xref: /openbmc/linux/drivers/md/md.c (revision 9fd0198f)
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 && 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 pair is percpu_ref_get() from md_flush_request() */
498 		percpu_ref_put(&mddev->active_io);
499 
500 		/* The pre-request flush has finished */
501 		queue_work(md_wq, &mddev->flush_work);
502 	}
503 }
504 
505 static void md_submit_flush_data(struct work_struct *ws);
506 
507 static void submit_flushes(struct work_struct *ws)
508 {
509 	struct mddev *mddev = container_of(ws, struct mddev, flush_work);
510 	struct md_rdev *rdev;
511 
512 	mddev->start_flush = ktime_get_boottime();
513 	INIT_WORK(&mddev->flush_work, md_submit_flush_data);
514 	atomic_set(&mddev->flush_pending, 1);
515 	rcu_read_lock();
516 	rdev_for_each_rcu(rdev, mddev)
517 		if (rdev->raid_disk >= 0 &&
518 		    !test_bit(Faulty, &rdev->flags)) {
519 			struct bio *bi;
520 
521 			atomic_inc(&rdev->nr_pending);
522 			rcu_read_unlock();
523 			bi = bio_alloc_bioset(rdev->bdev, 0,
524 					      REQ_OP_WRITE | REQ_PREFLUSH,
525 					      GFP_NOIO, &mddev->bio_set);
526 			bi->bi_end_io = md_end_flush;
527 			bi->bi_private = rdev;
528 			atomic_inc(&mddev->flush_pending);
529 			submit_bio(bi);
530 			rcu_read_lock();
531 		}
532 	rcu_read_unlock();
533 	if (atomic_dec_and_test(&mddev->flush_pending)) {
534 		/* The pair is percpu_ref_get() from md_flush_request() */
535 		percpu_ref_put(&mddev->active_io);
536 
537 		queue_work(md_wq, &mddev->flush_work);
538 	}
539 }
540 
541 static void md_submit_flush_data(struct work_struct *ws)
542 {
543 	struct mddev *mddev = container_of(ws, struct mddev, flush_work);
544 	struct bio *bio = mddev->flush_bio;
545 
546 	/*
547 	 * must reset flush_bio before calling into md_handle_request to avoid a
548 	 * deadlock, because other bios passed md_handle_request suspend check
549 	 * could wait for this and below md_handle_request could wait for those
550 	 * bios because of suspend check
551 	 */
552 	spin_lock_irq(&mddev->lock);
553 	mddev->prev_flush_start = mddev->start_flush;
554 	mddev->flush_bio = NULL;
555 	spin_unlock_irq(&mddev->lock);
556 	wake_up(&mddev->sb_wait);
557 
558 	if (bio->bi_iter.bi_size == 0) {
559 		/* an empty barrier - all done */
560 		bio_endio(bio);
561 	} else {
562 		bio->bi_opf &= ~REQ_PREFLUSH;
563 		md_handle_request(mddev, bio);
564 	}
565 }
566 
567 /*
568  * Manages consolidation of flushes and submitting any flushes needed for
569  * a bio with REQ_PREFLUSH.  Returns true if the bio is finished or is
570  * being finished in another context.  Returns false if the flushing is
571  * complete but still needs the I/O portion of the bio to be processed.
572  */
573 bool md_flush_request(struct mddev *mddev, struct bio *bio)
574 {
575 	ktime_t req_start = ktime_get_boottime();
576 	spin_lock_irq(&mddev->lock);
577 	/* flush requests wait until ongoing flush completes,
578 	 * hence coalescing all the pending requests.
579 	 */
580 	wait_event_lock_irq(mddev->sb_wait,
581 			    !mddev->flush_bio ||
582 			    ktime_before(req_start, mddev->prev_flush_start),
583 			    mddev->lock);
584 	/* new request after previous flush is completed */
585 	if (ktime_after(req_start, mddev->prev_flush_start)) {
586 		WARN_ON(mddev->flush_bio);
587 		/*
588 		 * Grab a reference to make sure mddev_suspend() will wait for
589 		 * this flush to be done.
590 		 *
591 		 * md_flush_reqeust() is called under md_handle_request() and
592 		 * 'active_io' is already grabbed, hence percpu_ref_is_zero()
593 		 * won't pass, percpu_ref_tryget_live() can't be used because
594 		 * percpu_ref_kill() can be called by mddev_suspend()
595 		 * concurrently.
596 		 */
597 		WARN_ON(percpu_ref_is_zero(&mddev->active_io));
598 		percpu_ref_get(&mddev->active_io);
599 		mddev->flush_bio = bio;
600 		bio = NULL;
601 	}
602 	spin_unlock_irq(&mddev->lock);
603 
604 	if (!bio) {
605 		INIT_WORK(&mddev->flush_work, submit_flushes);
606 		queue_work(md_wq, &mddev->flush_work);
607 	} else {
608 		/* flush was performed for some other bio while we waited. */
609 		if (bio->bi_iter.bi_size == 0)
610 			/* an empty barrier - all done */
611 			bio_endio(bio);
612 		else {
613 			bio->bi_opf &= ~REQ_PREFLUSH;
614 			return false;
615 		}
616 	}
617 	return true;
618 }
619 EXPORT_SYMBOL(md_flush_request);
620 
621 static inline struct mddev *mddev_get(struct mddev *mddev)
622 {
623 	lockdep_assert_held(&all_mddevs_lock);
624 
625 	if (test_bit(MD_DELETED, &mddev->flags))
626 		return NULL;
627 	atomic_inc(&mddev->active);
628 	return mddev;
629 }
630 
631 static void mddev_delayed_delete(struct work_struct *ws);
632 
633 void mddev_put(struct mddev *mddev)
634 {
635 	if (!atomic_dec_and_lock(&mddev->active, &all_mddevs_lock))
636 		return;
637 	if (!mddev->raid_disks && list_empty(&mddev->disks) &&
638 	    mddev->ctime == 0 && !mddev->hold_active) {
639 		/* Array is not configured at all, and not held active,
640 		 * so destroy it */
641 		set_bit(MD_DELETED, &mddev->flags);
642 
643 		/*
644 		 * Call queue_work inside the spinlock so that
645 		 * flush_workqueue() after mddev_find will succeed in waiting
646 		 * for the work to be done.
647 		 */
648 		INIT_WORK(&mddev->del_work, mddev_delayed_delete);
649 		queue_work(md_misc_wq, &mddev->del_work);
650 	}
651 	spin_unlock(&all_mddevs_lock);
652 }
653 
654 static void md_safemode_timeout(struct timer_list *t);
655 
656 void mddev_init(struct mddev *mddev)
657 {
658 	mutex_init(&mddev->open_mutex);
659 	mutex_init(&mddev->reconfig_mutex);
660 	mutex_init(&mddev->sync_mutex);
661 	mutex_init(&mddev->bitmap_info.mutex);
662 	INIT_LIST_HEAD(&mddev->disks);
663 	INIT_LIST_HEAD(&mddev->all_mddevs);
664 	INIT_LIST_HEAD(&mddev->deleting);
665 	timer_setup(&mddev->safemode_timer, md_safemode_timeout, 0);
666 	atomic_set(&mddev->active, 1);
667 	atomic_set(&mddev->openers, 0);
668 	atomic_set(&mddev->sync_seq, 0);
669 	spin_lock_init(&mddev->lock);
670 	atomic_set(&mddev->flush_pending, 0);
671 	init_waitqueue_head(&mddev->sb_wait);
672 	init_waitqueue_head(&mddev->recovery_wait);
673 	mddev->reshape_position = MaxSector;
674 	mddev->reshape_backwards = 0;
675 	mddev->last_sync_action = "none";
676 	mddev->resync_min = 0;
677 	mddev->resync_max = MaxSector;
678 	mddev->level = LEVEL_NONE;
679 }
680 EXPORT_SYMBOL_GPL(mddev_init);
681 
682 static struct mddev *mddev_find_locked(dev_t unit)
683 {
684 	struct mddev *mddev;
685 
686 	list_for_each_entry(mddev, &all_mddevs, all_mddevs)
687 		if (mddev->unit == unit)
688 			return mddev;
689 
690 	return NULL;
691 }
692 
693 /* find an unused unit number */
694 static dev_t mddev_alloc_unit(void)
695 {
696 	static int next_minor = 512;
697 	int start = next_minor;
698 	bool is_free = 0;
699 	dev_t dev = 0;
700 
701 	while (!is_free) {
702 		dev = MKDEV(MD_MAJOR, next_minor);
703 		next_minor++;
704 		if (next_minor > MINORMASK)
705 			next_minor = 0;
706 		if (next_minor == start)
707 			return 0;		/* Oh dear, all in use. */
708 		is_free = !mddev_find_locked(dev);
709 	}
710 
711 	return dev;
712 }
713 
714 static struct mddev *mddev_alloc(dev_t unit)
715 {
716 	struct mddev *new;
717 	int error;
718 
719 	if (unit && MAJOR(unit) != MD_MAJOR)
720 		unit &= ~((1 << MdpMinorShift) - 1);
721 
722 	new = kzalloc(sizeof(*new), GFP_KERNEL);
723 	if (!new)
724 		return ERR_PTR(-ENOMEM);
725 	mddev_init(new);
726 
727 	spin_lock(&all_mddevs_lock);
728 	if (unit) {
729 		error = -EEXIST;
730 		if (mddev_find_locked(unit))
731 			goto out_free_new;
732 		new->unit = unit;
733 		if (MAJOR(unit) == MD_MAJOR)
734 			new->md_minor = MINOR(unit);
735 		else
736 			new->md_minor = MINOR(unit) >> MdpMinorShift;
737 		new->hold_active = UNTIL_IOCTL;
738 	} else {
739 		error = -ENODEV;
740 		new->unit = mddev_alloc_unit();
741 		if (!new->unit)
742 			goto out_free_new;
743 		new->md_minor = MINOR(new->unit);
744 		new->hold_active = UNTIL_STOP;
745 	}
746 
747 	list_add(&new->all_mddevs, &all_mddevs);
748 	spin_unlock(&all_mddevs_lock);
749 	return new;
750 out_free_new:
751 	spin_unlock(&all_mddevs_lock);
752 	kfree(new);
753 	return ERR_PTR(error);
754 }
755 
756 static void mddev_free(struct mddev *mddev)
757 {
758 	spin_lock(&all_mddevs_lock);
759 	list_del(&mddev->all_mddevs);
760 	spin_unlock(&all_mddevs_lock);
761 
762 	kfree(mddev);
763 }
764 
765 static const struct attribute_group md_redundancy_group;
766 
767 void mddev_unlock(struct mddev *mddev)
768 {
769 	struct md_rdev *rdev;
770 	struct md_rdev *tmp;
771 	LIST_HEAD(delete);
772 
773 	if (!list_empty(&mddev->deleting))
774 		list_splice_init(&mddev->deleting, &delete);
775 
776 	if (mddev->to_remove) {
777 		/* These cannot be removed under reconfig_mutex as
778 		 * an access to the files will try to take reconfig_mutex
779 		 * while holding the file unremovable, which leads to
780 		 * a deadlock.
781 		 * So hold set sysfs_active while the remove in happeing,
782 		 * and anything else which might set ->to_remove or my
783 		 * otherwise change the sysfs namespace will fail with
784 		 * -EBUSY if sysfs_active is still set.
785 		 * We set sysfs_active under reconfig_mutex and elsewhere
786 		 * test it under the same mutex to ensure its correct value
787 		 * is seen.
788 		 */
789 		const struct attribute_group *to_remove = mddev->to_remove;
790 		mddev->to_remove = NULL;
791 		mddev->sysfs_active = 1;
792 		mutex_unlock(&mddev->reconfig_mutex);
793 
794 		if (mddev->kobj.sd) {
795 			if (to_remove != &md_redundancy_group)
796 				sysfs_remove_group(&mddev->kobj, to_remove);
797 			if (mddev->pers == NULL ||
798 			    mddev->pers->sync_request == NULL) {
799 				sysfs_remove_group(&mddev->kobj, &md_redundancy_group);
800 				if (mddev->sysfs_action)
801 					sysfs_put(mddev->sysfs_action);
802 				if (mddev->sysfs_completed)
803 					sysfs_put(mddev->sysfs_completed);
804 				if (mddev->sysfs_degraded)
805 					sysfs_put(mddev->sysfs_degraded);
806 				mddev->sysfs_action = NULL;
807 				mddev->sysfs_completed = NULL;
808 				mddev->sysfs_degraded = NULL;
809 			}
810 		}
811 		mddev->sysfs_active = 0;
812 	} else
813 		mutex_unlock(&mddev->reconfig_mutex);
814 
815 	md_wakeup_thread(mddev->thread);
816 	wake_up(&mddev->sb_wait);
817 
818 	list_for_each_entry_safe(rdev, tmp, &delete, same_set) {
819 		list_del_init(&rdev->same_set);
820 		kobject_del(&rdev->kobj);
821 		export_rdev(rdev, mddev);
822 	}
823 }
824 EXPORT_SYMBOL_GPL(mddev_unlock);
825 
826 struct md_rdev *md_find_rdev_nr_rcu(struct mddev *mddev, int nr)
827 {
828 	struct md_rdev *rdev;
829 
830 	rdev_for_each_rcu(rdev, mddev)
831 		if (rdev->desc_nr == nr)
832 			return rdev;
833 
834 	return NULL;
835 }
836 EXPORT_SYMBOL_GPL(md_find_rdev_nr_rcu);
837 
838 static struct md_rdev *find_rdev(struct mddev *mddev, dev_t dev)
839 {
840 	struct md_rdev *rdev;
841 
842 	rdev_for_each(rdev, mddev)
843 		if (rdev->bdev->bd_dev == dev)
844 			return rdev;
845 
846 	return NULL;
847 }
848 
849 struct md_rdev *md_find_rdev_rcu(struct mddev *mddev, dev_t dev)
850 {
851 	struct md_rdev *rdev;
852 
853 	rdev_for_each_rcu(rdev, mddev)
854 		if (rdev->bdev->bd_dev == dev)
855 			return rdev;
856 
857 	return NULL;
858 }
859 EXPORT_SYMBOL_GPL(md_find_rdev_rcu);
860 
861 static struct md_personality *find_pers(int level, char *clevel)
862 {
863 	struct md_personality *pers;
864 	list_for_each_entry(pers, &pers_list, list) {
865 		if (level != LEVEL_NONE && pers->level == level)
866 			return pers;
867 		if (strcmp(pers->name, clevel)==0)
868 			return pers;
869 	}
870 	return NULL;
871 }
872 
873 /* return the offset of the super block in 512byte sectors */
874 static inline sector_t calc_dev_sboffset(struct md_rdev *rdev)
875 {
876 	return MD_NEW_SIZE_SECTORS(bdev_nr_sectors(rdev->bdev));
877 }
878 
879 static int alloc_disk_sb(struct md_rdev *rdev)
880 {
881 	rdev->sb_page = alloc_page(GFP_KERNEL);
882 	if (!rdev->sb_page)
883 		return -ENOMEM;
884 	return 0;
885 }
886 
887 void md_rdev_clear(struct md_rdev *rdev)
888 {
889 	if (rdev->sb_page) {
890 		put_page(rdev->sb_page);
891 		rdev->sb_loaded = 0;
892 		rdev->sb_page = NULL;
893 		rdev->sb_start = 0;
894 		rdev->sectors = 0;
895 	}
896 	if (rdev->bb_page) {
897 		put_page(rdev->bb_page);
898 		rdev->bb_page = NULL;
899 	}
900 	badblocks_exit(&rdev->badblocks);
901 }
902 EXPORT_SYMBOL_GPL(md_rdev_clear);
903 
904 static void super_written(struct bio *bio)
905 {
906 	struct md_rdev *rdev = bio->bi_private;
907 	struct mddev *mddev = rdev->mddev;
908 
909 	if (bio->bi_status) {
910 		pr_err("md: %s gets error=%d\n", __func__,
911 		       blk_status_to_errno(bio->bi_status));
912 		md_error(mddev, rdev);
913 		if (!test_bit(Faulty, &rdev->flags)
914 		    && (bio->bi_opf & MD_FAILFAST)) {
915 			set_bit(MD_SB_NEED_REWRITE, &mddev->sb_flags);
916 			set_bit(LastDev, &rdev->flags);
917 		}
918 	} else
919 		clear_bit(LastDev, &rdev->flags);
920 
921 	bio_put(bio);
922 
923 	rdev_dec_pending(rdev, mddev);
924 
925 	if (atomic_dec_and_test(&mddev->pending_writes))
926 		wake_up(&mddev->sb_wait);
927 }
928 
929 void md_super_write(struct mddev *mddev, struct md_rdev *rdev,
930 		   sector_t sector, int size, struct page *page)
931 {
932 	/* write first size bytes of page to sector of rdev
933 	 * Increment mddev->pending_writes before returning
934 	 * and decrement it on completion, waking up sb_wait
935 	 * if zero is reached.
936 	 * If an error occurred, call md_error
937 	 */
938 	struct bio *bio;
939 
940 	if (!page)
941 		return;
942 
943 	if (test_bit(Faulty, &rdev->flags))
944 		return;
945 
946 	bio = bio_alloc_bioset(rdev->meta_bdev ? rdev->meta_bdev : rdev->bdev,
947 			      1,
948 			      REQ_OP_WRITE | REQ_SYNC | REQ_IDLE | REQ_META
949 				  | REQ_PREFLUSH | REQ_FUA,
950 			      GFP_NOIO, &mddev->sync_set);
951 
952 	atomic_inc(&rdev->nr_pending);
953 
954 	bio->bi_iter.bi_sector = sector;
955 	__bio_add_page(bio, page, size, 0);
956 	bio->bi_private = rdev;
957 	bio->bi_end_io = super_written;
958 
959 	if (test_bit(MD_FAILFAST_SUPPORTED, &mddev->flags) &&
960 	    test_bit(FailFast, &rdev->flags) &&
961 	    !test_bit(LastDev, &rdev->flags))
962 		bio->bi_opf |= MD_FAILFAST;
963 
964 	atomic_inc(&mddev->pending_writes);
965 	submit_bio(bio);
966 }
967 
968 int md_super_wait(struct mddev *mddev)
969 {
970 	/* wait for all superblock writes that were scheduled to complete */
971 	wait_event(mddev->sb_wait, atomic_read(&mddev->pending_writes)==0);
972 	if (test_and_clear_bit(MD_SB_NEED_REWRITE, &mddev->sb_flags))
973 		return -EAGAIN;
974 	return 0;
975 }
976 
977 int sync_page_io(struct md_rdev *rdev, sector_t sector, int size,
978 		 struct page *page, blk_opf_t opf, bool metadata_op)
979 {
980 	struct bio bio;
981 	struct bio_vec bvec;
982 
983 	if (metadata_op && rdev->meta_bdev)
984 		bio_init(&bio, rdev->meta_bdev, &bvec, 1, opf);
985 	else
986 		bio_init(&bio, rdev->bdev, &bvec, 1, opf);
987 
988 	if (metadata_op)
989 		bio.bi_iter.bi_sector = sector + rdev->sb_start;
990 	else if (rdev->mddev->reshape_position != MaxSector &&
991 		 (rdev->mddev->reshape_backwards ==
992 		  (sector >= rdev->mddev->reshape_position)))
993 		bio.bi_iter.bi_sector = sector + rdev->new_data_offset;
994 	else
995 		bio.bi_iter.bi_sector = sector + rdev->data_offset;
996 	__bio_add_page(&bio, page, size, 0);
997 
998 	submit_bio_wait(&bio);
999 
1000 	return !bio.bi_status;
1001 }
1002 EXPORT_SYMBOL_GPL(sync_page_io);
1003 
1004 static int read_disk_sb(struct md_rdev *rdev, int size)
1005 {
1006 	if (rdev->sb_loaded)
1007 		return 0;
1008 
1009 	if (!sync_page_io(rdev, 0, size, rdev->sb_page, REQ_OP_READ, true))
1010 		goto fail;
1011 	rdev->sb_loaded = 1;
1012 	return 0;
1013 
1014 fail:
1015 	pr_err("md: disabled device %pg, could not read superblock.\n",
1016 	       rdev->bdev);
1017 	return -EINVAL;
1018 }
1019 
1020 static int md_uuid_equal(mdp_super_t *sb1, mdp_super_t *sb2)
1021 {
1022 	return	sb1->set_uuid0 == sb2->set_uuid0 &&
1023 		sb1->set_uuid1 == sb2->set_uuid1 &&
1024 		sb1->set_uuid2 == sb2->set_uuid2 &&
1025 		sb1->set_uuid3 == sb2->set_uuid3;
1026 }
1027 
1028 static int md_sb_equal(mdp_super_t *sb1, mdp_super_t *sb2)
1029 {
1030 	int ret;
1031 	mdp_super_t *tmp1, *tmp2;
1032 
1033 	tmp1 = kmalloc(sizeof(*tmp1),GFP_KERNEL);
1034 	tmp2 = kmalloc(sizeof(*tmp2),GFP_KERNEL);
1035 
1036 	if (!tmp1 || !tmp2) {
1037 		ret = 0;
1038 		goto abort;
1039 	}
1040 
1041 	*tmp1 = *sb1;
1042 	*tmp2 = *sb2;
1043 
1044 	/*
1045 	 * nr_disks is not constant
1046 	 */
1047 	tmp1->nr_disks = 0;
1048 	tmp2->nr_disks = 0;
1049 
1050 	ret = (memcmp(tmp1, tmp2, MD_SB_GENERIC_CONSTANT_WORDS * 4) == 0);
1051 abort:
1052 	kfree(tmp1);
1053 	kfree(tmp2);
1054 	return ret;
1055 }
1056 
1057 static u32 md_csum_fold(u32 csum)
1058 {
1059 	csum = (csum & 0xffff) + (csum >> 16);
1060 	return (csum & 0xffff) + (csum >> 16);
1061 }
1062 
1063 static unsigned int calc_sb_csum(mdp_super_t *sb)
1064 {
1065 	u64 newcsum = 0;
1066 	u32 *sb32 = (u32*)sb;
1067 	int i;
1068 	unsigned int disk_csum, csum;
1069 
1070 	disk_csum = sb->sb_csum;
1071 	sb->sb_csum = 0;
1072 
1073 	for (i = 0; i < MD_SB_BYTES/4 ; i++)
1074 		newcsum += sb32[i];
1075 	csum = (newcsum & 0xffffffff) + (newcsum>>32);
1076 
1077 #ifdef CONFIG_ALPHA
1078 	/* This used to use csum_partial, which was wrong for several
1079 	 * reasons including that different results are returned on
1080 	 * different architectures.  It isn't critical that we get exactly
1081 	 * the same return value as before (we always csum_fold before
1082 	 * testing, and that removes any differences).  However as we
1083 	 * know that csum_partial always returned a 16bit value on
1084 	 * alphas, do a fold to maximise conformity to previous behaviour.
1085 	 */
1086 	sb->sb_csum = md_csum_fold(disk_csum);
1087 #else
1088 	sb->sb_csum = disk_csum;
1089 #endif
1090 	return csum;
1091 }
1092 
1093 /*
1094  * Handle superblock details.
1095  * We want to be able to handle multiple superblock formats
1096  * so we have a common interface to them all, and an array of
1097  * different handlers.
1098  * We rely on user-space to write the initial superblock, and support
1099  * reading and updating of superblocks.
1100  * Interface methods are:
1101  *   int load_super(struct md_rdev *dev, struct md_rdev *refdev, int minor_version)
1102  *      loads and validates a superblock on dev.
1103  *      if refdev != NULL, compare superblocks on both devices
1104  *    Return:
1105  *      0 - dev has a superblock that is compatible with refdev
1106  *      1 - dev has a superblock that is compatible and newer than refdev
1107  *          so dev should be used as the refdev in future
1108  *     -EINVAL superblock incompatible or invalid
1109  *     -othererror e.g. -EIO
1110  *
1111  *   int validate_super(struct mddev *mddev, struct md_rdev *dev)
1112  *      Verify that dev is acceptable into mddev.
1113  *       The first time, mddev->raid_disks will be 0, and data from
1114  *       dev should be merged in.  Subsequent calls check that dev
1115  *       is new enough.  Return 0 or -EINVAL
1116  *
1117  *   void sync_super(struct mddev *mddev, struct md_rdev *dev)
1118  *     Update the superblock for rdev with data in mddev
1119  *     This does not write to disc.
1120  *
1121  */
1122 
1123 struct super_type  {
1124 	char		    *name;
1125 	struct module	    *owner;
1126 	int		    (*load_super)(struct md_rdev *rdev,
1127 					  struct md_rdev *refdev,
1128 					  int minor_version);
1129 	int		    (*validate_super)(struct mddev *mddev,
1130 					      struct md_rdev *freshest,
1131 					      struct md_rdev *rdev);
1132 	void		    (*sync_super)(struct mddev *mddev,
1133 					  struct md_rdev *rdev);
1134 	unsigned long long  (*rdev_size_change)(struct md_rdev *rdev,
1135 						sector_t num_sectors);
1136 	int		    (*allow_new_offset)(struct md_rdev *rdev,
1137 						unsigned long long new_offset);
1138 };
1139 
1140 /*
1141  * Check that the given mddev has no bitmap.
1142  *
1143  * This function is called from the run method of all personalities that do not
1144  * support bitmaps. It prints an error message and returns non-zero if mddev
1145  * has a bitmap. Otherwise, it returns 0.
1146  *
1147  */
1148 int md_check_no_bitmap(struct mddev *mddev)
1149 {
1150 	if (!mddev->bitmap_info.file && !mddev->bitmap_info.offset)
1151 		return 0;
1152 	pr_warn("%s: bitmaps are not supported for %s\n",
1153 		mdname(mddev), mddev->pers->name);
1154 	return 1;
1155 }
1156 EXPORT_SYMBOL(md_check_no_bitmap);
1157 
1158 /*
1159  * load_super for 0.90.0
1160  */
1161 static int super_90_load(struct md_rdev *rdev, struct md_rdev *refdev, int minor_version)
1162 {
1163 	mdp_super_t *sb;
1164 	int ret;
1165 	bool spare_disk = true;
1166 
1167 	/*
1168 	 * Calculate the position of the superblock (512byte sectors),
1169 	 * it's at the end of the disk.
1170 	 *
1171 	 * It also happens to be a multiple of 4Kb.
1172 	 */
1173 	rdev->sb_start = calc_dev_sboffset(rdev);
1174 
1175 	ret = read_disk_sb(rdev, MD_SB_BYTES);
1176 	if (ret)
1177 		return ret;
1178 
1179 	ret = -EINVAL;
1180 
1181 	sb = page_address(rdev->sb_page);
1182 
1183 	if (sb->md_magic != MD_SB_MAGIC) {
1184 		pr_warn("md: invalid raid superblock magic on %pg\n",
1185 			rdev->bdev);
1186 		goto abort;
1187 	}
1188 
1189 	if (sb->major_version != 0 ||
1190 	    sb->minor_version < 90 ||
1191 	    sb->minor_version > 91) {
1192 		pr_warn("Bad version number %d.%d on %pg\n",
1193 			sb->major_version, sb->minor_version, rdev->bdev);
1194 		goto abort;
1195 	}
1196 
1197 	if (sb->raid_disks <= 0)
1198 		goto abort;
1199 
1200 	if (md_csum_fold(calc_sb_csum(sb)) != md_csum_fold(sb->sb_csum)) {
1201 		pr_warn("md: invalid superblock checksum on %pg\n", rdev->bdev);
1202 		goto abort;
1203 	}
1204 
1205 	rdev->preferred_minor = sb->md_minor;
1206 	rdev->data_offset = 0;
1207 	rdev->new_data_offset = 0;
1208 	rdev->sb_size = MD_SB_BYTES;
1209 	rdev->badblocks.shift = -1;
1210 
1211 	if (sb->level == LEVEL_MULTIPATH)
1212 		rdev->desc_nr = -1;
1213 	else
1214 		rdev->desc_nr = sb->this_disk.number;
1215 
1216 	/* not spare disk, or LEVEL_MULTIPATH */
1217 	if (sb->level == LEVEL_MULTIPATH ||
1218 		(rdev->desc_nr >= 0 &&
1219 		 rdev->desc_nr < MD_SB_DISKS &&
1220 		 sb->disks[rdev->desc_nr].state &
1221 		 ((1<<MD_DISK_SYNC) | (1 << MD_DISK_ACTIVE))))
1222 		spare_disk = false;
1223 
1224 	if (!refdev) {
1225 		if (!spare_disk)
1226 			ret = 1;
1227 		else
1228 			ret = 0;
1229 	} else {
1230 		__u64 ev1, ev2;
1231 		mdp_super_t *refsb = page_address(refdev->sb_page);
1232 		if (!md_uuid_equal(refsb, sb)) {
1233 			pr_warn("md: %pg has different UUID to %pg\n",
1234 				rdev->bdev, refdev->bdev);
1235 			goto abort;
1236 		}
1237 		if (!md_sb_equal(refsb, sb)) {
1238 			pr_warn("md: %pg has same UUID but different superblock to %pg\n",
1239 				rdev->bdev, refdev->bdev);
1240 			goto abort;
1241 		}
1242 		ev1 = md_event(sb);
1243 		ev2 = md_event(refsb);
1244 
1245 		if (!spare_disk && ev1 > ev2)
1246 			ret = 1;
1247 		else
1248 			ret = 0;
1249 	}
1250 	rdev->sectors = rdev->sb_start;
1251 	/* Limit to 4TB as metadata cannot record more than that.
1252 	 * (not needed for Linear and RAID0 as metadata doesn't
1253 	 * record this size)
1254 	 */
1255 	if ((u64)rdev->sectors >= (2ULL << 32) && sb->level >= 1)
1256 		rdev->sectors = (sector_t)(2ULL << 32) - 2;
1257 
1258 	if (rdev->sectors < ((sector_t)sb->size) * 2 && sb->level >= 1)
1259 		/* "this cannot possibly happen" ... */
1260 		ret = -EINVAL;
1261 
1262  abort:
1263 	return ret;
1264 }
1265 
1266 /*
1267  * validate_super for 0.90.0
1268  * note: we are not using "freshest" for 0.9 superblock
1269  */
1270 static int super_90_validate(struct mddev *mddev, struct md_rdev *freshest, struct md_rdev *rdev)
1271 {
1272 	mdp_disk_t *desc;
1273 	mdp_super_t *sb = page_address(rdev->sb_page);
1274 	__u64 ev1 = md_event(sb);
1275 
1276 	rdev->raid_disk = -1;
1277 	clear_bit(Faulty, &rdev->flags);
1278 	clear_bit(In_sync, &rdev->flags);
1279 	clear_bit(Bitmap_sync, &rdev->flags);
1280 	clear_bit(WriteMostly, &rdev->flags);
1281 
1282 	if (mddev->raid_disks == 0) {
1283 		mddev->major_version = 0;
1284 		mddev->minor_version = sb->minor_version;
1285 		mddev->patch_version = sb->patch_version;
1286 		mddev->external = 0;
1287 		mddev->chunk_sectors = sb->chunk_size >> 9;
1288 		mddev->ctime = sb->ctime;
1289 		mddev->utime = sb->utime;
1290 		mddev->level = sb->level;
1291 		mddev->clevel[0] = 0;
1292 		mddev->layout = sb->layout;
1293 		mddev->raid_disks = sb->raid_disks;
1294 		mddev->dev_sectors = ((sector_t)sb->size) * 2;
1295 		mddev->events = ev1;
1296 		mddev->bitmap_info.offset = 0;
1297 		mddev->bitmap_info.space = 0;
1298 		/* bitmap can use 60 K after the 4K superblocks */
1299 		mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9;
1300 		mddev->bitmap_info.default_space = 64*2 - (MD_SB_BYTES >> 9);
1301 		mddev->reshape_backwards = 0;
1302 
1303 		if (mddev->minor_version >= 91) {
1304 			mddev->reshape_position = sb->reshape_position;
1305 			mddev->delta_disks = sb->delta_disks;
1306 			mddev->new_level = sb->new_level;
1307 			mddev->new_layout = sb->new_layout;
1308 			mddev->new_chunk_sectors = sb->new_chunk >> 9;
1309 			if (mddev->delta_disks < 0)
1310 				mddev->reshape_backwards = 1;
1311 		} else {
1312 			mddev->reshape_position = MaxSector;
1313 			mddev->delta_disks = 0;
1314 			mddev->new_level = mddev->level;
1315 			mddev->new_layout = mddev->layout;
1316 			mddev->new_chunk_sectors = mddev->chunk_sectors;
1317 		}
1318 		if (mddev->level == 0)
1319 			mddev->layout = -1;
1320 
1321 		if (sb->state & (1<<MD_SB_CLEAN))
1322 			mddev->recovery_cp = MaxSector;
1323 		else {
1324 			if (sb->events_hi == sb->cp_events_hi &&
1325 				sb->events_lo == sb->cp_events_lo) {
1326 				mddev->recovery_cp = sb->recovery_cp;
1327 			} else
1328 				mddev->recovery_cp = 0;
1329 		}
1330 
1331 		memcpy(mddev->uuid+0, &sb->set_uuid0, 4);
1332 		memcpy(mddev->uuid+4, &sb->set_uuid1, 4);
1333 		memcpy(mddev->uuid+8, &sb->set_uuid2, 4);
1334 		memcpy(mddev->uuid+12,&sb->set_uuid3, 4);
1335 
1336 		mddev->max_disks = MD_SB_DISKS;
1337 
1338 		if (sb->state & (1<<MD_SB_BITMAP_PRESENT) &&
1339 		    mddev->bitmap_info.file == NULL) {
1340 			mddev->bitmap_info.offset =
1341 				mddev->bitmap_info.default_offset;
1342 			mddev->bitmap_info.space =
1343 				mddev->bitmap_info.default_space;
1344 		}
1345 
1346 	} else if (mddev->pers == NULL) {
1347 		/* Insist on good event counter while assembling, except
1348 		 * for spares (which don't need an event count) */
1349 		++ev1;
1350 		if (sb->disks[rdev->desc_nr].state & (
1351 			    (1<<MD_DISK_SYNC) | (1 << MD_DISK_ACTIVE)))
1352 			if (ev1 < mddev->events)
1353 				return -EINVAL;
1354 	} else if (mddev->bitmap) {
1355 		/* if adding to array with a bitmap, then we can accept an
1356 		 * older device ... but not too old.
1357 		 */
1358 		if (ev1 < mddev->bitmap->events_cleared)
1359 			return 0;
1360 		if (ev1 < mddev->events)
1361 			set_bit(Bitmap_sync, &rdev->flags);
1362 	} else {
1363 		if (ev1 < mddev->events)
1364 			/* just a hot-add of a new device, leave raid_disk at -1 */
1365 			return 0;
1366 	}
1367 
1368 	if (mddev->level != LEVEL_MULTIPATH) {
1369 		desc = sb->disks + rdev->desc_nr;
1370 
1371 		if (desc->state & (1<<MD_DISK_FAULTY))
1372 			set_bit(Faulty, &rdev->flags);
1373 		else if (desc->state & (1<<MD_DISK_SYNC) /* &&
1374 			    desc->raid_disk < mddev->raid_disks */) {
1375 			set_bit(In_sync, &rdev->flags);
1376 			rdev->raid_disk = desc->raid_disk;
1377 			rdev->saved_raid_disk = desc->raid_disk;
1378 		} else if (desc->state & (1<<MD_DISK_ACTIVE)) {
1379 			/* active but not in sync implies recovery up to
1380 			 * reshape position.  We don't know exactly where
1381 			 * that is, so set to zero for now */
1382 			if (mddev->minor_version >= 91) {
1383 				rdev->recovery_offset = 0;
1384 				rdev->raid_disk = desc->raid_disk;
1385 			}
1386 		}
1387 		if (desc->state & (1<<MD_DISK_WRITEMOSTLY))
1388 			set_bit(WriteMostly, &rdev->flags);
1389 		if (desc->state & (1<<MD_DISK_FAILFAST))
1390 			set_bit(FailFast, &rdev->flags);
1391 	} else /* MULTIPATH are always insync */
1392 		set_bit(In_sync, &rdev->flags);
1393 	return 0;
1394 }
1395 
1396 /*
1397  * sync_super for 0.90.0
1398  */
1399 static void super_90_sync(struct mddev *mddev, struct md_rdev *rdev)
1400 {
1401 	mdp_super_t *sb;
1402 	struct md_rdev *rdev2;
1403 	int next_spare = mddev->raid_disks;
1404 
1405 	/* make rdev->sb match mddev data..
1406 	 *
1407 	 * 1/ zero out disks
1408 	 * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
1409 	 * 3/ any empty disks < next_spare become removed
1410 	 *
1411 	 * disks[0] gets initialised to REMOVED because
1412 	 * we cannot be sure from other fields if it has
1413 	 * been initialised or not.
1414 	 */
1415 	int i;
1416 	int active=0, working=0,failed=0,spare=0,nr_disks=0;
1417 
1418 	rdev->sb_size = MD_SB_BYTES;
1419 
1420 	sb = page_address(rdev->sb_page);
1421 
1422 	memset(sb, 0, sizeof(*sb));
1423 
1424 	sb->md_magic = MD_SB_MAGIC;
1425 	sb->major_version = mddev->major_version;
1426 	sb->patch_version = mddev->patch_version;
1427 	sb->gvalid_words  = 0; /* ignored */
1428 	memcpy(&sb->set_uuid0, mddev->uuid+0, 4);
1429 	memcpy(&sb->set_uuid1, mddev->uuid+4, 4);
1430 	memcpy(&sb->set_uuid2, mddev->uuid+8, 4);
1431 	memcpy(&sb->set_uuid3, mddev->uuid+12,4);
1432 
1433 	sb->ctime = clamp_t(time64_t, mddev->ctime, 0, U32_MAX);
1434 	sb->level = mddev->level;
1435 	sb->size = mddev->dev_sectors / 2;
1436 	sb->raid_disks = mddev->raid_disks;
1437 	sb->md_minor = mddev->md_minor;
1438 	sb->not_persistent = 0;
1439 	sb->utime = clamp_t(time64_t, mddev->utime, 0, U32_MAX);
1440 	sb->state = 0;
1441 	sb->events_hi = (mddev->events>>32);
1442 	sb->events_lo = (u32)mddev->events;
1443 
1444 	if (mddev->reshape_position == MaxSector)
1445 		sb->minor_version = 90;
1446 	else {
1447 		sb->minor_version = 91;
1448 		sb->reshape_position = mddev->reshape_position;
1449 		sb->new_level = mddev->new_level;
1450 		sb->delta_disks = mddev->delta_disks;
1451 		sb->new_layout = mddev->new_layout;
1452 		sb->new_chunk = mddev->new_chunk_sectors << 9;
1453 	}
1454 	mddev->minor_version = sb->minor_version;
1455 	if (mddev->in_sync)
1456 	{
1457 		sb->recovery_cp = mddev->recovery_cp;
1458 		sb->cp_events_hi = (mddev->events>>32);
1459 		sb->cp_events_lo = (u32)mddev->events;
1460 		if (mddev->recovery_cp == MaxSector)
1461 			sb->state = (1<< MD_SB_CLEAN);
1462 	} else
1463 		sb->recovery_cp = 0;
1464 
1465 	sb->layout = mddev->layout;
1466 	sb->chunk_size = mddev->chunk_sectors << 9;
1467 
1468 	if (mddev->bitmap && mddev->bitmap_info.file == NULL)
1469 		sb->state |= (1<<MD_SB_BITMAP_PRESENT);
1470 
1471 	sb->disks[0].state = (1<<MD_DISK_REMOVED);
1472 	rdev_for_each(rdev2, mddev) {
1473 		mdp_disk_t *d;
1474 		int desc_nr;
1475 		int is_active = test_bit(In_sync, &rdev2->flags);
1476 
1477 		if (rdev2->raid_disk >= 0 &&
1478 		    sb->minor_version >= 91)
1479 			/* we have nowhere to store the recovery_offset,
1480 			 * but if it is not below the reshape_position,
1481 			 * we can piggy-back on that.
1482 			 */
1483 			is_active = 1;
1484 		if (rdev2->raid_disk < 0 ||
1485 		    test_bit(Faulty, &rdev2->flags))
1486 			is_active = 0;
1487 		if (is_active)
1488 			desc_nr = rdev2->raid_disk;
1489 		else
1490 			desc_nr = next_spare++;
1491 		rdev2->desc_nr = desc_nr;
1492 		d = &sb->disks[rdev2->desc_nr];
1493 		nr_disks++;
1494 		d->number = rdev2->desc_nr;
1495 		d->major = MAJOR(rdev2->bdev->bd_dev);
1496 		d->minor = MINOR(rdev2->bdev->bd_dev);
1497 		if (is_active)
1498 			d->raid_disk = rdev2->raid_disk;
1499 		else
1500 			d->raid_disk = rdev2->desc_nr; /* compatibility */
1501 		if (test_bit(Faulty, &rdev2->flags))
1502 			d->state = (1<<MD_DISK_FAULTY);
1503 		else if (is_active) {
1504 			d->state = (1<<MD_DISK_ACTIVE);
1505 			if (test_bit(In_sync, &rdev2->flags))
1506 				d->state |= (1<<MD_DISK_SYNC);
1507 			active++;
1508 			working++;
1509 		} else {
1510 			d->state = 0;
1511 			spare++;
1512 			working++;
1513 		}
1514 		if (test_bit(WriteMostly, &rdev2->flags))
1515 			d->state |= (1<<MD_DISK_WRITEMOSTLY);
1516 		if (test_bit(FailFast, &rdev2->flags))
1517 			d->state |= (1<<MD_DISK_FAILFAST);
1518 	}
1519 	/* now set the "removed" and "faulty" bits on any missing devices */
1520 	for (i=0 ; i < mddev->raid_disks ; i++) {
1521 		mdp_disk_t *d = &sb->disks[i];
1522 		if (d->state == 0 && d->number == 0) {
1523 			d->number = i;
1524 			d->raid_disk = i;
1525 			d->state = (1<<MD_DISK_REMOVED);
1526 			d->state |= (1<<MD_DISK_FAULTY);
1527 			failed++;
1528 		}
1529 	}
1530 	sb->nr_disks = nr_disks;
1531 	sb->active_disks = active;
1532 	sb->working_disks = working;
1533 	sb->failed_disks = failed;
1534 	sb->spare_disks = spare;
1535 
1536 	sb->this_disk = sb->disks[rdev->desc_nr];
1537 	sb->sb_csum = calc_sb_csum(sb);
1538 }
1539 
1540 /*
1541  * rdev_size_change for 0.90.0
1542  */
1543 static unsigned long long
1544 super_90_rdev_size_change(struct md_rdev *rdev, sector_t num_sectors)
1545 {
1546 	if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
1547 		return 0; /* component must fit device */
1548 	if (rdev->mddev->bitmap_info.offset)
1549 		return 0; /* can't move bitmap */
1550 	rdev->sb_start = calc_dev_sboffset(rdev);
1551 	if (!num_sectors || num_sectors > rdev->sb_start)
1552 		num_sectors = rdev->sb_start;
1553 	/* Limit to 4TB as metadata cannot record more than that.
1554 	 * 4TB == 2^32 KB, or 2*2^32 sectors.
1555 	 */
1556 	if ((u64)num_sectors >= (2ULL << 32) && rdev->mddev->level >= 1)
1557 		num_sectors = (sector_t)(2ULL << 32) - 2;
1558 	do {
1559 		md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
1560 		       rdev->sb_page);
1561 	} while (md_super_wait(rdev->mddev) < 0);
1562 	return num_sectors;
1563 }
1564 
1565 static int
1566 super_90_allow_new_offset(struct md_rdev *rdev, unsigned long long new_offset)
1567 {
1568 	/* non-zero offset changes not possible with v0.90 */
1569 	return new_offset == 0;
1570 }
1571 
1572 /*
1573  * version 1 superblock
1574  */
1575 
1576 static __le32 calc_sb_1_csum(struct mdp_superblock_1 *sb)
1577 {
1578 	__le32 disk_csum;
1579 	u32 csum;
1580 	unsigned long long newcsum;
1581 	int size = 256 + le32_to_cpu(sb->max_dev)*2;
1582 	__le32 *isuper = (__le32*)sb;
1583 
1584 	disk_csum = sb->sb_csum;
1585 	sb->sb_csum = 0;
1586 	newcsum = 0;
1587 	for (; size >= 4; size -= 4)
1588 		newcsum += le32_to_cpu(*isuper++);
1589 
1590 	if (size == 2)
1591 		newcsum += le16_to_cpu(*(__le16*) isuper);
1592 
1593 	csum = (newcsum & 0xffffffff) + (newcsum >> 32);
1594 	sb->sb_csum = disk_csum;
1595 	return cpu_to_le32(csum);
1596 }
1597 
1598 static int super_1_load(struct md_rdev *rdev, struct md_rdev *refdev, int minor_version)
1599 {
1600 	struct mdp_superblock_1 *sb;
1601 	int ret;
1602 	sector_t sb_start;
1603 	sector_t sectors;
1604 	int bmask;
1605 	bool spare_disk = true;
1606 
1607 	/*
1608 	 * Calculate the position of the superblock in 512byte sectors.
1609 	 * It is always aligned to a 4K boundary and
1610 	 * depeding on minor_version, it can be:
1611 	 * 0: At least 8K, but less than 12K, from end of device
1612 	 * 1: At start of device
1613 	 * 2: 4K from start of device.
1614 	 */
1615 	switch(minor_version) {
1616 	case 0:
1617 		sb_start = bdev_nr_sectors(rdev->bdev) - 8 * 2;
1618 		sb_start &= ~(sector_t)(4*2-1);
1619 		break;
1620 	case 1:
1621 		sb_start = 0;
1622 		break;
1623 	case 2:
1624 		sb_start = 8;
1625 		break;
1626 	default:
1627 		return -EINVAL;
1628 	}
1629 	rdev->sb_start = sb_start;
1630 
1631 	/* superblock is rarely larger than 1K, but it can be larger,
1632 	 * and it is safe to read 4k, so we do that
1633 	 */
1634 	ret = read_disk_sb(rdev, 4096);
1635 	if (ret) return ret;
1636 
1637 	sb = page_address(rdev->sb_page);
1638 
1639 	if (sb->magic != cpu_to_le32(MD_SB_MAGIC) ||
1640 	    sb->major_version != cpu_to_le32(1) ||
1641 	    le32_to_cpu(sb->max_dev) > (4096-256)/2 ||
1642 	    le64_to_cpu(sb->super_offset) != rdev->sb_start ||
1643 	    (le32_to_cpu(sb->feature_map) & ~MD_FEATURE_ALL) != 0)
1644 		return -EINVAL;
1645 
1646 	if (calc_sb_1_csum(sb) != sb->sb_csum) {
1647 		pr_warn("md: invalid superblock checksum on %pg\n",
1648 			rdev->bdev);
1649 		return -EINVAL;
1650 	}
1651 	if (le64_to_cpu(sb->data_size) < 10) {
1652 		pr_warn("md: data_size too small on %pg\n",
1653 			rdev->bdev);
1654 		return -EINVAL;
1655 	}
1656 	if (sb->pad0 ||
1657 	    sb->pad3[0] ||
1658 	    memcmp(sb->pad3, sb->pad3+1, sizeof(sb->pad3) - sizeof(sb->pad3[1])))
1659 		/* Some padding is non-zero, might be a new feature */
1660 		return -EINVAL;
1661 
1662 	rdev->preferred_minor = 0xffff;
1663 	rdev->data_offset = le64_to_cpu(sb->data_offset);
1664 	rdev->new_data_offset = rdev->data_offset;
1665 	if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE) &&
1666 	    (le32_to_cpu(sb->feature_map) & MD_FEATURE_NEW_OFFSET))
1667 		rdev->new_data_offset += (s32)le32_to_cpu(sb->new_offset);
1668 	atomic_set(&rdev->corrected_errors, le32_to_cpu(sb->cnt_corrected_read));
1669 
1670 	rdev->sb_size = le32_to_cpu(sb->max_dev) * 2 + 256;
1671 	bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
1672 	if (rdev->sb_size & bmask)
1673 		rdev->sb_size = (rdev->sb_size | bmask) + 1;
1674 
1675 	if (minor_version
1676 	    && rdev->data_offset < sb_start + (rdev->sb_size/512))
1677 		return -EINVAL;
1678 	if (minor_version
1679 	    && rdev->new_data_offset < sb_start + (rdev->sb_size/512))
1680 		return -EINVAL;
1681 
1682 	if (sb->level == cpu_to_le32(LEVEL_MULTIPATH))
1683 		rdev->desc_nr = -1;
1684 	else
1685 		rdev->desc_nr = le32_to_cpu(sb->dev_number);
1686 
1687 	if (!rdev->bb_page) {
1688 		rdev->bb_page = alloc_page(GFP_KERNEL);
1689 		if (!rdev->bb_page)
1690 			return -ENOMEM;
1691 	}
1692 	if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BAD_BLOCKS) &&
1693 	    rdev->badblocks.count == 0) {
1694 		/* need to load the bad block list.
