xref: /linux/drivers/md/md.c (revision 2363088eba2ecccfb643725e4864af73c4226a04)
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3    md.c : Multiple Devices driver for Linux
4      Copyright (C) 1998, 1999, 2000 Ingo Molnar
5 
6      completely rewritten, based on the MD driver code from Marc Zyngier
7 
8    Changes:
9 
10    - RAID-1/RAID-5 extensions by Miguel de Icaza, Gadi Oxman, Ingo Molnar
11    - RAID-6 extensions by H. Peter Anvin <hpa@zytor.com>
12    - boot support for linear and striped mode by Harald Hoyer <HarryH@Royal.Net>
13    - kerneld support by Boris Tobotras <boris@xtalk.msk.su>
14    - kmod support by: Cyrus Durgin
15    - RAID0 bugfixes: Mark Anthony Lisher <markal@iname.com>
16    - Devfs support by Richard Gooch <rgooch@atnf.csiro.au>
17 
18    - lots of fixes and improvements to the RAID1/RAID5 and generic
19      RAID code (such as request based resynchronization):
20 
21      Neil Brown <neilb@cse.unsw.edu.au>.
22 
23    - persistent bitmap code
24      Copyright (C) 2003-2004, Paul Clements, SteelEye Technology, Inc.
25 
26 
27    Errors, Warnings, etc.
28    Please use:
29      pr_crit() for error conditions that risk data loss
30      pr_err() for error conditions that are unexpected, like an IO error
31          or internal inconsistency
32      pr_warn() for error conditions that could have been predicated, like
33          adding a device to an array when it has incompatible metadata
34      pr_info() for every interesting, very rare events, like an array starting
35          or stopping, or resync starting or stopping
36      pr_debug() for everything else.
37 
38 */
39 
40 #include <linux/sched/mm.h>
41 #include <linux/sched/signal.h>
42 #include <linux/kthread.h>
43 #include <linux/blkdev.h>
44 #include <linux/blk-integrity.h>
45 #include <linux/badblocks.h>
46 #include <linux/sysctl.h>
47 #include <linux/seq_file.h>
48 #include <linux/fs.h>
49 #include <linux/poll.h>
50 #include <linux/ctype.h>
51 #include <linux/string.h>
52 #include <linux/hdreg.h>
53 #include <linux/proc_fs.h>
54 #include <linux/random.h>
55 #include <linux/major.h>
56 #include <linux/module.h>
57 #include <linux/reboot.h>
58 #include <linux/file.h>
59 #include <linux/compat.h>
60 #include <linux/delay.h>
61 #include <linux/raid/md_p.h>
62 #include <linux/raid/md_u.h>
63 #include <linux/raid/detect.h>
64 #include <linux/slab.h>
65 #include <linux/percpu-refcount.h>
66 #include <linux/part_stat.h>
67 
68 #include <trace/events/block.h>
69 #include "md.h"
70 #include "md-bitmap.h"
71 #include "md-cluster.h"
72 
73 /* pers_list is a list of registered personalities protected by pers_lock. */
74 static LIST_HEAD(pers_list);
75 static DEFINE_SPINLOCK(pers_lock);
76 
77 static const struct kobj_type md_ktype;
78 
79 struct md_cluster_operations *md_cluster_ops;
80 EXPORT_SYMBOL(md_cluster_ops);
81 static struct module *md_cluster_mod;
82 
83 static DECLARE_WAIT_QUEUE_HEAD(resync_wait);
84 static struct workqueue_struct *md_wq;
85 static struct workqueue_struct *md_misc_wq;
86 struct workqueue_struct *md_bitmap_wq;
87 
88 static int remove_and_add_spares(struct mddev *mddev,
89 				 struct md_rdev *this);
90 static void mddev_detach(struct mddev *mddev);
91 static void export_rdev(struct md_rdev *rdev, struct mddev *mddev);
92 static void md_wakeup_thread_directly(struct md_thread __rcu *thread);
93 
94 /*
95  * Default number of read corrections we'll attempt on an rdev
96  * before ejecting it from the array. We divide the read error
97  * count by 2 for every hour elapsed between read errors.
98  */
99 #define MD_DEFAULT_MAX_CORRECTED_READ_ERRORS 20
100 /* Default safemode delay: 200 msec */
101 #define DEFAULT_SAFEMODE_DELAY ((200 * HZ)/1000 +1)
102 /*
103  * Current RAID-1,4,5 parallel reconstruction 'guaranteed speed limit'
104  * is 1000 KB/sec, so the extra system load does not show up that much.
105  * Increase it if you want to have more _guaranteed_ speed. Note that
106  * the RAID driver will use the maximum available bandwidth if the IO
107  * subsystem is idle. There is also an 'absolute maximum' reconstruction
108  * speed limit - in case reconstruction slows down your system despite
109  * idle IO detection.
110  *
111  * you can change it via /proc/sys/dev/raid/speed_limit_min and _max.
112  * or /sys/block/mdX/md/sync_speed_{min,max}
113  */
114 
115 static int sysctl_speed_limit_min = 1000;
116 static int sysctl_speed_limit_max = 200000;
117 static inline int speed_min(struct mddev *mddev)
118 {
119 	return mddev->sync_speed_min ?
120 		mddev->sync_speed_min : sysctl_speed_limit_min;
121 }
122 
123 static inline int speed_max(struct mddev *mddev)
124 {
125 	return mddev->sync_speed_max ?
126 		mddev->sync_speed_max : sysctl_speed_limit_max;
127 }
128 
129 static void rdev_uninit_serial(struct md_rdev *rdev)
130 {
131 	if (!test_and_clear_bit(CollisionCheck, &rdev->flags))
132 		return;
133 
134 	kvfree(rdev->serial);
135 	rdev->serial = NULL;
136 }
137 
138 static void rdevs_uninit_serial(struct mddev *mddev)
139 {
140 	struct md_rdev *rdev;
141 
142 	rdev_for_each(rdev, mddev)
143 		rdev_uninit_serial(rdev);
144 }
145 
146 static int rdev_init_serial(struct md_rdev *rdev)
147 {
148 	/* serial_nums equals with BARRIER_BUCKETS_NR */
149 	int i, serial_nums = 1 << ((PAGE_SHIFT - ilog2(sizeof(atomic_t))));
150 	struct serial_in_rdev *serial = NULL;
151 
152 	if (test_bit(CollisionCheck, &rdev->flags))
153 		return 0;
154 
155 	serial = kvmalloc(sizeof(struct serial_in_rdev) * serial_nums,
156 			  GFP_KERNEL);
157 	if (!serial)
158 		return -ENOMEM;
159 
160 	for (i = 0; i < serial_nums; i++) {
161 		struct serial_in_rdev *serial_tmp = &serial[i];
162 
163 		spin_lock_init(&serial_tmp->serial_lock);
164 		serial_tmp->serial_rb = RB_ROOT_CACHED;
165 		init_waitqueue_head(&serial_tmp->serial_io_wait);
166 	}
167 
168 	rdev->serial = serial;
169 	set_bit(CollisionCheck, &rdev->flags);
170 
171 	return 0;
172 }
173 
174 static int rdevs_init_serial(struct mddev *mddev)
175 {
176 	struct md_rdev *rdev;
177 	int ret = 0;
178 
179 	rdev_for_each(rdev, mddev) {
180 		ret = rdev_init_serial(rdev);
181 		if (ret)
182 			break;
183 	}
184 
185 	/* Free all resources if pool is not existed */
186 	if (ret && !mddev->serial_info_pool)
187 		rdevs_uninit_serial(mddev);
188 
189 	return ret;
190 }
191 
192 /*
193  * rdev needs to enable serial stuffs if it meets the conditions:
194  * 1. it is multi-queue device flaged with writemostly.
195  * 2. the write-behind mode is enabled.
196  */
197 static int rdev_need_serial(struct md_rdev *rdev)
198 {
199 	return (rdev && rdev->mddev->bitmap_info.max_write_behind > 0 &&
200 		rdev->bdev->bd_disk->queue->nr_hw_queues != 1 &&
201 		test_bit(WriteMostly, &rdev->flags));
202 }
203 
204 /*
205  * Init resource for rdev(s), then create serial_info_pool if:
206  * 1. rdev is the first device which return true from rdev_enable_serial.
207  * 2. rdev is NULL, means we want to enable serialization for all rdevs.
208  */
209 void mddev_create_serial_pool(struct mddev *mddev, struct md_rdev *rdev,
210 			      bool is_suspend)
211 {
212 	int ret = 0;
213 
214 	if (rdev && !rdev_need_serial(rdev) &&
215 	    !test_bit(CollisionCheck, &rdev->flags))
216 		return;
217 
218 	if (!is_suspend)
219 		mddev_suspend(mddev);
220 
221 	if (!rdev)
222 		ret = rdevs_init_serial(mddev);
223 	else
224 		ret = rdev_init_serial(rdev);
225 	if (ret)
226 		goto abort;
227 
228 	if (mddev->serial_info_pool == NULL) {
229 		/*
230 		 * already in memalloc noio context by
231 		 * mddev_suspend()
232 		 */
233 		mddev->serial_info_pool =
234 			mempool_create_kmalloc_pool(NR_SERIAL_INFOS,
235 						sizeof(struct serial_info));
236 		if (!mddev->serial_info_pool) {
237 			rdevs_uninit_serial(mddev);
238 			pr_err("can't alloc memory pool for serialization\n");
239 		}
240 	}
241 
242 abort:
243 	if (!is_suspend)
244 		mddev_resume(mddev);
245 }
246 
247 /*
248  * Free resource from rdev(s), and destroy serial_info_pool under conditions:
249  * 1. rdev is the last device flaged with CollisionCheck.
250  * 2. when bitmap is destroyed while policy is not enabled.
251  * 3. for disable policy, the pool is destroyed only when no rdev needs it.
252  */
253 void mddev_destroy_serial_pool(struct mddev *mddev, struct md_rdev *rdev,
254 			       bool is_suspend)
255 {
256 	if (rdev && !test_bit(CollisionCheck, &rdev->flags))
257 		return;
258 
259 	if (mddev->serial_info_pool) {
260 		struct md_rdev *temp;
261 		int num = 0; /* used to track if other rdevs need the pool */
262 
263 		if (!is_suspend)
264 			mddev_suspend(mddev);
265 		rdev_for_each(temp, mddev) {
266 			if (!rdev) {
267 				if (!mddev->serialize_policy ||
268 				    !rdev_need_serial(temp))
269 					rdev_uninit_serial(temp);
270 				else
271 					num++;
272 			} else if (temp != rdev &&
273 				   test_bit(CollisionCheck, &temp->flags))
274 				num++;
275 		}
276 
277 		if (rdev)
278 			rdev_uninit_serial(rdev);
279 
280 		if (num)
281 			pr_info("The mempool could be used by other devices\n");
282 		else {
283 			mempool_destroy(mddev->serial_info_pool);
284 			mddev->serial_info_pool = NULL;
285 		}
286 		if (!is_suspend)
287 			mddev_resume(mddev);
288 	}
289 }
290 
291 static struct ctl_table_header *raid_table_header;
292 
293 static struct ctl_table raid_table[] = {
294 	{
295 		.procname	= "speed_limit_min",
296 		.data		= &sysctl_speed_limit_min,
297 		.maxlen		= sizeof(int),
298 		.mode		= S_IRUGO|S_IWUSR,
299 		.proc_handler	= proc_dointvec,
300 	},
301 	{
302 		.procname	= "speed_limit_max",
303 		.data		= &sysctl_speed_limit_max,
304 		.maxlen		= sizeof(int),
305 		.mode		= S_IRUGO|S_IWUSR,
306 		.proc_handler	= proc_dointvec,
307 	},
308 	{ }
309 };
310 
311 static int start_readonly;
312 
313 /*
314  * The original mechanism for creating an md device is to create
315  * a device node in /dev and to open it.  This causes races with device-close.
316  * The preferred method is to write to the "new_array" module parameter.
317  * This can avoid races.
318  * Setting create_on_open to false disables the original mechanism
319  * so all the races disappear.
320  */
321 static bool create_on_open = true;
322 
323 /*
324  * We have a system wide 'event count' that is incremented
325  * on any 'interesting' event, and readers of /proc/mdstat
326  * can use 'poll' or 'select' to find out when the event
327  * count increases.
328  *
329  * Events are:
330  *  start array, stop array, error, add device, remove device,
331  *  start build, activate spare
332  */
333 static DECLARE_WAIT_QUEUE_HEAD(md_event_waiters);
334 static atomic_t md_event_count;
335 void md_new_event(void)
336 {
337 	atomic_inc(&md_event_count);
338 	wake_up(&md_event_waiters);
339 }
340 EXPORT_SYMBOL_GPL(md_new_event);
341 
342 /*
343  * Enables to iterate over all existing md arrays
344  * all_mddevs_lock protects this list.
345  */
346 static LIST_HEAD(all_mddevs);
347 static DEFINE_SPINLOCK(all_mddevs_lock);
348 
349 /* Rather than calling directly into the personality make_request function,
350  * IO requests come here first so that we can check if the device is
351  * being suspended pending a reconfiguration.
352  * We hold a refcount over the call to ->make_request.  By the time that
353  * call has finished, the bio has been linked into some internal structure
354  * and so is visible to ->quiesce(), so we don't need the refcount any more.
355  */
356 static bool is_suspended(struct mddev *mddev, struct bio *bio)
357 {
358 	if (is_md_suspended(mddev))
359 		return true;
360 	if (bio_data_dir(bio) != WRITE)
361 		return false;
362 	if (mddev->suspend_lo >= mddev->suspend_hi)
363 		return false;
364 	if (bio->bi_iter.bi_sector >= mddev->suspend_hi)
365 		return false;
366 	if (bio_end_sector(bio) < mddev->suspend_lo)
367 		return false;
368 	return true;
369 }
370 
371 void md_handle_request(struct mddev *mddev, struct bio *bio)
372 {
373 check_suspended:
374 	if (is_suspended(mddev, bio)) {
375 		DEFINE_WAIT(__wait);
376 		/* Bail out if REQ_NOWAIT is set for the bio */
377 		if (bio->bi_opf & REQ_NOWAIT) {
378 			bio_wouldblock_error(bio);
379 			return;
380 		}
381 		for (;;) {
382 			prepare_to_wait(&mddev->sb_wait, &__wait,
383 					TASK_UNINTERRUPTIBLE);
384 			if (!is_suspended(mddev, bio))
385 				break;
386 			schedule();
387 		}
388 		finish_wait(&mddev->sb_wait, &__wait);
389 	}
390 	if (!percpu_ref_tryget_live(&mddev->active_io))
391 		goto check_suspended;
392 
393 	if (!mddev->pers->make_request(mddev, bio)) {
394 		percpu_ref_put(&mddev->active_io);
395 		goto check_suspended;
396 	}
397 
398 	percpu_ref_put(&mddev->active_io);
399 }
400 EXPORT_SYMBOL(md_handle_request);
401 
402 static void md_submit_bio(struct bio *bio)
403 {
404 	const int rw = bio_data_dir(bio);
405 	struct mddev *mddev = bio->bi_bdev->bd_disk->private_data;
406 
407 	if (mddev == NULL || mddev->pers == NULL) {
408 		bio_io_error(bio);
409 		return;
410 	}
411 
412 	if (unlikely(test_bit(MD_BROKEN, &mddev->flags)) && (rw == WRITE)) {
413 		bio_io_error(bio);
414 		return;
415 	}
416 
417 	bio = bio_split_to_limits(bio);
418 	if (!bio)
419 		return;
420 
421 	if (mddev->ro == MD_RDONLY && unlikely(rw == WRITE)) {
422 		if (bio_sectors(bio) != 0)
423 			bio->bi_status = BLK_STS_IOERR;
424 		bio_endio(bio);
425 		return;
426 	}
427 
428 	/* bio could be mergeable after passing to underlayer */
429 	bio->bi_opf &= ~REQ_NOMERGE;
430 
431 	md_handle_request(mddev, bio);
432 }
433 
434 /* mddev_suspend makes sure no new requests are submitted
435  * to the device, and that any requests that have been submitted
436  * are completely handled.
437  * Once mddev_detach() is called and completes, the module will be
438  * completely unused.
439  */
440 void mddev_suspend(struct mddev *mddev)
441 {
442 	struct md_thread *thread = rcu_dereference_protected(mddev->thread,
443 			lockdep_is_held(&mddev->reconfig_mutex));
444 
445 	WARN_ON_ONCE(thread && current == thread->tsk);
446 	if (mddev->suspended++)
447 		return;
448 	wake_up(&mddev->sb_wait);
449 	set_bit(MD_ALLOW_SB_UPDATE, &mddev->flags);
450 	percpu_ref_kill(&mddev->active_io);
451 
452 	if (mddev->pers->prepare_suspend)
453 		mddev->pers->prepare_suspend(mddev);
454 
455 	wait_event(mddev->sb_wait, percpu_ref_is_zero(&mddev->active_io));
456 	clear_bit_unlock(MD_ALLOW_SB_UPDATE, &mddev->flags);
457 	wait_event(mddev->sb_wait, !test_bit(MD_UPDATING_SB, &mddev->flags));
458 
459 	del_timer_sync(&mddev->safemode_timer);
460 	/* restrict memory reclaim I/O during raid array is suspend */
461 	mddev->noio_flag = memalloc_noio_save();
462 }
463 EXPORT_SYMBOL_GPL(mddev_suspend);
464 
465 void mddev_resume(struct mddev *mddev)
466 {
467 	lockdep_assert_held(&mddev->reconfig_mutex);
468 	if (--mddev->suspended)
469 		return;
470 
471 	/* entred the memalloc scope from mddev_suspend() */
472 	memalloc_noio_restore(mddev->noio_flag);
473 
474 	percpu_ref_resurrect(&mddev->active_io);
475 	wake_up(&mddev->sb_wait);
476 
477 	set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
478 	md_wakeup_thread(mddev->thread);
479 	md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
480 }
481 EXPORT_SYMBOL_GPL(mddev_resume);
482 
483 /*
484  * Generic flush handling for md
485  */
486 
487 static void md_end_flush(struct bio *bio)
488 {
489 	struct md_rdev *rdev = bio->bi_private;
490 	struct mddev *mddev = rdev->mddev;
491 
492 	bio_put(bio);
493 
494 	rdev_dec_pending(rdev, mddev);
495 
496 	if (atomic_dec_and_test(&mddev->flush_pending)) {
497 		/* The pre-request flush has finished */
498 		queue_work(md_wq, &mddev->flush_work);
499 	}
500 }
501 
502 static void md_submit_flush_data(struct work_struct *ws);
503 
504 static void submit_flushes(struct work_struct *ws)
505 {
506 	struct mddev *mddev = container_of(ws, struct mddev, flush_work);
507 	struct md_rdev *rdev;
508 
509 	mddev->start_flush = ktime_get_boottime();
510 	INIT_WORK(&mddev->flush_work, md_submit_flush_data);
511 	atomic_set(&mddev->flush_pending, 1);
512 	rcu_read_lock();
513 	rdev_for_each_rcu(rdev, mddev)
514 		if (rdev->raid_disk >= 0 &&
515 		    !test_bit(Faulty, &rdev->flags)) {
516 			/* Take two references, one is dropped
517 			 * when request finishes, one after
518 			 * we reclaim rcu_read_lock
519 			 */
520 			struct bio *bi;
521 			atomic_inc(&rdev->nr_pending);
522 			atomic_inc(&rdev->nr_pending);
523 			rcu_read_unlock();
524 			bi = bio_alloc_bioset(rdev->bdev, 0,
525 					      REQ_OP_WRITE | REQ_PREFLUSH,
526 					      GFP_NOIO, &mddev->bio_set);
527 			bi->bi_end_io = md_end_flush;
528 			bi->bi_private = rdev;
529 			atomic_inc(&mddev->flush_pending);
530 			submit_bio(bi);
531 			rcu_read_lock();
532 			rdev_dec_pending(rdev, mddev);
533 		}
534 	rcu_read_unlock();
535 	if (atomic_dec_and_test(&mddev->flush_pending))
536 		queue_work(md_wq, &mddev->flush_work);
537 }
538 
539 static void md_submit_flush_data(struct work_struct *ws)
540 {
541 	struct mddev *mddev = container_of(ws, struct mddev, flush_work);
542 	struct bio *bio = mddev->flush_bio;
543 
544 	/*
545 	 * must reset flush_bio before calling into md_handle_request to avoid a
546 	 * deadlock, because other bios passed md_handle_request suspend check
547 	 * could wait for this and below md_handle_request could wait for those
548 	 * bios because of suspend check
549 	 */
550 	spin_lock_irq(&mddev->lock);
551 	mddev->prev_flush_start = mddev->start_flush;
552 	mddev->flush_bio = NULL;
553 	spin_unlock_irq(&mddev->lock);
554 	wake_up(&mddev->sb_wait);
555 
556 	if (bio->bi_iter.bi_size == 0) {
557 		/* an empty barrier - all done */
558 		bio_endio(bio);
559 	} else {
560 		bio->bi_opf &= ~REQ_PREFLUSH;
561 		md_handle_request(mddev, bio);
562 	}
563 }
564 
565 /*
566  * Manages consolidation of flushes and submitting any flushes needed for
567  * a bio with REQ_PREFLUSH.  Returns true if the bio is finished or is
568  * being finished in another context.  Returns false if the flushing is
569  * complete but still needs the I/O portion of the bio to be processed.
570  */
571 bool md_flush_request(struct mddev *mddev, struct bio *bio)
572 {
573 	ktime_t req_start = ktime_get_boottime();
574 	spin_lock_irq(&mddev->lock);
575 	/* flush requests wait until ongoing flush completes,
576 	 * hence coalescing all the pending requests.
577 	 */
578 	wait_event_lock_irq(mddev->sb_wait,
579 			    !mddev->flush_bio ||
580 			    ktime_before(req_start, mddev->prev_flush_start),
581 			    mddev->lock);
582 	/* new request after previous flush is completed */
583 	if (ktime_after(req_start, mddev->prev_flush_start)) {
584 		WARN_ON(mddev->flush_bio);
585 		mddev->flush_bio = bio;
586 		bio = NULL;
587 	}
588 	spin_unlock_irq(&mddev->lock);
589 
590 	if (!bio) {
591 		INIT_WORK(&mddev->flush_work, submit_flushes);
592 		queue_work(md_wq, &mddev->flush_work);
593 	} else {
594 		/* flush was performed for some other bio while we waited. */
595 		if (bio->bi_iter.bi_size == 0)
596 			/* an empty barrier - all done */
597 			bio_endio(bio);
598 		else {
599 			bio->bi_opf &= ~REQ_PREFLUSH;
600 			return false;
601 		}
602 	}
603 	return true;
604 }
605 EXPORT_SYMBOL(md_flush_request);
606 
607 static inline struct mddev *mddev_get(struct mddev *mddev)
608 {
609 	lockdep_assert_held(&all_mddevs_lock);
610 
611 	if (test_bit(MD_DELETED, &mddev->flags))
612 		return NULL;
613 	atomic_inc(&mddev->active);
614 	return mddev;
615 }
616 
617 static void mddev_delayed_delete(struct work_struct *ws);
618 
619 void mddev_put(struct mddev *mddev)
620 {
621 	if (!atomic_dec_and_lock(&mddev->active, &all_mddevs_lock))
622 		return;
623 	if (!mddev->raid_disks && list_empty(&mddev->disks) &&
624 	    mddev->ctime == 0 && !mddev->hold_active) {
625 		/* Array is not configured at all, and not held active,
626 		 * so destroy it */
627 		set_bit(MD_DELETED, &mddev->flags);
628 
629 		/*
630 		 * Call queue_work inside the spinlock so that
631 		 * flush_workqueue() after mddev_find will succeed in waiting
632 		 * for the work to be done.
633 		 */
634 		INIT_WORK(&mddev->del_work, mddev_delayed_delete);
635 		queue_work(md_misc_wq, &mddev->del_work);
636 	}
637 	spin_unlock(&all_mddevs_lock);
638 }
639 
640 static void md_safemode_timeout(struct timer_list *t);
641 
642 void mddev_init(struct mddev *mddev)
643 {
644 	mutex_init(&mddev->open_mutex);
645 	mutex_init(&mddev->reconfig_mutex);
646 	mutex_init(&mddev->sync_mutex);
647 	mutex_init(&mddev->bitmap_info.mutex);
648 	INIT_LIST_HEAD(&mddev->disks);
649 	INIT_LIST_HEAD(&mddev->all_mddevs);
650 	INIT_LIST_HEAD(&mddev->deleting);
651 	timer_setup(&mddev->safemode_timer, md_safemode_timeout, 0);
652 	atomic_set(&mddev->active, 1);
653 	atomic_set(&mddev->openers, 0);
654 	atomic_set(&mddev->sync_seq, 0);
655 	spin_lock_init(&mddev->lock);
656 	atomic_set(&mddev->flush_pending, 0);
657 	init_waitqueue_head(&mddev->sb_wait);
658 	init_waitqueue_head(&mddev->recovery_wait);
659 	mddev->reshape_position = MaxSector;
660 	mddev->reshape_backwards = 0;
661 	mddev->last_sync_action = "none";
662 	mddev->resync_min = 0;
663 	mddev->resync_max = MaxSector;
664 	mddev->level = LEVEL_NONE;
665 }
666 EXPORT_SYMBOL_GPL(mddev_init);
667 
668 static struct mddev *mddev_find_locked(dev_t unit)
669 {
670 	struct mddev *mddev;
671 
672 	list_for_each_entry(mddev, &all_mddevs, all_mddevs)
673 		if (mddev->unit == unit)
674 			return mddev;
675 
676 	return NULL;
677 }
678 
679 /* find an unused unit number */
680 static dev_t mddev_alloc_unit(void)
681 {
682 	static int next_minor = 512;
683 	int start = next_minor;
684 	bool is_free = 0;
685 	dev_t dev = 0;
686 
687 	while (!is_free) {
688 		dev = MKDEV(MD_MAJOR, next_minor);
689 		next_minor++;
690 		if (next_minor > MINORMASK)
691 			next_minor = 0;
692 		if (next_minor == start)
693 			return 0;		/* Oh dear, all in use. */
694 		is_free = !mddev_find_locked(dev);
695 	}
696 
697 	return dev;
698 }
699 
700 static struct mddev *mddev_alloc(dev_t unit)
701 {
702 	struct mddev *new;
703 	int error;
704 
705 	if (unit && MAJOR(unit) != MD_MAJOR)
706 		unit &= ~((1 << MdpMinorShift) - 1);
707 
708 	new = kzalloc(sizeof(*new), GFP_KERNEL);
709 	if (!new)
710 		return ERR_PTR(-ENOMEM);
711 	mddev_init(new);
712 
713 	spin_lock(&all_mddevs_lock);
714 	if (unit) {
715 		error = -EEXIST;
716 		if (mddev_find_locked(unit))
717 			goto out_free_new;
718 		new->unit = unit;
719 		if (MAJOR(unit) == MD_MAJOR)
720 			new->md_minor = MINOR(unit);
721 		else
722 			new->md_minor = MINOR(unit) >> MdpMinorShift;
723 		new->hold_active = UNTIL_IOCTL;
724 	} else {
725 		error = -ENODEV;
726 		new->unit = mddev_alloc_unit();
727 		if (!new->unit)
728 			goto out_free_new;
729 		new->md_minor = MINOR(new->unit);
730 		new->hold_active = UNTIL_STOP;
731 	}
732 
733 	list_add(&new->all_mddevs, &all_mddevs);
734 	spin_unlock(&all_mddevs_lock);
735 	return new;
736 out_free_new:
737 	spin_unlock(&all_mddevs_lock);
738 	kfree(new);
739 	return ERR_PTR(error);
740 }
741 
742 static void mddev_free(struct mddev *mddev)
743 {
744 	spin_lock(&all_mddevs_lock);
745 	list_del(&mddev->all_mddevs);
746 	spin_unlock(&all_mddevs_lock);
747 
748 	kfree(mddev);
749 }
750 
751 static const struct attribute_group md_redundancy_group;
752 
753 void mddev_unlock(struct mddev *mddev)
754 {
755 	struct md_rdev *rdev;
756 	struct md_rdev *tmp;
757 	LIST_HEAD(delete);
758 
759 	if (!list_empty(&mddev->deleting))
760 		list_splice_init(&mddev->deleting, &delete);
761 
762 	if (mddev->to_remove) {
763 		/* These cannot be removed under reconfig_mutex as
764 		 * an access to the files will try to take reconfig_mutex
765 		 * while holding the file unremovable, which leads to
766 		 * a deadlock.
767 		 * So hold set sysfs_active while the remove in happeing,
768 		 * and anything else which might set ->to_remove or my
769 		 * otherwise change the sysfs namespace will fail with
770 		 * -EBUSY if sysfs_active is still set.
771 		 * We set sysfs_active under reconfig_mutex and elsewhere
772 		 * test it under the same mutex to ensure its correct value
773 		 * is seen.
774 		 */
775 		const struct attribute_group *to_remove = mddev->to_remove;
776 		mddev->to_remove = NULL;
777 		mddev->sysfs_active = 1;
778 		mutex_unlock(&mddev->reconfig_mutex);
779 
780 		if (mddev->kobj.sd) {
781 			if (to_remove != &md_redundancy_group)
782 				sysfs_remove_group(&mddev->kobj, to_remove);
783 			if (mddev->pers == NULL ||
784 			    mddev->pers->sync_request == NULL) {
785 				sysfs_remove_group(&mddev->kobj, &md_redundancy_group);
786 				if (mddev->sysfs_action)
787 					sysfs_put(mddev->sysfs_action);
788 				if (mddev->sysfs_completed)
789 					sysfs_put(mddev->sysfs_completed);
790 				if (mddev->sysfs_degraded)
791 					sysfs_put(mddev->sysfs_degraded);
792 				mddev->sysfs_action = NULL;
793 				mddev->sysfs_completed = NULL;
794 				mddev->sysfs_degraded = NULL;
795 			}
796 		}
797 		mddev->sysfs_active = 0;
798 	} else
799 		mutex_unlock(&mddev->reconfig_mutex);
800 
801 	list_for_each_entry_safe(rdev, tmp, &delete, same_set) {
802 		list_del_init(&rdev->same_set);
803 		kobject_del(&rdev->kobj);
804 		export_rdev(rdev, mddev);
805 	}
806 
807 	md_wakeup_thread(mddev->thread);
808 	wake_up(&mddev->sb_wait);
809 }
810 EXPORT_SYMBOL_GPL(mddev_unlock);
811 
812 struct md_rdev *md_find_rdev_nr_rcu(struct mddev *mddev, int nr)
813 {
814 	struct md_rdev *rdev;
815 
816 	rdev_for_each_rcu(rdev, mddev)
817 		if (rdev->desc_nr == nr)
818 			return rdev;
819 
820 	return NULL;
821 }
822 EXPORT_SYMBOL_GPL(md_find_rdev_nr_rcu);
823 
824 static struct md_rdev *find_rdev(struct mddev *mddev, dev_t dev)
825 {
826 	struct md_rdev *rdev;
827 
828 	rdev_for_each(rdev, mddev)
829 		if (rdev->bdev->bd_dev == dev)
830 			return rdev;
831 
832 	return NULL;
833 }
834 
835 struct md_rdev *md_find_rdev_rcu(struct mddev *mddev, dev_t dev)
836 {
837 	struct md_rdev *rdev;
838 
839 	rdev_for_each_rcu(rdev, mddev)
840 		if (rdev->bdev->bd_dev == dev)
841 			return rdev;
842 
843 	return NULL;
844 }
845 EXPORT_SYMBOL_GPL(md_find_rdev_rcu);
846 
847 static struct md_personality *find_pers(int level, char *clevel)
848 {
849 	struct md_personality *pers;
850 	list_for_each_entry(pers, &pers_list, list) {
851 		if (level != LEVEL_NONE && pers->level == level)
852 			return pers;
853 		if (strcmp(pers->name, clevel)==0)
854 			return pers;
855 	}
856 	return NULL;
857 }
858 
859 /* return the offset of the super block in 512byte sectors */
860 static inline sector_t calc_dev_sboffset(struct md_rdev *rdev)
861 {
862 	return MD_NEW_SIZE_SECTORS(bdev_nr_sectors(rdev->bdev));
863 }
864 
865 static int alloc_disk_sb(struct md_rdev *rdev)
866 {
867 	rdev->sb_page = alloc_page(GFP_KERNEL);
868 	if (!rdev->sb_page)
869 		return -ENOMEM;
870 	return 0;
871 }
872 
873 void md_rdev_clear(struct md_rdev *rdev)
874 {
875 	if (rdev->sb_page) {
876 		put_page(rdev->sb_page);
877 		rdev->sb_loaded = 0;
878 		rdev->sb_page = NULL;
879 		rdev->sb_start = 0;
880 		rdev->sectors = 0;
881 	}
882 	if (rdev->bb_page) {
883 		put_page(rdev->bb_page);
884 		rdev->bb_page = NULL;
885 	}
886 	badblocks_exit(&rdev->badblocks);
887 }
888 EXPORT_SYMBOL_GPL(md_rdev_clear);
889 
890 static void super_written(struct bio *bio)
891 {
892 	struct md_rdev *rdev = bio->bi_private;
893 	struct mddev *mddev = rdev->mddev;
894 
895 	if (bio->bi_status) {
896 		pr_err("md: %s gets error=%d\n", __func__,
897 		       blk_status_to_errno(bio->bi_status));
898 		md_error(mddev, rdev);
899 		if (!test_bit(Faulty, &rdev->flags)
900 		    && (bio->bi_opf & MD_FAILFAST)) {
901 			set_bit(MD_SB_NEED_REWRITE, &mddev->sb_flags);
902 			set_bit(LastDev, &rdev->flags);
903 		}
904 	} else
905 		clear_bit(LastDev, &rdev->flags);
906 
907 	bio_put(bio);
908 
909 	rdev_dec_pending(rdev, mddev);
910 
911 	if (atomic_dec_and_test(&mddev->pending_writes))
912 		wake_up(&mddev->sb_wait);
913 }
914 
915 void md_super_write(struct mddev *mddev, struct md_rdev *rdev,
916 		   sector_t sector, int size, struct page *page)
917 {
918 	/* write first size bytes of page to sector of rdev
919 	 * Increment mddev->pending_writes before returning
920 	 * and decrement it on completion, waking up sb_wait
921 	 * if zero is reached.
922 	 * If an error occurred, call md_error
923 	 */
924 	struct bio *bio;
925 
926 	if (!page)
927 		return;
928 
929 	if (test_bit(Faulty, &rdev->flags))
930 		return;
931 
932 	bio = bio_alloc_bioset(rdev->meta_bdev ? rdev->meta_bdev : rdev->bdev,
933 			       1,
934 			       REQ_OP_WRITE | REQ_SYNC | REQ_PREFLUSH | REQ_FUA,
935 			       GFP_NOIO, &mddev->sync_set);
936 
937 	atomic_inc(&rdev->nr_pending);
938 
939 	bio->bi_iter.bi_sector = sector;
940 	__bio_add_page(bio, page, size, 0);
941 	bio->bi_private = rdev;
942 	bio->bi_end_io = super_written;
943 
944 	if (test_bit(MD_FAILFAST_SUPPORTED, &mddev->flags) &&
945 	    test_bit(FailFast, &rdev->flags) &&
946 	    !test_bit(LastDev, &rdev->flags))
947 		bio->bi_opf |= MD_FAILFAST;
948 
949 	atomic_inc(&mddev->pending_writes);
950 	submit_bio(bio);
951 }
952 
953 int md_super_wait(struct mddev *mddev)
954 {
955 	/* wait for all superblock writes that were scheduled to complete */
956 	wait_event(mddev->sb_wait, atomic_read(&mddev->pending_writes)==0);
957 	if (test_and_clear_bit(MD_SB_NEED_REWRITE, &mddev->sb_flags))
958 		return -EAGAIN;
959 	return 0;
960 }
961 
962 int sync_page_io(struct md_rdev *rdev, sector_t sector, int size,
963 		 struct page *page, blk_opf_t opf, bool metadata_op)
964 {
965 	struct bio bio;
966 	struct bio_vec bvec;
967 
968 	if (metadata_op && rdev->meta_bdev)
969 		bio_init(&bio, rdev->meta_bdev, &bvec, 1, opf);
970 	else
971 		bio_init(&bio, rdev->bdev, &bvec, 1, opf);
972 
973 	if (metadata_op)
974 		bio.bi_iter.bi_sector = sector + rdev->sb_start;
975 	else if (rdev->mddev->reshape_position != MaxSector &&
976 		 (rdev->mddev->reshape_backwards ==
977 		  (sector >= rdev->mddev->reshape_position)))
978 		bio.bi_iter.bi_sector = sector + rdev->new_data_offset;
979 	else
980 		bio.bi_iter.bi_sector = sector + rdev->data_offset;
981 	__bio_add_page(&bio, page, size, 0);
982 
983 	submit_bio_wait(&bio);
984 
985 	return !bio.bi_status;
986 }
987 EXPORT_SYMBOL_GPL(sync_page_io);
988 
989 static int read_disk_sb(struct md_rdev *rdev, int size)
990 {
991 	if (rdev->sb_loaded)
992 		return 0;
993 
994 	if (!sync_page_io(rdev, 0, size, rdev->sb_page, REQ_OP_READ, true))
995 		goto fail;
996 	rdev->sb_loaded = 1;
997 	return 0;
998 
999 fail:
1000 	pr_err("md: disabled device %pg, could not read superblock.\n",
1001 	       rdev->bdev);
1002 	return -EINVAL;
1003 }
1004 
1005 static int md_uuid_equal(mdp_super_t *sb1, mdp_super_t *sb2)
1006 {
1007 	return	sb1->set_uuid0 == sb2->set_uuid0 &&
1008 		sb1->set_uuid1 == sb2->set_uuid1 &&
1009 		sb1->set_uuid2 == sb2->set_uuid2 &&
1010 		sb1->set_uuid3 == sb2->set_uuid3;
1011 }
1012 
1013 static int md_sb_equal(mdp_super_t *sb1, mdp_super_t *sb2)
1014 {
1015 	int ret;
1016 	mdp_super_t *tmp1, *tmp2;
1017 
1018 	tmp1 = kmalloc(sizeof(*tmp1),GFP_KERNEL);
1019 	tmp2 = kmalloc(sizeof(*tmp2),GFP_KERNEL);
1020 
1021 	if (!tmp1 || !tmp2) {
1022 		ret = 0;
1023 		goto abort;
1024 	}
1025 
1026 	*tmp1 = *sb1;
1027 	*tmp2 = *sb2;
1028 
1029 	/*
1030 	 * nr_disks is not constant
1031 	 */
1032 	tmp1->nr_disks = 0;
1033 	tmp2->nr_disks = 0;
1034 
1035 	ret = (memcmp(tmp1, tmp2, MD_SB_GENERIC_CONSTANT_WORDS * 4) == 0);
1036 abort:
1037 	kfree(tmp1);
1038 	kfree(tmp2);
1039 	return ret;
1040 }
1041 
1042 static u32 md_csum_fold(u32 csum)
1043 {
1044 	csum = (csum & 0xffff) + (csum >> 16);
1045 	return (csum & 0xffff) + (csum >> 16);
1046 }
1047 
1048 static unsigned int calc_sb_csum(mdp_super_t *sb)
1049 {
1050 	u64 newcsum = 0;
1051 	u32 *sb32 = (u32*)sb;
1052 	int i;
1053 	unsigned int disk_csum, csum;
1054 
1055 	disk_csum = sb->sb_csum;
1056 	sb->sb_csum = 0;
1057 
1058 	for (i = 0; i < MD_SB_BYTES/4 ; i++)
1059 		newcsum += sb32[i];
1060 	csum = (newcsum & 0xffffffff) + (newcsum>>32);
1061 
1062 #ifdef CONFIG_ALPHA
1063 	/* This used to use csum_partial, which was wrong for several
1064 	 * reasons including that different results are returned on
1065 	 * different architectures.  It isn't critical that we get exactly
1066 	 * the same return value as before (we always csum_fold before
1067 	 * testing, and that removes any differences).  However as we
1068 	 * know that csum_partial always returned a 16bit value on
1069 	 * alphas, do a fold to maximise conformity to previous behaviour.
1070 	 */
1071 	sb->sb_csum = md_csum_fold(disk_csum);
1072 #else
1073 	sb->sb_csum = disk_csum;
1074 #endif
1075 	return csum;
1076 }
1077 
1078 /*
1079  * Handle superblock details.
1080  * We want to be able to handle multiple superblock formats
1081  * so we have a common interface to them all, and an array of
1082  * different handlers.
1083  * We rely on user-space to write the initial superblock, and support
1084  * reading and updating of superblocks.
1085  * Interface methods are:
1086  *   int load_super(struct md_rdev *dev, struct md_rdev *refdev, int minor_version)
1087  *      loads and validates a superblock on dev.
1088  *      if refdev != NULL, compare superblocks on both devices
1089  *    Return:
1090  *      0 - dev has a superblock that is compatible with refdev
1091  *      1 - dev has a superblock that is compatible and newer than refdev
1092  *          so dev should be used as the refdev in future
1093  *     -EINVAL superblock incompatible or invalid
1094  *     -othererror e.g. -EIO
1095  *
1096  *   int validate_super(struct mddev *mddev, struct md_rdev *dev)
1097  *      Verify that dev is acceptable into mddev.
1098  *       The first time, mddev->raid_disks will be 0, and data from
1099  *       dev should be merged in.  Subsequent calls check that dev
1100  *       is new enough.  Return 0 or -EINVAL
1101  *
1102  *   void sync_super(struct mddev *mddev, struct md_rdev *dev)
1103  *     Update the superblock for rdev with data in mddev
1104  *     This does not write to disc.
1105  *
1106  */
1107 
1108 struct super_type  {
1109 	char		    *name;
1110 	struct module	    *owner;
1111 	int		    (*load_super)(struct md_rdev *rdev,
1112 					  struct md_rdev *refdev,
1113 					  int minor_version);
1114 	int		    (*validate_super)(struct mddev *mddev,
1115 					      struct md_rdev *rdev);
1116 	void		    (*sync_super)(struct mddev *mddev,
1117 					  struct md_rdev *rdev);
1118 	unsigned long long  (*rdev_size_change)(struct md_rdev *rdev,
1119 						sector_t num_sectors);
1120 	int		    (*allow_new_offset)(struct md_rdev *rdev,
1121 						unsigned long long new_offset);
1122 };
1123 
1124 /*
1125  * Check that the given mddev has no bitmap.
1126  *
1127  * This function is called from the run method of all personalities that do not
1128  * support bitmaps. It prints an error message and returns non-zero if mddev
1129  * has a bitmap. Otherwise, it returns 0.
1130  *
1131  */
1132 int md_check_no_bitmap(struct mddev *mddev)
1133 {
1134 	if (!mddev->bitmap_info.file && !mddev->bitmap_info.offset)
1135 		return 0;
1136 	pr_warn("%s: bitmaps are not supported for %s\n",
1137 		mdname(mddev), mddev->pers->name);
1138 	return 1;
1139 }
1140 EXPORT_SYMBOL(md_check_no_bitmap);
1141 
1142 /*
1143  * load_super for 0.90.0
1144  */
1145 static int super_90_load(struct md_rdev *rdev, struct md_rdev *refdev, int minor_version)
1146 {
1147 	mdp_super_t *sb;
1148 	int ret;
1149 	bool spare_disk = true;
1150 
1151 	/*
1152 	 * Calculate the position of the superblock (512byte sectors),
1153 	 * it's at the end of the disk.
1154 	 *
1155 	 * It also happens to be a multiple of 4Kb.
1156 	 */
1157 	rdev->sb_start = calc_dev_sboffset(rdev);
1158 
1159 	ret = read_disk_sb(rdev, MD_SB_BYTES);
1160 	if (ret)
1161 		return ret;
1162 
1163 	ret = -EINVAL;
1164 
1165 	sb = page_address(rdev->sb_page);
1166 
1167 	if (sb->md_magic != MD_SB_MAGIC) {
1168 		pr_warn("md: invalid raid superblock magic on %pg\n",
1169 			rdev->bdev);
1170 		goto abort;
1171 	}
1172 
1173 	if (sb->major_version != 0 ||
1174 	    sb->minor_version < 90 ||
1175 	    sb->minor_version > 91) {
1176 		pr_warn("Bad version number %d.%d on %pg\n",
1177 			sb->major_version, sb->minor_version, rdev->bdev);
1178 		goto abort;
1179 	}
1180 
1181 	if (sb->raid_disks <= 0)
1182 		goto abort;
1183 
1184 	if (md_csum_fold(calc_sb_csum(sb)) != md_csum_fold(sb->sb_csum)) {
1185 		pr_warn("md: invalid superblock checksum on %pg\n", rdev->bdev);
1186 		goto abort;
1187 	}
1188 
1189 	rdev->preferred_minor = sb->md_minor;
1190 	rdev->data_offset = 0;
1191 	rdev->new_data_offset = 0;
1192 	rdev->sb_size = MD_SB_BYTES;
1193 	rdev->badblocks.shift = -1;
1194 
1195 	if (sb->level == LEVEL_MULTIPATH)
1196 		rdev->desc_nr = -1;
1197 	else
1198 		rdev->desc_nr = sb->this_disk.number;
1199 
1200 	/* not spare disk, or LEVEL_MULTIPATH */
1201 	if (sb->level == LEVEL_MULTIPATH ||
1202 		(rdev->desc_nr >= 0 &&
1203 		 rdev->desc_nr < MD_SB_DISKS &&
1204 		 sb->disks[rdev->desc_nr].state &
1205 		 ((1<<MD_DISK_SYNC) | (1 << MD_DISK_ACTIVE))))
1206 		spare_disk = false;
1207 
1208 	if (!refdev) {
1209 		if (!spare_disk)
1210 			ret = 1;
1211 		else
1212 			ret = 0;
1213 	} else {
1214 		__u64 ev1, ev2;
1215 		mdp_super_t *refsb = page_address(refdev->sb_page);
1216 		if (!md_uuid_equal(refsb, sb)) {
1217 			pr_warn("md: %pg has different UUID to %pg\n",
1218 				rdev->bdev, refdev->bdev);
1219 			goto abort;
1220 		}
1221 		if (!md_sb_equal(refsb, sb)) {
1222 			pr_warn("md: %pg has same UUID but different superblock to %pg\n",
1223 				rdev->bdev, refdev->bdev);
1224 			goto abort;
1225 		}
1226 		ev1 = md_event(sb);
1227 		ev2 = md_event(refsb);
1228 
1229 		if (!spare_disk && ev1 > ev2)
1230 			ret = 1;
1231 		else
1232 			ret = 0;
1233 	}
1234 	rdev->sectors = rdev->sb_start;
1235 	/* Limit to 4TB as metadata cannot record more than that.
