlock;
4427 	}
4428 	major = simple_strtoul(buf, &e, 10);
4429 	err = -EINVAL;
4430 	if (e==buf || *e != '.')
4431 		goto out_unlock;
4432 	buf = e+1;
4433 	minor = simple_strtoul(buf, &e, 10);
4434 	if (e==buf || (*e && *e != '\n') )
4435 		goto out_unlock;
4436 	err = -ENOENT;
4437 	if (major >= ARRAY_SIZE(super_types) || super_types[major].name == NULL)
4438 		goto out_unlock;
4439 	mddev->major_version = major;
4440 	mddev->minor_version = minor;
4441 	mddev->persistent = 1;
4442 	mddev->external = 0;
4443 	err = 0;
4444 out_unlock:
4445 	mddev_unlock(mddev);
4446 	return err ?: len;
4447 }
4448 
4449 static struct md_sysfs_entry md_metadata =
4450 __ATTR_PREALLOC(metadata_version, S_IRUGO|S_IWUSR, metadata_show, metadata_store);
4451 
4452 static ssize_t
4453 action_show(struct mddev *mddev, char *page)
4454 {
4455 	char *type = "idle";
4456 	unsigned long recovery = mddev->recovery;
4457 	if (test_bit(MD_RECOVERY_FROZEN, &recovery))
4458 		type = "frozen";
4459 	else if (test_bit(MD_RECOVERY_RUNNING, &recovery) ||
4460 	    (!mddev->ro && test_bit(MD_RECOVERY_NEEDED, &recovery))) {
4461 		if (test_bit(MD_RECOVERY_RESHAPE, &recovery))
4462 			type = "reshape";
4463 		else if (test_bit(MD_RECOVERY_SYNC, &recovery)) {
4464 			if (!test_bit(MD_RECOVERY_REQUESTED, &recovery))
4465 				type = "resync";
4466 			else if (test_bit(MD_RECOVERY_CHECK, &recovery))
4467 				type = "check";
4468 			else
4469 				type = "repair";
4470 		} else if (test_bit(MD_RECOVERY_RECOVER, &recovery))
4471 			type = "recover";
4472 		else if (mddev->reshape_position != MaxSector)
4473 			type = "reshape";
4474 	}
4475 	return sprintf(page, "%s\n", type);
4476 }
4477 
4478 static ssize_t
4479 action_store(struct mddev *mddev, const char *page, size_t len)
4480 {
4481 	if (!mddev->pers || !mddev->pers->sync_request)
4482 		return -EINVAL;
4483 
4484 
4485 	if (cmd_match(page, "idle") || cmd_match(page, "frozen")) {
4486 		if (cmd_match(page, "frozen"))
4487 			set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4488 		else
4489 			clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4490 		if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
4491 		    mddev_lock(mddev) == 0) {
4492 			flush_workqueue(md_misc_wq);
4493 			if (mddev->sync_thread) {
4494 				set_bit(MD_RECOVERY_INTR, &mddev->recovery);
4495 				md_reap_sync_thread(mddev);
4496 			}
4497 			mddev_unlock(mddev);
4498 		}
4499 	} else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4500 		return -EBUSY;
4501 	else if (cmd_match(page, "resync"))
4502 		clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4503 	else if (cmd_match(page, "recover")) {
4504 		clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4505 		set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
4506 	} else if (cmd_match(page, "reshape")) {
4507 		int err;
4508 		if (mddev->pers->start_reshape == NULL)
4509 			return -EINVAL;
4510 		err = mddev_lock(mddev);
4511 		if (!err) {
4512 			if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4513 				err =  -EBUSY;
4514 			else {
4515 				clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4516 				err = mddev->pers->start_reshape(mddev);
4517 			}
4518 			mddev_unlock(mddev);
4519 		}
4520 		if (err)
4521 			return err;
4522 		sysfs_notify(&mddev->kobj, NULL, "degraded");
4523 	} else {
4524 		if (cmd_match(page, "check"))
4525 			set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
4526 		else if (!cmd_match(page, "repair"))
4527 			return -EINVAL;
4528 		clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4529 		set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
4530 		set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
4531 	}
4532 	if (mddev->ro == 2) {
4533 		/* A write to sync_action is enough to justify
4534 		 * canceling read-auto mode
4535 		 */
4536 		mddev->ro = 0;
4537 		md_wakeup_thread(mddev->sync_thread);
4538 	}
4539 	set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4540 	md_wakeup_thread(mddev->thread);
4541 	sysfs_notify_dirent_safe(mddev->sysfs_action);
4542 	return len;
4543 }
4544 
4545 static struct md_sysfs_entry md_scan_mode =
4546 __ATTR_PREALLOC(sync_action, S_IRUGO|S_IWUSR, action_show, action_store);
4547 
4548 static ssize_t
4549 last_sync_action_show(struct mddev *mddev, char *page)
4550 {
4551 	return sprintf(page, "%s\n", mddev->last_sync_action);
4552 }
4553 
4554 static struct md_sysfs_entry md_last_scan_mode = __ATTR_RO(last_sync_action);
4555 
4556 static ssize_t
4557 mismatch_cnt_show(struct mddev *mddev, char *page)
4558 {
4559 	return sprintf(page, "%llu\n",
4560 		       (unsigned long long)
4561 		       atomic64_read(&mddev->resync_mismatches));
4562 }
4563 
4564 static struct md_sysfs_entry md_mismatches = __ATTR_RO(mismatch_cnt);
4565 
4566 static ssize_t
4567 sync_min_show(struct mddev *mddev, char *page)
4568 {
4569 	return sprintf(page, "%d (%s)\n", speed_min(mddev),
4570 		       mddev->sync_speed_min ? "local": "system");
4571 }
4572 
4573 static ssize_t
4574 sync_min_store(struct mddev *mddev, const char *buf, size_t len)
4575 {
4576 	unsigned int min;
4577 	int rv;
4578 
4579 	if (strncmp(buf, "system", 6)==0) {
4580 		min = 0;
4581 	} else {
4582 		rv = kstrtouint(buf, 10, &min);
4583 		if (rv < 0)
4584 			return rv;
4585 		if (min == 0)
4586 			return -EINVAL;
4587 	}
4588 	mddev->sync_speed_min = min;
4589 	return len;
4590 }
4591 
4592 static struct md_sysfs_entry md_sync_min =
4593 __ATTR(sync_speed_min, S_IRUGO|S_IWUSR, sync_min_show, sync_min_store);
4594 
4595 static ssize_t
4596 sync_max_show(struct mddev *mddev, char *page)
4597 {
4598 	return sprintf(page, "%d (%s)\n", speed_max(mddev),
4599 		       mddev->sync_speed_max ? "local": "system");
4600 }
4601 
4602 static ssize_t
4603 sync_max_store(struct mddev *mddev, const char *buf, size_t len)
4604 {
4605 	unsigned int max;
4606 	int rv;
4607 
4608 	if (strncmp(buf, "system", 6)==0) {
4609 		max = 0;
4610 	} else {
4611 		rv = kstrtouint(buf, 10, &max);
4612 		if (rv < 0)
4613 			return rv;
4614 		if (max == 0)
4615 			return -EINVAL;
4616 	}
4617 	mddev->sync_speed_max = max;
4618 	return len;
4619 }
4620 
4621 static struct md_sysfs_entry md_sync_max =
4622 __ATTR(sync_speed_max, S_IRUGO|S_IWUSR, sync_max_show, sync_max_store);
4623 
4624 static ssize_t
4625 degraded_show(struct mddev *mddev, char *page)
4626 {
4627 	return sprintf(page, "%d\n", mddev->degraded);
4628 }
4629 static struct md_sysfs_entry md_degraded = __ATTR_RO(degraded);
4630 
4631 static ssize_t
4632 sync_force_parallel_show(struct mddev *mddev, char *page)
4633 {
4634 	return sprintf(page, "%d\n", mddev->parallel_resync);
4635 }
4636 
4637 static ssize_t
4638 sync_force_parallel_store(struct mddev *mddev, const char *buf, size_t len)
4639 {
4640 	long n;
4641 
4642 	if (kstrtol(buf, 10, &n))
4643 		return -EINVAL;
4644 
4645 	if (n != 0 && n != 1)
4646 		return -EINVAL;
4647 
4648 	mddev->parallel_resync = n;
4649 
4650 	if (mddev->sync_thread)
4651 		wake_up(&resync_wait);
4652 
4653 	return len;
4654 }
4655 
4656 /* force parallel resync, even with shared block devices */
4657 static struct md_sysfs_entry md_sync_force_parallel =
4658 __ATTR(sync_force_parallel, S_IRUGO|S_IWUSR,
4659        sync_force_parallel_show, sync_force_parallel_store);
4660 
4661 static ssize_t
4662 sync_speed_show(struct mddev *mddev, char *page)
4663 {
4664 	unsigned long resync, dt, db;
4665 	if (mddev->curr_resync == 0)
4666 		return sprintf(page, "none\n");
4667 	resync = mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active);
4668 	dt = (jiffies - mddev->resync_mark) / HZ;
4669 	if (!dt) dt++;
4670 	db = resync - mddev->resync_mark_cnt;
4671 	return sprintf(page, "%lu\n", db/dt/2); /* K/sec */
4672 }
4673 
4674 static struct md_sysfs_entry md_sync_speed = __ATTR_RO(sync_speed);
4675 
4676 static ssize_t
4677 sync_completed_show(struct mddev *mddev, char *page)
4678 {
4679 	unsigned long long max_sectors, resync;
4680 
4681 	if (!test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4682 		return sprintf(page, "none\n");
4683 
4684 	if (mddev->curr_resync == 1 ||
4685 	    mddev->curr_resync == 2)
4686 		return sprintf(page, "delayed\n");
4687 
4688 	if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ||
4689 	    test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
4690 		max_sectors = mddev->resync_max_sectors;
4691 	else
4692 		max_sectors = mddev->dev_sectors;
4693 
4694 	resync = mddev->curr_resync_completed;
4695 	return sprintf(page, "%llu / %llu\n", resync, max_sectors);
4696 }
4697 
4698 static struct md_sysfs_entry md_sync_completed =
4699 	__ATTR_PREALLOC(sync_completed, S_IRUGO, sync_completed_show, NULL);
4700 
4701 static ssize_t
4702 min_sync_show(struct mddev *mddev, char *page)
4703 {
4704 	return sprintf(page, "%llu\n",
4705 		       (unsigned long long)mddev->resync_min);
4706 }
4707 static ssize_t
4708 min_sync_store(struct mddev *mddev, const char *buf, size_t len)
4709 {
4710 	unsigned long long min;
4711 	int err;
4712 
4713 	if (kstrtoull(buf, 10, &min))
4714 		return -EINVAL;
4715 
4716 	spin_lock(&mddev->lock);
4717 	err = -EINVAL;
4718 	if (min > mddev->resync_max)
4719 		goto out_unlock;
4720 
4721 	err = -EBUSY;
4722 	if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4723 		goto out_unlock;
4724 
4725 	/* Round down to multiple of 4K for safety */
4726 	mddev->resync_min = round_down(min, 8);
4727 	err = 0;
4728 
4729 out_unlock:
4730 	spin_unlock(&mddev->lock);
4731 	return err ?: len;
4732 }
4733 
4734 static struct md_sysfs_entry md_min_sync =
4735 __ATTR(sync_min, S_IRUGO|S_IWUSR, min_sync_show, min_sync_store);
4736 
4737 static ssize_t
4738 max_sync_show(struct mddev *mddev, char *page)
4739 {
4740 	if (mddev->resync_max == MaxSector)
4741 		return sprintf(page, "max\n");
4742 	else
4743 		return sprintf(page, "%llu\n",
4744 			       (unsigned long long)mddev->resync_max);
4745 }
4746 static ssize_t
4747 max_sync_store(struct mddev *mddev, const char *buf, size_t len)
4748 {
4749 	int err;
4750 	spin_lock(&mddev->lock);
4751 	if (strncmp(buf, "max", 3) == 0)
4752 		mddev->resync_max = MaxSector;
4753 	else {
4754 		unsigned long long max;
4755 		int chunk;
4756 
4757 		err = -EINVAL;
4758 		if (kstrtoull(buf, 10, &max))
4759 			goto out_unlock;
4760 		if (max < mddev->resync_min)
4761 			goto out_unlock;
4762 
4763 		err = -EBUSY;
4764 		if (max < mddev->resync_max &&
4765 		    mddev->ro == 0 &&
4766 		    test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4767 			goto out_unlock;
4768 
4769 		/* Must be a multiple of chunk_size */
4770 		chunk = mddev->chunk_sectors;
4771 		if (chunk) {
4772 			sector_t temp = max;
4773 
4774 			err = -EINVAL;
4775 			if (sector_div(temp, chunk))
4776 				goto out_unlock;
4777 		}
4778 		mddev->resync_max = max;
4779 	}
4780 	wake_up(&mddev->recovery_wait);
4781 	err = 0;
4782 out_unlock:
4783 	spin_unlock(&mddev->lock);
4784 	return err ?: len;
4785 }
4786 
4787 static struct md_sysfs_entry md_max_sync =
4788 __ATTR(sync_max, S_IRUGO|S_IWUSR, max_sync_show, max_sync_store);
4789 
4790 static ssize_t
4791 suspend_lo_show(struct mddev *mddev, char *page)
4792 {
4793 	return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_lo);
4794 }
4795 
4796 static ssize_t
4797 suspend_lo_store(struct mddev *mddev, const char *buf, size_t len)
4798 {
4799 	unsigned long long old, new;
4800 	int err;
4801 
4802 	err = kstrtoull(buf, 10, &new);
4803 	if (err < 0)
4804 		return err;
4805 	if (new != (sector_t)new)
4806 		return -EINVAL;
4807 
4808 	err = mddev_lock(mddev);
4809 	if (err)
4810 		return err;
4811 	err = -EINVAL;
4812 	if (mddev->pers == NULL ||
4813 	    mddev->pers->quiesce == NULL)
4814 		goto unlock;
4815 	old = mddev->suspend_lo;
4816 	mddev->suspend_lo = new;
4817 	if (new >= old)
4818 		/* Shrinking suspended region */
4819 		mddev->pers->quiesce(mddev, 2);
4820 	else {
4821 		/* Expanding suspended region - need to wait */
4822 		mddev->pers->quiesce(mddev, 1);
4823 		mddev->pers->quiesce(mddev, 0);
4824 	}
4825 	err = 0;
4826 unlock:
4827 	mddev_unlock(mddev);
4828 	return err ?: len;
4829 }
4830 static struct md_sysfs_entry md_suspend_lo =
4831 __ATTR(suspend_lo, S_IRUGO|S_IWUSR, suspend_lo_show, suspend_lo_store);
4832 
4833 static ssize_t
4834 suspend_hi_show(struct mddev *mddev, char *page)
4835 {
4836 	return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_hi);
4837 }
4838 
4839 static ssize_t
4840 suspend_hi_store(struct mddev *mddev, const char *buf, size_t len)
4841 {
4842 	unsigned long long old, new;
4843 	int err;
4844 
4845 	err = kstrtoull(buf, 10, &new);
4846 	if (err < 0)
4847 		return err;
4848 	if (new != (sector_t)new)
4849 		return -EINVAL;
4850 
4851 	err = mddev_lock(mddev);
4852 	if (err)
4853 		return err;
4854 	err = -EINVAL;
4855 	if (mddev->pers == NULL ||
4856 	    mddev->pers->quiesce == NULL)
4857 		goto unlock;
4858 	old = mddev->suspend_hi;
4859 	mddev->suspend_hi = new;
4860 	if (new <= old)
4861 		/* Shrinking suspended region */
4862 		mddev->pers->quiesce(mddev, 2);
4863 	else {
4864 		/* Expanding suspended region - need to wait */
4865 		mddev->pers->quiesce(mddev, 1);
4866 		mddev->pers->quiesce(mddev, 0);
4867 	}
4868 	err = 0;
4869 unlock:
4870 	mddev_unlock(mddev);
4871 	return err ?: len;
4872 }
4873 static struct md_sysfs_entry md_suspend_hi =
4874 __ATTR(suspend_hi, S_IRUGO|S_IWUSR, suspend_hi_show, suspend_hi_store);
4875 
4876 static ssize_t
4877 reshape_position_show(struct mddev *mddev, char *page)
4878 {
4879 	if (mddev->reshape_position != MaxSector)
4880 		return sprintf(page, "%llu\n",
4881 			       (unsigned long long)mddev->reshape_position);
4882 	strcpy(page, "none\n");
4883 	return 5;
4884 }
4885 
4886 static ssize_t
4887 reshape_position_store(struct mddev *mddev, const char *buf, size_t len)
4888 {
4889 	struct md_rdev *rdev;
4890 	unsigned long long new;
4891 	int err;
4892 
4893 	err = kstrtoull(buf, 10, &new);
4894 	if (err < 0)
4895 		return err;
4896 	if (new != (sector_t)new)
4897 		return -EINVAL;
4898 	err = mddev_lock(mddev);
4899 	if (err)
4900 		return err;
4901 	err = -EBUSY;
4902 	if (mddev->pers)
4903 		goto unlock;
4904 	mddev->reshape_position = new;
4905 	mddev->delta_disks = 0;
4906 	mddev->reshape_backwards = 0;
4907 	mddev->new_level = mddev->level;
4908 	mddev->new_layout = mddev->layout;
4909 	mddev->new_chunk_sectors = mddev->chunk_sectors;
4910 	rdev_for_each(rdev, mddev)
4911 		rdev->new_data_offset = rdev->data_offset;
4912 	err = 0;
4913 unlock:
4914 	mddev_unlock(mddev);
4915 	return err ?: len;
4916 }
4917 
4918 static struct md_sysfs_entry md_reshape_position =
4919 __ATTR(reshape_position, S_IRUGO|S_IWUSR, reshape_position_show,
4920        reshape_position_store);
4921 
4922 static ssize_t
4923 reshape_direction_show(struct mddev *mddev, char *page)
4924 {
4925 	return sprintf(page, "%s\n",
4926 		       mddev->reshape_backwards ? "backwards" : "forwards");
4927 }
4928 
4929 static ssize_t
4930 reshape_direction_store(struct mddev *mddev, const char *buf, size_t len)
4931 {
4932 	int backwards = 0;
4933 	int err;
4934 
4935 	if (cmd_match(buf, "forwards"))
4936 		backwards = 0;
4937 	else if (cmd_match(buf, "backwards"))
4938 		backwards = 1;
4939 	else
4940 		return -EINVAL;
4941 	if (mddev->reshape_backwards == backwards)
4942 		return len;
4943 
4944 	err = mddev_lock(mddev);
4945 	if (err)
4946 		return err;
4947 	/* check if we are allowed to change */
4948 	if (mddev->delta_disks)
4949 		err = -EBUSY;
4950 	else if (mddev->persistent &&
4951 	    mddev->major_version == 0)
4952 		err =  -EINVAL;
4953 	else
4954 		mddev->reshape_backwards = backwards;
4955 	mddev_unlock(mddev);
4956 	return err ?: len;
4957 }
4958 
4959 static struct md_sysfs_entry md_reshape_direction =
4960 __ATTR(reshape_direction, S_IRUGO|S_IWUSR, reshape_direction_show,
4961        reshape_direction_store);
4962 
4963 static ssize_t
4964 array_size_show(struct mddev *mddev, char *page)
4965 {
4966 	if (mddev->external_size)
4967 		return sprintf(page, "%llu\n",
4968 			       (unsigned long long)mddev->array_sectors/2);
4969 	else
4970 		return sprintf(page, "default\n");
4971 }
4972 
4973 static ssize_t
4974 array_size_store(struct mddev *mddev, const char *buf, size_t len)
4975 {
4976 	sector_t sectors;
4977 	int err;
4978 
4979 	err = mddev_lock(mddev);
4980 	if (err)
4981 		return err;
4982 
4983 	/* cluster raid doesn't support change array_sectors */
4984 	if (mddev_is_clustered(mddev)) {
4985 		mddev_unlock(mddev);
4986 		return -EINVAL;
4987 	}
4988 
4989 	if (strncmp(buf, "default", 7) == 0) {
4990 		if (mddev->pers)
4991 			sectors = mddev->pers->size(mddev, 0, 0);
4992 		else
4993 			sectors = mddev->array_sectors;
4994 
4995 		mddev->external_size = 0;
4996 	} else {
4997 		if (strict_blocks_to_sectors(buf, &sectors) < 0)
4998 			err = -EINVAL;
4999 		else if (mddev->pers && mddev->pers->size(mddev, 0, 0) < sectors)
5000 			err = -E2BIG;
5001 		else
5002 			mddev->external_size = 1;
5003 	}
5004 
5005 	if (!err) {
5006 		mddev->array_sectors = sectors;
5007 		if (mddev->pers) {
5008 			set_capacity(mddev->gendisk, mddev->array_sectors);
5009 			revalidate_disk(mddev->gendisk);
5010 		}
5011 	}
5012 	mddev_unlock(mddev);
5013 	return err ?: len;
5014 }
5015 
5016 static struct md_sysfs_entry md_array_size =
5017 __ATTR(array_size, S_IRUGO|S_IWUSR, array_size_show,
5018        array_size_store);
5019 
5020 static ssize_t
5021 consistency_policy_show(struct mddev *mddev, char *page)
5022 {
5023 	int ret;
5024 
5025 	if (test_bit(MD_HAS_JOURNAL, &mddev->flags)) {
5026 		ret = sprintf(page, "journal\n");
5027 	} else if (test_bit(MD_HAS_PPL, &mddev->flags)) {
5028 		ret = sprintf(page, "ppl\n");
5029 	} else if (mddev->bitmap) {
5030 		ret = sprintf(page, "bitmap\n");
5031 	} else if (mddev->pers) {
5032 		if (mddev->pers->sync_request)
5033 			ret = sprintf(page, "resync\n");
5034 		else
5035 			ret = sprintf(page, "none\n");
5036 	} else {
5037 		ret = sprintf(page, "unknown\n");
5038 	}
5039 
5040 	return ret;
5041 }
5042 
5043 static ssize_t
5044 consistency_policy_store(struct mddev *mddev, const char *buf, size_t len)
5045 {
5046 	int err = 0;
5047 
5048 	if (mddev->pers) {
5049 		if (mddev->pers->change_consistency_policy)
5050 			err = mddev->pers->change_consistency_policy(mddev, buf);
5051 		else
5052 			err = -EBUSY;
5053 	} else if (mddev->external && strncmp(buf, "ppl", 3) == 0) {
5054 		set_bit(MD_HAS_PPL, &mddev->flags);
5055 	} else {
5056 		err = -EINVAL;
5057 	}
5058 
5059 	return err ? err : len;
5060 }
5061 
5062 static struct md_sysfs_entry md_consistency_policy =
5063 __ATTR(consistency_policy, S_IRUGO | S_IWUSR, consistency_policy_show,
5064        consistency_policy_store);
5065 
5066 static struct attribute *md_default_attrs[] = {
5067 	&md_level.attr,
5068 	&md_layout.attr,
5069 	&md_raid_disks.attr,
5070 	&md_chunk_size.attr,
5071 	&md_size.attr,
5072 	&md_resync_start.attr,
5073 	&md_metadata.attr,
5074 	&md_new_device.attr,
5075 	&md_safe_delay.attr,
5076 	&md_array_state.attr,
5077 	&md_reshape_position.attr,
5078 	&md_reshape_direction.attr,
5079 	&md_array_size.attr,
5080 	&max_corr_read_errors.attr,
5081 	&md_consistency_policy.attr,
5082 	NULL,
5083 };
5084 
5085 static struct attribute *md_redundancy_attrs[] = {
5086 	&md_scan_mode.attr,
5087 	&md_last_scan_mode.attr,
5088 	&md_mismatches.attr,
5089 	&md_sync_min.attr,
5090 	&md_sync_max.attr,
5091 	&md_sync_speed.attr,
5092 	&md_sync_force_parallel.attr,
5093 	&md_sync_completed.attr,
5094 	&md_min_sync.attr,
5095 	&md_max_sync.attr,
5096 	&md_suspend_lo.attr,
5097 	&md_suspend_hi.attr,
5098 	&md_bitmap.attr,
5099 	&md_degraded.attr,
5100 	NULL,
5101 };
5102 static struct attribute_group md_redundancy_group = {
5103 	.name = NULL,
5104 	.attrs = md_redundancy_attrs,
5105 };
5106 
5107 static ssize_t
5108 md_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
5109 {
5110 	struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
5111 	struct mddev *mddev = container_of(kobj, struct mddev, kobj);
5112 	ssize_t rv;
5113 
5114 	if (!entry->show)
5115 		return -EIO;
5116 	spin_lock(&all_mddevs_lock);
5117 	if (list_empty(&mddev->all_mddevs)) {
5118 		spin_unlock(&all_mddevs_lock);
5119 		return -EBUSY;
5120 	}
5121 	mddev_get(mddev);
5122 	spin_unlock(&all_mddevs_lock);
5123 
5124 	rv = entry->show(mddev, page);
5125 	mddev_put(mddev);
5126 	return rv;
5127 }
5128 
5129 static ssize_t
5130 md_attr_store(struct kobject *kobj, struct attribute *attr,
5131 	      const char *page, size_t length)
5132 {
5133 	struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
5134 	struct mddev *mddev = container_of(kobj, struct mddev, kobj);
5135 	ssize_t rv;
5136 
5137 	if (!entry->store)
5138 		return -EIO;
5139 	if (!capable(CAP_SYS_ADMIN))
5140 		return -EACCES;
5141 	spin_lock(&all_mddevs_lock);
5142 	if (list_empty(&mddev->all_mddevs)) {
5143 		spin_unlock(&all_mddevs_lock);
5144 		return -EBUSY;
5145 	}
5146 	mddev_get(mddev);
5147 	spin_unlock(&all_mddevs_lock);
5148 	rv = entry->store(mddev, page, length);
5149 	mddev_put(mddev);
5150 	return rv;
5151 }
5152 
5153 static void md_free(struct kobject *ko)
5154 {
5155 	struct mddev *mddev = container_of(ko, struct mddev, kobj);
5156 
5157 	if (mddev->sysfs_state)
5158 		sysfs_put(mddev->sysfs_state);
5159 
5160 	if (mddev->queue)
5161 		blk_cleanup_queue(mddev->queue);
5162 	if (mddev->gendisk) {
5163 		del_gendisk(mddev->gendisk);
5164 		put_disk(mddev->gendisk);
5165 	}
5166 	percpu_ref_exit(&mddev->writes_pending);
5167 
5168 	kfree(mddev);
5169 }
5170 
5171 static const struct sysfs_ops md_sysfs_ops = {
5172 	.show	= md_attr_show,
5173 	.store	= md_attr_store,
5174 };
5175 static struct kobj_type md_ktype = {
5176 	.release	= md_free,
5177 	.sysfs_ops	= &md_sysfs_ops,
5178 	.default_attrs	= md_default_attrs,
5179 };
5180 
5181 int mdp_major = 0;
5182 
5183 static void mddev_delayed_delete(struct work_struct *ws)
5184 {
5185 	struct mddev *mddev = container_of(ws, struct mddev, del_work);
5186 
5187 	sysfs_remove_group(&mddev->kobj, &md_bitmap_group);
5188 	kobject_del(&mddev->kobj);
5189 	kobject_put(&mddev->kobj);
5190 }
5191 
5192 static void no_op(struct percpu_ref *r) {}
5193 
5194 int mddev_init_writes_pending(struct mddev *mddev)
5195 {
5196 	if (mddev->writes_pending.percpu_count_ptr)
5197 		return 0;
5198 	if (percpu_ref_init(&mddev->writes_pending, no_op, 0, GFP_KERNEL) < 0)
5199 		return -ENOMEM;
5200 	/* We want to start with the refcount at zero */
5201 	percpu_ref_put(&mddev->writes_pending);
5202 	return 0;
5203 }
5204 EXPORT_SYMBOL_GPL(mddev_init_writes_pending);
5205 
5206 static int md_alloc(dev_t dev, char *name)
5207 {
5208 	/*
5209 	 * If dev is zero, name is the name of a device to allocate with
5210 	 * an arbitrary minor number.  It will be "md_???"