1695 		 * Currently we limit it to one page.
1696 		 */
1697 		s32 offset;
1698 		sector_t bb_sector;
1699 		__le64 *bbp;
1700 		int i;
1701 		int sectors = le16_to_cpu(sb->bblog_size);
1702 		if (sectors > (PAGE_SIZE / 512))
1703 			return -EINVAL;
1704 		offset = le32_to_cpu(sb->bblog_offset);
1705 		if (offset == 0)
1706 			return -EINVAL;
1707 		bb_sector = (long long)offset;
1708 		if (!sync_page_io(rdev, bb_sector, sectors << 9,
1709 				  rdev->bb_page, REQ_OP_READ, true))
1710 			return -EIO;
1711 		bbp = (__le64 *)page_address(rdev->bb_page);
1712 		rdev->badblocks.shift = sb->bblog_shift;
1713 		for (i = 0 ; i < (sectors << (9-3)) ; i++, bbp++) {
1714 			u64 bb = le64_to_cpu(*bbp);
1715 			int count = bb & (0x3ff);
1716 			u64 sector = bb >> 10;
1717 			sector <<= sb->bblog_shift;
1718 			count <<= sb->bblog_shift;
1719 			if (bb + 1 == 0)
1720 				break;
1721 			if (badblocks_set(&rdev->badblocks, sector, count, 1))
1722 				return -EINVAL;
1723 		}
1724 	} else if (sb->bblog_offset != 0)
1725 		rdev->badblocks.shift = 0;
1726 
1727 	if ((le32_to_cpu(sb->feature_map) &
1728 	    (MD_FEATURE_PPL | MD_FEATURE_MULTIPLE_PPLS))) {
1729 		rdev->ppl.offset = (__s16)le16_to_cpu(sb->ppl.offset);
1730 		rdev->ppl.size = le16_to_cpu(sb->ppl.size);
1731 		rdev->ppl.sector = rdev->sb_start + rdev->ppl.offset;
1732 	}
1733 
1734 	if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RAID0_LAYOUT) &&
1735 	    sb->level != 0)
1736 		return -EINVAL;
1737 
1738 	/* not spare disk, or LEVEL_MULTIPATH */
1739 	if (sb->level == cpu_to_le32(LEVEL_MULTIPATH) ||
1740 		(rdev->desc_nr >= 0 &&
1741 		rdev->desc_nr < le32_to_cpu(sb->max_dev) &&
1742 		(le16_to_cpu(sb->dev_roles[rdev->desc_nr]) < MD_DISK_ROLE_MAX ||
1743 		 le16_to_cpu(sb->dev_roles[rdev->desc_nr]) == MD_DISK_ROLE_JOURNAL)))
1744 		spare_disk = false;
1745 
1746 	if (!refdev) {
1747 		if (!spare_disk)
1748 			ret = 1;
1749 		else
1750 			ret = 0;
1751 	} else {
1752 		__u64 ev1, ev2;
1753 		struct mdp_superblock_1 *refsb = page_address(refdev->sb_page);
1754 
1755 		if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 ||
1756 		    sb->level != refsb->level ||
1757 		    sb->layout != refsb->layout ||
1758 		    sb->chunksize != refsb->chunksize) {
1759 			pr_warn("md: %pg has strangely different superblock to %pg\n",
1760 				rdev->bdev,
1761 				refdev->bdev);
1762 			return -EINVAL;
1763 		}
1764 		ev1 = le64_to_cpu(sb->events);
1765 		ev2 = le64_to_cpu(refsb->events);
1766 
1767 		if (!spare_disk && ev1 > ev2)
1768 			ret = 1;
1769 		else
1770 			ret = 0;
1771 	}
1772 	if (minor_version)
1773 		sectors = bdev_nr_sectors(rdev->bdev) - rdev->data_offset;
1774 	else
1775 		sectors = rdev->sb_start;
1776 	if (sectors < le64_to_cpu(sb->data_size))
1777 		return -EINVAL;
1778 	rdev->sectors = le64_to_cpu(sb->data_size);
1779 	return ret;
1780 }
1781 
1782 static int super_1_validate(struct mddev *mddev, struct md_rdev *freshest, struct md_rdev *rdev)
1783 {
1784 	struct mdp_superblock_1 *sb = page_address(rdev->sb_page);
1785 	__u64 ev1 = le64_to_cpu(sb->events);
1786 
1787 	rdev->raid_disk = -1;
1788 	clear_bit(Faulty, &rdev->flags);
1789 	clear_bit(In_sync, &rdev->flags);
1790 	clear_bit(Bitmap_sync, &rdev->flags);
1791 	clear_bit(WriteMostly, &rdev->flags);
1792 
1793 	if (mddev->raid_disks == 0) {
1794 		mddev->major_version = 1;
1795 		mddev->patch_version = 0;
1796 		mddev->external = 0;
1797 		mddev->chunk_sectors = le32_to_cpu(sb->chunksize);
1798 		mddev->ctime = le64_to_cpu(sb->ctime);
1799 		mddev->utime = le64_to_cpu(sb->utime);
1800 		mddev->level = le32_to_cpu(sb->level);
1801 		mddev->clevel[0] = 0;
1802 		mddev->layout = le32_to_cpu(sb->layout);
1803 		mddev->raid_disks = le32_to_cpu(sb->raid_disks);
1804 		mddev->dev_sectors = le64_to_cpu(sb->size);
1805 		mddev->events = ev1;
1806 		mddev->bitmap_info.offset = 0;
1807 		mddev->bitmap_info.space = 0;
1808 		/* Default location for bitmap is 1K after superblock
1809 		 * using 3K - total of 4K
1810 		 */
1811 		mddev->bitmap_info.default_offset = 1024 >> 9;
1812 		mddev->bitmap_info.default_space = (4096-1024) >> 9;
1813 		mddev->reshape_backwards = 0;
1814 
1815 		mddev->recovery_cp = le64_to_cpu(sb->resync_offset);
1816 		memcpy(mddev->uuid, sb->set_uuid, 16);
1817 
1818 		mddev->max_disks =  (4096-256)/2;
1819 
1820 		if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) &&
1821 		    mddev->bitmap_info.file == NULL) {
1822 			mddev->bitmap_info.offset =
1823 				(__s32)le32_to_cpu(sb->bitmap_offset);
1824 			/* Metadata doesn't record how much space is available.
1825 			 * For 1.0, we assume we can use up to the superblock
1826 			 * if before, else to 4K beyond superblock.
1827 			 * For others, assume no change is possible.
1828 			 */
1829 			if (mddev->minor_version > 0)
1830 				mddev->bitmap_info.space = 0;
1831 			else if (mddev->bitmap_info.offset > 0)
1832 				mddev->bitmap_info.space =
1833 					8 - mddev->bitmap_info.offset;
1834 			else
1835 				mddev->bitmap_info.space =
1836 					-mddev->bitmap_info.offset;
1837 		}
1838 
1839 		if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) {
1840 			mddev->reshape_position = le64_to_cpu(sb->reshape_position);
1841 			mddev->delta_disks = le32_to_cpu(sb->delta_disks);
1842 			mddev->new_level = le32_to_cpu(sb->new_level);
1843 			mddev->new_layout = le32_to_cpu(sb->new_layout);
1844 			mddev->new_chunk_sectors = le32_to_cpu(sb->new_chunk);
1845 			if (mddev->delta_disks < 0 ||
1846 			    (mddev->delta_disks == 0 &&
1847 			     (le32_to_cpu(sb->feature_map)
1848 			      & MD_FEATURE_RESHAPE_BACKWARDS)))
1849 				mddev->reshape_backwards = 1;
1850 		} else {
1851 			mddev->reshape_position = MaxSector;
1852 			mddev->delta_disks = 0;
1853 			mddev->new_level = mddev->level;
1854 			mddev->new_layout = mddev->layout;
1855 			mddev->new_chunk_sectors = mddev->chunk_sectors;
1856 		}
1857 
1858 		if (mddev->level == 0 &&
1859 		    !(le32_to_cpu(sb->feature_map) & MD_FEATURE_RAID0_LAYOUT))
1860 			mddev->layout = -1;
1861 
1862 		if (le32_to_cpu(sb->feature_map) & MD_FEATURE_JOURNAL)
1863 			set_bit(MD_HAS_JOURNAL, &mddev->flags);
1864 
1865 		if (le32_to_cpu(sb->feature_map) &
1866 		    (MD_FEATURE_PPL | MD_FEATURE_MULTIPLE_PPLS)) {
1867 			if (le32_to_cpu(sb->feature_map) &
1868 			    (MD_FEATURE_BITMAP_OFFSET | MD_FEATURE_JOURNAL))
1869 				return -EINVAL;
1870 			if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_PPL) &&
1871 			    (le32_to_cpu(sb->feature_map) &
1872 					    MD_FEATURE_MULTIPLE_PPLS))
1873 				return -EINVAL;
1874 			set_bit(MD_HAS_PPL, &mddev->flags);
1875 		}
1876 	} else if (mddev->pers == NULL) {
1877 		/* Insist of good event counter while assembling, except for
1878 		 * spares (which don't need an event count).
1879 		 * Similar to mdadm, we allow event counter difference of 1
1880 		 * from the freshest device.
1881 		 */
1882 		if (rdev->desc_nr >= 0 &&
1883 		    rdev->desc_nr < le32_to_cpu(sb->max_dev) &&
1884 		    (le16_to_cpu(sb->dev_roles[rdev->desc_nr]) < MD_DISK_ROLE_MAX ||
1885 		     le16_to_cpu(sb->dev_roles[rdev->desc_nr]) == MD_DISK_ROLE_JOURNAL))
1886 			if (ev1 + 1 < mddev->events)
1887 				return -EINVAL;
1888 	} else if (mddev->bitmap) {
1889 		/* If adding to array with a bitmap, then we can accept an
1890 		 * older device, but not too old.
1891 		 */
1892 		if (ev1 < mddev->bitmap->events_cleared)
1893 			return 0;
1894 		if (ev1 < mddev->events)
1895 			set_bit(Bitmap_sync, &rdev->flags);
1896 	} else {
1897 		if (ev1 < mddev->events)
1898 			/* just a hot-add of a new device, leave raid_disk at -1 */
1899 			return 0;
1900 	}
1901 	if (mddev->level != LEVEL_MULTIPATH) {
1902 		int role;
1903 		if (rdev->desc_nr < 0 ||
1904 		    rdev->desc_nr >= le32_to_cpu(sb->max_dev)) {
1905 			role = MD_DISK_ROLE_SPARE;
1906 			rdev->desc_nr = -1;
1907 		} else if (mddev->pers == NULL && freshest && ev1 < mddev->events) {
1908 			/*
1909 			 * If we are assembling, and our event counter is smaller than the
1910 			 * highest event counter, we cannot trust our superblock about the role.
1911 			 * It could happen that our rdev was marked as Faulty, and all other
1912 			 * superblocks were updated with +1 event counter.
1913 			 * Then, before the next superblock update, which typically happens when
1914 			 * remove_and_add_spares() removes the device from the array, there was
1915 			 * a crash or reboot.
1916 			 * If we allow current rdev without consulting the freshest superblock,
1917 			 * we could cause data corruption.
1918 			 * Note that in this case our event counter is smaller by 1 than the
1919 			 * highest, otherwise, this rdev would not be allowed into array;
1920 			 * both kernel and mdadm allow event counter difference of 1.
1921 			 */
1922 			struct mdp_superblock_1 *freshest_sb = page_address(freshest->sb_page);
1923 			u32 freshest_max_dev = le32_to_cpu(freshest_sb->max_dev);
1924 
1925 			if (rdev->desc_nr >= freshest_max_dev) {
1926 				/* this is unexpected, better not proceed */
1927 				pr_warn("md: %s: rdev[%pg]: desc_nr(%d) >= freshest(%pg)->sb->max_dev(%u)\n",
1928 						mdname(mddev), rdev->bdev, rdev->desc_nr,
1929 						freshest->bdev, freshest_max_dev);
1930 				return -EUCLEAN;
1931 			}
1932 
1933 			role = le16_to_cpu(freshest_sb->dev_roles[rdev->desc_nr]);
1934 			pr_debug("md: %s: rdev[%pg]: role=%d(0x%x) according to freshest %pg\n",
1935 				     mdname(mddev), rdev->bdev, role, role, freshest->bdev);
1936 		} else {
1937 			role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
1938 		}
1939 		switch(role) {
1940 		case MD_DISK_ROLE_SPARE: /* spare */
1941 			break;
1942 		case MD_DISK_ROLE_FAULTY: /* faulty */
1943 			set_bit(Faulty, &rdev->flags);
1944 			break;
1945 		case MD_DISK_ROLE_JOURNAL: /* journal device */
1946 			if (!(le32_to_cpu(sb->feature_map) & MD_FEATURE_JOURNAL)) {
1947 				/* journal device without journal feature */
1948 				pr_warn("md: journal device provided without journal feature, ignoring the device\n");
1949 				return -EINVAL;
1950 			}
1951 			set_bit(Journal, &rdev->flags);
1952 			rdev->journal_tail = le64_to_cpu(sb->journal_tail);
1953 			rdev->raid_disk = 0;
1954 			break;
1955 		default:
1956 			rdev->saved_raid_disk = role;
1957 			if ((le32_to_cpu(sb->feature_map) &
1958 			     MD_FEATURE_RECOVERY_OFFSET)) {
1959 				rdev->recovery_offset = le64_to_cpu(sb->recovery_offset);
1960 				if (!(le32_to_cpu(sb->feature_map) &
1961 				      MD_FEATURE_RECOVERY_BITMAP))
1962 					rdev->saved_raid_disk = -1;
1963 			} else {
1964 				/*
1965 				 * If the array is FROZEN, then the device can't
1966 				 * be in_sync with rest of array.
1967 				 */
1968 				if (!test_bit(MD_RECOVERY_FROZEN,
1969 					      &mddev->recovery))
1970 					set_bit(In_sync, &rdev->flags);
1971 			}
1972 			rdev->raid_disk = role;
1973 			break;
1974 		}
1975 		if (sb->devflags & WriteMostly1)
1976 			set_bit(WriteMostly, &rdev->flags);
1977 		if (sb->devflags & FailFast1)
1978 			set_bit(FailFast, &rdev->flags);
1979 		if (le32_to_cpu(sb->feature_map) & MD_FEATURE_REPLACEMENT)
1980 			set_bit(Replacement, &rdev->flags);
1981 	} else /* MULTIPATH are always insync */
1982 		set_bit(In_sync, &rdev->flags);
1983 
1984 	return 0;
1985 }
1986 
1987 static void super_1_sync(struct mddev *mddev, struct md_rdev *rdev)
1988 {
1989 	struct mdp_superblock_1 *sb;
1990 	struct md_rdev *rdev2;
1991 	int max_dev, i;
1992 	/* make rdev->sb match mddev and rdev data. */
1993 
1994 	sb = page_address(rdev->sb_page);
1995 
1996 	sb->feature_map = 0;
1997 	sb->pad0 = 0;
1998 	sb->recovery_offset = cpu_to_le64(0);
1999 	memset(sb->pad3, 0, sizeof(sb->pad3));
2000 
2001 	sb->utime = cpu_to_le64((__u64)mddev->utime);
2002 	sb->events = cpu_to_le64(mddev->events);
2003 	if (mddev->in_sync)
2004 		sb->resync_offset = cpu_to_le64(mddev->recovery_cp);
2005 	else if (test_bit(MD_JOURNAL_CLEAN, &mddev->flags))
2006 		sb->resync_offset = cpu_to_le64(MaxSector);
2007 	else
2008 		sb->resync_offset = cpu_to_le64(0);
2009 
2010 	sb->cnt_corrected_read = cpu_to_le32(atomic_read(&rdev->corrected_errors));
2011 
2012 	sb->raid_disks = cpu_to_le32(mddev->raid_disks);
2013 	sb->size = cpu_to_le64(mddev->dev_sectors);
2014 	sb->chunksize = cpu_to_le32(mddev->chunk_sectors);
2015 	sb->level = cpu_to_le32(mddev->level);
2016 	sb->layout = cpu_to_le32(mddev->layout);
2017 	if (test_bit(FailFast, &rdev->flags))
2018 		sb->devflags |= FailFast1;
2019 	else
2020 		sb->devflags &= ~FailFast1;
2021 
2022 	if (test_bit(WriteMostly, &rdev->flags))
2023 		sb->devflags |= WriteMostly1;
2024 	else
2025 		sb->devflags &= ~WriteMostly1;
2026 	sb->data_offset = cpu_to_le64(rdev->data_offset);
2027 	sb->data_size = cpu_to_le64(rdev->sectors);
2028 
2029 	if (mddev->bitmap && mddev->bitmap_info.file == NULL) {
2030 		sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_info.offset);
2031 		sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET);
2032 	}
2033 
2034 	if (rdev->raid_disk >= 0 && !test_bit(Journal, &rdev->flags) &&
2035 	    !test_bit(In_sync, &rdev->flags)) {
2036 		sb->feature_map |=
2037 			cpu_to_le32(MD_FEATURE_RECOVERY_OFFSET);
2038 		sb->recovery_offset =
2039 			cpu_to_le64(rdev->recovery_offset);
2040 		if (rdev->saved_raid_disk >= 0 && mddev->bitmap)
2041 			sb->feature_map |=
2042 				cpu_to_le32(MD_FEATURE_RECOVERY_BITMAP);
2043 	}
2044 	/* Note: recovery_offset and journal_tail share space  */
2045 	if (test_bit(Journal, &rdev->flags))
2046 		sb->journal_tail = cpu_to_le64(rdev->journal_tail);
2047 	if (test_bit(Replacement, &rdev->flags))
2048 		sb->feature_map |=
2049 			cpu_to_le32(MD_FEATURE_REPLACEMENT);
2050 
2051 	if (mddev->reshape_position != MaxSector) {
2052 		sb->feature_map |= cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE);
2053 		sb->reshape_position = cpu_to_le64(mddev->reshape_position);
2054 		sb->new_layout = cpu_to_le32(mddev->new_layout);
2055 		sb->delta_disks = cpu_to_le32(mddev->delta_disks);
2056 		sb->new_level = cpu_to_le32(mddev->new_level);
2057 		sb->new_chunk = cpu_to_le32(mddev->new_chunk_sectors);
2058 		if (mddev->delta_disks == 0 &&
2059 		    mddev->reshape_backwards)
2060 			sb->feature_map
2061 				|= cpu_to_le32(MD_FEATURE_RESHAPE_BACKWARDS);
2062 		if (rdev->new_data_offset != rdev->data_offset) {
2063 			sb->feature_map
2064 				|= cpu_to_le32(MD_FEATURE_NEW_OFFSET);
2065 			sb->new_offset = cpu_to_le32((__u32)(rdev->new_data_offset
2066 							     - rdev->data_offset));
2067 		}
2068 	}
2069 
2070 	if (mddev_is_clustered(mddev))
2071 		sb->feature_map |= cpu_to_le32(MD_FEATURE_CLUSTERED);
2072 
2073 	if (rdev->badblocks.count == 0)
2074 		/* Nothing to do for bad blocks*/ ;
2075 	else if (sb->bblog_offset == 0)
2076 		/* Cannot record bad blocks on this device */
2077 		md_error(mddev, rdev);
2078 	else {
2079 		struct badblocks *bb = &rdev->badblocks;
2080 		__le64 *bbp = (__le64 *)page_address(rdev->bb_page);
2081 		u64 *p = bb->page;
2082 		sb->feature_map |= cpu_to_le32(MD_FEATURE_BAD_BLOCKS);
2083 		if (bb->changed) {
2084 			unsigned seq;
2085 
2086 retry:
2087 			seq = read_seqbegin(&bb->lock);
2088 
2089 			memset(bbp, 0xff, PAGE_SIZE);
2090 
2091 			for (i = 0 ; i < bb->count ; i++) {
2092 				u64 internal_bb = p[i];
2093 				u64 store_bb = ((BB_OFFSET(internal_bb) << 10)
2094 						| BB_LEN(internal_bb));
2095 				bbp[i] = cpu_to_le64(store_bb);
2096 			}
2097 			bb->changed = 0;
2098 			if (read_seqretry(&bb->lock, seq))
2099 				goto retry;
2100 
2101 			bb->sector = (rdev->sb_start +
2102 				      (int)le32_to_cpu(sb->bblog_offset));
2103 			bb->size = le16_to_cpu(sb->bblog_size);
2104 		}
2105 	}
2106 
2107 	max_dev = 0;
2108 	rdev_for_each(rdev2, mddev)
2109 		if (rdev2->desc_nr+1 > max_dev)
2110 			max_dev = rdev2->desc_nr+1;
2111 
2112 	if (max_dev > le32_to_cpu(sb->max_dev)) {
2113 		int bmask;
2114 		sb->max_dev = cpu_to_le32(max_dev);
2115 		rdev->sb_size = max_dev * 2 + 256;
2116 		bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
2117 		if (rdev->sb_size & bmask)
2118 			rdev->sb_size = (rdev->sb_size | bmask) + 1;
2119 	} else
2120 		max_dev = le32_to_cpu(sb->max_dev);
2121 
2122 	for (i=0; i<max_dev;i++)
2123 		sb->dev_roles[i] = cpu_to_le16(MD_DISK_ROLE_SPARE);
2124 
2125 	if (test_bit(MD_HAS_JOURNAL, &mddev->flags))
2126 		sb->feature_map |= cpu_to_le32(MD_FEATURE_JOURNAL);
2127 
2128 	if (test_bit(MD_HAS_PPL, &mddev->flags)) {
2129 		if (test_bit(MD_HAS_MULTIPLE_PPLS, &mddev->flags))
2130 			sb->feature_map |=
2131 			    cpu_to_le32(MD_FEATURE_MULTIPLE_PPLS);
2132 		else
2133 			sb->feature_map |= cpu_to_le32(MD_FEATURE_PPL);
2134 		sb->ppl.offset = cpu_to_le16(rdev->ppl.offset);
2135 		sb->ppl.size = cpu_to_le16(rdev->ppl.size);
2136 	}
2137 
2138 	rdev_for_each(rdev2, mddev) {
2139 		i = rdev2->desc_nr;
2140 		if (test_bit(Faulty, &rdev2->flags))
2141 			sb->dev_roles[i] = cpu_to_le16(MD_DISK_ROLE_FAULTY);
2142 		else if (test_bit(In_sync, &rdev2->flags))
2143 			sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
2144 		else if (test_bit(Journal, &rdev2->flags))
2145 			sb->dev_roles[i] = cpu_to_le16(MD_DISK_ROLE_JOURNAL);
2146 		else if (rdev2->raid_disk >= 0)
2147 			sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
2148 		else
2149 			sb->dev_roles[i] = cpu_to_le16(MD_DISK_ROLE_SPARE);
2150 	}
2151 
2152 	sb->sb_csum = calc_sb_1_csum(sb);
2153 }
2154 
2155 static sector_t super_1_choose_bm_space(sector_t dev_size)
2156 {
2157 	sector_t bm_space;
2158 
2159 	/* if the device is bigger than 8Gig, save 64k for bitmap
2160 	 * usage, if bigger than 200Gig, save 128k
2161 	 */
2162 	if (dev_size < 64*2)
2163 		bm_space = 0;
2164 	else if (dev_size - 64*2 >= 200*1024*1024*2)
2165 		bm_space = 128*2;
2166 	else if (dev_size - 4*2 > 8*1024*1024*2)
2167 		bm_space = 64*2;
2168 	else
2169 		bm_space = 4*2;
2170 	return bm_space;
2171 }
2172 
2173 static unsigned long long
2174 super_1_rdev_size_change(struct md_rdev *rdev, sector_t num_sectors)
2175 {
2176 	struct mdp_superblock_1 *sb;
2177 	sector_t max_sectors;
2178 	if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
2179 		return 0; /* component must fit device */
2180 	if (rdev->data_offset != rdev->new_data_offset)
2181 		return 0; /* too confusing */
2182 	if (rdev->sb_start < rdev->data_offset) {
2183 		/* minor versions 1 and 2; superblock before data */
2184 		max_sectors = bdev_nr_sectors(rdev->bdev) - rdev->data_offset;
2185 		if (!num_sectors || num_sectors > max_sectors)
2186 			num_sectors = max_sectors;
2187 	} else if (rdev->mddev->bitmap_info.offset) {
2188 		/* minor version 0 with bitmap we can't move */
2189 		return 0;
2190 	} else {
2191 		/* minor version 0; superblock after data */
2192 		sector_t sb_start, bm_space;
2193 		sector_t dev_size = bdev_nr_sectors(rdev->bdev);
2194 
2195 		/* 8K is for superblock */
2196 		sb_start = dev_size - 8*2;
2197 		sb_start &= ~(sector_t)(4*2 - 1);
2198 
2199 		bm_space = super_1_choose_bm_space(dev_size);
2200 
2201 		/* Space that can be used to store date needs to decrease
2202 		 * superblock bitmap space and bad block space(4K)
2203 		 */
2204 		max_sectors = sb_start - bm_space - 4*2;
2205 
2206 		if (!num_sectors || num_sectors > max_sectors)
2207 			num_sectors = max_sectors;
2208 		rdev->sb_start = sb_start;
2209 	}
2210 	sb = page_address(rdev->sb_page);
2211 	sb->data_size = cpu_to_le64(num_sectors);
2212 	sb->super_offset = cpu_to_le64(rdev->sb_start);
2213 	sb->sb_csum = calc_sb_1_csum(sb);
2214 	do {
2215 		md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
2216 			       rdev->sb_page);
2217 	} while (md_super_wait(rdev->mddev) < 0);
2218 	return num_sectors;
2219 
2220 }
2221 
2222 static int
2223 super_1_allow_new_offset(struct md_rdev *rdev,
2224 			 unsigned long long new_offset)
2225 {
2226 	/* All necessary checks on new >= old have been done */
2227 	struct bitmap *bitmap;
2228 	if (new_offset >= rdev->data_offset)
2229 		return 1;
2230 
2231 	/* with 1.0 metadata, there is no metadata to tread on
2232 	 * so we can always move back */
2233 	if (rdev->mddev->minor_version == 0)
2234 		return 1;
2235 
2236 	/* otherwise we must be sure not to step on
2237 	 * any metadata, so stay:
2238 	 * 36K beyond start of superblock
2239 	 * beyond end of badblocks
2240 	 * beyond write-intent bitmap
2241 	 */
2242 	if (rdev->sb_start + (32+4)*2 > new_offset)
2243 		return 0;
2244 	bitmap = rdev->mddev->bitmap;
2245 	if (bitmap && !rdev->mddev->bitmap_info.file &&
2246 	    rdev->sb_start + rdev->mddev->bitmap_info.offset +
2247 	    bitmap->storage.file_pages * (PAGE_SIZE>>9) > new_offset)
2248 		return 0;
2249 	if (rdev->badblocks.sector + rdev->badblocks.size > new_offset)
2250 		return 0;
2251 
2252 	return 1;
2253 }
2254 
2255 static struct super_type super_types[] = {
2256 	[0] = {
2257 		.name	= "0.90.0",
2258 		.owner	= THIS_MODULE,
2259 		.load_super	    = super_90_load,
2260 		.validate_super	    = super_90_validate,
2261 		.sync_super	    = super_90_sync,
2262 		.rdev_size_change   = super_90_rdev_size_change,
2263 		.allow_new_offset   = super_90_allow_new_offset,
2264 	},
2265 	[1] = {
2266 		.name	= "md-1",
2267 		.owner	= THIS_MODULE,
2268 		.load_super	    = super_1_load,
2269 		.validate_super	    = super_1_validate,
2270 		.sync_super	    = super_1_sync,
2271 		.rdev_size_change   = super_1_rdev_size_change,
2272 		.allow_new_offset   = super_1_allow_new_offset,
2273 	},
2274 };
2275 
2276 static void sync_super(struct mddev *mddev, struct md_rdev *rdev)
2277 {
2278 	if (mddev->sync_super) {
2279 		mddev->sync_super(mddev, rdev);
2280 		return;
2281 	}
2282 
2283 	BUG_ON(mddev->major_version >= ARRAY_SIZE(super_types));
2284 
2285 	super_types[mddev->major_version].sync_super(mddev, rdev);
2286 }
2287 
2288 static int match_mddev_units(struct mddev *mddev1, struct mddev *mddev2)
2289 {
2290 	struct md_rdev *rdev, *rdev2;
2291 
2292 	rcu_read_lock();
2293 	rdev_for_each_rcu(rdev, mddev1) {
2294 		if (test_bit(Faulty, &rdev->flags) ||
2295 		    test_bit(Journal, &rdev->flags) ||
2296 		    rdev->raid_disk == -1)
2297 			continue;
2298 		rdev_for_each_rcu(rdev2, mddev2) {
2299 			if (test_bit(Faulty, &rdev2->flags) ||
2300 			    test_bit(Journal, &rdev2->flags) ||
2301 			    rdev2->raid_disk == -1)
2302 				continue;
2303 			if (rdev->bdev->bd_disk == rdev2->bdev->bd_disk) {
2304 				rcu_read_unlock();
2305 				return 1;
2306 			}
2307 		}
2308 	}
2309 	rcu_read_unlock();
2310 	return 0;
2311 }
2312 
2313 static LIST_HEAD(pending_raid_disks);
2314 
2315 /*
2316  * Try to register data integrity profile for an mddev
2317  *
2318  * This is called when an array is started and after a disk has been kicked
2319  * from the array. It only succeeds if all working and active component devices
2320  * are integrity capable with matching profiles.
2321  */
2322 int md_integrity_register(struct mddev *mddev)
2323 {
2324 	struct md_rdev *rdev, *reference = NULL;
2325 
2326 	if (list_empty(&mddev->disks))
2327 		return 0; /* nothing to do */
2328 	if (!mddev->gendisk || blk_get_integrity(mddev->gendisk))
2329 		return 0; /* shouldn't register, or already is */
2330 	rdev_for_each(rdev, mddev) {
2331 		/* skip spares and non-functional disks */
2332 		if (test_bit(Faulty, &rdev->flags))
2333 			continue;
2334 		if (rdev->raid_disk < 0)
2335 			continue;
2336 		if (!reference) {
2337 			/* Use the first rdev as the reference */
2338 			reference = rdev;
2339 			continue;
2340 		}
2341 		/* does this rdev's profile match the reference profile? */
2342 		if (blk_integrity_compare(reference->bdev->bd_disk,
2343 				rdev->bdev->bd_disk) < 0)
2344 			return -EINVAL;
2345 	}
2346 	if (!reference || !bdev_get_integrity(reference->bdev))
2347 		return 0;
2348 	/*
2349 	 * All component devices are integrity capable and have matching
2350 	 * profiles, register the common profile for the md device.
2351 	 */
2352 	blk_integrity_register(mddev->gendisk,
2353 			       bdev_get_integrity(reference->bdev));
2354 
2355 	pr_debug("md: data integrity enabled on %s\n", mdname(mddev));
2356 	if (bioset_integrity_create(&mddev->bio_set, BIO_POOL_SIZE) ||
2357 	    (mddev->level != 1 && mddev->level != 10 &&
2358 	     bioset_integrity_create(&mddev->io_clone_set, BIO_POOL_SIZE))) {
2359 		/*
2360 		 * No need to handle the failure of bioset_integrity_create,
2361 		 * because the function is called by md_run() -> pers->run(),
2362 		 * md_run calls bioset_exit -> bioset_integrity_free in case
2363 		 * of failure case.
2364 		 */
2365 		pr_err("md: failed to create integrity pool for %s\n",
2366 		       mdname(mddev));
2367 		return -EINVAL;
2368 	}
2369 	return 0;
2370 }
2371 EXPORT_SYMBOL(md_integrity_register);
2372 
2373 /*
2374  * Attempt to add an rdev, but only if it is consistent with the current
2375  * integrity profile
2376  */
2377 int md_integrity_add_rdev(struct md_rdev *rdev, struct mddev *mddev)
2378 {
2379 	struct blk_integrity *bi_mddev;
2380 
2381 	if (!mddev->gendisk)
2382 		return 0;
2383 
2384 	bi_mddev = blk_get_integrity(mddev->gendisk);
2385 
2386 	if (!bi_mddev) /* nothing to do */
2387 		return 0;
2388 
2389 	if (blk_integrity_compare(mddev->gendisk, rdev->bdev->bd_disk) != 0) {
2390 		pr_err("%s: incompatible integrity profile for %pg\n",
2391 		       mdname(mddev), rdev->bdev);
2392 		return -ENXIO;
2393 	}
2394 
2395 	return 0;
2396 }
2397 EXPORT_SYMBOL(md_integrity_add_rdev);
2398 
2399 static bool rdev_read_only(struct md_rdev *rdev)
2400 {
2401 	return bdev_read_only(rdev->bdev) ||
2402 		(rdev->meta_bdev && bdev_read_only(rdev->meta_bdev));
2403 }
2404 
2405 static int bind_rdev_to_array(struct md_rdev *rdev, struct mddev *mddev)
2406 {
2407 	char b[BDEVNAME_SIZE];
2408 	int err;
2409 
2410 	/* prevent duplicates */
2411 	if (find_rdev(mddev, rdev->bdev->bd_dev))
2412 		return -EEXIST;
2413 
2414 	if (rdev_read_only(rdev) && mddev->pers)
2415 		return -EROFS;
2416 
2417 	/* make sure rdev->sectors exceeds mddev->dev_sectors */
2418 	if (!test_bit(Journal, &rdev->flags) &&
2419 	    rdev->sectors &&
2420 	    (mddev->dev_sectors == 0 || rdev->sectors < mddev->dev_sectors)) {
2421 		if (mddev->pers) {
2422 			/* Cannot change size, so fail
2423 			 * If mddev->level <= 0, then we don't care
2424 			 * about aligning sizes (e.g. linear)
2425 			 */
2426 			if (mddev->level > 0)
2427 				return -ENOSPC;
2428 		} else
2429 			mddev->dev_sectors = rdev->sectors;
2430 	}
2431 
2432 	/* Verify rdev->desc_nr is unique.
2433 	 * If it is -1, assign a free number, else
2434 	 * check number is not in use
2435 	 */
2436 	rcu_read_lock();
2437 	if (rdev->desc_nr < 0) {
2438 		int choice = 0;
2439 		if (mddev->pers)
2440 			choice = mddev->raid_disks;
2441 		while (md_find_rdev_nr_rcu(mddev, choice))
2442 			choice++;
2443 		rdev->desc_nr = choice;
2444 	} else {
2445 		if (md_find_rdev_nr_rcu(mddev, rdev->desc_nr)) {
2446 			rcu_read_unlock();
2447 			return -EBUSY;
2448 		}
2449 	}
2450 	rcu_read_unlock();
2451 	if (!test_bit(Journal, &rdev->flags) &&
2452 	    mddev->max_disks && rdev->desc_nr >= mddev->max_disks) {
2453 		pr_warn("md: %s: array is limited to %d devices\n",
2454 			mdname(mddev), mddev->max_disks);
2455 		return -EBUSY;
2456 	}
2457 	snprintf(b, sizeof(b), "%pg", rdev->bdev);
2458 	strreplace(b, '/', '!');
2459 
2460 	rdev->mddev = mddev;
2461 	pr_debug("md: bind<%s>\n", b);
2462 
2463 	if (mddev->raid_disks)
2464 		mddev_create_serial_pool(mddev, rdev, false);
2465 
2466 	if ((err = kobject_add(&rdev->kobj, &mddev->kobj, "dev-%s", b)))
2467 		goto fail;
2468 
2469 	/* failure here is OK */
2470 	err = sysfs_create_link(&rdev->kobj, bdev_kobj(rdev->bdev), "block");
2471 	rdev->sysfs_state = sysfs_get_dirent_safe(rdev->kobj.sd, "state");
2472 	rdev->sysfs_unack_badblocks =
2473 		sysfs_get_dirent_safe(rdev->kobj.sd, "unacknowledged_bad_blocks");
2474 	rdev->sysfs_badblocks =
2475 		sysfs_get_dirent_safe(rdev->kobj.sd, "bad_blocks");
2476 
2477 	list_add_rcu(&rdev->same_set, &mddev->disks);
2478 	bd_link_disk_holder(rdev->bdev, mddev->gendisk);
2479 
2480 	/* May as well allow recovery to be retried once */
2481 	mddev->recovery_disabled++;
2482 
2483 	return 0;
2484 
2485  fail:
2486 	pr_warn("md: failed to register dev-%s for %s\n",
2487 		b, mdname(mddev));
2488 	mddev_destroy_serial_pool(mddev, rdev, false);
2489 	return err;
2490 }
2491 
2492 void md_autodetect_dev(dev_t dev);
2493 
2494 /* just for claiming the bdev */
2495 static struct md_rdev claim_rdev;
2496 
2497 static void export_rdev(struct md_rdev *rdev, struct mddev *mddev)
2498 {
2499 	pr_debug("md: export_rdev(%pg)\n", rdev->bdev);
2500 	md_rdev_clear(rdev);
2501 #ifndef MODULE
2502 	if (test_bit(AutoDetected, &rdev->flags))
2503 		md_autodetect_dev(rdev->bdev->bd_dev);
2504 #endif
2505 	blkdev_put(rdev->bdev,
2506 		   test_bit(Holder, &rdev->flags) ? rdev : &claim_rdev);
2507 	rdev->bdev = NULL;
2508 	kobject_put(&rdev->kobj);
2509 }
2510 
2511 static void md_kick_rdev_from_array(struct md_rdev *rdev)
2512 {
2513 	struct mddev *mddev = rdev->mddev;
2514 
2515 	bd_unlink_disk_holder(rdev->bdev, rdev->mddev->gendisk);
2516 	list_del_rcu(&rdev->same_set);
2517 	pr_debug("md: unbind<%pg>\n", rdev->bdev);
2518 	mddev_destroy_serial_pool(rdev->mddev, rdev, false);
2519 	rdev->mddev = NULL;
2520 	sysfs_remove_link(&rdev->kobj, "block");
2521 	sysfs_put(rdev->sysfs_state);
2522 	sysfs_put(rdev->sysfs_unack_badblocks);
2523 	sysfs_put(rdev->sysfs_badblocks);
2524 	rdev->sysfs_state = NULL;
2525 	rdev->sysfs_unack_badblocks = NULL;
2526 	rdev->sysfs_badblocks = NULL;
2527 	rdev->badblocks.count = 0;
2528 
2529 	synchronize_rcu();
2530 
2531 	/*
2532 	 * kobject_del() will wait for all in progress writers to be done, where
2533 	 * reconfig_mutex is held, hence it can't be called under
2534 	 * reconfig_mutex and it's delayed to mddev_unlock().
2535 	 */
2536 	list_add(&rdev->same_set, &mddev->deleting);
2537 }
2538 
2539 static void export_array(struct mddev *mddev)
2540 {
2541 	struct md_rdev *rdev;
2542 
2543 	while (!list_empty(&mddev->disks)) {
2544 		rdev = list_first_entry(&mddev->disks, struct md_rdev,
2545 					same_set);
2546 		md_kick_rdev_from_array(rdev);
2547 	}
2548 	mddev->raid_disks = 0;
2549 	mddev->major_version = 0;
2550 }
2551 
2552 static bool set_in_sync(struct mddev *mddev)
2553 {
2554 	lockdep_assert_held(&mddev->lock);
2555 	if (!mddev->in_sync) {
2556 		mddev->sync_checkers++;
2557 		spin_unlock(&mddev->lock);
2558 		percpu_ref_switch_to_atomic_sync(&mddev->writes_pending);
2559 		spin_lock(&mddev->lock);
2560 		if (!mddev->in_sync &&
2561 		    percpu_ref_is_zero(&mddev->writes_pending)) {
2562 			mddev->in_sync = 1;
2563 			/*
2564 			 * Ensure ->in_sync is visible before we clear
2565 			 * ->sync_checkers.
2566 			 */
2567 			smp_mb();
2568 			set_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags);
2569 			sysfs_notify_dirent_safe(mddev->sysfs_state);
2570 		}
2571 		if (--mddev->sync_checkers == 0)
2572 			percpu_ref_switch_to_percpu(&mddev->writes_pending);
2573 	}
2574 	if (mddev->safemode == 1)
2575 		mddev->safemode = 0;
2576 	return mddev->in_sync;
2577 }
2578 
2579 static void sync_sbs(struct mddev *mddev, int nospares)
2580 {
2581 	/* Update each superblock (in-memory image), but
2582 	 * if we are allowed to, skip spares which already
2583 	 * have the right event counter, or have one earlier
2584 	 * (which would mean they aren't being marked as dirty
2585 	 * with the rest of the array)
2586 	 */
2587 	struct md_rdev *rdev;
2588 	rdev_for_each(rdev, mddev) {
2589 		if (rdev->sb_events == mddev->events ||
2590 		    (nospares &&
2591 		     rdev->raid_disk < 0 &&
2592 		     rdev->sb_events+1 == mddev->events)) {
2593 			/* Don't update this superblock */
2594 			rdev->sb_loaded = 2;
2595 		} else {
2596 			sync_super(mddev, rdev);
2597 			rdev->sb_loaded = 1;
2598 		}
2599 	}
2600 }
2601 
2602 static bool does_sb_need_changing(struct mddev *mddev)
2603 {
2604 	struct md_rdev *rdev = NULL, *iter;
2605 	struct mdp_superblock_1 *sb;
2606 	int role;
2607 
2608 	/* Find a good rdev */
2609 	rdev_for_each(iter, mddev)
2610 		if ((iter->raid_disk >= 0) && !test_bit(Faulty, &iter->flags)) {
2611 			rdev = iter;
2612 			break;
2613 		}
2614 
2615 	/* No good device found. */
2616 	if (!rdev)
2617 		return false;
2618 
2619 	sb = page_address(rdev->sb_page);
2620 	/* Check if a device has become faulty or a spare become active */
2621 	rdev_for_each(rdev, mddev) {
2622 		role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
2623 		/* Device activated? */
2624 		if (role == MD_DISK_ROLE_SPARE && rdev->raid_disk >= 0 &&
2625 		    !test_bit(Faulty, &rdev->flags))
2626 			return true;
2627 		/* Device turned faulty? */
2628 		if (test_bit(Faulty, &rdev->flags) && (role < MD_DISK_ROLE_MAX))
2629 			return true;
2630 	}
2631 
2632 	/* Check if any mddev parameters have changed */
2633 	if ((mddev->dev_sectors != le64_to_cpu(sb->size)) ||
2634 	    (mddev->reshape_position != le64_to_cpu(sb->reshape_position)) ||
2635 	    (mddev->layout != le32_to_cpu(sb->layout)) ||
2636 	    (mddev->raid_disks != le32_to_cpu(sb->raid_disks)) ||
2637 	    (mddev->chunk_sectors != le32_to_cpu(sb->chunksize)))
2638 		return true;
2639 
2640 	return false;
2641 }
2642 
2643 void md_update_sb(struct mddev *mddev, int force_change)
2644 {
2645 	struct md_rdev *rdev;
2646 	int sync_req;
2647 	int nospares = 0;
2648 	int any_badblocks_changed = 0;
2649 	int ret = -1;
2650 
2651 	if (!md_is_rdwr(mddev)) {
2652 		if (force_change)
2653 			set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
2654 		return;
2655 	}
2656 
2657 repeat:
2658 	if (mddev_is_clustered(mddev)) {
2659 		if (test_and_clear_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags))
2660 			force_change = 1;
2661 		if (test_and_clear_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags))
2662 			nospares = 1;
2663 		ret = md_cluster_ops->metadata_update_start(mddev);
2664 		/* Has someone else has updated the sb */
2665 		if (!does_sb_need_changing(mddev)) {
2666 			if (ret == 0)
2667 				md_cluster_ops->metadata_update_cancel(mddev);
2668 			bit_clear_unless(&mddev->sb_flags, BIT(MD_SB_CHANGE_PENDING),
2669 							 BIT(MD_SB_CHANGE_DEVS) |
2670 							 BIT(MD_SB_CHANGE_CLEAN));
2671 			return;
2672 		}
2673 	}
2674 
2675 	/*
2676 	 * First make sure individual recovery_offsets are correct
2677 	 * curr_resync_completed can only be used during recovery.
2678 	 * During reshape/resync it might use array-addresses rather
2679 	 * that device addresses.
2680 	 */
2681 	rdev_for_each(rdev, mddev) {
2682 		if (rdev->raid_disk >= 0 &&
2683 		    mddev->delta_disks >= 0 &&
2684 		    test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
2685 		    test_bit(MD_RECOVERY_RECOVER, &mddev->recovery) &&
2686 		    !test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
2687 		    !test_bit(Journal, &rdev->flags) &&
2688 		    !test_bit(In_sync, &rdev->flags) &&
2689 		    mddev->curr_resync_completed > rdev->recovery_offset)
2690 				rdev->recovery_offset = mddev->curr_resync_completed;
2691 
2692 	}
2693 	if (!mddev->persistent) {
2694 		clear_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags);
2695 		clear_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
2696 		if (!mddev->external) {
2697 			clear_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags);
2698 			rdev_for_each(rdev, mddev) {
2699 				if (rdev->badblocks.changed) {
2700 					rdev->badblocks.changed = 0;
2701 					ack_all_badblocks(&rdev->badblocks);
2702 					md_error(mddev, rdev);
2703 				}
2704 				clear_bit(Blocked, &rdev->flags);
2705 				clear_bit(BlockedBadBlocks, &rdev->flags);
2706 				wake_up(&rdev->blocked_wait);
2707 			}
2708 		}
2709 		wake_up(&mddev->sb_wait);
2710 		return;
2711 	}
2712 
2713 	spin_lock(&mddev->lock);
2714 
2715 	mddev->utime = ktime_get_real_seconds();
2716 
2717 	if (test_and_clear_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags))
2718 		force_change = 1;
2719 	if (test_and_clear_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags))
2720 		/* just a clean<-> dirty transition, possibly leave spares alone,
2721 		 * though if events isn't the right even/odd, we will have to do
2722 		 * spares after all
2723 		 */
2724 		nospares = 1;
2725 	if (force_change)
2726 		nospares = 0;
2727 	if (mddev->degraded)
2728 		/* If the array is degraded, then skipping spares is both
2729 		 * dangerous and fairly pointless.
