xref: /openbmc/linux/drivers/md/md.c (revision d37cf9b63113f13d742713881ce691fc615d8b3b)
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;
speed_min(struct mddev * mddev)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 
speed_max(struct mddev * mddev)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 
rdev_uninit_serial(struct md_rdev * rdev)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 
rdevs_uninit_serial(struct mddev * mddev)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 
rdev_init_serial(struct md_rdev * rdev)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 
rdevs_init_serial(struct mddev * mddev)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  */
rdev_need_serial(struct md_rdev * rdev)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  */
mddev_create_serial_pool(struct mddev * mddev,struct md_rdev * rdev,bool is_suspend)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  */
mddev_destroy_serial_pool(struct mddev * mddev,struct md_rdev * rdev,bool is_suspend)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;
md_new_event(void)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  */
is_suspended(struct mddev * mddev,struct bio * bio)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 
md_handle_request(struct mddev * mddev,struct bio * bio)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 
md_submit_bio(struct bio * bio)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  */
mddev_suspend(struct mddev * mddev)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 	/* restrict memory reclaim I/O during raid array is suspend */
460 	mddev->noio_flag = memalloc_noio_save();
461 }
462 EXPORT_SYMBOL_GPL(mddev_suspend);
463 
mddev_resume(struct mddev * mddev)464 void mddev_resume(struct mddev *mddev)
465 {
466 	lockdep_assert_held(&mddev->reconfig_mutex);
467 	if (--mddev->suspended)
468 		return;
469 
470 	/* entred the memalloc scope from mddev_suspend() */
471 	memalloc_noio_restore(mddev->noio_flag);
472 
473 	percpu_ref_resurrect(&mddev->active_io);
474 	wake_up(&mddev->sb_wait);
475 
476 	set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
477 	md_wakeup_thread(mddev->thread);
478 	md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
479 }
480 EXPORT_SYMBOL_GPL(mddev_resume);
481 
482 /*
483  * Generic flush handling for md
484  */
485 
md_end_flush(struct bio * bio)486 static void md_end_flush(struct bio *bio)
487 {
488 	struct md_rdev *rdev = bio->bi_private;
489 	struct mddev *mddev = rdev->mddev;
490 
491 	bio_put(bio);
492 
493 	rdev_dec_pending(rdev, mddev);
494 
495 	if (atomic_dec_and_test(&mddev->flush_pending))
496 		/* The pre-request flush has finished */
497 		queue_work(md_wq, &mddev->flush_work);
498 }
499 
500 static void md_submit_flush_data(struct work_struct *ws);
501 
submit_flushes(struct work_struct * ws)502 static void submit_flushes(struct work_struct *ws)
503 {
504 	struct mddev *mddev = container_of(ws, struct mddev, flush_work);
505 	struct md_rdev *rdev;
506 
507 	mddev->start_flush = ktime_get_boottime();
508 	INIT_WORK(&mddev->flush_work, md_submit_flush_data);
509 	atomic_set(&mddev->flush_pending, 1);
510 	rcu_read_lock();
511 	rdev_for_each_rcu(rdev, mddev)
512 		if (rdev->raid_disk >= 0 &&
513 		    !test_bit(Faulty, &rdev->flags)) {
514 			struct bio *bi;
515 
516 			atomic_inc(&rdev->nr_pending);
517 			rcu_read_unlock();
518 			bi = bio_alloc_bioset(rdev->bdev, 0,
519 					      REQ_OP_WRITE | REQ_PREFLUSH,
520 					      GFP_NOIO, &mddev->bio_set);
521 			bi->bi_end_io = md_end_flush;
522 			bi->bi_private = rdev;
523 			atomic_inc(&mddev->flush_pending);
524 			submit_bio(bi);
525 			rcu_read_lock();
526 		}
527 	rcu_read_unlock();
528 	if (atomic_dec_and_test(&mddev->flush_pending))
529 		queue_work(md_wq, &mddev->flush_work);
530 }
531 
md_submit_flush_data(struct work_struct * ws)532 static void md_submit_flush_data(struct work_struct *ws)
533 {
534 	struct mddev *mddev = container_of(ws, struct mddev, flush_work);
535 	struct bio *bio = mddev->flush_bio;
536 
537 	/*
538 	 * must reset flush_bio before calling into md_handle_request to avoid a
539 	 * deadlock, because other bios passed md_handle_request suspend check
540 	 * could wait for this and below md_handle_request could wait for those
541 	 * bios because of suspend check
542 	 */
543 	spin_lock_irq(&mddev->lock);
544 	mddev->prev_flush_start = mddev->start_flush;
545 	mddev->flush_bio = NULL;
546 	spin_unlock_irq(&mddev->lock);
547 	wake_up(&mddev->sb_wait);
548 
549 	if (bio->bi_iter.bi_size == 0) {
550 		/* an empty barrier - all done */
551 		bio_endio(bio);
552 	} else {
553 		bio->bi_opf &= ~REQ_PREFLUSH;
554 
555 		/*
556 		 * make_requst() will never return error here, it only
557 		 * returns error in raid5_make_request() by dm-raid.
558 		 * Since dm always splits data and flush operation into
559 		 * two separate io, io size of flush submitted by dm
560 		 * always is 0, make_request() will not be called here.
561 		 */
562 		if (WARN_ON_ONCE(!mddev->pers->make_request(mddev, bio)))
563 			bio_io_error(bio);;
564 	}
565 
566 	/* The pair is percpu_ref_get() from md_flush_request() */
567 	percpu_ref_put(&mddev->active_io);
568 }
569 
570 /*
571  * Manages consolidation of flushes and submitting any flushes needed for
572  * a bio with REQ_PREFLUSH.  Returns true if the bio is finished or is
573  * being finished in another context.  Returns false if the flushing is
574  * complete but still needs the I/O portion of the bio to be processed.