5211 	 * If dev is non-zero it must be a device number with a MAJOR of
5212 	 * MD_MAJOR or mdp_major.  In this case, if "name" is NULL, then
5213 	 * the device is being created by opening a node in /dev.
5214 	 * If "name" is not NULL, the device is being created by
5215 	 * writing to /sys/module/md_mod/parameters/new_array.
5216 	 */
5217 	static DEFINE_MUTEX(disks_mutex);
5218 	struct mddev *mddev = mddev_find(dev);
5219 	struct gendisk *disk;
5220 	int partitioned;
5221 	int shift;
5222 	int unit;
5223 	int error;
5224 
5225 	if (!mddev)
5226 		return -ENODEV;
5227 
5228 	partitioned = (MAJOR(mddev->unit) != MD_MAJOR);
5229 	shift = partitioned ? MdpMinorShift : 0;
5230 	unit = MINOR(mddev->unit) >> shift;
5231 
5232 	/* wait for any previous instance of this device to be
5233 	 * completely removed (mddev_delayed_delete).
5234 	 */
5235 	flush_workqueue(md_misc_wq);
5236 
5237 	mutex_lock(&disks_mutex);
5238 	error = -EEXIST;
5239 	if (mddev->gendisk)
5240 		goto abort;
5241 
5242 	if (name && !dev) {
5243 		/* Need to ensure that 'name' is not a duplicate.
5244 		 */
5245 		struct mddev *mddev2;
5246 		spin_lock(&all_mddevs_lock);
5247 
5248 		list_for_each_entry(mddev2, &all_mddevs, all_mddevs)
5249 			if (mddev2->gendisk &&
5250 			    strcmp(mddev2->gendisk->disk_name, name) == 0) {
5251 				spin_unlock(&all_mddevs_lock);
5252 				goto abort;
5253 			}
5254 		spin_unlock(&all_mddevs_lock);
5255 	}
5256 	if (name && dev)
5257 		/*
5258 		 * Creating /dev/mdNNN via "newarray", so adjust hold_active.
5259 		 */
5260 		mddev->hold_active = UNTIL_STOP;
5261 
5262 	error = -ENOMEM;
5263 	mddev->queue = blk_alloc_queue(GFP_KERNEL);
5264 	if (!mddev->queue)
5265 		goto abort;
5266 	mddev->queue->queuedata = mddev;
5267 
5268 	blk_queue_make_request(mddev->queue, md_make_request);
5269 	blk_set_stacking_limits(&mddev->queue->limits);
5270 
5271 	disk = alloc_disk(1 << shift);
5272 	if (!disk) {
5273 		blk_cleanup_queue(mddev->queue);
5274 		mddev->queue = NULL;
5275 		goto abort;
5276 	}
5277 	disk->major = MAJOR(mddev->unit);
5278 	disk->first_minor = unit << shift;
5279 	if (name)
5280 		strcpy(disk->disk_name, name);
5281 	else if (partitioned)
5282 		sprintf(disk->disk_name, "md_d%d", unit);
5283 	else
5284 		sprintf(disk->disk_name, "md%d", unit);
5285 	disk->fops = &md_fops;
5286 	disk->private_data = mddev;
5287 	disk->queue = mddev->queue;
5288 	blk_queue_write_cache(mddev->queue, true, true);
5289 	/* Allow extended partitions.  This makes the
5290 	 * 'mdp' device redundant, but we can't really
5291 	 * remove it now.
5292 	 */
5293 	disk->flags |= GENHD_FL_EXT_DEVT;
5294 	mddev->gendisk = disk;
5295 	/* As soon as we call add_disk(), another thread could get
5296 	 * through to md_open, so make sure it doesn't get too far
5297 	 */
5298 	mutex_lock(&mddev->open_mutex);
5299 	add_disk(disk);
5300 
5301 	error = kobject_init_and_add(&mddev->kobj, &md_ktype,
5302 				     &disk_to_dev(disk)->kobj, "%s", "md");
5303 	if (error) {
5304 		/* This isn't possible, but as kobject_init_and_add is marked
5305 		 * __must_check, we must do something with the result
5306 		 */
5307 		pr_debug("md: cannot register %s/md - name in use\n",
5308 			 disk->disk_name);
5309 		error = 0;
5310 	}
5311 	if (mddev->kobj.sd &&
5312 	    sysfs_create_group(&mddev->kobj, &md_bitmap_group))
5313 		pr_debug("pointless warning\n");
5314 	mutex_unlock(&mddev->open_mutex);
5315  abort:
5316 	mutex_unlock(&disks_mutex);
5317 	if (!error && mddev->kobj.sd) {
5318 		kobject_uevent(&mddev->kobj, KOBJ_ADD);
5319 		mddev->sysfs_state = sysfs_get_dirent_safe(mddev->kobj.sd, "array_state");
5320 	}
5321 	mddev_put(mddev);
5322 	return error;
5323 }
5324 
5325 static struct kobject *md_probe(dev_t dev, int *part, void *data)
5326 {
5327 	if (create_on_open)
5328 		md_alloc(dev, NULL);
5329 	return NULL;
5330 }
5331 
5332 static int add_named_array(const char *val, struct kernel_param *kp)
5333 {
5334 	/*
5335 	 * val must be "md_*" or "mdNNN".
5336 	 * For "md_*" we allocate an array with a large free minor number, and
5337 	 * set the name to val.  val must not already be an active name.
5338 	 * For "mdNNN" we allocate an array with the minor number NNN
5339 	 * which must not already be in use.
5340 	 */
5341 	int len = strlen(val);
5342 	char buf[DISK_NAME_LEN];
5343 	unsigned long devnum;
5344 
5345 	while (len && val[len-1] == '\n')
5346 		len--;
5347 	if (len >= DISK_NAME_LEN)
5348 		return -E2BIG;
5349 	strlcpy(buf, val, len+1);
5350 	if (strncmp(buf, "md_", 3) == 0)
5351 		return md_alloc(0, buf);
5352 	if (strncmp(buf, "md", 2) == 0 &&
5353 	    isdigit(buf[2]) &&
5354 	    kstrtoul(buf+2, 10, &devnum) == 0 &&
5355 	    devnum <= MINORMASK)
5356 		return md_alloc(MKDEV(MD_MAJOR, devnum), NULL);
5357 
5358 	return -EINVAL;
5359 }
5360 
5361 static void md_safemode_timeout(unsigned long data)
5362 {
5363 	struct mddev *mddev = (struct mddev *) data;
5364 
5365 	mddev->safemode = 1;
5366 	if (mddev->external)
5367 		sysfs_notify_dirent_safe(mddev->sysfs_state);
5368 
5369 	md_wakeup_thread(mddev->thread);
5370 }
5371 
5372 static int start_dirty_degraded;
5373 
5374 int md_run(struct mddev *mddev)
5375 {
5376 	int err;
5377 	struct md_rdev *rdev;
5378 	struct md_personality *pers;
5379 
5380 	if (list_empty(&mddev->disks))
5381 		/* cannot run an array with no devices.. */
5382 		return -EINVAL;
5383 
5384 	if (mddev->pers)
5385 		return -EBUSY;
5386 	/* Cannot run until previous stop completes properly */
5387 	if (mddev->sysfs_active)
5388 		return -EBUSY;
5389 
5390 	/*
5391 	 * Analyze all RAID superblock(s)
5392 	 */
5393 	if (!mddev->raid_disks) {
5394 		if (!mddev->persistent)
5395 			return -EINVAL;
5396 		analyze_sbs(mddev);
5397 	}
5398 
5399 	if (mddev->level != LEVEL_NONE)
5400 		request_module("md-level-%d", mddev->level);
5401 	else if (mddev->clevel[0])
5402 		request_module("md-%s", mddev->clevel);
5403 
5404 	/*
5405 	 * Drop all container device buffers, from now on
5406 	 * the only valid external interface is through the md
5407 	 * device.
5408 	 */
5409 	rdev_for_each(rdev, mddev) {
5410 		if (test_bit(Faulty, &rdev->flags))
5411 			continue;
5412 		sync_blockdev(rdev->bdev);
5413 		invalidate_bdev(rdev->bdev);
5414 		if (mddev->ro != 1 &&
5415 		    (bdev_read_only(rdev->bdev) ||
5416 		     bdev_read_only(rdev->meta_bdev))) {
5417 			mddev->ro = 1;
5418 			if (mddev->gendisk)
5419 				set_disk_ro(mddev->gendisk, 1);
5420 		}
5421 
5422 		/* perform some consistency tests on the device.
5423 		 * We don't want the data to overlap the metadata,
5424 		 * Internal Bitmap issues have been handled elsewhere.
5425 		 */
5426 		if (rdev->meta_bdev) {
5427 			/* Nothing to check */;
5428 		} else if (rdev->data_offset < rdev->sb_start) {
5429 			if (mddev->dev_sectors &&
5430 			    rdev->data_offset + mddev->dev_sectors
5431 			    > rdev->sb_start) {
5432 				pr_warn("md: %s: data overlaps metadata\n",
5433 					mdname(mddev));
5434 				return -EINVAL;
5435 			}
5436 		} else {
5437 			if (rdev->sb_start + rdev->sb_size/512
5438 			    > rdev->data_offset) {
5439 				pr_warn("md: %s: metadata overlaps data\n",
5440 					mdname(mddev));
5441 				return -EINVAL;
5442 			}
5443 		}
5444 		sysfs_notify_dirent_safe(rdev->sysfs_state);
5445 	}
5446 
5447 	if (mddev->bio_set == NULL) {
5448 		mddev->bio_set = bioset_create(BIO_POOL_SIZE, 0, BIOSET_NEED_BVECS);
5449 		if (!mddev->bio_set)
5450 			return -ENOMEM;
5451 	}
5452 	if (mddev->sync_set == NULL) {
5453 		mddev->sync_set = bioset_create(BIO_POOL_SIZE, 0, BIOSET_NEED_BVECS);
5454 		if (!mddev->sync_set)
5455 			return -ENOMEM;
5456 	}
5457 
5458 	spin_lock(&pers_lock);
5459 	pers = find_pers(mddev->level, mddev->clevel);
5460 	if (!pers || !try_module_get(pers->owner)) {
5461 		spin_unlock(&pers_lock);
5462 		if (mddev->level != LEVEL_NONE)
5463 			pr_warn("md: personality for level %d is not loaded!\n",
5464 				mddev->level);
5465 		else
5466 			pr_warn("md: personality for level %s is not loaded!\n",
5467 				mddev->clevel);
5468 		return -EINVAL;
5469 	}
5470 	spin_unlock(&pers_lock);
5471 	if (mddev->level != pers->level) {
5472 		mddev->level = pers->level;
5473 		mddev->new_level = pers->level;
5474 	}
5475 	strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
5476 
5477 	if (mddev->reshape_position != MaxSector &&
5478 	    pers->start_reshape == NULL) {
5479 		/* This personality cannot handle reshaping... */
5480 		module_put(pers->owner);
5481 		return -EINVAL;
5482 	}
5483 
5484 	if (pers->sync_request) {
5485 		/* Warn if this is a potentially silly
5486 		 * configuration.
5487 		 */
5488 		char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
5489 		struct md_rdev *rdev2;
5490 		int warned = 0;
5491 
5492 		rdev_for_each(rdev, mddev)
5493 			rdev_for_each(rdev2, mddev) {
5494 				if (rdev < rdev2 &&
5495 				    rdev->bdev->bd_contains ==
5496 				    rdev2->bdev->bd_contains) {
5497 					pr_warn("%s: WARNING: %s appears to be on the same physical disk as %s.\n",
5498 						mdname(mddev),
5499 						bdevname(rdev->bdev,b),
5500 						bdevname(rdev2->bdev,b2));
5501 					warned = 1;
5502 				}
5503 			}
5504 
5505 		if (warned)
5506 			pr_warn("True protection against single-disk failure might be compromised.\n");
5507 	}
5508 
5509 	mddev->recovery = 0;
5510 	/* may be over-ridden by personality */
5511 	mddev->resync_max_sectors = mddev->dev_sectors;
5512 
5513 	mddev->ok_start_degraded = start_dirty_degraded;
5514 
5515 	if (start_readonly && mddev->ro == 0)
5516 		mddev->ro = 2; /* read-only, but switch on first write */
5517 
5518 	/*
5519 	 * NOTE: some pers->run(), for example r5l_recovery_log(), wakes
5520 	 * up mddev->thread. It is important to initialize critical
5521 	 * resources for mddev->thread BEFORE calling pers->run().
5522 	 */
5523 	err = pers->run(mddev);
5524 	if (err)
5525 		pr_warn("md: pers->run() failed ...\n");
5526 	else if (pers->size(mddev, 0, 0) < mddev->array_sectors) {
5527 		WARN_ONCE(!mddev->external_size,
5528 			  "%s: default size too small, but 'external_size' not in effect?\n",
5529 			  __func__);
5530 		pr_warn("md: invalid array_size %llu > default size %llu\n",
5531 			(unsigned long long)mddev->array_sectors / 2,
5532 			(unsigned long long)pers->size(mddev, 0, 0) / 2);
5533 		err = -EINVAL;
5534 	}
5535 	if (err == 0 && pers->sync_request &&
5536 	    (mddev->bitmap_info.file || mddev->bitmap_info.offset)) {
5537 		struct bitmap *bitmap;
5538 
5539 		bitmap = bitmap_create(mddev, -1);
5540 		if (IS_ERR(bitmap)) {
5541 			err = PTR_ERR(bitmap);
5542 			pr_warn("%s: failed to create bitmap (%d)\n",
5543 				mdname(mddev), err);
5544 		} else
5545 			mddev->bitmap = bitmap;
5546 
5547 	}
5548 	if (err) {
5549 		mddev_detach(mddev);
5550 		if (mddev->private)
5551 			pers->free(mddev, mddev->private);
5552 		mddev->private = NULL;
5553 		module_put(pers->owner);
5554 		bitmap_destroy(mddev);
5555 		return err;
5556 	}
5557 	if (mddev->queue) {
5558 		bool nonrot = true;
5559 
5560 		rdev_for_each(rdev, mddev) {
5561 			if (rdev->raid_disk >= 0 &&
5562 			    !blk_queue_nonrot(bdev_get_queue(rdev->bdev))) {
5563 				nonrot = false;
5564 				break;
5565 			}
5566 		}
5567 		if (mddev->degraded)
5568 			nonrot = false;
5569 		if (nonrot)
5570 			queue_flag_set_unlocked(QUEUE_FLAG_NONROT, mddev->queue);
5571 		else
5572 			queue_flag_clear_unlocked(QUEUE_FLAG_NONROT, mddev->queue);
5573 		mddev->queue->backing_dev_info->congested_data = mddev;
5574 		mddev->queue->backing_dev_info->congested_fn = md_congested;
5575 	}
5576 	if (pers->sync_request) {
5577 		if (mddev->kobj.sd &&
5578 		    sysfs_create_group(&mddev->kobj, &md_redundancy_group))
5579 			pr_warn("md: cannot register extra attributes for %s\n",
5580 				mdname(mddev));
5581 		mddev->sysfs_action = sysfs_get_dirent_safe(mddev->kobj.sd, "sync_action");
5582 	} else if (mddev->ro == 2) /* auto-readonly not meaningful */
5583 		mddev->ro = 0;
5584 
5585 	atomic_set(&mddev->max_corr_read_errors,
5586 		   MD_DEFAULT_MAX_CORRECTED_READ_ERRORS);
5587 	mddev->safemode = 0;
5588 	if (mddev_is_clustered(mddev))
5589 		mddev->safemode_delay = 0;
5590 	else
5591 		mddev->safemode_delay = (200 * HZ)/1000 +1; /* 200 msec delay */
5592 	mddev->in_sync = 1;
5593 	smp_wmb();
5594 	spin_lock(&mddev->lock);
5595 	mddev->pers = pers;
5596 	spin_unlock(&mddev->lock);
5597 	rdev_for_each(rdev, mddev)
5598 		if (rdev->raid_disk >= 0)
5599 			if (sysfs_link_rdev(mddev, rdev))
5600 				/* failure here is OK */;
5601 
5602 	if (mddev->degraded && !mddev->ro)
5603 		/* This ensures that recovering status is reported immediately
5604 		 * via sysfs - until a lack of spares is confirmed.