2730 		 * Dangerous because a device that was removed from the array
2731 		 * might have a event_count that still looks up-to-date,
2732 		 * so it can be re-added without a resync.
2733 		 * Pointless because if there are any spares to skip,
2734 		 * then a recovery will happen and soon that array won't
2735 		 * be degraded any more and the spare can go back to sleep then.
2736 		 */
2737 		nospares = 0;
2738 
2739 	sync_req = mddev->in_sync;
2740 
2741 	/* If this is just a dirty<->clean transition, and the array is clean
2742 	 * and 'events' is odd, we can roll back to the previous clean state */
2743 	if (nospares
2744 	    && (mddev->in_sync && mddev->recovery_cp == MaxSector)
2745 	    && mddev->can_decrease_events
2746 	    && mddev->events != 1) {
2747 		mddev->events--;
2748 		mddev->can_decrease_events = 0;
2749 	} else {
2750 		/* otherwise we have to go forward and ... */
2751 		mddev->events ++;
2752 		mddev->can_decrease_events = nospares;
2753 	}
2754 
2755 	/*
2756 	 * This 64-bit counter should never wrap.
2757 	 * Either we are in around ~1 trillion A.C., assuming
2758 	 * 1 reboot per second, or we have a bug...
2759 	 */
2760 	WARN_ON(mddev->events == 0);
2761 
2762 	rdev_for_each(rdev, mddev) {
2763 		if (rdev->badblocks.changed)
2764 			any_badblocks_changed++;
2765 		if (test_bit(Faulty, &rdev->flags))
2766 			set_bit(FaultRecorded, &rdev->flags);
2767 	}
2768 
2769 	sync_sbs(mddev, nospares);
2770 	spin_unlock(&mddev->lock);
2771 
2772 	pr_debug("md: updating %s RAID superblock on device (in sync %d)\n",
2773 		 mdname(mddev), mddev->in_sync);
2774 
2775 	if (mddev->queue)
2776 		blk_add_trace_msg(mddev->queue, "md md_update_sb");
2777 rewrite:
2778 	md_bitmap_update_sb(mddev->bitmap);
2779 	rdev_for_each(rdev, mddev) {
2780 		if (rdev->sb_loaded != 1)
2781 			continue; /* no noise on spare devices */
2782 
2783 		if (!test_bit(Faulty, &rdev->flags)) {
2784 			md_super_write(mddev,rdev,
2785 				       rdev->sb_start, rdev->sb_size,
2786 				       rdev->sb_page);
2787 			pr_debug("md: (write) %pg's sb offset: %llu\n",
2788 				 rdev->bdev,
2789 				 (unsigned long long)rdev->sb_start);
2790 			rdev->sb_events = mddev->events;
2791 			if (rdev->badblocks.size) {
2792 				md_super_write(mddev, rdev,
2793 					       rdev->badblocks.sector,
2794 					       rdev->badblocks.size << 9,
2795 					       rdev->bb_page);
2796 				rdev->badblocks.size = 0;
2797 			}
2798 
2799 		} else
2800 			pr_debug("md: %pg (skipping faulty)\n",
2801 				 rdev->bdev);
2802 
2803 		if (mddev->level == LEVEL_MULTIPATH)
2804 			/* only need to write one superblock... */
2805 			break;
2806 	}
2807 	if (md_super_wait(mddev) < 0)
2808 		goto rewrite;
2809 	/* if there was a failure, MD_SB_CHANGE_DEVS was set, and we re-write super */
2810 
2811 	if (mddev_is_clustered(mddev) && ret == 0)
2812 		md_cluster_ops->metadata_update_finish(mddev);
2813 
2814 	if (mddev->in_sync != sync_req ||
2815 	    !bit_clear_unless(&mddev->sb_flags, BIT(MD_SB_CHANGE_PENDING),
2816 			       BIT(MD_SB_CHANGE_DEVS) | BIT(MD_SB_CHANGE_CLEAN)))
2817 		/* have to write it out again */
2818 		goto repeat;
2819 	wake_up(&mddev->sb_wait);
2820 	if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
2821 		sysfs_notify_dirent_safe(mddev->sysfs_completed);
2822 
2823 	rdev_for_each(rdev, mddev) {
2824 		if (test_and_clear_bit(FaultRecorded, &rdev->flags))
2825 			clear_bit(Blocked, &rdev->flags);
2826 
2827 		if (any_badblocks_changed)
2828 			ack_all_badblocks(&rdev->badblocks);
2829 		clear_bit(BlockedBadBlocks, &rdev->flags);
2830 		wake_up(&rdev->blocked_wait);
2831 	}
2832 }
2833 EXPORT_SYMBOL(md_update_sb);
2834 
2835 static int add_bound_rdev(struct md_rdev *rdev)
2836 {
2837 	struct mddev *mddev = rdev->mddev;
2838 	int err = 0;
2839 	bool add_journal = test_bit(Journal, &rdev->flags);
2840 
2841 	if (!mddev->pers->hot_remove_disk || add_journal) {
2842 		/* If there is hot_add_disk but no hot_remove_disk
2843 		 * then added disks for geometry changes,
2844 		 * and should be added immediately.
2845 		 */
2846 		super_types[mddev->major_version].
2847 			validate_super(mddev, NULL/*freshest*/, rdev);
2848 		if (add_journal)
2849 			mddev_suspend(mddev);
2850 		err = mddev->pers->hot_add_disk(mddev, rdev);
2851 		if (add_journal)
2852 			mddev_resume(mddev);
2853 		if (err) {
2854 			md_kick_rdev_from_array(rdev);
2855 			return err;
2856 		}
2857 	}
2858 	sysfs_notify_dirent_safe(rdev->sysfs_state);
2859 
2860 	set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
2861 	if (mddev->degraded)
2862 		set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
2863 	set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2864 	md_new_event();
2865 	md_wakeup_thread(mddev->thread);
2866 	return 0;
2867 }
2868 
2869 /* words written to sysfs files may, or may not, be \n terminated.
2870  * We want to accept with case. For this we use cmd_match.
2871  */
2872 static int cmd_match(const char *cmd, const char *str)
2873 {
2874 	/* See if cmd, written into a sysfs file, matches
2875 	 * str.  They must either be the same, or cmd can
2876 	 * have a trailing newline
2877 	 */
2878 	while (*cmd && *str && *cmd == *str) {
2879 		cmd++;
2880 		str++;
2881 	}
2882 	if (*cmd == '\n')
2883 		cmd++;
2884 	if (*str || *cmd)
2885 		return 0;
2886 	return 1;
2887 }
2888 
2889 struct rdev_sysfs_entry {
2890 	struct attribute attr;
2891 	ssize_t (*show)(struct md_rdev *, char *);
2892 	ssize_t (*store)(struct md_rdev *, const char *, size_t);
2893 };
2894 
2895 static ssize_t
2896 state_show(struct md_rdev *rdev, char *page)
2897 {
2898 	char *sep = ",";
2899 	size_t len = 0;
2900 	unsigned long flags = READ_ONCE(rdev->flags);
2901 
2902 	if (test_bit(Faulty, &flags) ||
2903 	    (!test_bit(ExternalBbl, &flags) &&
2904 	    rdev->badblocks.unacked_exist))
2905 		len += sprintf(page+len, "faulty%s", sep);
2906 	if (test_bit(In_sync, &flags))
2907 		len += sprintf(page+len, "in_sync%s", sep);
2908 	if (test_bit(Journal, &flags))
2909 		len += sprintf(page+len, "journal%s", sep);
2910 	if (test_bit(WriteMostly, &flags))
2911 		len += sprintf(page+len, "write_mostly%s", sep);
2912 	if (test_bit(Blocked, &flags) ||
2913 	    (rdev->badblocks.unacked_exist
2914 	     && !test_bit(Faulty, &flags)))
2915 		len += sprintf(page+len, "blocked%s", sep);
2916 	if (!test_bit(Faulty, &flags) &&
2917 	    !test_bit(Journal, &flags) &&
2918 	    !test_bit(In_sync, &flags))
2919 		len += sprintf(page+len, "spare%s", sep);
2920 	if (test_bit(WriteErrorSeen, &flags))
2921 		len += sprintf(page+len, "write_error%s", sep);
2922 	if (test_bit(WantReplacement, &flags))
2923 		len += sprintf(page+len, "want_replacement%s", sep);
2924 	if (test_bit(Replacement, &flags))
2925 		len += sprintf(page+len, "replacement%s", sep);
2926 	if (test_bit(ExternalBbl, &flags))
2927 		len += sprintf(page+len, "external_bbl%s", sep);
2928 	if (test_bit(FailFast, &flags))
2929 		len += sprintf(page+len, "failfast%s", sep);
2930 
2931 	if (len)
2932 		len -= strlen(sep);
2933 
2934 	return len+sprintf(page+len, "\n");
2935 }
2936 
2937 static ssize_t
2938 state_store(struct md_rdev *rdev, const char *buf, size_t len)
2939 {
2940 	/* can write
2941 	 *  faulty  - simulates an error
2942 	 *  remove  - disconnects the device
2943 	 *  writemostly - sets write_mostly
2944 	 *  -writemostly - clears write_mostly
2945 	 *  blocked - sets the Blocked flags
2946 	 *  -blocked - clears the Blocked and possibly simulates an error
2947 	 *  insync - sets Insync providing device isn't active
2948 	 *  -insync - clear Insync for a device with a slot assigned,
2949 	 *            so that it gets rebuilt based on bitmap
2950 	 *  write_error - sets WriteErrorSeen
2951 	 *  -write_error - clears WriteErrorSeen
2952 	 *  {,-}failfast - set/clear FailFast
2953 	 */
2954 
2955 	struct mddev *mddev = rdev->mddev;
2956 	int err = -EINVAL;
2957 	bool need_update_sb = false;
2958 
2959 	if (cmd_match(buf, "faulty") && rdev->mddev->pers) {
2960 		md_error(rdev->mddev, rdev);
2961 
2962 		if (test_bit(MD_BROKEN, &rdev->mddev->flags))
2963 			err = -EBUSY;
2964 		else
2965 			err = 0;
2966 	} else if (cmd_match(buf, "remove")) {
2967 		if (rdev->mddev->pers) {
2968 			clear_bit(Blocked, &rdev->flags);
2969 			remove_and_add_spares(rdev->mddev, rdev);
2970 		}
2971 		if (rdev->raid_disk >= 0)
2972 			err = -EBUSY;
2973 		else {
2974 			err = 0;
2975 			if (mddev_is_clustered(mddev))
2976 				err = md_cluster_ops->remove_disk(mddev, rdev);
2977 
2978 			if (err == 0) {
2979 				md_kick_rdev_from_array(rdev);
2980 				if (mddev->pers) {
2981 					set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
2982 					md_wakeup_thread(mddev->thread);
2983 				}
2984 				md_new_event();
2985 			}
2986 		}
2987 	} else if (cmd_match(buf, "writemostly")) {
2988 		set_bit(WriteMostly, &rdev->flags);
2989 		mddev_create_serial_pool(rdev->mddev, rdev, false);
2990 		need_update_sb = true;
2991 		err = 0;
2992 	} else if (cmd_match(buf, "-writemostly")) {
2993 		mddev_destroy_serial_pool(rdev->mddev, rdev, false);
2994 		clear_bit(WriteMostly, &rdev->flags);
2995 		need_update_sb = true;
2996 		err = 0;
2997 	} else if (cmd_match(buf, "blocked")) {
2998 		set_bit(Blocked, &rdev->flags);
2999 		err = 0;
3000 	} else if (cmd_match(buf, "-blocked")) {
3001 		if (!test_bit(Faulty, &rdev->flags) &&
3002 		    !test_bit(ExternalBbl, &rdev->flags) &&
3003 		    rdev->badblocks.unacked_exist) {
3004 			/* metadata handler doesn't understand badblocks,
3005 			 * so we need to fail the device
3006 			 */
3007 			md_error(rdev->mddev, rdev);
3008 		}
3009 		clear_bit(Blocked, &rdev->flags);
3010 		clear_bit(BlockedBadBlocks, &rdev->flags);
3011 		wake_up(&rdev->blocked_wait);
3012 		set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
3013 		md_wakeup_thread(rdev->mddev->thread);
3014 
3015 		err = 0;
3016 	} else if (cmd_match(buf, "insync") && rdev->raid_disk == -1) {
3017 		set_bit(In_sync, &rdev->flags);
3018 		err = 0;
3019 	} else if (cmd_match(buf, "failfast")) {
3020 		set_bit(FailFast, &rdev->flags);
3021 		need_update_sb = true;
3022 		err = 0;
3023 	} else if (cmd_match(buf, "-failfast")) {
3024 		clear_bit(FailFast, &rdev->flags);
3025 		need_update_sb = true;
3026 		err = 0;
3027 	} else if (cmd_match(buf, "-insync") && rdev->raid_disk >= 0 &&
3028 		   !test_bit(Journal, &rdev->flags)) {
3029 		if (rdev->mddev->pers == NULL) {
3030 			clear_bit(In_sync, &rdev->flags);
3031 			rdev->saved_raid_disk = rdev->raid_disk;
3032 			rdev->raid_disk = -1;
3033 			err = 0;
3034 		}
3035 	} else if (cmd_match(buf, "write_error")) {
3036 		set_bit(WriteErrorSeen, &rdev->flags);
3037 		err = 0;
3038 	} else if (cmd_match(buf, "-write_error")) {
3039 		clear_bit(WriteErrorSeen, &rdev->flags);
3040 		err = 0;
3041 	} else if (cmd_match(buf, "want_replacement")) {
3042 		/* Any non-spare device that is not a replacement can
3043 		 * become want_replacement at any time, but we then need to
3044 		 * check if recovery is needed.
3045 		 */
3046 		if (rdev->raid_disk >= 0 &&
3047 		    !test_bit(Journal, &rdev->flags) &&
3048 		    !test_bit(Replacement, &rdev->flags))
3049 			set_bit(WantReplacement, &rdev->flags);
3050 		set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
3051 		md_wakeup_thread(rdev->mddev->thread);
3052 		err = 0;
3053 	} else if (cmd_match(buf, "-want_replacement")) {
3054 		/* Clearing 'want_replacement' is always allowed.
3055 		 * Once replacements starts it is too late though.
3056 		 */
3057 		err = 0;
3058 		clear_bit(WantReplacement, &rdev->flags);
3059 	} else if (cmd_match(buf, "replacement")) {
3060 		/* Can only set a device as a replacement when array has not
3061 		 * yet been started.  Once running, replacement is automatic
3062 		 * from spares, or by assigning 'slot'.
3063 		 */
3064 		if (rdev->mddev->pers)
3065 			err = -EBUSY;
3066 		else {
3067 			set_bit(Replacement, &rdev->flags);
3068 			err = 0;
3069 		}
3070 	} else if (cmd_match(buf, "-replacement")) {
3071 		/* Similarly, can only clear Replacement before start */
3072 		if (rdev->mddev->pers)
3073 			err = -EBUSY;
3074 		else {
3075 			clear_bit(Replacement, &rdev->flags);
3076 			err = 0;
3077 		}
3078 	} else if (cmd_match(buf, "re-add")) {
3079 		if (!rdev->mddev->pers)
3080 			err = -EINVAL;
3081 		else if (test_bit(Faulty, &rdev->flags) && (rdev->raid_disk == -1) &&
3082 				rdev->saved_raid_disk >= 0) {
3083 			/* clear_bit is performed _after_ all the devices
3084 			 * have their local Faulty bit cleared. If any writes
3085 			 * happen in the meantime in the local node, they
3086 			 * will land in the local bitmap, which will be synced
3087 			 * by this node eventually
3088 			 */
3089 			if (!mddev_is_clustered(rdev->mddev) ||
3090 			    (err = md_cluster_ops->gather_bitmaps(rdev)) == 0) {
3091 				clear_bit(Faulty, &rdev->flags);
3092 				err = add_bound_rdev(rdev);
3093 			}
3094 		} else
3095 			err = -EBUSY;
3096 	} else if (cmd_match(buf, "external_bbl") && (rdev->mddev->external)) {
3097 		set_bit(ExternalBbl, &rdev->flags);
3098 		rdev->badblocks.shift = 0;
3099 		err = 0;
3100 	} else if (cmd_match(buf, "-external_bbl") && (rdev->mddev->external)) {
3101 		clear_bit(ExternalBbl, &rdev->flags);
3102 		err = 0;
3103 	}
3104 	if (need_update_sb)
3105 		md_update_sb(mddev, 1);
3106 	if (!err)
3107 		sysfs_notify_dirent_safe(rdev->sysfs_state);
3108 	return err ? err : len;
3109 }
3110 static struct rdev_sysfs_entry rdev_state =
3111 __ATTR_PREALLOC(state, S_IRUGO|S_IWUSR, state_show, state_store);
3112 
3113 static ssize_t
3114 errors_show(struct md_rdev *rdev, char *page)
3115 {
3116 	return sprintf(page, "%d\n", atomic_read(&rdev->corrected_errors));
3117 }
3118 
3119 static ssize_t
3120 errors_store(struct md_rdev *rdev, const char *buf, size_t len)
3121 {
3122 	unsigned int n;
3123 	int rv;
3124 
3125 	rv = kstrtouint(buf, 10, &n);
3126 	if (rv < 0)
3127 		return rv;
3128 	atomic_set(&rdev->corrected_errors, n);
3129 	return len;
3130 }
3131 static struct rdev_sysfs_entry rdev_errors =
3132 __ATTR(errors, S_IRUGO|S_IWUSR, errors_show, errors_store);
3133 
3134 static ssize_t
3135 slot_show(struct md_rdev *rdev, char *page)
3136 {
3137 	if (test_bit(Journal, &rdev->flags))
3138 		return sprintf(page, "journal\n");
3139 	else if (rdev->raid_disk < 0)
3140 		return sprintf(page, "none\n");
3141 	else
3142 		return sprintf(page, "%d\n", rdev->raid_disk);
3143 }
3144 
3145 static ssize_t
3146 slot_store(struct md_rdev *rdev, const char *buf, size_t len)
3147 {
3148 	int slot;
3149 	int err;
3150 
3151 	if (test_bit(Journal, &rdev->flags))
3152 		return -EBUSY;
3153 	if (strncmp(buf, "none", 4)==0)
3154 		slot = -1;
3155 	else {
3156 		err = kstrtouint(buf, 10, (unsigned int *)&slot);
3157 		if (err < 0)
3158 			return err;
3159 		if (slot < 0)
3160 			/* overflow */
3161 			return -ENOSPC;
3162 	}
3163 	if (rdev->mddev->pers && slot == -1) {
3164 		/* Setting 'slot' on an active array requires also
3165 		 * updating the 'rd%d' link, and communicating
3166 		 * with the personality with ->hot_*_disk.
3167 		 * For now we only support removing
3168 		 * failed/spare devices.  This normally happens automatically,
3169 		 * but not when the metadata is externally managed.
3170 		 */
3171 		if (rdev->raid_disk == -1)
3172 			return -EEXIST;
3173 		/* personality does all needed checks */
3174 		if (rdev->mddev->pers->hot_remove_disk == NULL)
3175 			return -EINVAL;
3176 		clear_bit(Blocked, &rdev->flags);
3177 		remove_and_add_spares(rdev->mddev, rdev);
3178 		if (rdev->raid_disk >= 0)
3179 			return -EBUSY;
3180 		set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
3181 		md_wakeup_thread(rdev->mddev->thread);
3182 	} else if (rdev->mddev->pers) {
3183 		/* Activating a spare .. or possibly reactivating
3184 		 * if we ever get bitmaps working here.
3185 		 */
3186 		int err;
3187 
3188 		if (rdev->raid_disk != -1)
3189 			return -EBUSY;
3190 
3191 		if (test_bit(MD_RECOVERY_RUNNING, &rdev->mddev->recovery))
3192 			return -EBUSY;
3193 
3194 		if (rdev->mddev->pers->hot_add_disk == NULL)
3195 			return -EINVAL;
3196 
3197 		if (slot >= rdev->mddev->raid_disks &&
3198 		    slot >= rdev->mddev->raid_disks + rdev->mddev->delta_disks)
3199 			return -ENOSPC;
3200 
3201 		rdev->raid_disk = slot;
3202 		if (test_bit(In_sync, &rdev->flags))
3203 			rdev->saved_raid_disk = slot;
3204 		else
3205 			rdev->saved_raid_disk = -1;
3206 		clear_bit(In_sync, &rdev->flags);
3207 		clear_bit(Bitmap_sync, &rdev->flags);
3208 		err = rdev->mddev->pers->hot_add_disk(rdev->mddev, rdev);
3209 		if (err) {
3210 			rdev->raid_disk = -1;
3211 			return err;
3212 		} else
3213 			sysfs_notify_dirent_safe(rdev->sysfs_state);
3214 		/* failure here is OK */;
3215 		sysfs_link_rdev(rdev->mddev, rdev);
3216 		/* don't wakeup anyone, leave that to userspace. */
3217 	} else {
3218 		if (slot >= rdev->mddev->raid_disks &&
3219 		    slot >= rdev->mddev->raid_disks + rdev->mddev->delta_disks)
3220 			return -ENOSPC;
3221 		rdev->raid_disk = slot;
3222 		/* assume it is working */
3223 		clear_bit(Faulty, &rdev->flags);
3224 		clear_bit(WriteMostly, &rdev->flags);
3225 		set_bit(In_sync, &rdev->flags);
3226 		sysfs_notify_dirent_safe(rdev->sysfs_state);
3227 	}
3228 	return len;
3229 }
3230 
3231 static struct rdev_sysfs_entry rdev_slot =
3232 __ATTR(slot, S_IRUGO|S_IWUSR, slot_show, slot_store);
3233 
3234 static ssize_t
3235 offset_show(struct md_rdev *rdev, char *page)
3236 {
3237 	return sprintf(page, "%llu\n", (unsigned long long)rdev->data_offset);
3238 }
3239 
3240 static ssize_t
3241 offset_store(struct md_rdev *rdev, const char *buf, size_t len)
3242 {
3243 	unsigned long long offset;
3244 	if (kstrtoull(buf, 10, &offset) < 0)
3245 		return -EINVAL;
3246 	if (rdev->mddev->pers && rdev->raid_disk >= 0)
3247 		return -EBUSY;
3248 	if (rdev->sectors && rdev->mddev->external)
3249 		/* Must set offset before size, so overlap checks
3250 		 * can be sane */
3251 		return -EBUSY;
3252 	rdev->data_offset = offset;
3253 	rdev->new_data_offset = offset;
3254 	return len;
3255 }
3256 
3257 static struct rdev_sysfs_entry rdev_offset =
3258 __ATTR(offset, S_IRUGO|S_IWUSR, offset_show, offset_store);
3259 
3260 static ssize_t new_offset_show(struct md_rdev *rdev, char *page)
3261 {
3262 	return sprintf(page, "%llu\n",
3263 		       (unsigned long long)rdev->new_data_offset);
3264 }
3265 
3266 static ssize_t new_offset_store(struct md_rdev *rdev,
3267 				const char *buf, size_t len)
3268 {
3269 	unsigned long long new_offset;
3270 	struct mddev *mddev = rdev->mddev;
3271 
3272 	if (kstrtoull(buf, 10, &new_offset) < 0)
3273 		return -EINVAL;
3274 
3275 	if (mddev->sync_thread ||
3276 	    test_bit(MD_RECOVERY_RUNNING,&mddev->recovery))
3277 		return -EBUSY;
3278 	if (new_offset == rdev->data_offset)
3279 		/* reset is always permitted */
3280 		;
3281 	else if (new_offset > rdev->data_offset) {
3282 		/* must not push array size beyond rdev_sectors */
3283 		if (new_offset - rdev->data_offset
3284 		    + mddev->dev_sectors > rdev->sectors)
3285 				return -E2BIG;
3286 	}
3287 	/* Metadata worries about other space details. */
3288 
3289 	/* decreasing the offset is inconsistent with a backwards
3290 	 * reshape.
3291 	 */
3292 	if (new_offset < rdev->data_offset &&
3293 	    mddev->reshape_backwards)
3294 		return -EINVAL;
3295 	/* Increasing offset is inconsistent with forwards
3296 	 * reshape.  reshape_direction should be set to
3297 	 * 'backwards' first.
3298 	 */
3299 	if (new_offset > rdev->data_offset &&
3300 	    !mddev->reshape_backwards)
3301 		return -EINVAL;
3302 
3303 	if (mddev->pers && mddev->persistent &&
3304 	    !super_types[mddev->major_version]
3305 	    .allow_new_offset(rdev, new_offset))
3306 		return -E2BIG;
3307 	rdev->new_data_offset = new_offset;
3308 	if (new_offset > rdev->data_offset)
3309 		mddev->reshape_backwards = 1;
3310 	else if (new_offset < rdev->data_offset)
3311 		mddev->reshape_backwards = 0;
3312 
3313 	return len;
3314 }
3315 static struct rdev_sysfs_entry rdev_new_offset =
3316 __ATTR(new_offset, S_IRUGO|S_IWUSR, new_offset_show, new_offset_store);
3317 
3318 static ssize_t
3319 rdev_size_show(struct md_rdev *rdev, char *page)
3320 {
3321 	return sprintf(page, "%llu\n", (unsigned long long)rdev->sectors / 2);
3322 }
3323 
3324 static int md_rdevs_overlap(struct md_rdev *a, struct md_rdev *b)
3325 {
3326 	/* check if two start/length pairs overlap */
3327 	if (a->data_offset + a->sectors <= b->data_offset)
3328 		return false;
3329 	if (b->data_offset + b->sectors <= a->data_offset)
3330 		return false;
3331 	return true;
3332 }
3333 
3334 static bool md_rdev_overlaps(struct md_rdev *rdev)
3335 {
3336 	struct mddev *mddev;
3337 	struct md_rdev *rdev2;
3338 
3339 	spin_lock(&all_mddevs_lock);
3340 	list_for_each_entry(mddev, &all_mddevs, all_mddevs) {
3341 		if (test_bit(MD_DELETED, &mddev->flags))
3342 			continue;
3343 		rdev_for_each(rdev2, mddev) {
3344 			if (rdev != rdev2 && rdev->bdev == rdev2->bdev &&
3345 			    md_rdevs_overlap(rdev, rdev2)) {
3346 				spin_unlock(&all_mddevs_lock);
3347 				return true;
3348 			}
3349 		}
3350 	}
3351 	spin_unlock(&all_mddevs_lock);
3352 	return false;
3353 }
3354 
3355 static int strict_blocks_to_sectors(const char *buf, sector_t *sectors)
3356 {
3357 	unsigned long long blocks;
3358 	sector_t new;
3359 
3360 	if (kstrtoull(buf, 10, &blocks) < 0)
3361 		return -EINVAL;
3362 
3363 	if (blocks & 1ULL << (8 * sizeof(blocks) - 1))
3364 		return -EINVAL; /* sector conversion overflow */
3365 
3366 	new = blocks * 2;
3367 	if (new != blocks * 2)
3368 		return -EINVAL; /* unsigned long long to sector_t overflow */
3369 
3370 	*sectors = new;
3371 	return 0;
3372 }
3373 
3374 static ssize_t
3375 rdev_size_store(struct md_rdev *rdev, const char *buf, size_t len)
3376 {
3377 	struct mddev *my_mddev = rdev->mddev;
3378 	sector_t oldsectors = rdev->sectors;
3379 	sector_t sectors;
3380 
3381 	if (test_bit(Journal, &rdev->flags))
3382 		return -EBUSY;
3383 	if (strict_blocks_to_sectors(buf, &sectors) < 0)
3384 		return -EINVAL;
3385 	if (rdev->data_offset != rdev->new_data_offset)
3386 		return -EINVAL; /* too confusing */
3387 	if (my_mddev->pers && rdev->raid_disk >= 0) {
3388 		if (my_mddev->persistent) {
3389 			sectors = super_types[my_mddev->major_version].
3390 				rdev_size_change(rdev, sectors);
3391 			if (!sectors)
3392 				return -EBUSY;
3393 		} else if (!sectors)
3394 			sectors = bdev_nr_sectors(rdev->bdev) -
3395 				rdev->data_offset;
3396 		if (!my_mddev->pers->resize)
3397 			/* Cannot change size for RAID0 or Linear etc */
3398 			return -EINVAL;
3399 	}
3400 	if (sectors < my_mddev->dev_sectors)
3401 		return -EINVAL; /* component must fit device */
3402 
3403 	rdev->sectors = sectors;
3404 
3405 	/*
3406 	 * Check that all other rdevs with the same bdev do not overlap.  This
3407 	 * check does not provide a hard guarantee, it just helps avoid
3408 	 * dangerous mistakes.
3409 	 */
3410 	if (sectors > oldsectors && my_mddev->external &&
3411 	    md_rdev_overlaps(rdev)) {
3412 		/*
3413 		 * Someone else could have slipped in a size change here, but
3414 		 * doing so is just silly.  We put oldsectors back because we
3415 		 * know it is safe, and trust userspace not to race with itself.
3416 		 */
3417 		rdev->sectors = oldsectors;
3418 		return -EBUSY;
3419 	}
3420 	return len;
3421 }
3422 
3423 static struct rdev_sysfs_entry rdev_size =
3424 __ATTR(size, S_IRUGO|S_IWUSR, rdev_size_show, rdev_size_store);
3425 
3426 static ssize_t recovery_start_show(struct md_rdev *rdev, char *page)
3427 {
3428 	unsigned long long recovery_start = rdev->recovery_offset;
3429 
3430 	if (test_bit(In_sync, &rdev->flags) ||
3431 	    recovery_start == MaxSector)
3432 		return sprintf(page, "none\n");
3433 
3434 	return sprintf(page, "%llu\n", recovery_start);
3435 }
3436 
3437 static ssize_t recovery_start_store(struct md_rdev *rdev, const char *buf, size_t len)
3438 {
3439 	unsigned long long recovery_start;
3440 
3441 	if (cmd_match(buf, "none"))
3442 		recovery_start = MaxSector;
3443 	else if (kstrtoull(buf, 10, &recovery_start))
3444 		return -EINVAL;
3445 
3446 	if (rdev->mddev->pers &&
3447 	    rdev->raid_disk >= 0)
3448 		return -EBUSY;
3449 
3450 	rdev->recovery_offset = recovery_start;
3451 	if (recovery_start == MaxSector)
3452 		set_bit(In_sync, &rdev->flags);
3453 	else
3454 		clear_bit(In_sync, &rdev->flags);
3455 	return len;
3456 }
3457 
3458 static struct rdev_sysfs_entry rdev_recovery_start =
3459 __ATTR(recovery_start, S_IRUGO|S_IWUSR, recovery_start_show, recovery_start_store);
3460 
3461 /* sysfs access to bad-blocks list.
3462  * We present two files.
3463  * 'bad-blocks' lists sector numbers and lengths of ranges that
3464  *    are recorded as bad.  The list is truncated to fit within
3465  *    the one-page limit of sysfs.
3466  *    Writing "sector length" to this file adds an acknowledged
3467  *    bad block list.
3468  * 'unacknowledged-bad-blocks' lists bad blocks that have not yet
3469  *    been acknowledged.  Writing to this file adds bad blocks
3470  *    without acknowledging them.  This is largely for testing.
3471  */
3472 static ssize_t bb_show(struct md_rdev *rdev, char *page)
3473 {
3474 	return badblocks_show(&rdev->badblocks, page, 0);
3475 }
3476 static ssize_t bb_store(struct md_rdev *rdev, const char *page, size_t len)
3477 {
3478 	int rv = badblocks_store(&rdev->badblocks, page, len, 0);
3479 	/* Maybe that ack was all we needed */
3480 	if (test_and_clear_bit(BlockedBadBlocks, &rdev->flags))
3481 		wake_up(&rdev->blocked_wait);
3482 	return rv;
3483 }
3484 static struct rdev_sysfs_entry rdev_bad_blocks =
3485 __ATTR(bad_blocks, S_IRUGO|S_IWUSR, bb_show, bb_store);
3486 
3487 static ssize_t ubb_show(struct md_rdev *rdev, char *page)
3488 {
3489 	return badblocks_show(&rdev->badblocks, page, 1);
3490 }
3491 static ssize_t ubb_store(struct md_rdev *rdev, const char *page, size_t len)
3492 {
3493 	return badblocks_store(&rdev->badblocks, page, len, 1);
3494 }
3495 static struct rdev_sysfs_entry rdev_unack_bad_blocks =
3496 __ATTR(unacknowledged_bad_blocks, S_IRUGO|S_IWUSR, ubb_show, ubb_store);
3497 
3498 static ssize_t
3499 ppl_sector_show(struct md_rdev *rdev, char *page)
3500 {
3501 	return sprintf(page, "%llu\n", (unsigned long long)rdev->ppl.sector);
3502 }
3503 
3504 static ssize_t
3505 ppl_sector_store(struct md_rdev *rdev, const char *buf, size_t len)
3506 {
3507 	unsigned long long sector;
3508 
3509 	if (kstrtoull(buf, 10, &sector) < 0)
3510 		return -EINVAL;
3511 	if (sector != (sector_t)sector)
3512 		return -EINVAL;
3513 
3514 	if (rdev->mddev->pers && test_bit(MD_HAS_PPL, &rdev->mddev->flags) &&
3515 	    rdev->raid_disk >= 0)
3516 		return -EBUSY;
3517 
3518 	if (rdev->mddev->persistent) {
3519 		if (rdev->mddev->major_version == 0)
3520 			return -EINVAL;
3521 		if ((sector > rdev->sb_start &&
3522 		     sector - rdev->sb_start > S16_MAX) ||
3523 		    (sector < rdev->sb_start &&
3524 		     rdev->sb_start - sector > -S16_MIN))
3525 			return -EINVAL;
3526 		rdev->ppl.offset = sector - rdev->sb_start;
3527 	} else if (!rdev->mddev->external) {
3528 		return -EBUSY;
3529 	}
3530 	rdev->ppl.sector = sector;
3531 	return len;
3532 }
3533 
3534 static struct rdev_sysfs_entry rdev_ppl_sector =
3535 __ATTR(ppl_sector, S_IRUGO|S_IWUSR, ppl_sector_show, ppl_sector_store);
3536 
3537 static ssize_t
3538 ppl_size_show(struct md_rdev *rdev, char *page)
3539 {
3540 	return sprintf(page, "%u\n", rdev->ppl.size);
3541 }
3542 
3543 static ssize_t
3544 ppl_size_store(struct md_rdev *rdev, const char *buf, size_t len)
3545 {
3546 	unsigned int size;
3547 
3548 	if (kstrtouint(buf, 10, &size) < 0)
3549 		return -EINVAL;
3550 
3551 	if (rdev->mddev->pers && test_bit(MD_HAS_PPL, &rdev->mddev->flags) &&
3552 	    rdev->raid_disk >= 0)
3553 		return -EBUSY;
3554 
3555 	if (rdev->mddev->persistent) {
3556 		if (rdev->mddev->major_version == 0)
3557 			return -EINVAL;
3558 		if (size > U16_MAX)
3559 			return -EINVAL;
3560 	} else if (!rdev->mddev->external) {
3561 		return -EBUSY;
3562 	}
3563 	rdev->ppl.size = size;
3564 	return len;
3565 }
3566 
3567 static struct rdev_sysfs_entry rdev_ppl_size =
3568 __ATTR(ppl_size, S_IRUGO|S_IWUSR, ppl_size_show, ppl_size_store);
3569 
3570 static struct attribute *rdev_default_attrs[] = {
3571 	&rdev_state.attr,
3572 	&rdev_errors.attr,
3573 	&rdev_slot.attr,
3574 	&rdev_offset.attr,
3575 	&rdev_new_offset.attr,
3576 	&rdev_size.attr,
3577 	&rdev_recovery_start.attr,
3578 	&rdev_bad_blocks.attr,
3579 	&rdev_unack_bad_blocks.attr,
3580 	&rdev_ppl_sector.attr,
3581 	&rdev_ppl_size.attr,
3582 	NULL,
3583 };
3584 ATTRIBUTE_GROUPS(rdev_default);
3585 static ssize_t
3586 rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
3587 {
3588 	struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
3589 	struct md_rdev *rdev = container_of(kobj, struct md_rdev, kobj);
3590 
3591 	if (!entry->show)
3592 		return -EIO;
3593 	if (!rdev->mddev)
3594 		return -ENODEV;
3595 	return entry->show(rdev, page);
3596 }
3597 
3598 static ssize_t
3599 rdev_attr_store(struct kobject *kobj, struct attribute *attr,
3600 	      const char *page, size_t length)
3601 {
3602 	struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
3603 	struct md_rdev *rdev = container_of(kobj, struct md_rdev, kobj);
3604 	struct kernfs_node *kn = NULL;
3605 	ssize_t rv;
3606 	struct mddev *mddev = rdev->mddev;
3607 
3608 	if (!entry->store)
3609 		return -EIO;
3610 	if (!capable(CAP_SYS_ADMIN))
3611 		return -EACCES;
3612 
3613 	if (entry->store == state_store && cmd_match(page, "remove"))
3614 		kn = sysfs_break_active_protection(kobj, attr);
3615 
3616 	rv = mddev ? mddev_lock(mddev) : -ENODEV;
3617 	if (!rv) {
3618 		if (rdev->mddev == NULL)
3619 			rv = -ENODEV;
3620 		else
3621 			rv = entry->store(rdev, page, length);
3622 		mddev_unlock(mddev);
3623 	}
3624 
3625 	if (kn)
3626 		sysfs_unbreak_active_protection(kn);
3627 
3628 	return rv;
3629 }
3630 
3631 static void rdev_free(struct kobject *ko)
3632 {
3633 	struct md_rdev *rdev = container_of(ko, struct md_rdev, kobj);
3634 	kfree(rdev);
3635 }
3636 static const struct sysfs_ops rdev_sysfs_ops = {
3637 	.show		= rdev_attr_show,
3638 	.store		= rdev_attr_store,
3639 };
3640 static const struct kobj_type rdev_ktype = {
3641 	.release	= rdev_free,
3642 	.sysfs_ops	= &rdev_sysfs_ops,
3643 	.default_groups	= rdev_default_groups,
3644 };
3645 
3646 int md_rdev_init(struct md_rdev *rdev)
3647 {
3648 	rdev->desc_nr = -1;
3649 	rdev->saved_raid_disk = -1;
3650 	rdev->raid_disk = -1;
3651 	rdev->flags = 0;
3652 	rdev->data_offset = 0;
3653 	rdev->new_data_offset = 0;
3654 	rdev->sb_events = 0;
3655 	rdev->last_read_error = 0;
3656 	rdev->sb_loaded = 0;
3657 	rdev->bb_page = NULL;
3658 	atomic_set(&rdev->nr_pending, 0);
3659 	atomic_set(&rdev->read_errors, 0);
3660 	atomic_set(&rdev->corrected_errors, 0);
3661 
3662 	INIT_LIST_HEAD(&rdev->same_set);
3663 	init_waitqueue_head(&rdev->blocked_wait);
3664 
3665 	/* Add space to store bad block list.
3666 	 * This reserves the space even on arrays where it cannot
3667 	 * be used - I wonder if that matters
3668 	 */
3669 	return badblocks_init(&rdev->badblocks, 0);
3670 }
3671 EXPORT_SYMBOL_GPL(md_rdev_init);
3672 
3673 /*
3674  * Import a device. If 'super_format' >= 0, then sanity check the superblock
3675  *
3676  * mark the device faulty if:
3677  *
3678  *   - the device is nonexistent (zero size)
3679  *   - the device has no valid superblock
3680  *
3681  * a faulty rdev _never_ has rdev->sb set.
3682  */
3683 static struct md_rdev *md_import_device(dev_t newdev, int super_format, int super_minor)
3684 {
3685 	struct md_rdev *rdev;
3686 	struct md_rdev *holder;
3687 	sector_t size;
3688 	int err;
3689 
3690 	rdev = kzalloc(sizeof(*rdev), GFP_KERNEL);
3691 	if (!rdev)
3692 		return ERR_PTR(-ENOMEM);
3693 
3694 	err = md_rdev_init(rdev);
3695 	if (err)
3696 		goto out_free_rdev;
3697 	err = alloc_disk_sb(rdev);
3698 	if (err)
3699 		goto out_clear_rdev;
3700 
3701 	if (super_format == -2) {
3702 		holder = &claim_rdev;
3703 	} else {
3704 		holder = rdev;
3705 		set_bit(Holder, &rdev->flags);
3706 	}
3707 
3708 	rdev->bdev = blkdev_get_by_dev(newdev, BLK_OPEN_READ | BLK_OPEN_WRITE,
3709 				       holder, NULL);
3710 	if (IS_ERR(rdev->bdev)) {
3711 		pr_warn("md: could not open device unknown-block(%u,%u).\n",
3712 			MAJOR(newdev), MINOR(newdev));
3713 		err = PTR_ERR(rdev->bdev);
3714 		goto out_clear_rdev;
3715 	}
3716 
3717 	kobject_init(&rdev->kobj, &rdev_ktype);
3718 
3719 	size = bdev_nr_bytes(rdev->bdev) >> BLOCK_SIZE_BITS;
3720 	if (!size) {
3721 		pr_warn("md: %pg has zero or unknown size, marking faulty!\n",
3722 			rdev->bdev);
3723 		err = -EINVAL;
3724 		goto out_blkdev_put;
3725 	}
3726 
3727 	if (super_format >= 0) {
3728 		err = super_types[super_format].
3729 			load_super(rdev, NULL, super_minor);
3730 		if (err == -EINVAL) {
3731 			pr_warn("md: %pg does not have a valid v%d.%d superblock, not importing!\n",
3732 				rdev->bdev,
3733 				super_format, super_minor);
3734 			goto out_blkdev_put;
3735 		}
3736 		if (err < 0) {
3737 			pr_warn("md: could not read %pg's sb, not importing!\n",
3738 				rdev->bdev);
3739 			goto out_blkdev_put;
3740 		}
3741 	}
3742 
3743 	return rdev;
3744 
3745 out_blkdev_put:
3746 	blkdev_put(rdev->bdev, holder);
3747 out_clear_rdev:
3748 	md_rdev_clear(rdev);
3749 out_free_rdev:
3750 	kfree(rdev);
3751 	return ERR_PTR(err);
3752 }
3753 
3754 /*
3755  * Check a full RAID array for plausibility
3756  */
3757 
3758 static int analyze_sbs(struct mddev *mddev)
3759 {
3760 	int i;
3761 	struct md_rdev *rdev, *freshest, *tmp;
3762 
3763 	freshest = NULL;
3764 	rdev_for_each_safe(rdev, tmp, mddev)
3765 		switch (super_types[mddev->major_version].
3766 			load_super(rdev, freshest, mddev->minor_version)) {
3767 		case 1:
3768 			freshest = rdev;
3769 			break;
3770 		case 0:
3771 			break;
3772 		default:
3773 			pr_warn("md: fatal superblock inconsistency in %pg -- removing from array\n",
3774 				rdev->bdev);
3775 			md_kick_rdev_from_array(rdev);
3776 		}
3777 
3778 	/* Cannot find a valid fresh disk */
3779 	if (!freshest) {
3780 		pr_warn("md: cannot find a valid disk\n");
3781 		return -EINVAL;
3782 	}
3783 
3784 	super_types[mddev->major_version].
3785 		validate_super(mddev, NULL/*freshest*/, freshest);
3786 
3787 	i = 0;
3788 	rdev_for_each_safe(rdev, tmp, mddev) {
3789 		if (mddev->max_disks &&
3790 		    (rdev->desc_nr >= mddev->max_disks ||
3791 		     i > mddev->max_disks)) {
3792 			pr_warn("md: %s: %pg: only %d devices permitted\n",
3793 				mdname(mddev), rdev->bdev,
3794 				mddev->max_disks);
3795 			md_kick_rdev_from_array(rdev);
3796 			continue;
3797 		}
3798 		if (rdev != freshest) {
3799 			if (super_types[mddev->major_version].
3800 			    validate_super(mddev, freshest, rdev)) {
3801 				pr_warn("md: kicking non-fresh %pg from array!\n",
3802 					rdev->bdev);
3803 				md_kick_rdev_from_array(rdev);
3804 				continue;
3805 			}
3806 		}
3807 		if (mddev->level == LEVEL_MULTIPATH) {
3808 			rdev->desc_nr = i++;
3809 			rdev->raid_disk = rdev->desc_nr;
3810 			set_bit(In_sync, &rdev->flags);
3811 		} else if (rdev->raid_disk >=
3812 			    (mddev->raid_disks - min(0, mddev->delta_disks)) &&
3813 			   !test_bit(Journal, &rdev->flags)) {
3814 			rdev->raid_disk = -1;
3815 			clear_bit(In_sync, &rdev->flags);
3816 		}
3817 	}
3818 
3819 	return 0;
3820 }
3821 
3822 /* Read a fixed-point number.
3823  * Numbers in sysfs attributes should be in "standard" units where
3824  * possible, so time should be in seconds.
3825  * However we internally use a a much smaller unit such as
3826  * milliseconds or jiffies.
3827  * This function takes a decimal number with a possible fractional
3828  * component, and produces an integer which is the result of
3829  * multiplying that number by 10^'scale'.
3830  * all without any floating-point arithmetic.
3831  */
3832 int strict_strtoul_scaled(const char *cp, unsigned long *res, int scale)
3833 {
3834 	unsigned long result = 0;
3835 	long decimals = -1;
3836 	while (isdigit(*cp) || (*cp == '.' && decimals < 0)) {
3837 		if (*cp == '.')