1236 	 * (not needed for Linear and RAID0 as metadata doesn't
1237 	 * record this size)
1238 	 */
1239 	if ((u64)rdev->sectors >= (2ULL << 32) && sb->level >= 1)
1240 		rdev->sectors = (sector_t)(2ULL << 32) - 2;
1241 
1242 	if (rdev->sectors < ((sector_t)sb->size) * 2 && sb->level >= 1)
1243 		/* "this cannot possibly happen" ... */
1244 		ret = -EINVAL;
1245 
1246  abort:
1247 	return ret;
1248 }
1249 
1250 /*
1251  * validate_super for 0.90.0
1252  */
1253 static int super_90_validate(struct mddev *mddev, struct md_rdev *rdev)
1254 {
1255 	mdp_disk_t *desc;
1256 	mdp_super_t *sb = page_address(rdev->sb_page);
1257 	__u64 ev1 = md_event(sb);
1258 
1259 	rdev->raid_disk = -1;
1260 	clear_bit(Faulty, &rdev->flags);
1261 	clear_bit(In_sync, &rdev->flags);
1262 	clear_bit(Bitmap_sync, &rdev->flags);
1263 	clear_bit(WriteMostly, &rdev->flags);
1264 
1265 	if (mddev->raid_disks == 0) {
1266 		mddev->major_version = 0;
1267 		mddev->minor_version = sb->minor_version;
1268 		mddev->patch_version = sb->patch_version;
1269 		mddev->external = 0;
1270 		mddev->chunk_sectors = sb->chunk_size >> 9;
1271 		mddev->ctime = sb->ctime;
1272 		mddev->utime = sb->utime;
1273 		mddev->level = sb->level;
1274 		mddev->clevel[0] = 0;
1275 		mddev->layout = sb->layout;
1276 		mddev->raid_disks = sb->raid_disks;
1277 		mddev->dev_sectors = ((sector_t)sb->size) * 2;
1278 		mddev->events = ev1;
1279 		mddev->bitmap_info.offset = 0;
1280 		mddev->bitmap_info.space = 0;
1281 		/* bitmap can use 60 K after the 4K superblocks */
1282 		mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9;
1283 		mddev->bitmap_info.default_space = 64*2 - (MD_SB_BYTES >> 9);
1284 		mddev->reshape_backwards = 0;
1285 
1286 		if (mddev->minor_version >= 91) {
1287 			mddev->reshape_position = sb->reshape_position;
1288 			mddev->delta_disks = sb->delta_disks;
1289 			mddev->new_level = sb->new_level;
1290 			mddev->new_layout = sb->new_layout;
1291 			mddev->new_chunk_sectors = sb->new_chunk >> 9;
1292 			if (mddev->delta_disks < 0)
1293 				mddev->reshape_backwards = 1;
1294 		} else {
1295 			mddev->reshape_position = MaxSector;
1296 			mddev->delta_disks = 0;
1297 			mddev->new_level = mddev->level;
1298 			mddev->new_layout = mddev->layout;
1299 			mddev->new_chunk_sectors = mddev->chunk_sectors;
1300 		}
1301 		if (mddev->level == 0)
1302 			mddev->layout = -1;
1303 
1304 		if (sb->state & (1<<MD_SB_CLEAN))
1305 			mddev->recovery_cp = MaxSector;
1306 		else {
1307 			if (sb->events_hi == sb->cp_events_hi &&
1308 				sb->events_lo == sb->cp_events_lo) {
1309 				mddev->recovery_cp = sb->recovery_cp;
1310 			} else
1311 				mddev->recovery_cp = 0;
1312 		}
1313 
1314 		memcpy(mddev->uuid+0, &sb->set_uuid0, 4);
1315 		memcpy(mddev->uuid+4, &sb->set_uuid1, 4);
1316 		memcpy(mddev->uuid+8, &sb->set_uuid2, 4);
1317 		memcpy(mddev->uuid+12,&sb->set_uuid3, 4);
1318 
1319 		mddev->max_disks = MD_SB_DISKS;
1320 
1321 		if (sb->state & (1<<MD_SB_BITMAP_PRESENT) &&
1322 		    mddev->bitmap_info.file == NULL) {
1323 			mddev->bitmap_info.offset =
1324 				mddev->bitmap_info.default_offset;
1325 			mddev->bitmap_info.space =
1326 				mddev->bitmap_info.default_space;
1327 		}
1328 
1329 	} else if (mddev->pers == NULL) {
1330 		/* Insist on good event counter while assembling, except
1331 		 * for spares (which don't need an event count) */
1332 		++ev1;
1333 		if (sb->disks[rdev->desc_nr].state & (
1334 			    (1<<MD_DISK_SYNC) | (1 << MD_DISK_ACTIVE)))
1335 			if (ev1 < mddev->events)
1336 				return -EINVAL;
1337 	} else if (mddev->bitmap) {
1338 		/* if adding to array with a bitmap, then we can accept an
1339 		 * older device ... but not too old.
1340 		 */
1341 		if (ev1 < mddev->bitmap->events_cleared)
1342 			return 0;
1343 		if (ev1 < mddev->events)
1344 			set_bit(Bitmap_sync, &rdev->flags);
1345 	} else {
1346 		if (ev1 < mddev->events)
1347 			/* just a hot-add of a new device, leave raid_disk at -1 */
1348 			return 0;
1349 	}
1350 
1351 	if (mddev->level != LEVEL_MULTIPATH) {
1352 		desc = sb->disks + rdev->desc_nr;
1353 
1354 		if (desc->state & (1<<MD_DISK_FAULTY))
1355 			set_bit(Faulty, &rdev->flags);
1356 		else if (desc->state & (1<<MD_DISK_SYNC) /* &&
1357 			    desc->raid_disk < mddev->raid_disks */) {
1358 			set_bit(In_sync, &rdev->flags);
1359 			rdev->raid_disk = desc->raid_disk;
1360 			rdev->saved_raid_disk = desc->raid_disk;
1361 		} else if (desc->state & (1<<MD_DISK_ACTIVE)) {
1362 			/* active but not in sync implies recovery up to
1363 			 * reshape position.  We don't know exactly where
1364 			 * that is, so set to zero for now */
1365 			if (mddev->minor_version >= 91) {
1366 				rdev->recovery_offset = 0;
1367 				rdev->raid_disk = desc->raid_disk;
1368 			}
1369 		}
1370 		if (desc->state & (1<<MD_DISK_WRITEMOSTLY))
1371 			set_bit(WriteMostly, &rdev->flags);
1372 		if (desc->state & (1<<MD_DISK_FAILFAST))
1373 			set_bit(FailFast, &rdev->flags);
1374 	} else /* MULTIPATH are always insync */
1375 		set_bit(In_sync, &rdev->flags);
1376 	return 0;
1377 }
1378 
1379 /*
1380  * sync_super for 0.90.0
1381  */
1382 static void super_90_sync(struct mddev *mddev, struct md_rdev *rdev)
1383 {
1384 	mdp_super_t *sb;
1385 	struct md_rdev *rdev2;
1386 	int next_spare = mddev->raid_disks;
1387 
1388 	/* make rdev->sb match mddev data..
1389 	 *
1390 	 * 1/ zero out disks
1391 	 * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
1392 	 * 3/ any empty disks < next_spare become removed
1393 	 *
1394 	 * disks[0] gets initialised to REMOVED because
1395 	 * we cannot be sure from other fields if it has
1396 	 * been initialised or not.
1397 	 */
1398 	int i;
1399 	int active=0, working=0,failed=0,spare=0,nr_disks=0;
1400 
1401 	rdev->sb_size = MD_SB_BYTES;
1402 
1403 	sb = page_address(rdev->sb_page);
1404 
1405 	memset(sb, 0, sizeof(*sb));
1406 
1407 	sb->md_magic = MD_SB_MAGIC;
1408 	sb->major_version = mddev->major_version;
1409 	sb->patch_version = mddev->patch_version;
1410 	sb->gvalid_words  = 0; /* ignored */
1411 	memcpy(&sb->set_uuid0, mddev->uuid+0, 4);
1412 	memcpy(&sb->set_uuid1, mddev->uuid+4, 4);
1413 	memcpy(&sb->set_uuid2, mddev->uuid+8, 4);
1414 	memcpy(&sb->set_uuid3, mddev->uuid+12,4);
1415 
1416 	sb->ctime = clamp_t(time64_t, mddev->ctime, 0, U32_MAX);
1417 	sb->level = mddev->level;
1418 	sb->size = mddev->dev_sectors / 2;
1419 	sb->raid_disks = mddev->raid_disks;
1420 	sb->md_minor = mddev->md_minor;
1421 	sb->not_persistent = 0;
1422 	sb->utime = clamp_t(time64_t, mddev->utime, 0, U32_MAX);
1423 	sb->state = 0;
1424 	sb->events_hi = (mddev->events>>32);
1425 	sb->events_lo = (u32)mddev->events;
1426 
1427 	if (mddev->reshape_position == MaxSector)
1428 		sb->minor_version = 90;
1429 	else {
1430 		sb->minor_version = 91;
1431 		sb->reshape_position = mddev->reshape_position;
1432 		sb->new_level = mddev->new_level;
1433 		sb->delta_disks = mddev->delta_disks;
1434 		sb->new_layout = mddev->new_layout;
1435 		sb->new_chunk = mddev->new_chunk_sectors << 9;
1436 	}
1437 	mddev->minor_version = sb->minor_version;
1438 	if (mddev->in_sync)
1439 	{
1440 		sb->recovery_cp = mddev->recovery_cp;
1441 		sb->cp_events_hi = (mddev->events>>32);
1442 		sb->cp_events_lo = (u32)mddev->events;
1443 		if (mddev->recovery_cp == MaxSector)
1444 			sb->state = (1<< MD_SB_CLEAN);
1445 	} else
1446 		sb->recovery_cp = 0;
1447 
1448 	sb->layout = mddev->layout;
1449 	sb->chunk_size = mddev->chunk_sectors << 9;
1450 
1451 	if (mddev->bitmap && mddev->bitmap_info.file == NULL)
1452 		sb->state |= (1<<MD_SB_BITMAP_PRESENT);
1453 
1454 	sb->disks[0].state = (1<<MD_DISK_REMOVED);
1455 	rdev_for_each(rdev2, mddev) {
1456 		mdp_disk_t *d;
1457 		int desc_nr;
1458 		int is_active = test_bit(In_sync, &rdev2->flags);
1459 
1460 		if (rdev2->raid_disk >= 0 &&
1461 		    sb->minor_version >= 91)
1462 			/* we have nowhere to store the recovery_offset,
1463 			 * but if it is not below the reshape_position,
1464 			 * we can piggy-back on that.
1465 			 */
1466 			is_active = 1;
1467 		if (rdev2->raid_disk < 0 ||
1468 		    test_bit(Faulty, &rdev2->flags))
1469 			is_active = 0;
1470 		if (is_active)
1471 			desc_nr = rdev2->raid_disk;
1472 		else
1473 			desc_nr = next_spare++;
1474 		rdev2->desc_nr = desc_nr;
1475 		d = &sb->disks[rdev2->desc_nr];
1476 		nr_disks++;
1477 		d->number = rdev2->desc_nr;
1478 		d->major = MAJOR(rdev2->bdev->bd_dev);
1479 		d->minor = MINOR(rdev2->bdev->bd_dev);
1480 		if (is_active)
1481 			d->raid_disk = rdev2->raid_disk;
1482 		else
1483 			d->raid_disk = rdev2->desc_nr; /* compatibility */
1484 		if (test_bit(Faulty, &rdev2->flags))
1485 			d->state = (1<<MD_DISK_FAULTY);
1486 		else if (is_active) {
1487 			d->state = (1<<MD_DISK_ACTIVE);
1488 			if (test_bit(In_sync, &rdev2->flags))
1489 				d->state |= (1<<MD_DISK_SYNC);
1490 			active++;
1491 			working++;
1492 		} else {
1493 			d->state = 0;
1494 			spare++;
1495 			working++;
1496 		}
1497 		if (test_bit(WriteMostly, &rdev2->flags))
1498 			d->state |= (1<<MD_DISK_WRITEMOSTLY);
1499 		if (test_bit(FailFast, &rdev2->flags))
1500 			d->state |= (1<<MD_DISK_FAILFAST);
1501 	}
1502 	/* now set the "removed" and "faulty" bits on any missing devices */
1503 	for (i=0 ; i < mddev->raid_disks ; i++) {
1504 		mdp_disk_t *d = &sb->disks[i];
1505 		if (d->state == 0 && d->number == 0) {
1506 			d->number = i;
1507 			d->raid_disk = i;
1508 			d->state = (1<<MD_DISK_REMOVED);
1509 			d->state |= (1<<MD_DISK_FAULTY);
1510 			failed++;
1511 		}
1512 	}
1513 	sb->nr_disks = nr_disks;
1514 	sb->active_disks = active;
1515 	sb->working_disks = working;
1516 	sb->failed_disks = failed;
1517 	sb->spare_disks = spare;
1518 
1519 	sb->this_disk = sb->disks[rdev->desc_nr];
1520 	sb->sb_csum = calc_sb_csum(sb);
1521 }
1522 
1523 /*
1524  * rdev_size_change for 0.90.0
1525  */
1526 static unsigned long long
1527 super_90_rdev_size_change(struct md_rdev *rdev, sector_t num_sectors)
1528 {
1529 	if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
1530 		return 0; /* component must fit device */
1531 	if (rdev->mddev->bitmap_info.offset)
1532 		return 0; /* can't move bitmap */
1533 	rdev->sb_start = calc_dev_sboffset(rdev);
1534 	if (!num_sectors || num_sectors > rdev->sb_start)
1535 		num_sectors = rdev->sb_start;
1536 	/* Limit to 4TB as metadata cannot record more than that.
1537 	 * 4TB == 2^32 KB, or 2*2^32 sectors.
1538 	 */
1539 	if ((u64)num_sectors >= (2ULL << 32) && rdev->mddev->level >= 1)
1540 		num_sectors = (sector_t)(2ULL << 32) - 2;
1541 	do {
1542 		md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
1543 		       rdev->sb_page);
1544 	} while (md_super_wait(rdev->mddev) < 0);
1545 	return num_sectors;
1546 }
1547 
1548 static int
1549 super_90_allow_new_offset(struct md_rdev *rdev, unsigned long long new_offset)
1550 {
1551 	/* non-zero offset changes not possible with v0.90 */
1552 	return new_offset == 0;
1553 }
1554 
1555 /*
1556  * version 1 superblock
1557  */
1558 
1559 static __le32 calc_sb_1_csum(struct mdp_superblock_1 *sb)
1560 {
1561 	__le32 disk_csum;
1562 	u32 csum;
1563 	unsigned long long newcsum;
1564 	int size = 256 + le32_to_cpu(sb->max_dev)*2;
1565 	__le32 *isuper = (__le32*)sb;
1566 
1567 	disk_csum = sb->sb_csum;
1568 	sb->sb_csum = 0;
1569 	newcsum = 0;
1570 	for (; size >= 4; size -= 4)
1571 		newcsum += le32_to_cpu(*isuper++);
1572 
1573 	if (size == 2)
1574 		newcsum += le16_to_cpu(*(__le16*) isuper);
1575 
1576 	csum = (newcsum & 0xffffffff) + (newcsum >> 32);
1577 	sb->sb_csum = disk_csum;
1578 	return cpu_to_le32(csum);
1579 }
1580 
1581 static int super_1_load(struct md_rdev *rdev, struct md_rdev *refdev, int minor_version)
1582 {
1583 	struct mdp_superblock_1 *sb;
1584 	int ret;
1585 	sector_t sb_start;
1586 	sector_t sectors;
1587 	int bmask;
1588 	bool spare_disk = true;
1589 
1590 	/*
1591 	 * Calculate the position of the superblock in 512byte sectors.
1592 	 * It is always aligned to a 4K boundary and
1593 	 * depeding on minor_version, it can be:
1594 	 * 0: At least 8K, but less than 12K, from end of device
1595 	 * 1: At start of device
1596 	 * 2: 4K from start of device.
1597 	 */
1598 	switch(minor_version) {
1599 	case 0:
1600 		sb_start = bdev_nr_sectors(rdev->bdev) - 8 * 2;
1601 		sb_start &= ~(sector_t)(4*2-1);
1602 		break;
1603 	case 1:
1604 		sb_start = 0;
1605 		break;
1606 	case 2:
1607 		sb_start = 8;
1608 		break;
1609 	default:
1610 		return -EINVAL;
1611 	}
1612 	rdev->sb_start = sb_start;
1613 
1614 	/* superblock is rarely larger than 1K, but it can be larger,
1615 	 * and it is safe to read 4k, so we do that
1616 	 */
1617 	ret = read_disk_sb(rdev, 4096);
1618 	if (ret) return ret;
1619 
1620 	sb = page_address(rdev->sb_page);
1621 
1622 	if (sb->magic != cpu_to_le32(MD_SB_MAGIC) ||
1623 	    sb->major_version != cpu_to_le32(1) ||
1624 	    le32_to_cpu(sb->max_dev) > (4096-256)/2 ||
1625 	    le64_to_cpu(sb->super_offset) != rdev->sb_start ||
1626 	    (le32_to_cpu(sb->feature_map) & ~MD_FEATURE_ALL) != 0)
1627 		return -EINVAL;
1628 
1629 	if (calc_sb_1_csum(sb) != sb->sb_csum) {
1630 		pr_warn("md: invalid superblock checksum on %pg\n",
1631 			rdev->bdev);
1632 		return -EINVAL;
1633 	}
1634 	if (le64_to_cpu(sb->data_size) < 10) {
1635 		pr_warn("md: data_size too small on %pg\n",
1636 			rdev->bdev);
1637 		return -EINVAL;
1638 	}
1639 	if (sb->pad0 ||
1640 	    sb->pad3[0] ||
1641 	    memcmp(sb->pad3, sb->pad3+1, sizeof(sb->pad3) - sizeof(sb->pad3[1])))
1642 		/* Some padding is non-zero, might be a new feature */
1643 		return -EINVAL;
1644 
1645 	rdev->preferred_minor = 0xffff;
1646 	rdev->data_offset = le64_to_cpu(sb->data_offset);
1647 	rdev->new_data_offset = rdev->data_offset;
1648 	if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE) &&
1649 	    (le32_to_cpu(sb->feature_map) & MD_FEATURE_NEW_OFFSET))
1650 		rdev->new_data_offset += (s32)le32_to_cpu(sb->new_offset);
1651 	atomic_set(&rdev->corrected_errors, le32_to_cpu(sb->cnt_corrected_read));
1652 
1653 	rdev->sb_size = le32_to_cpu(sb->max_dev) * 2 + 256;
1654 	bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
1655 	if (rdev->sb_size & bmask)
1656 		rdev->sb_size = (rdev->sb_size | bmask) + 1;
1657 
1658 	if (minor_version
1659 	    && rdev->data_offset < sb_start + (rdev->sb_size/512))
1660 		return -EINVAL;
1661 	if (minor_version
1662 	    && rdev->new_data_offset < sb_start + (rdev->sb_size/512))
1663 		return -EINVAL;
1664 
1665 	if (sb->level == cpu_to_le32(LEVEL_MULTIPATH))
1666 		rdev->desc_nr = -1;
1667 	else
1668 		rdev->desc_nr = le32_to_cpu(sb->dev_number);
1669 
1670 	if (!rdev->bb_page) {
1671 		rdev->bb_page = alloc_page(GFP_KERNEL);
1672 		if (!rdev->bb_page)
1673 			return -ENOMEM;
1674 	}
1675 	if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BAD_BLOCKS) &&
1676 	    rdev->badblocks.count == 0) {
1677 		/* need to load the bad block list.
1678 		 * Currently we limit it to one page.
1679 		 */
1680 		s32 offset;
1681 		sector_t bb_sector;
1682 		__le64 *bbp;
1683 		int i;
1684 		int sectors = le16_to_cpu(sb->bblog_size);
1685 		if (sectors > (PAGE_SIZE / 512))
1686 			return -EINVAL;
1687 		offset = le32_to_cpu(sb->bblog_offset);
1688 		if (offset == 0)
1689 			return -EINVAL;
1690 		bb_sector = (long long)offset;
1691 		if (!sync_page_io(rdev, bb_sector, sectors << 9,
1692 				  rdev->bb_page, REQ_OP_READ, true))
1693 			return -EIO;
1694 		bbp = (__le64 *)page_address(rdev->bb_page);
1695 		rdev->badblocks.shift = sb->bblog_shift;
1696 		for (i = 0 ; i < (sectors << (9-3)) ; i++, bbp++) {
1697 			u64 bb = le64_to_cpu(*bbp);
1698 			int count = bb & (0x3ff);
1699 			u64 sector = bb >> 10;
1700 			sector <<= sb->bblog_shift;
1701 			count <<= sb->bblog_shift;
1702 			if (bb + 1 == 0)
1703 				break;
1704 			if (badblocks_set(&rdev->badblocks, sector, count, 1))
1705 				return -EINVAL;
1706 		}
1707 	} else if (sb->bblog_offset != 0)
1708 		rdev->badblocks.shift = 0;
1709 
1710 	if ((le32_to_cpu(sb->feature_map) &
1711 	    (MD_FEATURE_PPL | MD_FEATURE_MULTIPLE_PPLS))) {
1712 		rdev->ppl.offset = (__s16)le16_to_cpu(sb->ppl.offset);
1713 		rdev->ppl.size = le16_to_cpu(sb->ppl.size);
1714 		rdev->ppl.sector = rdev->sb_start + rdev->ppl.offset;
1715 	}
1716 
1717 	if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RAID0_LAYOUT) &&
1718 	    sb->level != 0)
1719 		return -EINVAL;
1720 
1721 	/* not spare disk, or LEVEL_MULTIPATH */
1722 	if (sb->level == cpu_to_le32(LEVEL_MULTIPATH) ||
1723 		(rdev->desc_nr >= 0 &&
1724 		rdev->desc_nr < le32_to_cpu(sb->max_dev) &&
1725 		(le16_to_cpu(sb->dev_roles[rdev->desc_nr]) < MD_DISK_ROLE_MAX ||
1726 		 le16_to_cpu(sb->dev_roles[rdev->desc_nr]) == MD_DISK_ROLE_JOURNAL)))
1727 		spare_disk = false;
1728 
1729 	if (!refdev) {
1730 		if (!spare_disk)
1731 			ret = 1;
1732 		else
1733 			ret = 0;
1734 	} else {
1735 		__u64 ev1, ev2;
1736 		struct mdp_superblock_1 *refsb = page_address(refdev->sb_page);
1737 
1738 		if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 ||
1739 		    sb->level != refsb->level ||
1740 		    sb->layout != refsb->layout ||
1741 		    sb->chunksize != refsb->chunksize) {
1742 			pr_warn("md: %pg has strangely different superblock to %pg\n",
1743 				rdev->bdev,
1744 				refdev->bdev);
1745 			return -EINVAL;
1746 		}
1747 		ev1 = le64_to_cpu(sb->events);
1748 		ev2 = le64_to_cpu(refsb->events);
1749 
1750 		if (!spare_disk && ev1 > ev2)
1751 			ret = 1;
1752 		else
1753 			ret = 0;
1754 	}
1755 	if (minor_version)
1756 		sectors = bdev_nr_sectors(rdev->bdev) - rdev->data_offset;
1757 	else
1758 		sectors = rdev->sb_start;
1759 	if (sectors < le64_to_cpu(sb->data_size))
1760 		return -EINVAL;
1761 	rdev->sectors = le64_to_cpu(sb->data_size);
1762 	return ret;
1763 }
1764 
1765 static int super_1_validate(struct mddev *mddev, struct md_rdev *rdev)
1766 {
1767 	struct mdp_superblock_1 *sb = page_address(rdev->sb_page);
1768 	__u64 ev1 = le64_to_cpu(sb->events);
1769 
1770 	rdev->raid_disk = -1;
1771 	clear_bit(Faulty, &rdev->flags);
1772 	clear_bit(In_sync, &rdev->flags);
1773 	clear_bit(Bitmap_sync, &rdev->flags);
1774 	clear_bit(WriteMostly, &rdev->flags);
1775 
1776 	if (mddev->raid_disks == 0) {
1777 		mddev->major_version = 1;
1778 		mddev->patch_version = 0;
1779 		mddev->external = 0;
1780 		mddev->chunk_sectors = le32_to_cpu(sb->chunksize);
1781 		mddev->ctime = le64_to_cpu(sb->ctime);
1782 		mddev->utime = le64_to_cpu(sb->utime);
1783 		mddev->level = le32_to_cpu(sb->level);
1784 		mddev->clevel[0] = 0;
1785 		mddev->layout = le32_to_cpu(sb->layout);
1786 		mddev->raid_disks = le32_to_cpu(sb->raid_disks);
1787 		mddev->dev_sectors = le64_to_cpu(sb->size);
1788 		mddev->events = ev1;
1789 		mddev->bitmap_info.offset = 0;
1790 		mddev->bitmap_info.space = 0;
1791 		/* Default location for bitmap is 1K after superblock
1792 		 * using 3K - total of 4K
1793 		 */
1794 		mddev->bitmap_info.default_offset = 1024 >> 9;
1795 		mddev->bitmap_info.default_space = (4096-1024) >> 9;
1796 		mddev->reshape_backwards = 0;
1797 
1798 		mddev->recovery_cp = le64_to_cpu(sb->resync_offset);
1799 		memcpy(mddev->uuid, sb->set_uuid, 16);
1800 
1801 		mddev->max_disks =  (4096-256)/2;
1802 
1803 		if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) &&
1804 		    mddev->bitmap_info.file == NULL) {
1805 			mddev->bitmap_info.offset =
1806 				(__s32)le32_to_cpu(sb->bitmap_offset);
1807 			/* Metadata doesn't record how much space is available.
1808 			 * For 1.0, we assume we can use up to the superblock
1809 			 * if before, else to 4K beyond superblock.
1810 			 * For others, assume no change is possible.
1811 			 */
1812 			if (mddev->minor_version > 0)
1813 				mddev->bitmap_info.space = 0;
1814 			else if (mddev->bitmap_info.offset > 0)
1815 				mddev->bitmap_info.space =
1816 					8 - mddev->bitmap_info.offset;
1817 			else
1818 				mddev->bitmap_info.space =
1819 					-mddev->bitmap_info.offset;
1820 		}
1821 
1822 		if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) {
1823 			mddev->reshape_position = le64_to_cpu(sb->reshape_position);
1824 			mddev->delta_disks = le32_to_cpu(sb->delta_disks);
1825 			mddev->new_level = le32_to_cpu(sb->new_level);
1826 			mddev->new_layout = le32_to_cpu(sb->new_layout);
1827 			mddev->new_chunk_sectors = le32_to_cpu(sb->new_chunk);
1828 			if (mddev->delta_disks < 0 ||
1829 			    (mddev->delta_disks == 0 &&
1830 			     (le32_to_cpu(sb->feature_map)
1831 			      & MD_FEATURE_RESHAPE_BACKWARDS)))
1832 				mddev->reshape_backwards = 1;
1833 		} else {
1834 			mddev->reshape_position = MaxSector;
1835 			mddev->delta_disks = 0;
1836 			mddev->new_level = mddev->level;
1837 			mddev->new_layout = mddev->layout;
1838 			mddev->new_chunk_sectors = mddev->chunk_sectors;
1839 		}
1840 
1841 		if (mddev->level == 0 &&
1842 		    !(le32_to_cpu(sb->feature_map) & MD_FEATURE_RAID0_LAYOUT))
1843 			mddev->layout = -1;
1844 
1845 		if (le32_to_cpu(sb->feature_map) & MD_FEATURE_JOURNAL)
1846 			set_bit(MD_HAS_JOURNAL, &mddev->flags);
1847 
1848 		if (le32_to_cpu(sb->feature_map) &
1849 		    (MD_FEATURE_PPL | MD_FEATURE_MULTIPLE_PPLS)) {
1850 			if (le32_to_cpu(sb->feature_map) &
1851 			    (MD_FEATURE_BITMAP_OFFSET | MD_FEATURE_JOURNAL))
1852 				return -EINVAL;
1853 			if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_PPL) &&
1854 			    (le32_to_cpu(sb->feature_map) &
1855 					    MD_FEATURE_MULTIPLE_PPLS))
1856 				return -EINVAL;
1857 			set_bit(MD_HAS_PPL, &mddev->flags);
1858 		}
1859 	} else if (mddev->pers == NULL) {
1860 		/* Insist of good event counter while assembling, except for
1861 		 * spares (which don't need an event count) */
1862 		++ev1;
1863 		if (rdev->desc_nr >= 0 &&
1864 		    rdev->desc_nr < le32_to_cpu(sb->max_dev) &&
1865 		    (le16_to_cpu(sb->dev_roles[rdev->desc_nr]) < MD_DISK_ROLE_MAX ||
1866 		     le16_to_cpu(sb->dev_roles[rdev->desc_nr]) == MD_DISK_ROLE_JOURNAL))
1867 			if (ev1 < mddev->events)
1868 				return -EINVAL;
1869 	} else if (mddev->bitmap) {
1870 		/* If adding to array with a bitmap, then we can accept an
1871 		 * older device, but not too old.
1872 		 */
1873 		if (ev1 < mddev->bitmap->events_cleared)
1874 			return 0;
1875 		if (ev1 < mddev->events)
1876 			set_bit(Bitmap_sync, &rdev->flags);
1877 	} else {
1878 		if (ev1 < mddev->events)
1879 			/* just a hot-add of a new device, leave raid_disk at -1 */
1880 			return 0;
1881 	}
1882 	if (mddev->level != LEVEL_MULTIPATH) {
1883 		int role;
1884 		if (rdev->desc_nr < 0 ||
1885 		    rdev->desc_nr >= le32_to_cpu(sb->max_dev)) {
1886 			role = MD_DISK_ROLE_SPARE;
1887 			rdev->desc_nr = -1;
1888 		} else
1889 			role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
1890 		switch(role) {
1891 		case MD_DISK_ROLE_SPARE: /* spare */
1892 			break;
1893 		case MD_DISK_ROLE_FAULTY: /* faulty */
1894 			set_bit(Faulty, &rdev->flags);
1895 			break;
1896 		case MD_DISK_ROLE_JOURNAL: /* journal device */
1897 			if (!(le32_to_cpu(sb->feature_map) & MD_FEATURE_JOURNAL)) {
1898 				/* journal device without journal feature */
1899 				pr_warn("md: journal device provided without journal feature, ignoring the device\n");
1900 				return -EINVAL;
1901 			}
1902 			set_bit(Journal, &rdev->flags);
1903 			rdev->journal_tail = le64_to_cpu(sb->journal_tail);
1904 			rdev->raid_disk = 0;
1905 			break;
1906 		default:
1907 			rdev->saved_raid_disk = role;
1908 			if ((le32_to_cpu(sb->feature_map) &
1909 			     MD_FEATURE_RECOVERY_OFFSET)) {
1910 				rdev->recovery_offset = le64_to_cpu(sb->recovery_offset);
1911 				if (!(le32_to_cpu(sb->feature_map) &
1912 				      MD_FEATURE_RECOVERY_BITMAP))
1913 					rdev->saved_raid_disk = -1;
1914 			} else {
1915 				/*
1916 				 * If the array is FROZEN, then the device can't
1917 				 * be in_sync with rest of array.
1918 				 */
1919 				if (!test_bit(MD_RECOVERY_FROZEN,
1920 					      &mddev->recovery))
1921 					set_bit(In_sync, &rdev->flags);
1922 			}
1923 			rdev->raid_disk = role;
1924 			break;
1925 		}
1926 		if (sb->devflags & WriteMostly1)
1927 			set_bit(WriteMostly, &rdev->flags);
1928 		if (sb->devflags & FailFast1)
1929 			set_bit(FailFast, &rdev->flags);
1930 		if (le32_to_cpu(sb->feature_map) & MD_FEATURE_REPLACEMENT)
1931 			set_bit(Replacement, &rdev->flags);
1932 	} else /* MULTIPATH are always insync */
1933 		set_bit(In_sync, &rdev->flags);
1934 
1935 	return 0;
1936 }
1937 
1938 static void super_1_sync(struct mddev *mddev, struct md_rdev *rdev)
1939 {
1940 	struct mdp_superblock_1 *sb;
1941 	struct md_rdev *rdev2;
1942 	int max_dev, i;
1943 	/* make rdev->sb match mddev and rdev data. */
1944 
1945 	sb = page_address(rdev->sb_page);
1946 
1947 	sb->feature_map = 0;
1948 	sb->pad0 = 0;
1949 	sb->recovery_offset = cpu_to_le64(0);
1950 	memset(sb->pad3, 0, sizeof(sb->pad3));
1951 
1952 	sb->utime = cpu_to_le64((__u64)mddev->utime);
1953 	sb->events = cpu_to_le64(mddev->events);
1954 	if (mddev->in_sync)
1955 		sb->resync_offset = cpu_to_le64(mddev->recovery_cp);
1956 	else if (test_bit(MD_JOURNAL_CLEAN, &mddev->flags))
1957 		sb->resync_offset = cpu_to_le64(MaxSector);
1958 	else
1959 		sb->resync_offset = cpu_to_le64(0);
1960 
1961 	sb->cnt_corrected_read = cpu_to_le32(atomic_read(&rdev->corrected_errors));
1962 
1963 	sb->raid_disks = cpu_to_le32(mddev->raid_disks);
1964 	sb->size = cpu_to_le64(mddev->dev_sectors);
1965 	sb->chunksize = cpu_to_le32(mddev->chunk_sectors);
1966 	sb->level = cpu_to_le32(mddev->level);
1967 	sb->layout = cpu_to_le32(mddev->layout);
1968 	if (test_bit(FailFast, &rdev->flags))
1969 		sb->devflags |= FailFast1;
1970 	else
1971 		sb->devflags &= ~FailFast1;
1972 
1973 	if (test_bit(WriteMostly, &rdev->flags))
1974 		sb->devflags |= WriteMostly1;
1975 	else
1976 		sb->devflags &= ~WriteMostly1;
1977 	sb->data_offset = cpu_to_le64(rdev->data_offset);
1978 	sb->data_size = cpu_to_le64(rdev->sectors);
1979 
1980 	if (mddev->bitmap && mddev->bitmap_info.file == NULL) {
1981 		sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_info.offset);
1982 		sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET);
1983 	}
1984 
1985 	if (rdev->raid_disk >= 0 && !test_bit(Journal, &rdev->flags) &&
1986 	    !test_bit(In_sync, &rdev->flags)) {
1987 		sb->feature_map |=
1988 			cpu_to_le32(MD_FEATURE_RECOVERY_OFFSET);
1989 		sb->recovery_offset =
1990 			cpu_to_le64(rdev->recovery_offset);
1991 		if (rdev->saved_raid_disk >= 0 && mddev->bitmap)
1992 			sb->feature_map |=
1993 				cpu_to_le32(MD_FEATURE_RECOVERY_BITMAP);
1994 	}
1995 	/* Note: recovery_offset and journal_tail share space  */
1996 	if (test_bit(Journal, &rdev->flags))
1997 		sb->journal_tail = cpu_to_le64(rdev->journal_tail);
1998 	if (test_bit(Replacement, &rdev->flags))
1999 		sb->feature_map |=
2000 			cpu_to_le32(MD_FEATURE_REPLACEMENT);
2001 
2002 	if (mddev->reshape_position != MaxSector) {
2003 		sb->feature_map |= cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE);
2004 		sb->reshape_position = cpu_to_le64(mddev->reshape_position);
2005 		sb->new_layout = cpu_to_le32(mddev->new_layout);
2006 		sb->delta_disks = cpu_to_le32(mddev->delta_disks);
2007 		sb->new_level = cpu_to_le32(mddev->new_level);
2008 		sb->new_chunk = cpu_to_le32(mddev->new_chunk_sectors);
2009 		if (mddev->delta_disks == 0 &&
2010 		    mddev->reshape_backwards)
2011 			sb->feature_map
2012 				|= cpu_to_le32(MD_FEATURE_RESHAPE_BACKWARDS);
2013 		if (rdev->new_data_offset != rdev->data_offset) {
2014 			sb->feature_map
2015 				|= cpu_to_le32(MD_FEATURE_NEW_OFFSET);
2016 			sb->new_offset = cpu_to_le32((__u32)(rdev->new_data_offset
2017 							     - rdev->data_offset));
2018 		}
2019 	}
2020 
2021 	if (mddev_is_clustered(mddev))
2022 		sb->feature_map |= cpu_to_le32(MD_FEATURE_CLUSTERED);
2023 
2024 	if (rdev->badblocks.count == 0)
2025 		/* Nothing to do for bad blocks*/ ;
2026 	else if (sb->bblog_offset == 0)
2027 		/* Cannot record bad blocks on this device */
2028 		md_error(mddev, rdev);
2029 	else {
2030 		struct badblocks *bb = &rdev->badblocks;
2031 		__le64 *bbp = (__le64 *)page_address(rdev->bb_page);
2032 		u64 *p = bb->page;
2033 		sb->feature_map |= cpu_to_le32(MD_FEATURE_BAD_BLOCKS);
2034 		if (bb->changed) {
2035 			unsigned seq;
2036 
2037 retry:
2038 			seq = read_seqbegin(&bb->lock);
2039 
2040 			memset(bbp, 0xff, PAGE_SIZE);
2041 
2042 			for (i = 0 ; i < bb->count ; i++) {
2043 				u64 internal_bb = p[i];
2044 				u64 store_bb = ((BB_OFFSET(internal_bb) << 10)
2045 						| BB_LEN(internal_bb));
2046 				bbp[i] = cpu_to_le64(store_bb);
2047 			}
2048 			bb->changed = 0;
2049 			if (read_seqretry(&bb->lock, seq))
2050 				goto retry;
2051 
2052 			bb->sector = (rdev->sb_start +
2053 				      (int)le32_to_cpu(sb->bblog_offset));
2054 			bb->size = le16_to_cpu(sb->bblog_size);
2055 		}
2056 	}
2057 
2058 	max_dev = 0;
2059 	rdev_for_each(rdev2, mddev)
2060 		if (rdev2->desc_nr+1 > max_dev)
2061 			max_dev = rdev2->desc_nr+1;
2062 
2063 	if (max_dev > le32_to_cpu(sb->max_dev)) {
2064 		int bmask;
2065 		sb->max_dev = cpu_to_le32(max_dev);
2066 		rdev->sb_size = max_dev * 2 + 256;
2067 		bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
2068 		if (rdev->sb_size & bmask)
2069 			rdev->sb_size = (rdev->sb_size | bmask) + 1;
2070 	} else
2071 		max_dev = le32_to_cpu(sb->max_dev);
2072 
2073 	for (i=0; i<max_dev;i++)
2074 		sb->dev_roles[i] = cpu_to_le16(MD_DISK_ROLE_SPARE);
2075 
2076 	if (test_bit(MD_HAS_JOURNAL, &mddev->flags))
2077 		sb->feature_map |= cpu_to_le32(MD_FEATURE_JOURNAL);
2078 
2079 	if (test_bit(MD_HAS_PPL, &mddev->flags)) {
2080 		if (test_bit(MD_HAS_MULTIPLE_PPLS, &mddev->flags))
2081 			sb->feature_map |=
2082 			    cpu_to_le32(MD_FEATURE_MULTIPLE_PPLS);
2083 		else
2084 			sb->feature_map |= cpu_to_le32(MD_FEATURE_PPL);
2085 		sb->ppl.offset = cpu_to_le16(rdev->ppl.offset);
2086 		sb->ppl.size = cpu_to_le16(rdev->ppl.size);
2087 	}
2088 
2089 	rdev_for_each(rdev2, mddev) {
2090 		i = rdev2->desc_nr;
2091 		if (test_bit(Faulty, &rdev2->flags))
2092 			sb->dev_roles[i] = cpu_to_le16(MD_DISK_ROLE_FAULTY);
2093 		else if (test_bit(In_sync, &rdev2->flags))
2094 			sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
2095 		else if (test_bit(Journal, &rdev2->flags))
2096 			sb->dev_roles[i] = cpu_to_le16(MD_DISK_ROLE_JOURNAL);
2097 		else if (rdev2->raid_disk >= 0)
2098 			sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
2099 		else
2100 			sb->dev_roles[i] = cpu_to_le16(MD_DISK_ROLE_SPARE);
2101 	}
2102 
2103 	sb->sb_csum = calc_sb_1_csum(sb);
2104 }
2105 
2106 static sector_t super_1_choose_bm_space(sector_t dev_size)
2107 {
2108 	sector_t bm_space;
2109 
2110 	/* if the device is bigger than 8Gig, save 64k for bitmap
2111 	 * usage, if bigger than 200Gig, save 128k
2112 	 */
2113 	if (dev_size < 64*2)
2114 		bm_space = 0;
2115 	else if (dev_size - 64*2 >= 200*1024*1024*2)
2116 		bm_space = 128*2;
2117 	else if (dev_size - 4*2 > 8*1024*1024*2)
2118 		bm_space = 64*2;
2119 	else
2120 		bm_space = 4*2;
2121 	return bm_space;
2122 }
2123 
2124 static unsigned long long
2125 super_1_rdev_size_change(struct md_rdev *rdev, sector_t num_sectors)
2126 {
2127 	struct mdp_superblock_1 *sb;
2128 	sector_t max_sectors;
2129 	if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
2130 		return 0; /* component must fit device */
2131 	if (rdev->data_offset != rdev->new_data_offset)
2132 		return 0; /* too confusing */
2133 	if (rdev->sb_start < rdev->data_offset) {
2134 		/* minor versions 1 and 2; superblock before data */
2135 		max_sectors = bdev_nr_sectors(rdev->bdev) - rdev->data_offset;
2136 		if (!num_sectors || num_sectors > max_sectors)
2137 			num_sectors = max_sectors;
2138 	} else if (rdev->mddev->bitmap_info.offset) {
2139 		/* minor version 0 with bitmap we can't move */
2140 		return 0;
2141 	} else {
2142 		/* minor version 0; superblock after data */
2143 		sector_t sb_start, bm_space;
2144 		sector_t dev_size = bdev_nr_sectors(rdev->bdev);
2145 
2146 		/* 8K is for superblock */
2147 		sb_start = dev_size - 8*2;
2148 		sb_start &= ~(sector_t)(4*2 - 1);
2149 
2150 		bm_space = super_1_choose_bm_space(dev_size);
2151 
2152 		/* Space that can be used to store date needs to decrease
2153 		 * superblock bitmap space and bad block space(4K)
2154 		 */
2155 		max_sectors = sb_start - bm_space - 4*2;
2156 
2157 		if (!num_sectors || num_sectors > max_sectors)
2158 			num_sectors = max_sectors;
2159 		rdev->sb_start = sb_start;
2160 	}
2161 	sb = page_address(rdev->sb_page);
2162 	sb->data_size = cpu_to_le64(num_sectors);
2163 	sb->super_offset = cpu_to_le64(rdev->sb_start);
2164 	sb->sb_csum = calc_sb_1_csum(sb);
2165 	do {
2166 		md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
2167 			       rdev->sb_page);
2168 	} while (md_super_wait(rdev->mddev) < 0);
2169 	return num_sectors;
2170 
2171 }
2172 
2173 static int
2174 super_1_allow_new_offset(struct md_rdev *rdev,
2175 			 unsigned long long new_offset)
2176 {
2177 	/* All necessary checks on new >= old have been done */
2178 	struct bitmap *bitmap;
2179 	if (new_offset >= rdev->data_offset)
2180 		return 1;
2181 
2182 	/* with 1.0 metadata, there is no metadata to tread on
2183 	 * so we can always move back */
2184 	if (rdev->mddev->minor_version == 0)
2185 		return 1;
2186 
2187 	/* otherwise we must be sure not to step on
2188 	 * any metadata, so stay:
2189 	 * 36K beyond start of superblock
2190 	 * beyond end of badblocks
2191 	 * beyond write-intent bitmap
2192 	 */
2193 	if (rdev->sb_start + (32+4)*2 > new_offset)
2194 		return 0;
2195 	bitmap = rdev->mddev->bitmap;
2196 	if (bitmap && !rdev->mddev->bitmap_info.file &&
2197 	    rdev->sb_start + rdev->mddev->bitmap_info.offset +
2198 	    bitmap->storage.file_pages * (PAGE_SIZE>>9) > new_offset)
2199 		return 0;
2200 	if (rdev->badblocks.sector + rdev->badblocks.size > new_offset)
2201 		return 0;
2202 
2203 	return 1;
2204 }
2205 
2206 static struct super_type super_types[] = {
2207 	[0] = {
2208 		.name	= "0.90.0",
2209 		.owner	= THIS_MODULE,
2210 		.load_super	    = super_90_load,
2211 		.validate_super	    = super_90_validate,
2212 		.sync_super	    = super_90_sync,
2213 		.rdev_size_change   = super_90_rdev_size_change,
2214 		.allow_new_offset   = super_90_allow_new_offset,
2215 	},
2216 	[1] = {
2217 		.name	= "md-1",
2218 		.owner	= THIS_MODULE,
2219 		.load_super	    = super_1_load,
2220 		.validate_super	    = super_1_validate,
2221 		.sync_super	    = super_1_sync,
2222 		.rdev_size_change   = super_1_rdev_size_change,
2223 		.allow_new_offset   = super_1_allow_new_offset,
2224 	},
2225 };
2226 
2227 static void sync_super(struct mddev *mddev, struct md_rdev *rdev)
2228 {
2229 	if (mddev->sync_super) {
2230 		mddev->sync_super(mddev, rdev);
2231 		return;
2232 	}
2233 
2234 	BUG_ON(mddev->major_version >= ARRAY_SIZE(super_types));
2235 
2236 	super_types[mddev->major_version].sync_super(mddev, rdev);
2237 }
2238 
2239 static int match_mddev_units(struct mddev *mddev1, struct mddev *mddev2)
2240 {
2241 	struct md_rdev *rdev, *rdev2;
2242 
2243 	rcu_read_lock();
2244 	rdev_for_each_rcu(rdev, mddev1) {
2245 		if (test_bit(Faulty, &rdev->flags) ||
2246 		    test_bit(Journal, &rdev->flags) ||
2247 		    rdev->raid_disk == -1)
2248 			continue;
2249 		rdev_for_each_rcu(rdev2, mddev2) {
2250 			if (test_bit(Faulty, &rdev2->flags) ||
2251 			    test_bit(Journal, &rdev2->flags) ||
2252 			    rdev2->raid_disk == -1)
2253 				continue;
2254 			if (rdev->bdev->bd_disk == rdev2->bdev->bd_disk) {
2255 				rcu_read_unlock();
2256 				return 1;
2257 			}
2258 		}
2259 	}
2260 	rcu_read_unlock();
2261 	return 0;
2262 }
2263 
2264 static LIST_HEAD(pending_raid_disks);
2265 
2266 /*
2267  * Try to register data integrity profile for an mddev
2268  *
2269  * This is called when an array is started and after a disk has been kicked
2270  * from the array. It only succeeds if all working and active component devices
2271  * are integrity capable with matching profiles.
2272  */
2273 int md_integrity_register(struct mddev *mddev)
2274 {
2275 	struct md_rdev *rdev, *reference = NULL;
2276 
2277 	if (list_empty(&mddev->disks))
2278 		return 0; /* nothing to do */
2279 	if (!mddev->gendisk || blk_get_integrity(mddev->gendisk))
2280 		return 0; /* shouldn't register, or already is */
2281 	rdev_for_each(rdev, mddev) {
2282 		/* skip spares and non-functional disks */
2283 		if (test_bit(Faulty, &rdev->flags))
2284 			continue;
2285 		if (rdev->raid_disk < 0)
2286 			continue;
2287 		if (!reference) {
2288 			/* Use the first rdev as the reference */
2289 			reference = rdev;
2290 			continue;
2291 		}
2292 		/* does this rdev's profile match the reference profile? */
2293 		if (blk_integrity_compare(reference->bdev->bd_disk,
2294 				rdev->bdev->bd_disk) < 0)
2295 			return -EINVAL;
2296 	}
2297 	if (!reference || !bdev_get_integrity(reference->bdev))
2298 		return 0;
2299 	/*
2300 	 * All component devices are integrity capable and have matching
2301 	 * profiles, register the common profile for the md device.
2302 	 */
2303 	blk_integrity_register(mddev->gendisk,
2304 			       bdev_get_integrity(reference->bdev));
2305 
2306 	pr_debug("md: data integrity enabled on %s\n", mdname(mddev));
2307 	if (bioset_integrity_create(&mddev->bio_set, BIO_POOL_SIZE) ||
2308 	    (mddev->level != 1 && mddev->level != 10 &&
2309 	     bioset_integrity_create(&mddev->io_clone_set, BIO_POOL_SIZE))) {
2310 		/*
2311 		 * No need to handle the failure of bioset_integrity_create,
2312 		 * because the function is called by md_run() -> pers->run(),
2313 		 * md_run calls bioset_exit -> bioset_integrity_free in case
2314 		 * of failure case.
2315 		 */
2316 		pr_err("md: failed to create integrity pool for %s\n",
2317 		       mdname(mddev));
2318 		return -EINVAL;
2319 	}
2320 	return 0;
2321 }
2322 EXPORT_SYMBOL(md_integrity_register);
2323 
2324 /*
2325  * Attempt to add an rdev, but only if it is consistent with the current
2326  * integrity profile
2327  */
2328 int md_integrity_add_rdev(struct md_rdev *rdev, struct mddev *mddev)
2329 {
2330 	struct blk_integrity *bi_mddev;
2331 
2332 	if (!mddev->gendisk)
2333 		return 0;
2334 
2335 	bi_mddev = blk_get_integrity(mddev->gendisk);
2336 
2337 	if (!bi_mddev) /* nothing to do */
2338 		return 0;
2339 
2340 	if (blk_integrity_compare(mddev->gendisk, rdev->bdev->bd_disk) != 0) {
2341 		pr_err("%s: incompatible integrity profile for %pg\n",
2342 		       mdname(mddev), rdev->bdev);
2343 		return -ENXIO;
2344 	}
2345 
2346 	return 0;
2347 }
2348 EXPORT_SYMBOL(md_integrity_add_rdev);
2349 
2350 static bool rdev_read_only(struct md_rdev *rdev)
2351 {
2352 	return bdev_read_only(rdev->bdev) ||
2353 		(rdev->meta_bdev && bdev_read_only(rdev->meta_bdev));
2354 }
2355 
2356 static int bind_rdev_to_array(struct md_rdev *rdev, struct mddev *mddev)
2357 {
2358 	char b[BDEVNAME_SIZE];
2359 	int err;
2360 
2361 	/* prevent duplicates */
2362 	if (find_rdev(mddev, rdev->bdev->bd_dev))
2363 		return -EEXIST;
2364 
2365 	if (rdev_read_only(rdev) && mddev->pers)
2366 		return -EROFS;
2367 
2368 	/* make sure rdev->sectors exceeds mddev->dev_sectors */
2369 	if (!test_bit(Journal, &rdev->flags) &&
2370 	    rdev->sectors &&
2371 	    (mddev->dev_sectors == 0 || rdev->sectors < mddev->dev_sectors)) {
2372 		if (mddev->pers) {
2373 			/* Cannot change size, so fail
2374 			 * If mddev->level <= 0, then we don't care
2375 			 * about aligning sizes (e.g. linear)
2376 			 */
2377 			if (mddev->level > 0)
2378 				return -ENOSPC;
2379 		} else
2380 			mddev->dev_sectors = rdev->sectors;
2381 	}
2382 
2383 	/* Verify rdev->desc_nr is unique.