5605 		 */
5606 		set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
5607 	set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5608 
5609 	if (mddev->sb_flags)
5610 		md_update_sb(mddev, 0);
5611 
5612 	md_new_event(mddev);
5613 	sysfs_notify_dirent_safe(mddev->sysfs_state);
5614 	sysfs_notify_dirent_safe(mddev->sysfs_action);
5615 	sysfs_notify(&mddev->kobj, NULL, "degraded");
5616 	return 0;
5617 }
5618 EXPORT_SYMBOL_GPL(md_run);
5619 
5620 static int do_md_run(struct mddev *mddev)
5621 {
5622 	int err;
5623 
5624 	err = md_run(mddev);
5625 	if (err)
5626 		goto out;
5627 	err = bitmap_load(mddev);
5628 	if (err) {
5629 		bitmap_destroy(mddev);
5630 		goto out;
5631 	}
5632 
5633 	if (mddev_is_clustered(mddev))
5634 		md_allow_write(mddev);
5635 
5636 	md_wakeup_thread(mddev->thread);
5637 	md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
5638 
5639 	set_capacity(mddev->gendisk, mddev->array_sectors);
5640 	revalidate_disk(mddev->gendisk);
5641 	mddev->changed = 1;
5642 	kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
5643 out:
5644 	return err;
5645 }
5646 
5647 static int restart_array(struct mddev *mddev)
5648 {
5649 	struct gendisk *disk = mddev->gendisk;
5650 	struct md_rdev *rdev;
5651 	bool has_journal = false;
5652 	bool has_readonly = false;
5653 
5654 	/* Complain if it has no devices */
5655 	if (list_empty(&mddev->disks))
5656 		return -ENXIO;
5657 	if (!mddev->pers)
5658 		return -EINVAL;
5659 	if (!mddev->ro)
5660 		return -EBUSY;
5661 
5662 	rcu_read_lock();
5663 	rdev_for_each_rcu(rdev, mddev) {
5664 		if (test_bit(Journal, &rdev->flags) &&
5665 		    !test_bit(Faulty, &rdev->flags))
5666 			has_journal = true;
5667 		if (bdev_read_only(rdev->bdev))
5668 			has_readonly = true;
5669 	}
5670 	rcu_read_unlock();
5671 	if (test_bit(MD_HAS_JOURNAL, &mddev->flags) && !has_journal)
5672 		/* Don't restart rw with journal missing/faulty */
5673 			return -EINVAL;
5674 	if (has_readonly)
5675 		return -EROFS;
5676 
5677 	mddev->safemode = 0;
5678 	mddev->ro = 0;
5679 	set_disk_ro(disk, 0);
5680 	pr_debug("md: %s switched to read-write mode.\n", mdname(mddev));
5681 	/* Kick recovery or resync if necessary */
5682 	set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5683 	md_wakeup_thread(mddev->thread);
5684 	md_wakeup_thread(mddev->sync_thread);
5685 	sysfs_notify_dirent_safe(mddev->sysfs_state);
5686 	return 0;
5687 }
5688 
5689 static void md_clean(struct mddev *mddev)
5690 {
5691 	mddev->array_sectors = 0;
5692 	mddev->external_size = 0;
5693 	mddev->dev_sectors = 0;
5694 	mddev->raid_disks = 0;
5695 	mddev->recovery_cp = 0;
5696 	mddev->resync_min = 0;
5697 	mddev->resync_max = MaxSector;
5698 	mddev->reshape_position = MaxSector;
5699 	mddev->external = 0;
5700 	mddev->persistent = 0;
5701 	mddev->level = LEVEL_NONE;
5702 	mddev->clevel[0] = 0;
5703 	mddev->flags = 0;
5704 	mddev->sb_flags = 0;
5705 	mddev->ro = 0;
5706 	mddev->metadata_type[0] = 0;
5707 	mddev->chunk_sectors = 0;
5708 	mddev->ctime = mddev->utime = 0;
5709 	mddev->layout = 0;
5710 	mddev->max_disks = 0;
5711 	mddev->events = 0;
5712 	mddev->can_decrease_events = 0;
5713 	mddev->delta_disks = 0;
5714 	mddev->reshape_backwards = 0;
5715 	mddev->new_level = LEVEL_NONE;
5716 	mddev->new_layout = 0;
5717 	mddev->new_chunk_sectors = 0;
5718 	mddev->curr_resync = 0;
5719 	atomic64_set(&mddev->resync_mismatches, 0);
5720 	mddev->suspend_lo = mddev->suspend_hi = 0;
5721 	mddev->sync_speed_min = mddev->sync_speed_max = 0;
5722 	mddev->recovery = 0;
5723 	mddev->in_sync = 0;
5724 	mddev->changed = 0;
5725 	mddev->degraded = 0;
5726 	mddev->safemode = 0;
5727 	mddev->private = NULL;
5728 	mddev->cluster_info = NULL;
5729 	mddev->bitmap_info.offset = 0;
5730 	mddev->bitmap_info.default_offset = 0;
5731 	mddev->bitmap_info.default_space = 0;
5732 	mddev->bitmap_info.chunksize = 0;
5733 	mddev->bitmap_info.daemon_sleep = 0;
5734 	mddev->bitmap_info.max_write_behind = 0;
5735 	mddev->bitmap_info.nodes = 0;
5736 }
5737 
5738 static void __md_stop_writes(struct mddev *mddev)
5739 {
5740 	set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5741 	flush_workqueue(md_misc_wq);
5742 	if (mddev->sync_thread) {
5743 		set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5744 		md_reap_sync_thread(mddev);
5745 	}
5746 
5747 	del_timer_sync(&mddev->safemode_timer);
5748 
5749 	if (mddev->pers && mddev->pers->quiesce) {
5750 		mddev->pers->quiesce(mddev, 1);
5751 		mddev->pers->quiesce(mddev, 0);
5752 	}
5753 	bitmap_flush(mddev);
5754 
5755 	if (mddev->ro == 0 &&
5756 	    ((!mddev->in_sync && !mddev_is_clustered(mddev)) ||
5757 	     mddev->sb_flags)) {
5758 		/* mark array as shutdown cleanly */
5759 		if (!mddev_is_clustered(mddev))
5760 			mddev->in_sync = 1;
5761 		md_update_sb(mddev, 1);
5762 	}
5763 }
5764 
5765 void md_stop_writes(struct mddev *mddev)
5766 {
5767 	mddev_lock_nointr(mddev);
5768 	__md_stop_writes(mddev);
5769 	mddev_unlock(mddev);
5770 }
5771 EXPORT_SYMBOL_GPL(md_stop_writes);
5772 
5773 static void mddev_detach(struct mddev *mddev)
5774 {
5775 	bitmap_wait_behind_writes(mddev);
5776 	if (mddev->pers && mddev->pers->quiesce) {
5777 		mddev->pers->quiesce(mddev, 1);
5778 		mddev->pers->quiesce(mddev, 0);
5779 	}
5780 	md_unregister_thread(&mddev->thread);
5781 	if (mddev->queue)
5782 		blk_sync_queue(mddev->queue); /* the unplug fn references 'conf'*/
5783 }
5784 
5785 static void __md_stop(struct mddev *mddev)
5786 {
5787 	struct md_personality *pers = mddev->pers;
5788 	bitmap_destroy(mddev);
5789 	mddev_detach(mddev);
5790 	/* Ensure ->event_work is done */
5791 	flush_workqueue(md_misc_wq);
5792 	spin_lock(&mddev->lock);
5793 	mddev->pers = NULL;
5794 	spin_unlock(&mddev->lock);
5795 	pers->free(mddev, mddev->private);
5796 	mddev->private = NULL;
5797 	if (pers->sync_request && mddev->to_remove == NULL)
5798 		mddev->to_remove = &md_redundancy_group;
5799 	module_put(pers->owner);
5800 	clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5801 }
5802 
5803 void md_stop(struct mddev *mddev)
5804 {
5805 	/* stop the array and free an attached data structures.
5806 	 * This is called from dm-raid
5807 	 */
5808 	__md_stop(mddev);
5809 	if (mddev->bio_set)
5810 		bioset_free(mddev->bio_set);
5811 }
5812 
5813 EXPORT_SYMBOL_GPL(md_stop);
5814 
5815 static int md_set_readonly(struct mddev *mddev, struct block_device *bdev)
5816 {
5817 	int err = 0;
5818 	int did_freeze = 0;
5819 
5820 	if (!test_bit(MD_RECOVERY_FROZEN, &mddev->recovery)) {
5821 		did_freeze = 1;
5822 		set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5823 		md_wakeup_thread(mddev->thread);
5824 	}
5825 	if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
5826 		set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5827 	if (mddev->sync_thread)
5828 		/* Thread might be blocked waiting for metadata update
5829 		 * which will now never happen */
5830 		wake_up_process(mddev->sync_thread->tsk);
5831 
5832 	if (mddev->external && test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags))
5833 		return -EBUSY;
5834 	mddev_unlock(mddev);
5835 	wait_event(resync_wait, !test_bit(MD_RECOVERY_RUNNING,
5836 					  &mddev->recovery));
5837 	wait_event(mddev->sb_wait,
5838 		   !test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags));
5839 	mddev_lock_nointr(mddev);
5840 
5841 	mutex_lock(&mddev->open_mutex);
5842 	if ((mddev->pers && atomic_read(&mddev->openers) > !!bdev) ||
5843 	    mddev->sync_thread ||
5844 	    test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) {
5845 		pr_warn("md: %s still in use.\n",mdname(mddev));
5846 		if (did_freeze) {
5847 			clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5848 			set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5849 			md_wakeup_thread(mddev->thread);
5850 		}
5851 		err = -EBUSY;
5852 		goto out;
5853 	}
5854 	if (mddev->pers) {
5855 		__md_stop_writes(mddev);
5856 
5857 		err  = -ENXIO;
5858 		if (mddev->ro==1)
5859 			goto out;
5860 		mddev->ro = 1;
5861 		set_disk_ro(mddev->gendisk, 1);
5862 		clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5863 		set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5864 		md_wakeup_thread(mddev->thread);
5865 		sysfs_notify_dirent_safe(mddev->sysfs_state);
5866 		err = 0;
5867 	}
5868 out:
5869 	mutex_unlock(&mddev->open_mutex);
5870 	return err;
5871 }
5872 
5873 /* mode:
5874  *   0 - completely stop and dis-assemble array
5875  *   2 - stop but do not disassemble array
5876  */
5877 static int do_md_stop(struct mddev *mddev, int mode,
5878 		      struct block_device *bdev)
5879 {
5880 	struct gendisk *disk = mddev->gendisk;
5881 	struct md_rdev *rdev;
5882 	int did_freeze = 0;
5883 
5884 	if (!test_bit(MD_RECOVERY_FROZEN, &mddev->recovery)) {
5885 		did_freeze = 1;
5886 		set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5887 		md_wakeup_thread(mddev->thread);
5888 	}
5889 	if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
5890 		set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5891 	if (mddev->sync_thread)
5892 		/* Thread might be blocked waiting for metadata update
5893 		 * which will now never happen */
5894 		wake_up_process(mddev->sync_thread->tsk);
5895 
5896 	mddev_unlock(mddev);
5897 	wait_event(resync_wait, (mddev->sync_thread == NULL &&
5898 				 !test_bit(MD_RECOVERY_RUNNING,
5899 					   &mddev->recovery)));
5900 	mddev_lock_nointr(mddev);
5901 
5902 	mutex_lock(&mddev->open_mutex);
5903 	if ((mddev->pers && atomic_read(&mddev->openers) > !!bdev) ||
5904 	    mddev->sysfs_active ||
5905 	    mddev->sync_thread ||
5906 	    test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) {
5907 		pr_warn("md: %s still in use.\n",mdname(mddev));
5908 		mutex_unlock(&mddev->open_mutex);
5909 		if (did_freeze) {
5910 			clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5911 			set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5912 			md_wakeup_thread(mddev->thread);
5913 		}
5914 		return -EBUSY;
5915 	}
5916 	if (mddev->pers) {
5917 		if (mddev->ro)
5918 			set_disk_ro(disk, 0);
5919 
5920 		__md_stop_writes(mddev);
5921 		__md_stop(mddev);
5922 		mddev->queue->backing_dev_info->congested_fn = NULL;
5923 
5924 		/* tell userspace to handle 'inactive' */
5925 		sysfs_notify_dirent_safe(mddev->sysfs_state);
5926 
5927 		rdev_for_each(rdev, mddev)
5928 			if (rdev->raid_disk >= 0)
5929 				sysfs_unlink_rdev(mddev, rdev);
5930 
5931 		set_capacity(disk, 0);
5932 		mutex_unlock(&mddev->open_mutex);
5933 		mddev->changed = 1;
5934 		revalidate_disk(disk);
5935 
5936 		if (mddev->ro)
5937 			mddev->ro = 0;
5938 	} else
5939 		mutex_unlock(&mddev->open_mutex);
5940 	/*
5941 	 * Free resources if final stop
5942 	 */
5943 	if (mode == 0) {
5944 		pr_info("md: %s stopped.\n", mdname(mddev));
5945 
5946 		if (mddev->bitmap_info.file) {
5947 			struct file *f = mddev->bitmap_info.file;
5948 			spin_lock(&mddev->lock);
5949 			mddev->bitmap_info.file = NULL;
5950 			spin_unlock(&mddev->lock);
5951 			fput(f);
5952 		}
5953 		mddev->bitmap_info.offset = 0;
5954 
5955 		export_array(mddev);
5956 
5957 		md_clean(mddev);
5958 		if (mddev->hold_active == UNTIL_STOP)
5959 			mddev->hold_active = 0;
5960 	}
5961 	md_new_event(mddev);
5962 	sysfs_notify_dirent_safe(mddev->sysfs_state);
5963 	return 0;
5964 }
5965 
5966 #ifndef MODULE
5967 static void autorun_array(struct mddev *mddev)
5968 {
5969 	struct md_rdev *rdev;
5970 	int err;
5971 
5972 	if (list_empty(&mddev->disks))
5973 		return;
5974 
5975 	pr_info("md: running: ");
5976 
5977 	rdev_for_each(rdev, mddev) {
5978 		char b[BDEVNAME_SIZE];
5979 		pr_cont("<%s>", bdevname(rdev->bdev,b));
5980 	}
5981 	pr_cont("\n");
5982 
5983 	err = do_md_run(mddev);
5984 	if (err) {
5985 		pr_warn("md: do_md_run() returned %d\n", err);
5986 		do_md_stop(mddev, 0, NULL);
5987 	}
5988 }
5989 
5990 /*
5991  * lets try to run arrays based on all disks that have arrived
5992  * until now. (those are in pending_raid_disks)
5993  *
5994  * the method: pick the first pending disk, collect all disks with
5995  * the same UUID, remove all from the pending list and put them into
5996  * the 'same_array' list. Then order this list based on superblock
5997  * update time (freshest comes first), kick out 'old' disks and
5998  * compare superblocks. If everything's fine then run it.
5999  *
6000  * If "unit" is allocated, then bump its reference count
6001  */
6002 static void autorun_devices(int part)
6003 {
6004 	struct md_rdev *rdev0, *rdev, *tmp;
6005 	struct mddev *mddev;
6006 	char b[BDEVNAME_SIZE];
6007 
6008 	pr_info("md: autorun ...\n");
6009 	while (!list_empty(&pending_raid_disks)) {
6010 		int unit;
6011 		dev_t dev;
6012 		LIST_HEAD(candidates);
6013 		rdev0 = list_entry(pending_raid_disks.next,
6014 					 struct md_rdev, same_set);
6015 
6016 		pr_debug("md: considering %s ...\n", bdevname(rdev0->bdev,b));
6017 		INIT_LIST_HEAD(&candidates);
6018 		rdev_for_each_list(rdev, tmp, &pending_raid_disks)
6019 			if (super_90_load(rdev, rdev0, 0) >= 0) {
6020 				pr_debug("md:  adding %s ...\n",
6021 					 bdevname(rdev->bdev,b));
6022 				list_move(&rdev->same_set, &candidates);
6023 			}
6024 		/*
6025 		 * now we have a set of devices, with all of them having
6026 		 * mostly sane superblocks. It's time to allocate the
6027 		 * mddev.
6028 		 */
6029 		if (part) {
6030 			dev = MKDEV(mdp_major,
6031 				    rdev0->preferred_minor << MdpMinorShift);
6032 			unit = MINOR(dev) >> MdpMinorShift;
6033 		} else {
6034 			dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
6035 			unit = MINOR(dev);
6036 		}
6037 		if (rdev0->preferred_minor != unit) {
6038 			pr_warn("md: unit number in %s is bad: %d\n",
6039 				bdevname(rdev0->bdev, b), rdev0->preferred_minor);
6040 			break;
6041 		}
6042 
6043 		md_probe(dev, NULL, NULL);
6044 		mddev = mddev_find(dev);
6045 		if (!mddev || !mddev->gendisk) {
6046 			if (mddev)
6047 				mddev_put(mddev);
6048 			break;
6049 		}
6050 		if (mddev_lock(mddev))
6051 			pr_warn("md: %s locked, cannot run\n", mdname(mddev));
6052 		else if (mddev->raid_disks || mddev->major_version
6053 			 || !list_empty(&mddev->disks)) {
6054 			pr_warn("md: %s already running, cannot run %s\n",
6055 				mdname(mddev), bdevname(rdev0->bdev,b));
6056 			mddev_unlock(mddev);
6057 		} else {
6058 			pr_debug("md: created %s\n", mdname(mddev));
6059 			mddev->persistent = 1;
6060 			rdev_for_each_list(rdev, tmp, &candidates) {
6061 				list_del_init(&rdev->same_set);
6062 				if (bind_rdev_to_array(rdev, mddev))
6063 					export_rdev(rdev);
6064 			}
6065 			autorun_array(mddev);
6066 			mddev_unlock(mddev);
6067 		}
6068 		/* on success, candidates will be empty, on error
6069 		 * it won't...
6070 		 */
6071 		rdev_for_each_list(rdev, tmp, &candidates) {
6072 			list_del_init(&rdev->same_set);
6073 			export_rdev(rdev);
6074 		}
6075 		mddev_put(mddev);
6076 	}
6077 	pr_info("md: ... autorun DONE.\n");
6078 }
6079 #endif /* !MODULE */
6080 
6081 static int get_version(void __user *arg)
6082 {
6083 	mdu_version_t ver;
6084 
6085 	ver.major = MD_MAJOR_VERSION;
6086 	ver.minor = MD_MINOR_VERSION;
6087 	ver.patchlevel = MD_PATCHLEVEL_VERSION;
6088 
6089 	if (copy_to_user(arg, &ver, sizeof(ver)))
6090 		return -EFAULT;
6091 
6092 	return 0;
6093 }
6094 
6095 static int get_array_info(struct mddev *mddev, void __user *arg)
6096 {
6097 	mdu_array_info_t info;
6098 	int nr,working,insync,failed,spare;
6099 	struct md_rdev *rdev;
6100 
6101 	nr = working = insync = failed = spare = 0;
6102 	rcu_read_lock();
6103 	rdev_for_each_rcu(rdev, mddev) {
6104 		nr++;
6105 		if (test_bit(Faulty, &rdev->flags))
6106 			failed++;
6107 		else {
6108 			working++;
6109 			if (test_bit(In_sync, &rdev->flags))
6110 				insync++;
6111 			else if (test_bit(Journal, &rdev->flags))
6112 				/* TODO: add journal count to md_u.h */
6113 				;
6114 			else
6115 				spare++;
6116 		}
6117 	}
6118 	rcu_read_unlock();
6119 
6120 	info.major_version = mddev->major_version;
6121 	info.minor_version = mddev->minor_version;
6122 	info.patch_version = MD_PATCHLEVEL_VERSION;
6123 	info.ctime         = clamp_t(time64_t, mddev->ctime, 0, U32_MAX);
6124 	info.level         = mddev->level;
6125 	info.size          = mddev->dev_sectors / 2;
6126 	if (info.size != mddev->dev_sectors / 2) /* overflow */
6127 		info.size = -1;
6128 	info.nr_disks      = nr;
6129 	info.raid_disks    = mddev->raid_disks;
6130 	info.md_minor      = mddev->md_minor;
6131 	info.not_persistent= !mddev->persistent;
6132 
6133 	info.utime         = clamp_t(time64_t, mddev->utime, 0, U32_MAX);
6134 	info.state         = 0;
6135 	if (mddev->in_sync)
6136 		info.state = (1<<MD_SB_CLEAN);
6137 	if (mddev->bitmap && mddev->bitmap_info.offset)
6138 		info.state |= (1<<MD_SB_BITMAP_PRESENT);
6139 	if (mddev_is_clustered(mddev))
6140 		info.state |= (1<<MD_SB_CLUSTERED);
6141 	info.active_disks  = insync;
6142 	info.working_disks = working;
6143 	info.failed_disks  = failed;
6144 	info.spare_disks   = spare;
6145 
6146 	info.layout        = mddev->layout;
6147 	info.chunk_size    = mddev->chunk_sectors << 9;
6148 
6149 	if (copy_to_user(arg, &info, sizeof(info)))
6150 		return -EFAULT;
6151 
6152 	return 0;
6153 }
6154 
6155 static int get_bitmap_file(struct mddev *mddev, void __user * arg)
6156 {
6157 	mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */
6158 	char *ptr;
6159 	int err;
6160 
6161 	file = kzalloc(sizeof(*file), GFP_NOIO);
6162 	if (!file)
6163 		return -ENOMEM;
6164 
6165 	err = 0;
6166 	spin_lock(&mddev->lock);
6167 	/* bitmap enabled */
6168 	if (mddev->bitmap_info.file) {
6169 		ptr = file_path(mddev->bitmap_info.file, file->pathname,
6170 				sizeof(file->pathname));
6171 		if (IS_ERR(ptr))
6172 			err = PTR_ERR(ptr);
6173 		else
6174 			memmove(file->pathname, ptr,
6175 				sizeof(file->pathname)-(ptr-file->pathname));
6176 	}
6177 	spin_unlock(&mddev->lock);
6178 
6179 	if (err == 0 &&
6180 	    copy_to_user(arg, file, sizeof(*file)))
6181 		err = -EFAULT;
6182 
6183 	kfree(file);
6184 	return err;
6185 }
6186 
6187 static int get_disk_info(struct mddev *mddev, void __user * arg)
6188 {
6189 	mdu_disk_info_t info;
6190 	struct md_rdev *rdev;
6191 
6192 	if (copy_from_user(&info, arg, sizeof(info)))
6193 		return -EFAULT;
6194 
6195 	rcu_read_lock();
6196 	rdev = md_find_rdev_nr_rcu(mddev, info.number);
6197 	if (rdev) {
6198 		info.major = MAJOR(rdev->bdev->bd_dev);
6199 		info.minor = MINOR(rdev->bdev->bd_dev);
6200 		info.raid_disk = rdev->raid_disk;
6201 		info.state = 0;
6202 		if (test_bit(Faulty, &rdev->flags))
6203 			info.state |= (1<<MD_DISK_FAULTY);
6204 		else if (test_bit(In_sync, &rdev->flags)) {
6205 			info.state |= (1<<MD_DISK_ACTIVE);
6206 			info.state |= (1<<MD_DISK_SYNC);
6207 		}
6208 		if (test_bit(Journal, &rdev->flags))
6209 			info.state |= (1<<MD_DISK_JOURNAL);
6210 		if (test_bit(WriteMostly, &rdev->flags))
6211 			info.state |= (1<<MD_DISK_WRITEMOSTLY);
6212 		if (test_bit(FailFast, &rdev->flags))
6213 			info.state |= (1<<MD_DISK_FAILFAST);
6214 	} else {
6215 		info.major = info.minor = 0;
6216 		info.raid_disk = -1;
6217 		info.state = (1<<MD_DISK_REMOVED);
6218 	}
6219 	rcu_read_unlock();
6220 
6221 	if (copy_to_user(arg, &info, sizeof(info)))
6222 		return -EFAULT;
6223 
6224 	return 0;
6225 }
6226 
6227 static int add_new_disk(struct mddev *mddev, mdu_disk_info_t *info)
6228 {
6229 	char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
6230 	struct md_rdev *rdev;
6231 	dev_t dev = MKDEV(info->major,info->minor);
6232 
6233 	if (mddev_is_clustered(mddev) &&
6234 		!(info->state & ((1 << MD_DISK_CLUSTER_ADD) | (1 << MD_DISK_CANDIDATE)))) {
6235 		pr_warn("%s: Cannot add to clustered mddev.\n",
6236 			mdname(mddev));
6237 		return -EINVAL;
6238 	}
6239 
6240 	if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
6241 		return -EOVERFLOW;
6242 
6243 	if (!mddev->raid_disks) {
6244 		int err;
6245 		/* expecting a device which has a superblock */
6246 		rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
6247 		if (IS_ERR(rdev)) {
6248 			pr_warn("md: md_import_device returned %ld\n",
6249 				PTR_ERR(rdev));
6250 			return PTR_ERR(rdev);
6251 		}
6252 		if (!list_empty(&mddev->disks)) {
6253 			struct md_rdev *rdev0
6254 				= list_entry(mddev->disks.next,
6255 					     struct md_rdev, same_set);
6256 			err = super_types[mddev->major_version]
6257 				.load_super(rdev, rdev0, mddev->minor_version);
6258 			if (err < 0) {
6259 				pr_warn("md: %s has different UUID to %s\n",
6260 					bdevname(rdev->bdev,b),
6261 					bdevname(rdev0->bdev,b2));
6262 				export_rdev(rdev);
6263 				return -EINVAL;
6264 			}
6265 		}
6266 		err = bind_rdev_to_array(rdev, mddev);
6267 		if (err)
6268 			export_rdev(rdev);
6269 		return err;
6270 	}
6271 
6272 	/*
6273 	 * add_new_disk can be used once the array is assembled
6274 	 * to add "hot spares".  They must already have a superblock
6275 	 * written
6276 	 */
6277 	if (mddev->pers) {
6278 		int err;
6279 		if (!mddev->pers->hot_add_disk) {
6280 			pr_warn("%s: personality does not support diskops!\n",
6281 				mdname(mddev));
6282 			return -EINVAL;
6283 		}
6284 		if (mddev->persistent)
6285 			rdev = md_import_device(dev, mddev->major_version,
6286 						mddev->minor_version);
6287 		else
6288 			rdev = md_import_device(dev, -1, -1);
6289 		if (IS_ERR(rdev)) {
6290 			pr_warn("md: md_import_device returned %ld\n",
6291 				PTR_ERR(rdev));
6292 			return PTR_ERR(rdev);
6293 		}
6294 		/* set saved_raid_disk if appropriate */
6295 		if (!mddev->persistent) {
6296 			if (info->state & (1<<MD_DISK_SYNC)  &&
6297 			    info->raid_disk < mddev->raid_disks) {
6298 				rdev->raid_disk = info->raid_disk;
6299 				set_bit(In_sync, &rdev->flags);
6300 				clear_bit(Bitmap_sync, &rdev->flags);
6301 			} else
6302 				rdev->raid_disk = -1;
6303 			rdev->saved_raid_disk = rdev->raid_disk;
6304 		} else
6305 			super_types[mddev->major_version].