575  */
md_flush_request(struct mddev * mddev,struct bio * bio)576 bool md_flush_request(struct mddev *mddev, struct bio *bio)
577 {
578 	ktime_t req_start = ktime_get_boottime();
579 	spin_lock_irq(&mddev->lock);
580 	/* flush requests wait until ongoing flush completes,
581 	 * hence coalescing all the pending requests.
582 	 */
583 	wait_event_lock_irq(mddev->sb_wait,
584 			    !mddev->flush_bio ||
585 			    ktime_before(req_start, mddev->prev_flush_start),
586 			    mddev->lock);
587 	/* new request after previous flush is completed */
588 	if (ktime_after(req_start, mddev->prev_flush_start)) {
589 		WARN_ON(mddev->flush_bio);
590 		/*
591 		 * Grab a reference to make sure mddev_suspend() will wait for
592 		 * this flush to be done.
593 		 *
594 		 * md_flush_reqeust() is called under md_handle_request() and
595 		 * 'active_io' is already grabbed, hence percpu_ref_is_zero()
596 		 * won't pass, percpu_ref_tryget_live() can't be used because
597 		 * percpu_ref_kill() can be called by mddev_suspend()
598 		 * concurrently.
599 		 */
600 		WARN_ON(percpu_ref_is_zero(&mddev->active_io));
601 		percpu_ref_get(&mddev->active_io);
602 		mddev->flush_bio = bio;
603 		bio = NULL;
604 	}
605 	spin_unlock_irq(&mddev->lock);
606 
607 	if (!bio) {
608 		INIT_WORK(&mddev->flush_work, submit_flushes);
609 		queue_work(md_wq, &mddev->flush_work);
610 	} else {
611 		/* flush was performed for some other bio while we waited. */
612 		if (bio->bi_iter.bi_size == 0)
613 			/* an empty barrier - all done */
614 			bio_endio(bio);
615 		else {
616 			bio->bi_opf &= ~REQ_PREFLUSH;
617 			return false;
618 		}
619 	}
620 	return true;
621 }
622 EXPORT_SYMBOL(md_flush_request);
623 
mddev_get(struct mddev * mddev)624 static inline struct mddev *mddev_get(struct mddev *mddev)
625 {
626 	lockdep_assert_held(&all_mddevs_lock);
627 
628 	if (test_bit(MD_DELETED, &mddev->flags))
629 		return NULL;
630 	atomic_inc(&mddev->active);
631 	return mddev;
632 }
633 
634 static void mddev_delayed_delete(struct work_struct *ws);
635 
__mddev_put(struct mddev * mddev)636 static void __mddev_put(struct mddev *mddev)
637 {
638 	if (mddev->raid_disks || !list_empty(&mddev->disks) ||
639 	    mddev->ctime || mddev->hold_active)
640 		return;
641 
642 	/* Array is not configured at all, and not held active, so destroy it */
643 	set_bit(MD_DELETED, &mddev->flags);
644 
645 	/*
646 	 * Call queue_work inside the spinlock so that flush_workqueue() after
647 	 * mddev_find will succeed in waiting for the work to be done.
648 	 */
649 	queue_work(md_misc_wq, &mddev->del_work);
650 }
651 
mddev_put(struct mddev * mddev)652 void mddev_put(struct mddev *mddev)
653 {
654 	if (!atomic_dec_and_lock(&mddev->active, &all_mddevs_lock))
655 		return;
656 
657 	__mddev_put(mddev);
658 	spin_unlock(&all_mddevs_lock);
659 }
660 
661 static void md_safemode_timeout(struct timer_list *t);
662 static void md_start_sync(struct work_struct *ws);
663 
mddev_init(struct mddev * mddev)664 void mddev_init(struct mddev *mddev)
665 {
666 	mutex_init(&mddev->open_mutex);
667 	mutex_init(&mddev->reconfig_mutex);
668 	mutex_init(&mddev->sync_mutex);
669 	mutex_init(&mddev->bitmap_info.mutex);
670 	INIT_LIST_HEAD(&mddev->disks);
671 	INIT_LIST_HEAD(&mddev->all_mddevs);
672 	INIT_LIST_HEAD(&mddev->deleting);
673 	timer_setup(&mddev->safemode_timer, md_safemode_timeout, 0);
674 	atomic_set(&mddev->active, 1);
675 	atomic_set(&mddev->openers, 0);
676 	atomic_set(&mddev->sync_seq, 0);
677 	spin_lock_init(&mddev->lock);
678 	atomic_set(&mddev->flush_pending, 0);
679 	init_waitqueue_head(&mddev->sb_wait);
680 	init_waitqueue_head(&mddev->recovery_wait);
681 	mddev->reshape_position = MaxSector;
682 	mddev->reshape_backwards = 0;
683 	mddev->last_sync_action = "none";
684 	mddev->resync_min = 0;
685 	mddev->resync_max = MaxSector;
686 	mddev->level = LEVEL_NONE;
687 
688 	INIT_WORK(&mddev->sync_work, md_start_sync);
689 	INIT_WORK(&mddev->del_work, mddev_delayed_delete);
690 }
691 EXPORT_SYMBOL_GPL(mddev_init);
692 
mddev_find_locked(dev_t unit)693 static struct mddev *mddev_find_locked(dev_t unit)
694 {
695 	struct mddev *mddev;
696 
697 	list_for_each_entry(mddev, &all_mddevs, all_mddevs)
698 		if (mddev->unit == unit)
699 			return mddev;
700 
701 	return NULL;
702 }
703 
704 /* find an unused unit number */
mddev_alloc_unit(void)705 static dev_t mddev_alloc_unit(void)