3838 			decimals = 0;
3839 		else if (decimals < scale) {
3840 			unsigned int value;
3841 			value = *cp - '0';
3842 			result = result * 10 + value;
3843 			if (decimals >= 0)
3844 				decimals++;
3845 		}
3846 		cp++;
3847 	}
3848 	if (*cp == '\n')
3849 		cp++;
3850 	if (*cp)
3851 		return -EINVAL;
3852 	if (decimals < 0)
3853 		decimals = 0;
3854 	*res = result * int_pow(10, scale - decimals);
3855 	return 0;
3856 }
3857 
3858 static ssize_t
3859 safe_delay_show(struct mddev *mddev, char *page)
3860 {
3861 	unsigned int msec = ((unsigned long)mddev->safemode_delay*1000)/HZ;
3862 
3863 	return sprintf(page, "%u.%03u\n", msec/1000, msec%1000);
3864 }
3865 static ssize_t
3866 safe_delay_store(struct mddev *mddev, const char *cbuf, size_t len)
3867 {
3868 	unsigned long msec;
3869 
3870 	if (mddev_is_clustered(mddev)) {
3871 		pr_warn("md: Safemode is disabled for clustered mode\n");
3872 		return -EINVAL;
3873 	}
3874 
3875 	if (strict_strtoul_scaled(cbuf, &msec, 3) < 0 || msec > UINT_MAX / HZ)
3876 		return -EINVAL;
3877 	if (msec == 0)
3878 		mddev->safemode_delay = 0;
3879 	else {
3880 		unsigned long old_delay = mddev->safemode_delay;
3881 		unsigned long new_delay = (msec*HZ)/1000;
3882 
3883 		if (new_delay == 0)
3884 			new_delay = 1;
3885 		mddev->safemode_delay = new_delay;
3886 		if (new_delay < old_delay || old_delay == 0)
3887 			mod_timer(&mddev->safemode_timer, jiffies+1);
3888 	}
3889 	return len;
3890 }
3891 static struct md_sysfs_entry md_safe_delay =
3892 __ATTR(safe_mode_delay, S_IRUGO|S_IWUSR,safe_delay_show, safe_delay_store);
3893 
3894 static ssize_t
3895 level_show(struct mddev *mddev, char *page)
3896 {
3897 	struct md_personality *p;
3898 	int ret;
3899 	spin_lock(&mddev->lock);
3900 	p = mddev->pers;
3901 	if (p)
3902 		ret = sprintf(page, "%s\n", p->name);
3903 	else if (mddev->clevel[0])
3904 		ret = sprintf(page, "%s\n", mddev->clevel);
3905 	else if (mddev->level != LEVEL_NONE)
3906 		ret = sprintf(page, "%d\n", mddev->level);
3907 	else
3908 		ret = 0;
3909 	spin_unlock(&mddev->lock);
3910 	return ret;
3911 }
3912 
3913 static ssize_t
3914 level_store(struct mddev *mddev, const char *buf, size_t len)
3915 {
3916 	char clevel[16];
3917 	ssize_t rv;
3918 	size_t slen = len;
3919 	struct md_personality *pers, *oldpers;
3920 	long level;
3921 	void *priv, *oldpriv;
3922 	struct md_rdev *rdev;
3923 
3924 	if (slen == 0 || slen >= sizeof(clevel))
3925 		return -EINVAL;
3926 
3927 	rv = mddev_lock(mddev);
3928 	if (rv)
3929 		return rv;
3930 
3931 	if (mddev->pers == NULL) {
3932 		strncpy(mddev->clevel, buf, slen);
3933 		if (mddev->clevel[slen-1] == '\n')
3934 			slen--;
3935 		mddev->clevel[slen] = 0;
3936 		mddev->level = LEVEL_NONE;
3937 		rv = len;
3938 		goto out_unlock;
3939 	}
3940 	rv = -EROFS;
3941 	if (!md_is_rdwr(mddev))
3942 		goto out_unlock;
3943 
3944 	/* request to change the personality.  Need to ensure:
3945 	 *  - array is not engaged in resync/recovery/reshape
3946 	 *  - old personality can be suspended
3947 	 *  - new personality will access other array.
3948 	 */
3949 
3950 	rv = -EBUSY;
3951 	if (mddev->sync_thread ||
3952 	    test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
3953 	    mddev->reshape_position != MaxSector ||
3954 	    mddev->sysfs_active)
3955 		goto out_unlock;
3956 
3957 	rv = -EINVAL;
3958 	if (!mddev->pers->quiesce) {
3959 		pr_warn("md: %s: %s does not support online personality change\n",
3960 			mdname(mddev), mddev->pers->name);
3961 		goto out_unlock;
3962 	}
3963 
3964 	/* Now find the new personality */
3965 	strncpy(clevel, buf, slen);
3966 	if (clevel[slen-1] == '\n')
3967 		slen--;
3968 	clevel[slen] = 0;
3969 	if (kstrtol(clevel, 10, &level))
3970 		level = LEVEL_NONE;
3971 
3972 	if (request_module("md-%s", clevel) != 0)
3973 		request_module("md-level-%s", clevel);
3974 	spin_lock(&pers_lock);
3975 	pers = find_pers(level, clevel);
3976 	if (!pers || !try_module_get(pers->owner)) {
3977 		spin_unlock(&pers_lock);
3978 		pr_warn("md: personality %s not loaded\n", clevel);
3979 		rv = -EINVAL;
3980 		goto out_unlock;
3981 	}
3982 	spin_unlock(&pers_lock);
3983 
3984 	if (pers == mddev->pers) {
3985 		/* Nothing to do! */
3986 		module_put(pers->owner);
3987 		rv = len;
3988 		goto out_unlock;
3989 	}
3990 	if (!pers->takeover) {
3991 		module_put(pers->owner);
3992 		pr_warn("md: %s: %s does not support personality takeover\n",
3993 			mdname(mddev), clevel);
3994 		rv = -EINVAL;
3995 		goto out_unlock;
3996 	}
3997 
3998 	rdev_for_each(rdev, mddev)
3999 		rdev->new_raid_disk = rdev->raid_disk;
4000 
4001 	/* ->takeover must set new_* and/or delta_disks
4002 	 * if it succeeds, and may set them when it fails.
4003 	 */
4004 	priv = pers->takeover(mddev);
4005 	if (IS_ERR(priv)) {
4006 		mddev->new_level = mddev->level;
4007 		mddev->new_layout = mddev->layout;
4008 		mddev->new_chunk_sectors = mddev->chunk_sectors;
4009 		mddev->raid_disks -= mddev->delta_disks;
4010 		mddev->delta_disks = 0;
4011 		mddev->reshape_backwards = 0;
4012 		module_put(pers->owner);
4013 		pr_warn("md: %s: %s would not accept array\n",
4014 			mdname(mddev), clevel);
4015 		rv = PTR_ERR(priv);
4016 		goto out_unlock;
4017 	}
4018 
4019 	/* Looks like we have a winner */
4020 	mddev_suspend(mddev);
4021 	mddev_detach(mddev);
4022 
4023 	spin_lock(&mddev->lock);
4024 	oldpers = mddev->pers;
4025 	oldpriv = mddev->private;
4026 	mddev->pers = pers;
4027 	mddev->private = priv;
4028 	strscpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
4029 	mddev->level = mddev->new_level;
4030 	mddev->layout = mddev->new_layout;
4031 	mddev->chunk_sectors = mddev->new_chunk_sectors;
4032 	mddev->delta_disks = 0;
4033 	mddev->reshape_backwards = 0;
4034 	mddev->degraded = 0;
4035 	spin_unlock(&mddev->lock);
4036 
4037 	if (oldpers->sync_request == NULL &&
4038 	    mddev->external) {
4039 		/* We are converting from a no-redundancy array
4040 		 * to a redundancy array and metadata is managed
4041 		 * externally so we need to be sure that writes
4042 		 * won't block due to a need to transition
4043 		 *      clean->dirty
4044 		 * until external management is started.
4045 		 */
4046 		mddev->in_sync = 0;
4047 		mddev->safemode_delay = 0;
4048 		mddev->safemode = 0;
4049 	}
4050 
4051 	oldpers->free(mddev, oldpriv);
4052 
4053 	if (oldpers->sync_request == NULL &&
4054 	    pers->sync_request != NULL) {
4055 		/* need to add the md_redundancy_group */
4056 		if (sysfs_create_group(&mddev->kobj, &md_redundancy_group))
4057 			pr_warn("md: cannot register extra attributes for %s\n",
4058 				mdname(mddev));
4059 		mddev->sysfs_action = sysfs_get_dirent(mddev->kobj.sd, "sync_action");
4060 		mddev->sysfs_completed = sysfs_get_dirent_safe(mddev->kobj.sd, "sync_completed");
4061 		mddev->sysfs_degraded = sysfs_get_dirent_safe(mddev->kobj.sd, "degraded");
4062 	}
4063 	if (oldpers->sync_request != NULL &&
4064 	    pers->sync_request == NULL) {
4065 		/* need to remove the md_redundancy_group */
4066 		if (mddev->to_remove == NULL)
4067 			mddev->to_remove = &md_redundancy_group;
4068 	}
4069 
4070 	module_put(oldpers->owner);
4071 
4072 	rdev_for_each(rdev, mddev) {
4073 		if (rdev->raid_disk < 0)
4074 			continue;
4075 		if (rdev->new_raid_disk >= mddev->raid_disks)
4076 			rdev->new_raid_disk = -1;
4077 		if (rdev->new_raid_disk == rdev->raid_disk)
4078 			continue;
4079 		sysfs_unlink_rdev(mddev, rdev);
4080 	}
4081 	rdev_for_each(rdev, mddev) {
4082 		if (rdev->raid_disk < 0)
4083 			continue;
4084 		if (rdev->new_raid_disk == rdev->raid_disk)
4085 			continue;
4086 		rdev->raid_disk = rdev->new_raid_disk;
4087 		if (rdev->raid_disk < 0)
4088 			clear_bit(In_sync, &rdev->flags);
4089 		else {
4090 			if (sysfs_link_rdev(mddev, rdev))
4091 				pr_warn("md: cannot register rd%d for %s after level change\n",
4092 					rdev->raid_disk, mdname(mddev));
4093 		}
4094 	}
4095 
4096 	if (pers->sync_request == NULL) {
4097 		/* this is now an array without redundancy, so
4098 		 * it must always be in_sync
4099 		 */
4100 		mddev->in_sync = 1;
4101 		del_timer_sync(&mddev->safemode_timer);
4102 	}
4103 	blk_set_stacking_limits(&mddev->queue->limits);
4104 	pers->run(mddev);
4105 	set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
4106 	mddev_resume(mddev);
4107 	if (!mddev->thread)
4108 		md_update_sb(mddev, 1);
4109 	sysfs_notify_dirent_safe(mddev->sysfs_level);
4110 	md_new_event();
4111 	rv = len;
4112 out_unlock:
4113 	mddev_unlock(mddev);
4114 	return rv;
4115 }
4116 
4117 static struct md_sysfs_entry md_level =
4118 __ATTR(level, S_IRUGO|S_IWUSR, level_show, level_store);
4119 
4120 static ssize_t
4121 layout_show(struct mddev *mddev, char *page)
4122 {
4123 	/* just a number, not meaningful for all levels */
4124 	if (mddev->reshape_position != MaxSector &&
4125 	    mddev->layout != mddev->new_layout)
4126 		return sprintf(page, "%d (%d)\n",
4127 			       mddev->new_layout, mddev->layout);
4128 	return sprintf(page, "%d\n", mddev->layout);
4129 }
4130 
4131 static ssize_t
4132 layout_store(struct mddev *mddev, const char *buf, size_t len)
4133 {
4134 	unsigned int n;
4135 	int err;
4136 
4137 	err = kstrtouint(buf, 10, &n);
4138 	if (err < 0)
4139 		return err;
4140 	err = mddev_lock(mddev);
4141 	if (err)
4142 		return err;
4143 
4144 	if (mddev->pers) {
4145 		if (mddev->pers->check_reshape == NULL)
4146 			err = -EBUSY;
4147 		else if (!md_is_rdwr(mddev))
4148 			err = -EROFS;
4149 		else {
4150 			mddev->new_layout = n;
4151 			err = mddev->pers->check_reshape(mddev);
4152 			if (err)
4153 				mddev->new_layout = mddev->layout;
4154 		}
4155 	} else {
4156 		mddev->new_layout = n;
4157 		if (mddev->reshape_position == MaxSector)
4158 			mddev->layout = n;
4159 	}
4160 	mddev_unlock(mddev);
4161 	return err ?: len;
4162 }
4163 static struct md_sysfs_entry md_layout =
4164 __ATTR(layout, S_IRUGO|S_IWUSR, layout_show, layout_store);
4165 
4166 static ssize_t
4167 raid_disks_show(struct mddev *mddev, char *page)
4168 {
4169 	if (mddev->raid_disks == 0)
4170 		return 0;
4171 	if (mddev->reshape_position != MaxSector &&
4172 	    mddev->delta_disks != 0)
4173 		return sprintf(page, "%d (%d)\n", mddev->raid_disks,
4174 			       mddev->raid_disks - mddev->delta_disks);
4175 	return sprintf(page, "%d\n", mddev->raid_disks);
4176 }
4177 
4178 static int update_raid_disks(struct mddev *mddev, int raid_disks);
4179 
4180 static ssize_t
4181 raid_disks_store(struct mddev *mddev, const char *buf, size_t len)
4182 {
4183 	unsigned int n;
4184 	int err;
4185 
4186 	err = kstrtouint(buf, 10, &n);
4187 	if (err < 0)
4188 		return err;
4189 
4190 	err = mddev_lock(mddev);
4191 	if (err)
4192 		return err;
4193 	if (mddev->pers)
4194 		err = update_raid_disks(mddev, n);
4195 	else if (mddev->reshape_position != MaxSector) {
4196 		struct md_rdev *rdev;
4197 		int olddisks = mddev->raid_disks - mddev->delta_disks;
4198 
4199 		err = -EINVAL;
4200 		rdev_for_each(rdev, mddev) {
4201 			if (olddisks < n &&
4202 			    rdev->data_offset < rdev->new_data_offset)
4203 				goto out_unlock;
4204 			if (olddisks > n &&
4205 			    rdev->data_offset > rdev->new_data_offset)
4206 				goto out_unlock;
4207 		}
4208 		err = 0;
4209 		mddev->delta_disks = n - olddisks;
4210 		mddev->raid_disks = n;
4211 		mddev->reshape_backwards = (mddev->delta_disks < 0);
4212 	} else
4213 		mddev->raid_disks = n;
4214 out_unlock:
4215 	mddev_unlock(mddev);
4216 	return err ? err : len;
4217 }
4218 static struct md_sysfs_entry md_raid_disks =
4219 __ATTR(raid_disks, S_IRUGO|S_IWUSR, raid_disks_show, raid_disks_store);
4220 
4221 static ssize_t
4222 uuid_show(struct mddev *mddev, char *page)
4223 {
4224 	return sprintf(page, "%pU\n", mddev->uuid);
4225 }
4226 static struct md_sysfs_entry md_uuid =
4227 __ATTR(uuid, S_IRUGO, uuid_show, NULL);
4228 
4229 static ssize_t
4230 chunk_size_show(struct mddev *mddev, char *page)
4231 {
4232 	if (mddev->reshape_position != MaxSector &&
4233 	    mddev->chunk_sectors != mddev->new_chunk_sectors)
4234 		return sprintf(page, "%d (%d)\n",
4235 			       mddev->new_chunk_sectors << 9,
4236 			       mddev->chunk_sectors << 9);
4237 	return sprintf(page, "%d\n", mddev->chunk_sectors << 9);
4238 }
4239 
4240 static ssize_t
4241 chunk_size_store(struct mddev *mddev, const char *buf, size_t len)
4242 {
4243 	unsigned long n;
4244 	int err;
4245 
4246 	err = kstrtoul(buf, 10, &n);
4247 	if (err < 0)
4248 		return err;
4249 
4250 	err = mddev_lock(mddev);
4251 	if (err)
4252 		return err;
4253 	if (mddev->pers) {
4254 		if (mddev->pers->check_reshape == NULL)
4255 			err = -EBUSY;
4256 		else if (!md_is_rdwr(mddev))
4257 			err = -EROFS;
4258 		else {
4259 			mddev->new_chunk_sectors = n >> 9;
4260 			err = mddev->pers->check_reshape(mddev);
4261 			if (err)
4262 				mddev->new_chunk_sectors = mddev->chunk_sectors;
4263 		}
4264 	} else {
4265 		mddev->new_chunk_sectors = n >> 9;
4266 		if (mddev->reshape_position == MaxSector)
4267 			mddev->chunk_sectors = n >> 9;
4268 	}
4269 	mddev_unlock(mddev);
4270 	return err ?: len;
4271 }
4272 static struct md_sysfs_entry md_chunk_size =
4273 __ATTR(chunk_size, S_IRUGO|S_IWUSR, chunk_size_show, chunk_size_store);
4274 
4275 static ssize_t
4276 resync_start_show(struct mddev *mddev, char *page)
4277 {
4278 	if (mddev->recovery_cp == MaxSector)
4279 		return sprintf(page, "none\n");
4280 	return sprintf(page, "%llu\n", (unsigned long long)mddev->recovery_cp);
4281 }
4282 
4283 static ssize_t
4284 resync_start_store(struct mddev *mddev, const char *buf, size_t len)
4285 {
4286 	unsigned long long n;
4287 	int err;
4288 
4289 	if (cmd_match(buf, "none"))
4290 		n = MaxSector;
4291 	else {
4292 		err = kstrtoull(buf, 10, &n);
4293 		if (err < 0)
4294 			return err;
4295 		if (n != (sector_t)n)
4296 			return -EINVAL;
4297 	}
4298 
4299 	err = mddev_lock(mddev);
4300 	if (err)
4301 		return err;
4302 	if (mddev->pers && !test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
4303 		err = -EBUSY;
4304 
4305 	if (!err) {
4306 		mddev->recovery_cp = n;
4307 		if (mddev->pers)
4308 			set_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags);
4309 	}
4310 	mddev_unlock(mddev);
4311 	return err ?: len;
4312 }
4313 static struct md_sysfs_entry md_resync_start =
4314 __ATTR_PREALLOC(resync_start, S_IRUGO|S_IWUSR,
4315 		resync_start_show, resync_start_store);
4316 
4317 /*
4318  * The array state can be:
4319  *
4320  * clear
4321  *     No devices, no size, no level
4322  *     Equivalent to STOP_ARRAY ioctl
4323  * inactive
4324  *     May have some settings, but array is not active
4325  *        all IO results in error
4326  *     When written, doesn't tear down array, but just stops it
4327  * suspended (not supported yet)
4328  *     All IO requests will block. The array can be reconfigured.
4329  *     Writing this, if accepted, will block until array is quiescent
4330  * readonly
4331  *     no resync can happen.  no superblocks get written.
4332  *     write requests fail
4333  * read-auto
4334  *     like readonly, but behaves like 'clean' on a write request.
4335  *
4336  * clean - no pending writes, but otherwise active.
4337  *     When written to inactive array, starts without resync
4338  *     If a write request arrives then
4339  *       if metadata is known, mark 'dirty' and switch to 'active'.
4340  *       if not known, block and switch to write-pending
4341  *     If written to an active array that has pending writes, then fails.
4342  * active
4343  *     fully active: IO and resync can be happening.
4344  *     When written to inactive array, starts with resync
4345  *
4346  * write-pending
4347  *     clean, but writes are blocked waiting for 'active' to be written.
4348  *
4349  * active-idle
4350  *     like active, but no writes have been seen for a while (100msec).
4351  *
4352  * broken
4353 *     Array is failed. It's useful because mounted-arrays aren't stopped
4354 *     when array is failed, so this state will at least alert the user that
4355 *     something is wrong.
4356  */
4357 enum array_state { clear, inactive, suspended, readonly, read_auto, clean, active,
4358 		   write_pending, active_idle, broken, bad_word};
4359 static char *array_states[] = {
4360 	"clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active",
4361 	"write-pending", "active-idle", "broken", NULL };
4362 
4363 static int match_word(const char *word, char **list)
4364 {
4365 	int n;
4366 	for (n=0; list[n]; n++)
4367 		if (cmd_match(word, list[n]))
4368 			break;
4369 	return n;
4370 }
4371 
4372 static ssize_t
4373 array_state_show(struct mddev *mddev, char *page)
4374 {
4375 	enum array_state st = inactive;
4376 
4377 	if (mddev->pers && !test_bit(MD_NOT_READY, &mddev->flags)) {
4378 		switch(mddev->ro) {
4379 		case MD_RDONLY:
4380 			st = readonly;
4381 			break;
4382 		case MD_AUTO_READ:
4383 			st = read_auto;
4384 			break;
4385 		case MD_RDWR:
4386 			spin_lock(&mddev->lock);
4387 			if (test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags))
4388 				st = write_pending;
4389 			else if (mddev->in_sync)
4390 				st = clean;
4391 			else if (mddev->safemode)
4392 				st = active_idle;
4393 			else
4394 				st = active;
4395 			spin_unlock(&mddev->lock);
4396 		}
4397 
4398 		if (test_bit(MD_BROKEN, &mddev->flags) && st == clean)
4399 			st = broken;
4400 	} else {
4401 		if (list_empty(&mddev->disks) &&
4402 		    mddev->raid_disks == 0 &&
4403 		    mddev->dev_sectors == 0)
4404 			st = clear;
4405 		else
4406 			st = inactive;
4407 	}
4408 	return sprintf(page, "%s\n", array_states[st]);
4409 }
4410 
4411 static int do_md_stop(struct mddev *mddev, int ro, struct block_device *bdev);
4412 static int md_set_readonly(struct mddev *mddev, struct block_device *bdev);
4413 static int restart_array(struct mddev *mddev);
4414 
4415 static ssize_t
4416 array_state_store(struct mddev *mddev, const char *buf, size_t len)
4417 {
4418 	int err = 0;
4419 	enum array_state st = match_word(buf, array_states);
4420 
4421 	if (mddev->pers && (st == active || st == clean) &&
4422 	    mddev->ro != MD_RDONLY) {
4423 		/* don't take reconfig_mutex when toggling between
4424 		 * clean and active
4425 		 */
4426 		spin_lock(&mddev->lock);
4427 		if (st == active) {
4428 			restart_array(mddev);
4429 			clear_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags);
4430 			md_wakeup_thread(mddev->thread);
4431 			wake_up(&mddev->sb_wait);
4432 		} else /* st == clean */ {
4433 			restart_array(mddev);
4434 			if (!set_in_sync(mddev))
4435 				err = -EBUSY;
4436 		}
4437 		if (!err)
4438 			sysfs_notify_dirent_safe(mddev->sysfs_state);
4439 		spin_unlock(&mddev->lock);
4440 		return err ?: len;
4441 	}
4442 	err = mddev_lock(mddev);
4443 	if (err)
4444 		return err;
4445 	err = -EINVAL;
4446 	switch(st) {
4447 	case bad_word:
4448 		break;
4449 	case clear:
4450 		/* stopping an active array */
4451 		err = do_md_stop(mddev, 0, NULL);
4452 		break;
4453 	case inactive:
4454 		/* stopping an active array */
4455 		if (mddev->pers)
4456 			err = do_md_stop(mddev, 2, NULL);
4457 		else
4458 			err = 0; /* already inactive */
4459 		break;
4460 	case suspended:
4461 		break; /* not supported yet */
4462 	case readonly:
4463 		if (mddev->pers)
4464 			err = md_set_readonly(mddev, NULL);
4465 		else {
4466 			mddev->ro = MD_RDONLY;
4467 			set_disk_ro(mddev->gendisk, 1);
4468 			err = do_md_run(mddev);
4469 		}
4470 		break;
4471 	case read_auto:
4472 		if (mddev->pers) {
4473 			if (md_is_rdwr(mddev))
4474 				err = md_set_readonly(mddev, NULL);
4475 			else if (mddev->ro == MD_RDONLY)
4476 				err = restart_array(mddev);
4477 			if (err == 0) {
4478 				mddev->ro = MD_AUTO_READ;
4479 				set_disk_ro(mddev->gendisk, 0);
4480 			}
4481 		} else {
4482 			mddev->ro = MD_AUTO_READ;
4483 			err = do_md_run(mddev);
4484 		}
4485 		break;
4486 	case clean:
4487 		if (mddev->pers) {
4488 			err = restart_array(mddev);
4489 			if (err)
4490 				break;
4491 			spin_lock(&mddev->lock);
4492 			if (!set_in_sync(mddev))
4493 				err = -EBUSY;
4494 			spin_unlock(&mddev->lock);
4495 		} else
4496 			err = -EINVAL;
4497 		break;
4498 	case active:
4499 		if (mddev->pers) {
4500 			err = restart_array(mddev);
4501 			if (err)
4502 				break;
4503 			clear_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags);
4504 			wake_up(&mddev->sb_wait);
4505 			err = 0;
4506 		} else {
4507 			mddev->ro = MD_RDWR;
4508 			set_disk_ro(mddev->gendisk, 0);
4509 			err = do_md_run(mddev);
4510 		}
4511 		break;
4512 	case write_pending:
4513 	case active_idle:
4514 	case broken:
4515 		/* these cannot be set */
4516 		break;
4517 	}
4518 
4519 	if (!err) {
4520 		if (mddev->hold_active == UNTIL_IOCTL)
4521 			mddev->hold_active = 0;
4522 		sysfs_notify_dirent_safe(mddev->sysfs_state);
4523 	}
4524 	mddev_unlock(mddev);
4525 	return err ?: len;
4526 }
4527 static struct md_sysfs_entry md_array_state =
4528 __ATTR_PREALLOC(array_state, S_IRUGO|S_IWUSR, array_state_show, array_state_store);
4529 
4530 static ssize_t
4531 max_corrected_read_errors_show(struct mddev *mddev, char *page) {
4532 	return sprintf(page, "%d\n",
4533 		       atomic_read(&mddev->max_corr_read_errors));
4534 }
4535 
4536 static ssize_t
4537 max_corrected_read_errors_store(struct mddev *mddev, const char *buf, size_t len)
4538 {
4539 	unsigned int n;
4540 	int rv;
4541 
4542 	rv = kstrtouint(buf, 10, &n);
4543 	if (rv < 0)
4544 		return rv;
4545 	if (n > INT_MAX)
4546 		return -EINVAL;
4547 	atomic_set(&mddev->max_corr_read_errors, n);
4548 	return len;
4549 }
4550 
4551 static struct md_sysfs_entry max_corr_read_errors =
4552 __ATTR(max_read_errors, S_IRUGO|S_IWUSR, max_corrected_read_errors_show,
4553 	max_corrected_read_errors_store);
4554 
4555 static ssize_t
4556 null_show(struct mddev *mddev, char *page)
4557 {
4558 	return -EINVAL;
4559 }
4560 
4561 static ssize_t
4562 new_dev_store(struct mddev *mddev, const char *buf, size_t len)
4563 {
4564 	/* buf must be %d:%d\n? giving major and minor numbers */
4565 	/* The new device is added to the array.
4566 	 * If the array has a persistent superblock, we read the
4567 	 * superblock to initialise info and check validity.
4568 	 * Otherwise, only checking done is that in bind_rdev_to_array,
4569 	 * which mainly checks size.
4570 	 */
4571 	char *e;
4572 	int major = simple_strtoul(buf, &e, 10);
4573 	int minor;
4574 	dev_t dev;
4575 	struct md_rdev *rdev;
4576 	int err;
4577 
4578 	if (!*buf || *e != ':' || !e[1] || e[1] == '\n')
4579 		return -EINVAL;
4580 	minor = simple_strtoul(e+1, &e, 10);
4581 	if (*e && *e != '\n')
4582 		return -EINVAL;
4583 	dev = MKDEV(major, minor);
4584 	if (major != MAJOR(dev) ||
4585 	    minor != MINOR(dev))
4586 		return -EOVERFLOW;
4587 
4588 	err = mddev_lock(mddev);
4589 	if (err)
4590 		return err;
4591 	if (mddev->persistent) {
4592 		rdev = md_import_device(dev, mddev->major_version,
4593 					mddev->minor_version);
4594 		if (!IS_ERR(rdev) && !list_empty(&mddev->disks)) {
4595 			struct md_rdev *rdev0
4596 				= list_entry(mddev->disks.next,
4597 					     struct md_rdev, same_set);
4598 			err = super_types[mddev->major_version]
4599 				.load_super(rdev, rdev0, mddev->minor_version);
4600 			if (err < 0)
4601 				goto out;
4602 		}
4603 	} else if (mddev->external)
4604 		rdev = md_import_device(dev, -2, -1);
4605 	else
4606 		rdev = md_import_device(dev, -1, -1);
4607 
4608 	if (IS_ERR(rdev)) {
4609 		mddev_unlock(mddev);
4610 		return PTR_ERR(rdev);
4611 	}
4612 	err = bind_rdev_to_array(rdev, mddev);
4613  out:
4614 	if (err)
4615 		export_rdev(rdev, mddev);
4616 	mddev_unlock(mddev);
4617 	if (!err)
4618 		md_new_event();
4619 	return err ? err : len;
4620 }
4621 
4622 static struct md_sysfs_entry md_new_device =
4623 __ATTR(new_dev, S_IWUSR, null_show, new_dev_store);
4624 
4625 static ssize_t
4626 bitmap_store(struct mddev *mddev, const char *buf, size_t len)
4627 {
4628 	char *end;
4629 	unsigned long chunk, end_chunk;
4630 	int err;
4631 
4632 	err = mddev_lock(mddev);
4633 	if (err)
4634 		return err;
4635 	if (!mddev->bitmap)
4636 		goto out;
4637 	/* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */
4638 	while (*buf) {
4639 		chunk = end_chunk = simple_strtoul(buf, &end, 0);
4640 		if (buf == end) break;
4641 		if (*end == '-') { /* range */
4642 			buf = end + 1;
4643 			end_chunk = simple_strtoul(buf, &end, 0);
4644 			if (buf == end) break;
4645 		}
4646 		if (*end && !isspace(*end)) break;
4647 		md_bitmap_dirty_bits(mddev->bitmap, chunk, end_chunk);
4648 		buf = skip_spaces(end);
4649 	}
4650 	md_bitmap_unplug(mddev->bitmap); /* flush the bits to disk */
4651 out:
4652 	mddev_unlock(mddev);
4653 	return len;
4654 }
4655 
4656 static struct md_sysfs_entry md_bitmap =
4657 __ATTR(bitmap_set_bits, S_IWUSR, null_show, bitmap_store);
4658 
4659 static ssize_t
4660 size_show(struct mddev *mddev, char *page)
4661 {
4662 	return sprintf(page, "%llu\n",
4663 		(unsigned long long)mddev->dev_sectors / 2);
4664 }
4665 
4666 static int update_size(struct mddev *mddev, sector_t num_sectors);
4667 
4668 static ssize_t
4669 size_store(struct mddev *mddev, const char *buf, size_t len)
4670 {
4671 	/* If array is inactive, we can reduce the component size, but
4672 	 * not increase it (except from 0).
4673 	 * If array is active, we can try an on-line resize
4674 	 */
4675 	sector_t sectors;
4676 	int err = strict_blocks_to_sectors(buf, &sectors);
4677 
4678 	if (err < 0)
4679 		return err;
4680 	err = mddev_lock(mddev);
4681 	if (err)
4682 		return err;
4683 	if (mddev->pers) {
4684 		err = update_size(mddev, sectors);
4685 		if (err == 0)
4686 			md_update_sb(mddev, 1);
4687 	} else {
4688 		if (mddev->dev_sectors == 0 ||
4689 		    mddev->dev_sectors > sectors)
4690 			mddev->dev_sectors = sectors;
4691 		else
4692 			err = -ENOSPC;
4693 	}
4694 	mddev_unlock(mddev);
4695 	return err ? err : len;
4696 }
4697 
4698 static struct md_sysfs_entry md_size =
4699 __ATTR(component_size, S_IRUGO|S_IWUSR, size_show, size_store);
4700 
4701 /* Metadata version.
4702  * This is one of
4703  *   'none' for arrays with no metadata (good luck...)
4704  *   'external' for arrays with externally managed metadata,
4705  * or N.M for internally known formats
4706  */
4707 static ssize_t
4708 metadata_show(struct mddev *mddev, char *page)
4709 {
4710 	if (mddev->persistent)
4711 		return sprintf(page, "%d.%d\n",
4712 			       mddev->major_version, mddev->minor_version);
4713 	else if (mddev->external)
4714 		return sprintf(page, "external:%s\n", mddev->metadata_type);
4715 	else
4716 		return sprintf(page, "none\n");
4717 }
4718 
4719 static ssize_t
4720 metadata_store(struct mddev *mddev, const char *buf, size_t len)
4721 {
4722 	int major, minor;
4723 	char *e;
4724 	int err;
4725 	/* Changing the details of 'external' metadata is
4726 	 * always permitted.  Otherwise there must be
4727 	 * no devices attached to the array.
4728 	 */
4729 
4730 	err = mddev_lock(mddev);
4731 	if (err)
4732 		return err;
4733 	err = -EBUSY;
4734 	if (mddev->external && strncmp(buf, "external:", 9) == 0)
4735 		;
4736 	else if (!list_empty(&mddev->disks))
4737 		goto out_unlock;
4738 
4739 	err = 0;
4740 	if (cmd_match(buf, "none")) {
4741 		mddev->persistent = 0;
4742 		mddev->external = 0;
4743 		mddev->major_version = 0;
4744 		mddev->minor_version = 90;
4745 		goto out_unlock;
4746 	}
4747 	if (strncmp(buf, "external:", 9) == 0) {
4748 		size_t namelen = len-9;
4749 		if (namelen >= sizeof(mddev->metadata_type))
4750 			namelen = sizeof(mddev->metadata_type)-1;
4751 		strncpy(mddev->metadata_type, buf+9, namelen);
4752 		mddev->metadata_type[namelen] = 0;
4753 		if (namelen && mddev->metadata_type[namelen-1] == '\n')
4754 			mddev->metadata_type[--namelen] = 0;
4755 		mddev->persistent = 0;
4756 		mddev->external = 1;
4757 		mddev->major_version = 0;
4758 		mddev->minor_version = 90;
4759 		goto out_unlock;
4760 	}
4761 	major = simple_strtoul(buf, &e, 10);
4762 	err = -EINVAL;
4763 	if (e==buf || *e != '.')
4764 		goto out_unlock;
4765 	buf = e+1;
4766 	minor = simple_strtoul(buf, &e, 10);
4767 	if (e==buf || (*e && *e != '\n') )
4768 		goto out_unlock;
4769 	err = -ENOENT;
4770 	if (major >= ARRAY_SIZE(super_types) || super_types[major].name == NULL)
4771 		goto out_unlock;
4772 	mddev->major_version = major;
4773 	mddev->minor_version = minor;
4774 	mddev->persistent = 1;
4775 	mddev->external = 0;
4776 	err = 0;
4777 out_unlock:
4778 	mddev_unlock(mddev);
4779 	return err ?: len;
4780 }
4781 
4782 static struct md_sysfs_entry md_metadata =
4783 __ATTR_PREALLOC(metadata_version, S_IRUGO|S_IWUSR, metadata_show, metadata_store);
4784 
4785 static ssize_t
4786 action_show(struct mddev *mddev, char *page)
4787 {
4788 	char *type = "idle";
4789 	unsigned long recovery = mddev->recovery;
4790 	if (test_bit(MD_RECOVERY_FROZEN, &recovery))
4791 		type = "frozen";
4792 	else if (test_bit(MD_RECOVERY_RUNNING, &recovery) ||
4793 	    (md_is_rdwr(mddev) && test_bit(MD_RECOVERY_NEEDED, &recovery))) {
4794 		if (test_bit(MD_RECOVERY_RESHAPE, &recovery))
4795 			type = "reshape";
4796 		else if (test_bit(MD_RECOVERY_SYNC, &recovery)) {
4797 			if (!test_bit(MD_RECOVERY_REQUESTED, &recovery))
4798 				type = "resync";
4799 			else if (test_bit(MD_RECOVERY_CHECK, &recovery))
4800 				type = "check";
4801 			else
4802 				type = "repair";
4803 		} else if (test_bit(MD_RECOVERY_RECOVER, &recovery))
4804 			type = "recover";
4805 		else if (mddev->reshape_position != MaxSector)
4806 			type = "reshape";
4807 	}
4808 	return sprintf(page, "%s\n", type);
4809 }
4810 
4811 static void stop_sync_thread(struct mddev *mddev)
4812 {
4813 	if (!test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4814 		return;
4815 
4816 	if (mddev_lock(mddev))
4817 		return;
4818 
4819 	/*
4820 	 * Check again in case MD_RECOVERY_RUNNING is cleared before lock is
4821 	 * held.
4822 	 */
4823 	if (!test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) {
4824 		mddev_unlock(mddev);
4825 		return;
4826 	}
4827 
4828 	if (work_pending(&mddev->del_work))
4829 		flush_workqueue(md_misc_wq);
4830 
4831 	set_bit(MD_RECOVERY_INTR, &mddev->recovery);
4832 	/*
4833 	 * Thread might be blocked waiting for metadata update which will now
4834 	 * never happen
4835 	 */
4836 	md_wakeup_thread_directly(mddev->sync_thread);
4837 
4838 	mddev_unlock(mddev);
4839 }
4840 
4841 static void idle_sync_thread(struct mddev *mddev)
4842 {
4843 	int sync_seq = atomic_read(&mddev->sync_seq);
4844 
4845 	mutex_lock(&mddev->sync_mutex);
4846 	clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4847 	stop_sync_thread(mddev);
4848 
4849 	wait_event(resync_wait, sync_seq != atomic_read(&mddev->sync_seq) ||
4850 			!test_bit(MD_RECOVERY_RUNNING, &mddev->recovery));
4851 
4852 	mutex_unlock(&mddev->sync_mutex);
4853 }
4854 
4855 static void frozen_sync_thread(struct mddev *mddev)
4856 {
4857 	mutex_lock(&mddev->sync_mutex);
4858 	set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4859 	stop_sync_thread(mddev);
4860 
4861 	wait_event(resync_wait, mddev->sync_thread == NULL &&
4862 			!test_bit(MD_RECOVERY_RUNNING, &mddev->recovery));
4863 
4864 	mutex_unlock(&mddev->sync_mutex);
4865 }
4866 
4867 static ssize_t
4868 action_store(struct mddev *mddev, const char *page, size_t len)
4869 {
4870 	if (!mddev->pers || !mddev->pers->sync_request)
4871 		return -EINVAL;
4872 
4873 
4874 	if (cmd_match(page, "idle"))
4875 		idle_sync_thread(mddev);
4876 	else if (cmd_match(page, "frozen"))
4877 		frozen_sync_thread(mddev);
4878 	else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4879 		return -EBUSY;
4880 	else if (cmd_match(page, "resync"))
4881 		clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4882 	else if (cmd_match(page, "recover")) {
4883 		clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4884 		set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
4885 	} else if (cmd_match(page, "reshape")) {
4886 		int err;
4887 		if (mddev->pers->start_reshape == NULL)
4888 			return -EINVAL;
4889 		err = mddev_lock(mddev);
4890 		if (!err) {
4891 			if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) {
4892 				err =  -EBUSY;
4893 			} else if (mddev->reshape_position == MaxSector ||
4894 				   mddev->pers->check_reshape == NULL ||
4895 				   mddev->pers->check_reshape(mddev)) {
4896 				clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4897 				err = mddev->pers->start_reshape(mddev);
4898 			} else {
4899 				/*
4900 				 * If reshape is still in progress, and
4901 				 * md_check_recovery() can continue to reshape,
4902 				 * don't restart reshape because data can be
4903 				 * corrupted for raid456.