6306 				validate_super(mddev, rdev);
6307 		if ((info->state & (1<<MD_DISK_SYNC)) &&
6308 		     rdev->raid_disk != info->raid_disk) {
6309 			/* This was a hot-add request, but events doesn't
6310 			 * match, so reject it.
6311 			 */
6312 			export_rdev(rdev);
6313 			return -EINVAL;
6314 		}
6315 
6316 		clear_bit(In_sync, &rdev->flags); /* just to be sure */
6317 		if (info->state & (1<<MD_DISK_WRITEMOSTLY))
6318 			set_bit(WriteMostly, &rdev->flags);
6319 		else
6320 			clear_bit(WriteMostly, &rdev->flags);
6321 		if (info->state & (1<<MD_DISK_FAILFAST))
6322 			set_bit(FailFast, &rdev->flags);
6323 		else
6324 			clear_bit(FailFast, &rdev->flags);
6325 
6326 		if (info->state & (1<<MD_DISK_JOURNAL)) {
6327 			struct md_rdev *rdev2;
6328 			bool has_journal = false;
6329 
6330 			/* make sure no existing journal disk */
6331 			rdev_for_each(rdev2, mddev) {
6332 				if (test_bit(Journal, &rdev2->flags)) {
6333 					has_journal = true;
6334 					break;
6335 				}
6336 			}
6337 			if (has_journal) {
6338 				export_rdev(rdev);
6339 				return -EBUSY;
6340 			}
6341 			set_bit(Journal, &rdev->flags);
6342 		}
6343 		/*
6344 		 * check whether the device shows up in other nodes
6345 		 */
6346 		if (mddev_is_clustered(mddev)) {
6347 			if (info->state & (1 << MD_DISK_CANDIDATE))
6348 				set_bit(Candidate, &rdev->flags);
6349 			else if (info->state & (1 << MD_DISK_CLUSTER_ADD)) {
6350 				/* --add initiated by this node */
6351 				err = md_cluster_ops->add_new_disk(mddev, rdev);
6352 				if (err) {
6353 					export_rdev(rdev);
6354 					return err;
6355 				}
6356 			}
6357 		}
6358 
6359 		rdev->raid_disk = -1;
6360 		err = bind_rdev_to_array(rdev, mddev);
6361 
6362 		if (err)
6363 			export_rdev(rdev);
6364 
6365 		if (mddev_is_clustered(mddev)) {
6366 			if (info->state & (1 << MD_DISK_CANDIDATE)) {
6367 				if (!err) {
6368 					err = md_cluster_ops->new_disk_ack(mddev,
6369 						err == 0);
6370 					if (err)
6371 						md_kick_rdev_from_array(rdev);
6372 				}
6373 			} else {
6374 				if (err)
6375 					md_cluster_ops->add_new_disk_cancel(mddev);
6376 				else
6377 					err = add_bound_rdev(rdev);
6378 			}
6379 
6380 		} else if (!err)
6381 			err = add_bound_rdev(rdev);
6382 
6383 		return err;
6384 	}
6385 
6386 	/* otherwise, add_new_disk is only allowed
6387 	 * for major_version==0 superblocks
6388 	 */
6389 	if (mddev->major_version != 0) {
6390 		pr_warn("%s: ADD_NEW_DISK not supported\n", mdname(mddev));
6391 		return -EINVAL;
6392 	}
6393 
6394 	if (!(info->state & (1<<MD_DISK_FAULTY))) {
6395 		int err;
6396 		rdev = md_import_device(dev, -1, 0);
6397 		if (IS_ERR(rdev)) {
6398 			pr_warn("md: error, md_import_device() returned %ld\n",
6399 				PTR_ERR(rdev));
6400 			return PTR_ERR(rdev);
6401 		}
6402 		rdev->desc_nr = info->number;
6403 		if (info->raid_disk < mddev->raid_disks)
6404 			rdev->raid_disk = info->raid_disk;
6405 		else
6406 			rdev->raid_disk = -1;
6407 
6408 		if (rdev->raid_disk < mddev->raid_disks)
6409 			if (info->state & (1<<MD_DISK_SYNC))
6410 				set_bit(In_sync, &rdev->flags);
6411 
6412 		if (info->state & (1<<MD_DISK_WRITEMOSTLY))
6413 			set_bit(WriteMostly, &rdev->flags);
6414 		if (info->state & (1<<MD_DISK_FAILFAST))
6415 			set_bit(FailFast, &rdev->flags);
6416 
6417 		if (!mddev->persistent) {
6418 			pr_debug("md: nonpersistent superblock ...\n");
6419 			rdev->sb_start = i_size_read(rdev->bdev->bd_inode) / 512;
6420 		} else
6421 			rdev->sb_start = calc_dev_sboffset(rdev);
6422 		rdev->sectors = rdev->sb_start;
6423 
6424 		err = bind_rdev_to_array(rdev, mddev);
6425 		if (err) {
6426 			export_rdev(rdev);
6427 			return err;
6428 		}
6429 	}
6430 
6431 	return 0;
6432 }
6433 
6434 static int hot_remove_disk(struct mddev *mddev, dev_t dev)
6435 {
6436 	char b[BDEVNAME_SIZE];
6437 	struct md_rdev *rdev;
6438 
6439 	rdev = find_rdev(mddev, dev);
6440 	if (!rdev)
6441 		return -ENXIO;
6442 
6443 	if (rdev->raid_disk < 0)
6444 		goto kick_rdev;
6445 
6446 	clear_bit(Blocked, &rdev->flags);
6447 	remove_and_add_spares(mddev, rdev);
6448 
6449 	if (rdev->raid_disk >= 0)
6450 		goto busy;
6451 
6452 kick_rdev:
6453 	if (mddev_is_clustered(mddev))
6454 		md_cluster_ops->remove_disk(mddev, rdev);
6455 
6456 	md_kick_rdev_from_array(rdev);
6457 	set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
6458 	if (mddev->thread)
6459 		md_wakeup_thread(mddev->thread);
6460 	else
6461 		md_update_sb(mddev, 1);
6462 	md_new_event(mddev);
6463 
6464 	return 0;
6465 busy:
6466 	pr_debug("md: cannot remove active disk %s from %s ...\n",
6467 		 bdevname(rdev->bdev,b), mdname(mddev));
6468 	return -EBUSY;
6469 }
6470 
6471 static int hot_add_disk(struct mddev *mddev, dev_t dev)
6472 {
6473 	char b[BDEVNAME_SIZE];
6474 	int err;
6475 	struct md_rdev *rdev;
6476 
6477 	if (!mddev->pers)
6478 		return -ENODEV;
6479 
6480 	if (mddev->major_version != 0) {
6481 		pr_warn("%s: HOT_ADD may only be used with version-0 superblocks.\n",
6482 			mdname(mddev));
6483 		return -EINVAL;
6484 	}
6485 	if (!mddev->pers->hot_add_disk) {
6486 		pr_warn("%s: personality does not support diskops!\n",
6487 			mdname(mddev));
6488 		return -EINVAL;
6489 	}
6490 
6491 	rdev = md_import_device(dev, -1, 0);
6492 	if (IS_ERR(rdev)) {
6493 		pr_warn("md: error, md_import_device() returned %ld\n",
6494 			PTR_ERR(rdev));
6495 		return -EINVAL;
6496 	}
6497 
6498 	if (mddev->persistent)
6499 		rdev->sb_start = calc_dev_sboffset(rdev);
6500 	else
6501 		rdev->sb_start = i_size_read(rdev->bdev->bd_inode) / 512;
6502 
6503 	rdev->sectors = rdev->sb_start;
6504 
6505 	if (test_bit(Faulty, &rdev->flags)) {
6506 		pr_warn("md: can not hot-add faulty %s disk to %s!\n",
6507 			bdevname(rdev->bdev,b), mdname(mddev));
6508 		err = -EINVAL;
6509 		goto abort_export;
6510 	}
6511 
6512 	clear_bit(In_sync, &rdev->flags);
6513 	rdev->desc_nr = -1;
6514 	rdev->saved_raid_disk = -1;
6515 	err = bind_rdev_to_array(rdev, mddev);
6516 	if (err)
6517 		goto abort_export;
6518 
6519 	/*
6520 	 * The rest should better be atomic, we can have disk failures
6521 	 * noticed in interrupt contexts ...
6522 	 */
6523 
6524 	rdev->raid_disk = -1;
6525 
6526 	set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
6527 	if (!mddev->thread)
6528 		md_update_sb(mddev, 1);
6529 	/*
6530 	 * Kick recovery, maybe this spare has to be added to the
6531 	 * array immediately.
6532 	 */
6533 	set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6534 	md_wakeup_thread(mddev->thread);
6535 	md_new_event(mddev);
6536 	return 0;
6537 
6538 abort_export:
6539 	export_rdev(rdev);
6540 	return err;
6541 }
6542 
6543 static int set_bitmap_file(struct mddev *mddev, int fd)
6544 {
6545 	int err = 0;
6546 
6547 	if (mddev->pers) {
6548 		if (!mddev->pers->quiesce || !mddev->thread)
6549 			return -EBUSY;
6550 		if (mddev->recovery || mddev->sync_thread)
6551 			return -EBUSY;
6552 		/* we should be able to change the bitmap.. */
6553 	}
6554 
6555 	if (fd >= 0) {
6556 		struct inode *inode;
6557 		struct file *f;
6558 
6559 		if (mddev->bitmap || mddev->bitmap_info.file)
6560 			return -EEXIST; /* cannot add when bitmap is present */
6561 		f = fget(fd);
6562 
6563 		if (f == NULL) {
6564 			pr_warn("%s: error: failed to get bitmap file\n",
6565 				mdname(mddev));
6566 			return -EBADF;
6567 		}
6568 
6569 		inode = f->f_mapping->host;
6570 		if (!S_ISREG(inode->i_mode)) {
6571 			pr_warn("%s: error: bitmap file must be a regular file\n",
6572 				mdname(mddev));
6573 			err = -EBADF;
6574 		} else if (!(f->f_mode & FMODE_WRITE)) {
6575 			pr_warn("%s: error: bitmap file must open for write\n",
6576 				mdname(mddev));
6577 			err = -EBADF;
6578 		} else if (atomic_read(&inode->i_writecount) != 1) {
6579 			pr_warn("%s: error: bitmap file is already in use\n",
6580 				mdname(mddev));
6581 			err = -EBUSY;
6582 		}
6583 		if (err) {
6584 			fput(f);
6585 			return err;
6586 		}
6587 		mddev->bitmap_info.file = f;
6588 		mddev->bitmap_info.offset = 0; /* file overrides offset */
6589 	} else if (mddev->bitmap == NULL)
6590 		return -ENOENT; /* cannot remove what isn't there */
6591 	err = 0;
6592 	if (mddev->pers) {
6593 		mddev->pers->quiesce(mddev, 1);
6594 		if (fd >= 0) {
6595 			struct bitmap *bitmap;
6596 
6597 			bitmap = bitmap_create(mddev, -1);
6598 			if (!IS_ERR(bitmap)) {
6599 				mddev->bitmap = bitmap;
6600 				err = bitmap_load(mddev);
6601 			} else
6602 				err = PTR_ERR(bitmap);
6603 		}
6604 		if (fd < 0 || err) {
6605 			bitmap_destroy(mddev);
6606 			fd = -1; /* make sure to put the file */
6607 		}
6608 		mddev->pers->quiesce(mddev, 0);
6609 	}
6610 	if (fd < 0) {
6611 		struct file *f = mddev->bitmap_info.file;
6612 		if (f) {
6613 			spin_lock(&mddev->lock);
6614 			mddev->bitmap_info.file = NULL;
6615 			spin_unlock(&mddev->lock);
6616 			fput(f);
6617 		}
6618 	}
6619 
6620 	return err;
6621 }
6622 
6623 /*
6624  * set_array_info is used two different ways
6625  * The original usage is when creating a new array.
6626  * In this usage, raid_disks is > 0 and it together with
6627  *  level, size, not_persistent,layout,chunksize determine the
6628  *  shape of the array.
6629  *  This will always create an array with a type-0.90.0 superblock.
6630  * The newer usage is when assembling an array.
6631  *  In this case raid_disks will be 0, and the major_version field is
6632  *  use to determine which style super-blocks are to be found on the devices.
6633  *  The minor and patch _version numbers are also kept incase the
6634  *  super_block handler wishes to interpret them.
6635  */
6636 static int set_array_info(struct mddev *mddev, mdu_array_info_t *info)
6637 {
6638 
6639 	if (info->raid_disks == 0) {
6640 		/* just setting version number for superblock loading */
6641 		if (info->major_version < 0 ||
6642 		    info->major_version >= ARRAY_SIZE(super_types) ||
6643 		    super_types[info->major_version].name == NULL) {
6644 			/* maybe try to auto-load a module? */
6645 			pr_warn("md: superblock version %d not known\n",
6646 				info->major_version);
6647 			return -EINVAL;
6648 		}
6649 		mddev->major_version = info->major_version;
6650 		mddev->minor_version = info->minor_version;
6651 		mddev->patch_version = info->patch_version;
6652 		mddev->persistent = !info->not_persistent;
6653 		/* ensure mddev_put doesn't delete this now that there
6654 		 * is some minimal configuration.
6655 		 */
6656 		mddev->ctime         = ktime_get_real_seconds();
6657 		return 0;
6658 	}
6659 	mddev->major_version = MD_MAJOR_VERSION;
6660 	mddev->minor_version = MD_MINOR_VERSION;
6661 	mddev->patch_version = MD_PATCHLEVEL_VERSION;
6662 	mddev->ctime         = ktime_get_real_seconds();
6663 
6664 	mddev->level         = info->level;
6665 	mddev->clevel[0]     = 0;
6666 	mddev->dev_sectors   = 2 * (sector_t)info->size;
6667 	mddev->raid_disks    = info->raid_disks;
6668 	/* don't set md_minor, it is determined by which /dev/md* was
6669 	 * openned
6670 	 */
6671 	if (info->state & (1<<MD_SB_CLEAN))
6672 		mddev->recovery_cp = MaxSector;
6673 	else
6674 		mddev->recovery_cp = 0;
6675 	mddev->persistent    = ! info->not_persistent;
6676 	mddev->external	     = 0;
6677 
6678 	mddev->layout        = info->layout;
6679 	mddev->chunk_sectors = info->chunk_size >> 9;
6680 
6681 	if (mddev->persistent) {
6682 		mddev->max_disks = MD_SB_DISKS;
6683 		mddev->flags = 0;
6684 		mddev->sb_flags = 0;
6685 	}
6686 	set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
6687 
6688 	mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9;
6689 	mddev->bitmap_info.default_space = 64*2 - (MD_SB_BYTES >> 9);
6690 	mddev->bitmap_info.offset = 0;
6691 
6692 	mddev->reshape_position = MaxSector;
6693 
6694 	/*
6695 	 * Generate a 128 bit UUID
6696 	 */
6697 	get_random_bytes(mddev->uuid, 16);
6698 
6699 	mddev->new_level = mddev->level;
6700 	mddev->new_chunk_sectors = mddev->chunk_sectors;
6701 	mddev->new_layout = mddev->layout;
6702 	mddev->delta_disks = 0;
6703 	mddev->reshape_backwards = 0;
6704 
6705 	return 0;
6706 }
6707 
6708 void md_set_array_sectors(struct mddev *mddev, sector_t array_sectors)
6709 {
6710 	WARN(!mddev_is_locked(mddev), "%s: unlocked mddev!\n", __func__);
6711 
6712 	if (mddev->external_size)
6713 		return;
6714 
6715 	mddev->array_sectors = array_sectors;
6716 }
6717 EXPORT_SYMBOL(md_set_array_sectors);
6718 
6719 static int update_size(struct mddev *mddev, sector_t num_sectors)
6720 {
6721 	struct md_rdev *rdev;
6722 	int rv;
6723 	int fit = (num_sectors == 0);
6724 	sector_t old_dev_sectors = mddev->dev_sectors;
6725 
6726 	if (mddev->pers->resize == NULL)
6727 		return -EINVAL;
6728 	/* The "num_sectors" is the number of sectors of each device that
6729 	 * is used.  This can only make sense for arrays with redundancy.
6730 	 * linear and raid0 always use whatever space is available. We can only
6731 	 * consider changing this number if no resync or reconstruction is
6732 	 * happening, and if the new size is acceptable. It must fit before the
6733 	 * sb_start or, if that is <data_offset, it must fit before the size
6734 	 * of each device.  If num_sectors is zero, we find the largest size
6735 	 * that fits.
6736 	 */
6737 	if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
6738 	    mddev->sync_thread)
6739 		return -EBUSY;
6740 	if (mddev->ro)
6741 		return -EROFS;
6742 
6743 	rdev_for_each(rdev, mddev) {
6744 		sector_t avail = rdev->sectors;
6745 
6746 		if (fit && (num_sectors == 0 || num_sectors > avail))
6747 			num_sectors = avail;
6748 		if (avail < num_sectors)
6749 			return -ENOSPC;
6750 	}
6751 	rv = mddev->pers->resize(mddev, num_sectors);
6752 	if (!rv) {
6753 		if (mddev_is_clustered(mddev))
6754 			md_cluster_ops->update_size(mddev, old_dev_sectors);
6755 		else if (mddev->queue) {
6756 			set_capacity(mddev->gendisk, mddev->array_sectors);
6757 			revalidate_disk(mddev->gendisk);
6758 		}
6759 	}
6760 	return rv;
6761 }
6762 
6763 static int update_raid_disks(struct mddev *mddev, int raid_disks)
6764 {
6765 	int rv;
6766 	struct md_rdev *rdev;
6767 	/* change the number of raid disks */
6768 	if (mddev->pers->check_reshape == NULL)
6769 		return -EINVAL;
6770 	if (mddev->ro)
6771 		return -EROFS;
6772 	if (raid_disks <= 0 ||
6773 	    (mddev->max_disks && raid_disks >= mddev->max_disks))
6774 		return -EINVAL;
6775 	if (mddev->sync_thread ||
6776 	    test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
6777 	    mddev->reshape_position != MaxSector)
6778 		return -EBUSY;
6779 
6780 	rdev_for_each(rdev, mddev) {
6781 		if (mddev->raid_disks < raid_disks &&
6782 		    rdev->data_offset < rdev->new_data_offset)
6783 			return -EINVAL;
6784 		if (mddev->raid_disks > raid_disks &&
6785 		    rdev->data_offset > rdev->new_data_offset)
6786 			return -EINVAL;
6787 	}
6788 
6789 	mddev->delta_disks = raid_disks - mddev->raid_disks;
6790 	if (mddev->delta_disks < 0)
6791 		mddev->reshape_backwards = 1;
6792 	else if (mddev->delta_disks > 0)
6793 		mddev->reshape_backwards = 0;
6794 
6795 	rv = mddev->pers->check_reshape(mddev);
6796 	if (rv < 0) {
6797 		mddev->delta_disks = 0;
6798 		mddev->reshape_backwards = 0;
6799 	}
6800 	return rv;
6801 }
6802 
6803 /*
6804  * update_array_info is used to change the configuration of an
6805  * on-line array.
6806  * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
6807  * fields in the info are checked against the array.
6808  * Any differences that cannot be handled will cause an error.
6809  * Normally, only one change can be managed at a time.
6810  */
6811 static int update_array_info(struct mddev *mddev, mdu_array_info_t *info)
6812 {
6813 	int rv = 0;
6814 	int cnt = 0;
6815 	int state = 0;
6816 
6817 	/* calculate expected state,ignoring low bits */
6818 	if (mddev->bitmap && mddev->bitmap_info.offset)
6819 		state |= (1 << MD_SB_BITMAP_PRESENT);
6820 
6821 	if (mddev->major_version != info->major_version ||
6822 	    mddev->minor_version != info->minor_version ||
6823 /*	    mddev->patch_version != info->patch_version || */
6824 	    mddev->ctime         != info->ctime         ||
6825 	    mddev->level         != info->level         ||
6826 /*	    mddev->layout        != info->layout        || */
6827 	    mddev->persistent	 != !info->not_persistent ||
6828 	    mddev->chunk_sectors != info->chunk_size >> 9 ||
6829 	    /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
6830 	    ((state^info->state) & 0xfffffe00)
6831 		)
6832 		return -EINVAL;
6833 	/* Check there is only one change */
6834 	if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
6835 		cnt++;
6836 	if (mddev->raid_disks != info->raid_disks)
6837 		cnt++;
6838 	if (mddev->layout != info->layout)
6839 		cnt++;
6840 	if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT))
6841 		cnt++;
6842 	if (cnt == 0)
6843 		return 0;
6844 	if (cnt > 1)
6845 		return -EINVAL;
6846 
6847 	if (mddev->layout != info->layout) {
6848 		/* Change layout
6849 		 * we don't need to do anything at the md level, the
6850 		 * personality will take care of it all.