706 {
707 	static int next_minor = 512;
708 	int start = next_minor;
709 	bool is_free = 0;
710 	dev_t dev = 0;
711 
712 	while (!is_free) {
713 		dev = MKDEV(MD_MAJOR, next_minor);
714 		next_minor++;
715 		if (next_minor > MINORMASK)
716 			next_minor = 0;
717 		if (next_minor == start)
718 			return 0;		/* Oh dear, all in use. */
719 		is_free = !mddev_find_locked(dev);
720 	}
721 
722 	return dev;
723 }
724 
mddev_alloc(dev_t unit)725 static struct mddev *mddev_alloc(dev_t unit)
726 {
727 	struct mddev *new;
728 	int error;
729 
730 	if (unit && MAJOR(unit) != MD_MAJOR)
731 		unit &= ~((1 << MdpMinorShift) - 1);
732 
733 	new = kzalloc(sizeof(*new), GFP_KERNEL);
734 	if (!new)
735 		return ERR_PTR(-ENOMEM);
736 	mddev_init(new);
737 
738 	spin_lock(&all_mddevs_lock);
739 	if (unit) {
740 		error = -EEXIST;
741 		if (mddev_find_locked(unit))
742 			goto out_free_new;
743 		new->unit = unit;
744 		if (MAJOR(unit) == MD_MAJOR)
745 			new->md_minor = MINOR(unit);
746 		else
747 			new->md_minor = MINOR(unit) >> MdpMinorShift;
748 		new->hold_active = UNTIL_IOCTL;
749 	} else {
750 		error = -ENODEV;
751 		new->unit = mddev_alloc_unit();
752 		if (!new->unit)
753 			goto out_free_new;
754 		new->md_minor = MINOR(new->unit);
755 		new->hold_active = UNTIL_STOP;
756 	}
757 
758 	list_add(&new->all_mddevs, &all_mddevs);
759 	spin_unlock(&all_mddevs_lock);
760 	return new;
761 out_free_new:
762 	spin_unlock(&all_mddevs_lock);
763 	kfree(new);
764 	return ERR_PTR(error);
765 }
766 
mddev_free(struct mddev * mddev)767 static void mddev_free(struct mddev *mddev)
768 {
769 	spin_lock(&all_mddevs_lock);
770 	list_del(&mddev->all_mddevs);
771 	spin_unlock(&all_mddevs_lock);
772 
773 	kfree(mddev);
774 }
775 
776 static const struct attribute_group md_redundancy_group;
777 
mddev_unlock(struct mddev * mddev)778 void mddev_unlock(struct mddev *mddev)
779 {
780 	struct md_rdev *rdev;
781 	struct md_rdev *tmp;
782 	LIST_HEAD(delete);
783 
784 	if (!list_empty(&mddev->deleting))
785 		list_splice_init(&mddev->deleting, &delete);
786 
787 	if (mddev->to_remove) {
788 		/* These cannot be removed under reconfig_mutex as
789 		 * an access to the files will try to take reconfig_mutex
790 		 * while holding the file unremovable, which leads to
791 		 * a deadlock.
792 		 * So hold set sysfs_active while the remove in happeing,
793 		 * and anything else which might set ->to_remove or my
794 		 * otherwise change the sysfs namespace will fail with
795 		 * -EBUSY if sysfs_active is still set.
796 		 * We set sysfs_active under reconfig_mutex and elsewhere
797 		 * test it under the same mutex to ensure its correct value
798 		 * is seen.
799 		 */
800 		const struct attribute_group *to_remove = mddev->to_remove;
801 		mddev->to_remove = NULL;
802 		mddev->sysfs_active = 1;
803 		mutex_unlock(&mddev->reconfig_mutex);
804 
805 		if (mddev->kobj.sd) {
806 			if (to_remove != &md_redundancy_group)
807 				sysfs_remove_group(&mddev->kobj, to_remove);
808 			if (mddev->pers == NULL ||
809 			    mddev->pers->sync_request == NULL) {
810 				sysfs_remove_group(&mddev->kobj, &md_redundancy_group);
811 				if (mddev->sysfs_action)
812 					sysfs_put(mddev->sysfs_action);
813 				if (mddev->sysfs_completed)
814 					sysfs_put(mddev->sysfs_completed);
815 				if (mddev->sysfs_degraded)
816 					sysfs_put(mddev->sysfs_degraded);
817 				mddev->sysfs_action = NULL;
818 				mddev->sysfs_completed = NULL;
819 				mddev->sysfs_degraded = NULL;
820 			}
821 		}
822 		mddev->sysfs_active = 0;
823 	} else
824 		mutex_unlock(&mddev->reconfig_mutex);
825 
826 	md_wakeup_thread(mddev->thread);
827 	wake_up(&mddev->sb_wait);
828 
829 	list_for_each_entry_safe(rdev, tmp, &delete, same_set) {
830 		list_del_init(&rdev->same_set);
831 		kobject_del(&rdev->kobj);
832 		export_rdev(rdev, mddev);
833 	}
834 }
835 EXPORT_SYMBOL_GPL(mddev_unlock);
836 
md_find_rdev_nr_rcu(struct mddev * mddev,int nr)837 struct md_rdev *md_find_rdev_nr_rcu(struct mddev *mddev, int nr)
838 {
839 	struct md_rdev *rdev;
840 
841 	rdev_for_each_rcu(rdev, mddev)
842 		if (rdev->desc_nr == nr)
843 			return rdev;
844 
845 	return NULL;
846 }
847 EXPORT_SYMBOL_GPL(md_find_rdev_nr_rcu);
848 
find_rdev(struct mddev * mddev,dev_t dev)849 static struct md_rdev *find_rdev(struct mddev *mddev, dev_t dev)
850 {
851 	struct md_rdev *rdev;