2384 	 * If it is -1, assign a free number, else
2385 	 * check number is not in use
2386 	 */
2387 	rcu_read_lock();
2388 	if (rdev->desc_nr < 0) {
2389 		int choice = 0;
2390 		if (mddev->pers)
2391 			choice = mddev->raid_disks;
2392 		while (md_find_rdev_nr_rcu(mddev, choice))
2393 			choice++;
2394 		rdev->desc_nr = choice;
2395 	} else {
2396 		if (md_find_rdev_nr_rcu(mddev, rdev->desc_nr)) {
2397 			rcu_read_unlock();
2398 			return -EBUSY;
2399 		}
2400 	}
2401 	rcu_read_unlock();
2402 	if (!test_bit(Journal, &rdev->flags) &&
2403 	    mddev->max_disks && rdev->desc_nr >= mddev->max_disks) {
2404 		pr_warn("md: %s: array is limited to %d devices\n",
2405 			mdname(mddev), mddev->max_disks);
2406 		return -EBUSY;
2407 	}
2408 	snprintf(b, sizeof(b), "%pg", rdev->bdev);
2409 	strreplace(b, '/', '!');
2410 
2411 	rdev->mddev = mddev;
2412 	pr_debug("md: bind<%s>\n", b);
2413 
2414 	if (mddev->raid_disks)
2415 		mddev_create_serial_pool(mddev, rdev, false);
2416 
2417 	if ((err = kobject_add(&rdev->kobj, &mddev->kobj, "dev-%s", b)))
2418 		goto fail;
2419 
2420 	/* failure here is OK */
2421 	err = sysfs_create_link(&rdev->kobj, bdev_kobj(rdev->bdev), "block");
2422 	rdev->sysfs_state = sysfs_get_dirent_safe(rdev->kobj.sd, "state");
2423 	rdev->sysfs_unack_badblocks =
2424 		sysfs_get_dirent_safe(rdev->kobj.sd, "unacknowledged_bad_blocks");
2425 	rdev->sysfs_badblocks =
2426 		sysfs_get_dirent_safe(rdev->kobj.sd, "bad_blocks");
2427 
2428 	list_add_rcu(&rdev->same_set, &mddev->disks);
2429 	bd_link_disk_holder(rdev->bdev, mddev->gendisk);
2430 
2431 	/* May as well allow recovery to be retried once */
2432 	mddev->recovery_disabled++;
2433 
2434 	return 0;
2435 
2436  fail:
2437 	pr_warn("md: failed to register dev-%s for %s\n",
2438 		b, mdname(mddev));
2439 	return err;
2440 }
2441 
2442 void md_autodetect_dev(dev_t dev);
2443 
2444 /* just for claiming the bdev */
2445 static struct md_rdev claim_rdev;
2446 
2447 static void export_rdev(struct md_rdev *rdev, struct mddev *mddev)
2448 {
2449 	pr_debug("md: export_rdev(%pg)\n", rdev->bdev);
2450 	md_rdev_clear(rdev);
2451 #ifndef MODULE
2452 	if (test_bit(AutoDetected, &rdev->flags))
2453 		md_autodetect_dev(rdev->bdev->bd_dev);
2454 #endif
2455 	blkdev_put(rdev->bdev, mddev->external ? &claim_rdev : rdev);
2456 	rdev->bdev = NULL;
2457 	kobject_put(&rdev->kobj);
2458 }
2459 
2460 static void md_kick_rdev_from_array(struct md_rdev *rdev)
2461 {
2462 	struct mddev *mddev = rdev->mddev;
2463 
2464 	bd_unlink_disk_holder(rdev->bdev, rdev->mddev->gendisk);
2465 	list_del_rcu(&rdev->same_set);
2466 	pr_debug("md: unbind<%pg>\n", rdev->bdev);
2467 	mddev_destroy_serial_pool(rdev->mddev, rdev, false);
2468 	rdev->mddev = NULL;
2469 	sysfs_remove_link(&rdev->kobj, "block");
2470 	sysfs_put(rdev->sysfs_state);
2471 	sysfs_put(rdev->sysfs_unack_badblocks);
2472 	sysfs_put(rdev->sysfs_badblocks);
2473 	rdev->sysfs_state = NULL;
2474 	rdev->sysfs_unack_badblocks = NULL;
2475 	rdev->sysfs_badblocks = NULL;
2476 	rdev->badblocks.count = 0;
2477 
2478 	synchronize_rcu();
2479 
2480 	/*
2481 	 * kobject_del() will wait for all in progress writers to be done, where
2482 	 * reconfig_mutex is held, hence it can't be called under
2483 	 * reconfig_mutex and it's delayed to mddev_unlock().
2484 	 */
2485 	list_add(&rdev->same_set, &mddev->deleting);
2486 }
2487 
2488 static void export_array(struct mddev *mddev)
2489 {
2490 	struct md_rdev *rdev;
2491 
2492 	while (!list_empty(&mddev->disks)) {
2493 		rdev = list_first_entry(&mddev->disks, struct md_rdev,
2494 					same_set);
2495 		md_kick_rdev_from_array(rdev);
2496 	}
2497 	mddev->raid_disks = 0;
2498 	mddev->major_version = 0;
2499 }
2500 
2501 static bool set_in_sync(struct mddev *mddev)
2502 {
2503 	lockdep_assert_held(&mddev->lock);
2504 	if (!mddev->in_sync) {
2505 		mddev->sync_checkers++;
2506 		spin_unlock(&mddev->lock);
2507 		percpu_ref_switch_to_atomic_sync(&mddev->writes_pending);
2508 		spin_lock(&mddev->lock);
2509 		if (!mddev->in_sync &&
2510 		    percpu_ref_is_zero(&mddev->writes_pending)) {
2511 			mddev->in_sync = 1;
2512 			/*
2513 			 * Ensure ->in_sync is visible before we clear
2514 			 * ->sync_checkers.
2515 			 */
2516 			smp_mb();
2517 			set_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags);
2518 			sysfs_notify_dirent_safe(mddev->sysfs_state);
2519 		}
2520 		if (--mddev->sync_checkers == 0)
2521 			percpu_ref_switch_to_percpu(&mddev->writes_pending);
2522 	}
2523 	if (mddev->safemode == 1)
2524 		mddev->safemode = 0;
2525 	return mddev->in_sync;
2526 }
2527 
2528 static void sync_sbs(struct mddev *mddev, int nospares)
2529 {
2530 	/* Update each superblock (in-memory image), but
2531 	 * if we are allowed to, skip spares which already
2532 	 * have the right event counter, or have one earlier
2533 	 * (which would mean they aren't being marked as dirty
2534 	 * with the rest of the array)
2535 	 */
2536 	struct md_rdev *rdev;
2537 	rdev_for_each(rdev, mddev) {
2538 		if (rdev->sb_events == mddev->events ||
2539 		    (nospares &&
2540 		     rdev->raid_disk < 0 &&
2541 		     rdev->sb_events+1 == mddev->events)) {
2542 			/* Don't update this superblock */
2543 			rdev->sb_loaded = 2;
2544 		} else {
2545 			sync_super(mddev, rdev);
2546 			rdev->sb_loaded = 1;
2547 		}
2548 	}
2549 }
2550 
2551 static bool does_sb_need_changing(struct mddev *mddev)
2552 {
2553 	struct md_rdev *rdev = NULL, *iter;
2554 	struct mdp_superblock_1 *sb;
2555 	int role;
2556 
2557 	/* Find a good rdev */
2558 	rdev_for_each(iter, mddev)
2559 		if ((iter->raid_disk >= 0) && !test_bit(Faulty, &iter->flags)) {
2560 			rdev = iter;
2561 			break;
2562 		}
2563 
2564 	/* No good device found. */
2565 	if (!rdev)
2566 		return false;
2567 
2568 	sb = page_address(rdev->sb_page);
2569 	/* Check if a device has become faulty or a spare become active */
2570 	rdev_for_each(rdev, mddev) {
2571 		role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
2572 		/* Device activated? */
2573 		if (role == MD_DISK_ROLE_SPARE && rdev->raid_disk >= 0 &&
2574 		    !test_bit(Faulty, &rdev->flags))
2575 			return true;
2576 		/* Device turned faulty? */
2577 		if (test_bit(Faulty, &rdev->flags) && (role < MD_DISK_ROLE_MAX))
2578 			return true;
2579 	}
2580 
2581 	/* Check if any mddev parameters have changed */
2582 	if ((mddev->dev_sectors != le64_to_cpu(sb->size)) ||
2583 	    (mddev->reshape_position != le64_to_cpu(sb->reshape_position)) ||
2584 	    (mddev->layout != le32_to_cpu(sb->layout)) ||
2585 	    (mddev->raid_disks != le32_to_cpu(sb->raid_disks)) ||
2586 	    (mddev->chunk_sectors != le32_to_cpu(sb->chunksize)))
2587 		return true;
2588 
2589 	return false;
2590 }
2591 
2592 void md_update_sb(struct mddev *mddev, int force_change)
2593 {
2594 	struct md_rdev *rdev;
2595 	int sync_req;
2596 	int nospares = 0;
2597 	int any_badblocks_changed = 0;
2598 	int ret = -1;
2599 
2600 	if (!md_is_rdwr(mddev)) {
2601 		if (force_change)
2602 			set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
2603 		return;
2604 	}
2605 
2606 repeat:
2607 	if (mddev_is_clustered(mddev)) {
2608 		if (test_and_clear_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags))
2609 			force_change = 1;
2610 		if (test_and_clear_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags))
2611 			nospares = 1;
2612 		ret = md_cluster_ops->metadata_update_start(mddev);
2613 		/* Has someone else has updated the sb */
2614 		if (!does_sb_need_changing(mddev)) {
2615 			if (ret == 0)
2616 				md_cluster_ops->metadata_update_cancel(mddev);
2617 			bit_clear_unless(&mddev->sb_flags, BIT(MD_SB_CHANGE_PENDING),
2618 							 BIT(MD_SB_CHANGE_DEVS) |
2619 							 BIT(MD_SB_CHANGE_CLEAN));
2620 			return;
2621 		}
2622 	}
2623 
2624 	/*
2625 	 * First make sure individual recovery_offsets are correct
2626 	 * curr_resync_completed can only be used during recovery.
2627 	 * During reshape/resync it might use array-addresses rather
2628 	 * that device addresses.
2629 	 */
2630 	rdev_for_each(rdev, mddev) {
2631 		if (rdev->raid_disk >= 0 &&
2632 		    mddev->delta_disks >= 0 &&
2633 		    test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
2634 		    test_bit(MD_RECOVERY_RECOVER, &mddev->recovery) &&
2635 		    !test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
2636 		    !test_bit(Journal, &rdev->flags) &&
2637 		    !test_bit(In_sync, &rdev->flags) &&
2638 		    mddev->curr_resync_completed > rdev->recovery_offset)
2639 				rdev->recovery_offset = mddev->curr_resync_completed;
2640 
2641 	}
2642 	if (!mddev->persistent) {
2643 		clear_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags);
2644 		clear_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
2645 		if (!mddev->external) {
2646 			clear_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags);
2647 			rdev_for_each(rdev, mddev) {
2648 				if (rdev->badblocks.changed) {
2649 					rdev->badblocks.changed = 0;
2650 					ack_all_badblocks(&rdev->badblocks);
2651 					md_error(mddev, rdev);
2652 				}
2653 				clear_bit(Blocked, &rdev->flags);
2654 				clear_bit(BlockedBadBlocks, &rdev->flags);
2655 				wake_up(&rdev->blocked_wait);
2656 			}
2657 		}
2658 		wake_up(&mddev->sb_wait);
2659 		return;
2660 	}
2661 
2662 	spin_lock(&mddev->lock);
2663 
2664 	mddev->utime = ktime_get_real_seconds();
2665 
2666 	if (test_and_clear_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags))
2667 		force_change = 1;
2668 	if (test_and_clear_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags))
2669 		/* just a clean<-> dirty transition, possibly leave spares alone,
2670 		 * though if events isn't the right even/odd, we will have to do
2671 		 * spares after all
2672 		 */
2673 		nospares = 1;
2674 	if (force_change)
2675 		nospares = 0;
2676 	if (mddev->degraded)
2677 		/* If the array is degraded, then skipping spares is both
2678 		 * dangerous and fairly pointless.
2679 		 * Dangerous because a device that was removed from the array
2680 		 * might have a event_count that still looks up-to-date,
2681 		 * so it can be re-added without a resync.
2682 		 * Pointless because if there are any spares to skip,
2683 		 * then a recovery will happen and soon that array won't
2684 		 * be degraded any more and the spare can go back to sleep then.
2685 		 */
2686 		nospares = 0;
2687 
2688 	sync_req = mddev->in_sync;
2689 
2690 	/* If this is just a dirty<->clean transition, and the array is clean
2691 	 * and 'events' is odd, we can roll back to the previous clean state */
2692 	if (nospares
2693 	    && (mddev->in_sync && mddev->recovery_cp == MaxSector)
2694 	    && mddev->can_decrease_events
2695 	    && mddev->events != 1) {
2696 		mddev->events--;
2697 		mddev->can_decrease_events = 0;
2698 	} else {
2699 		/* otherwise we have to go forward and ... */
2700 		mddev->events ++;
2701 		mddev->can_decrease_events = nospares;
2702 	}
2703 
2704 	/*
2705 	 * This 64-bit counter should never wrap.
2706 	 * Either we are in around ~1 trillion A.C., assuming
2707 	 * 1 reboot per second, or we have a bug...
2708 	 */
2709 	WARN_ON(mddev->events == 0);
2710 
2711 	rdev_for_each(rdev, mddev) {
2712 		if (rdev->badblocks.changed)
2713 			any_badblocks_changed++;
2714 		if (test_bit(Faulty, &rdev->flags))
2715 			set_bit(FaultRecorded, &rdev->flags);
2716 	}
2717 
2718 	sync_sbs(mddev, nospares);
2719 	spin_unlock(&mddev->lock);
2720 
2721 	pr_debug("md: updating %s RAID superblock on device (in sync %d)\n",
2722 		 mdname(mddev), mddev->in_sync);
2723 
2724 	if (mddev->queue)
2725 		blk_add_trace_msg(mddev->queue, "md md_update_sb");
2726 rewrite:
2727 	md_bitmap_update_sb(mddev->bitmap);
2728 	rdev_for_each(rdev, mddev) {
2729 		if (rdev->sb_loaded != 1)
2730 			continue; /* no noise on spare devices */
2731 
2732 		if (!test_bit(Faulty, &rdev->flags)) {
2733 			md_super_write(mddev,rdev,
2734 				       rdev->sb_start, rdev->sb_size,
2735 				       rdev->sb_page);
2736 			pr_debug("md: (write) %pg's sb offset: %llu\n",
2737 				 rdev->bdev,
2738 				 (unsigned long long)rdev->sb_start);
2739 			rdev->sb_events = mddev->events;
2740 			if (rdev->badblocks.size) {
2741 				md_super_write(mddev, rdev,
2742 					       rdev->badblocks.sector,
2743 					       rdev->badblocks.size << 9,
2744 					       rdev->bb_page);
2745 				rdev->badblocks.size = 0;
2746 			}
2747 
2748 		} else
2749 			pr_debug("md: %pg (skipping faulty)\n",
2750 				 rdev->bdev);
2751 
2752 		if (mddev->level == LEVEL_MULTIPATH)
2753 			/* only need to write one superblock... */
2754 			break;
2755 	}
2756 	if (md_super_wait(mddev) < 0)
2757 		goto rewrite;
2758 	/* if there was a failure, MD_SB_CHANGE_DEVS was set, and we re-write super */
2759 
2760 	if (mddev_is_clustered(mddev) && ret == 0)
2761 		md_cluster_ops->metadata_update_finish(mddev);
2762 
2763 	if (mddev->in_sync != sync_req ||
2764 	    !bit_clear_unless(&mddev->sb_flags, BIT(MD_SB_CHANGE_PENDING),
2765 			       BIT(MD_SB_CHANGE_DEVS) | BIT(MD_SB_CHANGE_CLEAN)))
2766 		/* have to write it out again */
2767 		goto repeat;
2768 	wake_up(&mddev->sb_wait);
2769 	if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
2770 		sysfs_notify_dirent_safe(mddev->sysfs_completed);
2771 
2772 	rdev_for_each(rdev, mddev) {
2773 		if (test_and_clear_bit(FaultRecorded, &rdev->flags))
2774 			clear_bit(Blocked, &rdev->flags);
2775 
2776 		if (any_badblocks_changed)
2777 			ack_all_badblocks(&rdev->badblocks);
2778 		clear_bit(BlockedBadBlocks, &rdev->flags);
2779 		wake_up(&rdev->blocked_wait);
2780 	}
2781 }
2782 EXPORT_SYMBOL(md_update_sb);
2783 
2784 static int add_bound_rdev(struct md_rdev *rdev)
2785 {
2786 	struct mddev *mddev = rdev->mddev;
2787 	int err = 0;
2788 	bool add_journal = test_bit(Journal, &rdev->flags);
2789 
2790 	if (!mddev->pers->hot_remove_disk || add_journal) {
2791 		/* If there is hot_add_disk but no hot_remove_disk
2792 		 * then added disks for geometry changes,
2793 		 * and should be added immediately.
2794 		 */
2795 		super_types[mddev->major_version].
2796 			validate_super(mddev, rdev);
2797 		if (add_journal)
2798 			mddev_suspend(mddev);
2799 		err = mddev->pers->hot_add_disk(mddev, rdev);
2800 		if (add_journal)
2801 			mddev_resume(mddev);
2802 		if (err) {
2803 			md_kick_rdev_from_array(rdev);
2804 			return err;
2805 		}
2806 	}
2807 	sysfs_notify_dirent_safe(rdev->sysfs_state);
2808 
2809 	set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
2810 	if (mddev->degraded)
2811 		set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
2812 	set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2813 	md_new_event();
2814 	md_wakeup_thread(mddev->thread);
2815 	return 0;
2816 }
2817 
2818 /* words written to sysfs files may, or may not, be \n terminated.
2819  * We want to accept with case. For this we use cmd_match.
2820  */
2821 static int cmd_match(const char *cmd, const char *str)
2822 {
2823 	/* See if cmd, written into a sysfs file, matches
2824 	 * str.  They must either be the same, or cmd can
2825 	 * have a trailing newline
2826 	 */
2827 	while (*cmd && *str && *cmd == *str) {
2828 		cmd++;
2829 		str++;
2830 	}
2831 	if (*cmd == '\n')
2832 		cmd++;
2833 	if (*str || *cmd)
2834 		return 0;
2835 	return 1;
2836 }
2837 
2838 struct rdev_sysfs_entry {
2839 	struct attribute attr;
2840 	ssize_t (*show)(struct md_rdev *, char *);
2841 	ssize_t (*store)(struct md_rdev *, const char *, size_t);
2842 };
2843 
2844 static ssize_t
2845 state_show(struct md_rdev *rdev, char *page)
2846 {
2847 	char *sep = ",";
2848 	size_t len = 0;
2849 	unsigned long flags = READ_ONCE(rdev->flags);
2850 
2851 	if (test_bit(Faulty, &flags) ||
2852 	    (!test_bit(ExternalBbl, &flags) &&
2853 	    rdev->badblocks.unacked_exist))
2854 		len += sprintf(page+len, "faulty%s", sep);
2855 	if (test_bit(In_sync, &flags))
2856 		len += sprintf(page+len, "in_sync%s", sep);
2857 	if (test_bit(Journal, &flags))
2858 		len += sprintf(page+len, "journal%s", sep);
2859 	if (test_bit(WriteMostly, &flags))
2860 		len += sprintf(page+len, "write_mostly%s", sep);
2861 	if (test_bit(Blocked, &flags) ||
2862 	    (rdev->badblocks.unacked_exist
2863 	     && !test_bit(Faulty, &flags)))
2864 		len += sprintf(page+len, "blocked%s", sep);
2865 	if (!test_bit(Faulty, &flags) &&
2866 	    !test_bit(Journal, &flags) &&
2867 	    !test_bit(In_sync, &flags))
2868 		len += sprintf(page+len, "spare%s", sep);
2869 	if (test_bit(WriteErrorSeen, &flags))
2870 		len += sprintf(page+len, "write_error%s", sep);
2871 	if (test_bit(WantReplacement, &flags))
2872 		len += sprintf(page+len, "want_replacement%s", sep);
2873 	if (test_bit(Replacement, &flags))
2874 		len += sprintf(page+len, "replacement%s", sep);
2875 	if (test_bit(ExternalBbl, &flags))
2876 		len += sprintf(page+len, "external_bbl%s", sep);
2877 	if (test_bit(FailFast, &flags))
2878 		len += sprintf(page+len, "failfast%s", sep);
2879 
2880 	if (len)
2881 		len -= strlen(sep);
2882 
2883 	return len+sprintf(page+len, "\n");
2884 }
2885 
2886 static ssize_t
2887 state_store(struct md_rdev *rdev, const char *buf, size_t len)
2888 {
2889 	/* can write
2890 	 *  faulty  - simulates an error
2891 	 *  remove  - disconnects the device
2892 	 *  writemostly - sets write_mostly
2893 	 *  -writemostly - clears write_mostly
2894 	 *  blocked - sets the Blocked flags
2895 	 *  -blocked - clears the Blocked and possibly simulates an error
2896 	 *  insync - sets Insync providing device isn't active
2897 	 *  -insync - clear Insync for a device with a slot assigned,
2898 	 *            so that it gets rebuilt based on bitmap
2899 	 *  write_error - sets WriteErrorSeen
2900 	 *  -write_error - clears WriteErrorSeen
2901 	 *  {,-}failfast - set/clear FailFast
2902 	 */
2903 
2904 	struct mddev *mddev = rdev->mddev;
2905 	int err = -EINVAL;
2906 	bool need_update_sb = false;
2907 
2908 	if (cmd_match(buf, "faulty") && rdev->mddev->pers) {
2909 		md_error(rdev->mddev, rdev);
2910 
2911 		if (test_bit(MD_BROKEN, &rdev->mddev->flags))
2912 			err = -EBUSY;
2913 		else
2914 			err = 0;
2915 	} else if (cmd_match(buf, "remove")) {
2916 		if (rdev->mddev->pers) {
2917 			clear_bit(Blocked, &rdev->flags);
2918 			remove_and_add_spares(rdev->mddev, rdev);
2919 		}
2920 		if (rdev->raid_disk >= 0)
2921 			err = -EBUSY;
2922 		else {
2923 			err = 0;
2924 			if (mddev_is_clustered(mddev))
2925 				err = md_cluster_ops->remove_disk(mddev, rdev);
2926 
2927 			if (err == 0) {
2928 				md_kick_rdev_from_array(rdev);
2929 				if (mddev->pers) {
2930 					set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
2931 					md_wakeup_thread(mddev->thread);
2932 				}
2933 				md_new_event();
2934 			}
2935 		}
2936 	} else if (cmd_match(buf, "writemostly")) {
2937 		set_bit(WriteMostly, &rdev->flags);
2938 		mddev_create_serial_pool(rdev->mddev, rdev, false);
2939 		need_update_sb = true;
2940 		err = 0;
2941 	} else if (cmd_match(buf, "-writemostly")) {
2942 		mddev_destroy_serial_pool(rdev->mddev, rdev, false);
2943 		clear_bit(WriteMostly, &rdev->flags);
2944 		need_update_sb = true;
2945 		err = 0;
2946 	} else if (cmd_match(buf, "blocked")) {
2947 		set_bit(Blocked, &rdev->flags);
2948 		err = 0;
2949 	} else if (cmd_match(buf, "-blocked")) {
2950 		if (!test_bit(Faulty, &rdev->flags) &&
2951 		    !test_bit(ExternalBbl, &rdev->flags) &&
2952 		    rdev->badblocks.unacked_exist) {
2953 			/* metadata handler doesn't understand badblocks,
2954 			 * so we need to fail the device
2955 			 */
2956 			md_error(rdev->mddev, rdev);
2957 		}
2958 		clear_bit(Blocked, &rdev->flags);
2959 		clear_bit(BlockedBadBlocks, &rdev->flags);
2960 		wake_up(&rdev->blocked_wait);
2961 		set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2962 		md_wakeup_thread(rdev->mddev->thread);
2963 
2964 		err = 0;
2965 	} else if (cmd_match(buf, "insync") && rdev->raid_disk == -1) {
2966 		set_bit(In_sync, &rdev->flags);
2967 		err = 0;
2968 	} else if (cmd_match(buf, "failfast")) {
2969 		set_bit(FailFast, &rdev->flags);
2970 		need_update_sb = true;
2971 		err = 0;
2972 	} else if (cmd_match(buf, "-failfast")) {
2973 		clear_bit(FailFast, &rdev->flags);
2974 		need_update_sb = true;
2975 		err = 0;
2976 	} else if (cmd_match(buf, "-insync") && rdev->raid_disk >= 0 &&
2977 		   !test_bit(Journal, &rdev->flags)) {
2978 		if (rdev->mddev->pers == NULL) {
2979 			clear_bit(In_sync, &rdev->flags);
2980 			rdev->saved_raid_disk = rdev->raid_disk;
2981 			rdev->raid_disk = -1;
2982 			err = 0;
2983 		}
2984 	} else if (cmd_match(buf, "write_error")) {
2985 		set_bit(WriteErrorSeen, &rdev->flags);
2986 		err = 0;
2987 	} else if (cmd_match(buf, "-write_error")) {
2988 		clear_bit(WriteErrorSeen, &rdev->flags);
2989 		err = 0;
2990 	} else if (cmd_match(buf, "want_replacement")) {
2991 		/* Any non-spare device that is not a replacement can
2992 		 * become want_replacement at any time, but we then need to
2993 		 * check if recovery is needed.
2994 		 */
2995 		if (rdev->raid_disk >= 0 &&
2996 		    !test_bit(Journal, &rdev->flags) &&
2997 		    !test_bit(Replacement, &rdev->flags))
2998 			set_bit(WantReplacement, &rdev->flags);
2999 		set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
3000 		md_wakeup_thread(rdev->mddev->thread);
3001 		err = 0;
3002 	} else if (cmd_match(buf, "-want_replacement")) {
3003 		/* Clearing 'want_replacement' is always allowed.
3004 		 * Once replacements starts it is too late though.
3005 		 */
3006 		err = 0;
3007 		clear_bit(WantReplacement, &rdev->flags);
3008 	} else if (cmd_match(buf, "replacement")) {
3009 		/* Can only set a device as a replacement when array has not
3010 		 * yet been started.  Once running, replacement is automatic
3011 		 * from spares, or by assigning 'slot'.
3012 		 */
3013 		if (rdev->mddev->pers)
3014 			err = -EBUSY;
3015 		else {
3016 			set_bit(Replacement, &rdev->flags);
3017 			err = 0;
3018 		}
3019 	} else if (cmd_match(buf, "-replacement")) {
3020 		/* Similarly, can only clear Replacement before start */
3021 		if (rdev->mddev->pers)
3022 			err = -EBUSY;
3023 		else {
3024 			clear_bit(Replacement, &rdev->flags);
3025 			err = 0;
3026 		}
3027 	} else if (cmd_match(buf, "re-add")) {
3028 		if (!rdev->mddev->pers)
3029 			err = -EINVAL;
3030 		else if (test_bit(Faulty, &rdev->flags) && (rdev->raid_disk == -1) &&
3031 				rdev->saved_raid_disk >= 0) {
3032 			/* clear_bit is performed _after_ all the devices
3033 			 * have their local Faulty bit cleared. If any writes
3034 			 * happen in the meantime in the local node, they
3035 			 * will land in the local bitmap, which will be synced
3036 			 * by this node eventually
3037 			 */
3038 			if (!mddev_is_clustered(rdev->mddev) ||
3039 			    (err = md_cluster_ops->gather_bitmaps(rdev)) == 0) {
3040 				clear_bit(Faulty, &rdev->flags);
3041 				err = add_bound_rdev(rdev);
3042 			}
3043 		} else
3044 			err = -EBUSY;
3045 	} else if (cmd_match(buf, "external_bbl") && (rdev->mddev->external)) {
3046 		set_bit(ExternalBbl, &rdev->flags);
3047 		rdev->badblocks.shift = 0;
3048 		err = 0;
3049 	} else if (cmd_match(buf, "-external_bbl") && (rdev->mddev->external)) {
3050 		clear_bit(ExternalBbl, &rdev->flags);
3051 		err = 0;
3052 	}
3053 	if (need_update_sb)
3054 		md_update_sb(mddev, 1);
3055 	if (!err)
3056 		sysfs_notify_dirent_safe(rdev->sysfs_state);
3057 	return err ? err : len;
3058 }
3059 static struct rdev_sysfs_entry rdev_state =
3060 __ATTR_PREALLOC(state, S_IRUGO|S_IWUSR, state_show, state_store);
3061 
3062 static ssize_t
3063 errors_show(struct md_rdev *rdev, char *page)
3064 {
3065 	return sprintf(page, "%d\n", atomic_read(&rdev->corrected_errors));
3066 }
3067 
3068 static ssize_t
3069 errors_store(struct md_rdev *rdev, const char *buf, size_t len)
3070 {
3071 	unsigned int n;
3072 	int rv;
3073 
3074 	rv = kstrtouint(buf, 10, &n);
3075 	if (rv < 0)
3076 		return rv;
3077 	atomic_set(&rdev->corrected_errors, n);
3078 	return len;
3079 }
3080 static struct rdev_sysfs_entry rdev_errors =
3081 __ATTR(errors, S_IRUGO|S_IWUSR, errors_show, errors_store);
3082 
3083 static ssize_t
3084 slot_show(struct md_rdev *rdev, char *page)
3085 {
3086 	if (test_bit(Journal, &rdev->flags))
3087 		return sprintf(page, "journal\n");
3088 	else if (rdev->raid_disk < 0)
3089 		return sprintf(page, "none\n");
3090 	else
3091 		return sprintf(page, "%d\n", rdev->raid_disk);
3092 }
3093 
3094 static ssize_t
3095 slot_store(struct md_rdev *rdev, const char *buf, size_t len)
3096 {
3097 	int slot;
3098 	int err;
3099 
3100 	if (test_bit(Journal, &rdev->flags))
3101 		return -EBUSY;
3102 	if (strncmp(buf, "none", 4)==0)
3103 		slot = -1;
3104 	else {
3105 		err = kstrtouint(buf, 10, (unsigned int *)&slot);
3106 		if (err < 0)
3107 			return err;
3108 		if (slot < 0)
3109 			/* overflow */
3110 			return -ENOSPC;
3111 	}
3112 	if (rdev->mddev->pers && slot == -1) {
3113 		/* Setting 'slot' on an active array requires also
3114 		 * updating the 'rd%d' link, and communicating
3115 		 * with the personality with ->hot_*_disk.
3116 		 * For now we only support removing
3117 		 * failed/spare devices.  This normally happens automatically,
3118 		 * but not when the metadata is externally managed.
3119 		 */
3120 		if (rdev->raid_disk == -1)
3121 			return -EEXIST;
3122 		/* personality does all needed checks */
3123 		if (rdev->mddev->pers->hot_remove_disk == NULL)
3124 			return -EINVAL;
3125 		clear_bit(Blocked, &rdev->flags);
3126 		remove_and_add_spares(rdev->mddev, rdev);
3127 		if (rdev->raid_disk >= 0)
3128 			return -EBUSY;
3129 		set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
3130 		md_wakeup_thread(rdev->mddev->thread);
3131 	} else if (rdev->mddev->pers) {
3132 		/* Activating a spare .. or possibly reactivating
3133 		 * if we ever get bitmaps working here.
3134 		 */
3135 		int err;
3136 
3137 		if (rdev->raid_disk != -1)
3138 			return -EBUSY;
3139 
3140 		if (test_bit(MD_RECOVERY_RUNNING, &rdev->mddev->recovery))
3141 			return -EBUSY;
3142 
3143 		if (rdev->mddev->pers->hot_add_disk == NULL)
3144 			return -EINVAL;
3145 
3146 		if (slot >= rdev->mddev->raid_disks &&
3147 		    slot >= rdev->mddev->raid_disks + rdev->mddev->delta_disks)
3148 			return -ENOSPC;
3149 
3150 		rdev->raid_disk = slot;
3151 		if (test_bit(In_sync, &rdev->flags))
3152 			rdev->saved_raid_disk = slot;
3153 		else
3154 			rdev->saved_raid_disk = -1;
3155 		clear_bit(In_sync, &rdev->flags);
3156 		clear_bit(Bitmap_sync, &rdev->flags);
3157 		err = rdev->mddev->pers->hot_add_disk(rdev->mddev, rdev);
3158 		if (err) {
3159 			rdev->raid_disk = -1;
3160 			return err;
3161 		} else
3162 			sysfs_notify_dirent_safe(rdev->sysfs_state);
3163 		/* failure here is OK */;
3164 		sysfs_link_rdev(rdev->mddev, rdev);
3165 		/* don't wakeup anyone, leave that to userspace. */
3166 	} else {
3167 		if (slot >= rdev->mddev->raid_disks &&
3168 		    slot >= rdev->mddev->raid_disks + rdev->mddev->delta_disks)
3169 			return -ENOSPC;
3170 		rdev->raid_disk = slot;
3171 		/* assume it is working */
3172 		clear_bit(Faulty, &rdev->flags);
3173 		clear_bit(WriteMostly, &rdev->flags);
3174 		set_bit(In_sync, &rdev->flags);
3175 		sysfs_notify_dirent_safe(rdev->sysfs_state);
3176 	}
3177 	return len;
3178 }
3179 
3180 static struct rdev_sysfs_entry rdev_slot =
3181 __ATTR(slot, S_IRUGO|S_IWUSR, slot_show, slot_store);
3182 
3183 static ssize_t
3184 offset_show(struct md_rdev *rdev, char *page)
3185 {
3186 	return sprintf(page, "%llu\n", (unsigned long long)rdev->data_offset);
3187 }
3188 
3189 static ssize_t
3190 offset_store(struct md_rdev *rdev, const char *buf, size_t len)
3191 {
3192 	unsigned long long offset;
3193 	if (kstrtoull(buf, 10, &offset) < 0)
3194 		return -EINVAL;
3195 	if (rdev->mddev->pers && rdev->raid_disk >= 0)
3196 		return -EBUSY;
3197 	if (rdev->sectors && rdev->mddev->external)
3198 		/* Must set offset before size, so overlap checks
3199 		 * can be sane */
3200 		return -EBUSY;
3201 	rdev->data_offset = offset;
3202 	rdev->new_data_offset = offset;
3203 	return len;
3204 }
3205 
3206 static struct rdev_sysfs_entry rdev_offset =
3207 __ATTR(offset, S_IRUGO|S_IWUSR, offset_show, offset_store);
3208 
3209 static ssize_t new_offset_show(struct md_rdev *rdev, char *page)
3210 {
3211 	return sprintf(page, "%llu\n",
3212 		       (unsigned long long)rdev->new_data_offset);
3213 }
3214 
3215 static ssize_t new_offset_store(struct md_rdev *rdev,
3216 				const char *buf, size_t len)
3217 {
3218 	unsigned long long new_offset;
3219 	struct mddev *mddev = rdev->mddev;
3220 
3221 	if (kstrtoull(buf, 10, &new_offset) < 0)
3222 		return -EINVAL;
3223 
3224 	if (mddev->sync_thread ||
3225 	    test_bit(MD_RECOVERY_RUNNING,&mddev->recovery))
3226 		return -EBUSY;
3227 	if (new_offset == rdev->data_offset)
3228 		/* reset is always permitted */
3229 		;
3230 	else if (new_offset > rdev->data_offset) {
3231 		/* must not push array size beyond rdev_sectors */
3232 		if (new_offset - rdev->data_offset
3233 		    + mddev->dev_sectors > rdev->sectors)
3234 				return -E2BIG;
3235 	}
3236 	/* Metadata worries about other space details. */
3237 
3238 	/* decreasing the offset is inconsistent with a backwards
3239 	 * reshape.
3240 	 */
3241 	if (new_offset < rdev->data_offset &&
3242 	    mddev->reshape_backwards)
3243 		return -EINVAL;
3244 	/* Increasing offset is inconsistent with forwards
3245 	 * reshape.  reshape_direction should be set to
3246 	 * 'backwards' first.
3247 	 */
3248 	if (new_offset > rdev->data_offset &&
3249 	    !mddev->reshape_backwards)
3250 		return -EINVAL;
3251 
3252 	if (mddev->pers && mddev->persistent &&
3253 	    !super_types[mddev->major_version]
3254 	    .allow_new_offset(rdev, new_offset))
3255 		return -E2BIG;
3256 	rdev->new_data_offset = new_offset;
3257 	if (new_offset > rdev->data_offset)
3258 		mddev->reshape_backwards = 1;
3259 	else if (new_offset < rdev->data_offset)
3260 		mddev->reshape_backwards = 0;
3261 
3262 	return len;
3263 }
3264 static struct rdev_sysfs_entry rdev_new_offset =
3265 __ATTR(new_offset, S_IRUGO|S_IWUSR, new_offset_show, new_offset_store);
3266 
3267 static ssize_t
3268 rdev_size_show(struct md_rdev *rdev, char *page)
3269 {
3270 	return sprintf(page, "%llu\n", (unsigned long long)rdev->sectors / 2);
3271 }
3272 
3273 static int md_rdevs_overlap(struct md_rdev *a, struct md_rdev *b)
3274 {
3275 	/* check if two start/length pairs overlap */
3276 	if (a->data_offset + a->sectors <= b->data_offset)
3277 		return false;
3278 	if (b->data_offset + b->sectors <= a->data_offset)
3279 		return false;
3280 	return true;
3281 }
3282 
3283 static bool md_rdev_overlaps(struct md_rdev *rdev)
3284 {
3285 	struct mddev *mddev;
3286 	struct md_rdev *rdev2;
3287 
3288 	spin_lock(&all_mddevs_lock);
3289 	list_for_each_entry(mddev, &all_mddevs, all_mddevs) {
3290 		if (test_bit(MD_DELETED, &mddev->flags))
3291 			continue;
3292 		rdev_for_each(rdev2, mddev) {
3293 			if (rdev != rdev2 && rdev->bdev == rdev2->bdev &&
3294 			    md_rdevs_overlap(rdev, rdev2)) {
3295 				spin_unlock(&all_mddevs_lock);
3296 				return true;
3297 			}
3298 		}
3299 	}
3300 	spin_unlock(&all_mddevs_lock);
3301 	return false;
3302 }
3303 
3304 static int strict_blocks_to_sectors(const char *buf, sector_t *sectors)
3305 {
3306 	unsigned long long blocks;
3307 	sector_t new;
3308 
3309 	if (kstrtoull(buf, 10, &blocks) < 0)
3310 		return -EINVAL;
3311 
3312 	if (blocks & 1ULL << (8 * sizeof(blocks) - 1))
3313 		return -EINVAL; /* sector conversion overflow */
3314 
3315 	new = blocks * 2;
3316 	if (new != blocks * 2)
3317 		return -EINVAL; /* unsigned long long to sector_t overflow */
3318 
3319 	*sectors = new;
3320 	return 0;
3321 }
3322 
3323 static ssize_t
3324 rdev_size_store(struct md_rdev *rdev, const char *buf, size_t len)
3325 {
3326 	struct mddev *my_mddev = rdev->mddev;
3327 	sector_t oldsectors = rdev->sectors;
3328 	sector_t sectors;
3329 
3330 	if (test_bit(Journal, &rdev->flags))
3331 		return -EBUSY;
3332 	if (strict_blocks_to_sectors(buf, &sectors) < 0)
3333 		return -EINVAL;
3334 	if (rdev->data_offset != rdev->new_data_offset)
3335 		return -EINVAL; /* too confusing */
3336 	if (my_mddev->pers && rdev->raid_disk >= 0) {
3337 		if (my_mddev->persistent) {
3338 			sectors = super_types[my_mddev->major_version].
3339 				rdev_size_change(rdev, sectors);
3340 			if (!sectors)
3341 				return -EBUSY;
3342 		} else if (!sectors)
3343 			sectors = bdev_nr_sectors(rdev->bdev) -
3344 				rdev->data_offset;
3345 		if (!my_mddev->pers->resize)
3346 			/* Cannot change size for RAID0 or Linear etc */
3347 			return -EINVAL;
3348 	}
3349 	if (sectors < my_mddev->dev_sectors)
3350 		return -EINVAL; /* component must fit device */
3351 
3352 	rdev->sectors = sectors;
3353 
3354 	/*
3355 	 * Check that all other rdevs with the same bdev do not overlap.  This
3356 	 * check does not provide a hard guarantee, it just helps avoid
3357 	 * dangerous mistakes.
3358 	 */
3359 	if (sectors > oldsectors && my_mddev->external &&
3360 	    md_rdev_overlaps(rdev)) {
3361 		/*
3362 		 * Someone else could have slipped in a size change here, but
3363 		 * doing so is just silly.  We put oldsectors back because we
3364 		 * know it is safe, and trust userspace not to race with itself.
3365 		 */
3366 		rdev->sectors = oldsectors;
3367 		return -EBUSY;
3368 	}
3369 	return len;
3370 }
3371 
3372 static struct rdev_sysfs_entry rdev_size =
3373 __ATTR(size, S_IRUGO|S_IWUSR, rdev_size_show, rdev_size_store);
3374 
3375 static ssize_t recovery_start_show(struct md_rdev *rdev, char *page)
3376 {
3377 	unsigned long long recovery_start = rdev->recovery_offset;
3378 
3379 	if (test_bit(In_sync, &rdev->flags) ||
3380 	    recovery_start == MaxSector)
3381 		return sprintf(page, "none\n");
3382 
3383 	return sprintf(page, "%llu\n", recovery_start);
3384 }
3385 
3386 static ssize_t recovery_start_store(struct md_rdev *rdev, const char *buf, size_t len)
3387 {
3388 	unsigned long long recovery_start;
3389 
3390 	if (cmd_match(buf, "none"))
3391 		recovery_start = MaxSector;
3392 	else if (kstrtoull(buf, 10, &recovery_start))
3393 		return -EINVAL;
3394 
3395 	if (rdev->mddev->pers &&
3396 	    rdev->raid_disk >= 0)
3397 		return -EBUSY;
3398 
3399 	rdev->recovery_offset = recovery_start;
3400 	if (recovery_start == MaxSector)
3401 		set_bit(In_sync, &rdev->flags);
3402 	else
3403 		clear_bit(In_sync, &rdev->flags);
3404 	return len;
3405 }
3406 
3407 static struct rdev_sysfs_entry rdev_recovery_start =
3408 __ATTR(recovery_start, S_IRUGO|S_IWUSR, recovery_start_show, recovery_start_store);
3409 
3410 /* sysfs access to bad-blocks list.
3411  * We present two files.
3412  * 'bad-blocks' lists sector numbers and lengths of ranges that
3413  *    are recorded as bad.  The list is truncated to fit within
3414  *    the one-page limit of sysfs.
3415  *    Writing "sector length" to this file adds an acknowledged
3416  *    bad block list.
3417  * 'unacknowledged-bad-blocks' lists bad blocks that have not yet
3418  *    been acknowledged.  Writing to this file adds bad blocks
3419  *    without acknowledging them.  This is largely for testing.
3420  */
3421 static ssize_t bb_show(struct md_rdev *rdev, char *page)
3422 {
3423 	return badblocks_show(&rdev->badblocks, page, 0);
3424 }
3425 static ssize_t bb_store(struct md_rdev *rdev, const char *page, size_t len)
3426 {
3427 	int rv = badblocks_store(&rdev->badblocks, page, len, 0);
3428 	/* Maybe that ack was all we needed */
3429 	if (test_and_clear_bit(BlockedBadBlocks, &rdev->flags))
3430 		wake_up(&rdev->blocked_wait);
3431 	return rv;
3432 }
3433 static struct rdev_sysfs_entry rdev_bad_blocks =
3434 __ATTR(bad_blocks, S_IRUGO|S_IWUSR, bb_show, bb_store);
3435 
3436 static ssize_t ubb_show(struct md_rdev *rdev, char *page)
3437 {
3438 	return badblocks_show(&rdev->badblocks, page, 1);
3439 }
3440 static ssize_t ubb_store(struct md_rdev *rdev, const char *page, size_t len)
3441 {
3442 	return badblocks_store(&rdev->badblocks, page, len, 1);
3443 }
3444 static struct rdev_sysfs_entry rdev_unack_bad_blocks =
3445 __ATTR(unacknowledged_bad_blocks, S_IRUGO|S_IWUSR, ubb_show, ubb_store);
3446 
3447 static ssize_t
3448 ppl_sector_show(struct md_rdev *rdev, char *page)
3449 {
3450 	return sprintf(page, "%llu\n", (unsigned long long)rdev->ppl.sector);
3451 }
3452 
3453 static ssize_t
3454 ppl_sector_store(struct md_rdev *rdev, const char *buf, size_t len)
3455 {
3456 	unsigned long long sector;
3457 
3458 	if (kstrtoull(buf, 10, &sector) < 0)
3459 		return -EINVAL;
3460 	if (sector != (sector_t)sector)
3461 		return -EINVAL;
3462 
3463 	if (rdev->mddev->pers && test_bit(MD_HAS_PPL, &rdev->mddev->flags) &&
3464 	    rdev->raid_disk >= 0)
3465 		return -EBUSY;
3466 
3467 	if (rdev->mddev->persistent) {
3468 		if (rdev->mddev->major_version == 0)
3469 			return -EINVAL;
3470 		if ((sector > rdev->sb_start &&
3471 		     sector - rdev->sb_start > S16_MAX) ||
3472 		    (sector < rdev->sb_start &&
3473 		     rdev->sb_start - sector > -S16_MIN))
3474 			return -EINVAL;
3475 		rdev->ppl.offset = sector - rdev->sb_start;
3476 	} else if (!rdev->mddev->external) {
3477 		return -EBUSY;
3478 	}
3479 	rdev->ppl.sector = sector;
3480 	return len;
3481 }
3482 
3483 static struct rdev_sysfs_entry rdev_ppl_sector =
3484 __ATTR(ppl_sector, S_IRUGO|S_IWUSR, ppl_sector_show, ppl_sector_store);
3485 
3486 static ssize_t
3487 ppl_size_show(struct md_rdev *rdev, char *page)
3488 {
3489 	return sprintf(page, "%u\n", rdev->ppl.size);
3490 }
3491 
3492 static ssize_t
3493 ppl_size_store(struct md_rdev *rdev, const char *buf, size_t len)
3494 {
3495 	unsigned int size;
3496 
3497 	if (kstrtouint(buf, 10, &size) < 0)
3498 		return -EINVAL;
3499 
3500 	if (rdev->mddev->pers && test_bit(MD_HAS_PPL, &rdev->mddev->flags) &&
3501 	    rdev->raid_disk >= 0)
3502 		return -EBUSY;
3503 
3504 	if (rdev->mddev->persistent) {
3505 		if (rdev->mddev->major_version == 0)
3506 			return -EINVAL;
3507 		if (size > U16_MAX)
3508 			return -EINVAL;
3509 	} else if (!rdev->mddev->external) {
3510 		return -EBUSY;
3511 	}
3512 	rdev->ppl.size = size;
3513 	return len;
3514 }
3515 
3516 static struct rdev_sysfs_entry rdev_ppl_size =
3517 __ATTR(ppl_size, S_IRUGO|S_IWUSR, ppl_size_show, ppl_size_store);
3518 
3519 static struct attribute *rdev_default_attrs[] = {
3520 	&rdev_state.attr,
3521 	&rdev_errors.attr,
3522 	&rdev_slot.attr,
3523 	&rdev_offset.attr,
3524 	&rdev_new_offset.attr,
3525 	&rdev_size.attr,
3526 	&rdev_recovery_start.attr,
3527 	&rdev_bad_blocks.attr,
3528 	&rdev_unack_bad_blocks.attr,
3529 	&rdev_ppl_sector.attr,
3530 	&rdev_ppl_size.attr,
3531 	NULL,
3532 };
3533 ATTRIBUTE_GROUPS(rdev_default);
3534 static ssize_t
3535 rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
3536 {
3537 	struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
3538 	struct md_rdev *rdev = container_of(kobj, struct md_rdev, kobj);
3539 
3540 	if (!entry->show)
3541 		return -EIO;
3542 	if (!rdev->mddev)
3543 		return -ENODEV;
3544 	return entry->show(rdev, page);
3545 }
3546 
3547 static ssize_t
3548 rdev_attr_store(struct kobject *kobj, struct attribute *attr,
3549 	      const char *page, size_t length)
3550 {
3551 	struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
3552 	struct md_rdev *rdev = container_of(kobj, struct md_rdev, kobj);
3553 	struct kernfs_node *kn = NULL;
3554 	ssize_t rv;
3555 	struct mddev *mddev = rdev->mddev;
3556 
3557 	if (!entry->store)
3558 		return -EIO;
3559 	if (!capable(CAP_SYS_ADMIN))
3560 		return -EACCES;
3561 
3562 	if (entry->store == state_store && cmd_match(page, "remove"))
3563 		kn = sysfs_break_active_protection(kobj, attr);
3564 
3565 	rv = mddev ? mddev_lock(mddev) : -ENODEV;
3566 	if (!rv) {
3567 		if (rdev->mddev == NULL)
3568 			rv = -ENODEV;
3569 		else
3570 			rv = entry->store(rdev, page, length);
3571 		mddev_unlock(mddev);
3572 	}
3573 
3574 	if (kn)
3575 		sysfs_unbreak_active_protection(kn);
3576 
3577 	return rv;
3578 }
3579 
3580 static void rdev_free(struct kobject *ko)
3581 {
3582 	struct md_rdev *rdev = container_of(ko, struct md_rdev, kobj);
3583 	kfree(rdev);
3584 }
3585 static const struct sysfs_ops rdev_sysfs_ops = {
3586 	.show		= rdev_attr_show,
3587 	.store		= rdev_attr_store,
3588 };
3589 static const struct kobj_type rdev_ktype = {
3590 	.release	= rdev_free,
3591 	.sysfs_ops	= &rdev_sysfs_ops,
3592 	.default_groups	= rdev_default_groups,
3593 };
3594 
3595 int md_rdev_init(struct md_rdev *rdev)
3596 {
3597 	rdev->desc_nr = -1;
3598 	rdev->saved_raid_disk = -1;
3599 	rdev->raid_disk = -1;
3600 	rdev->flags = 0;
3601 	rdev->data_offset = 0;
3602 	rdev->new_data_offset = 0;
3603 	rdev->sb_events = 0;
3604 	rdev->last_read_error = 0;
3605 	rdev->sb_loaded = 0;
3606 	rdev->bb_page = NULL;
3607 	atomic_set(&rdev->nr_pending, 0);
3608 	atomic_set(&rdev->read_errors, 0);
3609 	atomic_set(&rdev->corrected_errors, 0);
3610 
3611 	INIT_LIST_HEAD(&rdev->same_set);
3612 	init_waitqueue_head(&rdev->blocked_wait);
3613 
3614 	/* Add space to store bad block list.