6851 		 */
6852 		if (mddev->pers->check_reshape == NULL)
6853 			return -EINVAL;
6854 		else {
6855 			mddev->new_layout = info->layout;
6856 			rv = mddev->pers->check_reshape(mddev);
6857 			if (rv)
6858 				mddev->new_layout = mddev->layout;
6859 			return rv;
6860 		}
6861 	}
6862 	if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
6863 		rv = update_size(mddev, (sector_t)info->size * 2);
6864 
6865 	if (mddev->raid_disks    != info->raid_disks)
6866 		rv = update_raid_disks(mddev, info->raid_disks);
6867 
6868 	if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) {
6869 		if (mddev->pers->quiesce == NULL || mddev->thread == NULL) {
6870 			rv = -EINVAL;
6871 			goto err;
6872 		}
6873 		if (mddev->recovery || mddev->sync_thread) {
6874 			rv = -EBUSY;
6875 			goto err;
6876 		}
6877 		if (info->state & (1<<MD_SB_BITMAP_PRESENT)) {
6878 			struct bitmap *bitmap;
6879 			/* add the bitmap */
6880 			if (mddev->bitmap) {
6881 				rv = -EEXIST;
6882 				goto err;
6883 			}
6884 			if (mddev->bitmap_info.default_offset == 0) {
6885 				rv = -EINVAL;
6886 				goto err;
6887 			}
6888 			mddev->bitmap_info.offset =
6889 				mddev->bitmap_info.default_offset;
6890 			mddev->bitmap_info.space =
6891 				mddev->bitmap_info.default_space;
6892 			mddev->pers->quiesce(mddev, 1);
6893 			bitmap = bitmap_create(mddev, -1);
6894 			if (!IS_ERR(bitmap)) {
6895 				mddev->bitmap = bitmap;
6896 				rv = bitmap_load(mddev);
6897 			} else
6898 				rv = PTR_ERR(bitmap);
6899 			if (rv)
6900 				bitmap_destroy(mddev);
6901 			mddev->pers->quiesce(mddev, 0);
6902 		} else {
6903 			/* remove the bitmap */
6904 			if (!mddev->bitmap) {
6905 				rv = -ENOENT;
6906 				goto err;
6907 			}
6908 			if (mddev->bitmap->storage.file) {
6909 				rv = -EINVAL;
6910 				goto err;
6911 			}
6912 			if (mddev->bitmap_info.nodes) {
6913 				/* hold PW on all the bitmap lock */
6914 				if (md_cluster_ops->lock_all_bitmaps(mddev) <= 0) {
6915 					pr_warn("md: can't change bitmap to none since the array is in use by more than one node\n");
6916 					rv = -EPERM;
6917 					md_cluster_ops->unlock_all_bitmaps(mddev);
6918 					goto err;
6919 				}
6920 
6921 				mddev->bitmap_info.nodes = 0;
6922 				md_cluster_ops->leave(mddev);
6923 			}
6924 			mddev->pers->quiesce(mddev, 1);
6925 			bitmap_destroy(mddev);
6926 			mddev->pers->quiesce(mddev, 0);
6927 			mddev->bitmap_info.offset = 0;
6928 		}
6929 	}
6930 	md_update_sb(mddev, 1);
6931 	return rv;
6932 err:
6933 	return rv;
6934 }
6935 
6936 static int set_disk_faulty(struct mddev *mddev, dev_t dev)
6937 {
6938 	struct md_rdev *rdev;
6939 	int err = 0;
6940 
6941 	if (mddev->pers == NULL)
6942 		return -ENODEV;
6943 
6944 	rcu_read_lock();
6945 	rdev = find_rdev_rcu(mddev, dev);
6946 	if (!rdev)
6947 		err =  -ENODEV;
6948 	else {
6949 		md_error(mddev, rdev);
6950 		if (!test_bit(Faulty, &rdev->flags))
6951 			err = -EBUSY;
6952 	}
6953 	rcu_read_unlock();
6954 	return err;
6955 }
6956 
6957 /*
6958  * We have a problem here : there is no easy way to give a CHS
6959  * virtual geometry. We currently pretend that we have a 2 heads
6960  * 4 sectors (with a BIG number of cylinders...). This drives
6961  * dosfs just mad... ;-)
6962  */
6963 static int md_getgeo(struct block_device *bdev, struct hd_geometry *geo)
6964 {
6965 	struct mddev *mddev = bdev->bd_disk->private_data;
6966 
6967 	geo->heads = 2;
6968 	geo->sectors = 4;
6969 	geo->cylinders = mddev->array_sectors / 8;
6970 	return 0;
6971 }
6972 
6973 static inline bool md_ioctl_valid(unsigned int cmd)
6974 {
6975 	switch (cmd) {
6976 	case ADD_NEW_DISK:
6977 	case BLKROSET:
6978 	case GET_ARRAY_INFO:
6979 	case GET_BITMAP_FILE:
6980 	case GET_DISK_INFO:
6981 	case HOT_ADD_DISK:
6982 	case HOT_REMOVE_DISK:
6983 	case RAID_AUTORUN:
6984 	case RAID_VERSION:
6985 	case RESTART_ARRAY_RW:
6986 	case RUN_ARRAY:
6987 	case SET_ARRAY_INFO:
6988 	case SET_BITMAP_FILE:
6989 	case SET_DISK_FAULTY:
6990 	case STOP_ARRAY:
6991 	case STOP_ARRAY_RO:
6992 	case CLUSTERED_DISK_NACK:
6993 		return true;
6994 	default:
6995 		return false;
6996 	}
6997 }
6998 
6999 static int md_ioctl(struct block_device *bdev, fmode_t mode,
7000 			unsigned int cmd, unsigned long arg)
7001 {
7002 	int err = 0;
7003 	void __user *argp = (void __user *)arg;
7004 	struct mddev *mddev = NULL;
7005 	int ro;
7006 	bool did_set_md_closing = false;
7007 
7008 	if (!md_ioctl_valid(cmd))
7009 		return -ENOTTY;
7010 
7011 	switch (cmd) {
7012 	case RAID_VERSION:
7013 	case GET_ARRAY_INFO:
7014 	case GET_DISK_INFO:
7015 		break;
7016 	default:
7017 		if (!capable(CAP_SYS_ADMIN))
7018 			return -EACCES;
7019 	}
7020 
7021 	/*
7022 	 * Commands dealing with the RAID driver but not any
7023 	 * particular array:
7024 	 */
7025 	switch (cmd) {
7026 	case RAID_VERSION:
7027 		err = get_version(argp);
7028 		goto out;
7029 
7030 #ifndef MODULE
7031 	case RAID_AUTORUN:
7032 		err = 0;
7033 		autostart_arrays(arg);
7034 		goto out;
7035 #endif
7036 	default:;
7037 	}
7038 
7039 	/*
7040 	 * Commands creating/starting a new array:
7041 	 */
7042 
7043 	mddev = bdev->bd_disk->private_data;
7044 
7045 	if (!mddev) {
7046 		BUG();
7047 		goto out;
7048 	}
7049 
7050 	/* Some actions do not requires the mutex */
7051 	switch (cmd) {
7052 	case GET_ARRAY_INFO:
7053 		if (!mddev->raid_disks && !mddev->external)
7054 			err = -ENODEV;
7055 		else
7056 			err = get_array_info(mddev, argp);
7057 		goto out;
7058 
7059 	case GET_DISK_INFO:
7060 		if (!mddev->raid_disks && !mddev->external)
7061 			err = -ENODEV;
7062 		else
7063 			err = get_disk_info(mddev, argp);
7064 		goto out;
7065 
7066 	case SET_DISK_FAULTY:
7067 		err = set_disk_faulty(mddev, new_decode_dev(arg));
7068 		goto out;
7069 
7070 	case GET_BITMAP_FILE:
7071 		err = get_bitmap_file(mddev, argp);
7072 		goto out;
7073 
7074 	}
7075 
7076 	if (cmd == ADD_NEW_DISK)
7077 		/* need to ensure md_delayed_delete() has completed */
7078 		flush_workqueue(md_misc_wq);
7079 
7080 	if (cmd == HOT_REMOVE_DISK)
7081 		/* need to ensure recovery thread has run */
7082 		wait_event_interruptible_timeout(mddev->sb_wait,
7083 						 !test_bit(MD_RECOVERY_NEEDED,
7084 							   &mddev->recovery),
7085 						 msecs_to_jiffies(5000));
7086 	if (cmd == STOP_ARRAY || cmd == STOP_ARRAY_RO) {
7087 		/* Need to flush page cache, and ensure no-one else opens
7088 		 * and writes
7089 		 */
7090 		mutex_lock(&mddev->open_mutex);
7091 		if (mddev->pers && atomic_read(&mddev->openers) > 1) {
7092 			mutex_unlock(&mddev->open_mutex);
7093 			err = -EBUSY;
7094 			goto out;
7095 		}
7096 		WARN_ON_ONCE(test_bit(MD_CLOSING, &mddev->flags));
7097 		set_bit(MD_CLOSING, &mddev->flags);
7098 		did_set_md_closing = true;
7099 		mutex_unlock(&mddev->open_mutex);
7100 		sync_blockdev(bdev);
7101 	}
7102 	err = mddev_lock(mddev);
7103 	if (err) {
7104 		pr_debug("md: ioctl lock interrupted, reason %d, cmd %d\n",
7105 			 err, cmd);
7106 		goto out;
7107 	}
7108 
7109 	if (cmd == SET_ARRAY_INFO) {
7110 		mdu_array_info_t info;
7111 		if (!arg)
7112 			memset(&info, 0, sizeof(info));
7113 		else if (copy_from_user(&info, argp, sizeof(info))) {
7114 			err = -EFAULT;
7115 			goto unlock;
7116 		}
7117 		if (mddev->pers) {
7118 			err = update_array_info(mddev, &info);
7119 			if (err) {
7120 				pr_warn("md: couldn't update array info. %d\n", err);
7121 				goto unlock;
7122 			}
7123 			goto unlock;
7124 		}
7125 		if (!list_empty(&mddev->disks)) {
7126 			pr_warn("md: array %s already has disks!\n", mdname(mddev));
7127 			err = -EBUSY;
7128 			goto unlock;
7129 		}
7130 		if (mddev->raid_disks) {
7131 			pr_warn("md: array %s already initialised!\n", mdname(mddev));
7132 			err = -EBUSY;
7133 			goto unlock;
7134 		}
7135 		err = set_array_info(mddev, &info);
7136 		if (err) {
7137 			pr_warn("md: couldn't set array info. %d\n", err);
7138 			goto unlock;
7139 		}
7140 		goto unlock;
7141 	}
7142 
7143 	/*
7144 	 * Commands querying/configuring an existing array:
7145 	 */
7146 	/* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
7147 	 * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */
7148 	if ((!mddev->raid_disks && !mddev->external)
7149 	    && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY
7150 	    && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE
7151 	    && cmd != GET_BITMAP_FILE) {
7152 		err = -ENODEV;
7153 		goto unlock;
7154 	}
7155 
7156 	/*
7157 	 * Commands even a read-only array can execute:
7158 	 */
7159 	switch (cmd) {
7160 	case RESTART_ARRAY_RW:
7161 		err = restart_array(mddev);
7162 		goto unlock;
7163 
7164 	case STOP_ARRAY:
7165 		err = do_md_stop(mddev, 0, bdev);
7166 		goto unlock;
7167 
7168 	case STOP_ARRAY_RO:
7169 		err = md_set_readonly(mddev, bdev);
7170 		goto unlock;
7171 
7172 	case HOT_REMOVE_DISK:
7173 		err = hot_remove_disk(mddev, new_decode_dev(arg));
7174 		goto unlock;
7175 
7176 	case ADD_NEW_DISK:
7177 		/* We can support ADD_NEW_DISK on read-only arrays
7178 		 * only if we are re-adding a preexisting device.
7179 		 * So require mddev->pers and MD_DISK_SYNC.
7180 		 */
7181 		if (mddev->pers) {
7182 			mdu_disk_info_t info;
7183 			if (copy_from_user(&info, argp, sizeof(info)))
7184 				err = -EFAULT;
7185 			else if (!(info.state & (1<<MD_DISK_SYNC)))
7186 				/* Need to clear read-only for this */
7187 				break;
7188 			else
7189 				err = add_new_disk(mddev, &info);
7190 			goto unlock;
7191 		}
7192 		break;
7193 
7194 	case BLKROSET:
7195 		if (get_user(ro, (int __user *)(arg))) {
7196 			err = -EFAULT;
7197 			goto unlock;
7198 		}
7199 		err = -EINVAL;
7200 
7201 		/* if the bdev is going readonly the value of mddev->ro
7202 		 * does not matter, no writes are coming
7203 		 */
7204 		if (ro)
7205 			goto unlock;
7206 
7207 		/* are we are already prepared for writes? */
7208 		if (mddev->ro != 1)
7209 			goto unlock;
7210 
7211 		/* transitioning to readauto need only happen for
7212 		 * arrays that call md_write_start
7213 		 */
7214 		if (mddev->pers) {
7215 			err = restart_array(mddev);
7216 			if (err == 0) {
7217 				mddev->ro = 2;
7218 				set_disk_ro(mddev->gendisk, 0);
7219 			}
7220 		}
7221 		goto unlock;
7222 	}
7223 
7224 	/*
7225 	 * The remaining ioctls are changing the state of the
7226 	 * superblock, so we do not allow them on read-only arrays.
7227 	 */
7228 	if (mddev->ro && mddev->pers) {
7229 		if (mddev->ro == 2) {
7230 			mddev->ro = 0;
7231 			sysfs_notify_dirent_safe(mddev->sysfs_state);
7232 			set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7233 			/* mddev_unlock will wake thread */
7234 			/* If a device failed while we were read-only, we
7235 			 * need to make sure the metadata is updated now.
7236 			 */
7237 			if (test_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags)) {
7238 				mddev_unlock(mddev);
7239 				wait_event(mddev->sb_wait,
7240 					   !test_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags) &&
7241 					   !test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags));
7242 				mddev_lock_nointr(mddev);
7243 			}
7244 		} else {
7245 			err = -EROFS;
7246 			goto unlock;
7247 		}
7248 	}
7249 
7250 	switch (cmd) {
7251 	case ADD_NEW_DISK:
7252 	{
7253 		mdu_disk_info_t info;
7254 		if (copy_from_user(&info, argp, sizeof(info)))
7255 			err = -EFAULT;
7256 		else
7257 			err = add_new_disk(mddev, &info);
7258 		goto unlock;
7259 	}
7260 
7261 	case CLUSTERED_DISK_NACK:
7262 		if (mddev_is_clustered(mddev))
7263 			md_cluster_ops->new_disk_ack(mddev, false);
7264 		else
7265 			err = -EINVAL;
7266 		goto unlock;
7267 
7268 	case HOT_ADD_DISK:
7269 		err = hot_add_disk(mddev, new_decode_dev(arg));
7270 		goto unlock;
7271 
7272 	case RUN_ARRAY:
7273 		err = do_md_run(mddev);
7274 		goto unlock;
7275 
7276 	case SET_BITMAP_FILE:
7277 		err = set_bitmap_file(mddev, (int)arg);
7278 		goto unlock;
7279 
7280 	default:
7281 		err = -EINVAL;
7282 		goto unlock;
7283 	}
7284 
7285 unlock:
7286 	if (mddev->hold_active == UNTIL_IOCTL &&
7287 	    err != -EINVAL)
7288 		mddev->hold_active = 0;
7289 	mddev_unlock(mddev);
7290 out:
7291 	if(did_set_md_closing)
7292 		clear_bit(MD_CLOSING, &mddev->flags);
7293 	return err;
7294 }
7295 #ifdef CONFIG_COMPAT
7296 static int md_compat_ioctl(struct block_device *bdev, fmode_t mode,
7297 		    unsigned int cmd, unsigned long arg)
7298 {
7299 	switch (cmd) {
7300 	case HOT_REMOVE_DISK:
7301 	case HOT_ADD_DISK:
7302 	case SET_DISK_FAULTY:
7303 	case SET_BITMAP_FILE:
7304 		/* These take in integer arg, do not convert */
7305 		break;
7306 	default:
7307 		arg = (unsigned long)compat_ptr(arg);
7308 		break;
7309 	}
7310 
7311 	return md_ioctl(bdev, mode, cmd, arg);
7312 }
7313 #endif /* CONFIG_COMPAT */
7314 
7315 static int md_open(struct block_device *bdev, fmode_t mode)
7316 {
7317 	/*
7318 	 * Succeed if we can lock the mddev, which confirms that
7319 	 * it isn't being stopped right now.
7320 	 */
7321 	struct mddev *mddev = mddev_find(bdev->bd_dev);
7322 	int err;
7323 
7324 	if (!mddev)
7325 		return -ENODEV;
7326 
7327 	if (mddev->gendisk != bdev->bd_disk) {
7328 		/* we are racing with mddev_put which is discarding this
7329 		 * bd_disk.
7330 		 */
7331 		mddev_put(mddev);
7332 		/* Wait until bdev->bd_disk is definitely gone */
7333 		flush_workqueue(md_misc_wq);
7334 		/* Then retry the open from the top */
7335 		return -ERESTARTSYS;
7336 	}
7337 	BUG_ON(mddev != bdev->bd_disk->private_data);
7338 
7339 	if ((err = mutex_lock_interruptible(&mddev->open_mutex)))
7340 		goto out;
7341 
7342 	if (test_bit(MD_CLOSING, &mddev->flags)) {
7343 		mutex_unlock(&mddev->open_mutex);
7344 		err = -ENODEV;
7345 		goto out;
7346 	}
7347 
7348 	err = 0;
7349 	atomic_inc(&mddev->openers);
7350 	mutex_unlock(&mddev->open_mutex);
7351 
7352 	check_disk_change(bdev);
7353  out:
7354 	if (err)
7355 		mddev_put(mddev);
7356 	return err;
7357 }
7358 
7359 static void md_release(struct gendisk *disk, fmode_t mode)
7360 {
7361 	struct mddev *mddev = disk->private_data;
7362 
7363 	BUG_ON(!mddev);
7364 	atomic_dec(&mddev->openers);
7365 	mddev_put(mddev);
7366 }
7367 
7368 static int md_media_changed(struct gendisk *disk)
7369 {
7370 	struct mddev *mddev = disk->private_data;
7371 
7372 	return mddev->changed;
7373 }
7374 
7375 static int md_revalidate(struct gendisk *disk)
7376 {
7377 	struct mddev *mddev = disk->private_data;
7378 
7379 	mddev->changed = 0;
7380 	return 0;
7381 }
7382 static const struct block_device_operations md_fops =
7383 {
7384 	.owner		= THIS_MODULE,
7385 	.open		= md_open,
7386 	.release	= md_release,
7387 	.ioctl		= md_ioctl,
7388 #ifdef CONFIG_COMPAT
7389 	.compat_ioctl	= md_compat_ioctl,
7390 #endif
7391 	.getgeo		= md_getgeo,
7392 	.media_changed  = md_media_changed,
7393 	.revalidate_disk= md_revalidate,
7394 };
7395 
7396 static int md_thread(void *arg)
7397 {
7398 	struct md_thread *thread = arg;
7399 
7400 	/*
7401 	 * md_thread is a 'system-thread', it's priority should be very
7402 	 * high. We avoid resource deadlocks individually in each
7403 	 * raid personality. (RAID5 does preallocation) We also use RR and
7404 	 * the very same RT priority as kswapd, thus we will never get
7405 	 * into a priority inversion deadlock.
7406 	 *
7407 	 * we definitely have to have equal or higher priority than
7408 	 * bdflush, otherwise bdflush will deadlock if there are too
7409 	 * many dirty RAID5 blocks.
7410 	 */
7411 
7412 	allow_signal(SIGKILL);
7413 	while (!kthread_should_stop()) {
7414 
7415 		/* We need to wait INTERRUPTIBLE so that
7416 		 * we don't add to the load-average.