852 
853 	rdev_for_each(rdev, mddev)
854 		if (rdev->bdev->bd_dev == dev)
855 			return rdev;
856 
857 	return NULL;
858 }
859 
md_find_rdev_rcu(struct mddev * mddev,dev_t dev)860 struct md_rdev *md_find_rdev_rcu(struct mddev *mddev, dev_t dev)
861 {
862 	struct md_rdev *rdev;
863 
864 	rdev_for_each_rcu(rdev, mddev)
865 		if (rdev->bdev->bd_dev == dev)
866 			return rdev;
867 
868 	return NULL;
869 }
870 EXPORT_SYMBOL_GPL(md_find_rdev_rcu);
871 
find_pers(int level,char * clevel)872 static struct md_personality *find_pers(int level, char *clevel)
873 {
874 	struct md_personality *pers;
875 	list_for_each_entry(pers, &pers_list, list) {
876 		if (level != LEVEL_NONE && pers->level == level)
877 			return pers;
878 		if (strcmp(pers->name, clevel)==0)
879 			return pers;
880 	}
881 	return NULL;
882 }
883 
884 /* return the offset of the super block in 512byte sectors */
calc_dev_sboffset(struct md_rdev * rdev)885 static inline sector_t calc_dev_sboffset(struct md_rdev *rdev)
886 {
887 	return MD_NEW_SIZE_SECTORS(bdev_nr_sectors(rdev->bdev));
888 }
889 
alloc_disk_sb(struct md_rdev * rdev)890 static int alloc_disk_sb(struct md_rdev *rdev)
891 {
892 	rdev->sb_page = alloc_page(GFP_KERNEL);
893 	if (!rdev->sb_page)
894 		return -ENOMEM;
895 	return 0;
896 }
897 
md_rdev_clear(struct md_rdev * rdev)898 void md_rdev_clear(struct md_rdev *rdev)
899 {
900 	if (rdev->sb_page) {
901 		put_page(rdev->sb_page);
902 		rdev->sb_loaded = 0;
903 		rdev->sb_page = NULL;
904 		rdev->sb_start = 0;
905 		rdev->sectors = 0;
906 	}
907 	if (rdev->bb_page) {
908 		put_page(rdev->bb_page);
909 		rdev->bb_page = NULL;
910 	}
911 	badblocks_exit(&rdev->badblocks);
912 }
913 EXPORT_SYMBOL_GPL(md_rdev_clear);
914 
super_written(struct bio * bio)915 static void super_written(struct bio *bio)
916 {
917 	struct md_rdev *rdev = bio->bi_private;
918 	struct mddev *mddev = rdev->mddev;
919 
920 	if (bio->bi_status) {
921 		pr_err("md: %s gets error=%d\n", __func__,
922 		       blk_status_to_errno(bio->bi_status));
923 		md_error(mddev, rdev);
924 		if (!test_bit(Faulty, &rdev->flags)
925 		    && (bio->bi_opf & MD_FAILFAST)) {
926 			set_bit(MD_SB_NEED_REWRITE, &mddev->sb_flags);
927 			set_bit(LastDev, &rdev->flags);
928 		}
929 	} else
930 		clear_bit(LastDev, &rdev->flags);
931 
932 	bio_put(bio);
933 
934 	rdev_dec_pending(rdev, mddev);
935 
936 	if (atomic_dec_and_test(&mddev->pending_writes))
937 		wake_up(&mddev->sb_wait);
938 }
939 
md_super_write(struct mddev * mddev,struct md_rdev * rdev,sector_t sector,int size,struct page * page)940 void md_super_write(struct mddev *mddev, struct md_rdev *rdev,
941 		   sector_t sector, int size, struct page *page)
942 {
943 	/* write first size bytes of page to sector of rdev
944 	 * Increment mddev->pending_writes before returning
945 	 * and decrement it on completion, waking up sb_wait
946 	 * if zero is reached.
947 	 * If an error occurred, call md_error
948 	 */
949 	struct bio *bio;
950 
951 	if (!page)
952 		return;
953 
954 	if (test_bit(Faulty, &rdev->flags))
955 		return;
956 
957 	bio = bio_alloc_bioset(rdev->meta_bdev ? rdev->meta_bdev : rdev->bdev,
958 			      1,
959 			      REQ_OP_WRITE | REQ_SYNC | REQ_IDLE | REQ_META
960 				  | REQ_PREFLUSH | REQ_FUA,
961 			      GFP_NOIO, &mddev->sync_set);
962 
963 	atomic_inc(&rdev->nr_pending);
964 
965 	bio->bi_iter.bi_sector = sector;
966 	__bio_add_page(bio, page, size, 0);
967 	bio->bi_private = rdev;
968 	bio->bi_end_io = super_written;
969 
970 	if (test_bit(MD_FAILFAST_SUPPORTED, &mddev->flags) &&
971 	    test_bit(FailFast, &rdev->flags) &&
972 	    !test_bit(LastDev, &rdev->flags))
973 		bio->bi_opf |= MD_FAILFAST;
974 
975 	atomic_inc(&mddev->pending_writes);
976 	submit_bio(bio);
977 }
978 
md_super_wait(struct mddev * mddev)979 int md_super_wait(struct mddev *mddev)
980 {
981 	/* wait for all superblock writes that were scheduled to complete */
982 	wait_event(mddev->sb_wait, atomic_read(&mddev->pending_writes)==0);
983 	if (test_and_clear_bit(MD_SB_NEED_REWRITE, &mddev->sb_flags))
984 		return -EAGAIN;
985 	return 0;
986 }
987 
sync_page_io(struct md_rdev * rdev,sector_t sector,int size,struct page * page,blk_opf_t opf,bool metadata_op)988 int sync_page_io(struct md_rdev *rdev, sector_t sector, int size,
989 		 struct page *page, blk_opf_t opf, bool metadata_op)
990 {
991 	struct bio bio;