4904 				 */
4905 				clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4906 			}
4907 			mddev_unlock(mddev);
4908 		}
4909 		if (err)
4910 			return err;
4911 		sysfs_notify_dirent_safe(mddev->sysfs_degraded);
4912 	} else {
4913 		if (cmd_match(page, "check"))
4914 			set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
4915 		else if (!cmd_match(page, "repair"))
4916 			return -EINVAL;
4917 		clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4918 		set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
4919 		set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
4920 	}
4921 	if (mddev->ro == MD_AUTO_READ) {
4922 		/* A write to sync_action is enough to justify
4923 		 * canceling read-auto mode
4924 		 */
4925 		mddev->ro = MD_RDWR;
4926 		md_wakeup_thread(mddev->sync_thread);
4927 	}
4928 	set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4929 	md_wakeup_thread(mddev->thread);
4930 	sysfs_notify_dirent_safe(mddev->sysfs_action);
4931 	return len;
4932 }
4933 
4934 static struct md_sysfs_entry md_scan_mode =
4935 __ATTR_PREALLOC(sync_action, S_IRUGO|S_IWUSR, action_show, action_store);
4936 
4937 static ssize_t
4938 last_sync_action_show(struct mddev *mddev, char *page)
4939 {
4940 	return sprintf(page, "%s\n", mddev->last_sync_action);
4941 }
4942 
4943 static struct md_sysfs_entry md_last_scan_mode = __ATTR_RO(last_sync_action);
4944 
4945 static ssize_t
4946 mismatch_cnt_show(struct mddev *mddev, char *page)
4947 {
4948 	return sprintf(page, "%llu\n",
4949 		       (unsigned long long)
4950 		       atomic64_read(&mddev->resync_mismatches));
4951 }
4952 
4953 static struct md_sysfs_entry md_mismatches = __ATTR_RO(mismatch_cnt);
4954 
4955 static ssize_t
4956 sync_min_show(struct mddev *mddev, char *page)
4957 {
4958 	return sprintf(page, "%d (%s)\n", speed_min(mddev),
4959 		       mddev->sync_speed_min ? "local": "system");
4960 }
4961 
4962 static ssize_t
4963 sync_min_store(struct mddev *mddev, const char *buf, size_t len)
4964 {
4965 	unsigned int min;
4966 	int rv;
4967 
4968 	if (strncmp(buf, "system", 6)==0) {
4969 		min = 0;
4970 	} else {
4971 		rv = kstrtouint(buf, 10, &min);
4972 		if (rv < 0)
4973 			return rv;
4974 		if (min == 0)
4975 			return -EINVAL;
4976 	}
4977 	mddev->sync_speed_min = min;
4978 	return len;
4979 }
4980 
4981 static struct md_sysfs_entry md_sync_min =
4982 __ATTR(sync_speed_min, S_IRUGO|S_IWUSR, sync_min_show, sync_min_store);
4983 
4984 static ssize_t
4985 sync_max_show(struct mddev *mddev, char *page)
4986 {
4987 	return sprintf(page, "%d (%s)\n", speed_max(mddev),
4988 		       mddev->sync_speed_max ? "local": "system");
4989 }
4990 
4991 static ssize_t
4992 sync_max_store(struct mddev *mddev, const char *buf, size_t len)
4993 {
4994 	unsigned int max;
4995 	int rv;
4996 
4997 	if (strncmp(buf, "system", 6)==0) {
4998 		max = 0;
4999 	} else {
5000 		rv = kstrtouint(buf, 10, &max);
5001 		if (rv < 0)
5002 			return rv;
5003 		if (max == 0)
5004 			return -EINVAL;
5005 	}
5006 	mddev->sync_speed_max = max;
5007 	return len;
5008 }
5009 
5010 static struct md_sysfs_entry md_sync_max =
5011 __ATTR(sync_speed_max, S_IRUGO|S_IWUSR, sync_max_show, sync_max_store);
5012 
5013 static ssize_t
5014 degraded_show(struct mddev *mddev, char *page)
5015 {
5016 	return sprintf(page, "%d\n", mddev->degraded);
5017 }
5018 static struct md_sysfs_entry md_degraded = __ATTR_RO(degraded);
5019 
5020 static ssize_t
5021 sync_force_parallel_show(struct mddev *mddev, char *page)
5022 {
5023 	return sprintf(page, "%d\n", mddev->parallel_resync);
5024 }
5025 
5026 static ssize_t
5027 sync_force_parallel_store(struct mddev *mddev, const char *buf, size_t len)
5028 {
5029 	long n;
5030 
5031 	if (kstrtol(buf, 10, &n))
5032 		return -EINVAL;
5033 
5034 	if (n != 0 && n != 1)
5035 		return -EINVAL;
5036 
5037 	mddev->parallel_resync = n;
5038 
5039 	if (mddev->sync_thread)
5040 		wake_up(&resync_wait);
5041 
5042 	return len;
5043 }
5044 
5045 /* force parallel resync, even with shared block devices */
5046 static struct md_sysfs_entry md_sync_force_parallel =
5047 __ATTR(sync_force_parallel, S_IRUGO|S_IWUSR,
5048        sync_force_parallel_show, sync_force_parallel_store);
5049 
5050 static ssize_t
5051 sync_speed_show(struct mddev *mddev, char *page)
5052 {
5053 	unsigned long resync, dt, db;
5054 	if (mddev->curr_resync == MD_RESYNC_NONE)
5055 		return sprintf(page, "none\n");
5056 	resync = mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active);
5057 	dt = (jiffies - mddev->resync_mark) / HZ;
5058 	if (!dt) dt++;
5059 	db = resync - mddev->resync_mark_cnt;
5060 	return sprintf(page, "%lu\n", db/dt/2); /* K/sec */
5061 }
5062 
5063 static struct md_sysfs_entry md_sync_speed = __ATTR_RO(sync_speed);
5064 
5065 static ssize_t
5066 sync_completed_show(struct mddev *mddev, char *page)
5067 {
5068 	unsigned long long max_sectors, resync;
5069 
5070 	if (!test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
5071 		return sprintf(page, "none\n");
5072 
5073 	if (mddev->curr_resync == MD_RESYNC_YIELDED ||
5074 	    mddev->curr_resync == MD_RESYNC_DELAYED)
5075 		return sprintf(page, "delayed\n");
5076 
5077 	if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ||
5078 	    test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
5079 		max_sectors = mddev->resync_max_sectors;
5080 	else
5081 		max_sectors = mddev->dev_sectors;
5082 
5083 	resync = mddev->curr_resync_completed;
5084 	return sprintf(page, "%llu / %llu\n", resync, max_sectors);
5085 }
5086 
5087 static struct md_sysfs_entry md_sync_completed =
5088 	__ATTR_PREALLOC(sync_completed, S_IRUGO, sync_completed_show, NULL);
5089 
5090 static ssize_t
5091 min_sync_show(struct mddev *mddev, char *page)
5092 {
5093 	return sprintf(page, "%llu\n",
5094 		       (unsigned long long)mddev->resync_min);
5095 }
5096 static ssize_t
5097 min_sync_store(struct mddev *mddev, const char *buf, size_t len)
5098 {
5099 	unsigned long long min;
5100 	int err;
5101 
5102 	if (kstrtoull(buf, 10, &min))
5103 		return -EINVAL;
5104 
5105 	spin_lock(&mddev->lock);
5106 	err = -EINVAL;
5107 	if (min > mddev->resync_max)
5108 		goto out_unlock;
5109 
5110 	err = -EBUSY;
5111 	if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
5112 		goto out_unlock;
5113 
5114 	/* Round down to multiple of 4K for safety */
5115 	mddev->resync_min = round_down(min, 8);
5116 	err = 0;
5117 
5118 out_unlock:
5119 	spin_unlock(&mddev->lock);
5120 	return err ?: len;
5121 }
5122 
5123 static struct md_sysfs_entry md_min_sync =
5124 __ATTR(sync_min, S_IRUGO|S_IWUSR, min_sync_show, min_sync_store);
5125 
5126 static ssize_t
5127 max_sync_show(struct mddev *mddev, char *page)
5128 {
5129 	if (mddev->resync_max == MaxSector)
5130 		return sprintf(page, "max\n");
5131 	else
5132 		return sprintf(page, "%llu\n",
5133 			       (unsigned long long)mddev->resync_max);
5134 }
5135 static ssize_t
5136 max_sync_store(struct mddev *mddev, const char *buf, size_t len)
5137 {
5138 	int err;
5139 	spin_lock(&mddev->lock);
5140 	if (strncmp(buf, "max", 3) == 0)
5141 		mddev->resync_max = MaxSector;
5142 	else {
5143 		unsigned long long max;
5144 		int chunk;
5145 
5146 		err = -EINVAL;
5147 		if (kstrtoull(buf, 10, &max))
5148 			goto out_unlock;
5149 		if (max < mddev->resync_min)
5150 			goto out_unlock;
5151 
5152 		err = -EBUSY;
5153 		if (max < mddev->resync_max && md_is_rdwr(mddev) &&
5154 		    test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
5155 			goto out_unlock;
5156 
5157 		/* Must be a multiple of chunk_size */
5158 		chunk = mddev->chunk_sectors;
5159 		if (chunk) {
5160 			sector_t temp = max;
5161 
5162 			err = -EINVAL;
5163 			if (sector_div(temp, chunk))
5164 				goto out_unlock;
5165 		}
5166 		mddev->resync_max = max;
5167 	}
5168 	wake_up(&mddev->recovery_wait);
5169 	err = 0;
5170 out_unlock:
5171 	spin_unlock(&mddev->lock);
5172 	return err ?: len;
5173 }
5174 
5175 static struct md_sysfs_entry md_max_sync =
5176 __ATTR(sync_max, S_IRUGO|S_IWUSR, max_sync_show, max_sync_store);
5177 
5178 static ssize_t
5179 suspend_lo_show(struct mddev *mddev, char *page)
5180 {
5181 	return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_lo);
5182 }
5183 
5184 static ssize_t
5185 suspend_lo_store(struct mddev *mddev, const char *buf, size_t len)
5186 {
5187 	unsigned long long new;
5188 	int err;
5189 
5190 	err = kstrtoull(buf, 10, &new);
5191 	if (err < 0)
5192 		return err;
5193 	if (new != (sector_t)new)
5194 		return -EINVAL;
5195 
5196 	err = mddev_lock(mddev);
5197 	if (err)
5198 		return err;
5199 	err = -EINVAL;
5200 	if (mddev->pers == NULL ||
5201 	    mddev->pers->quiesce == NULL)
5202 		goto unlock;
5203 	mddev_suspend(mddev);
5204 	mddev->suspend_lo = new;
5205 	mddev_resume(mddev);
5206 
5207 	err = 0;
5208 unlock:
5209 	mddev_unlock(mddev);
5210 	return err ?: len;
5211 }
5212 static struct md_sysfs_entry md_suspend_lo =
5213 __ATTR(suspend_lo, S_IRUGO|S_IWUSR, suspend_lo_show, suspend_lo_store);
5214 
5215 static ssize_t
5216 suspend_hi_show(struct mddev *mddev, char *page)
5217 {
5218 	return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_hi);
5219 }
5220 
5221 static ssize_t
5222 suspend_hi_store(struct mddev *mddev, const char *buf, size_t len)
5223 {
5224 	unsigned long long new;
5225 	int err;
5226 
5227 	err = kstrtoull(buf, 10, &new);
5228 	if (err < 0)
5229 		return err;
5230 	if (new != (sector_t)new)
5231 		return -EINVAL;
5232 
5233 	err = mddev_lock(mddev);
5234 	if (err)
5235 		return err;
5236 	err = -EINVAL;
5237 	if (mddev->pers == NULL)
5238 		goto unlock;
5239 
5240 	mddev_suspend(mddev);
5241 	mddev->suspend_hi = new;
5242 	mddev_resume(mddev);
5243 
5244 	err = 0;
5245 unlock:
5246 	mddev_unlock(mddev);
5247 	return err ?: len;
5248 }
5249 static struct md_sysfs_entry md_suspend_hi =
5250 __ATTR(suspend_hi, S_IRUGO|S_IWUSR, suspend_hi_show, suspend_hi_store);
5251 
5252 static ssize_t
5253 reshape_position_show(struct mddev *mddev, char *page)
5254 {
5255 	if (mddev->reshape_position != MaxSector)
5256 		return sprintf(page, "%llu\n",
5257 			       (unsigned long long)mddev->reshape_position);
5258 	strcpy(page, "none\n");
5259 	return 5;
5260 }
5261 
5262 static ssize_t
5263 reshape_position_store(struct mddev *mddev, const char *buf, size_t len)
5264 {
5265 	struct md_rdev *rdev;
5266 	unsigned long long new;
5267 	int err;
5268 
5269 	err = kstrtoull(buf, 10, &new);
5270 	if (err < 0)
5271 		return err;
5272 	if (new != (sector_t)new)
5273 		return -EINVAL;
5274 	err = mddev_lock(mddev);
5275 	if (err)
5276 		return err;
5277 	err = -EBUSY;
5278 	if (mddev->pers)
5279 		goto unlock;
5280 	mddev->reshape_position = new;
5281 	mddev->delta_disks = 0;
5282 	mddev->reshape_backwards = 0;
5283 	mddev->new_level = mddev->level;
5284 	mddev->new_layout = mddev->layout;
5285 	mddev->new_chunk_sectors = mddev->chunk_sectors;
5286 	rdev_for_each(rdev, mddev)
5287 		rdev->new_data_offset = rdev->data_offset;
5288 	err = 0;
5289 unlock:
5290 	mddev_unlock(mddev);
5291 	return err ?: len;
5292 }
5293 
5294 static struct md_sysfs_entry md_reshape_position =
5295 __ATTR(reshape_position, S_IRUGO|S_IWUSR, reshape_position_show,
5296        reshape_position_store);
5297 
5298 static ssize_t
5299 reshape_direction_show(struct mddev *mddev, char *page)
5300 {
5301 	return sprintf(page, "%s\n",
5302 		       mddev->reshape_backwards ? "backwards" : "forwards");
5303 }
5304 
5305 static ssize_t
5306 reshape_direction_store(struct mddev *mddev, const char *buf, size_t len)
5307 {
5308 	int backwards = 0;
5309 	int err;
5310 
5311 	if (cmd_match(buf, "forwards"))
5312 		backwards = 0;
5313 	else if (cmd_match(buf, "backwards"))
5314 		backwards = 1;
5315 	else
5316 		return -EINVAL;
5317 	if (mddev->reshape_backwards == backwards)
5318 		return len;
5319 
5320 	err = mddev_lock(mddev);
5321 	if (err)
5322 		return err;
5323 	/* check if we are allowed to change */
5324 	if (mddev->delta_disks)
5325 		err = -EBUSY;
5326 	else if (mddev->persistent &&
5327 	    mddev->major_version == 0)
5328 		err =  -EINVAL;
5329 	else
5330 		mddev->reshape_backwards = backwards;
5331 	mddev_unlock(mddev);
5332 	return err ?: len;
5333 }
5334 
5335 static struct md_sysfs_entry md_reshape_direction =
5336 __ATTR(reshape_direction, S_IRUGO|S_IWUSR, reshape_direction_show,
5337        reshape_direction_store);
5338 
5339 static ssize_t
5340 array_size_show(struct mddev *mddev, char *page)
5341 {
5342 	if (mddev->external_size)
5343 		return sprintf(page, "%llu\n",
5344 			       (unsigned long long)mddev->array_sectors/2);
5345 	else
5346 		return sprintf(page, "default\n");
5347 }
5348 
5349 static ssize_t
5350 array_size_store(struct mddev *mddev, const char *buf, size_t len)
5351 {
5352 	sector_t sectors;
5353 	int err;
5354 
5355 	err = mddev_lock(mddev);
5356 	if (err)
5357 		return err;
5358 
5359 	/* cluster raid doesn't support change array_sectors */
5360 	if (mddev_is_clustered(mddev)) {
5361 		mddev_unlock(mddev);
5362 		return -EINVAL;
5363 	}
5364 
5365 	if (strncmp(buf, "default", 7) == 0) {
5366 		if (mddev->pers)
5367 			sectors = mddev->pers->size(mddev, 0, 0);
5368 		else
5369 			sectors = mddev->array_sectors;
5370 
5371 		mddev->external_size = 0;
5372 	} else {
5373 		if (strict_blocks_to_sectors(buf, &sectors) < 0)
5374 			err = -EINVAL;
5375 		else if (mddev->pers && mddev->pers->size(mddev, 0, 0) < sectors)
5376 			err = -E2BIG;
5377 		else
5378 			mddev->external_size = 1;
5379 	}
5380 
5381 	if (!err) {
5382 		mddev->array_sectors = sectors;
5383 		if (mddev->pers)
5384 			set_capacity_and_notify(mddev->gendisk,
5385 						mddev->array_sectors);
5386 	}
5387 	mddev_unlock(mddev);
5388 	return err ?: len;
5389 }
5390 
5391 static struct md_sysfs_entry md_array_size =
5392 __ATTR(array_size, S_IRUGO|S_IWUSR, array_size_show,
5393        array_size_store);
5394 
5395 static ssize_t
5396 consistency_policy_show(struct mddev *mddev, char *page)
5397 {
5398 	int ret;
5399 
5400 	if (test_bit(MD_HAS_JOURNAL, &mddev->flags)) {
5401 		ret = sprintf(page, "journal\n");
5402 	} else if (test_bit(MD_HAS_PPL, &mddev->flags)) {
5403 		ret = sprintf(page, "ppl\n");
5404 	} else if (mddev->bitmap) {
5405 		ret = sprintf(page, "bitmap\n");
5406 	} else if (mddev->pers) {
5407 		if (mddev->pers->sync_request)
5408 			ret = sprintf(page, "resync\n");
5409 		else
5410 			ret = sprintf(page, "none\n");
5411 	} else {
5412 		ret = sprintf(page, "unknown\n");
5413 	}
5414 
5415 	return ret;
5416 }
5417 
5418 static ssize_t
5419 consistency_policy_store(struct mddev *mddev, const char *buf, size_t len)
5420 {
5421 	int err = 0;
5422 
5423 	if (mddev->pers) {
5424 		if (mddev->pers->change_consistency_policy)
5425 			err = mddev->pers->change_consistency_policy(mddev, buf);
5426 		else
5427 			err = -EBUSY;
5428 	} else if (mddev->external && strncmp(buf, "ppl", 3) == 0) {
5429 		set_bit(MD_HAS_PPL, &mddev->flags);
5430 	} else {
5431 		err = -EINVAL;
5432 	}
5433 
5434 	return err ? err : len;
5435 }
5436 
5437 static struct md_sysfs_entry md_consistency_policy =
5438 __ATTR(consistency_policy, S_IRUGO | S_IWUSR, consistency_policy_show,
5439        consistency_policy_store);
5440 
5441 static ssize_t fail_last_dev_show(struct mddev *mddev, char *page)
5442 {
5443 	return sprintf(page, "%d\n", mddev->fail_last_dev);
5444 }
5445 
5446 /*
5447  * Setting fail_last_dev to true to allow last device to be forcibly removed
5448  * from RAID1/RAID10.
5449  */
5450 static ssize_t
5451 fail_last_dev_store(struct mddev *mddev, const char *buf, size_t len)
5452 {
5453 	int ret;
5454 	bool value;
5455 
5456 	ret = kstrtobool(buf, &value);
5457 	if (ret)
5458 		return ret;
5459 
5460 	if (value != mddev->fail_last_dev)
5461 		mddev->fail_last_dev = value;
5462 
5463 	return len;
5464 }
5465 static struct md_sysfs_entry md_fail_last_dev =
5466 __ATTR(fail_last_dev, S_IRUGO | S_IWUSR, fail_last_dev_show,
5467        fail_last_dev_store);
5468 
5469 static ssize_t serialize_policy_show(struct mddev *mddev, char *page)
5470 {
5471 	if (mddev->pers == NULL || (mddev->pers->level != 1))
5472 		return sprintf(page, "n/a\n");
5473 	else
5474 		return sprintf(page, "%d\n", mddev->serialize_policy);
5475 }
5476 
5477 /*
5478  * Setting serialize_policy to true to enforce write IO is not reordered
5479  * for raid1.
5480  */
5481 static ssize_t
5482 serialize_policy_store(struct mddev *mddev, const char *buf, size_t len)
5483 {
5484 	int err;
5485 	bool value;
5486 
5487 	err = kstrtobool(buf, &value);
5488 	if (err)
5489 		return err;
5490 
5491 	if (value == mddev->serialize_policy)
5492 		return len;
5493 
5494 	err = mddev_lock(mddev);
5495 	if (err)
5496 		return err;
5497 	if (mddev->pers == NULL || (mddev->pers->level != 1)) {
5498 		pr_err("md: serialize_policy is only effective for raid1\n");
5499 		err = -EINVAL;
5500 		goto unlock;
5501 	}
5502 
5503 	mddev_suspend(mddev);
5504 	if (value)
5505 		mddev_create_serial_pool(mddev, NULL, true);
5506 	else
5507 		mddev_destroy_serial_pool(mddev, NULL, true);
5508 	mddev->serialize_policy = value;
5509 	mddev_resume(mddev);
5510 unlock:
5511 	mddev_unlock(mddev);
5512 	return err ?: len;
5513 }
5514 
5515 static struct md_sysfs_entry md_serialize_policy =
5516 __ATTR(serialize_policy, S_IRUGO | S_IWUSR, serialize_policy_show,
5517        serialize_policy_store);
5518 
5519 
5520 static struct attribute *md_default_attrs[] = {
5521 	&md_level.attr,
5522 	&md_layout.attr,
5523 	&md_raid_disks.attr,
5524 	&md_uuid.attr,
5525 	&md_chunk_size.attr,
5526 	&md_size.attr,
5527 	&md_resync_start.attr,
5528 	&md_metadata.attr,
5529 	&md_new_device.attr,
5530 	&md_safe_delay.attr,
5531 	&md_array_state.attr,
5532 	&md_reshape_position.attr,
5533 	&md_reshape_direction.attr,
5534 	&md_array_size.attr,
5535 	&max_corr_read_errors.attr,
5536 	&md_consistency_policy.attr,
5537 	&md_fail_last_dev.attr,
5538 	&md_serialize_policy.attr,
5539 	NULL,
5540 };
5541 
5542 static const struct attribute_group md_default_group = {
5543 	.attrs = md_default_attrs,
5544 };
5545 
5546 static struct attribute *md_redundancy_attrs[] = {
5547 	&md_scan_mode.attr,
5548 	&md_last_scan_mode.attr,
5549 	&md_mismatches.attr,
5550 	&md_sync_min.attr,
5551 	&md_sync_max.attr,
5552 	&md_sync_speed.attr,
5553 	&md_sync_force_parallel.attr,
5554 	&md_sync_completed.attr,
5555 	&md_min_sync.attr,
5556 	&md_max_sync.attr,
5557 	&md_suspend_lo.attr,
5558 	&md_suspend_hi.attr,
5559 	&md_bitmap.attr,
5560 	&md_degraded.attr,
5561 	NULL,
5562 };
5563 static const struct attribute_group md_redundancy_group = {
5564 	.name = NULL,
5565 	.attrs = md_redundancy_attrs,
5566 };
5567 
5568 static const struct attribute_group *md_attr_groups[] = {
5569 	&md_default_group,
5570 	&md_bitmap_group,
5571 	NULL,
5572 };
5573 
5574 static ssize_t
5575 md_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
5576 {
5577 	struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
5578 	struct mddev *mddev = container_of(kobj, struct mddev, kobj);
5579 	ssize_t rv;
5580 
5581 	if (!entry->show)
5582 		return -EIO;
5583 	spin_lock(&all_mddevs_lock);
5584 	if (!mddev_get(mddev)) {
5585 		spin_unlock(&all_mddevs_lock);
5586 		return -EBUSY;
5587 	}
5588 	spin_unlock(&all_mddevs_lock);
5589 
5590 	rv = entry->show(mddev, page);
5591 	mddev_put(mddev);
5592 	return rv;
5593 }
5594 
5595 static ssize_t
5596 md_attr_store(struct kobject *kobj, struct attribute *attr,
5597 	      const char *page, size_t length)
5598 {
5599 	struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
5600 	struct mddev *mddev = container_of(kobj, struct mddev, kobj);
5601 	ssize_t rv;
5602 
5603 	if (!entry->store)
5604 		return -EIO;
5605 	if (!capable(CAP_SYS_ADMIN))
5606 		return -EACCES;
5607 	spin_lock(&all_mddevs_lock);
5608 	if (!mddev_get(mddev)) {
5609 		spin_unlock(&all_mddevs_lock);
5610 		return -EBUSY;
5611 	}
5612 	spin_unlock(&all_mddevs_lock);
5613 	rv = entry->store(mddev, page, length);
5614 	mddev_put(mddev);
5615 	return rv;
5616 }
5617 
5618 static void md_kobj_release(struct kobject *ko)
5619 {
5620 	struct mddev *mddev = container_of(ko, struct mddev, kobj);
5621 
5622 	if (mddev->sysfs_state)
5623 		sysfs_put(mddev->sysfs_state);
5624 	if (mddev->sysfs_level)
5625 		sysfs_put(mddev->sysfs_level);
5626 
5627 	del_gendisk(mddev->gendisk);
5628 	put_disk(mddev->gendisk);
5629 }
5630 
5631 static const struct sysfs_ops md_sysfs_ops = {
5632 	.show	= md_attr_show,
5633 	.store	= md_attr_store,
5634 };
5635 static const struct kobj_type md_ktype = {
5636 	.release	= md_kobj_release,
5637 	.sysfs_ops	= &md_sysfs_ops,
5638 	.default_groups	= md_attr_groups,
5639 };
5640 
5641 int mdp_major = 0;
5642 
5643 static void mddev_delayed_delete(struct work_struct *ws)
5644 {
5645 	struct mddev *mddev = container_of(ws, struct mddev, del_work);
5646 
5647 	kobject_put(&mddev->kobj);
5648 }
5649 
5650 static void no_op(struct percpu_ref *r) {}
5651 
5652 int mddev_init_writes_pending(struct mddev *mddev)
5653 {
5654 	if (mddev->writes_pending.percpu_count_ptr)
5655 		return 0;
5656 	if (percpu_ref_init(&mddev->writes_pending, no_op,
5657 			    PERCPU_REF_ALLOW_REINIT, GFP_KERNEL) < 0)
5658 		return -ENOMEM;
5659 	/* We want to start with the refcount at zero */
5660 	percpu_ref_put(&mddev->writes_pending);
5661 	return 0;
5662 }
5663 EXPORT_SYMBOL_GPL(mddev_init_writes_pending);
5664 
5665 struct mddev *md_alloc(dev_t dev, char *name)
5666 {
5667 	/*
5668 	 * If dev is zero, name is the name of a device to allocate with
5669 	 * an arbitrary minor number.  It will be "md_???"
5670 	 * If dev is non-zero it must be a device number with a MAJOR of
5671 	 * MD_MAJOR or mdp_major.  In this case, if "name" is NULL, then
5672 	 * the device is being created by opening a node in /dev.
5673 	 * If "name" is not NULL, the device is being created by
5674 	 * writing to /sys/module/md_mod/parameters/new_array.
5675 	 */
5676 	static DEFINE_MUTEX(disks_mutex);
5677 	struct mddev *mddev;
5678 	struct gendisk *disk;
5679 	int partitioned;
5680 	int shift;
5681 	int unit;
5682 	int error ;
5683 
5684 	/*
5685 	 * Wait for any previous instance of this device to be completely
5686 	 * removed (mddev_delayed_delete).
5687 	 */
5688 	flush_workqueue(md_misc_wq);
5689 
5690 	mutex_lock(&disks_mutex);
5691 	mddev = mddev_alloc(dev);
5692 	if (IS_ERR(mddev)) {
5693 		error = PTR_ERR(mddev);
5694 		goto out_unlock;
5695 	}
5696 
5697 	partitioned = (MAJOR(mddev->unit) != MD_MAJOR);
5698 	shift = partitioned ? MdpMinorShift : 0;
5699 	unit = MINOR(mddev->unit) >> shift;
5700 
5701 	if (name && !dev) {
5702 		/* Need to ensure that 'name' is not a duplicate.
5703 		 */
5704 		struct mddev *mddev2;
5705 		spin_lock(&all_mddevs_lock);
5706 
5707 		list_for_each_entry(mddev2, &all_mddevs, all_mddevs)
5708 			if (mddev2->gendisk &&
5709 			    strcmp(mddev2->gendisk->disk_name, name) == 0) {
5710 				spin_unlock(&all_mddevs_lock);
5711 				error = -EEXIST;
5712 				goto out_free_mddev;
5713 			}
5714 		spin_unlock(&all_mddevs_lock);
5715 	}
5716 	if (name && dev)
5717 		/*
5718 		 * Creating /dev/mdNNN via "newarray", so adjust hold_active.
5719 		 */
5720 		mddev->hold_active = UNTIL_STOP;
5721 
5722 	error = -ENOMEM;
5723 	disk = blk_alloc_disk(NUMA_NO_NODE);
5724 	if (!disk)
5725 		goto out_free_mddev;
5726 
5727 	disk->major = MAJOR(mddev->unit);
5728 	disk->first_minor = unit << shift;
5729 	disk->minors = 1 << shift;
5730 	if (name)
5731 		strcpy(disk->disk_name, name);
5732 	else if (partitioned)
5733 		sprintf(disk->disk_name, "md_d%d", unit);
5734 	else
5735 		sprintf(disk->disk_name, "md%d", unit);
5736 	disk->fops = &md_fops;
5737 	disk->private_data = mddev;
5738 
5739 	mddev->queue = disk->queue;
5740 	blk_set_stacking_limits(&mddev->queue->limits);
5741 	blk_queue_write_cache(mddev->queue, true, true);
5742 	disk->events |= DISK_EVENT_MEDIA_CHANGE;
5743 	mddev->gendisk = disk;
5744 	error = add_disk(disk);
5745 	if (error)
5746 		goto out_put_disk;
5747 
5748 	kobject_init(&mddev->kobj, &md_ktype);
5749 	error = kobject_add(&mddev->kobj, &disk_to_dev(disk)->kobj, "%s", "md");
5750 	if (error) {
5751 		/*
5752 		 * The disk is already live at this point.  Clear the hold flag
5753 		 * and let mddev_put take care of the deletion, as it isn't any
5754 		 * different from a normal close on last release now.
5755 		 */
5756 		mddev->hold_active = 0;
5757 		mutex_unlock(&disks_mutex);
5758 		mddev_put(mddev);
5759 		return ERR_PTR(error);
5760 	}
5761 
5762 	kobject_uevent(&mddev->kobj, KOBJ_ADD);
5763 	mddev->sysfs_state = sysfs_get_dirent_safe(mddev->kobj.sd, "array_state");
5764 	mddev->sysfs_level = sysfs_get_dirent_safe(mddev->kobj.sd, "level");
5765 	mutex_unlock(&disks_mutex);
5766 	return mddev;
5767 
5768 out_put_disk:
5769 	put_disk(disk);
5770 out_free_mddev:
5771 	mddev_free(mddev);
5772 out_unlock:
5773 	mutex_unlock(&disks_mutex);
5774 	return ERR_PTR(error);
5775 }
5776 
5777 static int md_alloc_and_put(dev_t dev, char *name)
5778 {
5779 	struct mddev *mddev = md_alloc(dev, name);
5780 
5781 	if (IS_ERR(mddev))
5782 		return PTR_ERR(mddev);
5783 	mddev_put(mddev);
5784 	return 0;
5785 }
5786 
5787 static void md_probe(dev_t dev)
5788 {
5789 	if (MAJOR(dev) == MD_MAJOR && MINOR(dev) >= 512)
5790 		return;
5791 	if (create_on_open)
5792 		md_alloc_and_put(dev, NULL);
5793 }
5794 
5795 static int add_named_array(const char *val, const struct kernel_param *kp)
5796 {
5797 	/*
5798 	 * val must be "md_*" or "mdNNN".
5799 	 * For "md_*" we allocate an array with a large free minor number, and
5800 	 * set the name to val.  val must not already be an active name.
5801 	 * For "mdNNN" we allocate an array with the minor number NNN
5802 	 * which must not already be in use.
5803 	 */
5804 	int len = strlen(val);
5805 	char buf[DISK_NAME_LEN];
5806 	unsigned long devnum;
5807 
5808 	while (len && val[len-1] == '\n')
5809 		len--;
5810 	if (len >= DISK_NAME_LEN)
5811 		return -E2BIG;
5812 	strscpy(buf, val, len+1);
5813 	if (strncmp(buf, "md_", 3) == 0)
5814 		return md_alloc_and_put(0, buf);
5815 	if (strncmp(buf, "md", 2) == 0 &&
5816 	    isdigit(buf[2]) &&
5817 	    kstrtoul(buf+2, 10, &devnum) == 0 &&
5818 	    devnum <= MINORMASK)
5819 		return md_alloc_and_put(MKDEV(MD_MAJOR, devnum), NULL);
5820 
5821 	return -EINVAL;
5822 }
5823 
5824 static void md_safemode_timeout(struct timer_list *t)
5825 {
5826 	struct mddev *mddev = from_timer(mddev, t, safemode_timer);
5827 
5828 	mddev->safemode = 1;
5829 	if (mddev->external)
5830 		sysfs_notify_dirent_safe(mddev->sysfs_state);
5831 
5832 	md_wakeup_thread(mddev->thread);
5833 }
5834 
5835 static int start_dirty_degraded;
5836 static void active_io_release(struct percpu_ref *ref)
5837 {
5838 	struct mddev *mddev = container_of(ref, struct mddev, active_io);
5839 
5840 	wake_up(&mddev->sb_wait);
5841 }
5842 
5843 int md_run(struct mddev *mddev)
5844 {
5845 	int err;
5846 	struct md_rdev *rdev;
5847 	struct md_personality *pers;
5848 	bool nowait = true;
5849 
5850 	if (list_empty(&mddev->disks))
5851 		/* cannot run an array with no devices.. */
5852 		return -EINVAL;
5853 
5854 	if (mddev->pers)
5855 		return -EBUSY;
5856 	/* Cannot run until previous stop completes properly */
5857 	if (mddev->sysfs_active)
5858 		return -EBUSY;
5859 
5860 	/*
5861 	 * Analyze all RAID superblock(s)
5862 	 */
5863 	if (!mddev->raid_disks) {
5864 		if (!mddev->persistent)
5865 			return -EINVAL;
5866 		err = analyze_sbs(mddev);
5867 		if (err)
5868 			return -EINVAL;
5869 	}
5870 
5871 	if (mddev->level != LEVEL_NONE)
5872 		request_module("md-level-%d", mddev->level);
5873 	else if (mddev->clevel[0])
5874 		request_module("md-%s", mddev->clevel);
5875 
5876 	/*
5877 	 * Drop all container device buffers, from now on
5878 	 * the only valid external interface is through the md
5879 	 * device.
5880 	 */
5881 	mddev->has_superblocks = false;
5882 	rdev_for_each(rdev, mddev) {
5883 		if (test_bit(Faulty, &rdev->flags))
5884 			continue;
5885 		sync_blockdev(rdev->bdev);
5886 		invalidate_bdev(rdev->bdev);
5887 		if (mddev->ro != MD_RDONLY && rdev_read_only(rdev)) {
5888 			mddev->ro = MD_RDONLY;
5889 			if (mddev->gendisk)
5890 				set_disk_ro(mddev->gendisk, 1);
5891 		}
5892 
5893 		if (rdev->sb_page)
5894 			mddev->has_superblocks = true;
5895 
5896 		/* perform some consistency tests on the device.
5897 		 * We don't want the data to overlap the metadata,
5898 		 * Internal Bitmap issues have been handled elsewhere.
5899 		 */
5900 		if (rdev->meta_bdev) {
5901 			/* Nothing to check */;
5902 		} else if (rdev->data_offset < rdev->sb_start) {
5903 			if (mddev->dev_sectors &&
5904 			    rdev->data_offset + mddev->dev_sectors
5905 			    > rdev->sb_start) {
5906 				pr_warn("md: %s: data overlaps metadata\n",
5907 					mdname(mddev));
5908 				return -EINVAL;
5909 			}
5910 		} else {
5911 			if (rdev->sb_start + rdev->sb_size/512
5912 			    > rdev->data_offset) {
5913 				pr_warn("md: %s: metadata overlaps data\n",
5914 					mdname(mddev));
5915 				return -EINVAL;
5916 			}
5917 		}
5918 		sysfs_notify_dirent_safe(rdev->sysfs_state);
5919 		nowait = nowait && bdev_nowait(rdev->bdev);
5920 	}
5921 
5922 	err = percpu_ref_init(&mddev->active_io, active_io_release,
5923 				PERCPU_REF_ALLOW_REINIT, GFP_KERNEL);
5924 	if (err)
5925 		return err;
5926 
5927 	if (!bioset_initialized(&mddev->bio_set)) {
5928 		err = bioset_init(&mddev->bio_set, BIO_POOL_SIZE, 0, BIOSET_NEED_BVECS);
5929 		if (err)
5930 			goto exit_active_io;
5931 	}
5932 	if (!bioset_initialized(&mddev->sync_set)) {
5933 		err = bioset_init(&mddev->sync_set, BIO_POOL_SIZE, 0, BIOSET_NEED_BVECS);
5934 		if (err)
5935 			goto exit_bio_set;
5936 	}
5937 
5938 	if (!bioset_initialized(&mddev->io_clone_set)) {
5939 		err = bioset_init(&mddev->io_clone_set, BIO_POOL_SIZE,
5940 				  offsetof(struct md_io_clone, bio_clone), 0);
5941 		if (err)
5942 			goto exit_sync_set;
5943 	}
5944 
5945 	spin_lock(&pers_lock);
5946 	pers = find_pers(mddev->level, mddev->clevel);
5947 	if (!pers || !try_module_get(pers->owner)) {
5948 		spin_unlock(&pers_lock);
5949 		if (mddev->level != LEVEL_NONE)
5950 			pr_warn("md: personality for level %d is not loaded!\n",
5951 				mddev->level);
5952 		else
5953 			pr_warn("md: personality for level %s is not loaded!\n",
5954 				mddev->clevel);
5955 		err = -EINVAL;
5956 		goto abort;
5957 	}
5958 	spin_unlock(&pers_lock);
5959 	if (mddev->level != pers->level) {
5960 		mddev->level = pers->level;
5961 		mddev->new_level = pers->level;
5962 	}
5963 	strscpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
5964 
5965 	if (mddev->reshape_position != MaxSector &&
5966 	    pers->start_reshape == NULL) {
5967 		/* This personality cannot handle reshaping... */
5968 		module_put(pers->owner);
5969 		err = -EINVAL;
5970 		goto abort;
5971 	}
5972 
5973 	if (pers->sync_request) {
5974 		/* Warn if this is a potentially silly
5975 		 * configuration.
5976 		 */
5977 		struct md_rdev *rdev2;
5978 		int warned = 0;
5979 
5980 		rdev_for_each(rdev, mddev)
5981 			rdev_for_each(rdev2, mddev) {
5982 				if (rdev < rdev2 &&
5983 				    rdev->bdev->bd_disk ==
5984 				    rdev2->bdev->bd_disk) {
5985 					pr_warn("%s: WARNING: %pg appears to be on the same physical disk as %pg.\n",
5986 						mdname(mddev),
5987 						rdev->bdev,
5988 						rdev2->bdev);
5989 					warned = 1;
5990 				}
5991 			}
5992 
5993 		if (warned)
5994 			pr_warn("True protection against single-disk failure might be compromised.\n");
5995 	}
5996 
5997 	mddev->recovery = 0;
5998 	/* may be over-ridden by personality */
5999 	mddev->resync_max_sectors = mddev->dev_sectors;
6000 
6001 	mddev->ok_start_degraded = start_dirty_degraded;
6002 
6003 	if (start_readonly && md_is_rdwr(mddev))
6004 		mddev->ro = MD_AUTO_READ; /* read-only, but switch on first write */
6005 
6006 	err = pers->run(mddev);
6007 	if (err)
6008 		pr_warn("md: pers->run() failed ...\n");
6009 	else if (pers->size(mddev, 0, 0) < mddev->array_sectors) {
6010 		WARN_ONCE(!mddev->external_size,
6011 			  "%s: default size too small, but 'external_size' not in effect?\n",
6012 			  __func__);
6013 		pr_warn("md: invalid array_size %llu > default size %llu\n",
6014 			(unsigned long long)mddev->array_sectors / 2,
6015 			(unsigned long long)pers->size(mddev, 0, 0) / 2);
6016 		err = -EINVAL;
6017 	}
6018 	if (err == 0 && pers->sync_request &&
6019 	    (mddev->bitmap_info.file || mddev->bitmap_info.offset)) {
6020 		struct bitmap *bitmap;
6021 
6022 		bitmap = md_bitmap_create(mddev, -1);
6023 		if (IS_ERR(bitmap)) {
6024 			err = PTR_ERR(bitmap);
6025 			pr_warn("%s: failed to create bitmap (%d)\n",
6026 				mdname(mddev), err);
6027 		} else
6028 			mddev->bitmap = bitmap;
6029 
6030 	}
6031 	if (err)
6032 		goto bitmap_abort;
6033 
6034 	if (mddev->bitmap_info.max_write_behind > 0) {
6035 		bool create_pool = false;
6036 
6037 		rdev_for_each(rdev, mddev) {
6038 			if (test_bit(WriteMostly, &rdev->flags) &&
6039 			    rdev_init_serial(rdev))
6040 				create_pool = true;
6041 		}
6042 		if (create_pool && mddev->serial_info_pool == NULL) {
6043 			mddev->serial_info_pool =
6044 				mempool_create_kmalloc_pool(NR_SERIAL_INFOS,
6045 						    sizeof(struct serial_info));
6046 			if (!mddev->serial_info_pool) {
6047 				err = -ENOMEM;
6048 				goto bitmap_abort;
6049 			}
6050 		}
6051 	}
6052 
6053 	if (mddev->queue) {
6054 		bool nonrot = true;
6055 
6056 		rdev_for_each(rdev, mddev) {
6057 			if (rdev->raid_disk >= 0 && !bdev_nonrot(rdev->bdev)) {
6058 				nonrot = false;
6059 				break;
6060 			}
6061 		}
6062 		if (mddev->degraded)
6063 			nonrot = false;
6064 		if (nonrot)
6065 			blk_queue_flag_set(QUEUE_FLAG_NONROT, mddev->queue);
6066 		else
6067 			blk_queue_flag_clear(QUEUE_FLAG_NONROT, mddev->queue);
6068 		blk_queue_flag_set(QUEUE_FLAG_IO_STAT, mddev->queue);
6069 
6070 		/* Set the NOWAIT flags if all underlying devices support it */
6071 		if (nowait)
6072 			blk_queue_flag_set(QUEUE_FLAG_NOWAIT, mddev->queue);
6073 	}
6074 	if (pers->sync_request) {
6075 		if (mddev->kobj.sd &&
6076 		    sysfs_create_group(&mddev->kobj, &md_redundancy_group))
6077 			pr_warn("md: cannot register extra attributes for %s\n",
6078 				mdname(mddev));
6079 		mddev->sysfs_action = sysfs_get_dirent_safe(mddev->kobj.sd, "sync_action");
6080 		mddev->sysfs_completed = sysfs_get_dirent_safe(mddev->kobj.sd, "sync_completed");
6081 		mddev->sysfs_degraded = sysfs_get_dirent_safe(mddev->kobj.sd, "degraded");
6082 	} else if (mddev->ro == MD_AUTO_READ)
6083 		mddev->ro = MD_RDWR;
6084 
6085 	atomic_set(&mddev->max_corr_read_errors,
6086 		   MD_DEFAULT_MAX_CORRECTED_READ_ERRORS);
6087 	mddev->safemode = 0;
6088 	if (mddev_is_clustered(mddev))
6089 		mddev->safemode_delay = 0;
6090 	else
6091 		mddev->safemode_delay = DEFAULT_SAFEMODE_DELAY;
6092 	mddev->in_sync = 1;
6093 	smp_wmb();
6094 	spin_lock(&mddev->lock);
6095 	mddev->pers = pers;
6096 	spin_unlock(&mddev->lock);
6097 	rdev_for_each(rdev, mddev)
6098 		if (rdev->raid_disk >= 0)
6099 			sysfs_link_rdev(mddev, rdev); /* failure here is OK */
6100 
6101 	if (mddev->degraded && md_is_rdwr(mddev))
6102 		/* This ensures that recovering status is reported immediately
6103 		 * via sysfs - until a lack of spares is confirmed.
6104 		 */
6105 		set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
6106 	set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6107 
6108 	if (mddev->sb_flags)
6109 		md_update_sb(mddev, 0);
6110 
6111 	md_new_event();
6112 	return 0;
6113 
6114 bitmap_abort:
6115 	mddev_detach(mddev);
6116 	if (mddev->private)
6117 		pers->free(mddev, mddev->private);
6118 	mddev->private = NULL;
6119 	module_put(pers->owner);
6120 	md_bitmap_destroy(mddev);
6121 abort:
6122 	bioset_exit(&mddev->io_clone_set);
6123 exit_sync_set:
6124 	bioset_exit(&mddev->sync_set);
6125 exit_bio_set:
6126 	bioset_exit(&mddev->bio_set);
6127 exit_active_io:
6128 	percpu_ref_exit(&mddev->active_io);
6129 	return err;
6130 }
6131 EXPORT_SYMBOL_GPL(md_run);
6132 
6133 int do_md_run(struct mddev *mddev)
6134 {
6135 	int err;
6136 
6137 	set_bit(MD_NOT_READY, &mddev->flags);
6138 	err = md_run(mddev);
6139 	if (err)
6140 		goto out;
6141 	err = md_bitmap_load(mddev);
6142 	if (err) {
6143 		md_bitmap_destroy(mddev);
6144 		goto out;
6145 	}
6146 
6147 	if (mddev_is_clustered(mddev))
6148 		md_allow_write(mddev);
6149 
6150 	/* run start up tasks that require md_thread */
6151 	md_start(mddev);
6152 
6153 	md_wakeup_thread(mddev->thread);
6154 	md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
6155 
6156 	set_capacity_and_notify(mddev->gendisk, mddev->array_sectors);
6157 	clear_bit(MD_NOT_READY, &mddev->flags);
6158 	mddev->changed = 1;
6159 	kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
6160 	sysfs_notify_dirent_safe(mddev->sysfs_state);
6161 	sysfs_notify_dirent_safe(mddev->sysfs_action);
6162 	sysfs_notify_dirent_safe(mddev->sysfs_degraded);
6163 out:
6164 	clear_bit(MD_NOT_READY, &mddev->flags);
6165 	return err;
6166 }
6167 
6168 int md_start(struct mddev *mddev)
6169 {
6170 	int ret = 0;
6171 
6172 	if (mddev->pers->start) {
6173 		set_bit(MD_RECOVERY_WAIT, &mddev->recovery);
6174 		md_wakeup_thread(mddev->thread);
6175 		ret = mddev->pers->start(mddev);
6176 		clear_bit(MD_RECOVERY_WAIT, &mddev->recovery);
6177 		md_wakeup_thread(mddev->sync_thread);
6178 	}
6179 	return ret;
6180 }
6181 EXPORT_SYMBOL_GPL(md_start);
6182 
6183 static int restart_array(struct mddev *mddev)
6184 {
6185 	struct gendisk *disk = mddev->gendisk;
6186 	struct md_rdev *rdev;
6187 	bool has_journal = false;
6188 	bool has_readonly = false;
6189 
6190 	/* Complain if it has no devices */
6191 	if (list_empty(&mddev->disks))
6192 		return -ENXIO;
6193 	if (!mddev->pers)
6194 		return -EINVAL;
6195 	if (md_is_rdwr(mddev))
6196 		return -EBUSY;
6197 
6198 	rcu_read_lock();
6199 	rdev_for_each_rcu(rdev, mddev) {
6200 		if (test_bit(Journal, &rdev->flags) &&
6201 		    !test_bit(Faulty, &rdev->flags))
6202 			has_journal = true;
6203 		if (rdev_read_only(rdev))
6204 			has_readonly = true;
6205 	}
6206 	rcu_read_unlock();
6207 	if (test_bit(MD_HAS_JOURNAL, &mddev->flags) && !has_journal)
6208 		/* Don't restart rw with journal missing/faulty */
6209 			return -EINVAL;
6210 	if (has_readonly)
6211 		return -EROFS;
6212 
6213 	mddev->safemode = 0;
6214 	mddev->ro = MD_RDWR;
6215 	set_disk_ro(disk, 0);
6216 	pr_debug("md: %s switched to read-write mode.\n", mdname(mddev));
6217 	/* Kick recovery or resync if necessary */
6218 	set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6219 	md_wakeup_thread(mddev->thread);
6220 	md_wakeup_thread(mddev->sync_thread);
6221 	sysfs_notify_dirent_safe(mddev->sysfs_state);
6222 	return 0;
6223 }
6224 
6225 static void md_clean(struct mddev *mddev)
6226 {
6227 	mddev->array_sectors = 0;
6228 	mddev->external_size = 0;
6229 	mddev->dev_sectors = 0;
6230 	mddev->raid_disks = 0;
6231 	mddev->recovery_cp = 0;
6232 	mddev->resync_min = 0;
6233 	mddev->resync_max = MaxSector;
6234 	mddev->reshape_position = MaxSector;
6235 	/* we still need mddev->external in export_rdev, do not clear it yet */
6236 	mddev->persistent = 0;
6237 	mddev->level = LEVEL_NONE;
6238 	mddev->clevel[0] = 0;
6239 	/*
6240 	 * Don't clear MD_CLOSING, or mddev can be opened again.
6241 	 * 'hold_active != 0' means mddev is still in the creation
6242 	 * process and will be used later.
6243 	 */
6244 	if (mddev->hold_active)
6245 		mddev->flags = 0;
6246 	else
6247 		mddev->flags &= BIT_ULL_MASK(MD_CLOSING);
6248 	mddev->sb_flags = 0;
6249 	mddev->ro = MD_RDWR;
6250 	mddev->metadata_type[0] = 0;
6251 	mddev->chunk_sectors = 0;
6252 	mddev->ctime = mddev->utime = 0;
6253 	mddev->layout = 0;
6254 	mddev->max_disks = 0;
6255 	mddev->events = 0;
6256 	mddev->can_decrease_events = 0;
6257 	mddev->delta_disks = 0;
6258 	mddev->reshape_backwards = 0;
6259 	mddev->new_level = LEVEL_NONE;
6260 	mddev->new_layout = 0;
6261 	mddev->new_chunk_sectors = 0;
6262 	mddev->curr_resync = MD_RESYNC_NONE;
6263 	atomic64_set(&mddev->resync_mismatches, 0);
6264 	mddev->suspend_lo = mddev->suspend_hi = 0;
6265 	mddev->sync_speed_min = mddev->sync_speed_max = 0;
6266 	mddev->recovery = 0;
6267 	mddev->in_sync = 0;
6268 	mddev->changed = 0;
6269 	mddev->degraded = 0;
6270 	mddev->safemode = 0;
6271 	mddev->private = NULL;
6272 	mddev->cluster_info = NULL;
6273 	mddev->bitmap_info.offset = 0;
6274 	mddev->bitmap_info.default_offset = 0;
6275 	mddev->bitmap_info.default_space = 0;
6276 	mddev->bitmap_info.chunksize = 0;
6277 	mddev->bitmap_info.daemon_sleep = 0;
6278 	mddev->bitmap_info.max_write_behind = 0;
6279 	mddev->bitmap_info.nodes = 0;
6280 }
6281 
6282 static void __md_stop_writes(struct mddev *mddev)
6283 {
6284 	set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
6285 	if (work_pending(&mddev->del_work))
6286 		flush_workqueue(md_misc_wq);
6287 	if (mddev->sync_thread) {
6288 		set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6289 		md_reap_sync_thread(mddev);
6290 	}
6291 
6292 	del_timer_sync(&mddev->safemode_timer);
6293 
6294 	if (mddev->pers && mddev->pers->quiesce) {
6295 		mddev->pers->quiesce(mddev, 1);
6296 		mddev->pers->quiesce(mddev, 0);
6297 	}
6298 	md_bitmap_flush(mddev);
6299 
6300 	if (md_is_rdwr(mddev) &&
6301 	    ((!mddev->in_sync && !mddev_is_clustered(mddev)) ||
6302 	     mddev->sb_flags)) {
6303 		/* mark array as shutdown cleanly */
6304 		if (!mddev_is_clustered(mddev))
6305 			mddev->in_sync = 1;
6306 		md_update_sb(mddev, 1);
6307 	}
6308 	/* disable policy to guarantee rdevs free resources for serialization */
6309 	mddev->serialize_policy = 0;
6310 	mddev_destroy_serial_pool(mddev, NULL, true);
6311 }
6312 
6313 void md_stop_writes(struct mddev *mddev)
6314 {
6315 	mddev_lock_nointr(mddev);
6316 	__md_stop_writes(mddev);
6317 	mddev_unlock(mddev);
6318 }
6319 EXPORT_SYMBOL_GPL(md_stop_writes);
6320 
6321 static void mddev_detach(struct mddev *mddev)
6322 {
6323 	md_bitmap_wait_behind_writes(mddev);
6324 	if (mddev->pers && mddev->pers->quiesce && !is_md_suspended(mddev)) {
6325 		mddev->pers->quiesce(mddev, 1);
6326 		mddev->pers->quiesce(mddev, 0);
6327 	}
6328 	md_unregister_thread(mddev, &mddev->thread);
6329 	if (mddev->queue)
6330 		blk_sync_queue(mddev->queue); /* the unplug fn references 'conf'*/
6331 }
6332 
6333 static void __md_stop(struct mddev *mddev)
6334 {
6335 	struct md_personality *pers = mddev->pers;
6336 	md_bitmap_destroy(mddev);
6337 	mddev_detach(mddev);
6338 	/* Ensure ->event_work is done */
6339 	if (mddev->event_work.func)
6340 		flush_workqueue(md_misc_wq);
6341 	spin_lock(&mddev->lock);
6342 	mddev->pers = NULL;
6343 	spin_unlock(&mddev->lock);
6344 	if (mddev->private)
6345 		pers->free(mddev, mddev->private);
6346 	mddev->private = NULL;
6347 	if (pers->sync_request && mddev->to_remove == NULL)
6348 		mddev->to_remove = &md_redundancy_group;
6349 	module_put(pers->owner);
6350 	clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
6351 
6352 	percpu_ref_exit(&mddev->active_io);
6353 	bioset_exit(&mddev->bio_set);
6354 	bioset_exit(&mddev->sync_set);
6355 	bioset_exit(&mddev->io_clone_set);
6356 }
6357 
6358 void md_stop(struct mddev *mddev)
6359 {
6360 	lockdep_assert_held(&mddev->reconfig_mutex);
6361 
6362 	/* stop the array and free an attached data structures.