7417 		 * That means we need to be sure no signals are
7418 		 * pending
7419 		 */
7420 		if (signal_pending(current))
7421 			flush_signals(current);
7422 
7423 		wait_event_interruptible_timeout
7424 			(thread->wqueue,
7425 			 test_bit(THREAD_WAKEUP, &thread->flags)
7426 			 || kthread_should_stop() || kthread_should_park(),
7427 			 thread->timeout);
7428 
7429 		clear_bit(THREAD_WAKEUP, &thread->flags);
7430 		if (kthread_should_park())
7431 			kthread_parkme();
7432 		if (!kthread_should_stop())
7433 			thread->run(thread);
7434 	}
7435 
7436 	return 0;
7437 }
7438 
7439 void md_wakeup_thread(struct md_thread *thread)
7440 {
7441 	if (thread) {
7442 		pr_debug("md: waking up MD thread %s.\n", thread->tsk->comm);
7443 		if (!test_and_set_bit(THREAD_WAKEUP, &thread->flags))
7444 			wake_up(&thread->wqueue);
7445 	}
7446 }
7447 EXPORT_SYMBOL(md_wakeup_thread);
7448 
7449 struct md_thread *md_register_thread(void (*run) (struct md_thread *),
7450 		struct mddev *mddev, const char *name)
7451 {
7452 	struct md_thread *thread;
7453 
7454 	thread = kzalloc(sizeof(struct md_thread), GFP_KERNEL);
7455 	if (!thread)
7456 		return NULL;
7457 
7458 	init_waitqueue_head(&thread->wqueue);
7459 
7460 	thread->run = run;
7461 	thread->mddev = mddev;
7462 	thread->timeout = MAX_SCHEDULE_TIMEOUT;
7463 	thread->tsk = kthread_run(md_thread, thread,
7464 				  "%s_%s",
7465 				  mdname(thread->mddev),
7466 				  name);
7467 	if (IS_ERR(thread->tsk)) {
7468 		kfree(thread);
7469 		return NULL;
7470 	}
7471 	return thread;
7472 }
7473 EXPORT_SYMBOL(md_register_thread);
7474 
7475 void md_unregister_thread(struct md_thread **threadp)
7476 {
7477 	struct md_thread *thread = *threadp;
7478 	if (!thread)
7479 		return;
7480 	pr_debug("interrupting MD-thread pid %d\n", task_pid_nr(thread->tsk));
7481 	/* Locking ensures that mddev_unlock does not wake_up a
7482 	 * non-existent thread
7483 	 */
7484 	spin_lock(&pers_lock);
7485 	*threadp = NULL;
7486 	spin_unlock(&pers_lock);
7487 
7488 	kthread_stop(thread->tsk);
7489 	kfree(thread);
7490 }
7491 EXPORT_SYMBOL(md_unregister_thread);
7492 
7493 void md_error(struct mddev *mddev, struct md_rdev *rdev)
7494 {
7495 	if (!rdev || test_bit(Faulty, &rdev->flags))
7496 		return;
7497 
7498 	if (!mddev->pers || !mddev->pers->error_handler)
7499 		return;
7500 	mddev->pers->error_handler(mddev,rdev);
7501 	if (mddev->degraded)
7502 		set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
7503 	sysfs_notify_dirent_safe(rdev->sysfs_state);
7504 	set_bit(MD_RECOVERY_INTR, &mddev->recovery);
7505 	set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7506 	md_wakeup_thread(mddev->thread);
7507 	if (mddev->event_work.func)
7508 		queue_work(md_misc_wq, &mddev->event_work);
7509 	md_new_event(mddev);
7510 }
7511 EXPORT_SYMBOL(md_error);
7512 
7513 /* seq_file implementation /proc/mdstat */
7514 
7515 static void status_unused(struct seq_file *seq)
7516 {
7517 	int i = 0;
7518 	struct md_rdev *rdev;
7519 
7520 	seq_printf(seq, "unused devices: ");
7521 
7522 	list_for_each_entry(rdev, &pending_raid_disks, same_set) {
7523 		char b[BDEVNAME_SIZE];
7524 		i++;
7525 		seq_printf(seq, "%s ",
7526 			      bdevname(rdev->bdev,b));
7527 	}
7528 	if (!i)
7529 		seq_printf(seq, "<none>");
7530 
7531 	seq_printf(seq, "\n");
7532 }
7533 
7534 static int status_resync(struct seq_file *seq, struct mddev *mddev)
7535 {
7536 	sector_t max_sectors, resync, res;
7537 	unsigned long dt, db;
7538 	sector_t rt;
7539 	int scale;
7540 	unsigned int per_milli;
7541 
7542 	if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ||
7543 	    test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
7544 		max_sectors = mddev->resync_max_sectors;
7545 	else
7546 		max_sectors = mddev->dev_sectors;
7547 
7548 	resync = mddev->curr_resync;
7549 	if (resync <= 3) {
7550 		if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
7551 			/* Still cleaning up */
7552 			resync = max_sectors;
7553 	} else
7554 		resync -= atomic_read(&mddev->recovery_active);
7555 
7556 	if (resync == 0) {
7557 		if (mddev->recovery_cp < MaxSector) {
7558 			seq_printf(seq, "\tresync=PENDING");
7559 			return 1;
7560 		}
7561 		return 0;
7562 	}
7563 	if (resync < 3) {
7564 		seq_printf(seq, "\tresync=DELAYED");
7565 		return 1;
7566 	}
7567 
7568 	WARN_ON(max_sectors == 0);
7569 	/* Pick 'scale' such that (resync>>scale)*1000 will fit
7570 	 * in a sector_t, and (max_sectors>>scale) will fit in a
7571 	 * u32, as those are the requirements for sector_div.
7572 	 * Thus 'scale' must be at least 10
7573 	 */
7574 	scale = 10;
7575 	if (sizeof(sector_t) > sizeof(unsigned long)) {
7576 		while ( max_sectors/2 > (1ULL<<(scale+32)))
7577 			scale++;
7578 	}
7579 	res = (resync>>scale)*1000;
7580 	sector_div(res, (u32)((max_sectors>>scale)+1));
7581 
7582 	per_milli = res;
7583 	{
7584 		int i, x = per_milli/50, y = 20-x;
7585 		seq_printf(seq, "[");
7586 		for (i = 0; i < x; i++)
7587 			seq_printf(seq, "=");
7588 		seq_printf(seq, ">");
7589 		for (i = 0; i < y; i++)
7590 			seq_printf(seq, ".");
7591 		seq_printf(seq, "] ");
7592 	}
7593 	seq_printf(seq, " %s =%3u.%u%% (%llu/%llu)",
7594 		   (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)?
7595 		    "reshape" :
7596 		    (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)?
7597 		     "check" :
7598 		     (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
7599 		      "resync" : "recovery"))),
7600 		   per_milli/10, per_milli % 10,
7601 		   (unsigned long long) resync/2,
7602 		   (unsigned long long) max_sectors/2);
7603 
7604 	/*
7605 	 * dt: time from mark until now
7606 	 * db: blocks written from mark until now
7607 	 * rt: remaining time
7608 	 *
7609 	 * rt is a sector_t, so could be 32bit or 64bit.
7610 	 * So we divide before multiply in case it is 32bit and close
7611 	 * to the limit.
7612 	 * We scale the divisor (db) by 32 to avoid losing precision
7613 	 * near the end of resync when the number of remaining sectors
7614 	 * is close to 'db'.
7615 	 * We then divide rt by 32 after multiplying by db to compensate.
7616 	 * The '+1' avoids division by zero if db is very small.
7617 	 */
7618 	dt = ((jiffies - mddev->resync_mark) / HZ);
7619 	if (!dt) dt++;
7620 	db = (mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active))
7621 		- mddev->resync_mark_cnt;
7622 
7623 	rt = max_sectors - resync;    /* number of remaining sectors */
7624 	sector_div(rt, db/32+1);
7625 	rt *= dt;
7626 	rt >>= 5;
7627 
7628 	seq_printf(seq, " finish=%lu.%lumin", (unsigned long)rt / 60,
7629 		   ((unsigned long)rt % 60)/6);
7630 
7631 	seq_printf(seq, " speed=%ldK/sec", db/2/dt);
7632 	return 1;
7633 }
7634 
7635 static void *md_seq_start(struct seq_file *seq, loff_t *pos)
7636 {
7637 	struct list_head *tmp;
7638 	loff_t l = *pos;
7639 	struct mddev *mddev;
7640 
7641 	if (l >= 0x10000)
7642 		return NULL;
7643 	if (!l--)
7644 		/* header */
7645 		return (void*)1;
7646 
7647 	spin_lock(&all_mddevs_lock);
7648 	list_for_each(tmp,&all_mddevs)
7649 		if (!l--) {
7650 			mddev = list_entry(tmp, struct mddev, all_mddevs);
7651 			mddev_get(mddev);
7652 			spin_unlock(&all_mddevs_lock);
7653 			return mddev;
7654 		}
7655 	spin_unlock(&all_mddevs_lock);
7656 	if (!l--)
7657 		return (void*)2;/* tail */
7658 	return NULL;
7659 }
7660 
7661 static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
7662 {
7663 	struct list_head *tmp;
7664 	struct mddev *next_mddev, *mddev = v;
7665 
7666 	++*pos;
7667 	if (v == (void*)2)
7668 		return NULL;
7669 
7670 	spin_lock(&all_mddevs_lock);
7671 	if (v == (void*)1)
7672 		tmp = all_mddevs.next;
7673 	else
7674 		tmp = mddev->all_mddevs.next;
7675 	if (tmp != &all_mddevs)
7676 		next_mddev = mddev_get(list_entry(tmp,struct mddev,all_mddevs));
7677 	else {
7678 		next_mddev = (void*)2;
7679 		*pos = 0x10000;
7680 	}
7681 	spin_unlock(&all_mddevs_lock);
7682 
7683 	if (v != (void*)1)
7684 		mddev_put(mddev);
7685 	return next_mddev;
7686 
7687 }
7688 
7689 static void md_seq_stop(struct seq_file *seq, void *v)
7690 {
7691 	struct mddev *mddev = v;
7692 
7693 	if (mddev && v != (void*)1 && v != (void*)2)
7694 		mddev_put(mddev);
7695 }
7696 
7697 static int md_seq_show(struct seq_file *seq, void *v)
7698 {
7699 	struct mddev *mddev = v;
7700 	sector_t sectors;
7701 	struct md_rdev *rdev;
7702 
7703 	if (v == (void*)1) {
7704 		struct md_personality *pers;
7705 		seq_printf(seq, "Personalities : ");
7706 		spin_lock(&pers_lock);
7707 		list_for_each_entry(pers, &pers_list, list)
7708 			seq_printf(seq, "[%s] ", pers->name);
7709 
7710 		spin_unlock(&pers_lock);
7711 		seq_printf(seq, "\n");
7712 		seq->poll_event = atomic_read(&md_event_count);
7713 		return 0;
7714 	}
7715 	if (v == (void*)2) {
7716 		status_unused(seq);
7717 		return 0;
7718 	}
7719 
7720 	spin_lock(&mddev->lock);
7721 	if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
7722 		seq_printf(seq, "%s : %sactive", mdname(mddev),
7723 						mddev->pers ? "" : "in");
7724 		if (mddev->pers) {
7725 			if (mddev->ro==1)
7726 				seq_printf(seq, " (read-only)");
7727 			if (mddev->ro==2)
7728 				seq_printf(seq, " (auto-read-only)");
7729 			seq_printf(seq, " %s", mddev->pers->name);
7730 		}
7731 
7732 		sectors = 0;
7733 		rcu_read_lock();
7734 		rdev_for_each_rcu(rdev, mddev) {
7735 			char b[BDEVNAME_SIZE];
7736 			seq_printf(seq, " %s[%d]",
7737 				bdevname(rdev->bdev,b), rdev->desc_nr);
7738 			if (test_bit(WriteMostly, &rdev->flags))
7739 				seq_printf(seq, "(W)");
7740 			if (test_bit(Journal, &rdev->flags))
7741 				seq_printf(seq, "(J)");
7742 			if (test_bit(Faulty, &rdev->flags)) {
7743 				seq_printf(seq, "(F)");
7744 				continue;
7745 			}
7746 			if (rdev->raid_disk < 0)
7747 				seq_printf(seq, "(S)"); /* spare */
7748 			if (test_bit(Replacement, &rdev->flags))
7749 				seq_printf(seq, "(R)");
7750 			sectors += rdev->sectors;
7751 		}
7752 		rcu_read_unlock();
7753 
7754 		if (!list_empty(&mddev->disks)) {
7755 			if (mddev->pers)
7756 				seq_printf(seq, "\n      %llu blocks",
7757 					   (unsigned long long)
7758 					   mddev->array_sectors / 2);
7759 			else
7760 				seq_printf(seq, "\n      %llu blocks",
7761 					   (unsigned long long)sectors / 2);
7762 		}
7763 		if (mddev->persistent) {
7764 			if (mddev->major_version != 0 ||
7765 			    mddev->minor_version != 90) {
7766 				seq_printf(seq," super %d.%d",
7767 					   mddev->major_version,
7768 					   mddev->minor_version);
7769 			}
7770 		} else if (mddev->external)
7771 			seq_printf(seq, " super external:%s",
7772 				   mddev->metadata_type);
7773 		else
7774 			seq_printf(seq, " super non-persistent");
7775 
7776 		if (mddev->pers) {
7777 			mddev->pers->status(seq, mddev);
7778 			seq_printf(seq, "\n      ");
7779 			if (mddev->pers->sync_request) {
7780 				if (status_resync(seq, mddev))
7781 					seq_printf(seq, "\n      ");
7782 			}
7783 		} else
7784 			seq_printf(seq, "\n       ");
7785 
7786 		bitmap_status(seq, mddev->bitmap);
7787 
7788 		seq_printf(seq, "\n");
7789 	}
7790 	spin_unlock(&mddev->lock);
7791 
7792 	return 0;
7793 }
7794 
7795 static const struct seq_operations md_seq_ops = {
7796 	.start  = md_seq_start,
7797 	.next   = md_seq_next,
7798 	.stop   = md_seq_stop,
7799 	.show   = md_seq_show,
7800 };
7801 
7802 static int md_seq_open(struct inode *inode, struct file *file)
7803 {
7804 	struct seq_file *seq;
7805 	int error;
7806 
7807 	error = seq_open(file, &md_seq_ops);
7808 	if (error)
7809 		return error;
7810 
7811 	seq = file->private_data;
7812 	seq->poll_event = atomic_read(&md_event_count);
7813 	return error;
7814 }
7815 
7816 static int md_unloading;
7817 static unsigned int mdstat_poll(struct file *filp, poll_table *wait)
7818 {
7819 	struct seq_file *seq = filp->private_data;
7820 	int mask;
7821 
7822 	if (md_unloading)
7823 		return POLLIN|POLLRDNORM|POLLERR|POLLPRI;
7824 	poll_wait(filp, &md_event_waiters, wait);
7825 
7826 	/* always allow read */
7827 	mask = POLLIN | POLLRDNORM;
7828 
7829 	if (seq->poll_event != atomic_read(&md_event_count))
7830 		mask |= POLLERR | POLLPRI;
7831 	return mask;
7832 }
7833 
7834 static const struct file_operations md_seq_fops = {
7835 	.owner		= THIS_MODULE,
7836 	.open           = md_seq_open,
7837 	.read           = seq_read,
7838 	.llseek         = seq_lseek,
7839 	.release	= seq_release_private,
7840 	.poll		= mdstat_poll,
7841 };
7842 
7843 int register_md_personality(struct md_personality *p)
7844 {
7845 	pr_debug("md: %s personality registered for level %d\n",
7846 		 p->name, p->level);
7847 	spin_lock(&pers_lock);
7848 	list_add_tail(&p->list, &pers_list);
7849 	spin_unlock(&pers_lock);
7850 	return 0;
7851 }
7852 EXPORT_SYMBOL(register_md_personality);
7853 
7854 int unregister_md_personality(struct md_personality *p)
7855 {
7856 	pr_debug("md: %s personality unregistered\n", p->name);
7857 	spin_lock(&pers_lock);
7858 	list_del_init(&p->list);
7859 	spin_unlock(&pers_lock);
7860 	return 0;
7861 }
7862 EXPORT_SYMBOL(unregister_md_personality);
7863 
7864 int register_md_cluster_operations(struct md_cluster_operations *ops,
7865 				   struct module *module)
7866 {
7867 	int ret = 0;
7868 	spin_lock(&pers_lock);
7869 	if (md_cluster_ops != NULL)
7870 		ret = -EALREADY;
7871 	else {
7872 		md_cluster_ops = ops;
7873 		md_cluster_mod = module;
7874 	}
7875 	spin_unlock(&pers_lock);
7876 	return ret;
7877 }
7878 EXPORT_SYMBOL(register_md_cluster_operations);
7879 
7880 int unregister_md_cluster_operations(void)
7881 {
7882 	spin_lock(&pers_lock);
7883 	md_cluster_ops = NULL;
7884 	spin_unlock(&pers_lock);
7885 	return 0;
7886 }
7887 EXPORT_SYMBOL(unregister_md_cluster_operations);
7888 
7889 int md_setup_cluster(struct mddev *mddev, int nodes)
7890 {
7891 	if (!md_cluster_ops)
7892 		request_module("md-cluster");
7893 	spin_lock(&pers_lock);
7894 	/* ensure module won't be unloaded */
7895 	if (!md_cluster_ops || !try_module_get(md_cluster_mod)) {
7896 		pr_warn("can't find md-cluster module or get it's reference.\n");
7897 		spin_unlock(&pers_lock);
7898 		return -ENOENT;
7899 	}
7900 	spin_unlock(&pers_lock);
7901 
7902 	return md_cluster_ops->join(mddev, nodes);
7903 }
7904 
7905 void md_cluster_stop(struct mddev *mddev)
7906 {
7907 	if (!md_cluster_ops)
7908 		return;
7909 	md_cluster_ops->leave(mddev);
7910 	module_put(md_cluster_mod);
7911 }
7912 
7913 static int is_mddev_idle(struct mddev *mddev, int init)
7914 {
7915 	struct md_rdev *rdev;
7916 	int idle;
7917 	int curr_events;
7918 
7919 	idle = 1;
7920 	rcu_read_lock();
7921 	rdev_for_each_rcu(rdev, mddev) {
7922 		struct gendisk *disk = rdev->bdev->bd_contains->bd_disk;
7923 		curr_events = (int)part_stat_read(&disk->part0, sectors[0]) +
7924 			      (int)part_stat_read(&disk->part0, sectors[1]) -
7925 			      atomic_read(&disk->sync_io);
7926 		/* sync IO will cause sync_io to increase before the disk_stats
7927 		 * as sync_io is counted when a request starts, and
7928 		 * disk_stats is counted when it completes.
7929 		 * So resync activity will cause curr_events to be smaller than
7930 		 * when there was no such activity.
7931 		 * non-sync IO will cause disk_stat to increase without
7932 		 * increasing sync_io so curr_events will (eventually)
7933 		 * be larger than it was before.  Once it becomes
7934 		 * substantially larger, the test below will cause
7935 		 * the array to appear non-idle, and resync will slow
7936 		 * down.
7937 		 * If there is a lot of outstanding resync activity when
7938 		 * we set last_event to curr_events, then all that activity
7939 		 * completing might cause the array to appear non-idle
7940 		 * and resync will be slowed down even though there might
7941 		 * not have been non-resync activity.  This will only
7942 		 * happen once though.  'last_events' will soon reflect
7943 		 * the state where there is little or no outstanding
7944 		 * resync requests, and further resync activity will
7945 		 * always make curr_events less than last_events.
7946 		 *
7947 		 */
7948 		if (init || curr_events - rdev->last_events > 64) {
7949 			rdev->last_events = curr_events;
7950 			idle = 0;
7951 		}
7952 	}
7953 	rcu_read_unlock();
7954 	return idle;
7955 }
7956 
7957 void md_done_sync(struct mddev *mddev, int blocks, int ok)
7958 {
7959 	/* another "blocks" (512byte) blocks have been synced */
7960 	atomic_sub(blocks, &mddev->recovery_active);
7961 	wake_up(&mddev->recovery_wait);
7962 	if (!ok) {
7963 		set_bit(MD_RECOVERY_INTR, &mddev->recovery);
7964 		set_bit(MD_RECOVERY_ERROR, &mddev->recovery);
7965 		md_wakeup_thread(mddev->thread);
7966 		// stop recovery, signal do_sync ....
7967 	}
7968 }
7969 EXPORT_SYMBOL(md_done_sync);
7970 
7971 /* md_write_start(mddev, bi)
7972  * If we need to update some array metadata (e.g. 'active' flag
7973  * in superblock) before writing, schedule a superblock update
7974  * and wait for it to complete.
7975  * A return value of 'false' means that the write wasn't recorded
7976  * and cannot proceed as the array is being suspend.
7977  */
7978 bool md_write_start(struct mddev *mddev, struct bio *bi)
7979 {
7980 	int did_change = 0;
7981 	if (bio_data_dir(bi) != WRITE)
7982 		return true;
7983 
7984 	BUG_ON(mddev->ro == 1);
7985 	if (mddev->ro == 2) {
7986 		/* need to switch to read/write */
7987 		mddev->ro = 0;
7988 		set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7989 		md_wakeup_thread(mddev->thread);
7990 		md_wakeup_thread(mddev->sync_thread);
7991 		did_change = 1;
7992 	}
7993 	rcu_read_lock();
7994 	percpu_ref_get(&mddev->writes_pending);
7995 	smp_mb(); /* Match smp_mb in set_in_sync() */
7996 	if (mddev->safemode == 1)
7997 		mddev->safemode = 0;
7998 	/* sync_checkers is always 0 when writes_pending is in per-cpu mode */
7999 	if (mddev->in_sync || !mddev->sync_checkers) {
8000 		spin_lock(&mddev->lock);
8001 		if (mddev->in_sync) {
8002 			mddev->in_sync = 0;
8003 			set_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags);
8004 			set_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags);
8005 			md_wakeup_thread(mddev->thread);
8006 			did_change = 1;
8007 		}
8008 		spin_unlock(&mddev->lock);
8009 	}
8010 	rcu_read_unlock();
8011 	if (did_change)
8012 		sysfs_notify_dirent_safe(mddev->sysfs_state);
8013 	wait_event(mddev->sb_wait,
8014 		   !test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags) && !mddev->suspended);
8015 	if (test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags)) {
8016 		percpu_ref_put(&mddev->writes_pending);
8017 		return false;
8018 	}
8019 	return true;
8020 }
8021 EXPORT_SYMBOL(md_write_start);
8022 
8023 /* md_write_inc can only be called when md_write_start() has
8024  * already been called at least once of the current request.
8025  * It increments the counter and is useful when a single request
8026  * is split into several parts.  Each part causes an increment and
8027  * so needs a matching md_write_end().
8028  * Unlike md_write_start(), it is safe to call md_write_inc() inside
8029  * a spinlocked region.
8030  */
8031 void md_write_inc(struct mddev *mddev, struct bio *bi)
8032 {
8033 	if (bio_data_dir(bi) != WRITE)
8034 		return;
8035 	WARN_ON_ONCE(mddev->in_sync || mddev->ro);
8036 	percpu_ref_get(&mddev->writes_pending);
8037 }
8038 EXPORT_SYMBOL(md_write_inc);
8039 
8040 void md_write_end(struct mddev *mddev)
8041 {
8042 	percpu_ref_put(&mddev->writes_pending);
8043 
8044 	if (mddev->safemode == 2)
8045 		md_wakeup_thread(mddev->thread);
8046 	else if (mddev->safemode_delay)
8047 		/* The roundup() ensures this only performs locking once
8048 		 * every ->safemode_delay jiffies
8049 		 */
8050 		mod_timer(&mddev->safemode_timer,
8051 			  roundup(jiffies, mddev->safemode_delay) +
8052 			  mddev->safemode_delay);
8053 }
8054 
8055 EXPORT_SYMBOL(md_write_end);
8056 
8057 /* md_allow_write(mddev)
8058  * Calling this ensures that the array is marked 'active' so that writes
8059  * may proceed without blocking.  It is important to call this before
8060  * attempting a GFP_KERNEL allocation while holding the mddev lock.
8061  * Must be called with mddev_lock held.