992 	struct bio_vec bvec;
993 
994 	if (metadata_op && rdev->meta_bdev)
995 		bio_init(&bio, rdev->meta_bdev, &bvec, 1, opf);
996 	else
997 		bio_init(&bio, rdev->bdev, &bvec, 1, opf);
998 
999 	if (metadata_op)
1000 		bio.bi_iter.bi_sector = sector + rdev->sb_start;
1001 	else if (rdev->mddev->reshape_position != MaxSector &&
1002 		 (rdev->mddev->reshape_backwards ==
1003 		  (sector >= rdev->mddev->reshape_position)))
1004 		bio.bi_iter.bi_sector = sector + rdev->new_data_offset;
1005 	else
1006 		bio.bi_iter.bi_sector = sector + rdev->data_offset;
1007 	__bio_add_page(&bio, page, size, 0);
1008 
1009 	submit_bio_wait(&bio);
1010 
1011 	return !bio.bi_status;
1012 }
1013 EXPORT_SYMBOL_GPL(sync_page_io);
1014 
read_disk_sb(struct md_rdev * rdev,int size)1015 static int read_disk_sb(struct md_rdev *rdev, int size)
1016 {
1017 	if (rdev->sb_loaded)
1018 		return 0;
1019 
1020 	if (!sync_page_io(rdev, 0, size, rdev->sb_page, REQ_OP_READ, true))
1021 		goto fail;
1022 	rdev->sb_loaded = 1;
1023 	return 0;
1024 
1025 fail:
1026 	pr_err("md: disabled device %pg, could not read superblock.\n",
1027 	       rdev->bdev);
1028 	return -EINVAL;
1029 }
1030 
md_uuid_equal(mdp_super_t * sb1,mdp_super_t * sb2)1031 static int md_uuid_equal(mdp_super_t *sb1, mdp_super_t *sb2)
1032 {
1033 	return	sb1->set_uuid0 == sb2->set_uuid0 &&
1034 		sb1->set_uuid1 == sb2->set_uuid1 &&
1035 		sb1->set_uuid2 == sb2->set_uuid2 &&
1036 		sb1->set_uuid3 == sb2->set_uuid3;
1037 }
1038 
md_sb_equal(mdp_super_t * sb1,mdp_super_t * sb2)1039 static int md_sb_equal(mdp_super_t *sb1, mdp_super_t *sb2)
1040 {
1041 	int ret;
1042 	mdp_super_t *tmp1, *tmp2;
1043 
1044 	tmp1 = kmalloc(sizeof(*tmp1),GFP_KERNEL);
1045 	tmp2 = kmalloc(sizeof(*tmp2),GFP_KERNEL);
1046 
1047 	if (!tmp1 || !tmp2) {
1048 		ret = 0;
1049 		goto abort;
1050 	}
1051 
1052 	*tmp1 = *sb1;
1053 	*tmp2 = *sb2;
1054 
1055 	/*
1056 	 * nr_disks is not constant
1057 	 */
1058 	tmp1->nr_disks = 0;
1059 	tmp2->nr_disks = 0;
1060 
1061 	ret = (memcmp(tmp1, tmp2, MD_SB_GENERIC_CONSTANT_WORDS * 4) == 0);
1062 abort:
1063 	kfree(tmp1);
1064 	kfree(tmp2);
1065 	return ret;
1066 }
1067 
md_csum_fold(u32 csum)1068 static u32 md_csum_fold(u32 csum)
1069 {
1070 	csum = (csum & 0xffff) + (csum >> 16);
1071 	return (csum & 0xffff) + (csum >> 16);
1072 }
1073 
calc_sb_csum(mdp_super_t * sb)1074 static unsigned int calc_sb_csum(mdp_super_t *sb)
1075 {
1076 	u64 newcsum = 0;
1077 	u32 *sb32 = (u32*)sb;
1078 	int i;
1079 	unsigned int disk_csum, csum;
1080 
1081 	disk_csum = sb->sb_csum;
1082 	sb->sb_csum = 0;
1083 
1084 	for (i = 0; i < MD_SB_BYTES/4 ; i++)
1085 		newcsum += sb32[i];
1086 	csum = (newcsum & 0xffffffff) + (newcsum>>32);
1087 
1088 #ifdef CONFIG_ALPHA
1089 	/* This used to use csum_partial, which was wrong for several
1090 	 * reasons including that different results are returned on
1091 	 * different architectures.  It isn't critical that we get exactly
1092 	 * the same return value as before (we always csum_fold before
1093 	 * testing, and that removes any differences).  However as we
1094 	 * know that csum_partial always returned a 16bit value on
1095 	 * alphas, do a fold to maximise conformity to previous behaviour.
1096 	 */
1097 	sb->sb_csum = md_csum_fold(disk_csum);
1098 #else
1099 	sb->sb_csum = disk_csum;
1100 #endif
1101 	return csum;
1102 }
1103 
1104 /*
1105  * Handle superblock details.
1106  * We want to be able to handle multiple superblock formats
1107  * so we have a common interface to them all, and an array of
1108  * different handlers.
1109  * We rely on user-space to write the initial superblock, and support
1110  * reading and updating of superblocks.
1111  * Interface methods are:
1112  *   int load_super(struct md_rdev *dev, struct md_rdev *refdev, int minor_version)
1113  *      loads and validates a superblock on dev.
1114  *      if refdev != NULL, compare superblocks on both devices
1115  *    Return:
1116  *      0 - dev has a superblock that is compatible with refdev
1117  *      1 - dev has a superblock that is compatible and newer than refdev
1118  *          so dev should be used as the refdev in future
1119  *     -EINVAL superblock incompatible or invalid
1120  *     -othererror e.g. -EIO
1121  *
1122  *   int validate_super(struct mddev *mddev, struct md_rdev *dev)
1123  *      Verify that dev is acceptable into mddev.