3615 	 * This reserves the space even on arrays where it cannot
3616 	 * be used - I wonder if that matters
3617 	 */
3618 	return badblocks_init(&rdev->badblocks, 0);
3619 }
3620 EXPORT_SYMBOL_GPL(md_rdev_init);
3621 
3622 /*
3623  * Import a device. If 'super_format' >= 0, then sanity check the superblock
3624  *
3625  * mark the device faulty if:
3626  *
3627  *   - the device is nonexistent (zero size)
3628  *   - the device has no valid superblock
3629  *
3630  * a faulty rdev _never_ has rdev->sb set.
3631  */
3632 static struct md_rdev *md_import_device(dev_t newdev, int super_format, int super_minor)
3633 {
3634 	struct md_rdev *rdev;
3635 	sector_t size;
3636 	int err;
3637 
3638 	rdev = kzalloc(sizeof(*rdev), GFP_KERNEL);
3639 	if (!rdev)
3640 		return ERR_PTR(-ENOMEM);
3641 
3642 	err = md_rdev_init(rdev);
3643 	if (err)
3644 		goto out_free_rdev;
3645 	err = alloc_disk_sb(rdev);
3646 	if (err)
3647 		goto out_clear_rdev;
3648 
3649 	rdev->bdev = blkdev_get_by_dev(newdev, BLK_OPEN_READ | BLK_OPEN_WRITE,
3650 			super_format == -2 ? &claim_rdev : rdev, NULL);
3651 	if (IS_ERR(rdev->bdev)) {
3652 		pr_warn("md: could not open device unknown-block(%u,%u).\n",
3653 			MAJOR(newdev), MINOR(newdev));
3654 		err = PTR_ERR(rdev->bdev);
3655 		goto out_clear_rdev;
3656 	}
3657 
3658 	kobject_init(&rdev->kobj, &rdev_ktype);
3659 
3660 	size = bdev_nr_bytes(rdev->bdev) >> BLOCK_SIZE_BITS;
3661 	if (!size) {
3662 		pr_warn("md: %pg has zero or unknown size, marking faulty!\n",
3663 			rdev->bdev);
3664 		err = -EINVAL;
3665 		goto out_blkdev_put;
3666 	}
3667 
3668 	if (super_format >= 0) {
3669 		err = super_types[super_format].
3670 			load_super(rdev, NULL, super_minor);
3671 		if (err == -EINVAL) {
3672 			pr_warn("md: %pg does not have a valid v%d.%d superblock, not importing!\n",
3673 				rdev->bdev,
3674 				super_format, super_minor);
3675 			goto out_blkdev_put;
3676 		}
3677 		if (err < 0) {
3678 			pr_warn("md: could not read %pg's sb, not importing!\n",
3679 				rdev->bdev);
3680 			goto out_blkdev_put;
3681 		}
3682 	}
3683 
3684 	return rdev;
3685 
3686 out_blkdev_put:
3687 	blkdev_put(rdev->bdev, super_format == -2 ? &claim_rdev : rdev);
3688 out_clear_rdev:
3689 	md_rdev_clear(rdev);
3690 out_free_rdev:
3691 	kfree(rdev);
3692 	return ERR_PTR(err);
3693 }
3694 
3695 /*
3696  * Check a full RAID array for plausibility
3697  */
3698 
3699 static int analyze_sbs(struct mddev *mddev)
3700 {
3701 	int i;
3702 	struct md_rdev *rdev, *freshest, *tmp;
3703 
3704 	freshest = NULL;
3705 	rdev_for_each_safe(rdev, tmp, mddev)
3706 		switch (super_types[mddev->major_version].
3707 			load_super(rdev, freshest, mddev->minor_version)) {
3708 		case 1:
3709 			freshest = rdev;
3710 			break;
3711 		case 0:
3712 			break;
3713 		default:
3714 			pr_warn("md: fatal superblock inconsistency in %pg -- removing from array\n",
3715 				rdev->bdev);
3716 			md_kick_rdev_from_array(rdev);
3717 		}
3718 
3719 	/* Cannot find a valid fresh disk */
3720 	if (!freshest) {
3721 		pr_warn("md: cannot find a valid disk\n");
3722 		return -EINVAL;
3723 	}
3724 
3725 	super_types[mddev->major_version].
3726 		validate_super(mddev, freshest);
3727 
3728 	i = 0;
3729 	rdev_for_each_safe(rdev, tmp, mddev) {
3730 		if (mddev->max_disks &&
3731 		    (rdev->desc_nr >= mddev->max_disks ||
3732 		     i > mddev->max_disks)) {
3733 			pr_warn("md: %s: %pg: only %d devices permitted\n",
3734 				mdname(mddev), rdev->bdev,
3735 				mddev->max_disks);
3736 			md_kick_rdev_from_array(rdev);
3737 			continue;
3738 		}
3739 		if (rdev != freshest) {
3740 			if (super_types[mddev->major_version].
3741 			    validate_super(mddev, rdev)) {
3742 				pr_warn("md: kicking non-fresh %pg from array!\n",
3743 					rdev->bdev);
3744 				md_kick_rdev_from_array(rdev);
3745 				continue;
3746 			}
3747 		}
3748 		if (mddev->level == LEVEL_MULTIPATH) {
3749 			rdev->desc_nr = i++;
3750 			rdev->raid_disk = rdev->desc_nr;
3751 			set_bit(In_sync, &rdev->flags);
3752 		} else if (rdev->raid_disk >=
3753 			    (mddev->raid_disks - min(0, mddev->delta_disks)) &&
3754 			   !test_bit(Journal, &rdev->flags)) {
3755 			rdev->raid_disk = -1;
3756 			clear_bit(In_sync, &rdev->flags);
3757 		}
3758 	}
3759 
3760 	return 0;
3761 }
3762 
3763 /* Read a fixed-point number.
3764  * Numbers in sysfs attributes should be in "standard" units where
3765  * possible, so time should be in seconds.
3766  * However we internally use a a much smaller unit such as
3767  * milliseconds or jiffies.
3768  * This function takes a decimal number with a possible fractional
3769  * component, and produces an integer which is the result of
3770  * multiplying that number by 10^'scale'.
3771  * all without any floating-point arithmetic.
3772  */
3773 int strict_strtoul_scaled(const char *cp, unsigned long *res, int scale)
3774 {
3775 	unsigned long result = 0;
3776 	long decimals = -1;
3777 	while (isdigit(*cp) || (*cp == '.' && decimals < 0)) {
3778 		if (*cp == '.')
3779 			decimals = 0;
3780 		else if (decimals < scale) {
3781 			unsigned int value;
3782 			value = *cp - '0';
3783 			result = result * 10 + value;
3784 			if (decimals >= 0)
3785 				decimals++;
3786 		}
3787 		cp++;
3788 	}
3789 	if (*cp == '\n')
3790 		cp++;
3791 	if (*cp)
3792 		return -EINVAL;
3793 	if (decimals < 0)
3794 		decimals = 0;
3795 	*res = result * int_pow(10, scale - decimals);
3796 	return 0;
3797 }
3798 
3799 static ssize_t
3800 safe_delay_show(struct mddev *mddev, char *page)
3801 {
3802 	unsigned int msec = ((unsigned long)mddev->safemode_delay*1000)/HZ;
3803 
3804 	return sprintf(page, "%u.%03u\n", msec/1000, msec%1000);
3805 }
3806 static ssize_t
3807 safe_delay_store(struct mddev *mddev, const char *cbuf, size_t len)
3808 {
3809 	unsigned long msec;
3810 
3811 	if (mddev_is_clustered(mddev)) {
3812 		pr_warn("md: Safemode is disabled for clustered mode\n");
3813 		return -EINVAL;
3814 	}
3815 
3816 	if (strict_strtoul_scaled(cbuf, &msec, 3) < 0 || msec > UINT_MAX / HZ)
3817 		return -EINVAL;
3818 	if (msec == 0)
3819 		mddev->safemode_delay = 0;
3820 	else {
3821 		unsigned long old_delay = mddev->safemode_delay;
3822 		unsigned long new_delay = (msec*HZ)/1000;
3823 
3824 		if (new_delay == 0)
3825 			new_delay = 1;
3826 		mddev->safemode_delay = new_delay;
3827 		if (new_delay < old_delay || old_delay == 0)
3828 			mod_timer(&mddev->safemode_timer, jiffies+1);
3829 	}
3830 	return len;
3831 }
3832 static struct md_sysfs_entry md_safe_delay =
3833 __ATTR(safe_mode_delay, S_IRUGO|S_IWUSR,safe_delay_show, safe_delay_store);
3834 
3835 static ssize_t
3836 level_show(struct mddev *mddev, char *page)
3837 {
3838 	struct md_personality *p;
3839 	int ret;
3840 	spin_lock(&mddev->lock);
3841 	p = mddev->pers;
3842 	if (p)
3843 		ret = sprintf(page, "%s\n", p->name);
3844 	else if (mddev->clevel[0])
3845 		ret = sprintf(page, "%s\n", mddev->clevel);
3846 	else if (mddev->level != LEVEL_NONE)
3847 		ret = sprintf(page, "%d\n", mddev->level);
3848 	else
3849 		ret = 0;
3850 	spin_unlock(&mddev->lock);
3851 	return ret;
3852 }
3853 
3854 static ssize_t
3855 level_store(struct mddev *mddev, const char *buf, size_t len)
3856 {
3857 	char clevel[16];
3858 	ssize_t rv;
3859 	size_t slen = len;
3860 	struct md_personality *pers, *oldpers;
3861 	long level;
3862 	void *priv, *oldpriv;
3863 	struct md_rdev *rdev;
3864 
3865 	if (slen == 0 || slen >= sizeof(clevel))
3866 		return -EINVAL;
3867 
3868 	rv = mddev_lock(mddev);
3869 	if (rv)
3870 		return rv;
3871 
3872 	if (mddev->pers == NULL) {
3873 		strncpy(mddev->clevel, buf, slen);
3874 		if (mddev->clevel[slen-1] == '\n')
3875 			slen--;
3876 		mddev->clevel[slen] = 0;
3877 		mddev->level = LEVEL_NONE;
3878 		rv = len;
3879 		goto out_unlock;
3880 	}
3881 	rv = -EROFS;
3882 	if (!md_is_rdwr(mddev))
3883 		goto out_unlock;
3884 
3885 	/* request to change the personality.  Need to ensure:
3886 	 *  - array is not engaged in resync/recovery/reshape
3887 	 *  - old personality can be suspended
3888 	 *  - new personality will access other array.
3889 	 */
3890 
3891 	rv = -EBUSY;
3892 	if (mddev->sync_thread ||
3893 	    test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
3894 	    mddev->reshape_position != MaxSector ||
3895 	    mddev->sysfs_active)
3896 		goto out_unlock;
3897 
3898 	rv = -EINVAL;
3899 	if (!mddev->pers->quiesce) {
3900 		pr_warn("md: %s: %s does not support online personality change\n",
3901 			mdname(mddev), mddev->pers->name);
3902 		goto out_unlock;
3903 	}
3904 
3905 	/* Now find the new personality */
3906 	strncpy(clevel, buf, slen);
3907 	if (clevel[slen-1] == '\n')
3908 		slen--;
3909 	clevel[slen] = 0;
3910 	if (kstrtol(clevel, 10, &level))
3911 		level = LEVEL_NONE;
3912 
3913 	if (request_module("md-%s", clevel) != 0)
3914 		request_module("md-level-%s", clevel);
3915 	spin_lock(&pers_lock);
3916 	pers = find_pers(level, clevel);
3917 	if (!pers || !try_module_get(pers->owner)) {
3918 		spin_unlock(&pers_lock);
3919 		pr_warn("md: personality %s not loaded\n", clevel);
3920 		rv = -EINVAL;
3921 		goto out_unlock;
3922 	}
3923 	spin_unlock(&pers_lock);
3924 
3925 	if (pers == mddev->pers) {
3926 		/* Nothing to do! */
3927 		module_put(pers->owner);
3928 		rv = len;
3929 		goto out_unlock;
3930 	}
3931 	if (!pers->takeover) {
3932 		module_put(pers->owner);
3933 		pr_warn("md: %s: %s does not support personality takeover\n",
3934 			mdname(mddev), clevel);
3935 		rv = -EINVAL;
3936 		goto out_unlock;
3937 	}
3938 
3939 	rdev_for_each(rdev, mddev)
3940 		rdev->new_raid_disk = rdev->raid_disk;
3941 
3942 	/* ->takeover must set new_* and/or delta_disks
3943 	 * if it succeeds, and may set them when it fails.
3944 	 */
3945 	priv = pers->takeover(mddev);
3946 	if (IS_ERR(priv)) {
3947 		mddev->new_level = mddev->level;
3948 		mddev->new_layout = mddev->layout;
3949 		mddev->new_chunk_sectors = mddev->chunk_sectors;
3950 		mddev->raid_disks -= mddev->delta_disks;
3951 		mddev->delta_disks = 0;
3952 		mddev->reshape_backwards = 0;
3953 		module_put(pers->owner);
3954 		pr_warn("md: %s: %s would not accept array\n",
3955 			mdname(mddev), clevel);
3956 		rv = PTR_ERR(priv);
3957 		goto out_unlock;
3958 	}
3959 
3960 	/* Looks like we have a winner */
3961 	mddev_suspend(mddev);
3962 	mddev_detach(mddev);
3963 
3964 	spin_lock(&mddev->lock);
3965 	oldpers = mddev->pers;
3966 	oldpriv = mddev->private;
3967 	mddev->pers = pers;
3968 	mddev->private = priv;
3969 	strscpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
3970 	mddev->level = mddev->new_level;
3971 	mddev->layout = mddev->new_layout;
3972 	mddev->chunk_sectors = mddev->new_chunk_sectors;
3973 	mddev->delta_disks = 0;
3974 	mddev->reshape_backwards = 0;
3975 	mddev->degraded = 0;
3976 	spin_unlock(&mddev->lock);
3977 
3978 	if (oldpers->sync_request == NULL &&
3979 	    mddev->external) {
3980 		/* We are converting from a no-redundancy array
3981 		 * to a redundancy array and metadata is managed
3982 		 * externally so we need to be sure that writes
3983 		 * won't block due to a need to transition
3984 		 *      clean->dirty
3985 		 * until external management is started.
3986 		 */
3987 		mddev->in_sync = 0;
3988 		mddev->safemode_delay = 0;
3989 		mddev->safemode = 0;
3990 	}
3991 
3992 	oldpers->free(mddev, oldpriv);
3993 
3994 	if (oldpers->sync_request == NULL &&
3995 	    pers->sync_request != NULL) {
3996 		/* need to add the md_redundancy_group */
3997 		if (sysfs_create_group(&mddev->kobj, &md_redundancy_group))
3998 			pr_warn("md: cannot register extra attributes for %s\n",
3999 				mdname(mddev));
4000 		mddev->sysfs_action = sysfs_get_dirent(mddev->kobj.sd, "sync_action");
4001 		mddev->sysfs_completed = sysfs_get_dirent_safe(mddev->kobj.sd, "sync_completed");
4002 		mddev->sysfs_degraded = sysfs_get_dirent_safe(mddev->kobj.sd, "degraded");
4003 	}
4004 	if (oldpers->sync_request != NULL &&
4005 	    pers->sync_request == NULL) {
4006 		/* need to remove the md_redundancy_group */
4007 		if (mddev->to_remove == NULL)
4008 			mddev->to_remove = &md_redundancy_group;
4009 	}
4010 
4011 	module_put(oldpers->owner);
4012 
4013 	rdev_for_each(rdev, mddev) {
4014 		if (rdev->raid_disk < 0)
4015 			continue;
4016 		if (rdev->new_raid_disk >= mddev->raid_disks)
4017 			rdev->new_raid_disk = -1;
4018 		if (rdev->new_raid_disk == rdev->raid_disk)
4019 			continue;
4020 		sysfs_unlink_rdev(mddev, rdev);
4021 	}
4022 	rdev_for_each(rdev, mddev) {
4023 		if (rdev->raid_disk < 0)
4024 			continue;
4025 		if (rdev->new_raid_disk == rdev->raid_disk)
4026 			continue;
4027 		rdev->raid_disk = rdev->new_raid_disk;
4028 		if (rdev->raid_disk < 0)
4029 			clear_bit(In_sync, &rdev->flags);
4030 		else {
4031 			if (sysfs_link_rdev(mddev, rdev))
4032 				pr_warn("md: cannot register rd%d for %s after level change\n",
4033 					rdev->raid_disk, mdname(mddev));
4034 		}
4035 	}
4036 
4037 	if (pers->sync_request == NULL) {
4038 		/* this is now an array without redundancy, so
4039 		 * it must always be in_sync
4040 		 */
4041 		mddev->in_sync = 1;
4042 		del_timer_sync(&mddev->safemode_timer);
4043 	}
4044 	blk_set_stacking_limits(&mddev->queue->limits);
4045 	pers->run(mddev);
4046 	set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
4047 	mddev_resume(mddev);
4048 	if (!mddev->thread)
4049 		md_update_sb(mddev, 1);
4050 	sysfs_notify_dirent_safe(mddev->sysfs_level);
4051 	md_new_event();
4052 	rv = len;
4053 out_unlock:
4054 	mddev_unlock(mddev);
4055 	return rv;
4056 }
4057 
4058 static struct md_sysfs_entry md_level =
4059 __ATTR(level, S_IRUGO|S_IWUSR, level_show, level_store);
4060 
4061 static ssize_t
4062 layout_show(struct mddev *mddev, char *page)
4063 {
4064 	/* just a number, not meaningful for all levels */
4065 	if (mddev->reshape_position != MaxSector &&
4066 	    mddev->layout != mddev->new_layout)
4067 		return sprintf(page, "%d (%d)\n",
4068 			       mddev->new_layout, mddev->layout);
4069 	return sprintf(page, "%d\n", mddev->layout);
4070 }
4071 
4072 static ssize_t
4073 layout_store(struct mddev *mddev, const char *buf, size_t len)
4074 {
4075 	unsigned int n;
4076 	int err;
4077 
4078 	err = kstrtouint(buf, 10, &n);
4079 	if (err < 0)
4080 		return err;
4081 	err = mddev_lock(mddev);
4082 	if (err)
4083 		return err;
4084 
4085 	if (mddev->pers) {
4086 		if (mddev->pers->check_reshape == NULL)
4087 			err = -EBUSY;
4088 		else if (!md_is_rdwr(mddev))
4089 			err = -EROFS;
4090 		else {
4091 			mddev->new_layout = n;
4092 			err = mddev->pers->check_reshape(mddev);
4093 			if (err)
4094 				mddev->new_layout = mddev->layout;
4095 		}
4096 	} else {
4097 		mddev->new_layout = n;
4098 		if (mddev->reshape_position == MaxSector)
4099 			mddev->layout = n;
4100 	}
4101 	mddev_unlock(mddev);
4102 	return err ?: len;
4103 }
4104 static struct md_sysfs_entry md_layout =
4105 __ATTR(layout, S_IRUGO|S_IWUSR, layout_show, layout_store);
4106 
4107 static ssize_t
4108 raid_disks_show(struct mddev *mddev, char *page)
4109 {
4110 	if (mddev->raid_disks == 0)
4111 		return 0;
4112 	if (mddev->reshape_position != MaxSector &&
4113 	    mddev->delta_disks != 0)
4114 		return sprintf(page, "%d (%d)\n", mddev->raid_disks,
4115 			       mddev->raid_disks - mddev->delta_disks);
4116 	return sprintf(page, "%d\n", mddev->raid_disks);
4117 }
4118 
4119 static int update_raid_disks(struct mddev *mddev, int raid_disks);
4120 
4121 static ssize_t
4122 raid_disks_store(struct mddev *mddev, const char *buf, size_t len)
4123 {
4124 	unsigned int n;
4125 	int err;
4126 
4127 	err = kstrtouint(buf, 10, &n);
4128 	if (err < 0)
4129 		return err;
4130 
4131 	err = mddev_lock(mddev);
4132 	if (err)
4133 		return err;
4134 	if (mddev->pers)
4135 		err = update_raid_disks(mddev, n);
4136 	else if (mddev->reshape_position != MaxSector) {
4137 		struct md_rdev *rdev;
4138 		int olddisks = mddev->raid_disks - mddev->delta_disks;
4139 
4140 		err = -EINVAL;
4141 		rdev_for_each(rdev, mddev) {
4142 			if (olddisks < n &&
4143 			    rdev->data_offset < rdev->new_data_offset)
4144 				goto out_unlock;
4145 			if (olddisks > n &&
4146 			    rdev->data_offset > rdev->new_data_offset)
4147 				goto out_unlock;
4148 		}
4149 		err = 0;
4150 		mddev->delta_disks = n - olddisks;
4151 		mddev->raid_disks = n;
4152 		mddev->reshape_backwards = (mddev->delta_disks < 0);
4153 	} else
4154 		mddev->raid_disks = n;
4155 out_unlock:
4156 	mddev_unlock(mddev);
4157 	return err ? err : len;
4158 }
4159 static struct md_sysfs_entry md_raid_disks =
4160 __ATTR(raid_disks, S_IRUGO|S_IWUSR, raid_disks_show, raid_disks_store);
4161 
4162 static ssize_t
4163 uuid_show(struct mddev *mddev, char *page)
4164 {
4165 	return sprintf(page, "%pU\n", mddev->uuid);
4166 }
4167 static struct md_sysfs_entry md_uuid =
4168 __ATTR(uuid, S_IRUGO, uuid_show, NULL);
4169 
4170 static ssize_t
4171 chunk_size_show(struct mddev *mddev, char *page)
4172 {
4173 	if (mddev->reshape_position != MaxSector &&
4174 	    mddev->chunk_sectors != mddev->new_chunk_sectors)
4175 		return sprintf(page, "%d (%d)\n",
4176 			       mddev->new_chunk_sectors << 9,
4177 			       mddev->chunk_sectors << 9);
4178 	return sprintf(page, "%d\n", mddev->chunk_sectors << 9);
4179 }
4180 
4181 static ssize_t
4182 chunk_size_store(struct mddev *mddev, const char *buf, size_t len)
4183 {
4184 	unsigned long n;
4185 	int err;
4186 
4187 	err = kstrtoul(buf, 10, &n);
4188 	if (err < 0)
4189 		return err;
4190 
4191 	err = mddev_lock(mddev);
4192 	if (err)
4193 		return err;
4194 	if (mddev->pers) {
4195 		if (mddev->pers->check_reshape == NULL)
4196 			err = -EBUSY;
4197 		else if (!md_is_rdwr(mddev))
4198 			err = -EROFS;
4199 		else {
4200 			mddev->new_chunk_sectors = n >> 9;
4201 			err = mddev->pers->check_reshape(mddev);
4202 			if (err)
4203 				mddev->new_chunk_sectors = mddev->chunk_sectors;
4204 		}
4205 	} else {
4206 		mddev->new_chunk_sectors = n >> 9;
4207 		if (mddev->reshape_position == MaxSector)
4208 			mddev->chunk_sectors = n >> 9;
4209 	}
4210 	mddev_unlock(mddev);
4211 	return err ?: len;
4212 }
4213 static struct md_sysfs_entry md_chunk_size =
4214 __ATTR(chunk_size, S_IRUGO|S_IWUSR, chunk_size_show, chunk_size_store);
4215 
4216 static ssize_t
4217 resync_start_show(struct mddev *mddev, char *page)
4218 {
4219 	if (mddev->recovery_cp == MaxSector)
4220 		return sprintf(page, "none\n");
4221 	return sprintf(page, "%llu\n", (unsigned long long)mddev->recovery_cp);
4222 }
4223 
4224 static ssize_t
4225 resync_start_store(struct mddev *mddev, const char *buf, size_t len)
4226 {
4227 	unsigned long long n;
4228 	int err;
4229 
4230 	if (cmd_match(buf, "none"))
4231 		n = MaxSector;
4232 	else {
4233 		err = kstrtoull(buf, 10, &n);
4234 		if (err < 0)
4235 			return err;
4236 		if (n != (sector_t)n)
4237 			return -EINVAL;
4238 	}
4239 
4240 	err = mddev_lock(mddev);
4241 	if (err)
4242 		return err;
4243 	if (mddev->pers && !test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
4244 		err = -EBUSY;
4245 
4246 	if (!err) {
4247 		mddev->recovery_cp = n;
4248 		if (mddev->pers)
4249 			set_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags);
4250 	}
4251 	mddev_unlock(mddev);
4252 	return err ?: len;
4253 }
4254 static struct md_sysfs_entry md_resync_start =
4255 __ATTR_PREALLOC(resync_start, S_IRUGO|S_IWUSR,
4256 		resync_start_show, resync_start_store);
4257 
4258 /*
4259  * The array state can be:
4260  *
4261  * clear
4262  *     No devices, no size, no level
4263  *     Equivalent to STOP_ARRAY ioctl
4264  * inactive
4265  *     May have some settings, but array is not active
4266  *        all IO results in error
4267  *     When written, doesn't tear down array, but just stops it
4268  * suspended (not supported yet)
4269  *     All IO requests will block. The array can be reconfigured.
4270  *     Writing this, if accepted, will block until array is quiescent
4271  * readonly
4272  *     no resync can happen.  no superblocks get written.
4273  *     write requests fail
4274  * read-auto
4275  *     like readonly, but behaves like 'clean' on a write request.
4276  *
4277  * clean - no pending writes, but otherwise active.
4278  *     When written to inactive array, starts without resync
4279  *     If a write request arrives then
4280  *       if metadata is known, mark 'dirty' and switch to 'active'.
4281  *       if not known, block and switch to write-pending
4282  *     If written to an active array that has pending writes, then fails.
4283  * active
4284  *     fully active: IO and resync can be happening.
4285  *     When written to inactive array, starts with resync
4286  *
4287  * write-pending
4288  *     clean, but writes are blocked waiting for 'active' to be written.
4289  *
4290  * active-idle
4291  *     like active, but no writes have been seen for a while (100msec).
4292  *
4293  * broken
4294 *     Array is failed. It's useful because mounted-arrays aren't stopped
4295 *     when array is failed, so this state will at least alert the user that
4296 *     something is wrong.
4297  */
4298 enum array_state { clear, inactive, suspended, readonly, read_auto, clean, active,
4299 		   write_pending, active_idle, broken, bad_word};
4300 static char *array_states[] = {
4301 	"clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active",
4302 	"write-pending", "active-idle", "broken", NULL };
4303 
4304 static int match_word(const char *word, char **list)
4305 {
4306 	int n;
4307 	for (n=0; list[n]; n++)
4308 		if (cmd_match(word, list[n]))
4309 			break;
4310 	return n;
4311 }
4312 
4313 static ssize_t
4314 array_state_show(struct mddev *mddev, char *page)
4315 {
4316 	enum array_state st = inactive;
4317 
4318 	if (mddev->pers && !test_bit(MD_NOT_READY, &mddev->flags)) {
4319 		switch(mddev->ro) {
4320 		case MD_RDONLY:
4321 			st = readonly;
4322 			break;
4323 		case MD_AUTO_READ:
4324 			st = read_auto;
4325 			break;
4326 		case MD_RDWR:
4327 			spin_lock(&mddev->lock);
4328 			if (test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags))
4329 				st = write_pending;
4330 			else if (mddev->in_sync)
4331 				st = clean;
4332 			else if (mddev->safemode)
4333 				st = active_idle;
4334 			else
4335 				st = active;
4336 			spin_unlock(&mddev->lock);
4337 		}
4338 
4339 		if (test_bit(MD_BROKEN, &mddev->flags) && st == clean)
4340 			st = broken;
4341 	} else {
4342 		if (list_empty(&mddev->disks) &&
4343 		    mddev->raid_disks == 0 &&
4344 		    mddev->dev_sectors == 0)
4345 			st = clear;
4346 		else
4347 			st = inactive;
4348 	}
4349 	return sprintf(page, "%s\n", array_states[st]);
4350 }
4351 
4352 static int do_md_stop(struct mddev *mddev, int ro, struct block_device *bdev);
4353 static int md_set_readonly(struct mddev *mddev, struct block_device *bdev);
4354 static int restart_array(struct mddev *mddev);
4355 
4356 static ssize_t
4357 array_state_store(struct mddev *mddev, const char *buf, size_t len)
4358 {
4359 	int err = 0;
4360 	enum array_state st = match_word(buf, array_states);
4361 
4362 	if (mddev->pers && (st == active || st == clean) &&
4363 	    mddev->ro != MD_RDONLY) {
4364 		/* don't take reconfig_mutex when toggling between
4365 		 * clean and active
4366 		 */
4367 		spin_lock(&mddev->lock);
4368 		if (st == active) {
4369 			restart_array(mddev);
4370 			clear_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags);
4371 			md_wakeup_thread(mddev->thread);
4372 			wake_up(&mddev->sb_wait);
4373 		} else /* st == clean */ {
4374 			restart_array(mddev);
4375 			if (!set_in_sync(mddev))
4376 				err = -EBUSY;
4377 		}
4378 		if (!err)
4379 			sysfs_notify_dirent_safe(mddev->sysfs_state);
4380 		spin_unlock(&mddev->lock);
4381 		return err ?: len;
4382 	}
4383 	err = mddev_lock(mddev);
4384 	if (err)
4385 		return err;
4386 	err = -EINVAL;
4387 	switch(st) {
4388 	case bad_word:
4389 		break;
4390 	case clear:
4391 		/* stopping an active array */
4392 		err = do_md_stop(mddev, 0, NULL);
4393 		break;
4394 	case inactive:
4395 		/* stopping an active array */
4396 		if (mddev->pers)
4397 			err = do_md_stop(mddev, 2, NULL);
4398 		else
4399 			err = 0; /* already inactive */
4400 		break;
4401 	case suspended:
4402 		break; /* not supported yet */
4403 	case readonly:
4404 		if (mddev->pers)
4405 			err = md_set_readonly(mddev, NULL);
4406 		else {
4407 			mddev->ro = MD_RDONLY;
4408 			set_disk_ro(mddev->gendisk, 1);
4409 			err = do_md_run(mddev);
4410 		}
4411 		break;
4412 	case read_auto:
4413 		if (mddev->pers) {
4414 			if (md_is_rdwr(mddev))
4415 				err = md_set_readonly(mddev, NULL);
4416 			else if (mddev->ro == MD_RDONLY)
4417 				err = restart_array(mddev);
4418 			if (err == 0) {
4419 				mddev->ro = MD_AUTO_READ;
4420 				set_disk_ro(mddev->gendisk, 0);
4421 			}
4422 		} else {
4423 			mddev->ro = MD_AUTO_READ;
4424 			err = do_md_run(mddev);
4425 		}
4426 		break;
4427 	case clean:
4428 		if (mddev->pers) {
4429 			err = restart_array(mddev);
4430 			if (err)
4431 				break;
4432 			spin_lock(&mddev->lock);
4433 			if (!set_in_sync(mddev))
4434 				err = -EBUSY;
4435 			spin_unlock(&mddev->lock);
4436 		} else
4437 			err = -EINVAL;
4438 		break;
4439 	case active:
4440 		if (mddev->pers) {
4441 			err = restart_array(mddev);
4442 			if (err)
4443 				break;
4444 			clear_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags);
4445 			wake_up(&mddev->sb_wait);
4446 			err = 0;
4447 		} else {
4448 			mddev->ro = MD_RDWR;
4449 			set_disk_ro(mddev->gendisk, 0);
4450 			err = do_md_run(mddev);
4451 		}
4452 		break;
4453 	case write_pending:
4454 	case active_idle:
4455 	case broken:
4456 		/* these cannot be set */
4457 		break;
4458 	}
4459 
4460 	if (!err) {
4461 		if (mddev->hold_active == UNTIL_IOCTL)
4462 			mddev->hold_active = 0;
4463 		sysfs_notify_dirent_safe(mddev->sysfs_state);
4464 	}
4465 	mddev_unlock(mddev);
4466 	return err ?: len;
4467 }
4468 static struct md_sysfs_entry md_array_state =
4469 __ATTR_PREALLOC(array_state, S_IRUGO|S_IWUSR, array_state_show, array_state_store);
4470 
4471 static ssize_t
4472 max_corrected_read_errors_show(struct mddev *mddev, char *page) {
4473 	return sprintf(page, "%d\n",
4474 		       atomic_read(&mddev->max_corr_read_errors));
4475 }
4476 
4477 static ssize_t
4478 max_corrected_read_errors_store(struct mddev *mddev, const char *buf, size_t len)
4479 {
4480 	unsigned int n;
4481 	int rv;
4482 
4483 	rv = kstrtouint(buf, 10, &n);
4484 	if (rv < 0)
4485 		return rv;
4486 	if (n > INT_MAX)
4487 		return -EINVAL;
4488 	atomic_set(&mddev->max_corr_read_errors, n);
4489 	return len;
4490 }
4491 
4492 static struct md_sysfs_entry max_corr_read_errors =
4493 __ATTR(max_read_errors, S_IRUGO|S_IWUSR, max_corrected_read_errors_show,
4494 	max_corrected_read_errors_store);
4495 
4496 static ssize_t
4497 null_show(struct mddev *mddev, char *page)
4498 {
4499 	return -EINVAL;
4500 }
4501 
4502 static ssize_t
4503 new_dev_store(struct mddev *mddev, const char *buf, size_t len)
4504 {
4505 	/* buf must be %d:%d\n? giving major and minor numbers */
4506 	/* The new device is added to the array.
4507 	 * If the array has a persistent superblock, we read the
4508 	 * superblock to initialise info and check validity.
4509 	 * Otherwise, only checking done is that in bind_rdev_to_array,
4510 	 * which mainly checks size.
4511 	 */
4512 	char *e;
4513 	int major = simple_strtoul(buf, &e, 10);
4514 	int minor;
4515 	dev_t dev;
4516 	struct md_rdev *rdev;
4517 	int err;
4518 
4519 	if (!*buf || *e != ':' || !e[1] || e[1] == '\n')
4520 		return -EINVAL;
4521 	minor = simple_strtoul(e+1, &e, 10);
4522 	if (*e && *e != '\n')
4523 		return -EINVAL;
4524 	dev = MKDEV(major, minor);
4525 	if (major != MAJOR(dev) ||
4526 	    minor != MINOR(dev))
4527 		return -EOVERFLOW;
4528 
4529 	err = mddev_lock(mddev);
4530 	if (err)
4531 		return err;
4532 	if (mddev->persistent) {
4533 		rdev = md_import_device(dev, mddev->major_version,
4534 					mddev->minor_version);
4535 		if (!IS_ERR(rdev) && !list_empty(&mddev->disks)) {
4536 			struct md_rdev *rdev0
4537 				= list_entry(mddev->disks.next,
4538 					     struct md_rdev, same_set);
4539 			err = super_types[mddev->major_version]
4540 				.load_super(rdev, rdev0, mddev->minor_version);
4541 			if (err < 0)
4542 				goto out;
4543 		}
4544 	} else if (mddev->external)
4545 		rdev = md_import_device(dev, -2, -1);
4546 	else
4547 		rdev = md_import_device(dev, -1, -1);
4548 
4549 	if (IS_ERR(rdev)) {
4550 		mddev_unlock(mddev);
4551 		return PTR_ERR(rdev);
4552 	}
4553 	err = bind_rdev_to_array(rdev, mddev);
4554  out:
4555 	if (err)
4556 		export_rdev(rdev, mddev);
4557 	mddev_unlock(mddev);
4558 	if (!err)
4559 		md_new_event();
4560 	return err ? err : len;
4561 }
4562 
4563 static struct md_sysfs_entry md_new_device =
4564 __ATTR(new_dev, S_IWUSR, null_show, new_dev_store);
4565 
4566 static ssize_t
4567 bitmap_store(struct mddev *mddev, const char *buf, size_t len)
4568 {
4569 	char *end;
4570 	unsigned long chunk, end_chunk;
4571 	int err;
4572 
4573 	err = mddev_lock(mddev);
4574 	if (err)
4575 		return err;
4576 	if (!mddev->bitmap)
4577 		goto out;
4578 	/* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */
4579 	while (*buf) {
4580 		chunk = end_chunk = simple_strtoul(buf, &end, 0);
4581 		if (buf == end) break;
4582 		if (*end == '-') { /* range */
4583 			buf = end + 1;
4584 			end_chunk = simple_strtoul(buf, &end, 0);
4585 			if (buf == end) break;
4586 		}
4587 		if (*end && !isspace(*end)) break;
4588 		md_bitmap_dirty_bits(mddev->bitmap, chunk, end_chunk);
4589 		buf = skip_spaces(end);
4590 	}
4591 	md_bitmap_unplug(mddev->bitmap); /* flush the bits to disk */
4592 out:
4593 	mddev_unlock(mddev);
4594 	return len;
4595 }
4596 
4597 static struct md_sysfs_entry md_bitmap =
4598 __ATTR(bitmap_set_bits, S_IWUSR, null_show, bitmap_store);
4599 
4600 static ssize_t
4601 size_show(struct mddev *mddev, char *page)
4602 {
4603 	return sprintf(page, "%llu\n",
4604 		(unsigned long long)mddev->dev_sectors / 2);
4605 }
4606 
4607 static int update_size(struct mddev *mddev, sector_t num_sectors);
4608 
4609 static ssize_t
4610 size_store(struct mddev *mddev, const char *buf, size_t len)
4611 {
4612 	/* If array is inactive, we can reduce the component size, but
4613 	 * not increase it (except from 0).
4614 	 * If array is active, we can try an on-line resize
4615 	 */
4616 	sector_t sectors;
4617 	int err = strict_blocks_to_sectors(buf, &sectors);
4618 
4619 	if (err < 0)
4620 		return err;
4621 	err = mddev_lock(mddev);
4622 	if (err)
4623 		return err;
4624 	if (mddev->pers) {
4625 		err = update_size(mddev, sectors);
4626 		if (err == 0)
4627 			md_update_sb(mddev, 1);
4628 	} else {
4629 		if (mddev->dev_sectors == 0 ||
4630 		    mddev->dev_sectors > sectors)
4631 			mddev->dev_sectors = sectors;
4632 		else
4633 			err = -ENOSPC;
4634 	}
4635 	mddev_unlock(mddev);
4636 	return err ? err : len;
4637 }
4638 
4639 static struct md_sysfs_entry md_size =
4640 __ATTR(component_size, S_IRUGO|S_IWUSR, size_show, size_store);
4641 
4642 /* Metadata version.
4643  * This is one of
4644  *   'none' for arrays with no metadata (good luck...)
4645  *   'external' for arrays with externally managed metadata,
4646  * or N.M for internally known formats
4647  */
4648 static ssize_t
4649 metadata_show(struct mddev *mddev, char *page)
4650 {
4651 	if (mddev->persistent)
4652 		return sprintf(page, "%d.%d\n",
4653 			       mddev->major_version, mddev->minor_version);
4654 	else if (mddev->external)
4655 		return sprintf(page, "external:%s\n", mddev->metadata_type);
4656 	else
4657 		return sprintf(page, "none\n");
4658 }
4659 
4660 static ssize_t
4661 metadata_store(struct mddev *mddev, const char *buf, size_t len)
4662 {
4663 	int major, minor;
4664 	char *e;
4665 	int err;
4666 	/* Changing the details of 'external' metadata is
4667 	 * always permitted.  Otherwise there must be
4668 	 * no devices attached to the array.
4669 	 */
4670 
4671 	err = mddev_lock(mddev);
4672 	if (err)
4673 		return err;
4674 	err = -EBUSY;
4675 	if (mddev->external && strncmp(buf, "external:", 9) == 0)
4676 		;
4677 	else if (!list_empty(&mddev->disks))
4678 		goto out_unlock;
4679 
4680 	err = 0;
4681 	if (cmd_match(buf, "none")) {
4682 		mddev->persistent = 0;
4683 		mddev->external = 0;
4684 		mddev->major_version = 0;
4685 		mddev->minor_version = 90;
4686 		goto out_unlock;
4687 	}
4688 	if (strncmp(buf, "external:", 9) == 0) {
4689 		size_t namelen = len-9;
4690 		if (namelen >= sizeof(mddev->metadata_type))
4691 			namelen = sizeof(mddev->metadata_type)-1;
4692 		strncpy(mddev->metadata_type, buf+9, namelen);
4693 		mddev->metadata_type[namelen] = 0;
4694 		if (namelen && mddev->metadata_type[namelen-1] == '\n')
4695 			mddev->metadata_type[--namelen] = 0;
4696 		mddev->persistent = 0;
4697 		mddev->external = 1;
4698 		mddev->major_version = 0;
4699 		mddev->minor_version = 90;
4700 		goto out_unlock;
4701 	}
4702 	major = simple_strtoul(buf, &e, 10);
4703 	err = -EINVAL;
4704 	if (e==buf || *e != '.')
4705 		goto out_unlock;
4706 	buf = e+1;
4707 	minor = simple_strtoul(buf, &e, 10);
4708 	if (e==buf || (*e && *e != '\n') )
4709 		goto out_unlock;
4710 	err = -ENOENT;
4711 	if (major >= ARRAY_SIZE(super_types) || super_types[major].name == NULL)
4712 		goto out_unlock;
4713 	mddev->major_version = major;
4714 	mddev->minor_version = minor;
4715 	mddev->persistent = 1;
4716 	mddev->external = 0;
4717 	err = 0;
4718 out_unlock:
4719 	mddev_unlock(mddev);
4720 	return err ?: len;
4721 }
4722 
4723 static struct md_sysfs_entry md_metadata =
4724 __ATTR_PREALLOC(metadata_version, S_IRUGO|S_IWUSR, metadata_show, metadata_store);
4725 
4726 static ssize_t
4727 action_show(struct mddev *mddev, char *page)
4728 {
4729 	char *type = "idle";
4730 	unsigned long recovery = mddev->recovery;
4731 	if (test_bit(MD_RECOVERY_FROZEN, &recovery))
4732 		type = "frozen";
4733 	else if (test_bit(MD_RECOVERY_RUNNING, &recovery) ||
4734 	    (md_is_rdwr(mddev) && test_bit(MD_RECOVERY_NEEDED, &recovery))) {
4735 		if (test_bit(MD_RECOVERY_RESHAPE, &recovery))
4736 			type = "reshape";
4737 		else if (test_bit(MD_RECOVERY_SYNC, &recovery)) {
4738 			if (!test_bit(MD_RECOVERY_REQUESTED, &recovery))
4739 				type = "resync";
4740 			else if (test_bit(MD_RECOVERY_CHECK, &recovery))
4741 				type = "check";
4742 			else
4743 				type = "repair";
4744 		} else if (test_bit(MD_RECOVERY_RECOVER, &recovery))
4745 			type = "recover";
4746 		else if (mddev->reshape_position != MaxSector)
4747 			type = "reshape";
4748 	}
4749 	return sprintf(page, "%s\n", type);
4750 }
4751 
4752 static void stop_sync_thread(struct mddev *mddev)
4753 {
4754 	if (!test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4755 		return;
4756 
4757 	if (mddev_lock(mddev))
4758 		return;
4759 
4760 	/*
4761 	 * Check again in case MD_RECOVERY_RUNNING is cleared before lock is
4762 	 * held.
4763 	 */
4764 	if (!test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) {
4765 		mddev_unlock(mddev);
4766 		return;
4767 	}
4768 
4769 	if (work_pending(&mddev->del_work))
4770 		flush_workqueue(md_misc_wq);
4771 
4772 	set_bit(MD_RECOVERY_INTR, &mddev->recovery);
4773 	/*
4774 	 * Thread might be blocked waiting for metadata update which will now
4775 	 * never happen
4776 	 */
4777 	md_wakeup_thread_directly(mddev->sync_thread);
4778 
4779 	mddev_unlock(mddev);
4780 }
4781 
4782 static void idle_sync_thread(struct mddev *mddev)
4783 {
4784 	int sync_seq = atomic_read(&mddev->sync_seq);
4785 
4786 	mutex_lock(&mddev->sync_mutex);
4787 	clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4788 	stop_sync_thread(mddev);
4789 
4790 	wait_event(resync_wait, sync_seq != atomic_read(&mddev->sync_seq) ||
4791 			!test_bit(MD_RECOVERY_RUNNING, &mddev->recovery));
4792 
4793 	mutex_unlock(&mddev->sync_mutex);
4794 }
4795 
4796 static void frozen_sync_thread(struct mddev *mddev)
4797 {
4798 	mutex_lock(&mddev->sync_mutex);
4799 	set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4800 	stop_sync_thread(mddev);
4801 
4802 	wait_event(resync_wait, mddev->sync_thread == NULL &&
4803 			!test_bit(MD_RECOVERY_RUNNING, &mddev->recovery));
4804 
4805 	mutex_unlock(&mddev->sync_mutex);
4806 }
4807 
4808 static ssize_t
4809 action_store(struct mddev *mddev, const char *page, size_t len)
4810 {
4811 	if (!mddev->pers || !mddev->pers->sync_request)
4812 		return -EINVAL;
4813 
4814 
4815 	if (cmd_match(page, "idle"))
4816 		idle_sync_thread(mddev);
4817 	else if (cmd_match(page, "frozen"))
4818 		frozen_sync_thread(mddev);
4819 	else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4820 		return -EBUSY;
4821 	else if (cmd_match(page, "resync"))
4822 		clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4823 	else if (cmd_match(page, "recover")) {
4824 		clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4825 		set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
4826 	} else if (cmd_match(page, "reshape")) {
4827 		int err;
4828 		if (mddev->pers->start_reshape == NULL)
4829 			return -EINVAL;
4830 		err = mddev_lock(mddev);
4831 		if (!err) {
4832 			if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) {
4833 				err =  -EBUSY;
4834 			} else if (mddev->reshape_position == MaxSector ||
4835 				   mddev->pers->check_reshape == NULL ||
4836 				   mddev->pers->check_reshape(mddev)) {
4837 				clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4838 				err = mddev->pers->start_reshape(mddev);
4839 			} else {
4840 				/*
4841 				 * If reshape is still in progress, and
4842 				 * md_check_recovery() can continue to reshape,
4843 				 * don't restart reshape because data can be
4844 				 * corrupted for raid456.