8062  */
8063 void md_allow_write(struct mddev *mddev)
8064 {
8065 	if (!mddev->pers)
8066 		return;
8067 	if (mddev->ro)
8068 		return;
8069 	if (!mddev->pers->sync_request)
8070 		return;
8071 
8072 	spin_lock(&mddev->lock);
8073 	if (mddev->in_sync) {
8074 		mddev->in_sync = 0;
8075 		set_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags);
8076 		set_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags);
8077 		if (mddev->safemode_delay &&
8078 		    mddev->safemode == 0)
8079 			mddev->safemode = 1;
8080 		spin_unlock(&mddev->lock);
8081 		md_update_sb(mddev, 0);
8082 		sysfs_notify_dirent_safe(mddev->sysfs_state);
8083 		/* wait for the dirty state to be recorded in the metadata */
8084 		wait_event(mddev->sb_wait,
8085 			   !test_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags) &&
8086 			   !test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags));
8087 	} else
8088 		spin_unlock(&mddev->lock);
8089 }
8090 EXPORT_SYMBOL_GPL(md_allow_write);
8091 
8092 #define SYNC_MARKS	10
8093 #define	SYNC_MARK_STEP	(3*HZ)
8094 #define UPDATE_FREQUENCY (5*60*HZ)
8095 void md_do_sync(struct md_thread *thread)
8096 {
8097 	struct mddev *mddev = thread->mddev;
8098 	struct mddev *mddev2;
8099 	unsigned int currspeed = 0,
8100 		 window;
8101 	sector_t max_sectors,j, io_sectors, recovery_done;
8102 	unsigned long mark[SYNC_MARKS];
8103 	unsigned long update_time;
8104 	sector_t mark_cnt[SYNC_MARKS];
8105 	int last_mark,m;
8106 	struct list_head *tmp;
8107 	sector_t last_check;
8108 	int skipped = 0;
8109 	struct md_rdev *rdev;
8110 	char *desc, *action = NULL;
8111 	struct blk_plug plug;
8112 	int ret;
8113 
8114 	/* just incase thread restarts... */
8115 	if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
8116 		return;
8117 	if (mddev->ro) {/* never try to sync a read-only array */
8118 		set_bit(MD_RECOVERY_INTR, &mddev->recovery);
8119 		return;
8120 	}
8121 
8122 	if (mddev_is_clustered(mddev)) {
8123 		ret = md_cluster_ops->resync_start(mddev);
8124 		if (ret)
8125 			goto skip;
8126 
8127 		set_bit(MD_CLUSTER_RESYNC_LOCKED, &mddev->flags);
8128 		if (!(test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ||
8129 			test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) ||
8130 			test_bit(MD_RECOVERY_RECOVER, &mddev->recovery))
8131 		     && ((unsigned long long)mddev->curr_resync_completed
8132 			 < (unsigned long long)mddev->resync_max_sectors))
8133 			goto skip;
8134 	}
8135 
8136 	if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
8137 		if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)) {
8138 			desc = "data-check";
8139 			action = "check";
8140 		} else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
8141 			desc = "requested-resync";
8142 			action = "repair";
8143 		} else
8144 			desc = "resync";
8145 	} else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
8146 		desc = "reshape";
8147 	else
8148 		desc = "recovery";
8149 
8150 	mddev->last_sync_action = action ?: desc;
8151 
8152 	/* we overload curr_resync somewhat here.
8153 	 * 0 == not engaged in resync at all
8154 	 * 2 == checking that there is no conflict with another sync
8155 	 * 1 == like 2, but have yielded to allow conflicting resync to
8156 	 *		commense
8157 	 * other == active in resync - this many blocks
8158 	 *
8159 	 * Before starting a resync we must have set curr_resync to
8160 	 * 2, and then checked that every "conflicting" array has curr_resync
8161 	 * less than ours.  When we find one that is the same or higher
8162 	 * we wait on resync_wait.  To avoid deadlock, we reduce curr_resync
8163 	 * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
8164 	 * This will mean we have to start checking from the beginning again.
8165 	 *
8166 	 */
8167 
8168 	do {
8169 		int mddev2_minor = -1;
8170 		mddev->curr_resync = 2;
8171 
8172 	try_again:
8173 		if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
8174 			goto skip;
8175 		for_each_mddev(mddev2, tmp) {
8176 			if (mddev2 == mddev)
8177 				continue;
8178 			if (!mddev->parallel_resync
8179 			&&  mddev2->curr_resync
8180 			&&  match_mddev_units(mddev, mddev2)) {
8181 				DEFINE_WAIT(wq);
8182 				if (mddev < mddev2 && mddev->curr_resync == 2) {
8183 					/* arbitrarily yield */
8184 					mddev->curr_resync = 1;
8185 					wake_up(&resync_wait);
8186 				}
8187 				if (mddev > mddev2 && mddev->curr_resync == 1)
8188 					/* no need to wait here, we can wait the next
8189 					 * time 'round when curr_resync == 2
8190 					 */
8191 					continue;
8192 				/* We need to wait 'interruptible' so as not to
8193 				 * contribute to the load average, and not to
8194 				 * be caught by 'softlockup'
8195 				 */
8196 				prepare_to_wait(&resync_wait, &wq, TASK_INTERRUPTIBLE);
8197 				if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
8198 				    mddev2->curr_resync >= mddev->curr_resync) {
8199 					if (mddev2_minor != mddev2->md_minor) {
8200 						mddev2_minor = mddev2->md_minor;
8201 						pr_info("md: delaying %s of %s until %s has finished (they share one or more physical units)\n",
8202 							desc, mdname(mddev),
8203 							mdname(mddev2));
8204 					}
8205 					mddev_put(mddev2);
8206 					if (signal_pending(current))
8207 						flush_signals(current);
8208 					schedule();
8209 					finish_wait(&resync_wait, &wq);
8210 					goto try_again;
8211 				}
8212 				finish_wait(&resync_wait, &wq);
8213 			}
8214 		}
8215 	} while (mddev->curr_resync < 2);
8216 
8217 	j = 0;
8218 	if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
8219 		/* resync follows the size requested by the personality,
8220 		 * which defaults to physical size, but can be virtual size
8221 		 */
8222 		max_sectors = mddev->resync_max_sectors;
8223 		atomic64_set(&mddev->resync_mismatches, 0);
8224 		/* we don't use the checkpoint if there's a bitmap */
8225 		if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
8226 			j = mddev->resync_min;
8227 		else if (!mddev->bitmap)
8228 			j = mddev->recovery_cp;
8229 
8230 	} else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
8231 		max_sectors = mddev->resync_max_sectors;
8232 	else {
8233 		/* recovery follows the physical size of devices */
8234 		max_sectors = mddev->dev_sectors;
8235 		j = MaxSector;
8236 		rcu_read_lock();
8237 		rdev_for_each_rcu(rdev, mddev)
8238 			if (rdev->raid_disk >= 0 &&
8239 			    !test_bit(Journal, &rdev->flags) &&
8240 			    !test_bit(Faulty, &rdev->flags) &&
8241 			    !test_bit(In_sync, &rdev->flags) &&
8242 			    rdev->recovery_offset < j)
8243 				j = rdev->recovery_offset;
8244 		rcu_read_unlock();
8245 
8246 		/* If there is a bitmap, we need to make sure all
8247 		 * writes that started before we added a spare
8248 		 * complete before we start doing a recovery.
8249 		 * Otherwise the write might complete and (via
8250 		 * bitmap_endwrite) set a bit in the bitmap after the
8251 		 * recovery has checked that bit and skipped that
8252 		 * region.
8253 		 */
8254 		if (mddev->bitmap) {
8255 			mddev->pers->quiesce(mddev, 1);
8256 			mddev->pers->quiesce(mddev, 0);
8257 		}
8258 	}
8259 
8260 	pr_info("md: %s of RAID array %s\n", desc, mdname(mddev));
8261 	pr_debug("md: minimum _guaranteed_  speed: %d KB/sec/disk.\n", speed_min(mddev));
8262 	pr_debug("md: using maximum available idle IO bandwidth (but not more than %d KB/sec) for %s.\n",
8263 		 speed_max(mddev), desc);
8264 
8265 	is_mddev_idle(mddev, 1); /* this initializes IO event counters */
8266 
8267 	io_sectors = 0;
8268 	for (m = 0; m < SYNC_MARKS; m++) {
8269 		mark[m] = jiffies;
8270 		mark_cnt[m] = io_sectors;
8271 	}
8272 	last_mark = 0;
8273 	mddev->resync_mark = mark[last_mark];
8274 	mddev->resync_mark_cnt = mark_cnt[last_mark];
8275 
8276 	/*
8277 	 * Tune reconstruction:
8278 	 */
8279 	window = 32*(PAGE_SIZE/512);
8280 	pr_debug("md: using %dk window, over a total of %lluk.\n",
8281 		 window/2, (unsigned long long)max_sectors/2);
8282 
8283 	atomic_set(&mddev->recovery_active, 0);
8284 	last_check = 0;
8285 
8286 	if (j>2) {
8287 		pr_debug("md: resuming %s of %s from checkpoint.\n",
8288 			 desc, mdname(mddev));
8289 		mddev->curr_resync = j;
8290 	} else
8291 		mddev->curr_resync = 3; /* no longer delayed */
8292 	mddev->curr_resync_completed = j;
8293 	sysfs_notify(&mddev->kobj, NULL, "sync_completed");
8294 	md_new_event(mddev);
8295 	update_time = jiffies;
8296 
8297 	blk_start_plug(&plug);
8298 	while (j < max_sectors) {
8299 		sector_t sectors;
8300 
8301 		skipped = 0;
8302 
8303 		if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
8304 		    ((mddev->curr_resync > mddev->curr_resync_completed &&
8305 		      (mddev->curr_resync - mddev->curr_resync_completed)
8306 		      > (max_sectors >> 4)) ||
8307 		     time_after_eq(jiffies, update_time + UPDATE_FREQUENCY) ||
8308 		     (j - mddev->curr_resync_completed)*2
8309 		     >= mddev->resync_max - mddev->curr_resync_completed ||
8310 		     mddev->curr_resync_completed > mddev->resync_max
8311 			    )) {
8312 			/* time to update curr_resync_completed */
8313 			wait_event(mddev->recovery_wait,
8314 				   atomic_read(&mddev->recovery_active) == 0);
8315 			mddev->curr_resync_completed = j;
8316 			if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) &&
8317 			    j > mddev->recovery_cp)
8318 				mddev->recovery_cp = j;
8319 			update_time = jiffies;
8320 			set_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags);
8321 			sysfs_notify(&mddev->kobj, NULL, "sync_completed");
8322 		}
8323 
8324 		while (j >= mddev->resync_max &&
8325 		       !test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
8326 			/* As this condition is controlled by user-space,
8327 			 * we can block indefinitely, so use '_interruptible'
8328 			 * to avoid triggering warnings.
8329 			 */
8330 			flush_signals(current); /* just in case */
8331 			wait_event_interruptible(mddev->recovery_wait,
8332 						 mddev->resync_max > j
8333 						 || test_bit(MD_RECOVERY_INTR,
8334 							     &mddev->recovery));
8335 		}
8336 
8337 		if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
8338 			break;
8339 
8340 		sectors = mddev->pers->sync_request(mddev, j, &skipped);
8341 		if (sectors == 0) {
8342 			set_bit(MD_RECOVERY_INTR, &mddev->recovery);
8343 			break;
8344 		}
8345 
8346 		if (!skipped) { /* actual IO requested */
8347 			io_sectors += sectors;
8348 			atomic_add(sectors, &mddev->recovery_active);
8349 		}
8350 
8351 		if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
8352 			break;
8353 
8354 		j += sectors;
8355 		if (j > max_sectors)
8356 			/* when skipping, extra large numbers can be returned. */
8357 			j = max_sectors;
8358 		if (j > 2)
8359 			mddev->curr_resync = j;
8360 		mddev->curr_mark_cnt = io_sectors;
8361 		if (last_check == 0)
8362 			/* this is the earliest that rebuild will be
8363 			 * visible in /proc/mdstat
8364 			 */
8365 			md_new_event(mddev);
8366 
8367 		if (last_check + window > io_sectors || j == max_sectors)
8368 			continue;
8369 
8370 		last_check = io_sectors;
8371 	repeat:
8372 		if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) {
8373 			/* step marks */
8374 			int next = (last_mark+1) % SYNC_MARKS;
8375 
8376 			mddev->resync_mark = mark[next];
8377 			mddev->resync_mark_cnt = mark_cnt[next];
8378 			mark[next] = jiffies;
8379 			mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active);
8380 			last_mark = next;
8381 		}
8382 
8383 		if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
8384 			break;
8385 
8386 		/*
8387 		 * this loop exits only if either when we are slower than
8388 		 * the 'hard' speed limit, or the system was IO-idle for
8389 		 * a jiffy.
8390 		 * the system might be non-idle CPU-wise, but we only care
8391 		 * about not overloading the IO subsystem. (things like an
8392 		 * e2fsck being done on the RAID array should execute fast)
8393 		 */
8394 		cond_resched();
8395 
8396 		recovery_done = io_sectors - atomic_read(&mddev->recovery_active);
8397 		currspeed = ((unsigned long)(recovery_done - mddev->resync_mark_cnt))/2
8398 			/((jiffies-mddev->resync_mark)/HZ +1) +1;
8399 
8400 		if (currspeed > speed_min(mddev)) {
8401 			if (currspeed > speed_max(mddev)) {
8402 				msleep(500);
8403 				goto repeat;
8404 			}
8405 			if (!is_mddev_idle(mddev, 0)) {
8406 				/*
8407 				 * Give other IO more of a chance.
8408 				 * The faster the devices, the less we wait.
8409 				 */
8410 				wait_event(mddev->recovery_wait,
8411 					   !atomic_read(&mddev->recovery_active));
8412 			}
8413 		}
8414 	}
8415 	pr_info("md: %s: %s %s.\n",mdname(mddev), desc,
8416 		test_bit(MD_RECOVERY_INTR, &mddev->recovery)
8417 		? "interrupted" : "done");
8418 	/*
8419 	 * this also signals 'finished resyncing' to md_stop
8420 	 */
8421 	blk_finish_plug(&plug);
8422 	wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
8423 
8424 	if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
8425 	    !test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
8426 	    mddev->curr_resync > 3) {
8427 		mddev->curr_resync_completed = mddev->curr_resync;
8428 		sysfs_notify(&mddev->kobj, NULL, "sync_completed");
8429 	}
8430 	mddev->pers->sync_request(mddev, max_sectors, &skipped);
8431 
8432 	if (!test_bit(MD_RECOVERY_CHECK, &mddev->recovery) &&
8433 	    mddev->curr_resync > 3) {
8434 		if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
8435 			if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
8436 				if (mddev->curr_resync >= mddev->recovery_cp) {
8437 					pr_debug("md: checkpointing %s of %s.\n",
8438 						 desc, mdname(mddev));
8439 					if (test_bit(MD_RECOVERY_ERROR,
8440 						&mddev->recovery))
8441 						mddev->recovery_cp =
8442 							mddev->curr_resync_completed;
8443 					else
8444 						mddev->recovery_cp =
8445 							mddev->curr_resync;
8446 				}
8447 			} else
8448 				mddev->recovery_cp = MaxSector;
8449 		} else {
8450 			if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery))
8451 				mddev->curr_resync = MaxSector;
8452 			rcu_read_lock();
8453 			rdev_for_each_rcu(rdev, mddev)
8454 				if (rdev->raid_disk >= 0 &&
8455 				    mddev->delta_disks >= 0 &&
8456 				    !test_bit(Journal, &rdev->flags) &&
8457 				    !test_bit(Faulty, &rdev->flags) &&
8458 				    !test_bit(In_sync, &rdev->flags) &&
8459 				    rdev->recovery_offset < mddev->curr_resync)
8460 					rdev->recovery_offset = mddev->curr_resync;
8461 			rcu_read_unlock();
8462 		}
8463 	}
8464  skip:
8465 	/* set CHANGE_PENDING here since maybe another update is needed,
8466 	 * so other nodes are informed. It should be harmless for normal
8467 	 * raid */
8468 	set_mask_bits(&mddev->sb_flags, 0,
8469 		      BIT(MD_SB_CHANGE_PENDING) | BIT(MD_SB_CHANGE_DEVS));
8470 
8471 	spin_lock(&mddev->lock);
8472 	if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
8473 		/* We completed so min/max setting can be forgotten if used. */
8474 		if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
8475 			mddev->resync_min = 0;
8476 		mddev->resync_max = MaxSector;
8477 	} else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
8478 		mddev->resync_min = mddev->curr_resync_completed;
8479 	set_bit(MD_RECOVERY_DONE, &mddev->recovery);
8480 	mddev->curr_resync = 0;
8481 	spin_unlock(&mddev->lock);
8482 
8483 	wake_up(&resync_wait);
8484 	md_wakeup_thread(mddev->thread);
8485 	return;
8486 }
8487 EXPORT_SYMBOL_GPL(md_do_sync);
8488 
8489 static int remove_and_add_spares(struct mddev *mddev,
8490 				 struct md_rdev *this)
8491 {
8492 	struct md_rdev *rdev;
8493 	int spares = 0;
8494 	int removed = 0;
8495 	bool remove_some = false;
8496 
8497 	rdev_for_each(rdev, mddev) {
8498 		if ((this == NULL || rdev == this) &&
8499 		    rdev->raid_disk >= 0 &&
8500 		    !test_bit(Blocked, &rdev->flags) &&
8501 		    test_bit(Faulty, &rdev->flags) &&
8502 		    atomic_read(&rdev->nr_pending)==0) {
8503 			/* Faulty non-Blocked devices with nr_pending == 0
8504 			 * never get nr_pending incremented,
8505 			 * never get Faulty cleared, and never get Blocked set.
8506 			 * So we can synchronize_rcu now rather than once per device
8507 			 */
8508 			remove_some = true;
8509 			set_bit(RemoveSynchronized, &rdev->flags);
8510 		}
8511 	}
8512 
8513 	if (remove_some)
8514 		synchronize_rcu();
8515 	rdev_for_each(rdev, mddev) {
8516 		if ((this == NULL || rdev == this) &&
8517 		    rdev->raid_disk >= 0 &&
8518 		    !test_bit(Blocked, &rdev->flags) &&
8519 		    ((test_bit(RemoveSynchronized, &rdev->flags) ||
8520 		     (!test_bit(In_sync, &rdev->flags) &&
8521 		      !test_bit(Journal, &rdev->flags))) &&
8522 		    atomic_read(&rdev->nr_pending)==0)) {
8523 			if (mddev->pers->hot_remove_disk(
8524 				    mddev, rdev) == 0) {
8525 				sysfs_unlink_rdev(mddev, rdev);
8526 				rdev->raid_disk = -1;
8527 				removed++;
8528 			}
8529 		}
8530 		if (remove_some && test_bit(RemoveSynchronized, &rdev->flags))
8531 			clear_bit(RemoveSynchronized, &rdev->flags);
8532 	}
8533 
8534 	if (removed && mddev->kobj.sd)
8535 		sysfs_notify(&mddev->kobj, NULL, "degraded");
8536 
8537 	if (this && removed)
8538 		goto no_add;
8539 
8540 	rdev_for_each(rdev, mddev) {
8541 		if (this && this != rdev)
8542 			continue;
8543 		if (test_bit(Candidate, &rdev->flags))
8544 			continue;
8545 		if (rdev->raid_disk >= 0 &&
8546 		    !test_bit(In_sync, &rdev->flags) &&
8547 		    !test_bit(Journal, &rdev->flags) &&
8548 		    !test_bit(Faulty, &rdev->flags))
8549 			spares++;
8550 		if (rdev->raid_disk >= 0)
8551 			continue;
8552 		if (test_bit(Faulty, &rdev->flags))
8553 			continue;
8554 		if (!test_bit(Journal, &rdev->flags)) {
8555 			if (mddev->ro &&
8556 			    ! (rdev->saved_raid_disk >= 0 &&
8557 			       !test_bit(Bitmap_sync, &rdev->flags)))
8558 				continue;
8559 
8560 			rdev->recovery_offset = 0;
8561 		}
8562 		if (mddev->pers->
8563 		    hot_add_disk(mddev, rdev) == 0) {
8564 			if (sysfs_link_rdev(mddev, rdev))
8565 				/* failure here is OK */;
8566 			if (!test_bit(Journal, &rdev->flags))
8567 				spares++;
8568 			md_new_event(mddev);
8569 			set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
8570 		}
8571 	}
8572 no_add:
8573 	if (removed)
8574 		set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
8575 	return spares;
8576 }
8577 
8578 static void md_start_sync(struct work_struct *ws)
8579 {
8580 	struct mddev *mddev = container_of(ws, struct mddev, del_work);
8581 
8582 	mddev->sync_thread = md_register_thread(md_do_sync,
8583 						mddev,
8584 						"resync");
8585 	if (!mddev->sync_thread) {
8586 		pr_warn("%s: could not start resync thread...\n",
8587 			mdname(mddev));
8588 		/* leave the spares where they are, it shouldn't hurt */
8589 		clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
8590 		clear_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
8591 		clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
8592 		clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
8593 		clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
8594 		wake_up(&resync_wait);
8595 		if (test_and_clear_bit(MD_RECOVERY_RECOVER,
8596 				       &mddev->recovery))
8597 			if (mddev->sysfs_action)
8598 				sysfs_notify_dirent_safe(mddev->sysfs_action);
8599 	} else
8600 		md_wakeup_thread(mddev->sync_thread);
8601 	sysfs_notify_dirent_safe(mddev->sysfs_action);
8602 	md_new_event(mddev);
8603 }
8604 
8605 /*
8606  * This routine is regularly called by all per-raid-array threads to
8607  * deal with generic issues like resync and super-block update.
8608  * Raid personalities that don't have a thread (linear/raid0) do not
8609  * need this as they never do any recovery or update the superblock.
8610  *
8611  * It does not do any resync itself, but rather "forks" off other threads
8612  * to do that as needed.
8613  * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
8614  * "->recovery" and create a thread at ->sync_thread.
8615  * When the thread finishes it sets MD_RECOVERY_DONE
8616  * and wakeups up this thread which will reap the thread and finish up.
8617  * This thread also removes any faulty devices (with nr_pending == 0).
8618  *
8619  * The overall approach is:
8620  *  1/ if the superblock needs updating, update it.
8621  *  2/ If a recovery thread is running, don't do anything else.
8622  *  3/ If recovery has finished, clean up, possibly marking spares active.
8623  *  4/ If there are any faulty devices, remove them.
8624  *  5/ If array is degraded, try to add spares devices
8625  *  6/ If array has spares or is not in-sync, start a resync thread.
8626  */
8627 void md_check_recovery(struct mddev *mddev)
8628 {
8629 	if (mddev->suspended)
8630 		return;
8631 
8632 	if (mddev->bitmap)
8633 		bitmap_daemon_work(mddev);
8634 
8635 	if (signal_pending(current)) {
8636 		if (mddev->pers->sync_request && !mddev->external) {
8637 			pr_debug("md: %s in immediate safe mode\n",
8638 				 mdname(mddev));
8639 			mddev->safemode = 2;
8640 		}
8641 		flush_signals(current);
8642 	}
8643 
8644 	if (mddev->ro && !test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
8645 		return;
8646 	if ( ! (
8647 		(mddev->sb_flags & ~ (1<<MD_SB_CHANGE_PENDING)) ||
8648 		test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
8649 		test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
8650 		(mddev->external == 0 && mddev->safemode == 1) ||
8651 		(mddev->safemode == 2
8652 		 && !mddev->in_sync && mddev->recovery_cp == MaxSector)
8653 		))
8654 		return;
8655 
8656 	if (mddev_trylock(mddev)) {
8657 		int spares = 0;
8658 
8659 		if (mddev->ro) {
8660 			struct md_rdev *rdev;
8661 			if (!mddev->external && mddev->in_sync)
8662 				/* 'Blocked' flag not needed as failed devices
8663 				 * will be recorded if array switched to read/write.
8664 				 * Leaving it set will prevent the device
8665 				 * from being removed.
8666 				 */
8667 				rdev_for_each(rdev, mddev)
8668 					clear_bit(Blocked, &rdev->flags);
8669 			/* On a read-only array we can:
8670 			 * - remove failed devices
8671 			 * - add already-in_sync devices if the array itself
8672 			 *   is in-sync.