1124  *       The first time, mddev->raid_disks will be 0, and data from
1125  *       dev should be merged in.  Subsequent calls check that dev
1126  *       is new enough.  Return 0 or -EINVAL
1127  *
1128  *   void sync_super(struct mddev *mddev, struct md_rdev *dev)
1129  *     Update the superblock for rdev with data in mddev
1130  *     This does not write to disc.
1131  *
1132  */
1133 
1134 struct super_type  {
1135 	char		    *name;
1136 	struct module	    *owner;
1137 	int		    (*load_super)(struct md_rdev *rdev,
1138 					  struct md_rdev *refdev,
1139 					  int minor_version);
1140 	int		    (*validate_super)(struct mddev *mddev,
1141 					      struct md_rdev *freshest,
1142 					      struct md_rdev *rdev);
1143 	void		    (*sync_super)(struct mddev *mddev,
1144 					  struct md_rdev *rdev);
1145 	unsigned long long  (*rdev_size_change)(struct md_rdev *rdev,
1146 						sector_t num_sectors);
1147 	int		    (*allow_new_offset)(struct md_rdev *rdev,
1148 						unsigned long long new_offset);
1149 };
1150 
1151 /*
1152  * Check that the given mddev has no bitmap.
1153  *
1154  * This function is called from the run method of all personalities that do not
1155  * support bitmaps. It prints an error message and returns non-zero if mddev
1156  * has a bitmap. Otherwise, it returns 0.
1157  *
1158  */
md_check_no_bitmap(struct mddev * mddev)1159 int md_check_no_bitmap(struct mddev *mddev)
1160 {
1161 	if (!mddev->bitmap_info.file && !mddev->bitmap_info.offset)
1162 		return 0;
1163 	pr_warn("%s: bitmaps are not supported for %s\n",
1164 		mdname(mddev), mddev->pers->name);
1165 	return 1;
1166 }
1167 EXPORT_SYMBOL(md_check_no_bitmap);
1168 
1169 /*
1170  * load_super for 0.90.0
1171  */
super_90_load(struct md_rdev * rdev,struct md_rdev * refdev,int minor_version)1172 static int super_90_load(struct md_rdev *rdev, struct md_rdev *refdev, int minor_version)
1173 {
1174 	mdp_super_t *sb;
1175 	int ret;
1176 	bool spare_disk = true;
1177 
1178 	/*
1179 	 * Calculate the position of the superblock (512byte sectors),
1180 	 * it's at the end of the disk.
1181 	 *
1182 	 * It also happens to be a multiple of 4Kb.
1183 	 */
1184 	rdev->sb_start = calc_dev_sboffset(rdev);
1185 
1186 	ret = read_disk_sb(rdev, MD_SB_BYTES);
1187 	if (ret)
1188 		return ret;
1189 
1190 	ret = -EINVAL;
1191 
1192 	sb = page_address(rdev->sb_page);
1193 
1194 	if (sb->md_magic != MD_SB_MAGIC) {
1195 		pr_warn("md: invalid raid superblock magic on %pg\n",
1196 			rdev->bdev);
1197 		goto abort;
1198 	}
1199 
1200 	if (sb->major_version != 0 ||
1201 	    sb->minor_version < 90 ||
1202 	    sb->minor_version > 91) {
1203 		pr_warn("Bad version number %d.%d on %pg\n",
1204 			sb->major_version, sb->minor_version, rdev->bdev);
1205 		goto abort;
1206 	}
1207 
1208 	if (sb->raid_disks <= 0)
1209 		goto abort;
1210 
1211 	if (md_csum_fold(calc_sb_csum(sb)) != md_csum_fold(sb->sb_csum)) {
1212 		pr_warn("md: invalid superblock checksum on %pg\n", rdev->bdev);
1213 		goto abort;
1214 	}
1215 
1216 	rdev->preferred_minor = sb->md_minor;
1217 	rdev->data_offset = 0;
1218 	rdev->new_data_offset = 0;
1219 	rdev->sb_size = MD_SB_BYTES;
1220 	rdev->badblocks.shift = -1;
1221 
1222 	if (sb->level == LEVEL_MULTIPATH)
1223 		rdev->desc_nr = -1;
1224 	else
1225 		rdev->desc_nr = sb->this_disk.number;
1226 
1227 	/* not spare disk, or LEVEL_MULTIPATH */
1228 	if (sb->level == LEVEL_MULTIPATH ||
1229 		(rdev->desc_nr >= 0 &&
1230 		 rdev->desc_nr < MD_SB_DISKS &&
1231 		 sb->disks[rdev->desc_nr].state &
1232 		 ((1<<MD_DISK_SYNC) | (1 << MD_DISK_ACTIVE))))
1233 		spare_disk = false;
1234 
1235 	if (!refdev) {
1236 		if (!spare_disk)
1237 			ret = 1;
1238 		else
1239 			ret = 0;
1240 	} else {
1241 		__u64 ev1, ev2;
1242 		mdp_super_t *refsb = page_address(refdev->sb_page);
1243 		if (!md_uuid_equal(refsb, sb)) {
1244 			pr_warn("md: %pg has different UUID to %pg\n",
1245 				rdev->bdev, refdev->bdev);
1246 			goto abort;
1247 		}
1248 		if (!md_sb_equal(refsb, sb)) {
1249 			pr_warn("md: %pg has same UUID but different superblock to %pg\n",
1250 				rdev->bdev, refdev->bdev);
1251 			goto abort;
1252 		}
1253 		ev1 = md_event(sb);
1254 		ev2 = md_event(refsb);
1255 
1256 		if (!spare_disk && ev1 > ev2)
1257 			ret = 1;
1258 		else
1259 			ret = 0;
1260 	}
1261 	rdev->sectors = rdev->sb_start;
1262 	/* Limit to 4TB as metadata cannot record more than that.