6363 	 * This is called from dm-raid
6364 	 */
6365 	__md_stop_writes(mddev);
6366 	__md_stop(mddev);
6367 	percpu_ref_exit(&mddev->writes_pending);
6368 }
6369 
6370 EXPORT_SYMBOL_GPL(md_stop);
6371 
6372 static int md_set_readonly(struct mddev *mddev, struct block_device *bdev)
6373 {
6374 	int err = 0;
6375 	int did_freeze = 0;
6376 
6377 	if (mddev->external && test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags))
6378 		return -EBUSY;
6379 
6380 	if (!test_bit(MD_RECOVERY_FROZEN, &mddev->recovery)) {
6381 		did_freeze = 1;
6382 		set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
6383 		md_wakeup_thread(mddev->thread);
6384 	}
6385 	if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
6386 		set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6387 
6388 	/*
6389 	 * Thread might be blocked waiting for metadata update which will now
6390 	 * never happen
6391 	 */
6392 	md_wakeup_thread_directly(mddev->sync_thread);
6393 
6394 	mddev_unlock(mddev);
6395 	wait_event(resync_wait, !test_bit(MD_RECOVERY_RUNNING,
6396 					  &mddev->recovery));
6397 	wait_event(mddev->sb_wait,
6398 		   !test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags));
6399 	mddev_lock_nointr(mddev);
6400 
6401 	mutex_lock(&mddev->open_mutex);
6402 	if ((mddev->pers && atomic_read(&mddev->openers) > !!bdev) ||
6403 	    mddev->sync_thread ||
6404 	    test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) {
6405 		pr_warn("md: %s still in use.\n",mdname(mddev));
6406 		err = -EBUSY;
6407 		goto out;
6408 	}
6409 
6410 	if (mddev->pers) {
6411 		__md_stop_writes(mddev);
6412 
6413 		if (mddev->ro == MD_RDONLY) {
6414 			err  = -ENXIO;
6415 			goto out;
6416 		}
6417 
6418 		mddev->ro = MD_RDONLY;
6419 		set_disk_ro(mddev->gendisk, 1);
6420 	}
6421 
6422 out:
6423 	if ((mddev->pers && !err) || did_freeze) {
6424 		clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
6425 		set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6426 		md_wakeup_thread(mddev->thread);
6427 		sysfs_notify_dirent_safe(mddev->sysfs_state);
6428 	}
6429 
6430 	mutex_unlock(&mddev->open_mutex);
6431 	return err;
6432 }
6433 
6434 /* mode:
6435  *   0 - completely stop and dis-assemble array
6436  *   2 - stop but do not disassemble array
6437  */
6438 static int do_md_stop(struct mddev *mddev, int mode,
6439 		      struct block_device *bdev)
6440 {
6441 	struct gendisk *disk = mddev->gendisk;
6442 	struct md_rdev *rdev;
6443 	int did_freeze = 0;
6444 
6445 	if (!test_bit(MD_RECOVERY_FROZEN, &mddev->recovery)) {
6446 		did_freeze = 1;
6447 		set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
6448 		md_wakeup_thread(mddev->thread);
6449 	}
6450 	if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
6451 		set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6452 
6453 	/*
6454 	 * Thread might be blocked waiting for metadata update which will now
6455 	 * never happen
6456 	 */
6457 	md_wakeup_thread_directly(mddev->sync_thread);
6458 
6459 	mddev_unlock(mddev);
6460 	wait_event(resync_wait, (mddev->sync_thread == NULL &&
6461 				 !test_bit(MD_RECOVERY_RUNNING,
6462 					   &mddev->recovery)));
6463 	mddev_lock_nointr(mddev);
6464 
6465 	mutex_lock(&mddev->open_mutex);
6466 	if ((mddev->pers && atomic_read(&mddev->openers) > !!bdev) ||
6467 	    mddev->sysfs_active ||
6468 	    mddev->sync_thread ||
6469 	    test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) {
6470 		pr_warn("md: %s still in use.\n",mdname(mddev));
6471 		mutex_unlock(&mddev->open_mutex);
6472 		if (did_freeze) {
6473 			clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
6474 			set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6475 			md_wakeup_thread(mddev->thread);
6476 		}
6477 		return -EBUSY;
6478 	}
6479 	if (mddev->pers) {
6480 		if (!md_is_rdwr(mddev))
6481 			set_disk_ro(disk, 0);
6482 
6483 		__md_stop_writes(mddev);
6484 		__md_stop(mddev);
6485 
6486 		/* tell userspace to handle 'inactive' */
6487 		sysfs_notify_dirent_safe(mddev->sysfs_state);
6488 
6489 		rdev_for_each(rdev, mddev)
6490 			if (rdev->raid_disk >= 0)
6491 				sysfs_unlink_rdev(mddev, rdev);
6492 
6493 		set_capacity_and_notify(disk, 0);
6494 		mutex_unlock(&mddev->open_mutex);
6495 		mddev->changed = 1;
6496 
6497 		if (!md_is_rdwr(mddev))
6498 			mddev->ro = MD_RDWR;
6499 	} else
6500 		mutex_unlock(&mddev->open_mutex);
6501 	/*
6502 	 * Free resources if final stop
6503 	 */
6504 	if (mode == 0) {
6505 		pr_info("md: %s stopped.\n", mdname(mddev));
6506 
6507 		if (mddev->bitmap_info.file) {
6508 			struct file *f = mddev->bitmap_info.file;
6509 			spin_lock(&mddev->lock);
6510 			mddev->bitmap_info.file = NULL;
6511 			spin_unlock(&mddev->lock);
6512 			fput(f);
6513 		}
6514 		mddev->bitmap_info.offset = 0;
6515 
6516 		export_array(mddev);
6517 
6518 		md_clean(mddev);
6519 		if (mddev->hold_active == UNTIL_STOP)
6520 			mddev->hold_active = 0;
6521 	}
6522 	md_new_event();
6523 	sysfs_notify_dirent_safe(mddev->sysfs_state);
6524 	return 0;
6525 }
6526 
6527 #ifndef MODULE
6528 static void autorun_array(struct mddev *mddev)
6529 {
6530 	struct md_rdev *rdev;
6531 	int err;
6532 
6533 	if (list_empty(&mddev->disks))
6534 		return;
6535 
6536 	pr_info("md: running: ");
6537 
6538 	rdev_for_each(rdev, mddev) {
6539 		pr_cont("<%pg>", rdev->bdev);
6540 	}
6541 	pr_cont("\n");
6542 
6543 	err = do_md_run(mddev);
6544 	if (err) {
6545 		pr_warn("md: do_md_run() returned %d\n", err);
6546 		do_md_stop(mddev, 0, NULL);
6547 	}
6548 }
6549 
6550 /*
6551  * lets try to run arrays based on all disks that have arrived
6552  * until now. (those are in pending_raid_disks)
6553  *
6554  * the method: pick the first pending disk, collect all disks with
6555  * the same UUID, remove all from the pending list and put them into
6556  * the 'same_array' list. Then order this list based on superblock
6557  * update time (freshest comes first), kick out 'old' disks and
6558  * compare superblocks. If everything's fine then run it.
6559  *
6560  * If "unit" is allocated, then bump its reference count
6561  */
6562 static void autorun_devices(int part)
6563 {
6564 	struct md_rdev *rdev0, *rdev, *tmp;
6565 	struct mddev *mddev;
6566 
6567 	pr_info("md: autorun ...\n");
6568 	while (!list_empty(&pending_raid_disks)) {
6569 		int unit;
6570 		dev_t dev;
6571 		LIST_HEAD(candidates);
6572 		rdev0 = list_entry(pending_raid_disks.next,
6573 					 struct md_rdev, same_set);
6574 
6575 		pr_debug("md: considering %pg ...\n", rdev0->bdev);
6576 		INIT_LIST_HEAD(&candidates);
6577 		rdev_for_each_list(rdev, tmp, &pending_raid_disks)
6578 			if (super_90_load(rdev, rdev0, 0) >= 0) {
6579 				pr_debug("md:  adding %pg ...\n",
6580 					 rdev->bdev);
6581 				list_move(&rdev->same_set, &candidates);
6582 			}
6583 		/*
6584 		 * now we have a set of devices, with all of them having
6585 		 * mostly sane superblocks. It's time to allocate the
6586 		 * mddev.
6587 		 */
6588 		if (part) {
6589 			dev = MKDEV(mdp_major,
6590 				    rdev0->preferred_minor << MdpMinorShift);
6591 			unit = MINOR(dev) >> MdpMinorShift;
6592 		} else {
6593 			dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
6594 			unit = MINOR(dev);
6595 		}
6596 		if (rdev0->preferred_minor != unit) {
6597 			pr_warn("md: unit number in %pg is bad: %d\n",
6598 				rdev0->bdev, rdev0->preferred_minor);
6599 			break;
6600 		}
6601 
6602 		mddev = md_alloc(dev, NULL);
6603 		if (IS_ERR(mddev))
6604 			break;
6605 
6606 		if (mddev_lock(mddev))
6607 			pr_warn("md: %s locked, cannot run\n", mdname(mddev));
6608 		else if (mddev->raid_disks || mddev->major_version
6609 			 || !list_empty(&mddev->disks)) {
6610 			pr_warn("md: %s already running, cannot run %pg\n",
6611 				mdname(mddev), rdev0->bdev);
6612 			mddev_unlock(mddev);
6613 		} else {
6614 			pr_debug("md: created %s\n", mdname(mddev));
6615 			mddev->persistent = 1;
6616 			rdev_for_each_list(rdev, tmp, &candidates) {
6617 				list_del_init(&rdev->same_set);
6618 				if (bind_rdev_to_array(rdev, mddev))
6619 					export_rdev(rdev, mddev);
6620 			}
6621 			autorun_array(mddev);
6622 			mddev_unlock(mddev);
6623 		}
6624 		/* on success, candidates will be empty, on error
6625 		 * it won't...
6626 		 */
6627 		rdev_for_each_list(rdev, tmp, &candidates) {
6628 			list_del_init(&rdev->same_set);
6629 			export_rdev(rdev, mddev);
6630 		}
6631 		mddev_put(mddev);
6632 	}
6633 	pr_info("md: ... autorun DONE.\n");
6634 }
6635 #endif /* !MODULE */
6636 
6637 static int get_version(void __user *arg)
6638 {
6639 	mdu_version_t ver;
6640 
6641 	ver.major = MD_MAJOR_VERSION;
6642 	ver.minor = MD_MINOR_VERSION;
6643 	ver.patchlevel = MD_PATCHLEVEL_VERSION;
6644 
6645 	if (copy_to_user(arg, &ver, sizeof(ver)))
6646 		return -EFAULT;
6647 
6648 	return 0;
6649 }
6650 
6651 static int get_array_info(struct mddev *mddev, void __user *arg)
6652 {
6653 	mdu_array_info_t info;
6654 	int nr,working,insync,failed,spare;
6655 	struct md_rdev *rdev;
6656 
6657 	nr = working = insync = failed = spare = 0;
6658 	rcu_read_lock();
6659 	rdev_for_each_rcu(rdev, mddev) {
6660 		nr++;
6661 		if (test_bit(Faulty, &rdev->flags))
6662 			failed++;
6663 		else {
6664 			working++;
6665 			if (test_bit(In_sync, &rdev->flags))
6666 				insync++;
6667 			else if (test_bit(Journal, &rdev->flags))
6668 				/* TODO: add journal count to md_u.h */
6669 				;
6670 			else
6671 				spare++;
6672 		}
6673 	}
6674 	rcu_read_unlock();
6675 
6676 	info.major_version = mddev->major_version;
6677 	info.minor_version = mddev->minor_version;
6678 	info.patch_version = MD_PATCHLEVEL_VERSION;
6679 	info.ctime         = clamp_t(time64_t, mddev->ctime, 0, U32_MAX);
6680 	info.level         = mddev->level;
6681 	info.size          = mddev->dev_sectors / 2;
6682 	if (info.size != mddev->dev_sectors / 2) /* overflow */
6683 		info.size = -1;
6684 	info.nr_disks      = nr;
6685 	info.raid_disks    = mddev->raid_disks;
6686 	info.md_minor      = mddev->md_minor;
6687 	info.not_persistent= !mddev->persistent;
6688 
6689 	info.utime         = clamp_t(time64_t, mddev->utime, 0, U32_MAX);
6690 	info.state         = 0;
6691 	if (mddev->in_sync)
6692 		info.state = (1<<MD_SB_CLEAN);
6693 	if (mddev->bitmap && mddev->bitmap_info.offset)
6694 		info.state |= (1<<MD_SB_BITMAP_PRESENT);
6695 	if (mddev_is_clustered(mddev))
6696 		info.state |= (1<<MD_SB_CLUSTERED);
6697 	info.active_disks  = insync;
6698 	info.working_disks = working;
6699 	info.failed_disks  = failed;
6700 	info.spare_disks   = spare;
6701 
6702 	info.layout        = mddev->layout;
6703 	info.chunk_size    = mddev->chunk_sectors << 9;
6704 
6705 	if (copy_to_user(arg, &info, sizeof(info)))
6706 		return -EFAULT;
6707 
6708 	return 0;
6709 }
6710 
6711 static int get_bitmap_file(struct mddev *mddev, void __user * arg)
6712 {
6713 	mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */
6714 	char *ptr;
6715 	int err;
6716 
6717 	file = kzalloc(sizeof(*file), GFP_NOIO);
6718 	if (!file)
6719 		return -ENOMEM;
6720 
6721 	err = 0;
6722 	spin_lock(&mddev->lock);
6723 	/* bitmap enabled */
6724 	if (mddev->bitmap_info.file) {
6725 		ptr = file_path(mddev->bitmap_info.file, file->pathname,
6726 				sizeof(file->pathname));
6727 		if (IS_ERR(ptr))
6728 			err = PTR_ERR(ptr);
6729 		else
6730 			memmove(file->pathname, ptr,
6731 				sizeof(file->pathname)-(ptr-file->pathname));
6732 	}
6733 	spin_unlock(&mddev->lock);
6734 
6735 	if (err == 0 &&
6736 	    copy_to_user(arg, file, sizeof(*file)))
6737 		err = -EFAULT;
6738 
6739 	kfree(file);
6740 	return err;
6741 }
6742 
6743 static int get_disk_info(struct mddev *mddev, void __user * arg)
6744 {
6745 	mdu_disk_info_t info;
6746 	struct md_rdev *rdev;
6747 
6748 	if (copy_from_user(&info, arg, sizeof(info)))
6749 		return -EFAULT;
6750 
6751 	rcu_read_lock();
6752 	rdev = md_find_rdev_nr_rcu(mddev, info.number);
6753 	if (rdev) {
6754 		info.major = MAJOR(rdev->bdev->bd_dev);
6755 		info.minor = MINOR(rdev->bdev->bd_dev);
6756 		info.raid_disk = rdev->raid_disk;
6757 		info.state = 0;
6758 		if (test_bit(Faulty, &rdev->flags))
6759 			info.state |= (1<<MD_DISK_FAULTY);
6760 		else if (test_bit(In_sync, &rdev->flags)) {
6761 			info.state |= (1<<MD_DISK_ACTIVE);
6762 			info.state |= (1<<MD_DISK_SYNC);
6763 		}
6764 		if (test_bit(Journal, &rdev->flags))
6765 			info.state |= (1<<MD_DISK_JOURNAL);
6766 		if (test_bit(WriteMostly, &rdev->flags))
6767 			info.state |= (1<<MD_DISK_WRITEMOSTLY);
6768 		if (test_bit(FailFast, &rdev->flags))
6769 			info.state |= (1<<MD_DISK_FAILFAST);
6770 	} else {
6771 		info.major = info.minor = 0;
6772 		info.raid_disk = -1;
6773 		info.state = (1<<MD_DISK_REMOVED);
6774 	}
6775 	rcu_read_unlock();
6776 
6777 	if (copy_to_user(arg, &info, sizeof(info)))
6778 		return -EFAULT;
6779 
6780 	return 0;
6781 }
6782 
6783 int md_add_new_disk(struct mddev *mddev, struct mdu_disk_info_s *info)
6784 {
6785 	struct md_rdev *rdev;
6786 	dev_t dev = MKDEV(info->major,info->minor);
6787 
6788 	if (mddev_is_clustered(mddev) &&
6789 		!(info->state & ((1 << MD_DISK_CLUSTER_ADD) | (1 << MD_DISK_CANDIDATE)))) {
6790 		pr_warn("%s: Cannot add to clustered mddev.\n",
6791 			mdname(mddev));
6792 		return -EINVAL;
6793 	}
6794 
6795 	if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
6796 		return -EOVERFLOW;
6797 
6798 	if (!mddev->raid_disks) {
6799 		int err;
6800 		/* expecting a device which has a superblock */
6801 		rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
6802 		if (IS_ERR(rdev)) {
6803 			pr_warn("md: md_import_device returned %ld\n",
6804 				PTR_ERR(rdev));
6805 			return PTR_ERR(rdev);
6806 		}
6807 		if (!list_empty(&mddev->disks)) {
6808 			struct md_rdev *rdev0
6809 				= list_entry(mddev->disks.next,
6810 					     struct md_rdev, same_set);
6811 			err = super_types[mddev->major_version]
6812 				.load_super(rdev, rdev0, mddev->minor_version);
6813 			if (err < 0) {
6814 				pr_warn("md: %pg has different UUID to %pg\n",
6815 					rdev->bdev,
6816 					rdev0->bdev);
6817 				export_rdev(rdev, mddev);
6818 				return -EINVAL;
6819 			}
6820 		}
6821 		err = bind_rdev_to_array(rdev, mddev);
6822 		if (err)
6823 			export_rdev(rdev, mddev);
6824 		return err;
6825 	}
6826 
6827 	/*
6828 	 * md_add_new_disk can be used once the array is assembled
6829 	 * to add "hot spares".  They must already have a superblock
6830 	 * written
6831 	 */
6832 	if (mddev->pers) {
6833 		int err;
6834 		if (!mddev->pers->hot_add_disk) {
6835 			pr_warn("%s: personality does not support diskops!\n",
6836 				mdname(mddev));
6837 			return -EINVAL;
6838 		}
6839 		if (mddev->persistent)
6840 			rdev = md_import_device(dev, mddev->major_version,
6841 						mddev->minor_version);
6842 		else
6843 			rdev = md_import_device(dev, -1, -1);
6844 		if (IS_ERR(rdev)) {
6845 			pr_warn("md: md_import_device returned %ld\n",
6846 				PTR_ERR(rdev));
6847 			return PTR_ERR(rdev);
6848 		}
6849 		/* set saved_raid_disk if appropriate */
6850 		if (!mddev->persistent) {
6851 			if (info->state & (1<<MD_DISK_SYNC)  &&
6852 			    info->raid_disk < mddev->raid_disks) {
6853 				rdev->raid_disk = info->raid_disk;
6854 				clear_bit(Bitmap_sync, &rdev->flags);
6855 			} else
6856 				rdev->raid_disk = -1;
6857 			rdev->saved_raid_disk = rdev->raid_disk;
6858 		} else
6859 			super_types[mddev->major_version].
6860 				validate_super(mddev, NULL/*freshest*/, rdev);
6861 		if ((info->state & (1<<MD_DISK_SYNC)) &&
6862 		     rdev->raid_disk != info->raid_disk) {
6863 			/* This was a hot-add request, but events doesn't
6864 			 * match, so reject it.
6865 			 */
6866 			export_rdev(rdev, mddev);
6867 			return -EINVAL;
6868 		}
6869 
6870 		clear_bit(In_sync, &rdev->flags); /* just to be sure */
6871 		if (info->state & (1<<MD_DISK_WRITEMOSTLY))
6872 			set_bit(WriteMostly, &rdev->flags);
6873 		else
6874 			clear_bit(WriteMostly, &rdev->flags);
6875 		if (info->state & (1<<MD_DISK_FAILFAST))
6876 			set_bit(FailFast, &rdev->flags);
6877 		else
6878 			clear_bit(FailFast, &rdev->flags);
6879 
6880 		if (info->state & (1<<MD_DISK_JOURNAL)) {
6881 			struct md_rdev *rdev2;
6882 			bool has_journal = false;
6883 
6884 			/* make sure no existing journal disk */
6885 			rdev_for_each(rdev2, mddev) {
6886 				if (test_bit(Journal, &rdev2->flags)) {
6887 					has_journal = true;
6888 					break;
6889 				}
6890 			}
6891 			if (has_journal || mddev->bitmap) {
6892 				export_rdev(rdev, mddev);
6893 				return -EBUSY;
6894 			}
6895 			set_bit(Journal, &rdev->flags);
6896 		}
6897 		/*
6898 		 * check whether the device shows up in other nodes
6899 		 */
6900 		if (mddev_is_clustered(mddev)) {
6901 			if (info->state & (1 << MD_DISK_CANDIDATE))
6902 				set_bit(Candidate, &rdev->flags);
6903 			else if (info->state & (1 << MD_DISK_CLUSTER_ADD)) {
6904 				/* --add initiated by this node */
6905 				err = md_cluster_ops->add_new_disk(mddev, rdev);
6906 				if (err) {
6907 					export_rdev(rdev, mddev);
6908 					return err;
6909 				}
6910 			}
6911 		}
6912 
6913 		rdev->raid_disk = -1;
6914 		err = bind_rdev_to_array(rdev, mddev);
6915 
6916 		if (err)
6917 			export_rdev(rdev, mddev);
6918 
6919 		if (mddev_is_clustered(mddev)) {
6920 			if (info->state & (1 << MD_DISK_CANDIDATE)) {
6921 				if (!err) {
6922 					err = md_cluster_ops->new_disk_ack(mddev,
6923 						err == 0);
6924 					if (err)
6925 						md_kick_rdev_from_array(rdev);
6926 				}
6927 			} else {
6928 				if (err)
6929 					md_cluster_ops->add_new_disk_cancel(mddev);
6930 				else
6931 					err = add_bound_rdev(rdev);
6932 			}
6933 
6934 		} else if (!err)
6935 			err = add_bound_rdev(rdev);
6936 
6937 		return err;
6938 	}
6939 
6940 	/* otherwise, md_add_new_disk is only allowed
6941 	 * for major_version==0 superblocks
6942 	 */
6943 	if (mddev->major_version != 0) {
6944 		pr_warn("%s: ADD_NEW_DISK not supported\n", mdname(mddev));
6945 		return -EINVAL;
6946 	}
6947 
6948 	if (!(info->state & (1<<MD_DISK_FAULTY))) {
6949 		int err;
6950 		rdev = md_import_device(dev, -1, 0);
6951 		if (IS_ERR(rdev)) {
6952 			pr_warn("md: error, md_import_device() returned %ld\n",
6953 				PTR_ERR(rdev));
6954 			return PTR_ERR(rdev);
6955 		}
6956 		rdev->desc_nr = info->number;
6957 		if (info->raid_disk < mddev->raid_disks)
6958 			rdev->raid_disk = info->raid_disk;
6959 		else
6960 			rdev->raid_disk = -1;
6961 
6962 		if (rdev->raid_disk < mddev->raid_disks)
6963 			if (info->state & (1<<MD_DISK_SYNC))
6964 				set_bit(In_sync, &rdev->flags);
6965 
6966 		if (info->state & (1<<MD_DISK_WRITEMOSTLY))
6967 			set_bit(WriteMostly, &rdev->flags);
6968 		if (info->state & (1<<MD_DISK_FAILFAST))
6969 			set_bit(FailFast, &rdev->flags);
6970 
6971 		if (!mddev->persistent) {
6972 			pr_debug("md: nonpersistent superblock ...\n");
6973 			rdev->sb_start = bdev_nr_sectors(rdev->bdev);
6974 		} else
6975 			rdev->sb_start = calc_dev_sboffset(rdev);
6976 		rdev->sectors = rdev->sb_start;
6977 
6978 		err = bind_rdev_to_array(rdev, mddev);
6979 		if (err) {
6980 			export_rdev(rdev, mddev);
6981 			return err;
6982 		}
6983 	}
6984 
6985 	return 0;
6986 }
6987 
6988 static int hot_remove_disk(struct mddev *mddev, dev_t dev)
6989 {
6990 	struct md_rdev *rdev;
6991 
6992 	if (!mddev->pers)
6993 		return -ENODEV;
6994 
6995 	rdev = find_rdev(mddev, dev);
6996 	if (!rdev)
6997 		return -ENXIO;
6998 
6999 	if (rdev->raid_disk < 0)
7000 		goto kick_rdev;
7001 
7002 	clear_bit(Blocked, &rdev->flags);
7003 	remove_and_add_spares(mddev, rdev);
7004 
7005 	if (rdev->raid_disk >= 0)
7006 		goto busy;
7007 
7008 kick_rdev:
7009 	if (mddev_is_clustered(mddev)) {
7010 		if (md_cluster_ops->remove_disk(mddev, rdev))
7011 			goto busy;
7012 	}
7013 
7014 	md_kick_rdev_from_array(rdev);
7015 	set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
7016 	if (mddev->thread)
7017 		md_wakeup_thread(mddev->thread);
7018 	else
7019 		md_update_sb(mddev, 1);
7020 	md_new_event();
7021 
7022 	return 0;
7023 busy:
7024 	pr_debug("md: cannot remove active disk %pg from %s ...\n",
7025 		 rdev->bdev, mdname(mddev));
7026 	return -EBUSY;
7027 }
7028 
7029 static int hot_add_disk(struct mddev *mddev, dev_t dev)
7030 {
7031 	int err;
7032 	struct md_rdev *rdev;
7033 
7034 	if (!mddev->pers)
7035 		return -ENODEV;
7036 
7037 	if (mddev->major_version != 0) {
7038 		pr_warn("%s: HOT_ADD may only be used with version-0 superblocks.\n",
7039 			mdname(mddev));
7040 		return -EINVAL;
7041 	}
7042 	if (!mddev->pers->hot_add_disk) {
7043 		pr_warn("%s: personality does not support diskops!\n",
7044 			mdname(mddev));
7045 		return -EINVAL;
7046 	}
7047 
7048 	rdev = md_import_device(dev, -1, 0);
7049 	if (IS_ERR(rdev)) {
7050 		pr_warn("md: error, md_import_device() returned %ld\n",
7051 			PTR_ERR(rdev));
7052 		return -EINVAL;
7053 	}
7054 
7055 	if (mddev->persistent)
7056 		rdev->sb_start = calc_dev_sboffset(rdev);
7057 	else
7058 		rdev->sb_start = bdev_nr_sectors(rdev->bdev);
7059 
7060 	rdev->sectors = rdev->sb_start;
7061 
7062 	if (test_bit(Faulty, &rdev->flags)) {
7063 		pr_warn("md: can not hot-add faulty %pg disk to %s!\n",
7064 			rdev->bdev, mdname(mddev));
7065 		err = -EINVAL;
7066 		goto abort_export;
7067 	}
7068 
7069 	clear_bit(In_sync, &rdev->flags);
7070 	rdev->desc_nr = -1;
7071 	rdev->saved_raid_disk = -1;
7072 	err = bind_rdev_to_array(rdev, mddev);
7073 	if (err)
7074 		goto abort_export;
7075 
7076 	/*
7077 	 * The rest should better be atomic, we can have disk failures
7078 	 * noticed in interrupt contexts ...
7079 	 */
7080 
7081 	rdev->raid_disk = -1;
7082 
7083 	set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
7084 	if (!mddev->thread)
7085 		md_update_sb(mddev, 1);
7086 	/*
7087 	 * If the new disk does not support REQ_NOWAIT,
7088 	 * disable on the whole MD.
7089 	 */
7090 	if (!bdev_nowait(rdev->bdev)) {
7091 		pr_info("%s: Disabling nowait because %pg does not support nowait\n",
7092 			mdname(mddev), rdev->bdev);
7093 		blk_queue_flag_clear(QUEUE_FLAG_NOWAIT, mddev->queue);
7094 	}
7095 	/*
7096 	 * Kick recovery, maybe this spare has to be added to the
7097 	 * array immediately.
7098 	 */
7099 	set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7100 	md_wakeup_thread(mddev->thread);
7101 	md_new_event();
7102 	return 0;
7103 
7104 abort_export:
7105 	export_rdev(rdev, mddev);
7106 	return err;
7107 }
7108 
7109 static int set_bitmap_file(struct mddev *mddev, int fd)
7110 {
7111 	int err = 0;
7112 
7113 	if (mddev->pers) {
7114 		if (!mddev->pers->quiesce || !mddev->thread)
7115 			return -EBUSY;
7116 		if (mddev->recovery || mddev->sync_thread)
7117 			return -EBUSY;
7118 		/* we should be able to change the bitmap.. */
7119 	}
7120 
7121 	if (fd >= 0) {
7122 		struct inode *inode;
7123 		struct file *f;
7124 
7125 		if (mddev->bitmap || mddev->bitmap_info.file)
7126 			return -EEXIST; /* cannot add when bitmap is present */
7127 
7128 		if (!IS_ENABLED(CONFIG_MD_BITMAP_FILE)) {
7129 			pr_warn("%s: bitmap files not supported by this kernel\n",
7130 				mdname(mddev));
7131 			return -EINVAL;
7132 		}
7133 		pr_warn("%s: using deprecated bitmap file support\n",
7134 			mdname(mddev));
7135 
7136 		f = fget(fd);
7137 
7138 		if (f == NULL) {
7139 			pr_warn("%s: error: failed to get bitmap file\n",
7140 				mdname(mddev));
7141 			return -EBADF;
7142 		}
7143 
7144 		inode = f->f_mapping->host;
7145 		if (!S_ISREG(inode->i_mode)) {
7146 			pr_warn("%s: error: bitmap file must be a regular file\n",
7147 				mdname(mddev));
7148 			err = -EBADF;
7149 		} else if (!(f->f_mode & FMODE_WRITE)) {
7150 			pr_warn("%s: error: bitmap file must open for write\n",
7151 				mdname(mddev));
7152 			err = -EBADF;
7153 		} else if (atomic_read(&inode->i_writecount) != 1) {
7154 			pr_warn("%s: error: bitmap file is already in use\n",
7155 				mdname(mddev));
7156 			err = -EBUSY;
7157 		}
7158 		if (err) {
7159 			fput(f);
7160 			return err;
7161 		}
7162 		mddev->bitmap_info.file = f;
7163 		mddev->bitmap_info.offset = 0; /* file overrides offset */
7164 	} else if (mddev->bitmap == NULL)
7165 		return -ENOENT; /* cannot remove what isn't there */
7166 	err = 0;
7167 	if (mddev->pers) {
7168 		if (fd >= 0) {
7169 			struct bitmap *bitmap;
7170 
7171 			bitmap = md_bitmap_create(mddev, -1);
7172 			mddev_suspend(mddev);
7173 			if (!IS_ERR(bitmap)) {
7174 				mddev->bitmap = bitmap;
7175 				err = md_bitmap_load(mddev);
7176 			} else
7177 				err = PTR_ERR(bitmap);
7178 			if (err) {
7179 				md_bitmap_destroy(mddev);
7180 				fd = -1;
7181 			}
7182 			mddev_resume(mddev);
7183 		} else if (fd < 0) {
7184 			mddev_suspend(mddev);
7185 			md_bitmap_destroy(mddev);
7186 			mddev_resume(mddev);
7187 		}
7188 	}
7189 	if (fd < 0) {
7190 		struct file *f = mddev->bitmap_info.file;
7191 		if (f) {
7192 			spin_lock(&mddev->lock);
7193 			mddev->bitmap_info.file = NULL;
7194 			spin_unlock(&mddev->lock);
7195 			fput(f);
7196 		}
7197 	}
7198 
7199 	return err;
7200 }
7201 
7202 /*
7203  * md_set_array_info is used two different ways
7204  * The original usage is when creating a new array.
7205  * In this usage, raid_disks is > 0 and it together with
7206  *  level, size, not_persistent,layout,chunksize determine the
7207  *  shape of the array.
7208  *  This will always create an array with a type-0.90.0 superblock.
7209  * The newer usage is when assembling an array.
7210  *  In this case raid_disks will be 0, and the major_version field is
7211  *  use to determine which style super-blocks are to be found on the devices.
7212  *  The minor and patch _version numbers are also kept incase the
7213  *  super_block handler wishes to interpret them.
7214  */
7215 int md_set_array_info(struct mddev *mddev, struct mdu_array_info_s *info)
7216 {
7217 	if (info->raid_disks == 0) {
7218 		/* just setting version number for superblock loading */
7219 		if (info->major_version < 0 ||
7220 		    info->major_version >= ARRAY_SIZE(super_types) ||
7221 		    super_types[info->major_version].name == NULL) {
7222 			/* maybe try to auto-load a module? */
7223 			pr_warn("md: superblock version %d not known\n",
7224 				info->major_version);
7225 			return -EINVAL;
7226 		}
7227 		mddev->major_version = info->major_version;
7228 		mddev->minor_version = info->minor_version;
7229 		mddev->patch_version = info->patch_version;
7230 		mddev->persistent = !info->not_persistent;
7231 		/* ensure mddev_put doesn't delete this now that there
7232 		 * is some minimal configuration.
7233 		 */
7234 		mddev->ctime         = ktime_get_real_seconds();
7235 		return 0;
7236 	}
7237 	mddev->major_version = MD_MAJOR_VERSION;
7238 	mddev->minor_version = MD_MINOR_VERSION;
7239 	mddev->patch_version = MD_PATCHLEVEL_VERSION;
7240 	mddev->ctime         = ktime_get_real_seconds();
7241 
7242 	mddev->level         = info->level;
7243 	mddev->clevel[0]     = 0;
7244 	mddev->dev_sectors   = 2 * (sector_t)info->size;
7245 	mddev->raid_disks    = info->raid_disks;
7246 	/* don't set md_minor, it is determined by which /dev/md* was
7247 	 * openned
7248 	 */
7249 	if (info->state & (1<<MD_SB_CLEAN))
7250 		mddev->recovery_cp = MaxSector;
7251 	else
7252 		mddev->recovery_cp = 0;
7253 	mddev->persistent    = ! info->not_persistent;
7254 	mddev->external	     = 0;
7255 
7256 	mddev->layout        = info->layout;
7257 	if (mddev->level == 0)
7258 		/* Cannot trust RAID0 layout info here */
7259 		mddev->layout = -1;
7260 	mddev->chunk_sectors = info->chunk_size >> 9;
7261 
7262 	if (mddev->persistent) {
7263 		mddev->max_disks = MD_SB_DISKS;
7264 		mddev->flags = 0;
7265 		mddev->sb_flags = 0;
7266 	}
7267 	set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
7268 
7269 	mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9;
7270 	mddev->bitmap_info.default_space = 64*2 - (MD_SB_BYTES >> 9);
7271 	mddev->bitmap_info.offset = 0;
7272 
7273 	mddev->reshape_position = MaxSector;
7274 
7275 	/*
7276 	 * Generate a 128 bit UUID
7277 	 */
7278 	get_random_bytes(mddev->uuid, 16);
7279 
7280 	mddev->new_level = mddev->level;
7281 	mddev->new_chunk_sectors = mddev->chunk_sectors;
7282 	mddev->new_layout = mddev->layout;
7283 	mddev->delta_disks = 0;
7284 	mddev->reshape_backwards = 0;
7285 
7286 	return 0;
7287 }
7288 
7289 void md_set_array_sectors(struct mddev *mddev, sector_t array_sectors)
7290 {
7291 	lockdep_assert_held(&mddev->reconfig_mutex);
7292 
7293 	if (mddev->external_size)
7294 		return;
7295 
7296 	mddev->array_sectors = array_sectors;
7297 }
7298 EXPORT_SYMBOL(md_set_array_sectors);
7299 
7300 static int update_size(struct mddev *mddev, sector_t num_sectors)
7301 {
7302 	struct md_rdev *rdev;
7303 	int rv;
7304 	int fit = (num_sectors == 0);
7305 	sector_t old_dev_sectors = mddev->dev_sectors;
7306 
7307 	if (mddev->pers->resize == NULL)
7308 		return -EINVAL;
7309 	/* The "num_sectors" is the number of sectors of each device that
7310 	 * is used.  This can only make sense for arrays with redundancy.
7311 	 * linear and raid0 always use whatever space is available. We can only
7312 	 * consider changing this number if no resync or reconstruction is
7313 	 * happening, and if the new size is acceptable. It must fit before the
7314 	 * sb_start or, if that is <data_offset, it must fit before the size
7315 	 * of each device.  If num_sectors is zero, we find the largest size
7316 	 * that fits.
7317 	 */
7318 	if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
7319 	    mddev->sync_thread)
7320 		return -EBUSY;
7321 	if (!md_is_rdwr(mddev))
7322 		return -EROFS;
7323 
7324 	rdev_for_each(rdev, mddev) {
7325 		sector_t avail = rdev->sectors;
7326 
7327 		if (fit && (num_sectors == 0 || num_sectors > avail))
7328 			num_sectors = avail;
7329 		if (avail < num_sectors)
7330 			return -ENOSPC;
7331 	}
7332 	rv = mddev->pers->resize(mddev, num_sectors);
7333 	if (!rv) {
7334 		if (mddev_is_clustered(mddev))
7335 			md_cluster_ops->update_size(mddev, old_dev_sectors);
7336 		else if (mddev->queue) {
7337 			set_capacity_and_notify(mddev->gendisk,
7338 						mddev->array_sectors);
7339 		}
7340 	}
7341 	return rv;
7342 }
7343 
7344 static int update_raid_disks(struct mddev *mddev, int raid_disks)
7345 {
7346 	int rv;
7347 	struct md_rdev *rdev;
7348 	/* change the number of raid disks */
7349 	if (mddev->pers->check_reshape == NULL)
7350 		return -EINVAL;
7351 	if (!md_is_rdwr(mddev))
7352 		return -EROFS;
7353 	if (raid_disks <= 0 ||
7354 	    (mddev->max_disks && raid_disks >= mddev->max_disks))
7355 		return -EINVAL;
7356 	if (mddev->sync_thread ||
7357 	    test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
7358 	    test_bit(MD_RESYNCING_REMOTE, &mddev->recovery) ||
7359 	    mddev->reshape_position != MaxSector)
7360 		return -EBUSY;
7361 
7362 	rdev_for_each(rdev, mddev) {
7363 		if (mddev->raid_disks < raid_disks &&
7364 		    rdev->data_offset < rdev->new_data_offset)
7365 			return -EINVAL;
7366 		if (mddev->raid_disks > raid_disks &&
7367 		    rdev->data_offset > rdev->new_data_offset)
7368 			return -EINVAL;
7369 	}
7370 
7371 	mddev->delta_disks = raid_disks - mddev->raid_disks;
7372 	if (mddev->delta_disks < 0)
7373 		mddev->reshape_backwards = 1;
7374 	else if (mddev->delta_disks > 0)
7375 		mddev->reshape_backwards = 0;
7376 
7377 	rv = mddev->pers->check_reshape(mddev);
7378 	if (rv < 0) {
7379 		mddev->delta_disks = 0;
7380 		mddev->reshape_backwards = 0;
7381 	}
7382 	return rv;
7383 }
7384 
7385 /*
7386  * update_array_info is used to change the configuration of an
7387  * on-line array.
7388  * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
7389  * fields in the info are checked against the array.
7390  * Any differences that cannot be handled will cause an error.
7391  * Normally, only one change can be managed at a time.
7392  */
7393 static int update_array_info(struct mddev *mddev, mdu_array_info_t *info)
7394 {
7395 	int rv = 0;
7396 	int cnt = 0;
7397 	int state = 0;
7398 
7399 	/* calculate expected state,ignoring low bits */
7400 	if (mddev->bitmap && mddev->bitmap_info.offset)
7401 		state |= (1 << MD_SB_BITMAP_PRESENT);
7402 
7403 	if (mddev->major_version != info->major_version ||
7404 	    mddev->minor_version != info->minor_version ||
7405 /*	    mddev->patch_version != info->patch_version || */
7406 	    mddev->ctime         != info->ctime         ||
7407 	    mddev->level         != info->level         ||
7408 /*	    mddev->layout        != info->layout        || */
7409 	    mddev->persistent	 != !info->not_persistent ||
7410 	    mddev->chunk_sectors != info->chunk_size >> 9 ||
7411 	    /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
7412 	    ((state^info->state) & 0xfffffe00)
7413 		)
7414 		return -EINVAL;
7415 	/* Check there is only one change */
7416 	if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
7417 		cnt++;
7418 	if (mddev->raid_disks != info->raid_disks)
7419 		cnt++;
7420 	if (mddev->layout != info->layout)
7421 		cnt++;
7422 	if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT))
7423 		cnt++;
7424 	if (cnt == 0)
7425 		return 0;
7426 	if (cnt > 1)
7427 		return -EINVAL;
7428 
7429 	if (mddev->layout != info->layout) {
7430 		/* Change layout
7431 		 * we don't need to do anything at the md level, the
7432 		 * personality will take care of it all.
7433 		 */
7434 		if (mddev->pers->check_reshape == NULL)
7435 			return -EINVAL;
7436 		else {
7437 			mddev->new_layout = info->layout;
7438 			rv = mddev->pers->check_reshape(mddev);
7439 			if (rv)
7440 				mddev->new_layout = mddev->layout;
7441 			return rv;
7442 		}
7443 	}
7444 	if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
7445 		rv = update_size(mddev, (sector_t)info->size * 2);
7446 
7447 	if (mddev->raid_disks    != info->raid_disks)
7448 		rv = update_raid_disks(mddev, info->raid_disks);
7449 
7450 	if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) {
7451 		if (mddev->pers->quiesce == NULL || mddev->thread == NULL) {
7452 			rv = -EINVAL;
7453 			goto err;
7454 		}
7455 		if (mddev->recovery || mddev->sync_thread) {
7456 			rv = -EBUSY;
7457 			goto err;
7458 		}
7459 		if (info->state & (1<<MD_SB_BITMAP_PRESENT)) {
7460 			struct bitmap *bitmap;
7461 			/* add the bitmap */
7462 			if (mddev->bitmap) {
7463 				rv = -EEXIST;
7464 				goto err;
7465 			}
7466 			if (mddev->bitmap_info.default_offset == 0) {
7467 				rv = -EINVAL;
7468 				goto err;
7469 			}
7470 			mddev->bitmap_info.offset =
7471 				mddev->bitmap_info.default_offset;
7472 			mddev->bitmap_info.space =
7473 				mddev->bitmap_info.default_space;
7474 			bitmap = md_bitmap_create(mddev, -1);
7475 			mddev_suspend(mddev);
7476 			if (!IS_ERR(bitmap)) {
7477 				mddev->bitmap = bitmap;
7478 				rv = md_bitmap_load(mddev);
7479 			} else
7480 				rv = PTR_ERR(bitmap);
7481 			if (rv)
7482 				md_bitmap_destroy(mddev);
7483 			mddev_resume(mddev);
7484 		} else {
7485 			/* remove the bitmap */
7486 			if (!mddev->bitmap) {
7487 				rv = -ENOENT;
7488 				goto err;
7489 			}
7490 			if (mddev->bitmap->storage.file) {
7491 				rv = -EINVAL;
7492 				goto err;
7493 			}
7494 			if (mddev->bitmap_info.nodes) {
7495 				/* hold PW on all the bitmap lock */
7496 				if (md_cluster_ops->lock_all_bitmaps(mddev) <= 0) {
7497 					pr_warn("md: can't change bitmap to none since the array is in use by more than one node\n");
7498 					rv = -EPERM;
7499 					md_cluster_ops->unlock_all_bitmaps(mddev);
7500 					goto err;
7501 				}
7502 
7503 				mddev->bitmap_info.nodes = 0;
7504 				md_cluster_ops->leave(mddev);
7505 				module_put(md_cluster_mod);
7506 				mddev->safemode_delay = DEFAULT_SAFEMODE_DELAY;
7507 			}
7508 			mddev_suspend(mddev);
7509 			md_bitmap_destroy(mddev);
7510 			mddev_resume(mddev);
7511 			mddev->bitmap_info.offset = 0;
7512 		}
7513 	}
7514 	md_update_sb(mddev, 1);
7515 	return rv;
7516 err:
7517 	return rv;
7518 }
7519 
7520 static int set_disk_faulty(struct mddev *mddev, dev_t dev)
7521 {
7522 	struct md_rdev *rdev;
7523 	int err = 0;
7524 
7525 	if (mddev->pers == NULL)
7526 		return -ENODEV;
7527 
7528 	rcu_read_lock();
7529 	rdev = md_find_rdev_rcu(mddev, dev);
7530 	if (!rdev)
7531 		err =  -ENODEV;
7532 	else {
7533 		md_error(mddev, rdev);
7534 		if (test_bit(MD_BROKEN, &mddev->flags))
7535 			err = -EBUSY;
7536 	}
7537 	rcu_read_unlock();
7538 	return err;
7539 }
7540 
7541 /*
7542  * We have a problem here : there is no easy way to give a CHS
7543  * virtual geometry. We currently pretend that we have a 2 heads
7544  * 4 sectors (with a BIG number of cylinders...). This drives
7545  * dosfs just mad... ;-)
7546  */
7547 static int md_getgeo(struct block_device *bdev, struct hd_geometry *geo)
7548 {
7549 	struct mddev *mddev = bdev->bd_disk->private_data;
7550 
7551 	geo->heads = 2;
7552 	geo->sectors = 4;
7553 	geo->cylinders = mddev->array_sectors / 8;
7554 	return 0;
7555 }
7556 
7557 static inline bool md_ioctl_valid(unsigned int cmd)
7558 {
7559 	switch (cmd) {
7560 	case ADD_NEW_DISK:
7561 	case GET_ARRAY_INFO:
7562 	case GET_BITMAP_FILE:
7563 	case GET_DISK_INFO:
7564 	case HOT_ADD_DISK:
7565 	case HOT_REMOVE_DISK:
7566 	case RAID_VERSION:
7567 	case RESTART_ARRAY_RW:
7568 	case RUN_ARRAY:
7569 	case SET_ARRAY_INFO:
7570 	case SET_BITMAP_FILE:
7571 	case SET_DISK_FAULTY:
7572 	case STOP_ARRAY:
7573 	case STOP_ARRAY_RO:
7574 	case CLUSTERED_DISK_NACK:
7575 		return true;
7576 	default:
7577 		return false;
7578 	}
7579 }
7580 
7581 static int __md_set_array_info(struct mddev *mddev, void __user *argp)
7582 {
7583 	mdu_array_info_t info;
7584 	int err;
7585 
7586 	if (!argp)
7587 		memset(&info, 0, sizeof(info));
7588 	else if (copy_from_user(&info, argp, sizeof(info)))
7589 		return -EFAULT;
7590 
7591 	if (mddev->pers) {
7592 		err = update_array_info(mddev, &info);
7593 		if (err)
7594 			pr_warn("md: couldn't update array info. %d\n", err);
7595 		return err;
7596 	}
7597 
7598 	if (!list_empty(&mddev->disks)) {
7599 		pr_warn("md: array %s already has disks!\n", mdname(mddev));
7600 		return -EBUSY;
7601 	}
7602 
7603 	if (mddev->raid_disks) {
7604 		pr_warn("md: array %s already initialised!\n", mdname(mddev));
7605 		return -EBUSY;
7606 	}
7607 
7608 	err = md_set_array_info(mddev, &info);
7609 	if (err)
7610 		pr_warn("md: couldn't set array info. %d\n", err);
7611 
7612 	return err;
7613 }
7614 
7615 static int md_ioctl(struct block_device *bdev, blk_mode_t mode,
7616 			unsigned int cmd, unsigned long arg)
7617 {
7618 	int err = 0;
7619 	void __user *argp = (void __user *)arg;
7620 	struct mddev *mddev = NULL;
7621 
7622 	if (!md_ioctl_valid(cmd))
7623 		return -ENOTTY;
7624 
7625 	switch (cmd) {
7626 	case RAID_VERSION:
7627 	case GET_ARRAY_INFO:
7628 	case GET_DISK_INFO:
7629 		break;
7630 	default:
7631 		if (!capable(CAP_SYS_ADMIN))
7632 			return -EACCES;
7633 	}
7634 
7635 	/*
7636 	 * Commands dealing with the RAID driver but not any
7637 	 * particular array:
7638 	 */
7639 	switch (cmd) {
7640 	case RAID_VERSION:
7641 		err = get_version(argp);
7642 		goto out;
7643 	default:;
7644 	}
7645 
7646 	/*
7647 	 * Commands creating/starting a new array:
7648 	 */
7649 
7650 	mddev = bdev->bd_disk->private_data;
7651 
7652 	if (!mddev) {
7653 		BUG();
7654 		goto out;
7655 	}
7656 
7657 	/* Some actions do not requires the mutex */
7658 	switch (cmd) {
7659 	case GET_ARRAY_INFO:
7660 		if (!mddev->raid_disks && !mddev->external)
7661 			err = -ENODEV;
7662 		else
7663 			err = get_array_info(mddev, argp);
7664 		goto out;
7665 
7666 	case GET_DISK_INFO:
7667 		if (!mddev->raid_disks && !mddev->external)
7668 			err = -ENODEV;
7669 		else
7670 			err = get_disk_info(mddev, argp);
7671 		goto out;
7672 
7673 	case SET_DISK_FAULTY:
7674 		err = set_disk_faulty(mddev, new_decode_dev(arg));
7675 		goto out;
7676 
7677 	case GET_BITMAP_FILE:
7678 		err = get_bitmap_file(mddev, argp);
7679 		goto out;
7680 
7681 	}
7682 
7683 	if (cmd == HOT_REMOVE_DISK)
7684 		/* need to ensure recovery thread has run */
7685 		wait_event_interruptible_timeout(mddev->sb_wait,
7686 						 !test_bit(MD_RECOVERY_NEEDED,
7687 							   &mddev->recovery),
7688 						 msecs_to_jiffies(5000));
7689 	if (cmd == STOP_ARRAY || cmd == STOP_ARRAY_RO) {
7690 		/* Need to flush page cache, and ensure no-one else opens
7691 		 * and writes
7692 		 */
7693 		mutex_lock(&mddev->open_mutex);
7694 		if (mddev->pers && atomic_read(&mddev->openers) > 1) {
7695 			mutex_unlock(&mddev->open_mutex);
7696 			err = -EBUSY;
7697 			goto out;
7698 		}
7699 		if (test_and_set_bit(MD_CLOSING, &mddev->flags)) {
7700 			mutex_unlock(&mddev->open_mutex);
7701 			err = -EBUSY;
7702 			goto out;
7703 		}
7704 		mutex_unlock(&mddev->open_mutex);
7705 		sync_blockdev(bdev);
7706 	}
7707 	err = mddev_lock(mddev);
7708 	if (err) {
7709 		pr_debug("md: ioctl lock interrupted, reason %d, cmd %d\n",
7710 			 err, cmd);
7711 		goto out;
7712 	}
7713 
7714 	if (cmd == SET_ARRAY_INFO) {
7715 		err = __md_set_array_info(mddev, argp);
7716 		goto unlock;
7717 	}
7718 
7719 	/*
7720 	 * Commands querying/configuring an existing array:
7721 	 */
7722 	/* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
7723 	 * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */
7724 	if ((!mddev->raid_disks && !mddev->external)
7725 	    && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY
7726 	    && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE
7727 	    && cmd != GET_BITMAP_FILE) {
7728 		err = -ENODEV;
7729 		goto unlock;
7730 	}
7731 
7732 	/*
7733 	 * Commands even a read-only array can execute:
7734 	 */
7735 	switch (cmd) {
7736 	case RESTART_ARRAY_RW:
7737 		err = restart_array(mddev);
7738 		goto unlock;
7739 
7740 	case STOP_ARRAY:
7741 		err = do_md_stop(mddev, 0, bdev);
7742 		goto unlock;
7743 
7744 	case STOP_ARRAY_RO:
7745 		err = md_set_readonly(mddev, bdev);
7746 		goto unlock;
7747 
7748 	case HOT_REMOVE_DISK:
7749 		err = hot_remove_disk(mddev, new_decode_dev(arg));
7750 		goto unlock;
7751 
7752 	case ADD_NEW_DISK:
7753 		/* We can support ADD_NEW_DISK on read-only arrays
7754 		 * only if we are re-adding a preexisting device.