4845 				 */
4846 				clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4847 			}
4848 			mddev_unlock(mddev);
4849 		}
4850 		if (err)
4851 			return err;
4852 		sysfs_notify_dirent_safe(mddev->sysfs_degraded);
4853 	} else {
4854 		if (cmd_match(page, "check"))
4855 			set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
4856 		else if (!cmd_match(page, "repair"))
4857 			return -EINVAL;
4858 		clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4859 		set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
4860 		set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
4861 	}
4862 	if (mddev->ro == MD_AUTO_READ) {
4863 		/* A write to sync_action is enough to justify
4864 		 * canceling read-auto mode
4865 		 */
4866 		mddev->ro = MD_RDWR;
4867 		md_wakeup_thread(mddev->sync_thread);
4868 	}
4869 	set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4870 	md_wakeup_thread(mddev->thread);
4871 	sysfs_notify_dirent_safe(mddev->sysfs_action);
4872 	return len;
4873 }
4874 
4875 static struct md_sysfs_entry md_scan_mode =
4876 __ATTR_PREALLOC(sync_action, S_IRUGO|S_IWUSR, action_show, action_store);
4877 
4878 static ssize_t
4879 last_sync_action_show(struct mddev *mddev, char *page)
4880 {
4881 	return sprintf(page, "%s\n", mddev->last_sync_action);
4882 }
4883 
4884 static struct md_sysfs_entry md_last_scan_mode = __ATTR_RO(last_sync_action);
4885 
4886 static ssize_t
4887 mismatch_cnt_show(struct mddev *mddev, char *page)
4888 {
4889 	return sprintf(page, "%llu\n",
4890 		       (unsigned long long)
4891 		       atomic64_read(&mddev->resync_mismatches));
4892 }
4893 
4894 static struct md_sysfs_entry md_mismatches = __ATTR_RO(mismatch_cnt);
4895 
4896 static ssize_t
4897 sync_min_show(struct mddev *mddev, char *page)
4898 {
4899 	return sprintf(page, "%d (%s)\n", speed_min(mddev),
4900 		       mddev->sync_speed_min ? "local": "system");
4901 }
4902 
4903 static ssize_t
4904 sync_min_store(struct mddev *mddev, const char *buf, size_t len)
4905 {
4906 	unsigned int min;
4907 	int rv;
4908 
4909 	if (strncmp(buf, "system", 6)==0) {
4910 		min = 0;
4911 	} else {
4912 		rv = kstrtouint(buf, 10, &min);
4913 		if (rv < 0)
4914 			return rv;
4915 		if (min == 0)
4916 			return -EINVAL;
4917 	}
4918 	mddev->sync_speed_min = min;
4919 	return len;
4920 }
4921 
4922 static struct md_sysfs_entry md_sync_min =
4923 __ATTR(sync_speed_min, S_IRUGO|S_IWUSR, sync_min_show, sync_min_store);
4924 
4925 static ssize_t
4926 sync_max_show(struct mddev *mddev, char *page)
4927 {
4928 	return sprintf(page, "%d (%s)\n", speed_max(mddev),
4929 		       mddev->sync_speed_max ? "local": "system");
4930 }
4931 
4932 static ssize_t
4933 sync_max_store(struct mddev *mddev, const char *buf, size_t len)
4934 {
4935 	unsigned int max;
4936 	int rv;
4937 
4938 	if (strncmp(buf, "system", 6)==0) {
4939 		max = 0;
4940 	} else {
4941 		rv = kstrtouint(buf, 10, &max);
4942 		if (rv < 0)
4943 			return rv;
4944 		if (max == 0)
4945 			return -EINVAL;
4946 	}
4947 	mddev->sync_speed_max = max;
4948 	return len;
4949 }
4950 
4951 static struct md_sysfs_entry md_sync_max =
4952 __ATTR(sync_speed_max, S_IRUGO|S_IWUSR, sync_max_show, sync_max_store);
4953 
4954 static ssize_t
4955 degraded_show(struct mddev *mddev, char *page)
4956 {
4957 	return sprintf(page, "%d\n", mddev->degraded);
4958 }
4959 static struct md_sysfs_entry md_degraded = __ATTR_RO(degraded);
4960 
4961 static ssize_t
4962 sync_force_parallel_show(struct mddev *mddev, char *page)
4963 {
4964 	return sprintf(page, "%d\n", mddev->parallel_resync);
4965 }
4966 
4967 static ssize_t
4968 sync_force_parallel_store(struct mddev *mddev, const char *buf, size_t len)
4969 {
4970 	long n;
4971 
4972 	if (kstrtol(buf, 10, &n))
4973 		return -EINVAL;
4974 
4975 	if (n != 0 && n != 1)
4976 		return -EINVAL;
4977 
4978 	mddev->parallel_resync = n;
4979 
4980 	if (mddev->sync_thread)
4981 		wake_up(&resync_wait);
4982 
4983 	return len;
4984 }
4985 
4986 /* force parallel resync, even with shared block devices */
4987 static struct md_sysfs_entry md_sync_force_parallel =
4988 __ATTR(sync_force_parallel, S_IRUGO|S_IWUSR,
4989        sync_force_parallel_show, sync_force_parallel_store);
4990 
4991 static ssize_t
4992 sync_speed_show(struct mddev *mddev, char *page)
4993 {
4994 	unsigned long resync, dt, db;
4995 	if (mddev->curr_resync == MD_RESYNC_NONE)
4996 		return sprintf(page, "none\n");
4997 	resync = mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active);
4998 	dt = (jiffies - mddev->resync_mark) / HZ;
4999 	if (!dt) dt++;
5000 	db = resync - mddev->resync_mark_cnt;
5001 	return sprintf(page, "%lu\n", db/dt/2); /* K/sec */
5002 }
5003 
5004 static struct md_sysfs_entry md_sync_speed = __ATTR_RO(sync_speed);
5005 
5006 static ssize_t
5007 sync_completed_show(struct mddev *mddev, char *page)
5008 {
5009 	unsigned long long max_sectors, resync;
5010 
5011 	if (!test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
5012 		return sprintf(page, "none\n");
5013 
5014 	if (mddev->curr_resync == MD_RESYNC_YIELDED ||
5015 	    mddev->curr_resync == MD_RESYNC_DELAYED)
5016 		return sprintf(page, "delayed\n");
5017 
5018 	if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ||
5019 	    test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
5020 		max_sectors = mddev->resync_max_sectors;
5021 	else
5022 		max_sectors = mddev->dev_sectors;
5023 
5024 	resync = mddev->curr_resync_completed;
5025 	return sprintf(page, "%llu / %llu\n", resync, max_sectors);
5026 }
5027 
5028 static struct md_sysfs_entry md_sync_completed =
5029 	__ATTR_PREALLOC(sync_completed, S_IRUGO, sync_completed_show, NULL);
5030 
5031 static ssize_t
5032 min_sync_show(struct mddev *mddev, char *page)
5033 {
5034 	return sprintf(page, "%llu\n",
5035 		       (unsigned long long)mddev->resync_min);
5036 }
5037 static ssize_t
5038 min_sync_store(struct mddev *mddev, const char *buf, size_t len)
5039 {
5040 	unsigned long long min;
5041 	int err;
5042 
5043 	if (kstrtoull(buf, 10, &min))
5044 		return -EINVAL;
5045 
5046 	spin_lock(&mddev->lock);
5047 	err = -EINVAL;
5048 	if (min > mddev->resync_max)
5049 		goto out_unlock;
5050 
5051 	err = -EBUSY;
5052 	if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
5053 		goto out_unlock;
5054 
5055 	/* Round down to multiple of 4K for safety */
5056 	mddev->resync_min = round_down(min, 8);
5057 	err = 0;
5058 
5059 out_unlock:
5060 	spin_unlock(&mddev->lock);
5061 	return err ?: len;
5062 }
5063 
5064 static struct md_sysfs_entry md_min_sync =
5065 __ATTR(sync_min, S_IRUGO|S_IWUSR, min_sync_show, min_sync_store);
5066 
5067 static ssize_t
5068 max_sync_show(struct mddev *mddev, char *page)
5069 {
5070 	if (mddev->resync_max == MaxSector)
5071 		return sprintf(page, "max\n");
5072 	else
5073 		return sprintf(page, "%llu\n",
5074 			       (unsigned long long)mddev->resync_max);
5075 }
5076 static ssize_t
5077 max_sync_store(struct mddev *mddev, const char *buf, size_t len)
5078 {
5079 	int err;
5080 	spin_lock(&mddev->lock);
5081 	if (strncmp(buf, "max", 3) == 0)
5082 		mddev->resync_max = MaxSector;
5083 	else {
5084 		unsigned long long max;
5085 		int chunk;
5086 
5087 		err = -EINVAL;
5088 		if (kstrtoull(buf, 10, &max))
5089 			goto out_unlock;
5090 		if (max < mddev->resync_min)
5091 			goto out_unlock;
5092 
5093 		err = -EBUSY;
5094 		if (max < mddev->resync_max && md_is_rdwr(mddev) &&
5095 		    test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
5096 			goto out_unlock;
5097 
5098 		/* Must be a multiple of chunk_size */
5099 		chunk = mddev->chunk_sectors;
5100 		if (chunk) {
5101 			sector_t temp = max;
5102 
5103 			err = -EINVAL;
5104 			if (sector_div(temp, chunk))
5105 				goto out_unlock;
5106 		}
5107 		mddev->resync_max = max;
5108 	}
5109 	wake_up(&mddev->recovery_wait);
5110 	err = 0;
5111 out_unlock:
5112 	spin_unlock(&mddev->lock);
5113 	return err ?: len;
5114 }
5115 
5116 static struct md_sysfs_entry md_max_sync =
5117 __ATTR(sync_max, S_IRUGO|S_IWUSR, max_sync_show, max_sync_store);
5118 
5119 static ssize_t
5120 suspend_lo_show(struct mddev *mddev, char *page)
5121 {
5122 	return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_lo);
5123 }
5124 
5125 static ssize_t
5126 suspend_lo_store(struct mddev *mddev, const char *buf, size_t len)
5127 {
5128 	unsigned long long new;
5129 	int err;
5130 
5131 	err = kstrtoull(buf, 10, &new);
5132 	if (err < 0)
5133 		return err;
5134 	if (new != (sector_t)new)
5135 		return -EINVAL;
5136 
5137 	err = mddev_lock(mddev);
5138 	if (err)
5139 		return err;
5140 	err = -EINVAL;
5141 	if (mddev->pers == NULL ||
5142 	    mddev->pers->quiesce == NULL)
5143 		goto unlock;
5144 	mddev_suspend(mddev);
5145 	mddev->suspend_lo = new;
5146 	mddev_resume(mddev);
5147 
5148 	err = 0;
5149 unlock:
5150 	mddev_unlock(mddev);
5151 	return err ?: len;
5152 }
5153 static struct md_sysfs_entry md_suspend_lo =
5154 __ATTR(suspend_lo, S_IRUGO|S_IWUSR, suspend_lo_show, suspend_lo_store);
5155 
5156 static ssize_t
5157 suspend_hi_show(struct mddev *mddev, char *page)
5158 {
5159 	return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_hi);
5160 }
5161 
5162 static ssize_t
5163 suspend_hi_store(struct mddev *mddev, const char *buf, size_t len)
5164 {
5165 	unsigned long long new;
5166 	int err;
5167 
5168 	err = kstrtoull(buf, 10, &new);
5169 	if (err < 0)
5170 		return err;
5171 	if (new != (sector_t)new)
5172 		return -EINVAL;
5173 
5174 	err = mddev_lock(mddev);
5175 	if (err)
5176 		return err;
5177 	err = -EINVAL;
5178 	if (mddev->pers == NULL)
5179 		goto unlock;
5180 
5181 	mddev_suspend(mddev);
5182 	mddev->suspend_hi = new;
5183 	mddev_resume(mddev);
5184 
5185 	err = 0;
5186 unlock:
5187 	mddev_unlock(mddev);
5188 	return err ?: len;
5189 }
5190 static struct md_sysfs_entry md_suspend_hi =
5191 __ATTR(suspend_hi, S_IRUGO|S_IWUSR, suspend_hi_show, suspend_hi_store);
5192 
5193 static ssize_t
5194 reshape_position_show(struct mddev *mddev, char *page)
5195 {
5196 	if (mddev->reshape_position != MaxSector)
5197 		return sprintf(page, "%llu\n",
5198 			       (unsigned long long)mddev->reshape_position);
5199 	strcpy(page, "none\n");
5200 	return 5;
5201 }
5202 
5203 static ssize_t
5204 reshape_position_store(struct mddev *mddev, const char *buf, size_t len)
5205 {
5206 	struct md_rdev *rdev;
5207 	unsigned long long new;
5208 	int err;
5209 
5210 	err = kstrtoull(buf, 10, &new);
5211 	if (err < 0)
5212 		return err;
5213 	if (new != (sector_t)new)
5214 		return -EINVAL;
5215 	err = mddev_lock(mddev);
5216 	if (err)
5217 		return err;
5218 	err = -EBUSY;
5219 	if (mddev->pers)
5220 		goto unlock;
5221 	mddev->reshape_position = new;
5222 	mddev->delta_disks = 0;
5223 	mddev->reshape_backwards = 0;
5224 	mddev->new_level = mddev->level;
5225 	mddev->new_layout = mddev->layout;
5226 	mddev->new_chunk_sectors = mddev->chunk_sectors;
5227 	rdev_for_each(rdev, mddev)
5228 		rdev->new_data_offset = rdev->data_offset;
5229 	err = 0;
5230 unlock:
5231 	mddev_unlock(mddev);
5232 	return err ?: len;
5233 }
5234 
5235 static struct md_sysfs_entry md_reshape_position =
5236 __ATTR(reshape_position, S_IRUGO|S_IWUSR, reshape_position_show,
5237        reshape_position_store);
5238 
5239 static ssize_t
5240 reshape_direction_show(struct mddev *mddev, char *page)
5241 {
5242 	return sprintf(page, "%s\n",
5243 		       mddev->reshape_backwards ? "backwards" : "forwards");
5244 }
5245 
5246 static ssize_t
5247 reshape_direction_store(struct mddev *mddev, const char *buf, size_t len)
5248 {
5249 	int backwards = 0;
5250 	int err;
5251 
5252 	if (cmd_match(buf, "forwards"))
5253 		backwards = 0;
5254 	else if (cmd_match(buf, "backwards"))
5255 		backwards = 1;
5256 	else
5257 		return -EINVAL;
5258 	if (mddev->reshape_backwards == backwards)
5259 		return len;
5260 
5261 	err = mddev_lock(mddev);
5262 	if (err)
5263 		return err;
5264 	/* check if we are allowed to change */
5265 	if (mddev->delta_disks)
5266 		err = -EBUSY;
5267 	else if (mddev->persistent &&
5268 	    mddev->major_version == 0)
5269 		err =  -EINVAL;
5270 	else
5271 		mddev->reshape_backwards = backwards;
5272 	mddev_unlock(mddev);
5273 	return err ?: len;
5274 }
5275 
5276 static struct md_sysfs_entry md_reshape_direction =
5277 __ATTR(reshape_direction, S_IRUGO|S_IWUSR, reshape_direction_show,
5278        reshape_direction_store);
5279 
5280 static ssize_t
5281 array_size_show(struct mddev *mddev, char *page)
5282 {
5283 	if (mddev->external_size)
5284 		return sprintf(page, "%llu\n",
5285 			       (unsigned long long)mddev->array_sectors/2);
5286 	else
5287 		return sprintf(page, "default\n");
5288 }
5289 
5290 static ssize_t
5291 array_size_store(struct mddev *mddev, const char *buf, size_t len)
5292 {
5293 	sector_t sectors;
5294 	int err;
5295 
5296 	err = mddev_lock(mddev);
5297 	if (err)
5298 		return err;
5299 
5300 	/* cluster raid doesn't support change array_sectors */
5301 	if (mddev_is_clustered(mddev)) {
5302 		mddev_unlock(mddev);
5303 		return -EINVAL;
5304 	}
5305 
5306 	if (strncmp(buf, "default", 7) == 0) {
5307 		if (mddev->pers)
5308 			sectors = mddev->pers->size(mddev, 0, 0);
5309 		else
5310 			sectors = mddev->array_sectors;
5311 
5312 		mddev->external_size = 0;
5313 	} else {
5314 		if (strict_blocks_to_sectors(buf, &sectors) < 0)
5315 			err = -EINVAL;
5316 		else if (mddev->pers && mddev->pers->size(mddev, 0, 0) < sectors)
5317 			err = -E2BIG;
5318 		else
5319 			mddev->external_size = 1;
5320 	}
5321 
5322 	if (!err) {
5323 		mddev->array_sectors = sectors;
5324 		if (mddev->pers)
5325 			set_capacity_and_notify(mddev->gendisk,
5326 						mddev->array_sectors);
5327 	}
5328 	mddev_unlock(mddev);
5329 	return err ?: len;
5330 }
5331 
5332 static struct md_sysfs_entry md_array_size =
5333 __ATTR(array_size, S_IRUGO|S_IWUSR, array_size_show,
5334        array_size_store);
5335 
5336 static ssize_t
5337 consistency_policy_show(struct mddev *mddev, char *page)
5338 {
5339 	int ret;
5340 
5341 	if (test_bit(MD_HAS_JOURNAL, &mddev->flags)) {
5342 		ret = sprintf(page, "journal\n");
5343 	} else if (test_bit(MD_HAS_PPL, &mddev->flags)) {
5344 		ret = sprintf(page, "ppl\n");
5345 	} else if (mddev->bitmap) {
5346 		ret = sprintf(page, "bitmap\n");
5347 	} else if (mddev->pers) {
5348 		if (mddev->pers->sync_request)
5349 			ret = sprintf(page, "resync\n");
5350 		else
5351 			ret = sprintf(page, "none\n");
5352 	} else {
5353 		ret = sprintf(page, "unknown\n");
5354 	}
5355 
5356 	return ret;
5357 }
5358 
5359 static ssize_t
5360 consistency_policy_store(struct mddev *mddev, const char *buf, size_t len)
5361 {
5362 	int err = 0;
5363 
5364 	if (mddev->pers) {
5365 		if (mddev->pers->change_consistency_policy)
5366 			err = mddev->pers->change_consistency_policy(mddev, buf);
5367 		else
5368 			err = -EBUSY;
5369 	} else if (mddev->external && strncmp(buf, "ppl", 3) == 0) {
5370 		set_bit(MD_HAS_PPL, &mddev->flags);
5371 	} else {
5372 		err = -EINVAL;
5373 	}
5374 
5375 	return err ? err : len;
5376 }
5377 
5378 static struct md_sysfs_entry md_consistency_policy =
5379 __ATTR(consistency_policy, S_IRUGO | S_IWUSR, consistency_policy_show,
5380        consistency_policy_store);
5381 
5382 static ssize_t fail_last_dev_show(struct mddev *mddev, char *page)
5383 {
5384 	return sprintf(page, "%d\n", mddev->fail_last_dev);
5385 }
5386 
5387 /*
5388  * Setting fail_last_dev to true to allow last device to be forcibly removed
5389  * from RAID1/RAID10.
5390  */
5391 static ssize_t
5392 fail_last_dev_store(struct mddev *mddev, const char *buf, size_t len)
5393 {
5394 	int ret;
5395 	bool value;
5396 
5397 	ret = kstrtobool(buf, &value);
5398 	if (ret)
5399 		return ret;
5400 
5401 	if (value != mddev->fail_last_dev)
5402 		mddev->fail_last_dev = value;
5403 
5404 	return len;
5405 }
5406 static struct md_sysfs_entry md_fail_last_dev =
5407 __ATTR(fail_last_dev, S_IRUGO | S_IWUSR, fail_last_dev_show,
5408        fail_last_dev_store);
5409 
5410 static ssize_t serialize_policy_show(struct mddev *mddev, char *page)
5411 {
5412 	if (mddev->pers == NULL || (mddev->pers->level != 1))
5413 		return sprintf(page, "n/a\n");
5414 	else
5415 		return sprintf(page, "%d\n", mddev->serialize_policy);
5416 }
5417 
5418 /*
5419  * Setting serialize_policy to true to enforce write IO is not reordered
5420  * for raid1.
5421  */
5422 static ssize_t
5423 serialize_policy_store(struct mddev *mddev, const char *buf, size_t len)
5424 {
5425 	int err;
5426 	bool value;
5427 
5428 	err = kstrtobool(buf, &value);
5429 	if (err)
5430 		return err;
5431 
5432 	if (value == mddev->serialize_policy)
5433 		return len;
5434 
5435 	err = mddev_lock(mddev);
5436 	if (err)
5437 		return err;
5438 	if (mddev->pers == NULL || (mddev->pers->level != 1)) {
5439 		pr_err("md: serialize_policy is only effective for raid1\n");
5440 		err = -EINVAL;
5441 		goto unlock;
5442 	}
5443 
5444 	mddev_suspend(mddev);
5445 	if (value)
5446 		mddev_create_serial_pool(mddev, NULL, true);
5447 	else
5448 		mddev_destroy_serial_pool(mddev, NULL, true);
5449 	mddev->serialize_policy = value;
5450 	mddev_resume(mddev);
5451 unlock:
5452 	mddev_unlock(mddev);
5453 	return err ?: len;
5454 }
5455 
5456 static struct md_sysfs_entry md_serialize_policy =
5457 __ATTR(serialize_policy, S_IRUGO | S_IWUSR, serialize_policy_show,
5458        serialize_policy_store);
5459 
5460 
5461 static struct attribute *md_default_attrs[] = {
5462 	&md_level.attr,
5463 	&md_layout.attr,
5464 	&md_raid_disks.attr,
5465 	&md_uuid.attr,
5466 	&md_chunk_size.attr,
5467 	&md_size.attr,
5468 	&md_resync_start.attr,
5469 	&md_metadata.attr,
5470 	&md_new_device.attr,
5471 	&md_safe_delay.attr,
5472 	&md_array_state.attr,
5473 	&md_reshape_position.attr,
5474 	&md_reshape_direction.attr,
5475 	&md_array_size.attr,
5476 	&max_corr_read_errors.attr,
5477 	&md_consistency_policy.attr,
5478 	&md_fail_last_dev.attr,
5479 	&md_serialize_policy.attr,
5480 	NULL,
5481 };
5482 
5483 static const struct attribute_group md_default_group = {
5484 	.attrs = md_default_attrs,
5485 };
5486 
5487 static struct attribute *md_redundancy_attrs[] = {
5488 	&md_scan_mode.attr,
5489 	&md_last_scan_mode.attr,
5490 	&md_mismatches.attr,
5491 	&md_sync_min.attr,
5492 	&md_sync_max.attr,
5493 	&md_sync_speed.attr,
5494 	&md_sync_force_parallel.attr,
5495 	&md_sync_completed.attr,
5496 	&md_min_sync.attr,
5497 	&md_max_sync.attr,
5498 	&md_suspend_lo.attr,
5499 	&md_suspend_hi.attr,
5500 	&md_bitmap.attr,
5501 	&md_degraded.attr,
5502 	NULL,
5503 };
5504 static const struct attribute_group md_redundancy_group = {
5505 	.name = NULL,
5506 	.attrs = md_redundancy_attrs,
5507 };
5508 
5509 static const struct attribute_group *md_attr_groups[] = {
5510 	&md_default_group,
5511 	&md_bitmap_group,
5512 	NULL,
5513 };
5514 
5515 static ssize_t
5516 md_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
5517 {
5518 	struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
5519 	struct mddev *mddev = container_of(kobj, struct mddev, kobj);
5520 	ssize_t rv;
5521 
5522 	if (!entry->show)
5523 		return -EIO;
5524 	spin_lock(&all_mddevs_lock);
5525 	if (!mddev_get(mddev)) {
5526 		spin_unlock(&all_mddevs_lock);
5527 		return -EBUSY;
5528 	}
5529 	spin_unlock(&all_mddevs_lock);
5530 
5531 	rv = entry->show(mddev, page);
5532 	mddev_put(mddev);
5533 	return rv;
5534 }
5535 
5536 static ssize_t
5537 md_attr_store(struct kobject *kobj, struct attribute *attr,
5538 	      const char *page, size_t length)
5539 {
5540 	struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
5541 	struct mddev *mddev = container_of(kobj, struct mddev, kobj);
5542 	ssize_t rv;
5543 
5544 	if (!entry->store)
5545 		return -EIO;
5546 	if (!capable(CAP_SYS_ADMIN))
5547 		return -EACCES;
5548 	spin_lock(&all_mddevs_lock);
5549 	if (!mddev_get(mddev)) {
5550 		spin_unlock(&all_mddevs_lock);
5551 		return -EBUSY;
5552 	}
5553 	spin_unlock(&all_mddevs_lock);
5554 	rv = entry->store(mddev, page, length);
5555 	mddev_put(mddev);
5556 	return rv;
5557 }
5558 
5559 static void md_kobj_release(struct kobject *ko)
5560 {
5561 	struct mddev *mddev = container_of(ko, struct mddev, kobj);
5562 
5563 	if (mddev->sysfs_state)
5564 		sysfs_put(mddev->sysfs_state);
5565 	if (mddev->sysfs_level)
5566 		sysfs_put(mddev->sysfs_level);
5567 
5568 	del_gendisk(mddev->gendisk);
5569 	put_disk(mddev->gendisk);
5570 }
5571 
5572 static const struct sysfs_ops md_sysfs_ops = {
5573 	.show	= md_attr_show,
5574 	.store	= md_attr_store,
5575 };
5576 static const struct kobj_type md_ktype = {
5577 	.release	= md_kobj_release,
5578 	.sysfs_ops	= &md_sysfs_ops,
5579 	.default_groups	= md_attr_groups,
5580 };
5581 
5582 int mdp_major = 0;
5583 
5584 static void mddev_delayed_delete(struct work_struct *ws)
5585 {
5586 	struct mddev *mddev = container_of(ws, struct mddev, del_work);
5587 
5588 	kobject_put(&mddev->kobj);
5589 }
5590 
5591 static void no_op(struct percpu_ref *r) {}
5592 
5593 int mddev_init_writes_pending(struct mddev *mddev)
5594 {
5595 	if (mddev->writes_pending.percpu_count_ptr)
5596 		return 0;
5597 	if (percpu_ref_init(&mddev->writes_pending, no_op,
5598 			    PERCPU_REF_ALLOW_REINIT, GFP_KERNEL) < 0)
5599 		return -ENOMEM;
5600 	/* We want to start with the refcount at zero */
5601 	percpu_ref_put(&mddev->writes_pending);
5602 	return 0;
5603 }
5604 EXPORT_SYMBOL_GPL(mddev_init_writes_pending);
5605 
5606 struct mddev *md_alloc(dev_t dev, char *name)
5607 {
5608 	/*
5609 	 * If dev is zero, name is the name of a device to allocate with
5610 	 * an arbitrary minor number.  It will be "md_???"
5611 	 * If dev is non-zero it must be a device number with a MAJOR of
5612 	 * MD_MAJOR or mdp_major.  In this case, if "name" is NULL, then
5613 	 * the device is being created by opening a node in /dev.
5614 	 * If "name" is not NULL, the device is being created by
5615 	 * writing to /sys/module/md_mod/parameters/new_array.
5616 	 */
5617 	static DEFINE_MUTEX(disks_mutex);
5618 	struct mddev *mddev;
5619 	struct gendisk *disk;
5620 	int partitioned;
5621 	int shift;
5622 	int unit;
5623 	int error ;
5624 
5625 	/*
5626 	 * Wait for any previous instance of this device to be completely
5627 	 * removed (mddev_delayed_delete).
5628 	 */
5629 	flush_workqueue(md_misc_wq);
5630 
5631 	mutex_lock(&disks_mutex);
5632 	mddev = mddev_alloc(dev);
5633 	if (IS_ERR(mddev)) {
5634 		error = PTR_ERR(mddev);
5635 		goto out_unlock;
5636 	}
5637 
5638 	partitioned = (MAJOR(mddev->unit) != MD_MAJOR);
5639 	shift = partitioned ? MdpMinorShift : 0;
5640 	unit = MINOR(mddev->unit) >> shift;
5641 
5642 	if (name && !dev) {
5643 		/* Need to ensure that 'name' is not a duplicate.
5644 		 */
5645 		struct mddev *mddev2;
5646 		spin_lock(&all_mddevs_lock);
5647 
5648 		list_for_each_entry(mddev2, &all_mddevs, all_mddevs)
5649 			if (mddev2->gendisk &&
5650 			    strcmp(mddev2->gendisk->disk_name, name) == 0) {
5651 				spin_unlock(&all_mddevs_lock);
5652 				error = -EEXIST;
5653 				goto out_free_mddev;
5654 			}
5655 		spin_unlock(&all_mddevs_lock);
5656 	}
5657 	if (name && dev)
5658 		/*
5659 		 * Creating /dev/mdNNN via "newarray", so adjust hold_active.
5660 		 */
5661 		mddev->hold_active = UNTIL_STOP;
5662 
5663 	error = -ENOMEM;
5664 	disk = blk_alloc_disk(NUMA_NO_NODE);
5665 	if (!disk)
5666 		goto out_free_mddev;
5667 
5668 	disk->major = MAJOR(mddev->unit);
5669 	disk->first_minor = unit << shift;
5670 	disk->minors = 1 << shift;
5671 	if (name)
5672 		strcpy(disk->disk_name, name);
5673 	else if (partitioned)
5674 		sprintf(disk->disk_name, "md_d%d", unit);
5675 	else
5676 		sprintf(disk->disk_name, "md%d", unit);
5677 	disk->fops = &md_fops;
5678 	disk->private_data = mddev;
5679 
5680 	mddev->queue = disk->queue;
5681 	blk_set_stacking_limits(&mddev->queue->limits);
5682 	blk_queue_write_cache(mddev->queue, true, true);
5683 	disk->events |= DISK_EVENT_MEDIA_CHANGE;
5684 	mddev->gendisk = disk;
5685 	error = add_disk(disk);
5686 	if (error)
5687 		goto out_put_disk;
5688 
5689 	kobject_init(&mddev->kobj, &md_ktype);
5690 	error = kobject_add(&mddev->kobj, &disk_to_dev(disk)->kobj, "%s", "md");
5691 	if (error) {
5692 		/*
5693 		 * The disk is already live at this point.  Clear the hold flag
5694 		 * and let mddev_put take care of the deletion, as it isn't any
5695 		 * different from a normal close on last release now.
5696 		 */
5697 		mddev->hold_active = 0;
5698 		mutex_unlock(&disks_mutex);
5699 		mddev_put(mddev);
5700 		return ERR_PTR(error);
5701 	}
5702 
5703 	kobject_uevent(&mddev->kobj, KOBJ_ADD);
5704 	mddev->sysfs_state = sysfs_get_dirent_safe(mddev->kobj.sd, "array_state");
5705 	mddev->sysfs_level = sysfs_get_dirent_safe(mddev->kobj.sd, "level");
5706 	mutex_unlock(&disks_mutex);
5707 	return mddev;
5708 
5709 out_put_disk:
5710 	put_disk(disk);
5711 out_free_mddev:
5712 	mddev_free(mddev);
5713 out_unlock:
5714 	mutex_unlock(&disks_mutex);
5715 	return ERR_PTR(error);
5716 }
5717 
5718 static int md_alloc_and_put(dev_t dev, char *name)
5719 {
5720 	struct mddev *mddev = md_alloc(dev, name);
5721 
5722 	if (IS_ERR(mddev))
5723 		return PTR_ERR(mddev);
5724 	mddev_put(mddev);
5725 	return 0;
5726 }
5727 
5728 static void md_probe(dev_t dev)
5729 {
5730 	if (MAJOR(dev) == MD_MAJOR && MINOR(dev) >= 512)
5731 		return;
5732 	if (create_on_open)
5733 		md_alloc_and_put(dev, NULL);
5734 }
5735 
5736 static int add_named_array(const char *val, const struct kernel_param *kp)
5737 {
5738 	/*
5739 	 * val must be "md_*" or "mdNNN".
5740 	 * For "md_*" we allocate an array with a large free minor number, and
5741 	 * set the name to val.  val must not already be an active name.
5742 	 * For "mdNNN" we allocate an array with the minor number NNN
5743 	 * which must not already be in use.
5744 	 */
5745 	int len = strlen(val);
5746 	char buf[DISK_NAME_LEN];
5747 	unsigned long devnum;
5748 
5749 	while (len && val[len-1] == '\n')
5750 		len--;
5751 	if (len >= DISK_NAME_LEN)
5752 		return -E2BIG;
5753 	strscpy(buf, val, len+1);
5754 	if (strncmp(buf, "md_", 3) == 0)
5755 		return md_alloc_and_put(0, buf);
5756 	if (strncmp(buf, "md", 2) == 0 &&
5757 	    isdigit(buf[2]) &&
5758 	    kstrtoul(buf+2, 10, &devnum) == 0 &&
5759 	    devnum <= MINORMASK)
5760 		return md_alloc_and_put(MKDEV(MD_MAJOR, devnum), NULL);
5761 
5762 	return -EINVAL;
5763 }
5764 
5765 static void md_safemode_timeout(struct timer_list *t)
5766 {
5767 	struct mddev *mddev = from_timer(mddev, t, safemode_timer);
5768 
5769 	mddev->safemode = 1;
5770 	if (mddev->external)
5771 		sysfs_notify_dirent_safe(mddev->sysfs_state);
5772 
5773 	md_wakeup_thread(mddev->thread);
5774 }
5775 
5776 static int start_dirty_degraded;
5777 static void active_io_release(struct percpu_ref *ref)
5778 {
5779 	struct mddev *mddev = container_of(ref, struct mddev, active_io);
5780 
5781 	wake_up(&mddev->sb_wait);
5782 }
5783 
5784 int md_run(struct mddev *mddev)
5785 {
5786 	int err;
5787 	struct md_rdev *rdev;
5788 	struct md_personality *pers;
5789 	bool nowait = true;
5790 
5791 	if (list_empty(&mddev->disks))
5792 		/* cannot run an array with no devices.. */
5793 		return -EINVAL;
5794 
5795 	if (mddev->pers)
5796 		return -EBUSY;
5797 	/* Cannot run until previous stop completes properly */
5798 	if (mddev->sysfs_active)
5799 		return -EBUSY;
5800 
5801 	/*
5802 	 * Analyze all RAID superblock(s)
5803 	 */
5804 	if (!mddev->raid_disks) {
5805 		if (!mddev->persistent)
5806 			return -EINVAL;
5807 		err = analyze_sbs(mddev);
5808 		if (err)
5809 			return -EINVAL;
5810 	}
5811 
5812 	if (mddev->level != LEVEL_NONE)
5813 		request_module("md-level-%d", mddev->level);
5814 	else if (mddev->clevel[0])
5815 		request_module("md-%s", mddev->clevel);
5816 
5817 	/*
5818 	 * Drop all container device buffers, from now on
5819 	 * the only valid external interface is through the md
5820 	 * device.
5821 	 */
5822 	mddev->has_superblocks = false;
5823 	rdev_for_each(rdev, mddev) {
5824 		if (test_bit(Faulty, &rdev->flags))
5825 			continue;
5826 		sync_blockdev(rdev->bdev);
5827 		invalidate_bdev(rdev->bdev);
5828 		if (mddev->ro != MD_RDONLY && rdev_read_only(rdev)) {
5829 			mddev->ro = MD_RDONLY;
5830 			if (mddev->gendisk)
5831 				set_disk_ro(mddev->gendisk, 1);
5832 		}
5833 
5834 		if (rdev->sb_page)
5835 			mddev->has_superblocks = true;
5836 
5837 		/* perform some consistency tests on the device.
5838 		 * We don't want the data to overlap the metadata,
5839 		 * Internal Bitmap issues have been handled elsewhere.
5840 		 */
5841 		if (rdev->meta_bdev) {
5842 			/* Nothing to check */;
5843 		} else if (rdev->data_offset < rdev->sb_start) {
5844 			if (mddev->dev_sectors &&
5845 			    rdev->data_offset + mddev->dev_sectors
5846 			    > rdev->sb_start) {
5847 				pr_warn("md: %s: data overlaps metadata\n",
5848 					mdname(mddev));
5849 				return -EINVAL;
5850 			}
5851 		} else {
5852 			if (rdev->sb_start + rdev->sb_size/512
5853 			    > rdev->data_offset) {
5854 				pr_warn("md: %s: metadata overlaps data\n",
5855 					mdname(mddev));
5856 				return -EINVAL;
5857 			}
5858 		}
5859 		sysfs_notify_dirent_safe(rdev->sysfs_state);
5860 		nowait = nowait && bdev_nowait(rdev->bdev);
5861 	}
5862 
5863 	err = percpu_ref_init(&mddev->active_io, active_io_release,
5864 				PERCPU_REF_ALLOW_REINIT, GFP_KERNEL);
5865 	if (err)
5866 		return err;
5867 
5868 	if (!bioset_initialized(&mddev->bio_set)) {
5869 		err = bioset_init(&mddev->bio_set, BIO_POOL_SIZE, 0, BIOSET_NEED_BVECS);
5870 		if (err)
5871 			goto exit_active_io;
5872 	}
5873 	if (!bioset_initialized(&mddev->sync_set)) {
5874 		err = bioset_init(&mddev->sync_set, BIO_POOL_SIZE, 0, BIOSET_NEED_BVECS);
5875 		if (err)
5876 			goto exit_bio_set;
5877 	}
5878 
5879 	if (!bioset_initialized(&mddev->io_clone_set)) {
5880 		err = bioset_init(&mddev->io_clone_set, BIO_POOL_SIZE,
5881 				  offsetof(struct md_io_clone, bio_clone), 0);
5882 		if (err)
5883 			goto exit_sync_set;
5884 	}
5885 
5886 	spin_lock(&pers_lock);
5887 	pers = find_pers(mddev->level, mddev->clevel);
5888 	if (!pers || !try_module_get(pers->owner)) {
5889 		spin_unlock(&pers_lock);
5890 		if (mddev->level != LEVEL_NONE)
5891 			pr_warn("md: personality for level %d is not loaded!\n",
5892 				mddev->level);
5893 		else
5894 			pr_warn("md: personality for level %s is not loaded!\n",
5895 				mddev->clevel);
5896 		err = -EINVAL;
5897 		goto abort;
5898 	}
5899 	spin_unlock(&pers_lock);
5900 	if (mddev->level != pers->level) {
5901 		mddev->level = pers->level;
5902 		mddev->new_level = pers->level;
5903 	}
5904 	strscpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
5905 
5906 	if (mddev->reshape_position != MaxSector &&
5907 	    pers->start_reshape == NULL) {
5908 		/* This personality cannot handle reshaping... */
5909 		module_put(pers->owner);
5910 		err = -EINVAL;
5911 		goto abort;
5912 	}
5913 
5914 	if (pers->sync_request) {
5915 		/* Warn if this is a potentially silly
5916 		 * configuration.
5917 		 */
5918 		struct md_rdev *rdev2;
5919 		int warned = 0;
5920 
5921 		rdev_for_each(rdev, mddev)
5922 			rdev_for_each(rdev2, mddev) {
5923 				if (rdev < rdev2 &&
5924 				    rdev->bdev->bd_disk ==
5925 				    rdev2->bdev->bd_disk) {
5926 					pr_warn("%s: WARNING: %pg appears to be on the same physical disk as %pg.\n",
5927 						mdname(mddev),
5928 						rdev->bdev,
5929 						rdev2->bdev);
5930 					warned = 1;
5931 				}
5932 			}
5933 
5934 		if (warned)
5935 			pr_warn("True protection against single-disk failure might be compromised.\n");
5936 	}
5937 
5938 	mddev->recovery = 0;
5939 	/* may be over-ridden by personality */
5940 	mddev->resync_max_sectors = mddev->dev_sectors;
5941 
5942 	mddev->ok_start_degraded = start_dirty_degraded;
5943 
5944 	if (start_readonly && md_is_rdwr(mddev))
5945 		mddev->ro = MD_AUTO_READ; /* read-only, but switch on first write */
5946 
5947 	err = pers->run(mddev);
5948 	if (err)
5949 		pr_warn("md: pers->run() failed ...\n");
5950 	else if (pers->size(mddev, 0, 0) < mddev->array_sectors) {
5951 		WARN_ONCE(!mddev->external_size,
5952 			  "%s: default size too small, but 'external_size' not in effect?\n",
5953 			  __func__);
5954 		pr_warn("md: invalid array_size %llu > default size %llu\n",
5955 			(unsigned long long)mddev->array_sectors / 2,
5956 			(unsigned long long)pers->size(mddev, 0, 0) / 2);
5957 		err = -EINVAL;
5958 	}
5959 	if (err == 0 && pers->sync_request &&
5960 	    (mddev->bitmap_info.file || mddev->bitmap_info.offset)) {
5961 		struct bitmap *bitmap;
5962 
5963 		bitmap = md_bitmap_create(mddev, -1);
5964 		if (IS_ERR(bitmap)) {
5965 			err = PTR_ERR(bitmap);
5966 			pr_warn("%s: failed to create bitmap (%d)\n",
5967 				mdname(mddev), err);
5968 		} else
5969 			mddev->bitmap = bitmap;
5970 
5971 	}
5972 	if (err)
5973 		goto bitmap_abort;
5974 
5975 	if (mddev->bitmap_info.max_write_behind > 0) {
5976 		bool create_pool = false;
5977 
5978 		rdev_for_each(rdev, mddev) {
5979 			if (test_bit(WriteMostly, &rdev->flags) &&
5980 			    rdev_init_serial(rdev))
5981 				create_pool = true;
5982 		}
5983 		if (create_pool && mddev->serial_info_pool == NULL) {
5984 			mddev->serial_info_pool =
5985 				mempool_create_kmalloc_pool(NR_SERIAL_INFOS,
5986 						    sizeof(struct serial_info));
5987 			if (!mddev->serial_info_pool) {
5988 				err = -ENOMEM;
5989 				goto bitmap_abort;
5990 			}
5991 		}
5992 	}
5993 
5994 	if (mddev->queue) {
5995 		bool nonrot = true;
5996 
5997 		rdev_for_each(rdev, mddev) {
5998 			if (rdev->raid_disk >= 0 && !bdev_nonrot(rdev->bdev)) {
5999 				nonrot = false;
6000 				break;
6001 			}
6002 		}
6003 		if (mddev->degraded)
6004 			nonrot = false;
6005 		if (nonrot)
6006 			blk_queue_flag_set(QUEUE_FLAG_NONROT, mddev->queue);
6007 		else
6008 			blk_queue_flag_clear(QUEUE_FLAG_NONROT, mddev->queue);
6009 		blk_queue_flag_set(QUEUE_FLAG_IO_STAT, mddev->queue);
6010 
6011 		/* Set the NOWAIT flags if all underlying devices support it */
6012 		if (nowait)
6013 			blk_queue_flag_set(QUEUE_FLAG_NOWAIT, mddev->queue);
6014 	}
6015 	if (pers->sync_request) {
6016 		if (mddev->kobj.sd &&
6017 		    sysfs_create_group(&mddev->kobj, &md_redundancy_group))
6018 			pr_warn("md: cannot register extra attributes for %s\n",
6019 				mdname(mddev));
6020 		mddev->sysfs_action = sysfs_get_dirent_safe(mddev->kobj.sd, "sync_action");
6021 		mddev->sysfs_completed = sysfs_get_dirent_safe(mddev->kobj.sd, "sync_completed");
6022 		mddev->sysfs_degraded = sysfs_get_dirent_safe(mddev->kobj.sd, "degraded");
6023 	} else if (mddev->ro == MD_AUTO_READ)
6024 		mddev->ro = MD_RDWR;
6025 
6026 	atomic_set(&mddev->max_corr_read_errors,
6027 		   MD_DEFAULT_MAX_CORRECTED_READ_ERRORS);
6028 	mddev->safemode = 0;
6029 	if (mddev_is_clustered(mddev))
6030 		mddev->safemode_delay = 0;
6031 	else
6032 		mddev->safemode_delay = DEFAULT_SAFEMODE_DELAY;
6033 	mddev->in_sync = 1;
6034 	smp_wmb();
6035 	spin_lock(&mddev->lock);
6036 	mddev->pers = pers;
6037 	spin_unlock(&mddev->lock);
6038 	rdev_for_each(rdev, mddev)
6039 		if (rdev->raid_disk >= 0)
6040 			sysfs_link_rdev(mddev, rdev); /* failure here is OK */
6041 
6042 	if (mddev->degraded && md_is_rdwr(mddev))
6043 		/* This ensures that recovering status is reported immediately
6044 		 * via sysfs - until a lack of spares is confirmed.
6045 		 */
6046 		set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
6047 	set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6048 
6049 	if (mddev->sb_flags)
6050 		md_update_sb(mddev, 0);
6051 
6052 	md_new_event();
6053 	return 0;
6054 
6055 bitmap_abort:
6056 	mddev_detach(mddev);
6057 	if (mddev->private)
6058 		pers->free(mddev, mddev->private);
6059 	mddev->private = NULL;
6060 	module_put(pers->owner);
6061 	md_bitmap_destroy(mddev);
6062 abort:
6063 	bioset_exit(&mddev->io_clone_set);
6064 exit_sync_set:
6065 	bioset_exit(&mddev->sync_set);
6066 exit_bio_set:
6067 	bioset_exit(&mddev->bio_set);
6068 exit_active_io:
6069 	percpu_ref_exit(&mddev->active_io);
6070 	return err;
6071 }
6072 EXPORT_SYMBOL_GPL(md_run);
6073 
6074 int do_md_run(struct mddev *mddev)
6075 {
6076 	int err;
6077 
6078 	set_bit(MD_NOT_READY, &mddev->flags);
6079 	err = md_run(mddev);
6080 	if (err)
6081 		goto out;
6082 	err = md_bitmap_load(mddev);
6083 	if (err) {
6084 		md_bitmap_destroy(mddev);
6085 		goto out;
6086 	}
6087 
6088 	if (mddev_is_clustered(mddev))
6089 		md_allow_write(mddev);
6090 
6091 	/* run start up tasks that require md_thread */
6092 	md_start(mddev);
6093 
6094 	md_wakeup_thread(mddev->thread);
6095 	md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
6096 
6097 	set_capacity_and_notify(mddev->gendisk, mddev->array_sectors);
6098 	clear_bit(MD_NOT_READY, &mddev->flags);
6099 	mddev->changed = 1;
6100 	kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
6101 	sysfs_notify_dirent_safe(mddev->sysfs_state);
6102 	sysfs_notify_dirent_safe(mddev->sysfs_action);
6103 	sysfs_notify_dirent_safe(mddev->sysfs_degraded);
6104 out:
6105 	clear_bit(MD_NOT_READY, &mddev->flags);
6106 	return err;
6107 }
6108 
6109 int md_start(struct mddev *mddev)
6110 {
6111 	int ret = 0;
6112 
6113 	if (mddev->pers->start) {
6114 		set_bit(MD_RECOVERY_WAIT, &mddev->recovery);
6115 		md_wakeup_thread(mddev->thread);
6116 		ret = mddev->pers->start(mddev);
6117 		clear_bit(MD_RECOVERY_WAIT, &mddev->recovery);
6118 		md_wakeup_thread(mddev->sync_thread);
6119 	}
6120 	return ret;
6121 }
6122 EXPORT_SYMBOL_GPL(md_start);
6123 
6124 static int restart_array(struct mddev *mddev)
6125 {
6126 	struct gendisk *disk = mddev->gendisk;
6127 	struct md_rdev *rdev;
6128 	bool has_journal = false;
6129 	bool has_readonly = false;
6130 
6131 	/* Complain if it has no devices */
6132 	if (list_empty(&mddev->disks))
6133 		return -ENXIO;
6134 	if (!mddev->pers)
6135 		return -EINVAL;
6136 	if (md_is_rdwr(mddev))
6137 		return -EBUSY;
6138 
6139 	rcu_read_lock();
6140 	rdev_for_each_rcu(rdev, mddev) {
6141 		if (test_bit(Journal, &rdev->flags) &&
6142 		    !test_bit(Faulty, &rdev->flags))
6143 			has_journal = true;
6144 		if (rdev_read_only(rdev))
6145 			has_readonly = true;
6146 	}
6147 	rcu_read_unlock();
6148 	if (test_bit(MD_HAS_JOURNAL, &mddev->flags) && !has_journal)
6149 		/* Don't restart rw with journal missing/faulty */
6150 			return -EINVAL;
6151 	if (has_readonly)
6152 		return -EROFS;
6153 
6154 	mddev->safemode = 0;
6155 	mddev->ro = MD_RDWR;
6156 	set_disk_ro(disk, 0);
6157 	pr_debug("md: %s switched to read-write mode.\n", mdname(mddev));
6158 	/* Kick recovery or resync if necessary */
6159 	set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6160 	md_wakeup_thread(mddev->thread);
6161 	md_wakeup_thread(mddev->sync_thread);
6162 	sysfs_notify_dirent_safe(mddev->sysfs_state);
6163 	return 0;
6164 }
6165 
6166 static void md_clean(struct mddev *mddev)
6167 {
6168 	mddev->array_sectors = 0;
6169 	mddev->external_size = 0;
6170 	mddev->dev_sectors = 0;
6171 	mddev->raid_disks = 0;
6172 	mddev->recovery_cp = 0;
6173 	mddev->resync_min = 0;
6174 	mddev->resync_max = MaxSector;
6175 	mddev->reshape_position = MaxSector;
6176 	/* we still need mddev->external in export_rdev, do not clear it yet */
6177 	mddev->persistent = 0;
6178 	mddev->level = LEVEL_NONE;
6179 	mddev->clevel[0] = 0;
6180 	mddev->flags = 0;
6181 	mddev->sb_flags = 0;
6182 	mddev->ro = MD_RDWR;
6183 	mddev->metadata_type[0] = 0;
6184 	mddev->chunk_sectors = 0;
6185 	mddev->ctime = mddev->utime = 0;
6186 	mddev->layout = 0;
6187 	mddev->max_disks = 0;
6188 	mddev->events = 0;
6189 	mddev->can_decrease_events = 0;
6190 	mddev->delta_disks = 0;
6191 	mddev->reshape_backwards = 0;
6192 	mddev->new_level = LEVEL_NONE;
6193 	mddev->new_layout = 0;
6194 	mddev->new_chunk_sectors = 0;
6195 	mddev->curr_resync = MD_RESYNC_NONE;
6196 	atomic64_set(&mddev->resync_mismatches, 0);
6197 	mddev->suspend_lo = mddev->suspend_hi = 0;
6198 	mddev->sync_speed_min = mddev->sync_speed_max = 0;
6199 	mddev->recovery = 0;
6200 	mddev->in_sync = 0;
6201 	mddev->changed = 0;
6202 	mddev->degraded = 0;
6203 	mddev->safemode = 0;
6204 	mddev->private = NULL;
6205 	mddev->cluster_info = NULL;
6206 	mddev->bitmap_info.offset = 0;
6207 	mddev->bitmap_info.default_offset = 0;
6208 	mddev->bitmap_info.default_space = 0;
6209 	mddev->bitmap_info.chunksize = 0;
6210 	mddev->bitmap_info.daemon_sleep = 0;
6211 	mddev->bitmap_info.max_write_behind = 0;
6212 	mddev->bitmap_info.nodes = 0;
6213 }
6214 
6215 static void __md_stop_writes(struct mddev *mddev)
6216 {
6217 	set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
6218 	if (work_pending(&mddev->del_work))
6219 		flush_workqueue(md_misc_wq);
6220 	if (mddev->sync_thread) {
6221 		set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6222 		md_reap_sync_thread(mddev);
6223 	}
6224 
6225 	del_timer_sync(&mddev->safemode_timer);
6226 
6227 	if (mddev->pers && mddev->pers->quiesce) {
6228 		mddev->pers->quiesce(mddev, 1);
6229 		mddev->pers->quiesce(mddev, 0);
6230 	}
6231 	md_bitmap_flush(mddev);
6232 
6233 	if (md_is_rdwr(mddev) &&
6234 	    ((!mddev->in_sync && !mddev_is_clustered(mddev)) ||
6235 	     mddev->sb_flags)) {
6236 		/* mark array as shutdown cleanly */
6237 		if (!mddev_is_clustered(mddev))
6238 			mddev->in_sync = 1;
6239 		md_update_sb(mddev, 1);
6240 	}
6241 	/* disable policy to guarantee rdevs free resources for serialization */
6242 	mddev->serialize_policy = 0;
6243 	mddev_destroy_serial_pool(mddev, NULL, true);
6244 }
6245 
6246 void md_stop_writes(struct mddev *mddev)
6247 {
6248 	mddev_lock_nointr(mddev);
6249 	__md_stop_writes(mddev);
6250 	mddev_unlock(mddev);
6251 }
6252 EXPORT_SYMBOL_GPL(md_stop_writes);
6253 
6254 static void mddev_detach(struct mddev *mddev)
6255 {
6256 	md_bitmap_wait_behind_writes(mddev);
6257 	if (mddev->pers && mddev->pers->quiesce && !is_md_suspended(mddev)) {
6258 		mddev->pers->quiesce(mddev, 1);
6259 		mddev->pers->quiesce(mddev, 0);
6260 	}
6261 	md_unregister_thread(mddev, &mddev->thread);
6262 	if (mddev->queue)
6263 		blk_sync_queue(mddev->queue); /* the unplug fn references 'conf'*/
6264 }
6265 
6266 static void __md_stop(struct mddev *mddev)
6267 {
6268 	struct md_personality *pers = mddev->pers;
6269 	md_bitmap_destroy(mddev);
6270 	mddev_detach(mddev);
6271 	/* Ensure ->event_work is done */
6272 	if (mddev->event_work.func)
6273 		flush_workqueue(md_misc_wq);
6274 	spin_lock(&mddev->lock);
6275 	mddev->pers = NULL;
6276 	spin_unlock(&mddev->lock);
6277 	if (mddev->private)
6278 		pers->free(mddev, mddev->private);
6279 	mddev->private = NULL;
6280 	if (pers->sync_request && mddev->to_remove == NULL)
6281 		mddev->to_remove = &md_redundancy_group;
6282 	module_put(pers->owner);
6283 	clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
6284 
6285 	percpu_ref_exit(&mddev->active_io);
6286 	bioset_exit(&mddev->bio_set);
6287 	bioset_exit(&mddev->sync_set);
6288 	bioset_exit(&mddev->io_clone_set);
6289 }
6290 
6291 void md_stop(struct mddev *mddev)
6292 {
6293 	lockdep_assert_held(&mddev->reconfig_mutex);
6294 
6295 	/* stop the array and free an attached data structures.