8673 			 * As we only add devices that are already in-sync,
8674 			 * we can activate the spares immediately.
8675 			 */
8676 			remove_and_add_spares(mddev, NULL);
8677 			/* There is no thread, but we need to call
8678 			 * ->spare_active and clear saved_raid_disk
8679 			 */
8680 			set_bit(MD_RECOVERY_INTR, &mddev->recovery);
8681 			md_reap_sync_thread(mddev);
8682 			clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
8683 			clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
8684 			clear_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags);
8685 			goto unlock;
8686 		}
8687 
8688 		if (mddev_is_clustered(mddev)) {
8689 			struct md_rdev *rdev;
8690 			/* kick the device if another node issued a
8691 			 * remove disk.
8692 			 */
8693 			rdev_for_each(rdev, mddev) {
8694 				if (test_and_clear_bit(ClusterRemove, &rdev->flags) &&
8695 						rdev->raid_disk < 0)
8696 					md_kick_rdev_from_array(rdev);
8697 			}
8698 		}
8699 
8700 		if (!mddev->external && !mddev->in_sync) {
8701 			spin_lock(&mddev->lock);
8702 			set_in_sync(mddev);
8703 			spin_unlock(&mddev->lock);
8704 		}
8705 
8706 		if (mddev->sb_flags)
8707 			md_update_sb(mddev, 0);
8708 
8709 		if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
8710 		    !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
8711 			/* resync/recovery still happening */
8712 			clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
8713 			goto unlock;
8714 		}
8715 		if (mddev->sync_thread) {
8716 			md_reap_sync_thread(mddev);
8717 			goto unlock;
8718 		}
8719 		/* Set RUNNING before clearing NEEDED to avoid
8720 		 * any transients in the value of "sync_action".
8721 		 */
8722 		mddev->curr_resync_completed = 0;
8723 		spin_lock(&mddev->lock);
8724 		set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
8725 		spin_unlock(&mddev->lock);
8726 		/* Clear some bits that don't mean anything, but
8727 		 * might be left set
8728 		 */
8729 		clear_bit(MD_RECOVERY_INTR, &mddev->recovery);
8730 		clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
8731 
8732 		if (!test_and_clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
8733 		    test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
8734 			goto not_running;
8735 		/* no recovery is running.
8736 		 * remove any failed drives, then
8737 		 * add spares if possible.
8738 		 * Spares are also removed and re-added, to allow
8739 		 * the personality to fail the re-add.
8740 		 */
8741 
8742 		if (mddev->reshape_position != MaxSector) {
8743 			if (mddev->pers->check_reshape == NULL ||
8744 			    mddev->pers->check_reshape(mddev) != 0)
8745 				/* Cannot proceed */
8746 				goto not_running;
8747 			set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
8748 			clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
8749 		} else if ((spares = remove_and_add_spares(mddev, NULL))) {
8750 			clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
8751 			clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
8752 			clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
8753 			set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
8754 		} else if (mddev->recovery_cp < MaxSector) {
8755 			set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
8756 			clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
8757 		} else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
8758 			/* nothing to be done ... */
8759 			goto not_running;
8760 
8761 		if (mddev->pers->sync_request) {
8762 			if (spares) {
8763 				/* We are adding a device or devices to an array
8764 				 * which has the bitmap stored on all devices.
8765 				 * So make sure all bitmap pages get written
8766 				 */
8767 				bitmap_write_all(mddev->bitmap);
8768 			}
8769 			INIT_WORK(&mddev->del_work, md_start_sync);
8770 			queue_work(md_misc_wq, &mddev->del_work);
8771 			goto unlock;
8772 		}
8773 	not_running:
8774 		if (!mddev->sync_thread) {
8775 			clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
8776 			wake_up(&resync_wait);
8777 			if (test_and_clear_bit(MD_RECOVERY_RECOVER,
8778 					       &mddev->recovery))
8779 				if (mddev->sysfs_action)
8780 					sysfs_notify_dirent_safe(mddev->sysfs_action);
8781 		}
8782 	unlock:
8783 		wake_up(&mddev->sb_wait);
8784 		mddev_unlock(mddev);
8785 	}
8786 }
8787 EXPORT_SYMBOL(md_check_recovery);
8788 
8789 void md_reap_sync_thread(struct mddev *mddev)
8790 {
8791 	struct md_rdev *rdev;
8792 
8793 	/* resync has finished, collect result */
8794 	md_unregister_thread(&mddev->sync_thread);
8795 	if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
8796 	    !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
8797 		/* success...*/
8798 		/* activate any spares */
8799 		if (mddev->pers->spare_active(mddev)) {
8800 			sysfs_notify(&mddev->kobj, NULL,
8801 				     "degraded");
8802 			set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
8803 		}
8804 	}
8805 	if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
8806 	    mddev->pers->finish_reshape)
8807 		mddev->pers->finish_reshape(mddev);
8808 
8809 	/* If array is no-longer degraded, then any saved_raid_disk
8810 	 * information must be scrapped.
8811 	 */
8812 	if (!mddev->degraded)
8813 		rdev_for_each(rdev, mddev)
8814 			rdev->saved_raid_disk = -1;
8815 
8816 	md_update_sb(mddev, 1);
8817 	/* MD_SB_CHANGE_PENDING should be cleared by md_update_sb, so we can
8818 	 * call resync_finish here if MD_CLUSTER_RESYNC_LOCKED is set by
8819 	 * clustered raid */
8820 	if (test_and_clear_bit(MD_CLUSTER_RESYNC_LOCKED, &mddev->flags))
8821 		md_cluster_ops->resync_finish(mddev);
8822 	clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
8823 	clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
8824 	clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
8825 	clear_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
8826 	clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
8827 	clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
8828 	wake_up(&resync_wait);
8829 	/* flag recovery needed just to double check */
8830 	set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
8831 	sysfs_notify_dirent_safe(mddev->sysfs_action);
8832 	md_new_event(mddev);
8833 	if (mddev->event_work.func)
8834 		queue_work(md_misc_wq, &mddev->event_work);
8835 }
8836 EXPORT_SYMBOL(md_reap_sync_thread);
8837 
8838 void md_wait_for_blocked_rdev(struct md_rdev *rdev, struct mddev *mddev)
8839 {
8840 	sysfs_notify_dirent_safe(rdev->sysfs_state);
8841 	wait_event_timeout(rdev->blocked_wait,
8842 			   !test_bit(Blocked, &rdev->flags) &&
8843 			   !test_bit(BlockedBadBlocks, &rdev->flags),
8844 			   msecs_to_jiffies(5000));
8845 	rdev_dec_pending(rdev, mddev);
8846 }
8847 EXPORT_SYMBOL(md_wait_for_blocked_rdev);
8848 
8849 void md_finish_reshape(struct mddev *mddev)
8850 {
8851 	/* called be personality module when reshape completes. */
8852 	struct md_rdev *rdev;
8853 
8854 	rdev_for_each(rdev, mddev) {
8855 		if (rdev->data_offset > rdev->new_data_offset)
8856 			rdev->sectors += rdev->data_offset - rdev->new_data_offset;
8857 		else
8858 			rdev->sectors -= rdev->new_data_offset - rdev->data_offset;
8859 		rdev->data_offset = rdev->new_data_offset;
8860 	}
8861 }
8862 EXPORT_SYMBOL(md_finish_reshape);
8863 
8864 /* Bad block management */
8865 
8866 /* Returns 1 on success, 0 on failure */
8867 int rdev_set_badblocks(struct md_rdev *rdev, sector_t s, int sectors,
8868 		       int is_new)
8869 {
8870 	struct mddev *mddev = rdev->mddev;
8871 	int rv;
8872 	if (is_new)
8873 		s += rdev->new_data_offset;
8874 	else
8875 		s += rdev->data_offset;
8876 	rv = badblocks_set(&rdev->badblocks, s, sectors, 0);
8877 	if (rv == 0) {
8878 		/* Make sure they get written out promptly */
8879 		if (test_bit(ExternalBbl, &rdev->flags))
8880 			sysfs_notify(&rdev->kobj, NULL,
8881 				     "unacknowledged_bad_blocks");
8882 		sysfs_notify_dirent_safe(rdev->sysfs_state);
8883 		set_mask_bits(&mddev->sb_flags, 0,
8884 			      BIT(MD_SB_CHANGE_CLEAN) | BIT(MD_SB_CHANGE_PENDING));
8885 		md_wakeup_thread(rdev->mddev->thread);
8886 		return 1;
8887 	} else
8888 		return 0;
8889 }
8890 EXPORT_SYMBOL_GPL(rdev_set_badblocks);
8891 
8892 int rdev_clear_badblocks(struct md_rdev *rdev, sector_t s, int sectors,
8893 			 int is_new)
8894 {
8895 	int rv;
8896 	if (is_new)
8897 		s += rdev->new_data_offset;
8898 	else
8899 		s += rdev->data_offset;
8900 	rv = badblocks_clear(&rdev->badblocks, s, sectors);
8901 	if ((rv == 0) && test_bit(ExternalBbl, &rdev->flags))
8902 		sysfs_notify(&rdev->kobj, NULL, "bad_blocks");
8903 	return rv;
8904 }
8905 EXPORT_SYMBOL_GPL(rdev_clear_badblocks);
8906 
8907 static int md_notify_reboot(struct notifier_block *this,
8908 			    unsigned long code, void *x)
8909 {
8910 	struct list_head *tmp;
8911 	struct mddev *mddev;
8912 	int need_delay = 0;
8913 
8914 	for_each_mddev(mddev, tmp) {
8915 		if (mddev_trylock(mddev)) {
8916 			if (mddev->pers)
8917 				__md_stop_writes(mddev);
8918 			if (mddev->persistent)
8919 				mddev->safemode = 2;
8920 			mddev_unlock(mddev);
8921 		}
8922 		need_delay = 1;
8923 	}
8924 	/*
8925 	 * certain more exotic SCSI devices are known to be
8926 	 * volatile wrt too early system reboots. While the
8927 	 * right place to handle this issue is the given
8928 	 * driver, we do want to have a safe RAID driver ...
8929 	 */
8930 	if (need_delay)
8931 		mdelay(1000*1);
8932 
8933 	return NOTIFY_DONE;
8934 }
8935 
8936 static struct notifier_block md_notifier = {
8937 	.notifier_call	= md_notify_reboot,
8938 	.next		= NULL,
8939 	.priority	= INT_MAX, /* before any real devices */
8940 };
8941 
8942 static void md_geninit(void)
8943 {
8944 	pr_debug("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
8945 
8946 	proc_create("mdstat", S_IRUGO, NULL, &md_seq_fops);
8947 }
8948 
8949 static int __init md_init(void)
8950 {
8951 	int ret = -ENOMEM;
8952 
8953 	md_wq = alloc_workqueue("md", WQ_MEM_RECLAIM, 0);
8954 	if (!md_wq)
8955 		goto err_wq;
8956 
8957 	md_misc_wq = alloc_workqueue("md_misc", 0, 0);
8958 	if (!md_misc_wq)
8959 		goto err_misc_wq;
8960 
8961 	if ((ret = register_blkdev(MD_MAJOR, "md")) < 0)
8962 		goto err_md;
8963 
8964 	if ((ret = register_blkdev(0, "mdp")) < 0)
8965 		goto err_mdp;
8966 	mdp_major = ret;
8967 
8968 	blk_register_region(MKDEV(MD_MAJOR, 0), 512, THIS_MODULE,
8969 			    md_probe, NULL, NULL);
8970 	blk_register_region(MKDEV(mdp_major, 0), 1UL<<MINORBITS, THIS_MODULE,
8971 			    md_probe, NULL, NULL);
8972 
8973 	register_reboot_notifier(&md_notifier);
8974 	raid_table_header = register_sysctl_table(raid_root_table);
8975 
8976 	md_geninit();
8977 	return 0;
8978 
8979 err_mdp:
8980 	unregister_blkdev(MD_MAJOR, "md");
8981 err_md:
8982 	destroy_workqueue(md_misc_wq);
8983 err_misc_wq:
8984 	destroy_workqueue(md_wq);
8985 err_wq:
8986 	return ret;
8987 }
8988 
8989 static void check_sb_changes(struct mddev *mddev, struct md_rdev *rdev)
8990 {
8991 	struct mdp_superblock_1 *sb = page_address(rdev->sb_page);
8992 	struct md_rdev *rdev2;
8993 	int role, ret;
8994 	char b[BDEVNAME_SIZE];
8995 
8996 	/*
8997 	 * If size is changed in another node then we need to
8998 	 * do resize as well.
8999 	 */
9000 	if (mddev->dev_sectors != le64_to_cpu(sb->size)) {
9001 		ret = mddev->pers->resize(mddev, le64_to_cpu(sb->size));
9002 		if (ret)
9003 			pr_info("md-cluster: resize failed\n");
9004 		else
9005 			bitmap_update_sb(mddev->bitmap);
9006 	}
9007 
9008 	/* Check for change of roles in the active devices */
9009 	rdev_for_each(rdev2, mddev) {
9010 		if (test_bit(Faulty, &rdev2->flags))
9011 			continue;
9012 
9013 		/* Check if the roles changed */
9014 		role = le16_to_cpu(sb->dev_roles[rdev2->desc_nr]);
9015 
9016 		if (test_bit(Candidate, &rdev2->flags)) {
9017 			if (role == 0xfffe) {
9018 				pr_info("md: Removing Candidate device %s because add failed\n", bdevname(rdev2->bdev,b));
9019 				md_kick_rdev_from_array(rdev2);
9020 				continue;
9021 			}
9022 			else
9023 				clear_bit(Candidate, &rdev2->flags);
9024 		}
9025 
9026 		if (role != rdev2->raid_disk) {
9027 			/* got activated */
9028 			if (rdev2->raid_disk == -1 && role != 0xffff) {
9029 				rdev2->saved_raid_disk = role;
9030 				ret = remove_and_add_spares(mddev, rdev2);
9031 				pr_info("Activated spare: %s\n",
9032 					bdevname(rdev2->bdev,b));
9033 				/* wakeup mddev->thread here, so array could
9034 				 * perform resync with the new activated disk */
9035 				set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
9036 				md_wakeup_thread(mddev->thread);
9037 
9038 			}
9039 			/* device faulty
9040 			 * We just want to do the minimum to mark the disk
9041 			 * as faulty. The recovery is performed by the
9042 			 * one who initiated the error.
9043 			 */
9044 			if ((role == 0xfffe) || (role == 0xfffd)) {
9045 				md_error(mddev, rdev2);
9046 				clear_bit(Blocked, &rdev2->flags);
9047 			}
9048 		}
9049 	}
9050 
9051 	if (mddev->raid_disks != le32_to_cpu(sb->raid_disks))
9052 		update_raid_disks(mddev, le32_to_cpu(sb->raid_disks));
9053 
9054 	/* Finally set the event to be up to date */
9055 	mddev->events = le64_to_cpu(sb->events);
9056 }
9057 
9058 static int read_rdev(struct mddev *mddev, struct md_rdev *rdev)
9059 {
9060 	int err;
9061 	struct page *swapout = rdev->sb_page;
9062 	struct mdp_superblock_1 *sb;
9063 
9064 	/* Store the sb page of the rdev in the swapout temporary
9065 	 * variable in case we err in the future
9066 	 */
9067 	rdev->sb_page = NULL;
9068 	err = alloc_disk_sb(rdev);
9069 	if (err == 0) {
9070 		ClearPageUptodate(rdev->sb_page);
9071 		rdev->sb_loaded = 0;
9072 		err = super_types[mddev->major_version].
9073 			load_super(rdev, NULL, mddev->minor_version);
9074 	}
9075 	if (err < 0) {
9076 		pr_warn("%s: %d Could not reload rdev(%d) err: %d. Restoring old values\n",
9077 				__func__, __LINE__, rdev->desc_nr, err);
9078 		if (rdev->sb_page)
9079 			put_page(rdev->sb_page);
9080 		rdev->sb_page = swapout;
9081 		rdev->sb_loaded = 1;
9082 		return err;
9083 	}
9084 
9085 	sb = page_address(rdev->sb_page);
9086 	/* Read the offset unconditionally, even if MD_FEATURE_RECOVERY_OFFSET
9087 	 * is not set
9088 	 */
9089 
9090 	if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RECOVERY_OFFSET))
9091 		rdev->recovery_offset = le64_to_cpu(sb->recovery_offset);
9092 
9093 	/* The other node finished recovery, call spare_active to set
9094 	 * device In_sync and mddev->degraded
9095 	 */
9096 	if (rdev->recovery_offset == MaxSector &&
9097 	    !test_bit(In_sync, &rdev->flags) &&
9098 	    mddev->pers->spare_active(mddev))
9099 		sysfs_notify(&mddev->kobj, NULL, "degraded");
9100 
9101 	put_page(swapout);
9102 	return 0;
9103 }
9104 
9105 void md_reload_sb(struct mddev *mddev, int nr)
9106 {
9107 	struct md_rdev *rdev;
9108 	int err;
9109 
9110 	/* Find the rdev */
9111 	rdev_for_each_rcu(rdev, mddev) {
9112 		if (rdev->desc_nr == nr)
9113 			break;
9114 	}
9115 
9116 	if (!rdev || rdev->desc_nr != nr) {
9117 		pr_warn("%s: %d Could not find rdev with nr %d\n", __func__, __LINE__, nr);
9118 		return;
9119 	}
9120 
9121 	err = read_rdev(mddev, rdev);
9122 	if (err < 0)
9123 		return;
9124 
9125 	check_sb_changes(mddev, rdev);
9126 
9127 	/* Read all rdev's to update recovery_offset */
9128 	rdev_for_each_rcu(rdev, mddev)
9129 		read_rdev(mddev, rdev);
9130 }
9131 EXPORT_SYMBOL(md_reload_sb);
9132 
9133 #ifndef MODULE
9134 
9135 /*
9136  * Searches all registered partitions for autorun RAID arrays
9137  * at boot time.
9138  */
9139 
9140 static DEFINE_MUTEX(detected_devices_mutex);
9141 static LIST_HEAD(all_detected_devices);
9142 struct detected_devices_node {
9143 	struct list_head list;
9144 	dev_t dev;
9145 };
9146 
9147 void md_autodetect_dev(dev_t dev)
9148 {
9149 	struct detected_devices_node *node_detected_dev;
9150 
9151 	node_detected_dev = kzalloc(sizeof(*node_detected_dev), GFP_KERNEL);
9152 	if (node_detected_dev) {
9153 		node_detected_dev->dev = dev;
9154 		mutex_lock(&detected_devices_mutex);
9155 		list_add_tail(&node_detected_dev->list, &all_detected_devices);
9156 		mutex_unlock(&detected_devices_mutex);
9157 	}
9158 }
9159 
9160 static void autostart_arrays(int part)
9161 {
9162 	struct md_rdev *rdev;
9163 	struct detected_devices_node *node_detected_dev;
9164 	dev_t dev;
9165 	int i_scanned, i_passed;
9166 
9167 	i_scanned = 0;
9168 	i_passed = 0;
9169 
9170 	pr_info("md: Autodetecting RAID arrays.\n");
9171 
9172 	mutex_lock(&detected_devices_mutex);
9173 	while (!list_empty(&all_detected_devices) && i_scanned < INT_MAX) {
9174 		i_scanned++;
9175 		node_detected_dev = list_entry(all_detected_devices.next,
9176 					struct detected_devices_node, list);
9177 		list_del(&node_detected_dev->list);
9178 		dev = node_detected_dev->dev;
9179 		kfree(node_detected_dev);
9180 		mutex_unlock(&detected_devices_mutex);
9181 		rdev = md_import_device(dev,0, 90);
9182 		mutex_lock(&detected_devices_mutex);
9183 		if (IS_ERR(rdev))
9184 			continue;
9185 
9186 		if (test_bit(Faulty, &rdev->flags))
9187 			continue;
9188 
9189 		set_bit(AutoDetected, &rdev->flags);
9190 		list_add(&rdev->same_set, &pending_raid_disks);
9191 		i_passed++;
9192 	}
9193 	mutex_unlock(&detected_devices_mutex);
9194 
9195 	pr_debug("md: Scanned %d and added %d devices.\n", i_scanned, i_passed);
9196 
9197 	autorun_devices(part);
9198 }
9199 
9200 #endif /* !MODULE */
9201 
9202 static __exit void md_exit(void)
9203 {
9204 	struct mddev *mddev;
9205 	struct list_head *tmp;
9206 	int delay = 1;
9207 
9208 	blk_unregister_region(MKDEV(MD_MAJOR,0), 512);
9209 	blk_unregister_region(MKDEV(mdp_major,0), 1U << MINORBITS);
9210 
9211 	unregister_blkdev(MD_MAJOR,"md");
9212 	unregister_blkdev(mdp_major, "mdp");
9213 	unregister_reboot_notifier(&md_notifier);
9214 	unregister_sysctl_table(raid_table_header);
9215 
9216 	/* We cannot unload the modules while some process is
9217 	 * waiting for us in select() or poll() - wake them up
9218 	 */
9219 	md_unloading = 1;
9220 	while (waitqueue_active(&md_event_waiters)) {
9221 		/* not safe to leave yet */
9222 		wake_up(&md_event_waiters);
9223 		msleep(delay);
9224 		delay += delay;
9225 	}
9226 	remove_proc_entry("mdstat", NULL);
9227 
9228 	for_each_mddev(mddev, tmp) {
9229 		export_array(mddev);
9230 		mddev->ctime = 0;
9231 		mddev->hold_active = 0;
9232 		/*
9233 		 * for_each_mddev() will call mddev_put() at the end of each
9234 		 * iteration.  As the mddev is now fully clear, this will
9235 		 * schedule the mddev for destruction by a workqueue, and the
9236 		 * destroy_workqueue() below will wait for that to complete.
9237 		 */
9238 	}
9239 	destroy_workqueue(md_misc_wq);
9240 	destroy_workqueue(md_wq);
9241 }
9242 
9243 subsys_initcall(md_init);
9244 module_exit(md_exit)
9245 
9246 static int get_ro(char *buffer, struct kernel_param *kp)
9247 {
9248 	return sprintf(buffer, "%d", start_readonly);
9249 }
9250 static int set_ro(const char *val, struct kernel_param *kp)
9251 {
9252 	return kstrtouint(val, 10, (unsigned int *)&start_readonly);
9253 }
9254 
9255 module_param_call(start_ro, set_ro, get_ro, NULL, S_IRUSR|S_IWUSR);
9256 module_param(start_dirty_degraded, int, S_IRUGO|S_IWUSR);
9257 module_param_call(new_array, add_named_array, NULL, NULL, S_IWUSR);
9258 module_param(create_on_open, bool, S_IRUSR|S_IWUSR);
9259 
9260 MODULE_LICENSE("GPL");
9261 MODULE_DESCRIPTION("MD RAID framework");
9262 MODULE_ALIAS("md");
9263 MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR);
9264