7755 		 * So require mddev->pers and MD_DISK_SYNC.
7756 		 */
7757 		if (mddev->pers) {
7758 			mdu_disk_info_t info;
7759 			if (copy_from_user(&info, argp, sizeof(info)))
7760 				err = -EFAULT;
7761 			else if (!(info.state & (1<<MD_DISK_SYNC)))
7762 				/* Need to clear read-only for this */
7763 				break;
7764 			else
7765 				err = md_add_new_disk(mddev, &info);
7766 			goto unlock;
7767 		}
7768 		break;
7769 	}
7770 
7771 	/*
7772 	 * The remaining ioctls are changing the state of the
7773 	 * superblock, so we do not allow them on read-only arrays.
7774 	 */
7775 	if (!md_is_rdwr(mddev) && mddev->pers) {
7776 		if (mddev->ro != MD_AUTO_READ) {
7777 			err = -EROFS;
7778 			goto unlock;
7779 		}
7780 		mddev->ro = MD_RDWR;
7781 		sysfs_notify_dirent_safe(mddev->sysfs_state);
7782 		set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7783 		/* mddev_unlock will wake thread */
7784 		/* If a device failed while we were read-only, we
7785 		 * need to make sure the metadata is updated now.
7786 		 */
7787 		if (test_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags)) {
7788 			mddev_unlock(mddev);
7789 			wait_event(mddev->sb_wait,
7790 				   !test_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags) &&
7791 				   !test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags));
7792 			mddev_lock_nointr(mddev);
7793 		}
7794 	}
7795 
7796 	switch (cmd) {
7797 	case ADD_NEW_DISK:
7798 	{
7799 		mdu_disk_info_t info;
7800 		if (copy_from_user(&info, argp, sizeof(info)))
7801 			err = -EFAULT;
7802 		else
7803 			err = md_add_new_disk(mddev, &info);
7804 		goto unlock;
7805 	}
7806 
7807 	case CLUSTERED_DISK_NACK:
7808 		if (mddev_is_clustered(mddev))
7809 			md_cluster_ops->new_disk_ack(mddev, false);
7810 		else
7811 			err = -EINVAL;
7812 		goto unlock;
7813 
7814 	case HOT_ADD_DISK:
7815 		err = hot_add_disk(mddev, new_decode_dev(arg));
7816 		goto unlock;
7817 
7818 	case RUN_ARRAY:
7819 		err = do_md_run(mddev);
7820 		goto unlock;
7821 
7822 	case SET_BITMAP_FILE:
7823 		err = set_bitmap_file(mddev, (int)arg);
7824 		goto unlock;
7825 
7826 	default:
7827 		err = -EINVAL;
7828 		goto unlock;
7829 	}
7830 
7831 unlock:
7832 	if (mddev->hold_active == UNTIL_IOCTL &&
7833 	    err != -EINVAL)
7834 		mddev->hold_active = 0;
7835 	mddev_unlock(mddev);
7836 out:
7837 	if (cmd == STOP_ARRAY_RO || (err && cmd == STOP_ARRAY))
7838 		clear_bit(MD_CLOSING, &mddev->flags);
7839 	return err;
7840 }
7841 #ifdef CONFIG_COMPAT
7842 static int md_compat_ioctl(struct block_device *bdev, blk_mode_t mode,
7843 		    unsigned int cmd, unsigned long arg)
7844 {
7845 	switch (cmd) {
7846 	case HOT_REMOVE_DISK:
7847 	case HOT_ADD_DISK:
7848 	case SET_DISK_FAULTY:
7849 	case SET_BITMAP_FILE:
7850 		/* These take in integer arg, do not convert */
7851 		break;
7852 	default:
7853 		arg = (unsigned long)compat_ptr(arg);
7854 		break;
7855 	}
7856 
7857 	return md_ioctl(bdev, mode, cmd, arg);
7858 }
7859 #endif /* CONFIG_COMPAT */
7860 
7861 static int md_set_read_only(struct block_device *bdev, bool ro)
7862 {
7863 	struct mddev *mddev = bdev->bd_disk->private_data;
7864 	int err;
7865 
7866 	err = mddev_lock(mddev);
7867 	if (err)
7868 		return err;
7869 
7870 	if (!mddev->raid_disks && !mddev->external) {
7871 		err = -ENODEV;
7872 		goto out_unlock;
7873 	}
7874 
7875 	/*
7876 	 * Transitioning to read-auto need only happen for arrays that call
7877 	 * md_write_start and which are not ready for writes yet.
7878 	 */
7879 	if (!ro && mddev->ro == MD_RDONLY && mddev->pers) {
7880 		err = restart_array(mddev);
7881 		if (err)
7882 			goto out_unlock;
7883 		mddev->ro = MD_AUTO_READ;
7884 	}
7885 
7886 out_unlock:
7887 	mddev_unlock(mddev);
7888 	return err;
7889 }
7890 
7891 static int md_open(struct gendisk *disk, blk_mode_t mode)
7892 {
7893 	struct mddev *mddev;
7894 	int err;
7895 
7896 	spin_lock(&all_mddevs_lock);
7897 	mddev = mddev_get(disk->private_data);
7898 	spin_unlock(&all_mddevs_lock);
7899 	if (!mddev)
7900 		return -ENODEV;
7901 
7902 	err = mutex_lock_interruptible(&mddev->open_mutex);
7903 	if (err)
7904 		goto out;
7905 
7906 	err = -ENODEV;
7907 	if (test_bit(MD_CLOSING, &mddev->flags))
7908 		goto out_unlock;
7909 
7910 	atomic_inc(&mddev->openers);
7911 	mutex_unlock(&mddev->open_mutex);
7912 
7913 	disk_check_media_change(disk);
7914 	return 0;
7915 
7916 out_unlock:
7917 	mutex_unlock(&mddev->open_mutex);
7918 out:
7919 	mddev_put(mddev);
7920 	return err;
7921 }
7922 
7923 static void md_release(struct gendisk *disk)
7924 {
7925 	struct mddev *mddev = disk->private_data;
7926 
7927 	BUG_ON(!mddev);
7928 	atomic_dec(&mddev->openers);
7929 	mddev_put(mddev);
7930 }
7931 
7932 static unsigned int md_check_events(struct gendisk *disk, unsigned int clearing)
7933 {
7934 	struct mddev *mddev = disk->private_data;
7935 	unsigned int ret = 0;
7936 
7937 	if (mddev->changed)
7938 		ret = DISK_EVENT_MEDIA_CHANGE;
7939 	mddev->changed = 0;
7940 	return ret;
7941 }
7942 
7943 static void md_free_disk(struct gendisk *disk)
7944 {
7945 	struct mddev *mddev = disk->private_data;
7946 
7947 	percpu_ref_exit(&mddev->writes_pending);
7948 	mddev_free(mddev);
7949 }
7950 
7951 const struct block_device_operations md_fops =
7952 {
7953 	.owner		= THIS_MODULE,
7954 	.submit_bio	= md_submit_bio,
7955 	.open		= md_open,
7956 	.release	= md_release,
7957 	.ioctl		= md_ioctl,
7958 #ifdef CONFIG_COMPAT
7959 	.compat_ioctl	= md_compat_ioctl,
7960 #endif
7961 	.getgeo		= md_getgeo,
7962 	.check_events	= md_check_events,
7963 	.set_read_only	= md_set_read_only,
7964 	.free_disk	= md_free_disk,
7965 };
7966 
7967 static int md_thread(void *arg)
7968 {
7969 	struct md_thread *thread = arg;
7970 
7971 	/*
7972 	 * md_thread is a 'system-thread', it's priority should be very
7973 	 * high. We avoid resource deadlocks individually in each
7974 	 * raid personality. (RAID5 does preallocation) We also use RR and
7975 	 * the very same RT priority as kswapd, thus we will never get
7976 	 * into a priority inversion deadlock.
7977 	 *
7978 	 * we definitely have to have equal or higher priority than
7979 	 * bdflush, otherwise bdflush will deadlock if there are too
7980 	 * many dirty RAID5 blocks.
7981 	 */
7982 
7983 	allow_signal(SIGKILL);
7984 	while (!kthread_should_stop()) {
7985 
7986 		/* We need to wait INTERRUPTIBLE so that
7987 		 * we don't add to the load-average.
7988 		 * That means we need to be sure no signals are
7989 		 * pending
7990 		 */
7991 		if (signal_pending(current))
7992 			flush_signals(current);
7993 
7994 		wait_event_interruptible_timeout
7995 			(thread->wqueue,
7996 			 test_bit(THREAD_WAKEUP, &thread->flags)
7997 			 || kthread_should_stop() || kthread_should_park(),
7998 			 thread->timeout);
7999 
8000 		clear_bit(THREAD_WAKEUP, &thread->flags);
8001 		if (kthread_should_park())
8002 			kthread_parkme();
8003 		if (!kthread_should_stop())
8004 			thread->run(thread);
8005 	}
8006 
8007 	return 0;
8008 }
8009 
8010 static void md_wakeup_thread_directly(struct md_thread __rcu *thread)
8011 {
8012 	struct md_thread *t;
8013 
8014 	rcu_read_lock();
8015 	t = rcu_dereference(thread);
8016 	if (t)
8017 		wake_up_process(t->tsk);
8018 	rcu_read_unlock();
8019 }
8020 
8021 void md_wakeup_thread(struct md_thread __rcu *thread)
8022 {
8023 	struct md_thread *t;
8024 
8025 	rcu_read_lock();
8026 	t = rcu_dereference(thread);
8027 	if (t) {
8028 		pr_debug("md: waking up MD thread %s.\n", t->tsk->comm);
8029 		set_bit(THREAD_WAKEUP, &t->flags);
8030 		wake_up(&t->wqueue);
8031 	}
8032 	rcu_read_unlock();
8033 }
8034 EXPORT_SYMBOL(md_wakeup_thread);
8035 
8036 struct md_thread *md_register_thread(void (*run) (struct md_thread *),
8037 		struct mddev *mddev, const char *name)
8038 {
8039 	struct md_thread *thread;
8040 
8041 	thread = kzalloc(sizeof(struct md_thread), GFP_KERNEL);
8042 	if (!thread)
8043 		return NULL;
8044 
8045 	init_waitqueue_head(&thread->wqueue);
8046 
8047 	thread->run = run;
8048 	thread->mddev = mddev;
8049 	thread->timeout = MAX_SCHEDULE_TIMEOUT;
8050 	thread->tsk = kthread_run(md_thread, thread,
8051 				  "%s_%s",
8052 				  mdname(thread->mddev),
8053 				  name);
8054 	if (IS_ERR(thread->tsk)) {
8055 		kfree(thread);
8056 		return NULL;
8057 	}
8058 	return thread;
8059 }
8060 EXPORT_SYMBOL(md_register_thread);
8061 
8062 void md_unregister_thread(struct mddev *mddev, struct md_thread __rcu **threadp)
8063 {
8064 	struct md_thread *thread = rcu_dereference_protected(*threadp,
8065 					lockdep_is_held(&mddev->reconfig_mutex));
8066 
8067 	if (!thread)
8068 		return;
8069 
8070 	rcu_assign_pointer(*threadp, NULL);
8071 	synchronize_rcu();
8072 
8073 	pr_debug("interrupting MD-thread pid %d\n", task_pid_nr(thread->tsk));
8074 	kthread_stop(thread->tsk);
8075 	kfree(thread);
8076 }
8077 EXPORT_SYMBOL(md_unregister_thread);
8078 
8079 void md_error(struct mddev *mddev, struct md_rdev *rdev)
8080 {
8081 	if (!rdev || test_bit(Faulty, &rdev->flags))
8082 		return;
8083 
8084 	if (!mddev->pers || !mddev->pers->error_handler)
8085 		return;
8086 	mddev->pers->error_handler(mddev, rdev);
8087 
8088 	if (mddev->pers->level == 0 || mddev->pers->level == LEVEL_LINEAR)
8089 		return;
8090 
8091 	if (mddev->degraded && !test_bit(MD_BROKEN, &mddev->flags))
8092 		set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
8093 	sysfs_notify_dirent_safe(rdev->sysfs_state);
8094 	set_bit(MD_RECOVERY_INTR, &mddev->recovery);
8095 	if (!test_bit(MD_BROKEN, &mddev->flags)) {
8096 		set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
8097 		md_wakeup_thread(mddev->thread);
8098 	}
8099 	if (mddev->event_work.func)
8100 		queue_work(md_misc_wq, &mddev->event_work);
8101 	md_new_event();
8102 }
8103 EXPORT_SYMBOL(md_error);
8104 
8105 /* seq_file implementation /proc/mdstat */
8106 
8107 static void status_unused(struct seq_file *seq)
8108 {
8109 	int i = 0;
8110 	struct md_rdev *rdev;
8111 
8112 	seq_printf(seq, "unused devices: ");
8113 
8114 	list_for_each_entry(rdev, &pending_raid_disks, same_set) {
8115 		i++;
8116 		seq_printf(seq, "%pg ", rdev->bdev);
8117 	}
8118 	if (!i)
8119 		seq_printf(seq, "<none>");
8120 
8121 	seq_printf(seq, "\n");
8122 }
8123 
8124 static int status_resync(struct seq_file *seq, struct mddev *mddev)
8125 {
8126 	sector_t max_sectors, resync, res;
8127 	unsigned long dt, db = 0;
8128 	sector_t rt, curr_mark_cnt, resync_mark_cnt;
8129 	int scale, recovery_active;
8130 	unsigned int per_milli;
8131 
8132 	if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ||
8133 	    test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
8134 		max_sectors = mddev->resync_max_sectors;
8135 	else
8136 		max_sectors = mddev->dev_sectors;
8137 
8138 	resync = mddev->curr_resync;
8139 	if (resync < MD_RESYNC_ACTIVE) {
8140 		if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
8141 			/* Still cleaning up */
8142 			resync = max_sectors;
8143 	} else if (resync > max_sectors) {
8144 		resync = max_sectors;
8145 	} else {
8146 		res = atomic_read(&mddev->recovery_active);
8147 		/*
8148 		 * Resync has started, but the subtraction has overflowed or
8149 		 * yielded one of the special values. Force it to active to
8150 		 * ensure the status reports an active resync.
8151 		 */
8152 		if (resync < res || resync - res < MD_RESYNC_ACTIVE)
8153 			resync = MD_RESYNC_ACTIVE;
8154 		else
8155 			resync -= res;
8156 	}
8157 
8158 	if (resync == MD_RESYNC_NONE) {
8159 		if (test_bit(MD_RESYNCING_REMOTE, &mddev->recovery)) {
8160 			struct md_rdev *rdev;
8161 
8162 			rdev_for_each(rdev, mddev)
8163 				if (rdev->raid_disk >= 0 &&
8164 				    !test_bit(Faulty, &rdev->flags) &&
8165 				    rdev->recovery_offset != MaxSector &&
8166 				    rdev->recovery_offset) {
8167 					seq_printf(seq, "\trecover=REMOTE");
8168 					return 1;
8169 				}
8170 			if (mddev->reshape_position != MaxSector)
8171 				seq_printf(seq, "\treshape=REMOTE");
8172 			else
8173 				seq_printf(seq, "\tresync=REMOTE");
8174 			return 1;
8175 		}
8176 		if (mddev->recovery_cp < MaxSector) {
8177 			seq_printf(seq, "\tresync=PENDING");
8178 			return 1;
8179 		}
8180 		return 0;
8181 	}
8182 	if (resync < MD_RESYNC_ACTIVE) {
8183 		seq_printf(seq, "\tresync=DELAYED");
8184 		return 1;
8185 	}
8186 
8187 	WARN_ON(max_sectors == 0);
8188 	/* Pick 'scale' such that (resync>>scale)*1000 will fit
8189 	 * in a sector_t, and (max_sectors>>scale) will fit in a
8190 	 * u32, as those are the requirements for sector_div.
8191 	 * Thus 'scale' must be at least 10
8192 	 */
8193 	scale = 10;
8194 	if (sizeof(sector_t) > sizeof(unsigned long)) {
8195 		while ( max_sectors/2 > (1ULL<<(scale+32)))
8196 			scale++;
8197 	}
8198 	res = (resync>>scale)*1000;
8199 	sector_div(res, (u32)((max_sectors>>scale)+1));
8200 
8201 	per_milli = res;
8202 	{
8203 		int i, x = per_milli/50, y = 20-x;
8204 		seq_printf(seq, "[");
8205 		for (i = 0; i < x; i++)
8206 			seq_printf(seq, "=");
8207 		seq_printf(seq, ">");
8208 		for (i = 0; i < y; i++)
8209 			seq_printf(seq, ".");
8210 		seq_printf(seq, "] ");
8211 	}
8212 	seq_printf(seq, " %s =%3u.%u%% (%llu/%llu)",
8213 		   (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)?
8214 		    "reshape" :
8215 		    (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)?
8216 		     "check" :
8217 		     (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
8218 		      "resync" : "recovery"))),
8219 		   per_milli/10, per_milli % 10,
8220 		   (unsigned long long) resync/2,
8221 		   (unsigned long long) max_sectors/2);
8222 
8223 	/*
8224 	 * dt: time from mark until now
8225 	 * db: blocks written from mark until now
8226 	 * rt: remaining time
8227 	 *
8228 	 * rt is a sector_t, which is always 64bit now. We are keeping
8229 	 * the original algorithm, but it is not really necessary.
8230 	 *
8231 	 * Original algorithm:
8232 	 *   So we divide before multiply in case it is 32bit and close
8233 	 *   to the limit.
8234 	 *   We scale the divisor (db) by 32 to avoid losing precision
8235 	 *   near the end of resync when the number of remaining sectors
8236 	 *   is close to 'db'.
8237 	 *   We then divide rt by 32 after multiplying by db to compensate.
8238 	 *   The '+1' avoids division by zero if db is very small.
8239 	 */
8240 	dt = ((jiffies - mddev->resync_mark) / HZ);
8241 	if (!dt) dt++;
8242 
8243 	curr_mark_cnt = mddev->curr_mark_cnt;
8244 	recovery_active = atomic_read(&mddev->recovery_active);
8245 	resync_mark_cnt = mddev->resync_mark_cnt;
8246 
8247 	if (curr_mark_cnt >= (recovery_active + resync_mark_cnt))
8248 		db = curr_mark_cnt - (recovery_active + resync_mark_cnt);
8249 
8250 	rt = max_sectors - resync;    /* number of remaining sectors */
8251 	rt = div64_u64(rt, db/32+1);
8252 	rt *= dt;
8253 	rt >>= 5;
8254 
8255 	seq_printf(seq, " finish=%lu.%lumin", (unsigned long)rt / 60,
8256 		   ((unsigned long)rt % 60)/6);
8257 
8258 	seq_printf(seq, " speed=%ldK/sec", db/2/dt);
8259 	return 1;
8260 }
8261 
8262 static void *md_seq_start(struct seq_file *seq, loff_t *pos)
8263 {
8264 	struct list_head *tmp;
8265 	loff_t l = *pos;
8266 	struct mddev *mddev;
8267 
8268 	if (l == 0x10000) {
8269 		++*pos;
8270 		return (void *)2;
8271 	}
8272 	if (l > 0x10000)
8273 		return NULL;
8274 	if (!l--)
8275 		/* header */
8276 		return (void*)1;
8277 
8278 	spin_lock(&all_mddevs_lock);
8279 	list_for_each(tmp,&all_mddevs)
8280 		if (!l--) {
8281 			mddev = list_entry(tmp, struct mddev, all_mddevs);
8282 			if (!mddev_get(mddev))
8283 				continue;
8284 			spin_unlock(&all_mddevs_lock);
8285 			return mddev;
8286 		}
8287 	spin_unlock(&all_mddevs_lock);
8288 	if (!l--)
8289 		return (void*)2;/* tail */
8290 	return NULL;
8291 }
8292 
8293 static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
8294 {
8295 	struct list_head *tmp;
8296 	struct mddev *next_mddev, *mddev = v;
8297 	struct mddev *to_put = NULL;
8298 
8299 	++*pos;
8300 	if (v == (void*)2)
8301 		return NULL;
8302 
8303 	spin_lock(&all_mddevs_lock);
8304 	if (v == (void*)1) {
8305 		tmp = all_mddevs.next;
8306 	} else {
8307 		to_put = mddev;
8308 		tmp = mddev->all_mddevs.next;
8309 	}
8310 
8311 	for (;;) {
8312 		if (tmp == &all_mddevs) {
8313 			next_mddev = (void*)2;
8314 			*pos = 0x10000;
8315 			break;
8316 		}
8317 		next_mddev = list_entry(tmp, struct mddev, all_mddevs);
8318 		if (mddev_get(next_mddev))
8319 			break;
8320 		mddev = next_mddev;
8321 		tmp = mddev->all_mddevs.next;
8322 	}
8323 	spin_unlock(&all_mddevs_lock);
8324 
8325 	if (to_put)
8326 		mddev_put(to_put);
8327 	return next_mddev;
8328 
8329 }
8330 
8331 static void md_seq_stop(struct seq_file *seq, void *v)
8332 {
8333 	struct mddev *mddev = v;
8334 
8335 	if (mddev && v != (void*)1 && v != (void*)2)
8336 		mddev_put(mddev);
8337 }
8338 
8339 static int md_seq_show(struct seq_file *seq, void *v)
8340 {
8341 	struct mddev *mddev = v;
8342 	sector_t sectors;
8343 	struct md_rdev *rdev;
8344 
8345 	if (v == (void*)1) {
8346 		struct md_personality *pers;
8347 		seq_printf(seq, "Personalities : ");
8348 		spin_lock(&pers_lock);
8349 		list_for_each_entry(pers, &pers_list, list)
8350 			seq_printf(seq, "[%s] ", pers->name);
8351 
8352 		spin_unlock(&pers_lock);
8353 		seq_printf(seq, "\n");
8354 		seq->poll_event = atomic_read(&md_event_count);
8355 		return 0;
8356 	}
8357 	if (v == (void*)2) {
8358 		status_unused(seq);
8359 		return 0;
8360 	}
8361 
8362 	spin_lock(&mddev->lock);
8363 	if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
8364 		seq_printf(seq, "%s : %sactive", mdname(mddev),
8365 						mddev->pers ? "" : "in");
8366 		if (mddev->pers) {
8367 			if (mddev->ro == MD_RDONLY)
8368 				seq_printf(seq, " (read-only)");
8369 			if (mddev->ro == MD_AUTO_READ)
8370 				seq_printf(seq, " (auto-read-only)");
8371 			seq_printf(seq, " %s", mddev->pers->name);
8372 		}
8373 
8374 		sectors = 0;
8375 		rcu_read_lock();
8376 		rdev_for_each_rcu(rdev, mddev) {
8377 			seq_printf(seq, " %pg[%d]", rdev->bdev, rdev->desc_nr);
8378 
8379 			if (test_bit(WriteMostly, &rdev->flags))
8380 				seq_printf(seq, "(W)");
8381 			if (test_bit(Journal, &rdev->flags))
8382 				seq_printf(seq, "(J)");
8383 			if (test_bit(Faulty, &rdev->flags)) {
8384 				seq_printf(seq, "(F)");
8385 				continue;
8386 			}
8387 			if (rdev->raid_disk < 0)
8388 				seq_printf(seq, "(S)"); /* spare */
8389 			if (test_bit(Replacement, &rdev->flags))
8390 				seq_printf(seq, "(R)");
8391 			sectors += rdev->sectors;
8392 		}
8393 		rcu_read_unlock();
8394 
8395 		if (!list_empty(&mddev->disks)) {
8396 			if (mddev->pers)
8397 				seq_printf(seq, "\n      %llu blocks",
8398 					   (unsigned long long)
8399 					   mddev->array_sectors / 2);
8400 			else
8401 				seq_printf(seq, "\n      %llu blocks",
8402 					   (unsigned long long)sectors / 2);
8403 		}
8404 		if (mddev->persistent) {
8405 			if (mddev->major_version != 0 ||
8406 			    mddev->minor_version != 90) {
8407 				seq_printf(seq," super %d.%d",
8408 					   mddev->major_version,
8409 					   mddev->minor_version);
8410 			}
8411 		} else if (mddev->external)
8412 			seq_printf(seq, " super external:%s",
8413 				   mddev->metadata_type);
8414 		else
8415 			seq_printf(seq, " super non-persistent");
8416 
8417 		if (mddev->pers) {
8418 			mddev->pers->status(seq, mddev);
8419 			seq_printf(seq, "\n      ");
8420 			if (mddev->pers->sync_request) {
8421 				if (status_resync(seq, mddev))
8422 					seq_printf(seq, "\n      ");
8423 			}
8424 		} else
8425 			seq_printf(seq, "\n       ");
8426 
8427 		md_bitmap_status(seq, mddev->bitmap);
8428 
8429 		seq_printf(seq, "\n");
8430 	}
8431 	spin_unlock(&mddev->lock);
8432 
8433 	return 0;
8434 }
8435 
8436 static const struct seq_operations md_seq_ops = {
8437 	.start  = md_seq_start,
8438 	.next   = md_seq_next,
8439 	.stop   = md_seq_stop,
8440 	.show   = md_seq_show,
8441 };
8442 
8443 static int md_seq_open(struct inode *inode, struct file *file)
8444 {
8445 	struct seq_file *seq;
8446 	int error;
8447 
8448 	error = seq_open(file, &md_seq_ops);
8449 	if (error)
8450 		return error;
8451 
8452 	seq = file->private_data;
8453 	seq->poll_event = atomic_read(&md_event_count);
8454 	return error;
8455 }
8456 
8457 static int md_unloading;
8458 static __poll_t mdstat_poll(struct file *filp, poll_table *wait)
8459 {
8460 	struct seq_file *seq = filp->private_data;
8461 	__poll_t mask;
8462 
8463 	if (md_unloading)
8464 		return EPOLLIN|EPOLLRDNORM|EPOLLERR|EPOLLPRI;
8465 	poll_wait(filp, &md_event_waiters, wait);
8466 
8467 	/* always allow read */
8468 	mask = EPOLLIN | EPOLLRDNORM;
8469 
8470 	if (seq->poll_event != atomic_read(&md_event_count))
8471 		mask |= EPOLLERR | EPOLLPRI;
8472 	return mask;
8473 }
8474 
8475 static const struct proc_ops mdstat_proc_ops = {
8476 	.proc_open	= md_seq_open,
8477 	.proc_read	= seq_read,
8478 	.proc_lseek	= seq_lseek,
8479 	.proc_release	= seq_release,
8480 	.proc_poll	= mdstat_poll,
8481 };
8482 
8483 int register_md_personality(struct md_personality *p)
8484 {
8485 	pr_debug("md: %s personality registered for level %d\n",
8486 		 p->name, p->level);
8487 	spin_lock(&pers_lock);
8488 	list_add_tail(&p->list, &pers_list);
8489 	spin_unlock(&pers_lock);
8490 	return 0;
8491 }
8492 EXPORT_SYMBOL(register_md_personality);
8493 
8494 int unregister_md_personality(struct md_personality *p)
8495 {
8496 	pr_debug("md: %s personality unregistered\n", p->name);
8497 	spin_lock(&pers_lock);
8498 	list_del_init(&p->list);
8499 	spin_unlock(&pers_lock);
8500 	return 0;
8501 }
8502 EXPORT_SYMBOL(unregister_md_personality);
8503 
8504 int register_md_cluster_operations(struct md_cluster_operations *ops,
8505 				   struct module *module)
8506 {
8507 	int ret = 0;
8508 	spin_lock(&pers_lock);
8509 	if (md_cluster_ops != NULL)
8510 		ret = -EALREADY;
8511 	else {
8512 		md_cluster_ops = ops;
8513 		md_cluster_mod = module;
8514 	}
8515 	spin_unlock(&pers_lock);
8516 	return ret;
8517 }
8518 EXPORT_SYMBOL(register_md_cluster_operations);
8519 
8520 int unregister_md_cluster_operations(void)
8521 {
8522 	spin_lock(&pers_lock);
8523 	md_cluster_ops = NULL;
8524 	spin_unlock(&pers_lock);
8525 	return 0;
8526 }
8527 EXPORT_SYMBOL(unregister_md_cluster_operations);
8528 
8529 int md_setup_cluster(struct mddev *mddev, int nodes)
8530 {
8531 	int ret;
8532 	if (!md_cluster_ops)
8533 		request_module("md-cluster");
8534 	spin_lock(&pers_lock);
8535 	/* ensure module won't be unloaded */
8536 	if (!md_cluster_ops || !try_module_get(md_cluster_mod)) {
8537 		pr_warn("can't find md-cluster module or get its reference.\n");
8538 		spin_unlock(&pers_lock);
8539 		return -ENOENT;
8540 	}
8541 	spin_unlock(&pers_lock);
8542 
8543 	ret = md_cluster_ops->join(mddev, nodes);
8544 	if (!ret)
8545 		mddev->safemode_delay = 0;
8546 	return ret;
8547 }
8548 
8549 void md_cluster_stop(struct mddev *mddev)
8550 {
8551 	if (!md_cluster_ops)
8552 		return;
8553 	md_cluster_ops->leave(mddev);
8554 	module_put(md_cluster_mod);
8555 }
8556 
8557 static int is_mddev_idle(struct mddev *mddev, int init)
8558 {
8559 	struct md_rdev *rdev;
8560 	int idle;
8561 	int curr_events;
8562 
8563 	idle = 1;
8564 	rcu_read_lock();
8565 	rdev_for_each_rcu(rdev, mddev) {
8566 		struct gendisk *disk = rdev->bdev->bd_disk;
8567 		curr_events = (int)part_stat_read_accum(disk->part0, sectors) -
8568 			      atomic_read(&disk->sync_io);
8569 		/* sync IO will cause sync_io to increase before the disk_stats
8570 		 * as sync_io is counted when a request starts, and
8571 		 * disk_stats is counted when it completes.
8572 		 * So resync activity will cause curr_events to be smaller than
8573 		 * when there was no such activity.
8574 		 * non-sync IO will cause disk_stat to increase without
8575 		 * increasing sync_io so curr_events will (eventually)
8576 		 * be larger than it was before.  Once it becomes
8577 		 * substantially larger, the test below will cause
8578 		 * the array to appear non-idle, and resync will slow
8579 		 * down.
8580 		 * If there is a lot of outstanding resync activity when
8581 		 * we set last_event to curr_events, then all that activity
8582 		 * completing might cause the array to appear non-idle
8583 		 * and resync will be slowed down even though there might
8584 		 * not have been non-resync activity.  This will only
8585 		 * happen once though.  'last_events' will soon reflect
8586 		 * the state where there is little or no outstanding
8587 		 * resync requests, and further resync activity will
8588 		 * always make curr_events less than last_events.
8589 		 *
8590 		 */
8591 		if (init || curr_events - rdev->last_events > 64) {
8592 			rdev->last_events = curr_events;
8593 			idle = 0;
8594 		}
8595 	}
8596 	rcu_read_unlock();
8597 	return idle;
8598 }
8599 
8600 void md_done_sync(struct mddev *mddev, int blocks, int ok)
8601 {
8602 	/* another "blocks" (512byte) blocks have been synced */
8603 	atomic_sub(blocks, &mddev->recovery_active);
8604 	wake_up(&mddev->recovery_wait);
8605 	if (!ok) {
8606 		set_bit(MD_RECOVERY_INTR, &mddev->recovery);
8607 		set_bit(MD_RECOVERY_ERROR, &mddev->recovery);
8608 		md_wakeup_thread(mddev->thread);
8609 		// stop recovery, signal do_sync ....
8610 	}
8611 }
8612 EXPORT_SYMBOL(md_done_sync);
8613 
8614 /* md_write_start(mddev, bi)
8615  * If we need to update some array metadata (e.g. 'active' flag
8616  * in superblock) before writing, schedule a superblock update
8617  * and wait for it to complete.
8618  * A return value of 'false' means that the write wasn't recorded
8619  * and cannot proceed as the array is being suspend.
8620  */
8621 bool md_write_start(struct mddev *mddev, struct bio *bi)
8622 {
8623 	int did_change = 0;
8624 
8625 	if (bio_data_dir(bi) != WRITE)
8626 		return true;
8627 
8628 	BUG_ON(mddev->ro == MD_RDONLY);
8629 	if (mddev->ro == MD_AUTO_READ) {
8630 		/* need to switch to read/write */
8631 		mddev->ro = MD_RDWR;
8632 		set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
8633 		md_wakeup_thread(mddev->thread);
8634 		md_wakeup_thread(mddev->sync_thread);
8635 		did_change = 1;
8636 	}
8637 	rcu_read_lock();
8638 	percpu_ref_get(&mddev->writes_pending);
8639 	smp_mb(); /* Match smp_mb in set_in_sync() */
8640 	if (mddev->safemode == 1)
8641 		mddev->safemode = 0;
8642 	/* sync_checkers is always 0 when writes_pending is in per-cpu mode */
8643 	if (mddev->in_sync || mddev->sync_checkers) {
8644 		spin_lock(&mddev->lock);
8645 		if (mddev->in_sync) {
8646 			mddev->in_sync = 0;
8647 			set_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags);
8648 			set_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags);
8649 			md_wakeup_thread(mddev->thread);
8650 			did_change = 1;
8651 		}
8652 		spin_unlock(&mddev->lock);
8653 	}
8654 	rcu_read_unlock();
8655 	if (did_change)
8656 		sysfs_notify_dirent_safe(mddev->sysfs_state);
8657 	if (!mddev->has_superblocks)
8658 		return true;
8659 	wait_event(mddev->sb_wait,
8660 		   !test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags) ||
8661 		   is_md_suspended(mddev));
8662 	if (test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags)) {
8663 		percpu_ref_put(&mddev->writes_pending);
8664 		return false;
8665 	}
8666 	return true;
8667 }
8668 EXPORT_SYMBOL(md_write_start);
8669 
8670 /* md_write_inc can only be called when md_write_start() has
8671  * already been called at least once of the current request.
8672  * It increments the counter and is useful when a single request
8673  * is split into several parts.  Each part causes an increment and
8674  * so needs a matching md_write_end().
8675  * Unlike md_write_start(), it is safe to call md_write_inc() inside
8676  * a spinlocked region.
8677  */
8678 void md_write_inc(struct mddev *mddev, struct bio *bi)
8679 {
8680 	if (bio_data_dir(bi) != WRITE)
8681 		return;
8682 	WARN_ON_ONCE(mddev->in_sync || !md_is_rdwr(mddev));
8683 	percpu_ref_get(&mddev->writes_pending);
8684 }
8685 EXPORT_SYMBOL(md_write_inc);
8686 
8687 void md_write_end(struct mddev *mddev)
8688 {
8689 	percpu_ref_put(&mddev->writes_pending);
8690 
8691 	if (mddev->safemode == 2)
8692 		md_wakeup_thread(mddev->thread);
8693 	else if (mddev->safemode_delay)
8694 		/* The roundup() ensures this only performs locking once
8695 		 * every ->safemode_delay jiffies
8696 		 */
8697 		mod_timer(&mddev->safemode_timer,
8698 			  roundup(jiffies, mddev->safemode_delay) +
8699 			  mddev->safemode_delay);
8700 }
8701 
8702 EXPORT_SYMBOL(md_write_end);
8703 
8704 /* This is used by raid0 and raid10 */
8705 void md_submit_discard_bio(struct mddev *mddev, struct md_rdev *rdev,
8706 			struct bio *bio, sector_t start, sector_t size)
8707 {
8708 	struct bio *discard_bio = NULL;
8709 
8710 	if (__blkdev_issue_discard(rdev->bdev, start, size, GFP_NOIO,
8711 			&discard_bio) || !discard_bio)
8712 		return;
8713 
8714 	bio_chain(discard_bio, bio);
8715 	bio_clone_blkg_association(discard_bio, bio);
8716 	if (mddev->gendisk)
8717 		trace_block_bio_remap(discard_bio,
8718 				disk_devt(mddev->gendisk),
8719 				bio->bi_iter.bi_sector);
8720 	submit_bio_noacct(discard_bio);
8721 }
8722 EXPORT_SYMBOL_GPL(md_submit_discard_bio);
8723 
8724 static void md_end_clone_io(struct bio *bio)
8725 {
8726 	struct md_io_clone *md_io_clone = bio->bi_private;
8727 	struct bio *orig_bio = md_io_clone->orig_bio;
8728 	struct mddev *mddev = md_io_clone->mddev;
8729 
8730 	if (bio->bi_status && !orig_bio->bi_status)
8731 		orig_bio->bi_status = bio->bi_status;
8732 
8733 	if (md_io_clone->start_time)
8734 		bio_end_io_acct(orig_bio, md_io_clone->start_time);
8735 
8736 	bio_put(bio);
8737 	bio_endio(orig_bio);
8738 	percpu_ref_put(&mddev->active_io);
8739 }
8740 
8741 static void md_clone_bio(struct mddev *mddev, struct bio **bio)
8742 {
8743 	struct block_device *bdev = (*bio)->bi_bdev;
8744 	struct md_io_clone *md_io_clone;
8745 	struct bio *clone =
8746 		bio_alloc_clone(bdev, *bio, GFP_NOIO, &mddev->io_clone_set);
8747 
8748 	md_io_clone = container_of(clone, struct md_io_clone, bio_clone);
8749 	md_io_clone->orig_bio = *bio;
8750 	md_io_clone->mddev = mddev;
8751 	if (blk_queue_io_stat(bdev->bd_disk->queue))
8752 		md_io_clone->start_time = bio_start_io_acct(*bio);
8753 
8754 	clone->bi_end_io = md_end_clone_io;
8755 	clone->bi_private = md_io_clone;
8756 	*bio = clone;
8757 }
8758 
8759 void md_account_bio(struct mddev *mddev, struct bio **bio)
8760 {
8761 	percpu_ref_get(&mddev->active_io);
8762 	md_clone_bio(mddev, bio);
8763 }
8764 EXPORT_SYMBOL_GPL(md_account_bio);
8765 
8766 /* md_allow_write(mddev)
8767  * Calling this ensures that the array is marked 'active' so that writes
8768  * may proceed without blocking.  It is important to call this before
8769  * attempting a GFP_KERNEL allocation while holding the mddev lock.
8770  * Must be called with mddev_lock held.
8771  */
8772 void md_allow_write(struct mddev *mddev)
8773 {
8774 	if (!mddev->pers)
8775 		return;
8776 	if (!md_is_rdwr(mddev))
8777 		return;
8778 	if (!mddev->pers->sync_request)
8779 		return;
8780 
8781 	spin_lock(&mddev->lock);
8782 	if (mddev->in_sync) {
8783 		mddev->in_sync = 0;
8784 		set_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags);
8785 		set_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags);
8786 		if (mddev->safemode_delay &&
8787 		    mddev->safemode == 0)
8788 			mddev->safemode = 1;
8789 		spin_unlock(&mddev->lock);
8790 		md_update_sb(mddev, 0);
8791 		sysfs_notify_dirent_safe(mddev->sysfs_state);
8792 		/* wait for the dirty state to be recorded in the metadata */
8793 		wait_event(mddev->sb_wait,
8794 			   !test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags));
8795 	} else
8796 		spin_unlock(&mddev->lock);
8797 }
8798 EXPORT_SYMBOL_GPL(md_allow_write);
8799 
8800 #define SYNC_MARKS	10
8801 #define	SYNC_MARK_STEP	(3*HZ)
8802 #define UPDATE_FREQUENCY (5*60*HZ)
8803 void md_do_sync(struct md_thread *thread)
8804 {
8805 	struct mddev *mddev = thread->mddev;
8806 	struct mddev *mddev2;
8807 	unsigned int currspeed = 0, window;
8808 	sector_t max_sectors,j, io_sectors, recovery_done;
8809 	unsigned long mark[SYNC_MARKS];
8810 	unsigned long update_time;
8811 	sector_t mark_cnt[SYNC_MARKS];
8812 	int last_mark,m;
8813 	sector_t last_check;
8814 	int skipped = 0;
8815 	struct md_rdev *rdev;
8816 	char *desc, *action = NULL;
8817 	struct blk_plug plug;
8818 	int ret;
8819 
8820 	/* just incase thread restarts... */
8821 	if (test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
8822 	    test_bit(MD_RECOVERY_WAIT, &mddev->recovery))
8823 		return;
8824 	if (!md_is_rdwr(mddev)) {/* never try to sync a read-only array */
8825 		set_bit(MD_RECOVERY_INTR, &mddev->recovery);
8826 		return;
8827 	}
8828 
8829 	if (mddev_is_clustered(mddev)) {
8830 		ret = md_cluster_ops->resync_start(mddev);
8831 		if (ret)
8832 			goto skip;
8833 
8834 		set_bit(MD_CLUSTER_RESYNC_LOCKED, &mddev->flags);
8835 		if (!(test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ||
8836 			test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) ||
8837 			test_bit(MD_RECOVERY_RECOVER, &mddev->recovery))
8838 		     && ((unsigned long long)mddev->curr_resync_completed
8839 			 < (unsigned long long)mddev->resync_max_sectors))
8840 			goto skip;
8841 	}
8842 
8843 	if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
8844 		if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)) {
8845 			desc = "data-check";
8846 			action = "check";
8847 		} else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
8848 			desc = "requested-resync";
8849 			action = "repair";
8850 		} else
8851 			desc = "resync";
8852 	} else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
8853 		desc = "reshape";
8854 	else
8855 		desc = "recovery";
8856 
8857 	mddev->last_sync_action = action ?: desc;
8858 
8859 	/*
8860 	 * Before starting a resync we must have set curr_resync to
8861 	 * 2, and then checked that every "conflicting" array has curr_resync
8862 	 * less than ours.  When we find one that is the same or higher
8863 	 * we wait on resync_wait.  To avoid deadlock, we reduce curr_resync
8864 	 * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
8865 	 * This will mean we have to start checking from the beginning again.