1263 	 * (not needed for Linear and RAID0 as metadata doesn't
1264 	 * record this size)
1265 	 */
1266 	if ((u64)rdev->sectors >= (2ULL << 32) && sb->level >= 1)
1267 		rdev->sectors = (sector_t)(2ULL << 32) - 2;
1268 
1269 	if (rdev->sectors < ((sector_t)sb->size) * 2 && sb->level >= 1)
1270 		/* "this cannot possibly happen" ... */
1271 		ret = -EINVAL;
1272 
1273  abort:
1274 	return ret;
1275 }
1276 
1277 /*
1278  * validate_super for 0.90.0
1279  * note: we are not using "freshest" for 0.9 superblock
1280  */
super_90_validate(struct mddev * mddev,struct md_rdev * freshest,struct md_rdev * rdev)1281 static int super_90_validate(struct mddev *mddev, struct md_rdev *freshest, struct md_rdev *rdev)
1282 {
1283 	mdp_disk_t *desc;
1284 	mdp_super_t *sb = page_address(rdev->sb_page);
1285 	__u64 ev1 = md_event(sb);
1286 
1287 	rdev->raid_disk = -1;
1288 	clear_bit(Faulty, &rdev->flags);
1289 	clear_bit(In_sync, &rdev->flags);
1290 	clear_bit(Bitmap_sync, &rdev->flags);
1291 	clear_bit(WriteMostly, &rdev->flags);
1292 
1293 	if (mddev->raid_disks == 0) {
1294 		mddev->major_version = 0;
1295 		mddev->minor_version = sb->minor_version;
1296 		mddev->patch_version = sb->patch_version;
1297 		mddev->external = 0;
1298 		mddev->chunk_sectors = sb->chunk_size >> 9;
1299 		mddev->ctime = sb->ctime;
1300 		mddev->utime = sb->utime;
1301 		mddev->level = sb->level;
1302 		mddev->clevel[0] = 0;
1303 		mddev->layout = sb->layout;
1304 		mddev->raid_disks = sb->raid_disks;
1305 		mddev->dev_sectors = ((sector_t)sb->size) * 2;
1306 		mddev->events = ev1;
1307 		mddev->bitmap_info.offset = 0;
1308 		mddev->bitmap_info.space = 0;
1309 		/* bitmap can use 60 K after the 4K superblocks */
1310 		mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9;
1311 		mddev->bitmap_info.default_space = 64*2 - (MD_SB_BYTES >> 9);
1312 		mddev->reshape_backwards = 0;
1313 
1314 		if (mddev->minor_version >= 91) {
1315 			mddev->reshape_position = sb->reshape_position;
1316 			mddev->delta_disks = sb->delta_disks;
1317 			mddev->new_level = sb->new_level;
1318 			mddev->new_layout = sb->new_layout;
1319 			mddev->new_chunk_sectors = sb->new_chunk >> 9;
1320 			if (mddev->delta_disks < 0)
1321 				mddev->reshape_backwards = 1;
1322 		} else {
1323 			mddev->reshape_position = MaxSector;
1324 			mddev->delta_disks = 0;
1325 			mddev->new_level = mddev->level;
1326 			mddev->new_layout = mddev->layout;
1327 			mddev->new_chunk_sectors = mddev->chunk_sectors;
1328 		}
1329 		if (mddev->level == 0)
1330 			mddev->layout = -1;
1331 
1332 		if (sb->state & (1<<MD_SB_CLEAN))
1333 			mddev->recovery_cp = MaxSector;
1334 		else {
1335 			if (sb->events_hi == sb->cp_events_hi &&
1336 				sb->events_lo == sb->cp_events_lo) {
1337 				mddev->recovery_cp = sb->recovery_cp;
1338 			} else
1339 				mddev->recovery_cp = 0;
1340 		}
1341 
1342 		memcpy(mddev->uuid+0, &sb->set_uuid0, 4);
1343 		memcpy(mddev->uuid+4, &sb->set_uuid1, 4);
1344 		memcpy(mddev->uuid+8, &sb->set_uuid2, 4);
1345 		memcpy(mddev->uuid+12,&sb->set_uuid3, 4);
1346 
1347 		mddev->max_disks = MD_SB_DISKS;
1348 
1349 		if (sb->state & (1<<MD_SB_BITMAP_PRESENT) &&
1350 		    mddev->bitmap_info.file == NULL) {
1351 			mddev->bitmap_info.offset =
1352 				mddev->bitmap_info.default_offset;
1353 			mddev->bitmap_info.space =
1354 				mddev->bitmap_info.default_space;
1355 		}
1356 
1357 	} else if (mddev->pers == NULL) {
1358 		/* Insist on good event counter while assembling, except
1359 		 * for spares (which don't need an event count) */
1360 		++ev1;
1361 		if (sb->disks[rdev->desc_nr].state & (
1362 			    (1<<MD_DISK_SYNC) | (1 << MD_DISK_ACTIVE)))
1363 			if (ev1 < mddev->events)
1364 				return -EINVAL;
1365 	} else if (mddev->bitmap) {
1366 		/* if adding to array with a bitmap, then we can accept an
1367 		 * older device ... but not too old.