6296 	 * This is called from dm-raid
6297 	 */
6298 	__md_stop_writes(mddev);
6299 	__md_stop(mddev);
6300 	percpu_ref_exit(&mddev->writes_pending);
6301 }
6302 
6303 EXPORT_SYMBOL_GPL(md_stop);
6304 
6305 static int md_set_readonly(struct mddev *mddev, struct block_device *bdev)
6306 {
6307 	int err = 0;
6308 	int did_freeze = 0;
6309 
6310 	if (!test_bit(MD_RECOVERY_FROZEN, &mddev->recovery)) {
6311 		did_freeze = 1;
6312 		set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
6313 		md_wakeup_thread(mddev->thread);
6314 	}
6315 	if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
6316 		set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6317 
6318 	/*
6319 	 * Thread might be blocked waiting for metadata update which will now
6320 	 * never happen
6321 	 */
6322 	md_wakeup_thread_directly(mddev->sync_thread);
6323 
6324 	if (mddev->external && test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags))
6325 		return -EBUSY;
6326 	mddev_unlock(mddev);
6327 	wait_event(resync_wait, !test_bit(MD_RECOVERY_RUNNING,
6328 					  &mddev->recovery));
6329 	wait_event(mddev->sb_wait,
6330 		   !test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags));
6331 	mddev_lock_nointr(mddev);
6332 
6333 	mutex_lock(&mddev->open_mutex);
6334 	if ((mddev->pers && atomic_read(&mddev->openers) > !!bdev) ||
6335 	    mddev->sync_thread ||
6336 	    test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) {
6337 		pr_warn("md: %s still in use.\n",mdname(mddev));
6338 		if (did_freeze) {
6339 			clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
6340 			set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6341 			md_wakeup_thread(mddev->thread);
6342 		}
6343 		err = -EBUSY;
6344 		goto out;
6345 	}
6346 	if (mddev->pers) {
6347 		__md_stop_writes(mddev);
6348 
6349 		err  = -ENXIO;
6350 		if (mddev->ro == MD_RDONLY)
6351 			goto out;
6352 		mddev->ro = MD_RDONLY;
6353 		set_disk_ro(mddev->gendisk, 1);
6354 		clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
6355 		set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6356 		md_wakeup_thread(mddev->thread);
6357 		sysfs_notify_dirent_safe(mddev->sysfs_state);
6358 		err = 0;
6359 	}
6360 out:
6361 	mutex_unlock(&mddev->open_mutex);
6362 	return err;
6363 }
6364 
6365 /* mode:
6366  *   0 - completely stop and dis-assemble array
6367  *   2 - stop but do not disassemble array
6368  */
6369 static int do_md_stop(struct mddev *mddev, int mode,
6370 		      struct block_device *bdev)
6371 {
6372 	struct gendisk *disk = mddev->gendisk;
6373 	struct md_rdev *rdev;
6374 	int did_freeze = 0;
6375 
6376 	if (!test_bit(MD_RECOVERY_FROZEN, &mddev->recovery)) {
6377 		did_freeze = 1;
6378 		set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
6379 		md_wakeup_thread(mddev->thread);
6380 	}
6381 	if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
6382 		set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6383 
6384 	/*
6385 	 * Thread might be blocked waiting for metadata update which will now
6386 	 * never happen
6387 	 */
6388 	md_wakeup_thread_directly(mddev->sync_thread);
6389 
6390 	mddev_unlock(mddev);
6391 	wait_event(resync_wait, (mddev->sync_thread == NULL &&
6392 				 !test_bit(MD_RECOVERY_RUNNING,
6393 					   &mddev->recovery)));
6394 	mddev_lock_nointr(mddev);
6395 
6396 	mutex_lock(&mddev->open_mutex);
6397 	if ((mddev->pers && atomic_read(&mddev->openers) > !!bdev) ||
6398 	    mddev->sysfs_active ||
6399 	    mddev->sync_thread ||
6400 	    test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) {
6401 		pr_warn("md: %s still in use.\n",mdname(mddev));
6402 		mutex_unlock(&mddev->open_mutex);
6403 		if (did_freeze) {
6404 			clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
6405 			set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6406 			md_wakeup_thread(mddev->thread);
6407 		}
6408 		return -EBUSY;
6409 	}
6410 	if (mddev->pers) {
6411 		if (!md_is_rdwr(mddev))
6412 			set_disk_ro(disk, 0);
6413 
6414 		__md_stop_writes(mddev);
6415 		__md_stop(mddev);
6416 
6417 		/* tell userspace to handle 'inactive' */
6418 		sysfs_notify_dirent_safe(mddev->sysfs_state);
6419 
6420 		rdev_for_each(rdev, mddev)
6421 			if (rdev->raid_disk >= 0)
6422 				sysfs_unlink_rdev(mddev, rdev);
6423 
6424 		set_capacity_and_notify(disk, 0);
6425 		mutex_unlock(&mddev->open_mutex);
6426 		mddev->changed = 1;
6427 
6428 		if (!md_is_rdwr(mddev))
6429 			mddev->ro = MD_RDWR;
6430 	} else
6431 		mutex_unlock(&mddev->open_mutex);
6432 	/*
6433 	 * Free resources if final stop
6434 	 */
6435 	if (mode == 0) {
6436 		pr_info("md: %s stopped.\n", mdname(mddev));
6437 
6438 		if (mddev->bitmap_info.file) {
6439 			struct file *f = mddev->bitmap_info.file;
6440 			spin_lock(&mddev->lock);
6441 			mddev->bitmap_info.file = NULL;
6442 			spin_unlock(&mddev->lock);
6443 			fput(f);
6444 		}
6445 		mddev->bitmap_info.offset = 0;
6446 
6447 		export_array(mddev);
6448 
6449 		md_clean(mddev);
6450 		if (mddev->hold_active == UNTIL_STOP)
6451 			mddev->hold_active = 0;
6452 	}
6453 	md_new_event();
6454 	sysfs_notify_dirent_safe(mddev->sysfs_state);
6455 	return 0;
6456 }
6457 
6458 #ifndef MODULE
6459 static void autorun_array(struct mddev *mddev)
6460 {
6461 	struct md_rdev *rdev;
6462 	int err;
6463 
6464 	if (list_empty(&mddev->disks))
6465 		return;
6466 
6467 	pr_info("md: running: ");
6468 
6469 	rdev_for_each(rdev, mddev) {
6470 		pr_cont("<%pg>", rdev->bdev);
6471 	}
6472 	pr_cont("\n");
6473 
6474 	err = do_md_run(mddev);
6475 	if (err) {
6476 		pr_warn("md: do_md_run() returned %d\n", err);
6477 		do_md_stop(mddev, 0, NULL);
6478 	}
6479 }
6480 
6481 /*
6482  * lets try to run arrays based on all disks that have arrived
6483  * until now. (those are in pending_raid_disks)
6484  *
6485  * the method: pick the first pending disk, collect all disks with
6486  * the same UUID, remove all from the pending list and put them into
6487  * the 'same_array' list. Then order this list based on superblock
6488  * update time (freshest comes first), kick out 'old' disks and
6489  * compare superblocks. If everything's fine then run it.
6490  *
6491  * If "unit" is allocated, then bump its reference count
6492  */
6493 static void autorun_devices(int part)
6494 {
6495 	struct md_rdev *rdev0, *rdev, *tmp;
6496 	struct mddev *mddev;
6497 
6498 	pr_info("md: autorun ...\n");
6499 	while (!list_empty(&pending_raid_disks)) {
6500 		int unit;
6501 		dev_t dev;
6502 		LIST_HEAD(candidates);
6503 		rdev0 = list_entry(pending_raid_disks.next,
6504 					 struct md_rdev, same_set);
6505 
6506 		pr_debug("md: considering %pg ...\n", rdev0->bdev);
6507 		INIT_LIST_HEAD(&candidates);
6508 		rdev_for_each_list(rdev, tmp, &pending_raid_disks)
6509 			if (super_90_load(rdev, rdev0, 0) >= 0) {
6510 				pr_debug("md:  adding %pg ...\n",
6511 					 rdev->bdev);
6512 				list_move(&rdev->same_set, &candidates);
6513 			}
6514 		/*
6515 		 * now we have a set of devices, with all of them having
6516 		 * mostly sane superblocks. It's time to allocate the
6517 		 * mddev.
6518 		 */
6519 		if (part) {
6520 			dev = MKDEV(mdp_major,
6521 				    rdev0->preferred_minor << MdpMinorShift);
6522 			unit = MINOR(dev) >> MdpMinorShift;
6523 		} else {
6524 			dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
6525 			unit = MINOR(dev);
6526 		}
6527 		if (rdev0->preferred_minor != unit) {
6528 			pr_warn("md: unit number in %pg is bad: %d\n",
6529 				rdev0->bdev, rdev0->preferred_minor);
6530 			break;
6531 		}
6532 
6533 		mddev = md_alloc(dev, NULL);
6534 		if (IS_ERR(mddev))
6535 			break;
6536 
6537 		if (mddev_lock(mddev))
6538 			pr_warn("md: %s locked, cannot run\n", mdname(mddev));
6539 		else if (mddev->raid_disks || mddev->major_version
6540 			 || !list_empty(&mddev->disks)) {
6541 			pr_warn("md: %s already running, cannot run %pg\n",
6542 				mdname(mddev), rdev0->bdev);
6543 			mddev_unlock(mddev);
6544 		} else {
6545 			pr_debug("md: created %s\n", mdname(mddev));
6546 			mddev->persistent = 1;
6547 			rdev_for_each_list(rdev, tmp, &candidates) {
6548 				list_del_init(&rdev->same_set);
6549 				if (bind_rdev_to_array(rdev, mddev))
6550 					export_rdev(rdev, mddev);
6551 			}
6552 			autorun_array(mddev);
6553 			mddev_unlock(mddev);
6554 		}
6555 		/* on success, candidates will be empty, on error
6556 		 * it won't...
6557 		 */
6558 		rdev_for_each_list(rdev, tmp, &candidates) {
6559 			list_del_init(&rdev->same_set);
6560 			export_rdev(rdev, mddev);
6561 		}
6562 		mddev_put(mddev);
6563 	}
6564 	pr_info("md: ... autorun DONE.\n");
6565 }
6566 #endif /* !MODULE */
6567 
6568 static int get_version(void __user *arg)
6569 {
6570 	mdu_version_t ver;
6571 
6572 	ver.major = MD_MAJOR_VERSION;
6573 	ver.minor = MD_MINOR_VERSION;
6574 	ver.patchlevel = MD_PATCHLEVEL_VERSION;
6575 
6576 	if (copy_to_user(arg, &ver, sizeof(ver)))
6577 		return -EFAULT;
6578 
6579 	return 0;
6580 }
6581 
6582 static int get_array_info(struct mddev *mddev, void __user *arg)
6583 {
6584 	mdu_array_info_t info;
6585 	int nr,working,insync,failed,spare;
6586 	struct md_rdev *rdev;
6587 
6588 	nr = working = insync = failed = spare = 0;
6589 	rcu_read_lock();
6590 	rdev_for_each_rcu(rdev, mddev) {
6591 		nr++;
6592 		if (test_bit(Faulty, &rdev->flags))
6593 			failed++;
6594 		else {
6595 			working++;
6596 			if (test_bit(In_sync, &rdev->flags))
6597 				insync++;
6598 			else if (test_bit(Journal, &rdev->flags))
6599 				/* TODO: add journal count to md_u.h */
6600 				;
6601 			else
6602 				spare++;
6603 		}
6604 	}
6605 	rcu_read_unlock();
6606 
6607 	info.major_version = mddev->major_version;
6608 	info.minor_version = mddev->minor_version;
6609 	info.patch_version = MD_PATCHLEVEL_VERSION;
6610 	info.ctime         = clamp_t(time64_t, mddev->ctime, 0, U32_MAX);
6611 	info.level         = mddev->level;
6612 	info.size          = mddev->dev_sectors / 2;
6613 	if (info.size != mddev->dev_sectors / 2) /* overflow */
6614 		info.size = -1;
6615 	info.nr_disks      = nr;
6616 	info.raid_disks    = mddev->raid_disks;
6617 	info.md_minor      = mddev->md_minor;
6618 	info.not_persistent= !mddev->persistent;
6619 
6620 	info.utime         = clamp_t(time64_t, mddev->utime, 0, U32_MAX);
6621 	info.state         = 0;
6622 	if (mddev->in_sync)
6623 		info.state = (1<<MD_SB_CLEAN);
6624 	if (mddev->bitmap && mddev->bitmap_info.offset)
6625 		info.state |= (1<<MD_SB_BITMAP_PRESENT);
6626 	if (mddev_is_clustered(mddev))
6627 		info.state |= (1<<MD_SB_CLUSTERED);
6628 	info.active_disks  = insync;
6629 	info.working_disks = working;
6630 	info.failed_disks  = failed;
6631 	info.spare_disks   = spare;
6632 
6633 	info.layout        = mddev->layout;
6634 	info.chunk_size    = mddev->chunk_sectors << 9;
6635 
6636 	if (copy_to_user(arg, &info, sizeof(info)))
6637 		return -EFAULT;
6638 
6639 	return 0;
6640 }
6641 
6642 static int get_bitmap_file(struct mddev *mddev, void __user * arg)
6643 {
6644 	mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */
6645 	char *ptr;
6646 	int err;
6647 
6648 	file = kzalloc(sizeof(*file), GFP_NOIO);
6649 	if (!file)
6650 		return -ENOMEM;
6651 
6652 	err = 0;
6653 	spin_lock(&mddev->lock);
6654 	/* bitmap enabled */
6655 	if (mddev->bitmap_info.file) {
6656 		ptr = file_path(mddev->bitmap_info.file, file->pathname,
6657 				sizeof(file->pathname));
6658 		if (IS_ERR(ptr))
6659 			err = PTR_ERR(ptr);
6660 		else
6661 			memmove(file->pathname, ptr,
6662 				sizeof(file->pathname)-(ptr-file->pathname));
6663 	}
6664 	spin_unlock(&mddev->lock);
6665 
6666 	if (err == 0 &&
6667 	    copy_to_user(arg, file, sizeof(*file)))
6668 		err = -EFAULT;
6669 
6670 	kfree(file);
6671 	return err;
6672 }
6673 
6674 static int get_disk_info(struct mddev *mddev, void __user * arg)
6675 {
6676 	mdu_disk_info_t info;
6677 	struct md_rdev *rdev;
6678 
6679 	if (copy_from_user(&info, arg, sizeof(info)))
6680 		return -EFAULT;
6681 
6682 	rcu_read_lock();
6683 	rdev = md_find_rdev_nr_rcu(mddev, info.number);
6684 	if (rdev) {
6685 		info.major = MAJOR(rdev->bdev->bd_dev);
6686 		info.minor = MINOR(rdev->bdev->bd_dev);
6687 		info.raid_disk = rdev->raid_disk;
6688 		info.state = 0;
6689 		if (test_bit(Faulty, &rdev->flags))
6690 			info.state |= (1<<MD_DISK_FAULTY);
6691 		else if (test_bit(In_sync, &rdev->flags)) {
6692 			info.state |= (1<<MD_DISK_ACTIVE);
6693 			info.state |= (1<<MD_DISK_SYNC);
6694 		}
6695 		if (test_bit(Journal, &rdev->flags))
6696 			info.state |= (1<<MD_DISK_JOURNAL);
6697 		if (test_bit(WriteMostly, &rdev->flags))
6698 			info.state |= (1<<MD_DISK_WRITEMOSTLY);
6699 		if (test_bit(FailFast, &rdev->flags))
6700 			info.state |= (1<<MD_DISK_FAILFAST);
6701 	} else {
6702 		info.major = info.minor = 0;
6703 		info.raid_disk = -1;
6704 		info.state = (1<<MD_DISK_REMOVED);
6705 	}
6706 	rcu_read_unlock();
6707 
6708 	if (copy_to_user(arg, &info, sizeof(info)))
6709 		return -EFAULT;
6710 
6711 	return 0;
6712 }
6713 
6714 int md_add_new_disk(struct mddev *mddev, struct mdu_disk_info_s *info)
6715 {
6716 	struct md_rdev *rdev;
6717 	dev_t dev = MKDEV(info->major,info->minor);
6718 
6719 	if (mddev_is_clustered(mddev) &&
6720 		!(info->state & ((1 << MD_DISK_CLUSTER_ADD) | (1 << MD_DISK_CANDIDATE)))) {
6721 		pr_warn("%s: Cannot add to clustered mddev.\n",
6722 			mdname(mddev));
6723 		return -EINVAL;
6724 	}
6725 
6726 	if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
6727 		return -EOVERFLOW;
6728 
6729 	if (!mddev->raid_disks) {
6730 		int err;
6731 		/* expecting a device which has a superblock */
6732 		rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
6733 		if (IS_ERR(rdev)) {
6734 			pr_warn("md: md_import_device returned %ld\n",
6735 				PTR_ERR(rdev));
6736 			return PTR_ERR(rdev);
6737 		}
6738 		if (!list_empty(&mddev->disks)) {
6739 			struct md_rdev *rdev0
6740 				= list_entry(mddev->disks.next,
6741 					     struct md_rdev, same_set);
6742 			err = super_types[mddev->major_version]
6743 				.load_super(rdev, rdev0, mddev->minor_version);
6744 			if (err < 0) {
6745 				pr_warn("md: %pg has different UUID to %pg\n",
6746 					rdev->bdev,
6747 					rdev0->bdev);
6748 				export_rdev(rdev, mddev);
6749 				return -EINVAL;
6750 			}
6751 		}
6752 		err = bind_rdev_to_array(rdev, mddev);
6753 		if (err)
6754 			export_rdev(rdev, mddev);
6755 		return err;
6756 	}
6757 
6758 	/*
6759 	 * md_add_new_disk can be used once the array is assembled
6760 	 * to add "hot spares".  They must already have a superblock
6761 	 * written
6762 	 */
6763 	if (mddev->pers) {
6764 		int err;
6765 		if (!mddev->pers->hot_add_disk) {
6766 			pr_warn("%s: personality does not support diskops!\n",
6767 				mdname(mddev));
6768 			return -EINVAL;
6769 		}
6770 		if (mddev->persistent)
6771 			rdev = md_import_device(dev, mddev->major_version,
6772 						mddev->minor_version);
6773 		else
6774 			rdev = md_import_device(dev, -1, -1);
6775 		if (IS_ERR(rdev)) {
6776 			pr_warn("md: md_import_device returned %ld\n",
6777 				PTR_ERR(rdev));
6778 			return PTR_ERR(rdev);
6779 		}
6780 		/* set saved_raid_disk if appropriate */
6781 		if (!mddev->persistent) {
6782 			if (info->state & (1<<MD_DISK_SYNC)  &&
6783 			    info->raid_disk < mddev->raid_disks) {
6784 				rdev->raid_disk = info->raid_disk;
6785 				clear_bit(Bitmap_sync, &rdev->flags);
6786 			} else
6787 				rdev->raid_disk = -1;
6788 			rdev->saved_raid_disk = rdev->raid_disk;
6789 		} else
6790 			super_types[mddev->major_version].
6791 				validate_super(mddev, rdev);
6792 		if ((info->state & (1<<MD_DISK_SYNC)) &&
6793 		     rdev->raid_disk != info->raid_disk) {
6794 			/* This was a hot-add request, but events doesn't
6795 			 * match, so reject it.
6796 			 */
6797 			export_rdev(rdev, mddev);
6798 			return -EINVAL;
6799 		}
6800 
6801 		clear_bit(In_sync, &rdev->flags); /* just to be sure */
6802 		if (info->state & (1<<MD_DISK_WRITEMOSTLY))
6803 			set_bit(WriteMostly, &rdev->flags);
6804 		else
6805 			clear_bit(WriteMostly, &rdev->flags);
6806 		if (info->state & (1<<MD_DISK_FAILFAST))
6807 			set_bit(FailFast, &rdev->flags);
6808 		else
6809 			clear_bit(FailFast, &rdev->flags);
6810 
6811 		if (info->state & (1<<MD_DISK_JOURNAL)) {
6812 			struct md_rdev *rdev2;
6813 			bool has_journal = false;
6814 
6815 			/* make sure no existing journal disk */
6816 			rdev_for_each(rdev2, mddev) {
6817 				if (test_bit(Journal, &rdev2->flags)) {
6818 					has_journal = true;
6819 					break;
6820 				}
6821 			}
6822 			if (has_journal || mddev->bitmap) {
6823 				export_rdev(rdev, mddev);
6824 				return -EBUSY;
6825 			}
6826 			set_bit(Journal, &rdev->flags);
6827 		}
6828 		/*
6829 		 * check whether the device shows up in other nodes
6830 		 */
6831 		if (mddev_is_clustered(mddev)) {
6832 			if (info->state & (1 << MD_DISK_CANDIDATE))
6833 				set_bit(Candidate, &rdev->flags);
6834 			else if (info->state & (1 << MD_DISK_CLUSTER_ADD)) {
6835 				/* --add initiated by this node */
6836 				err = md_cluster_ops->add_new_disk(mddev, rdev);
6837 				if (err) {
6838 					export_rdev(rdev, mddev);
6839 					return err;
6840 				}
6841 			}
6842 		}
6843 
6844 		rdev->raid_disk = -1;
6845 		err = bind_rdev_to_array(rdev, mddev);
6846 
6847 		if (err)
6848 			export_rdev(rdev, mddev);
6849 
6850 		if (mddev_is_clustered(mddev)) {
6851 			if (info->state & (1 << MD_DISK_CANDIDATE)) {
6852 				if (!err) {
6853 					err = md_cluster_ops->new_disk_ack(mddev,
6854 						err == 0);
6855 					if (err)
6856 						md_kick_rdev_from_array(rdev);
6857 				}
6858 			} else {
6859 				if (err)
6860 					md_cluster_ops->add_new_disk_cancel(mddev);
6861 				else
6862 					err = add_bound_rdev(rdev);
6863 			}
6864 
6865 		} else if (!err)
6866 			err = add_bound_rdev(rdev);
6867 
6868 		return err;
6869 	}
6870 
6871 	/* otherwise, md_add_new_disk is only allowed
6872 	 * for major_version==0 superblocks
6873 	 */
6874 	if (mddev->major_version != 0) {
6875 		pr_warn("%s: ADD_NEW_DISK not supported\n", mdname(mddev));
6876 		return -EINVAL;
6877 	}
6878 
6879 	if (!(info->state & (1<<MD_DISK_FAULTY))) {
6880 		int err;
6881 		rdev = md_import_device(dev, -1, 0);
6882 		if (IS_ERR(rdev)) {
6883 			pr_warn("md: error, md_import_device() returned %ld\n",
6884 				PTR_ERR(rdev));
6885 			return PTR_ERR(rdev);
6886 		}
6887 		rdev->desc_nr = info->number;
6888 		if (info->raid_disk < mddev->raid_disks)
6889 			rdev->raid_disk = info->raid_disk;
6890 		else
6891 			rdev->raid_disk = -1;
6892 
6893 		if (rdev->raid_disk < mddev->raid_disks)
6894 			if (info->state & (1<<MD_DISK_SYNC))
6895 				set_bit(In_sync, &rdev->flags);
6896 
6897 		if (info->state & (1<<MD_DISK_WRITEMOSTLY))
6898 			set_bit(WriteMostly, &rdev->flags);
6899 		if (info->state & (1<<MD_DISK_FAILFAST))
6900 			set_bit(FailFast, &rdev->flags);
6901 
6902 		if (!mddev->persistent) {
6903 			pr_debug("md: nonpersistent superblock ...\n");
6904 			rdev->sb_start = bdev_nr_sectors(rdev->bdev);
6905 		} else
6906 			rdev->sb_start = calc_dev_sboffset(rdev);
6907 		rdev->sectors = rdev->sb_start;
6908 
6909 		err = bind_rdev_to_array(rdev, mddev);
6910 		if (err) {
6911 			export_rdev(rdev, mddev);
6912 			return err;
6913 		}
6914 	}
6915 
6916 	return 0;
6917 }
6918 
6919 static int hot_remove_disk(struct mddev *mddev, dev_t dev)
6920 {
6921 	struct md_rdev *rdev;
6922 
6923 	if (!mddev->pers)
6924 		return -ENODEV;
6925 
6926 	rdev = find_rdev(mddev, dev);
6927 	if (!rdev)
6928 		return -ENXIO;
6929 
6930 	if (rdev->raid_disk < 0)
6931 		goto kick_rdev;
6932 
6933 	clear_bit(Blocked, &rdev->flags);
6934 	remove_and_add_spares(mddev, rdev);
6935 
6936 	if (rdev->raid_disk >= 0)
6937 		goto busy;
6938 
6939 kick_rdev:
6940 	if (mddev_is_clustered(mddev)) {
6941 		if (md_cluster_ops->remove_disk(mddev, rdev))
6942 			goto busy;
6943 	}
6944 
6945 	md_kick_rdev_from_array(rdev);
6946 	set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
6947 	if (mddev->thread)
6948 		md_wakeup_thread(mddev->thread);
6949 	else
6950 		md_update_sb(mddev, 1);
6951 	md_new_event();
6952 
6953 	return 0;
6954 busy:
6955 	pr_debug("md: cannot remove active disk %pg from %s ...\n",
6956 		 rdev->bdev, mdname(mddev));
6957 	return -EBUSY;
6958 }
6959 
6960 static int hot_add_disk(struct mddev *mddev, dev_t dev)
6961 {
6962 	int err;
6963 	struct md_rdev *rdev;
6964 
6965 	if (!mddev->pers)
6966 		return -ENODEV;
6967 
6968 	if (mddev->major_version != 0) {
6969 		pr_warn("%s: HOT_ADD may only be used with version-0 superblocks.\n",
6970 			mdname(mddev));
6971 		return -EINVAL;
6972 	}
6973 	if (!mddev->pers->hot_add_disk) {
6974 		pr_warn("%s: personality does not support diskops!\n",
6975 			mdname(mddev));
6976 		return -EINVAL;
6977 	}
6978 
6979 	rdev = md_import_device(dev, -1, 0);
6980 	if (IS_ERR(rdev)) {
6981 		pr_warn("md: error, md_import_device() returned %ld\n",
6982 			PTR_ERR(rdev));
6983 		return -EINVAL;
6984 	}
6985 
6986 	if (mddev->persistent)
6987 		rdev->sb_start = calc_dev_sboffset(rdev);
6988 	else
6989 		rdev->sb_start = bdev_nr_sectors(rdev->bdev);
6990 
6991 	rdev->sectors = rdev->sb_start;
6992 
6993 	if (test_bit(Faulty, &rdev->flags)) {
6994 		pr_warn("md: can not hot-add faulty %pg disk to %s!\n",
6995 			rdev->bdev, mdname(mddev));
6996 		err = -EINVAL;
6997 		goto abort_export;
6998 	}
6999 
7000 	clear_bit(In_sync, &rdev->flags);
7001 	rdev->desc_nr = -1;
7002 	rdev->saved_raid_disk = -1;
7003 	err = bind_rdev_to_array(rdev, mddev);
7004 	if (err)
7005 		goto abort_export;
7006 
7007 	/*
7008 	 * The rest should better be atomic, we can have disk failures
7009 	 * noticed in interrupt contexts ...
7010 	 */
7011 
7012 	rdev->raid_disk = -1;
7013 
7014 	set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
7015 	if (!mddev->thread)
7016 		md_update_sb(mddev, 1);
7017 	/*
7018 	 * If the new disk does not support REQ_NOWAIT,
7019 	 * disable on the whole MD.
7020 	 */
7021 	if (!bdev_nowait(rdev->bdev)) {
7022 		pr_info("%s: Disabling nowait because %pg does not support nowait\n",
7023 			mdname(mddev), rdev->bdev);
7024 		blk_queue_flag_clear(QUEUE_FLAG_NOWAIT, mddev->queue);
7025 	}
7026 	/*
7027 	 * Kick recovery, maybe this spare has to be added to the
7028 	 * array immediately.
7029 	 */
7030 	set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7031 	md_wakeup_thread(mddev->thread);
7032 	md_new_event();
7033 	return 0;
7034 
7035 abort_export:
7036 	export_rdev(rdev, mddev);
7037 	return err;
7038 }
7039 
7040 static int set_bitmap_file(struct mddev *mddev, int fd)
7041 {
7042 	int err = 0;
7043 
7044 	if (mddev->pers) {
7045 		if (!mddev->pers->quiesce || !mddev->thread)
7046 			return -EBUSY;
7047 		if (mddev->recovery || mddev->sync_thread)
7048 			return -EBUSY;
7049 		/* we should be able to change the bitmap.. */
7050 	}
7051 
7052 	if (fd >= 0) {
7053 		struct inode *inode;
7054 		struct file *f;
7055 
7056 		if (mddev->bitmap || mddev->bitmap_info.file)
7057 			return -EEXIST; /* cannot add when bitmap is present */
7058 
7059 		if (!IS_ENABLED(CONFIG_MD_BITMAP_FILE)) {
7060 			pr_warn("%s: bitmap files not supported by this kernel\n",
7061 				mdname(mddev));
7062 			return -EINVAL;
7063 		}
7064 		pr_warn("%s: using deprecated bitmap file support\n",
7065 			mdname(mddev));
7066 
7067 		f = fget(fd);
7068 
7069 		if (f == NULL) {
7070 			pr_warn("%s: error: failed to get bitmap file\n",
7071 				mdname(mddev));
7072 			return -EBADF;
7073 		}
7074 
7075 		inode = f->f_mapping->host;
7076 		if (!S_ISREG(inode->i_mode)) {
7077 			pr_warn("%s: error: bitmap file must be a regular file\n",
7078 				mdname(mddev));
7079 			err = -EBADF;
7080 		} else if (!(f->f_mode & FMODE_WRITE)) {
7081 			pr_warn("%s: error: bitmap file must open for write\n",
7082 				mdname(mddev));
7083 			err = -EBADF;
7084 		} else if (atomic_read(&inode->i_writecount) != 1) {
7085 			pr_warn("%s: error: bitmap file is already in use\n",
7086 				mdname(mddev));
7087 			err = -EBUSY;
7088 		}
7089 		if (err) {
7090 			fput(f);
7091 			return err;
7092 		}
7093 		mddev->bitmap_info.file = f;
7094 		mddev->bitmap_info.offset = 0; /* file overrides offset */
7095 	} else if (mddev->bitmap == NULL)
7096 		return -ENOENT; /* cannot remove what isn't there */
7097 	err = 0;
7098 	if (mddev->pers) {
7099 		if (fd >= 0) {
7100 			struct bitmap *bitmap;
7101 
7102 			bitmap = md_bitmap_create(mddev, -1);
7103 			mddev_suspend(mddev);
7104 			if (!IS_ERR(bitmap)) {
7105 				mddev->bitmap = bitmap;
7106 				err = md_bitmap_load(mddev);
7107 			} else
7108 				err = PTR_ERR(bitmap);
7109 			if (err) {
7110 				md_bitmap_destroy(mddev);
7111 				fd = -1;
7112 			}
7113 			mddev_resume(mddev);
7114 		} else if (fd < 0) {
7115 			mddev_suspend(mddev);
7116 			md_bitmap_destroy(mddev);
7117 			mddev_resume(mddev);
7118 		}
7119 	}
7120 	if (fd < 0) {
7121 		struct file *f = mddev->bitmap_info.file;
7122 		if (f) {
7123 			spin_lock(&mddev->lock);
7124 			mddev->bitmap_info.file = NULL;
7125 			spin_unlock(&mddev->lock);
7126 			fput(f);
7127 		}
7128 	}
7129 
7130 	return err;
7131 }
7132 
7133 /*
7134  * md_set_array_info is used two different ways
7135  * The original usage is when creating a new array.
7136  * In this usage, raid_disks is > 0 and it together with
7137  *  level, size, not_persistent,layout,chunksize determine the
7138  *  shape of the array.
7139  *  This will always create an array with a type-0.90.0 superblock.
7140  * The newer usage is when assembling an array.
7141  *  In this case raid_disks will be 0, and the major_version field is
7142  *  use to determine which style super-blocks are to be found on the devices.
7143  *  The minor and patch _version numbers are also kept incase the
7144  *  super_block handler wishes to interpret them.
7145  */
7146 int md_set_array_info(struct mddev *mddev, struct mdu_array_info_s *info)
7147 {
7148 	if (info->raid_disks == 0) {
7149 		/* just setting version number for superblock loading */
7150 		if (info->major_version < 0 ||
7151 		    info->major_version >= ARRAY_SIZE(super_types) ||
7152 		    super_types[info->major_version].name == NULL) {
7153 			/* maybe try to auto-load a module? */
7154 			pr_warn("md: superblock version %d not known\n",
7155 				info->major_version);
7156 			return -EINVAL;
7157 		}
7158 		mddev->major_version = info->major_version;
7159 		mddev->minor_version = info->minor_version;
7160 		mddev->patch_version = info->patch_version;
7161 		mddev->persistent = !info->not_persistent;
7162 		/* ensure mddev_put doesn't delete this now that there
7163 		 * is some minimal configuration.
7164 		 */
7165 		mddev->ctime         = ktime_get_real_seconds();
7166 		return 0;
7167 	}
7168 	mddev->major_version = MD_MAJOR_VERSION;
7169 	mddev->minor_version = MD_MINOR_VERSION;
7170 	mddev->patch_version = MD_PATCHLEVEL_VERSION;
7171 	mddev->ctime         = ktime_get_real_seconds();
7172 
7173 	mddev->level         = info->level;
7174 	mddev->clevel[0]     = 0;
7175 	mddev->dev_sectors   = 2 * (sector_t)info->size;
7176 	mddev->raid_disks    = info->raid_disks;
7177 	/* don't set md_minor, it is determined by which /dev/md* was
7178 	 * openned
7179 	 */
7180 	if (info->state & (1<<MD_SB_CLEAN))
7181 		mddev->recovery_cp = MaxSector;
7182 	else
7183 		mddev->recovery_cp = 0;
7184 	mddev->persistent    = ! info->not_persistent;
7185 	mddev->external	     = 0;
7186 
7187 	mddev->layout        = info->layout;
7188 	if (mddev->level == 0)
7189 		/* Cannot trust RAID0 layout info here */
7190 		mddev->layout = -1;
7191 	mddev->chunk_sectors = info->chunk_size >> 9;
7192 
7193 	if (mddev->persistent) {
7194 		mddev->max_disks = MD_SB_DISKS;
7195 		mddev->flags = 0;
7196 		mddev->sb_flags = 0;
7197 	}
7198 	set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
7199 
7200 	mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9;
7201 	mddev->bitmap_info.default_space = 64*2 - (MD_SB_BYTES >> 9);
7202 	mddev->bitmap_info.offset = 0;
7203 
7204 	mddev->reshape_position = MaxSector;
7205 
7206 	/*
7207 	 * Generate a 128 bit UUID
7208 	 */
7209 	get_random_bytes(mddev->uuid, 16);
7210 
7211 	mddev->new_level = mddev->level;
7212 	mddev->new_chunk_sectors = mddev->chunk_sectors;
7213 	mddev->new_layout = mddev->layout;
7214 	mddev->delta_disks = 0;
7215 	mddev->reshape_backwards = 0;
7216 
7217 	return 0;
7218 }
7219 
7220 void md_set_array_sectors(struct mddev *mddev, sector_t array_sectors)
7221 {
7222 	lockdep_assert_held(&mddev->reconfig_mutex);
7223 
7224 	if (mddev->external_size)
7225 		return;
7226 
7227 	mddev->array_sectors = array_sectors;
7228 }
7229 EXPORT_SYMBOL(md_set_array_sectors);
7230 
7231 static int update_size(struct mddev *mddev, sector_t num_sectors)
7232 {
7233 	struct md_rdev *rdev;
7234 	int rv;
7235 	int fit = (num_sectors == 0);
7236 	sector_t old_dev_sectors = mddev->dev_sectors;
7237 
7238 	if (mddev->pers->resize == NULL)
7239 		return -EINVAL;
7240 	/* The "num_sectors" is the number of sectors of each device that
7241 	 * is used.  This can only make sense for arrays with redundancy.
7242 	 * linear and raid0 always use whatever space is available. We can only
7243 	 * consider changing this number if no resync or reconstruction is
7244 	 * happening, and if the new size is acceptable. It must fit before the
7245 	 * sb_start or, if that is <data_offset, it must fit before the size
7246 	 * of each device.  If num_sectors is zero, we find the largest size
7247 	 * that fits.
7248 	 */
7249 	if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
7250 	    mddev->sync_thread)
7251 		return -EBUSY;
7252 	if (!md_is_rdwr(mddev))
7253 		return -EROFS;
7254 
7255 	rdev_for_each(rdev, mddev) {
7256 		sector_t avail = rdev->sectors;
7257 
7258 		if (fit && (num_sectors == 0 || num_sectors > avail))
7259 			num_sectors = avail;
7260 		if (avail < num_sectors)
7261 			return -ENOSPC;
7262 	}
7263 	rv = mddev->pers->resize(mddev, num_sectors);
7264 	if (!rv) {
7265 		if (mddev_is_clustered(mddev))
7266 			md_cluster_ops->update_size(mddev, old_dev_sectors);
7267 		else if (mddev->queue) {
7268 			set_capacity_and_notify(mddev->gendisk,
7269 						mddev->array_sectors);
7270 		}
7271 	}
7272 	return rv;
7273 }
7274 
7275 static int update_raid_disks(struct mddev *mddev, int raid_disks)
7276 {
7277 	int rv;
7278 	struct md_rdev *rdev;
7279 	/* change the number of raid disks */
7280 	if (mddev->pers->check_reshape == NULL)
7281 		return -EINVAL;
7282 	if (!md_is_rdwr(mddev))
7283 		return -EROFS;
7284 	if (raid_disks <= 0 ||
7285 	    (mddev->max_disks && raid_disks >= mddev->max_disks))
7286 		return -EINVAL;
7287 	if (mddev->sync_thread ||
7288 	    test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
7289 	    test_bit(MD_RESYNCING_REMOTE, &mddev->recovery) ||
7290 	    mddev->reshape_position != MaxSector)
7291 		return -EBUSY;
7292 
7293 	rdev_for_each(rdev, mddev) {
7294 		if (mddev->raid_disks < raid_disks &&
7295 		    rdev->data_offset < rdev->new_data_offset)
7296 			return -EINVAL;
7297 		if (mddev->raid_disks > raid_disks &&
7298 		    rdev->data_offset > rdev->new_data_offset)
7299 			return -EINVAL;
7300 	}
7301 
7302 	mddev->delta_disks = raid_disks - mddev->raid_disks;
7303 	if (mddev->delta_disks < 0)
7304 		mddev->reshape_backwards = 1;
7305 	else if (mddev->delta_disks > 0)
7306 		mddev->reshape_backwards = 0;
7307 
7308 	rv = mddev->pers->check_reshape(mddev);
7309 	if (rv < 0) {
7310 		mddev->delta_disks = 0;
7311 		mddev->reshape_backwards = 0;
7312 	}
7313 	return rv;
7314 }
7315 
7316 /*
7317  * update_array_info is used to change the configuration of an
7318  * on-line array.
7319  * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
7320  * fields in the info are checked against the array.
7321  * Any differences that cannot be handled will cause an error.
7322  * Normally, only one change can be managed at a time.
7323  */
7324 static int update_array_info(struct mddev *mddev, mdu_array_info_t *info)
7325 {
7326 	int rv = 0;
7327 	int cnt = 0;
7328 	int state = 0;
7329 
7330 	/* calculate expected state,ignoring low bits */
7331 	if (mddev->bitmap && mddev->bitmap_info.offset)
7332 		state |= (1 << MD_SB_BITMAP_PRESENT);
7333 
7334 	if (mddev->major_version != info->major_version ||
7335 	    mddev->minor_version != info->minor_version ||
7336 /*	    mddev->patch_version != info->patch_version || */
7337 	    mddev->ctime         != info->ctime         ||
7338 	    mddev->level         != info->level         ||
7339 /*	    mddev->layout        != info->layout        || */
7340 	    mddev->persistent	 != !info->not_persistent ||
7341 	    mddev->chunk_sectors != info->chunk_size >> 9 ||
7342 	    /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
7343 	    ((state^info->state) & 0xfffffe00)
7344 		)
7345 		return -EINVAL;
7346 	/* Check there is only one change */
7347 	if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
7348 		cnt++;
7349 	if (mddev->raid_disks != info->raid_disks)
7350 		cnt++;
7351 	if (mddev->layout != info->layout)
7352 		cnt++;
7353 	if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT))
7354 		cnt++;
7355 	if (cnt == 0)
7356 		return 0;
7357 	if (cnt > 1)
7358 		return -EINVAL;
7359 
7360 	if (mddev->layout != info->layout) {
7361 		/* Change layout
7362 		 * we don't need to do anything at the md level, the
7363 		 * personality will take care of it all.