8866 	 *
8867 	 */
8868 
8869 	do {
8870 		int mddev2_minor = -1;
8871 		mddev->curr_resync = MD_RESYNC_DELAYED;
8872 
8873 	try_again:
8874 		if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
8875 			goto skip;
8876 		spin_lock(&all_mddevs_lock);
8877 		list_for_each_entry(mddev2, &all_mddevs, all_mddevs) {
8878 			if (test_bit(MD_DELETED, &mddev2->flags))
8879 				continue;
8880 			if (mddev2 == mddev)
8881 				continue;
8882 			if (!mddev->parallel_resync
8883 			&&  mddev2->curr_resync
8884 			&&  match_mddev_units(mddev, mddev2)) {
8885 				DEFINE_WAIT(wq);
8886 				if (mddev < mddev2 &&
8887 				    mddev->curr_resync == MD_RESYNC_DELAYED) {
8888 					/* arbitrarily yield */
8889 					mddev->curr_resync = MD_RESYNC_YIELDED;
8890 					wake_up(&resync_wait);
8891 				}
8892 				if (mddev > mddev2 &&
8893 				    mddev->curr_resync == MD_RESYNC_YIELDED)
8894 					/* no need to wait here, we can wait the next
8895 					 * time 'round when curr_resync == 2
8896 					 */
8897 					continue;
8898 				/* We need to wait 'interruptible' so as not to
8899 				 * contribute to the load average, and not to
8900 				 * be caught by 'softlockup'
8901 				 */
8902 				prepare_to_wait(&resync_wait, &wq, TASK_INTERRUPTIBLE);
8903 				if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
8904 				    mddev2->curr_resync >= mddev->curr_resync) {
8905 					if (mddev2_minor != mddev2->md_minor) {
8906 						mddev2_minor = mddev2->md_minor;
8907 						pr_info("md: delaying %s of %s until %s has finished (they share one or more physical units)\n",
8908 							desc, mdname(mddev),
8909 							mdname(mddev2));
8910 					}
8911 					spin_unlock(&all_mddevs_lock);
8912 
8913 					if (signal_pending(current))
8914 						flush_signals(current);
8915 					schedule();
8916 					finish_wait(&resync_wait, &wq);
8917 					goto try_again;
8918 				}
8919 				finish_wait(&resync_wait, &wq);
8920 			}
8921 		}
8922 		spin_unlock(&all_mddevs_lock);
8923 	} while (mddev->curr_resync < MD_RESYNC_DELAYED);
8924 
8925 	j = 0;
8926 	if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
8927 		/* resync follows the size requested by the personality,
8928 		 * which defaults to physical size, but can be virtual size
8929 		 */
8930 		max_sectors = mddev->resync_max_sectors;
8931 		atomic64_set(&mddev->resync_mismatches, 0);
8932 		/* we don't use the checkpoint if there's a bitmap */
8933 		if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
8934 			j = mddev->resync_min;
8935 		else if (!mddev->bitmap)
8936 			j = mddev->recovery_cp;
8937 
8938 	} else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) {
8939 		max_sectors = mddev->resync_max_sectors;
8940 		/*
8941 		 * If the original node aborts reshaping then we continue the
8942 		 * reshaping, so set j again to avoid restart reshape from the
8943 		 * first beginning
8944 		 */
8945 		if (mddev_is_clustered(mddev) &&
8946 		    mddev->reshape_position != MaxSector)
8947 			j = mddev->reshape_position;
8948 	} else {
8949 		/* recovery follows the physical size of devices */
8950 		max_sectors = mddev->dev_sectors;
8951 		j = MaxSector;
8952 		rcu_read_lock();
8953 		rdev_for_each_rcu(rdev, mddev)
8954 			if (rdev->raid_disk >= 0 &&
8955 			    !test_bit(Journal, &rdev->flags) &&
8956 			    !test_bit(Faulty, &rdev->flags) &&
8957 			    !test_bit(In_sync, &rdev->flags) &&
8958 			    rdev->recovery_offset < j)
8959 				j = rdev->recovery_offset;
8960 		rcu_read_unlock();
8961 
8962 		/* If there is a bitmap, we need to make sure all
8963 		 * writes that started before we added a spare
8964 		 * complete before we start doing a recovery.
8965 		 * Otherwise the write might complete and (via
8966 		 * bitmap_endwrite) set a bit in the bitmap after the
8967 		 * recovery has checked that bit and skipped that
8968 		 * region.
8969 		 */
8970 		if (mddev->bitmap) {
8971 			mddev->pers->quiesce(mddev, 1);
8972 			mddev->pers->quiesce(mddev, 0);
8973 		}
8974 	}
8975 
8976 	pr_info("md: %s of RAID array %s\n", desc, mdname(mddev));
8977 	pr_debug("md: minimum _guaranteed_  speed: %d KB/sec/disk.\n", speed_min(mddev));
8978 	pr_debug("md: using maximum available idle IO bandwidth (but not more than %d KB/sec) for %s.\n",
8979 		 speed_max(mddev), desc);
8980 
8981 	is_mddev_idle(mddev, 1); /* this initializes IO event counters */
8982 
8983 	io_sectors = 0;
8984 	for (m = 0; m < SYNC_MARKS; m++) {
8985 		mark[m] = jiffies;
8986 		mark_cnt[m] = io_sectors;
8987 	}
8988 	last_mark = 0;
8989 	mddev->resync_mark = mark[last_mark];
8990 	mddev->resync_mark_cnt = mark_cnt[last_mark];
8991 
8992 	/*
8993 	 * Tune reconstruction:
8994 	 */
8995 	window = 32 * (PAGE_SIZE / 512);
8996 	pr_debug("md: using %dk window, over a total of %lluk.\n",
8997 		 window/2, (unsigned long long)max_sectors/2);
8998 
8999 	atomic_set(&mddev->recovery_active, 0);
9000 	last_check = 0;
9001 
9002 	if (j >= MD_RESYNC_ACTIVE) {
9003 		pr_debug("md: resuming %s of %s from checkpoint.\n",
9004 			 desc, mdname(mddev));
9005 		mddev->curr_resync = j;
9006 	} else
9007 		mddev->curr_resync = MD_RESYNC_ACTIVE; /* no longer delayed */
9008 	mddev->curr_resync_completed = j;
9009 	sysfs_notify_dirent_safe(mddev->sysfs_completed);
9010 	md_new_event();
9011 	update_time = jiffies;
9012 
9013 	blk_start_plug(&plug);
9014 	while (j < max_sectors) {
9015 		sector_t sectors;
9016 
9017 		skipped = 0;
9018 
9019 		if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
9020 		    ((mddev->curr_resync > mddev->curr_resync_completed &&
9021 		      (mddev->curr_resync - mddev->curr_resync_completed)
9022 		      > (max_sectors >> 4)) ||
9023 		     time_after_eq(jiffies, update_time + UPDATE_FREQUENCY) ||
9024 		     (j - mddev->curr_resync_completed)*2
9025 		     >= mddev->resync_max - mddev->curr_resync_completed ||
9026 		     mddev->curr_resync_completed > mddev->resync_max
9027 			    )) {
9028 			/* time to update curr_resync_completed */
9029 			wait_event(mddev->recovery_wait,
9030 				   atomic_read(&mddev->recovery_active) == 0);
9031 			mddev->curr_resync_completed = j;
9032 			if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) &&
9033 			    j > mddev->recovery_cp)
9034 				mddev->recovery_cp = j;
9035 			update_time = jiffies;
9036 			set_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags);
9037 			sysfs_notify_dirent_safe(mddev->sysfs_completed);
9038 		}
9039 
9040 		while (j >= mddev->resync_max &&
9041 		       !test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
9042 			/* As this condition is controlled by user-space,
9043 			 * we can block indefinitely, so use '_interruptible'
9044 			 * to avoid triggering warnings.
9045 			 */
9046 			flush_signals(current); /* just in case */
9047 			wait_event_interruptible(mddev->recovery_wait,
9048 						 mddev->resync_max > j
9049 						 || test_bit(MD_RECOVERY_INTR,
9050 							     &mddev->recovery));
9051 		}
9052 
9053 		if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
9054 			break;
9055 
9056 		sectors = mddev->pers->sync_request(mddev, j, &skipped);
9057 		if (sectors == 0) {
9058 			set_bit(MD_RECOVERY_INTR, &mddev->recovery);
9059 			break;
9060 		}
9061 
9062 		if (!skipped) { /* actual IO requested */
9063 			io_sectors += sectors;
9064 			atomic_add(sectors, &mddev->recovery_active);
9065 		}
9066 
9067 		if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
9068 			break;
9069 
9070 		j += sectors;
9071 		if (j > max_sectors)
9072 			/* when skipping, extra large numbers can be returned. */
9073 			j = max_sectors;
9074 		if (j >= MD_RESYNC_ACTIVE)
9075 			mddev->curr_resync = j;
9076 		mddev->curr_mark_cnt = io_sectors;
9077 		if (last_check == 0)
9078 			/* this is the earliest that rebuild will be
9079 			 * visible in /proc/mdstat
9080 			 */
9081 			md_new_event();
9082 
9083 		if (last_check + window > io_sectors || j == max_sectors)
9084 			continue;
9085 
9086 		last_check = io_sectors;
9087 	repeat:
9088 		if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) {
9089 			/* step marks */
9090 			int next = (last_mark+1) % SYNC_MARKS;
9091 
9092 			mddev->resync_mark = mark[next];
9093 			mddev->resync_mark_cnt = mark_cnt[next];
9094 			mark[next] = jiffies;
9095 			mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active);
9096 			last_mark = next;
9097 		}
9098 
9099 		if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
9100 			break;
9101 
9102 		/*
9103 		 * this loop exits only if either when we are slower than
9104 		 * the 'hard' speed limit, or the system was IO-idle for
9105 		 * a jiffy.
9106 		 * the system might be non-idle CPU-wise, but we only care
9107 		 * about not overloading the IO subsystem. (things like an
9108 		 * e2fsck being done on the RAID array should execute fast)
9109 		 */
9110 		cond_resched();
9111 
9112 		recovery_done = io_sectors - atomic_read(&mddev->recovery_active);
9113 		currspeed = ((unsigned long)(recovery_done - mddev->resync_mark_cnt))/2
9114 			/((jiffies-mddev->resync_mark)/HZ +1) +1;
9115 
9116 		if (currspeed > speed_min(mddev)) {
9117 			if (currspeed > speed_max(mddev)) {
9118 				msleep(500);
9119 				goto repeat;
9120 			}
9121 			if (!is_mddev_idle(mddev, 0)) {
9122 				/*
9123 				 * Give other IO more of a chance.
9124 				 * The faster the devices, the less we wait.
9125 				 */
9126 				wait_event(mddev->recovery_wait,
9127 					   !atomic_read(&mddev->recovery_active));
9128 			}
9129 		}
9130 	}
9131 	pr_info("md: %s: %s %s.\n",mdname(mddev), desc,
9132 		test_bit(MD_RECOVERY_INTR, &mddev->recovery)
9133 		? "interrupted" : "done");
9134 	/*
9135 	 * this also signals 'finished resyncing' to md_stop
9136 	 */
9137 	blk_finish_plug(&plug);
9138 	wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
9139 
9140 	if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
9141 	    !test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
9142 	    mddev->curr_resync >= MD_RESYNC_ACTIVE) {
9143 		mddev->curr_resync_completed = mddev->curr_resync;
9144 		sysfs_notify_dirent_safe(mddev->sysfs_completed);
9145 	}
9146 	mddev->pers->sync_request(mddev, max_sectors, &skipped);
9147 
9148 	if (!test_bit(MD_RECOVERY_CHECK, &mddev->recovery) &&
9149 	    mddev->curr_resync > MD_RESYNC_ACTIVE) {
9150 		if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
9151 			if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
9152 				if (mddev->curr_resync >= mddev->recovery_cp) {
9153 					pr_debug("md: checkpointing %s of %s.\n",
9154 						 desc, mdname(mddev));
9155 					if (test_bit(MD_RECOVERY_ERROR,
9156 						&mddev->recovery))
9157 						mddev->recovery_cp =
9158 							mddev->curr_resync_completed;
9159 					else
9160 						mddev->recovery_cp =
9161 							mddev->curr_resync;
9162 				}
9163 			} else
9164 				mddev->recovery_cp = MaxSector;
9165 		} else {
9166 			if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery))
9167 				mddev->curr_resync = MaxSector;
9168 			if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
9169 			    test_bit(MD_RECOVERY_RECOVER, &mddev->recovery)) {
9170 				rcu_read_lock();
9171 				rdev_for_each_rcu(rdev, mddev)
9172 					if (rdev->raid_disk >= 0 &&
9173 					    mddev->delta_disks >= 0 &&
9174 					    !test_bit(Journal, &rdev->flags) &&
9175 					    !test_bit(Faulty, &rdev->flags) &&
9176 					    !test_bit(In_sync, &rdev->flags) &&
9177 					    rdev->recovery_offset < mddev->curr_resync)
9178 						rdev->recovery_offset = mddev->curr_resync;
9179 				rcu_read_unlock();
9180 			}
9181 		}
9182 	}
9183  skip:
9184 	/* set CHANGE_PENDING here since maybe another update is needed,
9185 	 * so other nodes are informed. It should be harmless for normal
9186 	 * raid */
9187 	set_mask_bits(&mddev->sb_flags, 0,
9188 		      BIT(MD_SB_CHANGE_PENDING) | BIT(MD_SB_CHANGE_DEVS));
9189 
9190 	if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
9191 			!test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
9192 			mddev->delta_disks > 0 &&
9193 			mddev->pers->finish_reshape &&
9194 			mddev->pers->size &&
9195 			mddev->queue) {
9196 		mddev_lock_nointr(mddev);
9197 		md_set_array_sectors(mddev, mddev->pers->size(mddev, 0, 0));
9198 		mddev_unlock(mddev);
9199 		if (!mddev_is_clustered(mddev))
9200 			set_capacity_and_notify(mddev->gendisk,
9201 						mddev->array_sectors);
9202 	}
9203 
9204 	spin_lock(&mddev->lock);
9205 	if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
9206 		/* We completed so min/max setting can be forgotten if used. */
9207 		if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
9208 			mddev->resync_min = 0;
9209 		mddev->resync_max = MaxSector;
9210 	} else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
9211 		mddev->resync_min = mddev->curr_resync_completed;
9212 	set_bit(MD_RECOVERY_DONE, &mddev->recovery);
9213 	mddev->curr_resync = MD_RESYNC_NONE;
9214 	spin_unlock(&mddev->lock);
9215 
9216 	wake_up(&resync_wait);
9217 	wake_up(&mddev->sb_wait);
9218 	md_wakeup_thread(mddev->thread);
9219 	return;
9220 }
9221 EXPORT_SYMBOL_GPL(md_do_sync);
9222 
9223 static int remove_and_add_spares(struct mddev *mddev,
9224 				 struct md_rdev *this)
9225 {
9226 	struct md_rdev *rdev;
9227 	int spares = 0;
9228 	int removed = 0;
9229 	bool remove_some = false;
9230 
9231 	if (this && test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
9232 		/* Mustn't remove devices when resync thread is running */
9233 		return 0;
9234 
9235 	rdev_for_each(rdev, mddev) {
9236 		if ((this == NULL || rdev == this) &&
9237 		    rdev->raid_disk >= 0 &&
9238 		    !test_bit(Blocked, &rdev->flags) &&
9239 		    test_bit(Faulty, &rdev->flags) &&
9240 		    atomic_read(&rdev->nr_pending)==0) {
9241 			/* Faulty non-Blocked devices with nr_pending == 0
9242 			 * never get nr_pending incremented,
9243 			 * never get Faulty cleared, and never get Blocked set.
9244 			 * So we can synchronize_rcu now rather than once per device
9245 			 */
9246 			remove_some = true;
9247 			set_bit(RemoveSynchronized, &rdev->flags);
9248 		}
9249 	}
9250 
9251 	if (remove_some)
9252 		synchronize_rcu();
9253 	rdev_for_each(rdev, mddev) {
9254 		if ((this == NULL || rdev == this) &&
9255 		    rdev->raid_disk >= 0 &&
9256 		    !test_bit(Blocked, &rdev->flags) &&
9257 		    ((test_bit(RemoveSynchronized, &rdev->flags) ||
9258 		     (!test_bit(In_sync, &rdev->flags) &&
9259 		      !test_bit(Journal, &rdev->flags))) &&
9260 		    atomic_read(&rdev->nr_pending)==0)) {
9261 			if (mddev->pers->hot_remove_disk(
9262 				    mddev, rdev) == 0) {
9263 				sysfs_unlink_rdev(mddev, rdev);
9264 				rdev->saved_raid_disk = rdev->raid_disk;
9265 				rdev->raid_disk = -1;
9266 				removed++;
9267 			}
9268 		}
9269 		if (remove_some && test_bit(RemoveSynchronized, &rdev->flags))
9270 			clear_bit(RemoveSynchronized, &rdev->flags);
9271 	}
9272 
9273 	if (removed && mddev->kobj.sd)
9274 		sysfs_notify_dirent_safe(mddev->sysfs_degraded);
9275 
9276 	if (this && removed)
9277 		goto no_add;
9278 
9279 	rdev_for_each(rdev, mddev) {
9280 		if (this && this != rdev)
9281 			continue;
9282 		if (test_bit(Candidate, &rdev->flags))
9283 			continue;
9284 		if (rdev->raid_disk >= 0 &&
9285 		    !test_bit(In_sync, &rdev->flags) &&
9286 		    !test_bit(Journal, &rdev->flags) &&
9287 		    !test_bit(Faulty, &rdev->flags))
9288 			spares++;
9289 		if (rdev->raid_disk >= 0)
9290 			continue;
9291 		if (test_bit(Faulty, &rdev->flags))
9292 			continue;
9293 		if (!test_bit(Journal, &rdev->flags)) {
9294 			if (!md_is_rdwr(mddev) &&
9295 			    !(rdev->saved_raid_disk >= 0 &&
9296 			      !test_bit(Bitmap_sync, &rdev->flags)))
9297 				continue;
9298 
9299 			rdev->recovery_offset = 0;
9300 		}
9301 		if (mddev->pers->hot_add_disk(mddev, rdev) == 0) {
9302 			/* failure here is OK */
9303 			sysfs_link_rdev(mddev, rdev);
9304 			if (!test_bit(Journal, &rdev->flags))
9305 				spares++;
9306 			md_new_event();
9307 			set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
9308 		}
9309 	}
9310 no_add:
9311 	if (removed)
9312 		set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
9313 	return spares;
9314 }
9315 
9316 static void md_start_sync(struct work_struct *ws)
9317 {
9318 	struct mddev *mddev = container_of(ws, struct mddev, del_work);
9319 
9320 	rcu_assign_pointer(mddev->sync_thread,
9321 			   md_register_thread(md_do_sync, mddev, "resync"));
9322 	if (!mddev->sync_thread) {
9323 		pr_warn("%s: could not start resync thread...\n",
9324 			mdname(mddev));
9325 		/* leave the spares where they are, it shouldn't hurt */
9326 		clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
9327 		clear_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
9328 		clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
9329 		clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
9330 		clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
9331 		wake_up(&resync_wait);
9332 		if (test_and_clear_bit(MD_RECOVERY_RECOVER,
9333 				       &mddev->recovery))
9334 			if (mddev->sysfs_action)
9335 				sysfs_notify_dirent_safe(mddev->sysfs_action);
9336 	} else
9337 		md_wakeup_thread(mddev->sync_thread);
9338 	sysfs_notify_dirent_safe(mddev->sysfs_action);
9339 	md_new_event();
9340 }
9341 
9342 /*
9343  * This routine is regularly called by all per-raid-array threads to
9344  * deal with generic issues like resync and super-block update.
9345  * Raid personalities that don't have a thread (linear/raid0) do not
9346  * need this as they never do any recovery or update the superblock.
9347  *
9348  * It does not do any resync itself, but rather "forks" off other threads
9349  * to do that as needed.
9350  * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
9351  * "->recovery" and create a thread at ->sync_thread.
9352  * When the thread finishes it sets MD_RECOVERY_DONE
9353  * and wakeups up this thread which will reap the thread and finish up.
9354  * This thread also removes any faulty devices (with nr_pending == 0).
9355  *
9356  * The overall approach is:
9357  *  1/ if the superblock needs updating, update it.
9358  *  2/ If a recovery thread is running, don't do anything else.
9359  *  3/ If recovery has finished, clean up, possibly marking spares active.
9360  *  4/ If there are any faulty devices, remove them.
9361  *  5/ If array is degraded, try to add spares devices
9362  *  6/ If array has spares or is not in-sync, start a resync thread.
9363  */
9364 void md_check_recovery(struct mddev *mddev)
9365 {
9366 	if (test_bit(MD_ALLOW_SB_UPDATE, &mddev->flags) && mddev->sb_flags) {
9367 		/* Write superblock - thread that called mddev_suspend()
9368 		 * holds reconfig_mutex for us.
9369 		 */
9370 		set_bit(MD_UPDATING_SB, &mddev->flags);
9371 		smp_mb__after_atomic();
9372 		if (test_bit(MD_ALLOW_SB_UPDATE, &mddev->flags))
9373 			md_update_sb(mddev, 0);
9374 		clear_bit_unlock(MD_UPDATING_SB, &mddev->flags);
9375 		wake_up(&mddev->sb_wait);
9376 	}
9377 
9378 	if (is_md_suspended(mddev))
9379 		return;
9380 
9381 	if (mddev->bitmap)
9382 		md_bitmap_daemon_work(mddev);
9383 
9384 	if (signal_pending(current)) {
9385 		if (mddev->pers->sync_request && !mddev->external) {
9386 			pr_debug("md: %s in immediate safe mode\n",
9387 				 mdname(mddev));
9388 			mddev->safemode = 2;
9389 		}
9390 		flush_signals(current);
9391 	}
9392 
9393 	if (!md_is_rdwr(mddev) &&
9394 	    !test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
9395 		return;
9396 	if ( ! (
9397 		(mddev->sb_flags & ~ (1<<MD_SB_CHANGE_PENDING)) ||
9398 		test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
9399 		test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
9400 		(mddev->external == 0 && mddev->safemode == 1) ||
9401 		(mddev->safemode == 2
9402 		 && !mddev->in_sync && mddev->recovery_cp == MaxSector)
9403 		))
9404 		return;
9405 
9406 	if (mddev_trylock(mddev)) {
9407 		int spares = 0;
9408 		bool try_set_sync = mddev->safemode != 0;
9409 
9410 		if (!mddev->external && mddev->safemode == 1)
9411 			mddev->safemode = 0;
9412 
9413 		if (!md_is_rdwr(mddev)) {
9414 			struct md_rdev *rdev;
9415 			if (!mddev->external && mddev->in_sync)
9416 				/* 'Blocked' flag not needed as failed devices
9417 				 * will be recorded if array switched to read/write.
9418 				 * Leaving it set will prevent the device
9419 				 * from being removed.
9420 				 */
9421 				rdev_for_each(rdev, mddev)
9422 					clear_bit(Blocked, &rdev->flags);
9423 			/* On a read-only array we can:
9424 			 * - remove failed devices
9425 			 * - add already-in_sync devices if the array itself
9426 			 *   is in-sync.
9427 			 * As we only add devices that are already in-sync,
9428 			 * we can activate the spares immediately.
9429 			 */
9430 			remove_and_add_spares(mddev, NULL);
9431 			/* There is no thread, but we need to call
9432 			 * ->spare_active and clear saved_raid_disk
9433 			 */
9434 			set_bit(MD_RECOVERY_INTR, &mddev->recovery);
9435 			md_reap_sync_thread(mddev);
9436 			clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
9437 			clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
9438 			clear_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags);
9439 			goto unlock;
9440 		}
9441 
9442 		if (mddev_is_clustered(mddev)) {
9443 			struct md_rdev *rdev, *tmp;
9444 			/* kick the device if another node issued a
9445 			 * remove disk.
9446 			 */
9447 			rdev_for_each_safe(rdev, tmp, mddev) {
9448 				if (test_and_clear_bit(ClusterRemove, &rdev->flags) &&
9449 						rdev->raid_disk < 0)
9450 					md_kick_rdev_from_array(rdev);
9451 			}
9452 		}
9453 
9454 		if (try_set_sync && !mddev->external && !mddev->in_sync) {
9455 			spin_lock(&mddev->lock);
9456 			set_in_sync(mddev);
9457 			spin_unlock(&mddev->lock);
9458 		}
9459 
9460 		if (mddev->sb_flags)
9461 			md_update_sb(mddev, 0);
9462 
9463 		/*
9464 		 * Never start a new sync thread if MD_RECOVERY_RUNNING is
9465 		 * still set.
9466 		 */
9467 		if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) {
9468 			if (!test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
9469 				/* resync/recovery still happening */
9470 				clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
9471 				goto unlock;
9472 			}
9473 
9474 			if (WARN_ON_ONCE(!mddev->sync_thread))
9475 				goto unlock;
9476 
9477 			md_reap_sync_thread(mddev);
9478 			goto unlock;
9479 		}
9480 
9481 		/* Set RUNNING before clearing NEEDED to avoid
9482 		 * any transients in the value of "sync_action".
9483 		 */
9484 		mddev->curr_resync_completed = 0;
9485 		spin_lock(&mddev->lock);
9486 		set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
9487 		spin_unlock(&mddev->lock);
9488 		/* Clear some bits that don't mean anything, but
9489 		 * might be left set
9490 		 */
9491 		clear_bit(MD_RECOVERY_INTR, &mddev->recovery);
9492 		clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
9493 
9494 		if (!test_and_clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
9495 		    test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
9496 			goto not_running;
9497 		/* no recovery is running.
9498 		 * remove any failed drives, then
9499 		 * add spares if possible.
9500 		 * Spares are also removed and re-added, to allow
9501 		 * the personality to fail the re-add.
9502 		 */
9503 
9504 		if (mddev->reshape_position != MaxSector) {
9505 			if (mddev->pers->check_reshape == NULL ||
9506 			    mddev->pers->check_reshape(mddev) != 0)
9507 				/* Cannot proceed */
9508 				goto not_running;
9509 			set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
9510 			clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
9511 		} else if ((spares = remove_and_add_spares(mddev, NULL))) {
9512 			clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
9513 			clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
9514 			clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
9515 			set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
9516 		} else if (mddev->recovery_cp < MaxSector) {
9517 			set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
9518 			clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
9519 		} else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
9520 			/* nothing to be done ... */
9521 			goto not_running;
9522 
9523 		if (mddev->pers->sync_request) {
9524 			if (spares) {
9525 				/* We are adding a device or devices to an array
9526 				 * which has the bitmap stored on all devices.
9527 				 * So make sure all bitmap pages get written
9528 				 */
9529 				md_bitmap_write_all(mddev->bitmap);
9530 			}
9531 			INIT_WORK(&mddev->del_work, md_start_sync);
9532 			queue_work(md_misc_wq, &mddev->del_work);
9533 			goto unlock;
9534 		}
9535 	not_running:
9536 		if (!mddev->sync_thread) {
9537 			clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
9538 			wake_up(&resync_wait);
9539 			if (test_and_clear_bit(MD_RECOVERY_RECOVER,
9540 					       &mddev->recovery))
9541 				if (mddev->sysfs_action)
9542 					sysfs_notify_dirent_safe(mddev->sysfs_action);
9543 		}
9544 	unlock:
9545 		wake_up(&mddev->sb_wait);
9546 		mddev_unlock(mddev);
9547 	}
9548 }
9549 EXPORT_SYMBOL(md_check_recovery);
9550 
9551 void md_reap_sync_thread(struct mddev *mddev)
9552 {
9553 	struct md_rdev *rdev;
9554 	sector_t old_dev_sectors = mddev->dev_sectors;
9555 	bool is_reshaped = false;
9556 
9557 	/* resync has finished, collect result */
9558 	md_unregister_thread(mddev, &mddev->sync_thread);
9559 	atomic_inc(&mddev->sync_seq);
9560 
9561 	if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
9562 	    !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery) &&
9563 	    mddev->degraded != mddev->raid_disks) {
9564 		/* success...*/
9565 		/* activate any spares */
9566 		if (mddev->pers->spare_active(mddev)) {
9567 			sysfs_notify_dirent_safe(mddev->sysfs_degraded);
9568 			set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
9569 		}
9570 	}
9571 	if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
9572 	    mddev->pers->finish_reshape) {
9573 		mddev->pers->finish_reshape(mddev);
9574 		if (mddev_is_clustered(mddev))
9575 			is_reshaped = true;
9576 	}
9577 
9578 	/* If array is no-longer degraded, then any saved_raid_disk
9579 	 * information must be scrapped.
9580 	 */
9581 	if (!mddev->degraded)
9582 		rdev_for_each(rdev, mddev)
9583 			rdev->saved_raid_disk = -1;
9584 
9585 	md_update_sb(mddev, 1);
9586 	/* MD_SB_CHANGE_PENDING should be cleared by md_update_sb, so we can
9587 	 * call resync_finish here if MD_CLUSTER_RESYNC_LOCKED is set by
9588 	 * clustered raid */
9589 	if (test_and_clear_bit(MD_CLUSTER_RESYNC_LOCKED, &mddev->flags))
9590 		md_cluster_ops->resync_finish(mddev);
9591 	clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
9592 	clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
9593 	clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
9594 	clear_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
9595 	clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
9596 	clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
9597 	/*
9598 	 * We call md_cluster_ops->update_size here because sync_size could
9599 	 * be changed by md_update_sb, and MD_RECOVERY_RESHAPE is cleared,
9600 	 * so it is time to update size across cluster.
9601 	 */
9602 	if (mddev_is_clustered(mddev) && is_reshaped
9603 				      && !test_bit(MD_CLOSING, &mddev->flags))
9604 		md_cluster_ops->update_size(mddev, old_dev_sectors);
9605 	/* flag recovery needed just to double check */
9606 	set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
9607 	sysfs_notify_dirent_safe(mddev->sysfs_completed);
9608 	sysfs_notify_dirent_safe(mddev->sysfs_action);
9609 	md_new_event();
9610 	if (mddev->event_work.func)
9611 		queue_work(md_misc_wq, &mddev->event_work);
9612 	wake_up(&resync_wait);
9613 }
9614 EXPORT_SYMBOL(md_reap_sync_thread);
9615 
9616 void md_wait_for_blocked_rdev(struct md_rdev *rdev, struct mddev *mddev)
9617 {
9618 	sysfs_notify_dirent_safe(rdev->sysfs_state);
9619 	wait_event_timeout(rdev->blocked_wait,
9620 			   !test_bit(Blocked, &rdev->flags) &&
9621 			   !test_bit(BlockedBadBlocks, &rdev->flags),
9622 			   msecs_to_jiffies(5000));
9623 	rdev_dec_pending(rdev, mddev);
9624 }
9625 EXPORT_SYMBOL(md_wait_for_blocked_rdev);
9626 
9627 void md_finish_reshape(struct mddev *mddev)
9628 {
9629 	/* called be personality module when reshape completes. */
9630 	struct md_rdev *rdev;
9631 
9632 	rdev_for_each(rdev, mddev) {
9633 		if (rdev->data_offset > rdev->new_data_offset)
9634 			rdev->sectors += rdev->data_offset - rdev->new_data_offset;
9635 		else
9636 			rdev->sectors -= rdev->new_data_offset - rdev->data_offset;
9637 		rdev->data_offset = rdev->new_data_offset;
9638 	}
9639 }
9640 EXPORT_SYMBOL(md_finish_reshape);
9641 
9642 /* Bad block management */
9643 
9644 /* Returns 1 on success, 0 on failure */
9645 int rdev_set_badblocks(struct md_rdev *rdev, sector_t s, int sectors,
9646 		       int is_new)
9647 {
9648 	struct mddev *mddev = rdev->mddev;
9649 	int rv;
9650 	if (is_new)
9651 		s += rdev->new_data_offset;
9652 	else
9653 		s += rdev->data_offset;
9654 	rv = badblocks_set(&rdev->badblocks, s, sectors, 0);
9655 	if (rv == 0) {
9656 		/* Make sure they get written out promptly */
9657 		if (test_bit(ExternalBbl, &rdev->flags))
9658 			sysfs_notify_dirent_safe(rdev->sysfs_unack_badblocks);
9659 		sysfs_notify_dirent_safe(rdev->sysfs_state);
9660 		set_mask_bits(&mddev->sb_flags, 0,
9661 			      BIT(MD_SB_CHANGE_CLEAN) | BIT(MD_SB_CHANGE_PENDING));
9662 		md_wakeup_thread(rdev->mddev->thread);
9663 		return 1;
9664 	} else
9665 		return 0;
9666 }
9667 EXPORT_SYMBOL_GPL(rdev_set_badblocks);
9668 
9669 int rdev_clear_badblocks(struct md_rdev *rdev, sector_t s, int sectors,
9670 			 int is_new)
9671 {
9672 	int rv;
9673 	if (is_new)
9674 		s += rdev->new_data_offset;
9675 	else
9676 		s += rdev->data_offset;
9677 	rv = badblocks_clear(&rdev->badblocks, s, sectors);
9678 	if ((rv == 0) && test_bit(ExternalBbl, &rdev->flags))
9679 		sysfs_notify_dirent_safe(rdev->sysfs_badblocks);
9680 	return rv;
9681 }
9682 EXPORT_SYMBOL_GPL(rdev_clear_badblocks);
9683 
9684 static int md_notify_reboot(struct notifier_block *this,
9685 			    unsigned long code, void *x)
9686 {
9687 	struct mddev *mddev, *n;
9688 	int need_delay = 0;
9689 
9690 	spin_lock(&all_mddevs_lock);
9691 	list_for_each_entry_safe(mddev, n, &all_mddevs, all_mddevs) {
9692 		if (!mddev_get(mddev))
9693 			continue;
9694 		spin_unlock(&all_mddevs_lock);
9695 		if (mddev_trylock(mddev)) {
9696 			if (mddev->pers)
9697 				__md_stop_writes(mddev);
9698 			if (mddev->persistent)
9699 				mddev->safemode = 2;
9700 			mddev_unlock(mddev);
9701 		}
9702 		need_delay = 1;
9703 		mddev_put(mddev);
9704 		spin_lock(&all_mddevs_lock);
9705 	}
9706 	spin_unlock(&all_mddevs_lock);
9707 
9708 	/*
9709 	 * certain more exotic SCSI devices are known to be
9710 	 * volatile wrt too early system reboots. While the
9711 	 * right place to handle this issue is the given
9712 	 * driver, we do want to have a safe RAID driver ...
9713 	 */
9714 	if (need_delay)
9715 		msleep(1000);
9716 
9717 	return NOTIFY_DONE;
9718 }
9719 
9720 static struct notifier_block md_notifier = {
9721 	.notifier_call	= md_notify_reboot,
9722 	.next		= NULL,
9723 	.priority	= INT_MAX, /* before any real devices */
9724 };
9725 
9726 static void md_geninit(void)
9727 {
9728 	pr_debug("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
9729 
9730 	proc_create("mdstat", S_IRUGO, NULL, &mdstat_proc_ops);
9731 }
9732 
9733 static int __init md_init(void)
9734 {
9735 	int ret = -ENOMEM;
9736 
9737 	md_wq = alloc_workqueue("md", WQ_MEM_RECLAIM, 0);
9738 	if (!md_wq)
9739 		goto err_wq;
9740 
9741 	md_misc_wq = alloc_workqueue("md_misc", 0, 0);
9742 	if (!md_misc_wq)
9743 		goto err_misc_wq;
9744 
9745 	md_bitmap_wq = alloc_workqueue("md_bitmap", WQ_MEM_RECLAIM | WQ_UNBOUND,
9746 				       0);
9747 	if (!md_bitmap_wq)
9748 		goto err_bitmap_wq;
9749 
9750 	ret = __register_blkdev(MD_MAJOR, "md", md_probe);
9751 	if (ret < 0)
9752 		goto err_md;
9753 
9754 	ret = __register_blkdev(0, "mdp", md_probe);
9755 	if (ret < 0)
9756 		goto err_mdp;
9757 	mdp_major = ret;
9758 
9759 	register_reboot_notifier(&md_notifier);
9760 	raid_table_header = register_sysctl("dev/raid", raid_table);
9761 
9762 	md_geninit();
9763 	return 0;
9764 
9765 err_mdp:
9766 	unregister_blkdev(MD_MAJOR, "md");
9767 err_md:
9768 	destroy_workqueue(md_bitmap_wq);
9769 err_bitmap_wq:
9770 	destroy_workqueue(md_misc_wq);
9771 err_misc_wq:
9772 	destroy_workqueue(md_wq);
9773 err_wq:
9774 	return ret;
9775 }
9776 
9777 static void check_sb_changes(struct mddev *mddev, struct md_rdev *rdev)
9778 {
9779 	struct mdp_superblock_1 *sb = page_address(rdev->sb_page);
9780 	struct md_rdev *rdev2, *tmp;
9781 	int role, ret;
9782 
9783 	/*
9784 	 * If size is changed in another node then we need to
9785 	 * do resize as well.
9786 	 */
9787 	if (mddev->dev_sectors != le64_to_cpu(sb->size)) {
9788 		ret = mddev->pers->resize(mddev, le64_to_cpu(sb->size));
9789 		if (ret)
9790 			pr_info("md-cluster: resize failed\n");
9791 		else
9792 			md_bitmap_update_sb(mddev->bitmap);
9793 	}
9794 
9795 	/* Check for change of roles in the active devices */
9796 	rdev_for_each_safe(rdev2, tmp, mddev) {
9797 		if (test_bit(Faulty, &rdev2->flags))
9798 			continue;
9799 
9800 		/* Check if the roles changed */
9801 		role = le16_to_cpu(sb->dev_roles[rdev2->desc_nr]);
9802 
9803 		if (test_bit(Candidate, &rdev2->flags)) {
9804 			if (role == MD_DISK_ROLE_FAULTY) {
9805 				pr_info("md: Removing Candidate device %pg because add failed\n",
9806 					rdev2->bdev);
9807 				md_kick_rdev_from_array(rdev2);
9808 				continue;
9809 			}
9810 			else
9811 				clear_bit(Candidate, &rdev2->flags);
9812 		}
9813 
9814 		if (role != rdev2->raid_disk) {
9815 			/*
9816 			 * got activated except reshape is happening.
9817 			 */
9818 			if (rdev2->raid_disk == -1 && role != MD_DISK_ROLE_SPARE &&
9819 			    !(le32_to_cpu(sb->feature_map) &
9820 			      MD_FEATURE_RESHAPE_ACTIVE)) {
9821 				rdev2->saved_raid_disk = role;
9822 				ret = remove_and_add_spares(mddev, rdev2);
9823 				pr_info("Activated spare: %pg\n",
9824 					rdev2->bdev);
9825 				/* wakeup mddev->thread here, so array could
9826 				 * perform resync with the new activated disk */
9827 				set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
9828 				md_wakeup_thread(mddev->thread);
9829 			}
9830 			/* device faulty
9831 			 * We just want to do the minimum to mark the disk
9832 			 * as faulty. The recovery is performed by the
9833 			 * one who initiated the error.
9834 			 */
9835 			if (role == MD_DISK_ROLE_FAULTY ||
9836 			    role == MD_DISK_ROLE_JOURNAL) {
9837 				md_error(mddev, rdev2);
9838 				clear_bit(Blocked, &rdev2->flags);
9839 			}
9840 		}
9841 	}
9842 
9843 	if (mddev->raid_disks != le32_to_cpu(sb->raid_disks)) {
9844 		ret = update_raid_disks(mddev, le32_to_cpu(sb->raid_disks));
9845 		if (ret)
9846 			pr_warn("md: updating array disks failed. %d\n", ret);
9847 	}
9848 
9849 	/*
9850 	 * Since mddev->delta_disks has already updated in update_raid_disks,
9851 	 * so it is time to check reshape.
9852 	 */
9853 	if (test_bit(MD_RESYNCING_REMOTE, &mddev->recovery) &&
9854 	    (le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) {
9855 		/*
9856 		 * reshape is happening in the remote node, we need to
9857 		 * update reshape_position and call start_reshape.
9858 		 */
9859 		mddev->reshape_position = le64_to_cpu(sb->reshape_position);
9860 		if (mddev->pers->update_reshape_pos)
9861 			mddev->pers->update_reshape_pos(mddev);
9862 		if (mddev->pers->start_reshape)
9863 			mddev->pers->start_reshape(mddev);
9864 	} else if (test_bit(MD_RESYNCING_REMOTE, &mddev->recovery) &&
9865 		   mddev->reshape_position != MaxSector &&
9866 		   !(le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) {
9867 		/* reshape is just done in another node. */
9868 		mddev->reshape_position = MaxSector;
9869 		if (mddev->pers->update_reshape_pos)
9870 			mddev->pers->update_reshape_pos(mddev);
9871 	}
9872 
9873 	/* Finally set the event to be up to date */
9874 	mddev->events = le64_to_cpu(sb->events);
9875 }
9876 
9877 static int read_rdev(struct mddev *mddev, struct md_rdev *rdev)
9878 {
9879 	int err;
9880 	struct page *swapout = rdev->sb_page;
9881 	struct mdp_superblock_1 *sb;
9882 
9883 	/* Store the sb page of the rdev in the swapout temporary
9884 	 * variable in case we err in the future
9885 	 */
9886 	rdev->sb_page = NULL;
9887 	err = alloc_disk_sb(rdev);
9888 	if (err == 0) {
9889 		ClearPageUptodate(rdev->sb_page);
9890 		rdev->sb_loaded = 0;
9891 		err = super_types[mddev->major_version].
9892 			load_super(rdev, NULL, mddev->minor_version);
9893 	}
9894 	if (err < 0) {
9895 		pr_warn("%s: %d Could not reload rdev(%d) err: %d. Restoring old values\n",
9896 				__func__, __LINE__, rdev->desc_nr, err);
9897 		if (rdev->sb_page)
9898 			put_page(rdev->sb_page);
9899 		rdev->sb_page = swapout;
9900 		rdev->sb_loaded = 1;
9901 		return err;
9902 	}
9903 
9904 	sb = page_address(rdev->sb_page);
9905 	/* Read the offset unconditionally, even if MD_FEATURE_RECOVERY_OFFSET
9906 	 * is not set
9907 	 */
9908 
9909 	if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RECOVERY_OFFSET))
9910 		rdev->recovery_offset = le64_to_cpu(sb->recovery_offset);
9911 
9912 	/* The other node finished recovery, call spare_active to set
9913 	 * device In_sync and mddev->degraded
9914 	 */
9915 	if (rdev->recovery_offset == MaxSector &&
9916 	    !test_bit(In_sync, &rdev->flags) &&
9917 	    mddev->pers->spare_active(mddev))
9918 		sysfs_notify_dirent_safe(mddev->sysfs_degraded);
9919 
9920 	put_page(swapout);
9921 	return 0;
9922 }
9923 
9924 void md_reload_sb(struct mddev *mddev, int nr)
9925 {
9926 	struct md_rdev *rdev = NULL, *iter;
9927 	int err;
9928 
9929 	/* Find the rdev */
9930 	rdev_for_each_rcu(iter, mddev) {
9931 		if (iter->desc_nr == nr) {
9932 			rdev = iter;
9933 			break;
9934 		}
9935 	}
9936 
9937 	if (!rdev) {
9938 		pr_warn("%s: %d Could not find rdev with nr %d\n", __func__, __LINE__, nr);
9939 		return;
9940 	}
9941 
9942 	err = read_rdev(mddev, rdev);
9943 	if (err < 0)
9944 		return;
9945 
9946 	check_sb_changes(mddev, rdev);
9947 
9948 	/* Read all rdev's to update recovery_offset */
9949 	rdev_for_each_rcu(rdev, mddev) {
9950 		if (!test_bit(Faulty, &rdev->flags))
9951 			read_rdev(mddev, rdev);
9952 	}
9953 }
9954 EXPORT_SYMBOL(md_reload_sb);
9955 
9956 #ifndef MODULE
9957 
9958 /*
9959  * Searches all registered partitions for autorun RAID arrays
9960  * at boot time.
9961  */
9962 
9963 static DEFINE_MUTEX(detected_devices_mutex);
9964 static LIST_HEAD(all_detected_devices);
9965 struct detected_devices_node {
9966 	struct list_head list;
9967 	dev_t dev;
9968 };
9969 
9970 void md_autodetect_dev(dev_t dev)
9971 {
9972 	struct detected_devices_node *node_detected_dev;
9973 
9974 	node_detected_dev = kzalloc(sizeof(*node_detected_dev), GFP_KERNEL);
9975 	if (node_detected_dev) {
9976 		node_detected_dev->dev = dev;
9977 		mutex_lock(&detected_devices_mutex);
9978 		list_add_tail(&node_detected_dev->list, &all_detected_devices);
9979 		mutex_unlock(&detected_devices_mutex);
9980 	}
9981 }
9982 
9983 void md_autostart_arrays(int part)
9984 {
9985 	struct md_rdev *rdev;
9986 	struct detected_devices_node *node_detected_dev;
9987 	dev_t dev;
9988 	int i_scanned, i_passed;
9989 
9990 	i_scanned = 0;
9991 	i_passed = 0;
9992 
9993 	pr_info("md: Autodetecting RAID arrays.\n");
9994 
9995 	mutex_lock(&detected_devices_mutex);
9996 	while (!list_empty(&all_detected_devices) && i_scanned < INT_MAX) {
9997 		i_scanned++;
9998 		node_detected_dev = list_entry(all_detected_devices.next,
9999 					struct detected_devices_node, list);
10000 		list_del(&node_detected_dev->list);
10001 		dev = node_detected_dev->dev;
10002 		kfree(node_detected_dev);
10003 		mutex_unlock(&detected_devices_mutex);
10004 		rdev = md_import_device(dev,0, 90);
10005 		mutex_lock(&detected_devices_mutex);
10006 		if (IS_ERR(rdev))
10007 			continue;
10008 
10009 		if (test_bit(Faulty, &rdev->flags))
10010 			continue;
10011 
10012 		set_bit(AutoDetected, &rdev->flags);
10013 		list_add(&rdev->same_set, &pending_raid_disks);
10014 		i_passed++;
10015 	}
10016 	mutex_unlock(&detected_devices_mutex);
10017 
10018 	pr_debug("md: Scanned %d and added %d devices.\n", i_scanned, i_passed);
10019 
10020 	autorun_devices(part);
10021 }
10022 
10023 #endif /* !MODULE */
10024 
10025 static __exit void md_exit(void)
10026 {
10027 	struct mddev *mddev, *n;
10028 	int delay = 1;
10029 
10030 	unregister_blkdev(MD_MAJOR,"md");
10031 	unregister_blkdev(mdp_major, "mdp");
10032 	unregister_reboot_notifier(&md_notifier);
10033 	unregister_sysctl_table(raid_table_header);
10034 
10035 	/* We cannot unload the modules while some process is
10036 	 * waiting for us in select() or poll() - wake them up
10037 	 */
10038 	md_unloading = 1;
10039 	while (waitqueue_active(&md_event_waiters)) {
10040 		/* not safe to leave yet */
10041 		wake_up(&md_event_waiters);
10042 		msleep(delay);
10043 		delay += delay;
10044 	}
10045 	remove_proc_entry("mdstat", NULL);
10046 
10047 	spin_lock(&all_mddevs_lock);
10048 	list_for_each_entry_safe(mddev, n, &all_mddevs, all_mddevs) {
10049 		if (!mddev_get(mddev))
10050 			continue;
10051 		spin_unlock(&all_mddevs_lock);
10052 		export_array(mddev);
10053 		mddev->ctime = 0;
10054 		mddev->hold_active = 0;
10055 		/*
10056 		 * As the mddev is now fully clear, mddev_put will schedule
10057 		 * the mddev for destruction by a workqueue, and the
10058 		 * destroy_workqueue() below will wait for that to complete.
10059 		 */
10060 		mddev_put(mddev);
10061 		spin_lock(&all_mddevs_lock);
10062 	}
10063 	spin_unlock(&all_mddevs_lock);
10064 
10065 	destroy_workqueue(md_misc_wq);
10066 	destroy_workqueue(md_bitmap_wq);
10067 	destroy_workqueue(md_wq);
10068 }
10069 
10070 subsys_initcall(md_init);
10071 module_exit(md_exit)
10072 
10073 static int get_ro(char *buffer, const struct kernel_param *kp)
10074 {
10075 	return sprintf(buffer, "%d\n", start_readonly);
10076 }
10077 static int set_ro(const char *val, const struct kernel_param *kp)
10078 {
10079 	return kstrtouint(val, 10, (unsigned int *)&start_readonly);
10080 }
10081 
10082 module_param_call(start_ro, set_ro, get_ro, NULL, S_IRUSR|S_IWUSR);
10083 module_param(start_dirty_degraded, int, S_IRUGO|S_IWUSR);
10084 module_param_call(new_array, add_named_array, NULL, NULL, S_IWUSR);
10085 module_param(create_on_open, bool, S_IRUSR|S_IWUSR);
10086 
10087 MODULE_LICENSE("GPL");
10088 MODULE_DESCRIPTION("MD RAID framework");
10089 MODULE_ALIAS("md");
10090 MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR);
10091