7364 		 */
7365 		if (mddev->pers->check_reshape == NULL)
7366 			return -EINVAL;
7367 		else {
7368 			mddev->new_layout = info->layout;
7369 			rv = mddev->pers->check_reshape(mddev);
7370 			if (rv)
7371 				mddev->new_layout = mddev->layout;
7372 			return rv;
7373 		}
7374 	}
7375 	if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
7376 		rv = update_size(mddev, (sector_t)info->size * 2);
7377 
7378 	if (mddev->raid_disks    != info->raid_disks)
7379 		rv = update_raid_disks(mddev, info->raid_disks);
7380 
7381 	if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) {
7382 		if (mddev->pers->quiesce == NULL || mddev->thread == NULL) {
7383 			rv = -EINVAL;
7384 			goto err;
7385 		}
7386 		if (mddev->recovery || mddev->sync_thread) {
7387 			rv = -EBUSY;
7388 			goto err;
7389 		}
7390 		if (info->state & (1<<MD_SB_BITMAP_PRESENT)) {
7391 			struct bitmap *bitmap;
7392 			/* add the bitmap */
7393 			if (mddev->bitmap) {
7394 				rv = -EEXIST;
7395 				goto err;
7396 			}
7397 			if (mddev->bitmap_info.default_offset == 0) {
7398 				rv = -EINVAL;
7399 				goto err;
7400 			}
7401 			mddev->bitmap_info.offset =
7402 				mddev->bitmap_info.default_offset;
7403 			mddev->bitmap_info.space =
7404 				mddev->bitmap_info.default_space;
7405 			bitmap = md_bitmap_create(mddev, -1);
7406 			mddev_suspend(mddev);
7407 			if (!IS_ERR(bitmap)) {
7408 				mddev->bitmap = bitmap;
7409 				rv = md_bitmap_load(mddev);
7410 			} else
7411 				rv = PTR_ERR(bitmap);
7412 			if (rv)
7413 				md_bitmap_destroy(mddev);
7414 			mddev_resume(mddev);
7415 		} else {
7416 			/* remove the bitmap */
7417 			if (!mddev->bitmap) {
7418 				rv = -ENOENT;
7419 				goto err;
7420 			}
7421 			if (mddev->bitmap->storage.file) {
7422 				rv = -EINVAL;
7423 				goto err;
7424 			}
7425 			if (mddev->bitmap_info.nodes) {
7426 				/* hold PW on all the bitmap lock */
7427 				if (md_cluster_ops->lock_all_bitmaps(mddev) <= 0) {
7428 					pr_warn("md: can't change bitmap to none since the array is in use by more than one node\n");
7429 					rv = -EPERM;
7430 					md_cluster_ops->unlock_all_bitmaps(mddev);
7431 					goto err;
7432 				}
7433 
7434 				mddev->bitmap_info.nodes = 0;
7435 				md_cluster_ops->leave(mddev);
7436 				module_put(md_cluster_mod);
7437 				mddev->safemode_delay = DEFAULT_SAFEMODE_DELAY;
7438 			}
7439 			mddev_suspend(mddev);
7440 			md_bitmap_destroy(mddev);
7441 			mddev_resume(mddev);
7442 			mddev->bitmap_info.offset = 0;
7443 		}
7444 	}
7445 	md_update_sb(mddev, 1);
7446 	return rv;
7447 err:
7448 	return rv;
7449 }
7450 
7451 static int set_disk_faulty(struct mddev *mddev, dev_t dev)
7452 {
7453 	struct md_rdev *rdev;
7454 	int err = 0;
7455 
7456 	if (mddev->pers == NULL)
7457 		return -ENODEV;
7458 
7459 	rcu_read_lock();
7460 	rdev = md_find_rdev_rcu(mddev, dev);
7461 	if (!rdev)
7462 		err =  -ENODEV;
7463 	else {
7464 		md_error(mddev, rdev);
7465 		if (test_bit(MD_BROKEN, &mddev->flags))
7466 			err = -EBUSY;
7467 	}
7468 	rcu_read_unlock();
7469 	return err;
7470 }
7471 
7472 /*
7473  * We have a problem here : there is no easy way to give a CHS
7474  * virtual geometry. We currently pretend that we have a 2 heads
7475  * 4 sectors (with a BIG number of cylinders...). This drives
7476  * dosfs just mad... ;-)
7477  */
7478 static int md_getgeo(struct block_device *bdev, struct hd_geometry *geo)
7479 {
7480 	struct mddev *mddev = bdev->bd_disk->private_data;
7481 
7482 	geo->heads = 2;
7483 	geo->sectors = 4;
7484 	geo->cylinders = mddev->array_sectors / 8;
7485 	return 0;
7486 }
7487 
7488 static inline bool md_ioctl_valid(unsigned int cmd)
7489 {
7490 	switch (cmd) {
7491 	case ADD_NEW_DISK:
7492 	case GET_ARRAY_INFO:
7493 	case GET_BITMAP_FILE:
7494 	case GET_DISK_INFO:
7495 	case HOT_ADD_DISK:
7496 	case HOT_REMOVE_DISK:
7497 	case RAID_VERSION:
7498 	case RESTART_ARRAY_RW:
7499 	case RUN_ARRAY:
7500 	case SET_ARRAY_INFO:
7501 	case SET_BITMAP_FILE:
7502 	case SET_DISK_FAULTY:
7503 	case STOP_ARRAY:
7504 	case STOP_ARRAY_RO:
7505 	case CLUSTERED_DISK_NACK:
7506 		return true;
7507 	default:
7508 		return false;
7509 	}
7510 }
7511 
7512 static int __md_set_array_info(struct mddev *mddev, void __user *argp)
7513 {
7514 	mdu_array_info_t info;
7515 	int err;
7516 
7517 	if (!argp)
7518 		memset(&info, 0, sizeof(info));
7519 	else if (copy_from_user(&info, argp, sizeof(info)))
7520 		return -EFAULT;
7521 
7522 	if (mddev->pers) {
7523 		err = update_array_info(mddev, &info);
7524 		if (err)
7525 			pr_warn("md: couldn't update array info. %d\n", err);
7526 		return err;
7527 	}
7528 
7529 	if (!list_empty(&mddev->disks)) {
7530 		pr_warn("md: array %s already has disks!\n", mdname(mddev));
7531 		return -EBUSY;
7532 	}
7533 
7534 	if (mddev->raid_disks) {
7535 		pr_warn("md: array %s already initialised!\n", mdname(mddev));
7536 		return -EBUSY;
7537 	}
7538 
7539 	err = md_set_array_info(mddev, &info);
7540 	if (err)
7541 		pr_warn("md: couldn't set array info. %d\n", err);
7542 
7543 	return err;
7544 }
7545 
7546 static int md_ioctl(struct block_device *bdev, blk_mode_t mode,
7547 			unsigned int cmd, unsigned long arg)
7548 {
7549 	int err = 0;
7550 	void __user *argp = (void __user *)arg;
7551 	struct mddev *mddev = NULL;
7552 	bool did_set_md_closing = false;
7553 
7554 	if (!md_ioctl_valid(cmd))
7555 		return -ENOTTY;
7556 
7557 	switch (cmd) {
7558 	case RAID_VERSION:
7559 	case GET_ARRAY_INFO:
7560 	case GET_DISK_INFO:
7561 		break;
7562 	default:
7563 		if (!capable(CAP_SYS_ADMIN))
7564 			return -EACCES;
7565 	}
7566 
7567 	/*
7568 	 * Commands dealing with the RAID driver but not any
7569 	 * particular array:
7570 	 */
7571 	switch (cmd) {
7572 	case RAID_VERSION:
7573 		err = get_version(argp);
7574 		goto out;
7575 	default:;
7576 	}
7577 
7578 	/*
7579 	 * Commands creating/starting a new array:
7580 	 */
7581 
7582 	mddev = bdev->bd_disk->private_data;
7583 
7584 	if (!mddev) {
7585 		BUG();
7586 		goto out;
7587 	}
7588 
7589 	/* Some actions do not requires the mutex */
7590 	switch (cmd) {
7591 	case GET_ARRAY_INFO:
7592 		if (!mddev->raid_disks && !mddev->external)
7593 			err = -ENODEV;
7594 		else
7595 			err = get_array_info(mddev, argp);
7596 		goto out;
7597 
7598 	case GET_DISK_INFO:
7599 		if (!mddev->raid_disks && !mddev->external)
7600 			err = -ENODEV;
7601 		else
7602 			err = get_disk_info(mddev, argp);
7603 		goto out;
7604 
7605 	case SET_DISK_FAULTY:
7606 		err = set_disk_faulty(mddev, new_decode_dev(arg));
7607 		goto out;
7608 
7609 	case GET_BITMAP_FILE:
7610 		err = get_bitmap_file(mddev, argp);
7611 		goto out;
7612 
7613 	}
7614 
7615 	if (cmd == HOT_REMOVE_DISK)
7616 		/* need to ensure recovery thread has run */
7617 		wait_event_interruptible_timeout(mddev->sb_wait,
7618 						 !test_bit(MD_RECOVERY_NEEDED,
7619 							   &mddev->recovery),
7620 						 msecs_to_jiffies(5000));
7621 	if (cmd == STOP_ARRAY || cmd == STOP_ARRAY_RO) {
7622 		/* Need to flush page cache, and ensure no-one else opens
7623 		 * and writes
7624 		 */
7625 		mutex_lock(&mddev->open_mutex);
7626 		if (mddev->pers && atomic_read(&mddev->openers) > 1) {
7627 			mutex_unlock(&mddev->open_mutex);
7628 			err = -EBUSY;
7629 			goto out;
7630 		}
7631 		if (test_and_set_bit(MD_CLOSING, &mddev->flags)) {
7632 			mutex_unlock(&mddev->open_mutex);
7633 			err = -EBUSY;
7634 			goto out;
7635 		}
7636 		did_set_md_closing = true;
7637 		mutex_unlock(&mddev->open_mutex);
7638 		sync_blockdev(bdev);
7639 	}
7640 	err = mddev_lock(mddev);
7641 	if (err) {
7642 		pr_debug("md: ioctl lock interrupted, reason %d, cmd %d\n",
7643 			 err, cmd);
7644 		goto out;
7645 	}
7646 
7647 	if (cmd == SET_ARRAY_INFO) {
7648 		err = __md_set_array_info(mddev, argp);
7649 		goto unlock;
7650 	}
7651 
7652 	/*
7653 	 * Commands querying/configuring an existing array:
7654 	 */
7655 	/* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
7656 	 * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */
7657 	if ((!mddev->raid_disks && !mddev->external)
7658 	    && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY
7659 	    && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE
7660 	    && cmd != GET_BITMAP_FILE) {
7661 		err = -ENODEV;
7662 		goto unlock;
7663 	}
7664 
7665 	/*
7666 	 * Commands even a read-only array can execute:
7667 	 */
7668 	switch (cmd) {
7669 	case RESTART_ARRAY_RW:
7670 		err = restart_array(mddev);
7671 		goto unlock;
7672 
7673 	case STOP_ARRAY:
7674 		err = do_md_stop(mddev, 0, bdev);
7675 		goto unlock;
7676 
7677 	case STOP_ARRAY_RO:
7678 		err = md_set_readonly(mddev, bdev);
7679 		goto unlock;
7680 
7681 	case HOT_REMOVE_DISK:
7682 		err = hot_remove_disk(mddev, new_decode_dev(arg));
7683 		goto unlock;
7684 
7685 	case ADD_NEW_DISK:
7686 		/* We can support ADD_NEW_DISK on read-only arrays
7687 		 * only if we are re-adding a preexisting device.
7688 		 * So require mddev->pers and MD_DISK_SYNC.
7689 		 */
7690 		if (mddev->pers) {
7691 			mdu_disk_info_t info;
7692 			if (copy_from_user(&info, argp, sizeof(info)))
7693 				err = -EFAULT;
7694 			else if (!(info.state & (1<<MD_DISK_SYNC)))
7695 				/* Need to clear read-only for this */
7696 				break;
7697 			else
7698 				err = md_add_new_disk(mddev, &info);
7699 			goto unlock;
7700 		}
7701 		break;
7702 	}
7703 
7704 	/*
7705 	 * The remaining ioctls are changing the state of the
7706 	 * superblock, so we do not allow them on read-only arrays.
7707 	 */
7708 	if (!md_is_rdwr(mddev) && mddev->pers) {
7709 		if (mddev->ro != MD_AUTO_READ) {
7710 			err = -EROFS;
7711 			goto unlock;
7712 		}
7713 		mddev->ro = MD_RDWR;
7714 		sysfs_notify_dirent_safe(mddev->sysfs_state);
7715 		set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7716 		/* mddev_unlock will wake thread */
7717 		/* If a device failed while we were read-only, we
7718 		 * need to make sure the metadata is updated now.
7719 		 */
7720 		if (test_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags)) {
7721 			mddev_unlock(mddev);
7722 			wait_event(mddev->sb_wait,
7723 				   !test_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags) &&
7724 				   !test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags));
7725 			mddev_lock_nointr(mddev);
7726 		}
7727 	}
7728 
7729 	switch (cmd) {
7730 	case ADD_NEW_DISK:
7731 	{
7732 		mdu_disk_info_t info;
7733 		if (copy_from_user(&info, argp, sizeof(info)))
7734 			err = -EFAULT;
7735 		else
7736 			err = md_add_new_disk(mddev, &info);
7737 		goto unlock;
7738 	}
7739 
7740 	case CLUSTERED_DISK_NACK:
7741 		if (mddev_is_clustered(mddev))
7742 			md_cluster_ops->new_disk_ack(mddev, false);
7743 		else
7744 			err = -EINVAL;
7745 		goto unlock;
7746 
7747 	case HOT_ADD_DISK:
7748 		err = hot_add_disk(mddev, new_decode_dev(arg));
7749 		goto unlock;
7750 
7751 	case RUN_ARRAY:
7752 		err = do_md_run(mddev);
7753 		goto unlock;
7754 
7755 	case SET_BITMAP_FILE:
7756 		err = set_bitmap_file(mddev, (int)arg);
7757 		goto unlock;
7758 
7759 	default:
7760 		err = -EINVAL;
7761 		goto unlock;
7762 	}
7763 
7764 unlock:
7765 	if (mddev->hold_active == UNTIL_IOCTL &&
7766 	    err != -EINVAL)
7767 		mddev->hold_active = 0;
7768 	mddev_unlock(mddev);
7769 out:
7770 	if(did_set_md_closing)
7771 		clear_bit(MD_CLOSING, &mddev->flags);
7772 	return err;
7773 }
7774 #ifdef CONFIG_COMPAT
7775 static int md_compat_ioctl(struct block_device *bdev, blk_mode_t mode,
7776 		    unsigned int cmd, unsigned long arg)
7777 {
7778 	switch (cmd) {
7779 	case HOT_REMOVE_DISK:
7780 	case HOT_ADD_DISK:
7781 	case SET_DISK_FAULTY:
7782 	case SET_BITMAP_FILE:
7783 		/* These take in integer arg, do not convert */
7784 		break;
7785 	default:
7786 		arg = (unsigned long)compat_ptr(arg);
7787 		break;
7788 	}
7789 
7790 	return md_ioctl(bdev, mode, cmd, arg);
7791 }
7792 #endif /* CONFIG_COMPAT */
7793 
7794 static int md_set_read_only(struct block_device *bdev, bool ro)
7795 {
7796 	struct mddev *mddev = bdev->bd_disk->private_data;
7797 	int err;
7798 
7799 	err = mddev_lock(mddev);
7800 	if (err)
7801 		return err;
7802 
7803 	if (!mddev->raid_disks && !mddev->external) {
7804 		err = -ENODEV;
7805 		goto out_unlock;
7806 	}
7807 
7808 	/*
7809 	 * Transitioning to read-auto need only happen for arrays that call
7810 	 * md_write_start and which are not ready for writes yet.
7811 	 */
7812 	if (!ro && mddev->ro == MD_RDONLY && mddev->pers) {
7813 		err = restart_array(mddev);
7814 		if (err)
7815 			goto out_unlock;
7816 		mddev->ro = MD_AUTO_READ;
7817 	}
7818 
7819 out_unlock:
7820 	mddev_unlock(mddev);
7821 	return err;
7822 }
7823 
7824 static int md_open(struct gendisk *disk, blk_mode_t mode)
7825 {
7826 	struct mddev *mddev;
7827 	int err;
7828 
7829 	spin_lock(&all_mddevs_lock);
7830 	mddev = mddev_get(disk->private_data);
7831 	spin_unlock(&all_mddevs_lock);
7832 	if (!mddev)
7833 		return -ENODEV;
7834 
7835 	err = mutex_lock_interruptible(&mddev->open_mutex);
7836 	if (err)
7837 		goto out;
7838 
7839 	err = -ENODEV;
7840 	if (test_bit(MD_CLOSING, &mddev->flags))
7841 		goto out_unlock;
7842 
7843 	atomic_inc(&mddev->openers);
7844 	mutex_unlock(&mddev->open_mutex);
7845 
7846 	disk_check_media_change(disk);
7847 	return 0;
7848 
7849 out_unlock:
7850 	mutex_unlock(&mddev->open_mutex);
7851 out:
7852 	mddev_put(mddev);
7853 	return err;
7854 }
7855 
7856 static void md_release(struct gendisk *disk)
7857 {
7858 	struct mddev *mddev = disk->private_data;
7859 
7860 	BUG_ON(!mddev);
7861 	atomic_dec(&mddev->openers);
7862 	mddev_put(mddev);
7863 }
7864 
7865 static unsigned int md_check_events(struct gendisk *disk, unsigned int clearing)
7866 {
7867 	struct mddev *mddev = disk->private_data;
7868 	unsigned int ret = 0;
7869 
7870 	if (mddev->changed)
7871 		ret = DISK_EVENT_MEDIA_CHANGE;
7872 	mddev->changed = 0;
7873 	return ret;
7874 }
7875 
7876 static void md_free_disk(struct gendisk *disk)
7877 {
7878 	struct mddev *mddev = disk->private_data;
7879 
7880 	percpu_ref_exit(&mddev->writes_pending);
7881 	mddev_free(mddev);
7882 }
7883 
7884 const struct block_device_operations md_fops =
7885 {
7886 	.owner		= THIS_MODULE,
7887 	.submit_bio	= md_submit_bio,
7888 	.open		= md_open,
7889 	.release	= md_release,
7890 	.ioctl		= md_ioctl,
7891 #ifdef CONFIG_COMPAT
7892 	.compat_ioctl	= md_compat_ioctl,
7893 #endif
7894 	.getgeo		= md_getgeo,
7895 	.check_events	= md_check_events,
7896 	.set_read_only	= md_set_read_only,
7897 	.free_disk	= md_free_disk,
7898 };
7899 
7900 static int md_thread(void *arg)
7901 {
7902 	struct md_thread *thread = arg;
7903 
7904 	/*
7905 	 * md_thread is a 'system-thread', it's priority should be very
7906 	 * high. We avoid resource deadlocks individually in each
7907 	 * raid personality. (RAID5 does preallocation) We also use RR and
7908 	 * the very same RT priority as kswapd, thus we will never get
7909 	 * into a priority inversion deadlock.
7910 	 *
7911 	 * we definitely have to have equal or higher priority than
7912 	 * bdflush, otherwise bdflush will deadlock if there are too
7913 	 * many dirty RAID5 blocks.
7914 	 */
7915 
7916 	allow_signal(SIGKILL);
7917 	while (!kthread_should_stop()) {
7918 
7919 		/* We need to wait INTERRUPTIBLE so that
7920 		 * we don't add to the load-average.
7921 		 * That means we need to be sure no signals are
7922 		 * pending
7923 		 */
7924 		if (signal_pending(current))
7925 			flush_signals(current);
7926 
7927 		wait_event_interruptible_timeout
7928 			(thread->wqueue,
7929 			 test_bit(THREAD_WAKEUP, &thread->flags)
7930 			 || kthread_should_stop() || kthread_should_park(),
7931 			 thread->timeout);
7932 
7933 		clear_bit(THREAD_WAKEUP, &thread->flags);
7934 		if (kthread_should_park())
7935 			kthread_parkme();
7936 		if (!kthread_should_stop())
7937 			thread->run(thread);
7938 	}
7939 
7940 	return 0;
7941 }
7942 
7943 static void md_wakeup_thread_directly(struct md_thread __rcu *thread)
7944 {
7945 	struct md_thread *t;
7946 
7947 	rcu_read_lock();
7948 	t = rcu_dereference(thread);
7949 	if (t)
7950 		wake_up_process(t->tsk);
7951 	rcu_read_unlock();
7952 }
7953 
7954 void md_wakeup_thread(struct md_thread __rcu *thread)
7955 {
7956 	struct md_thread *t;
7957 
7958 	rcu_read_lock();
7959 	t = rcu_dereference(thread);
7960 	if (t) {
7961 		pr_debug("md: waking up MD thread %s.\n", t->tsk->comm);
7962 		set_bit(THREAD_WAKEUP, &t->flags);
7963 		wake_up(&t->wqueue);
7964 	}
7965 	rcu_read_unlock();
7966 }
7967 EXPORT_SYMBOL(md_wakeup_thread);
7968 
7969 struct md_thread *md_register_thread(void (*run) (struct md_thread *),
7970 		struct mddev *mddev, const char *name)
7971 {
7972 	struct md_thread *thread;
7973 
7974 	thread = kzalloc(sizeof(struct md_thread), GFP_KERNEL);
7975 	if (!thread)
7976 		return NULL;
7977 
7978 	init_waitqueue_head(&thread->wqueue);
7979 
7980 	thread->run = run;
7981 	thread->mddev = mddev;
7982 	thread->timeout = MAX_SCHEDULE_TIMEOUT;
7983 	thread->tsk = kthread_run(md_thread, thread,
7984 				  "%s_%s",
7985 				  mdname(thread->mddev),
7986 				  name);
7987 	if (IS_ERR(thread->tsk)) {
7988 		kfree(thread);
7989 		return NULL;
7990 	}
7991 	return thread;
7992 }
7993 EXPORT_SYMBOL(md_register_thread);
7994 
7995 void md_unregister_thread(struct mddev *mddev, struct md_thread __rcu **threadp)
7996 {
7997 	struct md_thread *thread = rcu_dereference_protected(*threadp,
7998 					lockdep_is_held(&mddev->reconfig_mutex));
7999 
8000 	if (!thread)
8001 		return;
8002 
8003 	rcu_assign_pointer(*threadp, NULL);
8004 	synchronize_rcu();
8005 
8006 	pr_debug("interrupting MD-thread pid %d\n", task_pid_nr(thread->tsk));
8007 	kthread_stop(thread->tsk);
8008 	kfree(thread);
8009 }
8010 EXPORT_SYMBOL(md_unregister_thread);
8011 
8012 void md_error(struct mddev *mddev, struct md_rdev *rdev)
8013 {
8014 	if (!rdev || test_bit(Faulty, &rdev->flags))
8015 		return;
8016 
8017 	if (!mddev->pers || !mddev->pers->error_handler)
8018 		return;
8019 	mddev->pers->error_handler(mddev, rdev);
8020 
8021 	if (mddev->pers->level == 0 || mddev->pers->level == LEVEL_LINEAR)
8022 		return;
8023 
8024 	if (mddev->degraded && !test_bit(MD_BROKEN, &mddev->flags))
8025 		set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
8026 	sysfs_notify_dirent_safe(rdev->sysfs_state);
8027 	set_bit(MD_RECOVERY_INTR, &mddev->recovery);
8028 	if (!test_bit(MD_BROKEN, &mddev->flags)) {
8029 		set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
8030 		md_wakeup_thread(mddev->thread);
8031 	}
8032 	if (mddev->event_work.func)
8033 		queue_work(md_misc_wq, &mddev->event_work);
8034 	md_new_event();
8035 }
8036 EXPORT_SYMBOL(md_error);
8037 
8038 /* seq_file implementation /proc/mdstat */
8039 
8040 static void status_unused(struct seq_file *seq)
8041 {
8042 	int i = 0;
8043 	struct md_rdev *rdev;
8044 
8045 	seq_printf(seq, "unused devices: ");
8046 
8047 	list_for_each_entry(rdev, &pending_raid_disks, same_set) {
8048 		i++;
8049 		seq_printf(seq, "%pg ", rdev->bdev);
8050 	}
8051 	if (!i)
8052 		seq_printf(seq, "<none>");
8053 
8054 	seq_printf(seq, "\n");
8055 }
8056 
8057 static int status_resync(struct seq_file *seq, struct mddev *mddev)
8058 {
8059 	sector_t max_sectors, resync, res;
8060 	unsigned long dt, db = 0;
8061 	sector_t rt, curr_mark_cnt, resync_mark_cnt;
8062 	int scale, recovery_active;
8063 	unsigned int per_milli;
8064 
8065 	if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ||
8066 	    test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
8067 		max_sectors = mddev->resync_max_sectors;
8068 	else
8069 		max_sectors = mddev->dev_sectors;
8070 
8071 	resync = mddev->curr_resync;
8072 	if (resync < MD_RESYNC_ACTIVE) {
8073 		if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
8074 			/* Still cleaning up */
8075 			resync = max_sectors;
8076 	} else if (resync > max_sectors) {
8077 		resync = max_sectors;
8078 	} else {
8079 		res = atomic_read(&mddev->recovery_active);
8080 		/*
8081 		 * Resync has started, but the subtraction has overflowed or
8082 		 * yielded one of the special values. Force it to active to
8083 		 * ensure the status reports an active resync.
8084 		 */
8085 		if (resync < res || resync - res < MD_RESYNC_ACTIVE)
8086 			resync = MD_RESYNC_ACTIVE;
8087 		else
8088 			resync -= res;
8089 	}
8090 
8091 	if (resync == MD_RESYNC_NONE) {
8092 		if (test_bit(MD_RESYNCING_REMOTE, &mddev->recovery)) {
8093 			struct md_rdev *rdev;
8094 
8095 			rdev_for_each(rdev, mddev)
8096 				if (rdev->raid_disk >= 0 &&
8097 				    !test_bit(Faulty, &rdev->flags) &&
8098 				    rdev->recovery_offset != MaxSector &&
8099 				    rdev->recovery_offset) {
8100 					seq_printf(seq, "\trecover=REMOTE");
8101 					return 1;
8102 				}
8103 			if (mddev->reshape_position != MaxSector)
8104 				seq_printf(seq, "\treshape=REMOTE");
8105 			else
8106 				seq_printf(seq, "\tresync=REMOTE");
8107 			return 1;
8108 		}
8109 		if (mddev->recovery_cp < MaxSector) {
8110 			seq_printf(seq, "\tresync=PENDING");
8111 			return 1;
8112 		}
8113 		return 0;
8114 	}
8115 	if (resync < MD_RESYNC_ACTIVE) {
8116 		seq_printf(seq, "\tresync=DELAYED");
8117 		return 1;
8118 	}
8119 
8120 	WARN_ON(max_sectors == 0);
8121 	/* Pick 'scale' such that (resync>>scale)*1000 will fit
8122 	 * in a sector_t, and (max_sectors>>scale) will fit in a
8123 	 * u32, as those are the requirements for sector_div.
8124 	 * Thus 'scale' must be at least 10
8125 	 */
8126 	scale = 10;
8127 	if (sizeof(sector_t) > sizeof(unsigned long)) {
8128 		while ( max_sectors/2 > (1ULL<<(scale+32)))
8129 			scale++;
8130 	}
8131 	res = (resync>>scale)*1000;
8132 	sector_div(res, (u32)((max_sectors>>scale)+1));
8133 
8134 	per_milli = res;
8135 	{
8136 		int i, x = per_milli/50, y = 20-x;
8137 		seq_printf(seq, "[");
8138 		for (i = 0; i < x; i++)
8139 			seq_printf(seq, "=");
8140 		seq_printf(seq, ">");
8141 		for (i = 0; i < y; i++)
8142 			seq_printf(seq, ".");
8143 		seq_printf(seq, "] ");
8144 	}
8145 	seq_printf(seq, " %s =%3u.%u%% (%llu/%llu)",
8146 		   (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)?
8147 		    "reshape" :
8148 		    (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)?
8149 		     "check" :
8150 		     (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
8151 		      "resync" : "recovery"))),
8152 		   per_milli/10, per_milli % 10,
8153 		   (unsigned long long) resync/2,
8154 		   (unsigned long long) max_sectors/2);
8155 
8156 	/*
8157 	 * dt: time from mark until now
8158 	 * db: blocks written from mark until now
8159 	 * rt: remaining time
8160 	 *
8161 	 * rt is a sector_t, which is always 64bit now. We are keeping
8162 	 * the original algorithm, but it is not really necessary.
8163 	 *
8164 	 * Original algorithm:
8165 	 *   So we divide before multiply in case it is 32bit and close
8166 	 *   to the limit.
8167 	 *   We scale the divisor (db) by 32 to avoid losing precision
8168 	 *   near the end of resync when the number of remaining sectors
8169 	 *   is close to 'db'.
8170 	 *   We then divide rt by 32 after multiplying by db to compensate.
8171 	 *   The '+1' avoids division by zero if db is very small.
8172 	 */
8173 	dt = ((jiffies - mddev->resync_mark) / HZ);
8174 	if (!dt) dt++;
8175 
8176 	curr_mark_cnt = mddev->curr_mark_cnt;
8177 	recovery_active = atomic_read(&mddev->recovery_active);
8178 	resync_mark_cnt = mddev->resync_mark_cnt;
8179 
8180 	if (curr_mark_cnt >= (recovery_active + resync_mark_cnt))
8181 		db = curr_mark_cnt - (recovery_active + resync_mark_cnt);
8182 
8183 	rt = max_sectors - resync;    /* number of remaining sectors */
8184 	rt = div64_u64(rt, db/32+1);
8185 	rt *= dt;
8186 	rt >>= 5;
8187 
8188 	seq_printf(seq, " finish=%lu.%lumin", (unsigned long)rt / 60,
8189 		   ((unsigned long)rt % 60)/6);
8190 
8191 	seq_printf(seq, " speed=%ldK/sec", db/2/dt);
8192 	return 1;
8193 }
8194 
8195 static void *md_seq_start(struct seq_file *seq, loff_t *pos)
8196 {
8197 	struct list_head *tmp;
8198 	loff_t l = *pos;
8199 	struct mddev *mddev;
8200 
8201 	if (l == 0x10000) {
8202 		++*pos;
8203 		return (void *)2;
8204 	}
8205 	if (l > 0x10000)
8206 		return NULL;
8207 	if (!l--)
8208 		/* header */
8209 		return (void*)1;
8210 
8211 	spin_lock(&all_mddevs_lock);
8212 	list_for_each(tmp,&all_mddevs)
8213 		if (!l--) {
8214 			mddev = list_entry(tmp, struct mddev, all_mddevs);
8215 			if (!mddev_get(mddev))
8216 				continue;
8217 			spin_unlock(&all_mddevs_lock);
8218 			return mddev;
8219 		}
8220 	spin_unlock(&all_mddevs_lock);
8221 	if (!l--)
8222 		return (void*)2;/* tail */
8223 	return NULL;
8224 }
8225 
8226 static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
8227 {
8228 	struct list_head *tmp;
8229 	struct mddev *next_mddev, *mddev = v;
8230 	struct mddev *to_put = NULL;
8231 
8232 	++*pos;
8233 	if (v == (void*)2)
8234 		return NULL;
8235 
8236 	spin_lock(&all_mddevs_lock);
8237 	if (v == (void*)1) {
8238 		tmp = all_mddevs.next;
8239 	} else {
8240 		to_put = mddev;
8241 		tmp = mddev->all_mddevs.next;
8242 	}
8243 
8244 	for (;;) {
8245 		if (tmp == &all_mddevs) {
8246 			next_mddev = (void*)2;
8247 			*pos = 0x10000;
8248 			break;
8249 		}
8250 		next_mddev = list_entry(tmp, struct mddev, all_mddevs);
8251 		if (mddev_get(next_mddev))
8252 			break;
8253 		mddev = next_mddev;
8254 		tmp = mddev->all_mddevs.next;
8255 	}
8256 	spin_unlock(&all_mddevs_lock);
8257 
8258 	if (to_put)
8259 		mddev_put(mddev);
8260 	return next_mddev;
8261 
8262 }
8263 
8264 static void md_seq_stop(struct seq_file *seq, void *v)
8265 {
8266 	struct mddev *mddev = v;
8267 
8268 	if (mddev && v != (void*)1 && v != (void*)2)
8269 		mddev_put(mddev);
8270 }
8271 
8272 static int md_seq_show(struct seq_file *seq, void *v)
8273 {
8274 	struct mddev *mddev = v;
8275 	sector_t sectors;
8276 	struct md_rdev *rdev;
8277 
8278 	if (v == (void*)1) {
8279 		struct md_personality *pers;
8280 		seq_printf(seq, "Personalities : ");
8281 		spin_lock(&pers_lock);
8282 		list_for_each_entry(pers, &pers_list, list)
8283 			seq_printf(seq, "[%s] ", pers->name);
8284 
8285 		spin_unlock(&pers_lock);
8286 		seq_printf(seq, "\n");
8287 		seq->poll_event = atomic_read(&md_event_count);
8288 		return 0;
8289 	}
8290 	if (v == (void*)2) {
8291 		status_unused(seq);
8292 		return 0;
8293 	}
8294 
8295 	spin_lock(&mddev->lock);
8296 	if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
8297 		seq_printf(seq, "%s : %sactive", mdname(mddev),
8298 						mddev->pers ? "" : "in");
8299 		if (mddev->pers) {
8300 			if (mddev->ro == MD_RDONLY)
8301 				seq_printf(seq, " (read-only)");
8302 			if (mddev->ro == MD_AUTO_READ)
8303 				seq_printf(seq, " (auto-read-only)");
8304 			seq_printf(seq, " %s", mddev->pers->name);
8305 		}
8306 
8307 		sectors = 0;
8308 		rcu_read_lock();
8309 		rdev_for_each_rcu(rdev, mddev) {
8310 			seq_printf(seq, " %pg[%d]", rdev->bdev, rdev->desc_nr);
8311 
8312 			if (test_bit(WriteMostly, &rdev->flags))
8313 				seq_printf(seq, "(W)");
8314 			if (test_bit(Journal, &rdev->flags))
8315 				seq_printf(seq, "(J)");
8316 			if (test_bit(Faulty, &rdev->flags)) {
8317 				seq_printf(seq, "(F)");
8318 				continue;
8319 			}
8320 			if (rdev->raid_disk < 0)
8321 				seq_printf(seq, "(S)"); /* spare */
8322 			if (test_bit(Replacement, &rdev->flags))
8323 				seq_printf(seq, "(R)");
8324 			sectors += rdev->sectors;
8325 		}
8326 		rcu_read_unlock();
8327 
8328 		if (!list_empty(&mddev->disks)) {
8329 			if (mddev->pers)
8330 				seq_printf(seq, "\n      %llu blocks",
8331 					   (unsigned long long)
8332 					   mddev->array_sectors / 2);
8333 			else
8334 				seq_printf(seq, "\n      %llu blocks",
8335 					   (unsigned long long)sectors / 2);
8336 		}
8337 		if (mddev->persistent) {
8338 			if (mddev->major_version != 0 ||
8339 			    mddev->minor_version != 90) {
8340 				seq_printf(seq," super %d.%d",
8341 					   mddev->major_version,
8342 					   mddev->minor_version);
8343 			}
8344 		} else if (mddev->external)
8345 			seq_printf(seq, " super external:%s",
8346 				   mddev->metadata_type);
8347 		else
8348 			seq_printf(seq, " super non-persistent");
8349 
8350 		if (mddev->pers) {
8351 			mddev->pers->status(seq, mddev);
8352 			seq_printf(seq, "\n      ");
8353 			if (mddev->pers->sync_request) {
8354 				if (status_resync(seq, mddev))
8355 					seq_printf(seq, "\n      ");
8356 			}
8357 		} else
8358 			seq_printf(seq, "\n       ");
8359 
8360 		md_bitmap_status(seq, mddev->bitmap);
8361 
8362 		seq_printf(seq, "\n");
8363 	}
8364 	spin_unlock(&mddev->lock);
8365 
8366 	return 0;
8367 }
8368 
8369 static const struct seq_operations md_seq_ops = {
8370 	.start  = md_seq_start,
8371 	.next   = md_seq_next,
8372 	.stop   = md_seq_stop,
8373 	.show   = md_seq_show,
8374 };
8375 
8376 static int md_seq_open(struct inode *inode, struct file *file)
8377 {
8378 	struct seq_file *seq;
8379 	int error;
8380 
8381 	error = seq_open(file, &md_seq_ops);
8382 	if (error)
8383 		return error;
8384 
8385 	seq = file->private_data;
8386 	seq->poll_event = atomic_read(&md_event_count);
8387 	return error;
8388 }
8389 
8390 static int md_unloading;
8391 static __poll_t mdstat_poll(struct file *filp, poll_table *wait)
8392 {
8393 	struct seq_file *seq = filp->private_data;
8394 	__poll_t mask;
8395 
8396 	if (md_unloading)
8397 		return EPOLLIN|EPOLLRDNORM|EPOLLERR|EPOLLPRI;
8398 	poll_wait(filp, &md_event_waiters, wait);
8399 
8400 	/* always allow read */
8401 	mask = EPOLLIN | EPOLLRDNORM;
8402 
8403 	if (seq->poll_event != atomic_read(&md_event_count))
8404 		mask |= EPOLLERR | EPOLLPRI;
8405 	return mask;
8406 }
8407 
8408 static const struct proc_ops mdstat_proc_ops = {
8409 	.proc_open	= md_seq_open,
8410 	.proc_read	= seq_read,
8411 	.proc_lseek	= seq_lseek,
8412 	.proc_release	= seq_release,
8413 	.proc_poll	= mdstat_poll,
8414 };
8415 
8416 int register_md_personality(struct md_personality *p)
8417 {
8418 	pr_debug("md: %s personality registered for level %d\n",
8419 		 p->name, p->level);
8420 	spin_lock(&pers_lock);
8421 	list_add_tail(&p->list, &pers_list);
8422 	spin_unlock(&pers_lock);
8423 	return 0;
8424 }
8425 EXPORT_SYMBOL(register_md_personality);
8426 
8427 int unregister_md_personality(struct md_personality *p)
8428 {
8429 	pr_debug("md: %s personality unregistered\n", p->name);
8430 	spin_lock(&pers_lock);
8431 	list_del_init(&p->list);
8432 	spin_unlock(&pers_lock);
8433 	return 0;
8434 }
8435 EXPORT_SYMBOL(unregister_md_personality);
8436 
8437 int register_md_cluster_operations(struct md_cluster_operations *ops,
8438 				   struct module *module)
8439 {
8440 	int ret = 0;
8441 	spin_lock(&pers_lock);
8442 	if (md_cluster_ops != NULL)
8443 		ret = -EALREADY;
8444 	else {
8445 		md_cluster_ops = ops;
8446 		md_cluster_mod = module;
8447 	}
8448 	spin_unlock(&pers_lock);
8449 	return ret;
8450 }
8451 EXPORT_SYMBOL(register_md_cluster_operations);
8452 
8453 int unregister_md_cluster_operations(void)
8454 {
8455 	spin_lock(&pers_lock);
8456 	md_cluster_ops = NULL;
8457 	spin_unlock(&pers_lock);
8458 	return 0;
8459 }
8460 EXPORT_SYMBOL(unregister_md_cluster_operations);
8461 
8462 int md_setup_cluster(struct mddev *mddev, int nodes)
8463 {
8464 	int ret;
8465 	if (!md_cluster_ops)
8466 		request_module("md-cluster");
8467 	spin_lock(&pers_lock);
8468 	/* ensure module won't be unloaded */
8469 	if (!md_cluster_ops || !try_module_get(md_cluster_mod)) {
8470 		pr_warn("can't find md-cluster module or get its reference.\n");
8471 		spin_unlock(&pers_lock);
8472 		return -ENOENT;
8473 	}
8474 	spin_unlock(&pers_lock);
8475 
8476 	ret = md_cluster_ops->join(mddev, nodes);
8477 	if (!ret)
8478 		mddev->safemode_delay = 0;
8479 	return ret;
8480 }
8481 
8482 void md_cluster_stop(struct mddev *mddev)
8483 {
8484 	if (!md_cluster_ops)
8485 		return;
8486 	md_cluster_ops->leave(mddev);
8487 	module_put(md_cluster_mod);
8488 }
8489 
8490 static int is_mddev_idle(struct mddev *mddev, int init)
8491 {
8492 	struct md_rdev *rdev;
8493 	int idle;
8494 	int curr_events;
8495 
8496 	idle = 1;
8497 	rcu_read_lock();
8498 	rdev_for_each_rcu(rdev, mddev) {
8499 		struct gendisk *disk = rdev->bdev->bd_disk;
8500 		curr_events = (int)part_stat_read_accum(disk->part0, sectors) -
8501 			      atomic_read(&disk->sync_io);
8502 		/* sync IO will cause sync_io to increase before the disk_stats
8503 		 * as sync_io is counted when a request starts, and
8504 		 * disk_stats is counted when it completes.
8505 		 * So resync activity will cause curr_events to be smaller than
8506 		 * when there was no such activity.
8507 		 * non-sync IO will cause disk_stat to increase without
8508 		 * increasing sync_io so curr_events will (eventually)
8509 		 * be larger than it was before.  Once it becomes
8510 		 * substantially larger, the test below will cause
8511 		 * the array to appear non-idle, and resync will slow
8512 		 * down.
8513 		 * If there is a lot of outstanding resync activity when
8514 		 * we set last_event to curr_events, then all that activity
8515 		 * completing might cause the array to appear non-idle
8516 		 * and resync will be slowed down even though there might
8517 		 * not have been non-resync activity.  This will only
8518 		 * happen once though.  'last_events' will soon reflect
8519 		 * the state where there is little or no outstanding
8520 		 * resync requests, and further resync activity will
8521 		 * always make curr_events less than last_events.
8522 		 *
8523 		 */
8524 		if (init || curr_events - rdev->last_events > 64) {
8525 			rdev->last_events = curr_events;
8526 			idle = 0;
8527 		}
8528 	}
8529 	rcu_read_unlock();
8530 	return idle;
8531 }
8532 
8533 void md_done_sync(struct mddev *mddev, int blocks, int ok)
8534 {
8535 	/* another "blocks" (512byte) blocks have been synced */
8536 	atomic_sub(blocks, &mddev->recovery_active);
8537 	wake_up(&mddev->recovery_wait);
8538 	if (!ok) {
8539 		set_bit(MD_RECOVERY_INTR, &mddev->recovery);
8540 		set_bit(MD_RECOVERY_ERROR, &mddev->recovery);
8541 		md_wakeup_thread(mddev->thread);
8542 		// stop recovery, signal do_sync ....
8543 	}
8544 }
8545 EXPORT_SYMBOL(md_done_sync);
8546 
8547 /* md_write_start(mddev, bi)
8548  * If we need to update some array metadata (e.g. 'active' flag
8549  * in superblock) before writing, schedule a superblock update
8550  * and wait for it to complete.
8551  * A return value of 'false' means that the write wasn't recorded
8552  * and cannot proceed as the array is being suspend.
8553  */
8554 bool md_write_start(struct mddev *mddev, struct bio *bi)
8555 {
8556 	int did_change = 0;
8557 
8558 	if (bio_data_dir(bi) != WRITE)
8559 		return true;
8560 
8561 	BUG_ON(mddev->ro == MD_RDONLY);
8562 	if (mddev->ro == MD_AUTO_READ) {
8563 		/* need to switch to read/write */
8564 		mddev->ro = MD_RDWR;
8565 		set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
8566 		md_wakeup_thread(mddev->thread);
8567 		md_wakeup_thread(mddev->sync_thread);
8568 		did_change = 1;
8569 	}
8570 	rcu_read_lock();
8571 	percpu_ref_get(&mddev->writes_pending);
8572 	smp_mb(); /* Match smp_mb in set_in_sync() */
8573 	if (mddev->safemode == 1)
8574 		mddev->safemode = 0;
8575 	/* sync_checkers is always 0 when writes_pending is in per-cpu mode */
8576 	if (mddev->in_sync || mddev->sync_checkers) {
8577 		spin_lock(&mddev->lock);
8578 		if (mddev->in_sync) {
8579 			mddev->in_sync = 0;
8580 			set_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags);
8581 			set_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags);
8582 			md_wakeup_thread(mddev->thread);
8583 			did_change = 1;
8584 		}
8585 		spin_unlock(&mddev->lock);
8586 	}
8587 	rcu_read_unlock();
8588 	if (did_change)
8589 		sysfs_notify_dirent_safe(mddev->sysfs_state);
8590 	if (!mddev->has_superblocks)
8591 		return true;
8592 	wait_event(mddev->sb_wait,
8593 		   !test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags) ||
8594 		   is_md_suspended(mddev));
8595 	if (test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags)) {
8596 		percpu_ref_put(&mddev->writes_pending);
8597 		return false;
8598 	}
8599 	return true;
8600 }
8601 EXPORT_SYMBOL(md_write_start);
8602 
8603 /* md_write_inc can only be called when md_write_start() has
8604  * already been called at least once of the current request.
8605  * It increments the counter and is useful when a single request
8606  * is split into several parts.  Each part causes an increment and
8607  * so needs a matching md_write_end().
8608  * Unlike md_write_start(), it is safe to call md_write_inc() inside
8609  * a spinlocked region.
8610  */
8611 void md_write_inc(struct mddev *mddev, struct bio *bi)
8612 {
8613 	if (bio_data_dir(bi) != WRITE)
8614 		return;
8615 	WARN_ON_ONCE(mddev->in_sync || !md_is_rdwr(mddev));
8616 	percpu_ref_get(&mddev->writes_pending);
8617 }
8618 EXPORT_SYMBOL(md_write_inc);
8619 
8620 void md_write_end(struct mddev *mddev)
8621 {
8622 	percpu_ref_put(&mddev->writes_pending);
8623 
8624 	if (mddev->safemode == 2)
8625 		md_wakeup_thread(mddev->thread);
8626 	else if (mddev->safemode_delay)
8627 		/* The roundup() ensures this only performs locking once
8628 		 * every ->safemode_delay jiffies
8629 		 */
8630 		mod_timer(&mddev->safemode_timer,
8631 			  roundup(jiffies, mddev->safemode_delay) +
8632 			  mddev->safemode_delay);
8633 }
8634 
8635 EXPORT_SYMBOL(md_write_end);
8636 
8637 /* This is used by raid0 and raid10 */
8638 void md_submit_discard_bio(struct mddev *mddev, struct md_rdev *rdev,
8639 			struct bio *bio, sector_t start, sector_t size)
8640 {
8641 	struct bio *discard_bio = NULL;
8642 
8643 	if (__blkdev_issue_discard(rdev->bdev, start, size, GFP_NOIO,
8644 			&discard_bio) || !discard_bio)
8645 		return;
8646 
8647 	bio_chain(discard_bio, bio);
8648 	bio_clone_blkg_association(discard_bio, bio);
8649 	if (mddev->gendisk)
8650 		trace_block_bio_remap(discard_bio,
8651 				disk_devt(mddev->gendisk),
8652 				bio->bi_iter.bi_sector);
8653 	submit_bio_noacct(discard_bio);
8654 }
8655 EXPORT_SYMBOL_GPL(md_submit_discard_bio);
8656 
8657 static void md_end_clone_io(struct bio *bio)
8658 {
8659 	struct md_io_clone *md_io_clone = bio->bi_private;
8660 	struct bio *orig_bio = md_io_clone->orig_bio;
8661 	struct mddev *mddev = md_io_clone->mddev;
8662 
8663 	orig_bio->bi_status = bio->bi_status;
8664 
8665 	if (md_io_clone->start_time)
8666 		bio_end_io_acct(orig_bio, md_io_clone->start_time);
8667 
8668 	bio_put(bio);
8669 	bio_endio(orig_bio);
8670 	percpu_ref_put(&mddev->active_io);
8671 }
8672 
8673 static void md_clone_bio(struct mddev *mddev, struct bio **bio)
8674 {
8675 	struct block_device *bdev = (*bio)->bi_bdev;
8676 	struct md_io_clone *md_io_clone;
8677 	struct bio *clone =
8678 		bio_alloc_clone(bdev, *bio, GFP_NOIO, &mddev->io_clone_set);
8679 
8680 	md_io_clone = container_of(clone, struct md_io_clone, bio_clone);
8681 	md_io_clone->orig_bio = *bio;
8682 	md_io_clone->mddev = mddev;
8683 	if (blk_queue_io_stat(bdev->bd_disk->queue))
8684 		md_io_clone->start_time = bio_start_io_acct(*bio);
8685 
8686 	clone->bi_end_io = md_end_clone_io;
8687 	clone->bi_private = md_io_clone;
8688 	*bio = clone;
8689 }
8690 
8691 void md_account_bio(struct mddev *mddev, struct bio **bio)
8692 {
8693 	percpu_ref_get(&mddev->active_io);
8694 	md_clone_bio(mddev, bio);
8695 }
8696 EXPORT_SYMBOL_GPL(md_account_bio);
8697 
8698 /* md_allow_write(mddev)
8699  * Calling this ensures that the array is marked 'active' so that writes
8700  * may proceed without blocking.  It is important to call this before
8701  * attempting a GFP_KERNEL allocation while holding the mddev lock.
8702  * Must be called with mddev_lock held.
8703  */
8704 void md_allow_write(struct mddev *mddev)
8705 {
8706 	if (!mddev->pers)
8707 		return;
8708 	if (!md_is_rdwr(mddev))
8709 		return;
8710 	if (!mddev->pers->sync_request)
8711 		return;
8712 
8713 	spin_lock(&mddev->lock);
8714 	if (mddev->in_sync) {
8715 		mddev->in_sync = 0;
8716 		set_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags);
8717 		set_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags);
8718 		if (mddev->safemode_delay &&
8719 		    mddev->safemode == 0)
8720 			mddev->safemode = 1;
8721 		spin_unlock(&mddev->lock);
8722 		md_update_sb(mddev, 0);
8723 		sysfs_notify_dirent_safe(mddev->sysfs_state);
8724 		/* wait for the dirty state to be recorded in the metadata */
8725 		wait_event(mddev->sb_wait,
8726 			   !test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags));
8727 	} else
8728 		spin_unlock(&mddev->lock);
8729 }
8730 EXPORT_SYMBOL_GPL(md_allow_write);
8731 
8732 #define SYNC_MARKS	10
8733 #define	SYNC_MARK_STEP	(3*HZ)
8734 #define UPDATE_FREQUENCY (5*60*HZ)
8735 void md_do_sync(struct md_thread *thread)
8736 {
8737 	struct mddev *mddev = thread->mddev;
8738 	struct mddev *mddev2;
8739 	unsigned int currspeed = 0, window;
8740 	sector_t max_sectors,j, io_sectors, recovery_done;
8741 	unsigned long mark[SYNC_MARKS];
8742 	unsigned long update_time;
8743 	sector_t mark_cnt[SYNC_MARKS];
8744 	int last_mark,m;
8745 	sector_t last_check;
8746 	int skipped = 0;
8747 	struct md_rdev *rdev;
8748 	char *desc, *action = NULL;
8749 	struct blk_plug plug;
8750 	int ret;
8751 
8752 	/* just incase thread restarts... */
8753 	if (test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
8754 	    test_bit(MD_RECOVERY_WAIT, &mddev->recovery))
8755 		return;
8756 	if (!md_is_rdwr(mddev)) {/* never try to sync a read-only array */
8757 		set_bit(MD_RECOVERY_INTR, &mddev->recovery);
8758 		return;
8759 	}
8760 
8761 	if (mddev_is_clustered(mddev)) {
8762 		ret = md_cluster_ops->resync_start(mddev);
8763 		if (ret)
8764 			goto skip;
8765 
8766 		set_bit(MD_CLUSTER_RESYNC_LOCKED, &mddev->flags);
8767 		if (!(test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ||
8768 			test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) ||
8769 			test_bit(MD_RECOVERY_RECOVER, &mddev->recovery))
8770 		     && ((unsigned long long)mddev->curr_resync_completed
8771 			 < (unsigned long long)mddev->resync_max_sectors))
8772 			goto skip;
8773 	}
8774 
8775 	if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
8776 		if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)) {
8777 			desc = "data-check";
8778 			action = "check";
8779 		} else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
8780 			desc = "requested-resync";
8781 			action = "repair";
8782 		} else
8783 			desc = "resync";
8784 	} else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
8785 		desc = "reshape";
8786 	else
8787 		desc = "recovery";
8788 
8789 	mddev->last_sync_action = action ?: desc;
8790 
8791 	/*
8792 	 * Before starting a resync we must have set curr_resync to
8793 	 * 2, and then checked that every "conflicting" array has curr_resync
8794 	 * less than ours.  When we find one that is the same or higher
8795 	 * we wait on resync_wait.  To avoid deadlock, we reduce curr_resync
8796 	 * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
8797 	 * This will mean we have to start checking from the beginning again.
8798 	 *
8799 	 */
8800 
8801 	do {
8802 		int mddev2_minor = -1;
8803 		mddev->curr_resync = MD_RESYNC_DELAYED;
8804 
8805 	try_again:
8806 		if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
8807 			goto skip;
8808 		spin_lock(&all_mddevs_lock);
8809 		list_for_each_entry(mddev2, &all_mddevs, all_mddevs) {
8810 			if (test_bit(MD_DELETED, &mddev2->flags))
8811 				continue;
8812 			if (mddev2 == mddev)
8813 				continue;
8814 			if (!mddev->parallel_resync
8815 			&&  mddev2->curr_resync
8816 			&&  match_mddev_units(mddev, mddev2)) {
8817 				DEFINE_WAIT(wq);
8818 				if (mddev < mddev2 &&
8819 				    mddev->curr_resync == MD_RESYNC_DELAYED) {
8820 					/* arbitrarily yield */
8821 					mddev->curr_resync = MD_RESYNC_YIELDED;
8822 					wake_up(&resync_wait);
8823 				}
8824 				if (mddev > mddev2 &&
8825 				    mddev->curr_resync == MD_RESYNC_YIELDED)
8826 					/* no need to wait here, we can wait the next
8827 					 * time 'round when curr_resync == 2
8828 					 */
8829 					continue;
8830 				/* We need to wait 'interruptible' so as not to
8831 				 * contribute to the load average, and not to
8832 				 * be caught by 'softlockup'
8833 				 */
8834 				prepare_to_wait(&resync_wait, &wq, TASK_INTERRUPTIBLE);
8835 				if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
8836 				    mddev2->curr_resync >= mddev->curr_resync) {
8837 					if (mddev2_minor != mddev2->md_minor) {
8838 						mddev2_minor = mddev2->md_minor;
8839 						pr_info("md: delaying %s of %s until %s has finished (they share one or more physical units)\n",
8840 							desc, mdname(mddev),
8841 							mdname(mddev2));
8842 					}
8843 					spin_unlock(&all_mddevs_lock);
8844 
8845 					if (signal_pending(current))
8846 						flush_signals(current);
8847 					schedule();
8848 					finish_wait(&resync_wait, &wq);
8849 					goto try_again;
8850 				}
8851 				finish_wait(&resync_wait, &wq);
8852 			}
8853 		}
8854 		spin_unlock(&all_mddevs_lock);
8855 	} while (mddev->curr_resync < MD_RESYNC_DELAYED);
8856 
8857 	j = 0;
8858 	if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
8859 		/* resync follows the size requested by the personality,
8860 		 * which defaults to physical size, but can be virtual size
8861 		 */
8862 		max_sectors = mddev->resync_max_sectors;
8863 		atomic64_set(&mddev->resync_mismatches, 0);
8864 		/* we don't use the checkpoint if there's a bitmap */
8865 		if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
8866 			j = mddev->resync_min;
8867 		else if (!mddev->bitmap)
8868 			j = mddev->recovery_cp;
8869 
8870 	} else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) {
8871 		max_sectors = mddev->resync_max_sectors;
8872 		/*
8873 		 * If the original node aborts reshaping then we continue the
8874 		 * reshaping, so set j again to avoid restart reshape from the
8875 		 * first beginning
8876 		 */
8877 		if (mddev_is_clustered(mddev) &&
8878 		    mddev->reshape_position != MaxSector)
8879 			j = mddev->reshape_position;
8880 	} else {
8881 		/* recovery follows the physical size of devices */
8882 		max_sectors = mddev->dev_sectors;
8883 		j = MaxSector;
8884 		rcu_read_lock();
8885 		rdev_for_each_rcu(rdev, mddev)
8886 			if (rdev->raid_disk >= 0 &&
8887 			    !test_bit(Journal, &rdev->flags) &&
8888 			    !test_bit(Faulty, &rdev->flags) &&
8889 			    !test_bit(In_sync, &rdev->flags) &&
8890 			    rdev->recovery_offset < j)
8891 				j = rdev->recovery_offset;
8892 		rcu_read_unlock();
8893 
8894 		/* If there is a bitmap, we need to make sure all
8895 		 * writes that started before we added a spare
8896 		 * complete before we start doing a recovery.
8897 		 * Otherwise the write might complete and (via
8898 		 * bitmap_endwrite) set a bit in the bitmap after the
8899 		 * recovery has checked that bit and skipped that
8900 		 * region.
8901 		 */
8902 		if (mddev->bitmap) {
8903 			mddev->pers->quiesce(mddev, 1);
8904 			mddev->pers->quiesce(mddev, 0);
8905 		}
8906 	}
8907 
8908 	pr_info("md: %s of RAID array %s\n", desc, mdname(mddev));
8909 	pr_debug("md: minimum _guaranteed_  speed: %d KB/sec/disk.\n", speed_min(mddev));
8910 	pr_debug("md: using maximum available idle IO bandwidth (but not more than %d KB/sec) for %s.\n",
8911 		 speed_max(mddev), desc);
8912 
8913 	is_mddev_idle(mddev, 1); /* this initializes IO event counters */
8914 
8915 	io_sectors = 0;
8916 	for (m = 0; m < SYNC_MARKS; m++) {
8917 		mark[m] = jiffies;
8918 		mark_cnt[m] = io_sectors;
8919 	}
8920 	last_mark = 0;
8921 	mddev->resync_mark = mark[last_mark];
8922 	mddev->resync_mark_cnt = mark_cnt[last_mark];
8923 
8924 	/*
8925 	 * Tune reconstruction:
8926 	 */
8927 	window = 32 * (PAGE_SIZE / 512);
8928 	pr_debug("md: using %dk window, over a total of %lluk.\n",
8929 		 window/2, (unsigned long long)max_sectors/2);
8930 
8931 	atomic_set(&mddev->recovery_active, 0);
8932 	last_check = 0;
8933 
8934 	if (j >= MD_RESYNC_ACTIVE) {
8935 		pr_debug("md: resuming %s of %s from checkpoint.\n",
8936 			 desc, mdname(mddev));
8937 		mddev->curr_resync = j;
8938 	} else
8939 		mddev->curr_resync = MD_RESYNC_ACTIVE; /* no longer delayed */
8940 	mddev->curr_resync_completed = j;
8941 	sysfs_notify_dirent_safe(mddev->sysfs_completed);
8942 	md_new_event();
8943 	update_time = jiffies;
8944 
8945 	blk_start_plug(&plug);
8946 	while (j < max_sectors) {
8947 		sector_t sectors;
8948 
8949 		skipped = 0;
8950 
8951 		if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
8952 		    ((mddev->curr_resync > mddev->curr_resync_completed &&
8953 		      (mddev->curr_resync - mddev->curr_resync_completed)
8954 		      > (max_sectors >> 4)) ||
8955 		     time_after_eq(jiffies, update_time + UPDATE_FREQUENCY) ||
8956 		     (j - mddev->curr_resync_completed)*2
8957 		     >= mddev->resync_max - mddev->curr_resync_completed ||
8958 		     mddev->curr_resync_completed > mddev->resync_max
8959 			    )) {
8960 			/* time to update curr_resync_completed */
8961 			wait_event(mddev->recovery_wait,
8962 				   atomic_read(&mddev->recovery_active) == 0);
8963 			mddev->curr_resync_completed = j;
8964 			if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) &&
8965 			    j > mddev->recovery_cp)
8966 				mddev->recovery_cp = j;
8967 			update_time = jiffies;
8968 			set_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags);
8969 			sysfs_notify_dirent_safe(mddev->sysfs_completed);
8970 		}
8971 
8972 		while (j >= mddev->resync_max &&
8973 		       !test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
8974 			/* As this condition is controlled by user-space,
8975 			 * we can block indefinitely, so use '_interruptible'
8976 			 * to avoid triggering warnings.
8977 			 */
8978 			flush_signals(current); /* just in case */
8979 			wait_event_interruptible(mddev->recovery_wait,
8980 						 mddev->resync_max > j
8981 						 || test_bit(MD_RECOVERY_INTR,
8982 							     &mddev->recovery));
8983 		}
8984 
8985 		if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
8986 			break;
8987 
8988 		sectors = mddev->pers->sync_request(mddev, j, &skipped);
8989 		if (sectors == 0) {
8990 			set_bit(MD_RECOVERY_INTR, &mddev->recovery);
8991 			break;
8992 		}
8993 
8994 		if (!skipped) { /* actual IO requested */
8995 			io_sectors += sectors;
8996 			atomic_add(sectors, &mddev->recovery_active);
8997 		}
8998 
8999 		if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
9000 			break;
9001 
9002 		j += sectors;
9003 		if (j > max_sectors)
9004 			/* when skipping, extra large numbers can be returned. */
9005 			j = max_sectors;
9006 		if (j >= MD_RESYNC_ACTIVE)
9007 			mddev->curr_resync = j;
9008 		mddev->curr_mark_cnt = io_sectors;
9009 		if (last_check == 0)
9010 			/* this is the earliest that rebuild will be
9011 			 * visible in /proc/mdstat
9012 			 */
9013 			md_new_event();
9014 
9015 		if (last_check + window > io_sectors || j == max_sectors)
9016 			continue;
9017 
9018 		last_check = io_sectors;
9019 	repeat:
9020 		if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) {
9021 			/* step marks */
9022 			int next = (last_mark+1) % SYNC_MARKS;
9023 
9024 			mddev->resync_mark = mark[next];
9025 			mddev->resync_mark_cnt = mark_cnt[next];
9026 			mark[next] = jiffies;
9027 			mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active);
9028 			last_mark = next;
9029 		}
9030 
9031 		if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
9032 			break;
9033 
9034 		/*
9035 		 * this loop exits only if either when we are slower than
9036 		 * the 'hard' speed limit, or the system was IO-idle for
9037 		 * a jiffy.
9038 		 * the system might be non-idle CPU-wise, but we only care
9039 		 * about not overloading the IO subsystem. (things like an
9040 		 * e2fsck being done on the RAID array should execute fast)
9041 		 */
9042 		cond_resched();
9043 
9044 		recovery_done = io_sectors - atomic_read(&mddev->recovery_active);
9045 		currspeed = ((unsigned long)(recovery_done - mddev->resync_mark_cnt))/2
9046 			/((jiffies-mddev->resync_mark)/HZ +1) +1;
9047 
9048 		if (currspeed > speed_min(mddev)) {
9049 			if (currspeed > speed_max(mddev)) {
9050 				msleep(500);
9051 				goto repeat;
9052 			}
9053 			if (!is_mddev_idle(mddev, 0)) {
9054 				/*
9055 				 * Give other IO more of a chance.
9056 				 * The faster the devices, the less we wait.
9057 				 */
9058 				wait_event(mddev->recovery_wait,
9059 					   !atomic_read(&mddev->recovery_active));
9060 			}
9061 		}
9062 	}
9063 	pr_info("md: %s: %s %s.\n",mdname(mddev), desc,
9064 		test_bit(MD_RECOVERY_INTR, &mddev->recovery)
9065 		? "interrupted" : "done");
9066 	/*
9067 	 * this also signals 'finished resyncing' to md_stop
9068 	 */
9069 	blk_finish_plug(&plug);
9070 	wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
9071 
9072 	if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
9073 	    !test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
9074 	    mddev->curr_resync >= MD_RESYNC_ACTIVE) {
9075 		mddev->curr_resync_completed = mddev->curr_resync;
9076 		sysfs_notify_dirent_safe(mddev->sysfs_completed);
9077 	}
9078 	mddev->pers->sync_request(mddev, max_sectors, &skipped);
9079 
9080 	if (!test_bit(MD_RECOVERY_CHECK, &mddev->recovery) &&
9081 	    mddev->curr_resync > MD_RESYNC_ACTIVE) {
9082 		if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
9083 			if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
9084 				if (mddev->curr_resync >= mddev->recovery_cp) {
9085 					pr_debug("md: checkpointing %s of %s.\n",
9086 						 desc, mdname(mddev));
9087 					if (test_bit(MD_RECOVERY_ERROR,
9088 						&mddev->recovery))
9089 						mddev->recovery_cp =
9090 							mddev->curr_resync_completed;
9091 					else
9092 						mddev->recovery_cp =
9093 							mddev->curr_resync;
9094 				}
9095 			} else
9096 				mddev->recovery_cp = MaxSector;
9097 		} else {
9098 			if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery))
9099 				mddev->curr_resync = MaxSector;
9100 			if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
9101 			    test_bit(MD_RECOVERY_RECOVER, &mddev->recovery)) {
9102 				rcu_read_lock();
9103 				rdev_for_each_rcu(rdev, mddev)
9104 					if (rdev->raid_disk >= 0 &&
9105 					    mddev->delta_disks >= 0 &&
9106 					    !test_bit(Journal, &rdev->flags) &&
9107 					    !test_bit(Faulty, &rdev->flags) &&
9108 					    !test_bit(In_sync, &rdev->flags) &&
9109 					    rdev->recovery_offset < mddev->curr_resync)
9110 						rdev->recovery_offset = mddev->curr_resync;
9111 				rcu_read_unlock();
9112 			}
9113 		}
9114 	}
9115  skip:
9116 	/* set CHANGE_PENDING here since maybe another update is needed,
9117 	 * so other nodes are informed. It should be harmless for normal
9118 	 * raid */
9119 	set_mask_bits(&mddev->sb_flags, 0,
9120 		      BIT(MD_SB_CHANGE_PENDING) | BIT(MD_SB_CHANGE_DEVS));
9121 
9122 	if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
9123 			!test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
9124 			mddev->delta_disks > 0 &&
9125 			mddev->pers->finish_reshape &&
9126 			mddev->pers->size &&
9127 			mddev->queue) {
9128 		mddev_lock_nointr(mddev);
9129 		md_set_array_sectors(mddev, mddev->pers->size(mddev, 0, 0));
9130 		mddev_unlock(mddev);
9131 		if (!mddev_is_clustered(mddev))
9132 			set_capacity_and_notify(mddev->gendisk,
9133 						mddev->array_sectors);
9134 	}
9135 
9136 	spin_lock(&mddev->lock);
9137 	if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
9138 		/* We completed so min/max setting can be forgotten if used. */
9139 		if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
9140 			mddev->resync_min = 0;
9141 		mddev->resync_max = MaxSector;
9142 	} else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
9143 		mddev->resync_min = mddev->curr_resync_completed;
9144 	set_bit(MD_RECOVERY_DONE, &mddev->recovery);
9145 	mddev->curr_resync = MD_RESYNC_NONE;
9146 	spin_unlock(&mddev->lock);
9147 
9148 	wake_up(&resync_wait);
9149 	wake_up(&mddev->sb_wait);
9150 	md_wakeup_thread(mddev->thread);
9151 	return;
9152 }
9153 EXPORT_SYMBOL_GPL(md_do_sync);
9154 
9155 static int remove_and_add_spares(struct mddev *mddev,
9156 				 struct md_rdev *this)
9157 {
9158 	struct md_rdev *rdev;
9159 	int spares = 0;
9160 	int removed = 0;
9161 	bool remove_some = false;
9162 
9163 	if (this && test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
9164 		/* Mustn't remove devices when resync thread is running */
9165 		return 0;
9166 
9167 	rdev_for_each(rdev, mddev) {
9168 		if ((this == NULL || rdev == this) &&
9169 		    rdev->raid_disk >= 0 &&
9170 		    !test_bit(Blocked, &rdev->flags) &&
9171 		    test_bit(Faulty, &rdev->flags) &&
9172 		    atomic_read(&rdev->nr_pending)==0) {
9173 			/* Faulty non-Blocked devices with nr_pending == 0
9174 			 * never get nr_pending incremented,
9175 			 * never get Faulty cleared, and never get Blocked set.
9176 			 * So we can synchronize_rcu now rather than once per device
9177 			 */
9178 			remove_some = true;
9179 			set_bit(RemoveSynchronized, &rdev->flags);
9180 		}
9181 	}
9182 
9183 	if (remove_some)
9184 		synchronize_rcu();
9185 	rdev_for_each(rdev, mddev) {
9186 		if ((this == NULL || rdev == this) &&
9187 		    rdev->raid_disk >= 0 &&
9188 		    !test_bit(Blocked, &rdev->flags) &&
9189 		    ((test_bit(RemoveSynchronized, &rdev->flags) ||
9190 		     (!test_bit(In_sync, &rdev->flags) &&
9191 		      !test_bit(Journal, &rdev->flags))) &&
9192 		    atomic_read(&rdev->nr_pending)==0)) {
9193 			if (mddev->pers->hot_remove_disk(
9194 				    mddev, rdev) == 0) {
9195 				sysfs_unlink_rdev(mddev, rdev);
9196 				rdev->saved_raid_disk = rdev->raid_disk;
9197 				rdev->raid_disk = -1;
9198 				removed++;
9199 			}
9200 		}
9201 		if (remove_some && test_bit(RemoveSynchronized, &rdev->flags))
9202 			clear_bit(RemoveSynchronized, &rdev->flags);
9203 	}
9204 
9205 	if (removed && mddev->kobj.sd)
9206 		sysfs_notify_dirent_safe(mddev->sysfs_degraded);
9207 
9208 	if (this && removed)
9209 		goto no_add;
9210 
9211 	rdev_for_each(rdev, mddev) {
9212 		if (this && this != rdev)
9213 			continue;
9214 		if (test_bit(Candidate, &rdev->flags))
9215 			continue;
9216 		if (rdev->raid_disk >= 0 &&
9217 		    !test_bit(In_sync, &rdev->flags) &&
9218 		    !test_bit(Journal, &rdev->flags) &&
9219 		    !test_bit(Faulty, &rdev->flags))
9220 			spares++;
9221 		if (rdev->raid_disk >= 0)
9222 			continue;
9223 		if (test_bit(Faulty, &rdev->flags))
9224 			continue;
9225 		if (!test_bit(Journal, &rdev->flags)) {
9226 			if (!md_is_rdwr(mddev) &&
9227 			    !(rdev->saved_raid_disk >= 0 &&
9228 			      !test_bit(Bitmap_sync, &rdev->flags)))
9229 				continue;
9230 
9231 			rdev->recovery_offset = 0;
9232 		}
9233 		if (mddev->pers->hot_add_disk(mddev, rdev) == 0) {
9234 			/* failure here is OK */
9235 			sysfs_link_rdev(mddev, rdev);
9236 			if (!test_bit(Journal, &rdev->flags))
9237 				spares++;
9238 			md_new_event();
9239 			set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
9240 		}
9241 	}
9242 no_add:
9243 	if (removed)
9244 		set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
9245 	return spares;
9246 }
9247 
9248 static void md_start_sync(struct work_struct *ws)
9249 {
9250 	struct mddev *mddev = container_of(ws, struct mddev, del_work);
9251 
9252 	rcu_assign_pointer(mddev->sync_thread,
9253 			   md_register_thread(md_do_sync, mddev, "resync"));
9254 	if (!mddev->sync_thread) {
9255 		pr_warn("%s: could not start resync thread...\n",
9256 			mdname(mddev));
9257 		/* leave the spares where they are, it shouldn't hurt */
9258 		clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
9259 		clear_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
9260 		clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
9261 		clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
9262 		clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
9263 		wake_up(&resync_wait);
9264 		if (test_and_clear_bit(MD_RECOVERY_RECOVER,
9265 				       &mddev->recovery))
9266 			if (mddev->sysfs_action)
9267 				sysfs_notify_dirent_safe(mddev->sysfs_action);
9268 	} else
9269 		md_wakeup_thread(mddev->sync_thread);
9270 	sysfs_notify_dirent_safe(mddev->sysfs_action);
9271 	md_new_event();
9272 }
9273 
9274 /*
9275  * This routine is regularly called by all per-raid-array threads to
9276  * deal with generic issues like resync and super-block update.
9277  * Raid personalities that don't have a thread (linear/raid0) do not
9278  * need this as they never do any recovery or update the superblock.
9279  *
9280  * It does not do any resync itself, but rather "forks" off other threads
9281  * to do that as needed.
9282  * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
9283  * "->recovery" and create a thread at ->sync_thread.
9284  * When the thread finishes it sets MD_RECOVERY_DONE
9285  * and wakeups up this thread which will reap the thread and finish up.
9286  * This thread also removes any faulty devices (with nr_pending == 0).
9287  *
9288  * The overall approach is:
9289  *  1/ if the superblock needs updating, update it.
9290  *  2/ If a recovery thread is running, don't do anything else.
9291  *  3/ If recovery has finished, clean up, possibly marking spares active.
9292  *  4/ If there are any faulty devices, remove them.
9293  *  5/ If array is degraded, try to add spares devices
9294  *  6/ If array has spares or is not in-sync, start a resync thread.
9295  */
9296 void md_check_recovery(struct mddev *mddev)
9297 {
9298 	if (test_bit(MD_ALLOW_SB_UPDATE, &mddev->flags) && mddev->sb_flags) {
9299 		/* Write superblock - thread that called mddev_suspend()
9300 		 * holds reconfig_mutex for us.
9301 		 */
9302 		set_bit(MD_UPDATING_SB, &mddev->flags);
9303 		smp_mb__after_atomic();
9304 		if (test_bit(MD_ALLOW_SB_UPDATE, &mddev->flags))
9305 			md_update_sb(mddev, 0);
9306 		clear_bit_unlock(MD_UPDATING_SB, &mddev->flags);
9307 		wake_up(&mddev->sb_wait);
9308 	}
9309 
9310 	if (is_md_suspended(mddev))
9311 		return;
9312 
9313 	if (mddev->bitmap)
9314 		md_bitmap_daemon_work(mddev);
9315 
9316 	if (signal_pending(current)) {
9317 		if (mddev->pers->sync_request && !mddev->external) {
9318 			pr_debug("md: %s in immediate safe mode\n",
9319 				 mdname(mddev));
9320 			mddev->safemode = 2;
9321 		}
9322 		flush_signals(current);
9323 	}
9324 
9325 	if (!md_is_rdwr(mddev) &&
9326 	    !test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
9327 		return;
9328 	if ( ! (
9329 		(mddev->sb_flags & ~ (1<<MD_SB_CHANGE_PENDING)) ||
9330 		test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
9331 		test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
9332 		(mddev->external == 0 && mddev->safemode == 1) ||
9333 		(mddev->safemode == 2
9334 		 && !mddev->in_sync && mddev->recovery_cp == MaxSector)
9335 		))
9336 		return;
9337 
9338 	if (mddev_trylock(mddev)) {
9339 		int spares = 0;
9340 		bool try_set_sync = mddev->safemode != 0;
9341 
9342 		if (!mddev->external && mddev->safemode == 1)
9343 			mddev->safemode = 0;
9344 
9345 		if (!md_is_rdwr(mddev)) {
9346 			struct md_rdev *rdev;
9347 			if (!mddev->external && mddev->in_sync)
9348 				/* 'Blocked' flag not needed as failed devices
9349 				 * will be recorded if array switched to read/write.
9350 				 * Leaving it set will prevent the device
9351 				 * from being removed.
9352 				 */
9353 				rdev_for_each(rdev, mddev)
9354 					clear_bit(Blocked, &rdev->flags);
9355 			/* On a read-only array we can:
9356 			 * - remove failed devices
9357 			 * - add already-in_sync devices if the array itself
9358 			 *   is in-sync.
9359 			 * As we only add devices that are already in-sync,
9360 			 * we can activate the spares immediately.
9361 			 */
9362 			remove_and_add_spares(mddev, NULL);
9363 			/* There is no thread, but we need to call
9364 			 * ->spare_active and clear saved_raid_disk
9365 			 */
9366 			set_bit(MD_RECOVERY_INTR, &mddev->recovery);
9367 			md_reap_sync_thread(mddev);
9368 			clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
9369 			clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
9370 			clear_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags);
9371 			goto unlock;
9372 		}
9373 
9374 		if (mddev_is_clustered(mddev)) {
9375 			struct md_rdev *rdev, *tmp;
9376 			/* kick the device if another node issued a
9377 			 * remove disk.
9378 			 */
9379 			rdev_for_each_safe(rdev, tmp, mddev) {
9380 				if (test_and_clear_bit(ClusterRemove, &rdev->flags) &&
9381 						rdev->raid_disk < 0)
9382 					md_kick_rdev_from_array(rdev);
9383 			}
9384 		}
9385 
9386 		if (try_set_sync && !mddev->external && !mddev->in_sync) {
9387 			spin_lock(&mddev->lock);
9388 			set_in_sync(mddev);
9389 			spin_unlock(&mddev->lock);
9390 		}
9391 
9392 		if (mddev->sb_flags)
9393 			md_update_sb(mddev, 0);
9394 
9395 		/*
9396 		 * Never start a new sync thread if MD_RECOVERY_RUNNING is
9397 		 * still set.
9398 		 */
9399 		if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) {
9400 			if (!test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
9401 				/* resync/recovery still happening */
9402 				clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
9403 				goto unlock;
9404 			}
9405 
9406 			if (WARN_ON_ONCE(!mddev->sync_thread))
9407 				goto unlock;
9408 
9409 			md_reap_sync_thread(mddev);
9410 			goto unlock;
9411 		}
9412 
9413 		/* Set RUNNING before clearing NEEDED to avoid
9414 		 * any transients in the value of "sync_action".
9415 		 */
9416 		mddev->curr_resync_completed = 0;
9417 		spin_lock(&mddev->lock);
9418 		set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
9419 		spin_unlock(&mddev->lock);
9420 		/* Clear some bits that don't mean anything, but
9421 		 * might be left set
9422 		 */
9423 		clear_bit(MD_RECOVERY_INTR, &mddev->recovery);
9424 		clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
9425 
9426 		if (!test_and_clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
9427 		    test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
9428 			goto not_running;
9429 		/* no recovery is running.
9430 		 * remove any failed drives, then
9431 		 * add spares if possible.
9432 		 * Spares are also removed and re-added, to allow
9433 		 * the personality to fail the re-add.
9434 		 */
9435 
9436 		if (mddev->reshape_position != MaxSector) {
9437 			if (mddev->pers->check_reshape == NULL ||
9438 			    mddev->pers->check_reshape(mddev) != 0)
9439 				/* Cannot proceed */
9440 				goto not_running;
9441 			set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
9442 			clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
9443 		} else if ((spares = remove_and_add_spares(mddev, NULL))) {
9444 			clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
9445 			clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
9446 			clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
9447 			set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
9448 		} else if (mddev->recovery_cp < MaxSector) {
9449 			set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
9450 			clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
9451 		} else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
9452 			/* nothing to be done ... */
9453 			goto not_running;
9454 
9455 		if (mddev->pers->sync_request) {
9456 			if (spares) {
9457 				/* We are adding a device or devices to an array
9458 				 * which has the bitmap stored on all devices.
9459 				 * So make sure all bitmap pages get written
9460 				 */
9461 				md_bitmap_write_all(mddev->bitmap);
9462 			}
9463 			INIT_WORK(&mddev->del_work, md_start_sync);
9464 			queue_work(md_misc_wq, &mddev->del_work);
9465 			goto unlock;
9466 		}
9467 	not_running:
9468 		if (!mddev->sync_thread) {
9469 			clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
9470 			wake_up(&resync_wait);
9471 			if (test_and_clear_bit(MD_RECOVERY_RECOVER,
9472 					       &mddev->recovery))
9473 				if (mddev->sysfs_action)
9474 					sysfs_notify_dirent_safe(mddev->sysfs_action);
9475 		}
9476 	unlock:
9477 		wake_up(&mddev->sb_wait);
9478 		mddev_unlock(mddev);
9479 	}
9480 }
9481 EXPORT_SYMBOL(md_check_recovery);
9482 
9483 void md_reap_sync_thread(struct mddev *mddev)
9484 {
9485 	struct md_rdev *rdev;
9486 	sector_t old_dev_sectors = mddev->dev_sectors;
9487 	bool is_reshaped = false;
9488 
9489 	/* resync has finished, collect result */
9490 	md_unregister_thread(mddev, &mddev->sync_thread);
9491 	atomic_inc(&mddev->sync_seq);
9492 
9493 	if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
9494 	    !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery) &&
9495 	    mddev->degraded != mddev->raid_disks) {
9496 		/* success...*/
9497 		/* activate any spares */
9498 		if (mddev->pers->spare_active(mddev)) {
9499 			sysfs_notify_dirent_safe(mddev->sysfs_degraded);
9500 			set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
9501 		}
9502 	}
9503 	if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
9504 	    mddev->pers->finish_reshape) {
9505 		mddev->pers->finish_reshape(mddev);
9506 		if (mddev_is_clustered(mddev))
9507 			is_reshaped = true;
9508 	}
9509 
9510 	/* If array is no-longer degraded, then any saved_raid_disk
9511 	 * information must be scrapped.
9512 	 */
9513 	if (!mddev->degraded)
9514 		rdev_for_each(rdev, mddev)
9515 			rdev->saved_raid_disk = -1;
9516 
9517 	md_update_sb(mddev, 1);
9518 	/* MD_SB_CHANGE_PENDING should be cleared by md_update_sb, so we can
9519 	 * call resync_finish here if MD_CLUSTER_RESYNC_LOCKED is set by
9520 	 * clustered raid */
9521 	if (test_and_clear_bit(MD_CLUSTER_RESYNC_LOCKED, &mddev->flags))
9522 		md_cluster_ops->resync_finish(mddev);
9523 	clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
9524 	clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
9525 	clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
9526 	clear_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
9527 	clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
9528 	clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
9529 	/*
9530 	 * We call md_cluster_ops->update_size here because sync_size could
9531 	 * be changed by md_update_sb, and MD_RECOVERY_RESHAPE is cleared,
9532 	 * so it is time to update size across cluster.
9533 	 */
9534 	if (mddev_is_clustered(mddev) && is_reshaped
9535 				      && !test_bit(MD_CLOSING, &mddev->flags))
9536 		md_cluster_ops->update_size(mddev, old_dev_sectors);
9537 	/* flag recovery needed just to double check */
9538 	set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
9539 	sysfs_notify_dirent_safe(mddev->sysfs_completed);
9540 	sysfs_notify_dirent_safe(mddev->sysfs_action);
9541 	md_new_event();
9542 	if (mddev->event_work.func)
9543 		queue_work(md_misc_wq, &mddev->event_work);
9544 	wake_up(&resync_wait);
9545 }
9546 EXPORT_SYMBOL(md_reap_sync_thread);
9547 
9548 void md_wait_for_blocked_rdev(struct md_rdev *rdev, struct mddev *mddev)
9549 {
9550 	sysfs_notify_dirent_safe(rdev->sysfs_state);
9551 	wait_event_timeout(rdev->blocked_wait,
9552 			   !test_bit(Blocked, &rdev->flags) &&
9553 			   !test_bit(BlockedBadBlocks, &rdev->flags),
9554 			   msecs_to_jiffies(5000));
9555 	rdev_dec_pending(rdev, mddev);
9556 }
9557 EXPORT_SYMBOL(md_wait_for_blocked_rdev);
9558 
9559 void md_finish_reshape(struct mddev *mddev)
9560 {
9561 	/* called be personality module when reshape completes. */
9562 	struct md_rdev *rdev;
9563 
9564 	rdev_for_each(rdev, mddev) {
9565 		if (rdev->data_offset > rdev->new_data_offset)
9566 			rdev->sectors += rdev->data_offset - rdev->new_data_offset;
9567 		else
9568 			rdev->sectors -= rdev->new_data_offset - rdev->data_offset;
9569 		rdev->data_offset = rdev->new_data_offset;
9570 	}
9571 }
9572 EXPORT_SYMBOL(md_finish_reshape);
9573 
9574 /* Bad block management */
9575 
9576 /* Returns 1 on success, 0 on failure */
9577 int rdev_set_badblocks(struct md_rdev *rdev, sector_t s, int sectors,
9578 		       int is_new)
9579 {
9580 	struct mddev *mddev = rdev->mddev;
9581 	int rv;
9582 	if (is_new)
9583 		s += rdev->new_data_offset;
9584 	else
9585 		s += rdev->data_offset;
9586 	rv = badblocks_set(&rdev->badblocks, s, sectors, 0);
9587 	if (rv == 0) {
9588 		/* Make sure they get written out promptly */
9589 		if (test_bit(ExternalBbl, &rdev->flags))
9590 			sysfs_notify_dirent_safe(rdev->sysfs_unack_badblocks);
9591 		sysfs_notify_dirent_safe(rdev->sysfs_state);
9592 		set_mask_bits(&mddev->sb_flags, 0,
9593 			      BIT(MD_SB_CHANGE_CLEAN) | BIT(MD_SB_CHANGE_PENDING));
9594 		md_wakeup_thread(rdev->mddev->thread);
9595 		return 1;
9596 	} else
9597 		return 0;
9598 }
9599 EXPORT_SYMBOL_GPL(rdev_set_badblocks);
9600 
9601 int rdev_clear_badblocks(struct md_rdev *rdev, sector_t s, int sectors,
9602 			 int is_new)
9603 {
9604 	int rv;
9605 	if (is_new)
9606 		s += rdev->new_data_offset;
9607 	else
9608 		s += rdev->data_offset;
9609 	rv = badblocks_clear(&rdev->badblocks, s, sectors);
9610 	if ((rv == 0) && test_bit(ExternalBbl, &rdev->flags))
9611 		sysfs_notify_dirent_safe(rdev->sysfs_badblocks);
9612 	return rv;
9613 }
9614 EXPORT_SYMBOL_GPL(rdev_clear_badblocks);
9615 
9616 static int md_notify_reboot(struct notifier_block *this,
9617 			    unsigned long code, void *x)
9618 {
9619 	struct mddev *mddev, *n;
9620 	int need_delay = 0;
9621 
9622 	spin_lock(&all_mddevs_lock);
9623 	list_for_each_entry_safe(mddev, n, &all_mddevs, all_mddevs) {
9624 		if (!mddev_get(mddev))
9625 			continue;
9626 		spin_unlock(&all_mddevs_lock);
9627 		if (mddev_trylock(mddev)) {
9628 			if (mddev->pers)
9629 				__md_stop_writes(mddev);
9630 			if (mddev->persistent)
9631 				mddev->safemode = 2;
9632 			mddev_unlock(mddev);
9633 		}
9634 		need_delay = 1;
9635 		mddev_put(mddev);
9636 		spin_lock(&all_mddevs_lock);
9637 	}
9638 	spin_unlock(&all_mddevs_lock);
9639 
9640 	/*
9641 	 * certain more exotic SCSI devices are known to be
9642 	 * volatile wrt too early system reboots. While the
9643 	 * right place to handle this issue is the given
9644 	 * driver, we do want to have a safe RAID driver ...
9645 	 */
9646 	if (need_delay)
9647 		msleep(1000);
9648 
9649 	return NOTIFY_DONE;
9650 }
9651 
9652 static struct notifier_block md_notifier = {
9653 	.notifier_call	= md_notify_reboot,
9654 	.next		= NULL,
9655 	.priority	= INT_MAX, /* before any real devices */
9656 };
9657 
9658 static void md_geninit(void)
9659 {
9660 	pr_debug("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
9661 
9662 	proc_create("mdstat", S_IRUGO, NULL, &mdstat_proc_ops);
9663 }
9664 
9665 static int __init md_init(void)
9666 {
9667 	int ret = -ENOMEM;
9668 
9669 	md_wq = alloc_workqueue("md", WQ_MEM_RECLAIM, 0);
9670 	if (!md_wq)
9671 		goto err_wq;
9672 
9673 	md_misc_wq = alloc_workqueue("md_misc", 0, 0);
9674 	if (!md_misc_wq)
9675 		goto err_misc_wq;
9676 
9677 	md_bitmap_wq = alloc_workqueue("md_bitmap", WQ_MEM_RECLAIM | WQ_UNBOUND,
9678 				       0);
9679 	if (!md_bitmap_wq)
9680 		goto err_bitmap_wq;
9681 
9682 	ret = __register_blkdev(MD_MAJOR, "md", md_probe);
9683 	if (ret < 0)
9684 		goto err_md;
9685 
9686 	ret = __register_blkdev(0, "mdp", md_probe);
9687 	if (ret < 0)
9688 		goto err_mdp;
9689 	mdp_major = ret;
9690 
9691 	register_reboot_notifier(&md_notifier);
9692 	raid_table_header = register_sysctl("dev/raid", raid_table);
9693 
9694 	md_geninit();
9695 	return 0;
9696 
9697 err_mdp:
9698 	unregister_blkdev(MD_MAJOR, "md");
9699 err_md:
9700 	destroy_workqueue(md_bitmap_wq);
9701 err_bitmap_wq:
9702 	destroy_workqueue(md_misc_wq);
9703 err_misc_wq:
9704 	destroy_workqueue(md_wq);
9705 err_wq:
9706 	return ret;
9707 }
9708 
9709 static void check_sb_changes(struct mddev *mddev, struct md_rdev *rdev)
9710 {
9711 	struct mdp_superblock_1 *sb = page_address(rdev->sb_page);
9712 	struct md_rdev *rdev2, *tmp;
9713 	int role, ret;
9714 
9715 	/*
9716 	 * If size is changed in another node then we need to
9717 	 * do resize as well.
9718 	 */
9719 	if (mddev->dev_sectors != le64_to_cpu(sb->size)) {
9720 		ret = mddev->pers->resize(mddev, le64_to_cpu(sb->size));
9721 		if (ret)
9722 			pr_info("md-cluster: resize failed\n");
9723 		else
9724 			md_bitmap_update_sb(mddev->bitmap);
9725 	}
9726 
9727 	/* Check for change of roles in the active devices */
9728 	rdev_for_each_safe(rdev2, tmp, mddev) {
9729 		if (test_bit(Faulty, &rdev2->flags))
9730 			continue;
9731 
9732 		/* Check if the roles changed */
9733 		role = le16_to_cpu(sb->dev_roles[rdev2->desc_nr]);
9734 
9735 		if (test_bit(Candidate, &rdev2->flags)) {
9736 			if (role == MD_DISK_ROLE_FAULTY) {
9737 				pr_info("md: Removing Candidate device %pg because add failed\n",
9738 					rdev2->bdev);
9739 				md_kick_rdev_from_array(rdev2);
9740 				continue;
9741 			}
9742 			else
9743 				clear_bit(Candidate, &rdev2->flags);
9744 		}
9745 
9746 		if (role != rdev2->raid_disk) {
9747 			/*
9748 			 * got activated except reshape is happening.
9749 			 */
9750 			if (rdev2->raid_disk == -1 && role != MD_DISK_ROLE_SPARE &&
9751 			    !(le32_to_cpu(sb->feature_map) &
9752 			      MD_FEATURE_RESHAPE_ACTIVE)) {
9753 				rdev2->saved_raid_disk = role;
9754 				ret = remove_and_add_spares(mddev, rdev2);
9755 				pr_info("Activated spare: %pg\n",
9756 					rdev2->bdev);
9757 				/* wakeup mddev->thread here, so array could
9758 				 * perform resync with the new activated disk */
9759 				set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
9760 				md_wakeup_thread(mddev->thread);
9761 			}
9762 			/* device faulty
9763 			 * We just want to do the minimum to mark the disk
9764 			 * as faulty. The recovery is performed by the
9765 			 * one who initiated the error.
9766 			 */
9767 			if (role == MD_DISK_ROLE_FAULTY ||
9768 			    role == MD_DISK_ROLE_JOURNAL) {
9769 				md_error(mddev, rdev2);
9770 				clear_bit(Blocked, &rdev2->flags);
9771 			}
9772 		}
9773 	}
9774 
9775 	if (mddev->raid_disks != le32_to_cpu(sb->raid_disks)) {
9776 		ret = update_raid_disks(mddev, le32_to_cpu(sb->raid_disks));
9777 		if (ret)
9778 			pr_warn("md: updating array disks failed. %d\n", ret);
9779 	}
9780 
9781 	/*
9782 	 * Since mddev->delta_disks has already updated in update_raid_disks,
9783 	 * so it is time to check reshape.
9784 	 */
9785 	if (test_bit(MD_RESYNCING_REMOTE, &mddev->recovery) &&
9786 	    (le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) {
9787 		/*
9788 		 * reshape is happening in the remote node, we need to
9789 		 * update reshape_position and call start_reshape.
9790 		 */
9791 		mddev->reshape_position = le64_to_cpu(sb->reshape_position);
9792 		if (mddev->pers->update_reshape_pos)
9793 			mddev->pers->update_reshape_pos(mddev);
9794 		if (mddev->pers->start_reshape)
9795 			mddev->pers->start_reshape(mddev);
9796 	} else if (test_bit(MD_RESYNCING_REMOTE, &mddev->recovery) &&
9797 		   mddev->reshape_position != MaxSector &&
9798 		   !(le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) {
9799 		/* reshape is just done in another node. */
9800 		mddev->reshape_position = MaxSector;
9801 		if (mddev->pers->update_reshape_pos)
9802 			mddev->pers->update_reshape_pos(mddev);
9803 	}
9804 
9805 	/* Finally set the event to be up to date */
9806 	mddev->events = le64_to_cpu(sb->events);
9807 }
9808 
9809 static int read_rdev(struct mddev *mddev, struct md_rdev *rdev)
9810 {
9811 	int err;
9812 	struct page *swapout = rdev->sb_page;
9813 	struct mdp_superblock_1 *sb;
9814 
9815 	/* Store the sb page of the rdev in the swapout temporary
9816 	 * variable in case we err in the future
9817 	 */
9818 	rdev->sb_page = NULL;
9819 	err = alloc_disk_sb(rdev);
9820 	if (err == 0) {
9821 		ClearPageUptodate(rdev->sb_page);
9822 		rdev->sb_loaded = 0;
9823 		err = super_types[mddev->major_version].
9824 			load_super(rdev, NULL, mddev->minor_version);
9825 	}
9826 	if (err < 0) {
9827 		pr_warn("%s: %d Could not reload rdev(%d) err: %d. Restoring old values\n",
9828 				__func__, __LINE__, rdev->desc_nr, err);
9829 		if (rdev->sb_page)
9830 			put_page(rdev->sb_page);
9831 		rdev->sb_page = swapout;
9832 		rdev->sb_loaded = 1;
9833 		return err;
9834 	}
9835 
9836 	sb = page_address(rdev->sb_page);
9837 	/* Read the offset unconditionally, even if MD_FEATURE_RECOVERY_OFFSET
9838 	 * is not set
9839 	 */
9840 
9841 	if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RECOVERY_OFFSET))
9842 		rdev->recovery_offset = le64_to_cpu(sb->recovery_offset);
9843 
9844 	/* The other node finished recovery, call spare_active to set
9845 	 * device In_sync and mddev->degraded
9846 	 */
9847 	if (rdev->recovery_offset == MaxSector &&
9848 	    !test_bit(In_sync, &rdev->flags) &&
9849 	    mddev->pers->spare_active(mddev))
9850 		sysfs_notify_dirent_safe(mddev->sysfs_degraded);
9851 
9852 	put_page(swapout);
9853 	return 0;
9854 }
9855 
9856 void md_reload_sb(struct mddev *mddev, int nr)
9857 {
9858 	struct md_rdev *rdev = NULL, *iter;
9859 	int err;
9860 
9861 	/* Find the rdev */
9862 	rdev_for_each_rcu(iter, mddev) {
9863 		if (iter->desc_nr == nr) {
9864 			rdev = iter;
9865 			break;
9866 		}
9867 	}
9868 
9869 	if (!rdev) {
9870 		pr_warn("%s: %d Could not find rdev with nr %d\n", __func__, __LINE__, nr);
9871 		return;
9872 	}
9873 
9874 	err = read_rdev(mddev, rdev);
9875 	if (err < 0)
9876 		return;
9877 
9878 	check_sb_changes(mddev, rdev);
9879 
9880 	/* Read all rdev's to update recovery_offset */
9881 	rdev_for_each_rcu(rdev, mddev) {
9882 		if (!test_bit(Faulty, &rdev->flags))
9883 			read_rdev(mddev, rdev);
9884 	}
9885 }
9886 EXPORT_SYMBOL(md_reload_sb);
9887 
9888 #ifndef MODULE
9889 
9890 /*
9891  * Searches all registered partitions for autorun RAID arrays
9892  * at boot time.
9893  */
9894 
9895 static DEFINE_MUTEX(detected_devices_mutex);
9896 static LIST_HEAD(all_detected_devices);
9897 struct detected_devices_node {
9898 	struct list_head list;
9899 	dev_t dev;
9900 };
9901 
9902 void md_autodetect_dev(dev_t dev)
9903 {
9904 	struct detected_devices_node *node_detected_dev;
9905 
9906 	node_detected_dev = kzalloc(sizeof(*node_detected_dev), GFP_KERNEL);
9907 	if (node_detected_dev) {
9908 		node_detected_dev->dev = dev;
9909 		mutex_lock(&detected_devices_mutex);
9910 		list_add_tail(&node_detected_dev->list, &all_detected_devices);
9911 		mutex_unlock(&detected_devices_mutex);
9912 	}
9913 }
9914 
9915 void md_autostart_arrays(int part)
9916 {
9917 	struct md_rdev *rdev;
9918 	struct detected_devices_node *node_detected_dev;
9919 	dev_t dev;
9920 	int i_scanned, i_passed;
9921 
9922 	i_scanned = 0;
9923 	i_passed = 0;
9924 
9925 	pr_info("md: Autodetecting RAID arrays.\n");
9926 
9927 	mutex_lock(&detected_devices_mutex);
9928 	while (!list_empty(&all_detected_devices) && i_scanned < INT_MAX) {
9929 		i_scanned++;
9930 		node_detected_dev = list_entry(all_detected_devices.next,
9931 					struct detected_devices_node, list);
9932 		list_del(&node_detected_dev->list);
9933 		dev = node_detected_dev->dev;
9934 		kfree(node_detected_dev);
9935 		mutex_unlock(&detected_devices_mutex);
9936 		rdev = md_import_device(dev,0, 90);
9937 		mutex_lock(&detected_devices_mutex);
9938 		if (IS_ERR(rdev))
9939 			continue;
9940 
9941 		if (test_bit(Faulty, &rdev->flags))
9942 			continue;
9943 
9944 		set_bit(AutoDetected, &rdev->flags);
9945 		list_add(&rdev->same_set, &pending_raid_disks);
9946 		i_passed++;
9947 	}
9948 	mutex_unlock(&detected_devices_mutex);
9949 
9950 	pr_debug("md: Scanned %d and added %d devices.\n", i_scanned, i_passed);
9951 
9952 	autorun_devices(part);
9953 }
9954 
9955 #endif /* !MODULE */
9956 
9957 static __exit void md_exit(void)
9958 {
9959 	struct mddev *mddev, *n;
9960 	int delay = 1;
9961 
9962 	unregister_blkdev(MD_MAJOR,"md");
9963 	unregister_blkdev(mdp_major, "mdp");
9964 	unregister_reboot_notifier(&md_notifier);
9965 	unregister_sysctl_table(raid_table_header);
9966 
9967 	/* We cannot unload the modules while some process is
9968 	 * waiting for us in select() or poll() - wake them up
9969 	 */
9970 	md_unloading = 1;
9971 	while (waitqueue_active(&md_event_waiters)) {
9972 		/* not safe to leave yet */
9973 		wake_up(&md_event_waiters);
9974 		msleep(delay);
9975 		delay += delay;
9976 	}
9977 	remove_proc_entry("mdstat", NULL);
9978 
9979 	spin_lock(&all_mddevs_lock);
9980 	list_for_each_entry_safe(mddev, n, &all_mddevs, all_mddevs) {
9981 		if (!mddev_get(mddev))
9982 			continue;
9983 		spin_unlock(&all_mddevs_lock);
9984 		export_array(mddev);
9985 		mddev->ctime = 0;
9986 		mddev->hold_active = 0;
9987 		/*
9988 		 * As the mddev is now fully clear, mddev_put will schedule
9989 		 * the mddev for destruction by a workqueue, and the
9990 		 * destroy_workqueue() below will wait for that to complete.
9991 		 */
9992 		mddev_put(mddev);
9993 		spin_lock(&all_mddevs_lock);
9994 	}
9995 	spin_unlock(&all_mddevs_lock);
9996 
9997 	destroy_workqueue(md_misc_wq);
9998 	destroy_workqueue(md_bitmap_wq);
9999 	destroy_workqueue(md_wq);
10000 }
10001 
10002 subsys_initcall(md_init);
10003 module_exit(md_exit)
10004 
10005 static int get_ro(char *buffer, const struct kernel_param *kp)
10006 {
10007 	return sprintf(buffer, "%d\n", start_readonly);
10008 }
10009 static int set_ro(const char *val, const struct kernel_param *kp)
10010 {
10011 	return kstrtouint(val, 10, (unsigned int *)&start_readonly);
10012 }
10013 
10014 module_param_call(start_ro, set_ro, get_ro, NULL, S_IRUSR|S_IWUSR);
10015 module_param(start_dirty_degraded, int, S_IRUGO|S_IWUSR);
10016 module_param_call(new_array, add_named_array, NULL, NULL, S_IWUSR);
10017 module_param(create_on_open, bool, S_IRUSR|S_IWUSR);
10018 
10019 MODULE_LICENSE("GPL");
10020 MODULE_DESCRIPTION("MD RAID framework");
10021 MODULE_ALIAS("md");
10022 MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR);
10023