1368 		 */
1369 		if (ev1 < mddev->bitmap->events_cleared)
1370 			return 0;
1371 		if (ev1 < mddev->events)
1372 			set_bit(Bitmap_sync, &rdev->flags);
1373 	} else {
1374 		if (ev1 < mddev->events)
1375 			/* just a hot-add of a new device, leave raid_disk at -1 */
1376 			return 0;
1377 	}
1378 
1379 	if (mddev->level != LEVEL_MULTIPATH) {
1380 		desc = sb->disks + rdev->desc_nr;
1381 
1382 		if (desc->state & (1<<MD_DISK_FAULTY))
1383 			set_bit(Faulty, &rdev->flags);
1384 		else if (desc->state & (1<<MD_DISK_SYNC) /* &&
1385 			    desc->raid_disk < mddev->raid_disks */) {
1386 			set_bit(In_sync, &rdev->flags);
1387 			rdev->raid_disk = desc->raid_disk;
1388 			rdev->saved_raid_disk = desc->raid_disk;
1389 		} else if (desc->state & (1<<MD_DISK_ACTIVE)) {
1390 			/* active but not in sync implies recovery up to
1391 			 * reshape position.  We don't know exactly where
1392 			 * that is, so set to zero for now */
1393 			if (mddev->minor_version >= 91) {
1394 				rdev->recovery_offset = 0;
1395 				rdev->raid_disk = desc->raid_disk;
1396 			}
1397 		}
1398 		if (desc->state & (1<<MD_DISK_WRITEMOSTLY))
1399 			set_bit(WriteMostly, &rdev->flags);
1400 		if (desc->state & (1<<MD_DISK_FAILFAST))
1401 			set_bit(FailFast, &rdev->flags);
1402 	} else /* MULTIPATH are always insync */
1403 		set_bit(In_sync, &rdev->flags);
1404 	return 0;
1405 }
1406 
1407 /*
1408  * sync_super for 0.90.0
1409  */
super_90_sync(struct mddev * mddev,struct md_rdev * rdev)1410 static void super_90_sync(struct mddev *mddev, struct md_rdev *rdev)
1411 {
1412 	mdp_super_t *sb;
1413 	struct md_rdev *rdev2;
1414 	int next_spare = mddev->raid_disks;
1415 
1416 	/* make rdev->sb match mddev data..
1417 	 *
1418 	 * 1/ zero out disks
1419 	 * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
1420 	 * 3/ any empty disks < next_spare become removed
1421 	 *
1422 	 * disks[0] gets initialised to REMOVED because
1423 	 * we cannot be sure from other fields if it has
1424 	 * been initialised or not.
1425 	 */
1426 	int i;
1427 	int active=0, working=0,failed=0,spare=0,nr_disks=0;
1428 
1429 	rdev->sb_size = MD_SB_BYTES;
1430 
1431 	sb = page_address(rdev->sb_page);
1432 
1433 	memset(sb, 0, sizeof(*sb));
1434 
1435 	sb->md_magic = MD_SB_MAGIC;
1436 	sb->major_version = mddev->major_version;
1437 	sb->patch_version = mddev->patch_version;
1438 	sb->gvalid_words  = 0; /* ignored */
1439 	memcpy(&sb->set_uuid0, mddev->uuid+0, 4);
1440 	memcpy(&sb->set_uuid1, mddev->uuid+4, 4);
1441 	memcpy(&sb->set_uuid2, mddev->uuid+8, 4);
1442 	memcpy(&sb->set_uuid3, mddev->uuid+12,4);
1443 
1444 	sb->ctime = clamp_t(time64_t, mddev->ctime, 0, U32_MAX);
1445 	sb->level = mddev->level;
1446 	sb->size = mddev->dev_sectors / 2;
1447 	sb->raid_disks = mddev->raid_disks;
1448 	sb->md_minor = mddev->md_minor;
1449 	sb->not_persistent = 0;
1450 	sb->utime = clamp_t(time64_t, mddev->utime, 0, U32_MAX);
1451 	sb->state = 0;
1452 	sb->events_hi = (mddev->events>>32);
1453 	sb->events_lo = (u32)mddev->events;
1454 
1455 	if (mddev->reshape_position == MaxSector)
1456 		sb->minor_version = 90;
1457 	else {
1458 		sb->minor_version = 91;
1459 		sb->reshape_position = mddev->reshape_position;
1460 		sb->new_level = mddev->new_level;
1461 		sb->delta_disks = mddev->delta_disks;
1462 		sb->new_layout = mddev->new_layout;
1463 		sb->new_chunk = mddev->new_chunk_sectors << 9;
1464 	}
1465 	mddev->minor_version = sb->minor_version;
1466 	if (mddev->in_sync)
1467 	{
1468 		sb->recovery_cp = mddev->recovery_cp;
1469 		sb->cp_events_hi = (mddev->events>>32);
1470 		sb->cp_events_lo = (u32)mddev->events;
1471 		if (mddev->recovery_cp == MaxSector)
1472 			sb->state = (1<< MD_SB_CLEAN);
1473 	} else
1474 		sb->recovery_cp = 0;
1475 
1476 	sb->layout = mddev->layout;
1477 	sb->chunk_size = mddev->chunk_sectors << 9;
1478 
1479 	if (mddev->bitmap && mddev->bitmap_info.file == NULL)
1480 		sb->state |= (1<<MD_SB_BITMAP_PRESENT);
1481 
1482 	sb->disks[0].state = (1<<MD_DISK_REMOVED);
1483 	rdev_for_each(rdev2, mddev) {
1484 		mdp_disk_t *d;
1485 		int desc_nr;
1486 		int is_active = test_bit(In_sync, &rdev2->flags);
1487 
1488 		if (rdev2->raid_disk >= 0 &&
1489 		    sb->minor_version >= 91)
1490 			/* we have nowhere to store the recovery_offset,
1491 			 * but if it is not below the reshape_position,
1492 			 * we can piggy-back on that.
1493 			 */
1494 			is_active = 1;
1495 		if (rdev2->raid_disk < 0 ||
1496 		    test_bit(Faulty, &rdev2->flags))
1497 			is_active = 0;
1498 		if (is_active)
1499 			desc_nr = rdev2->raid_disk;
1500 		else
1501 			desc_nr = next_spare++;
1502 		rdev2->desc_nr = desc_nr;
1503 		d = &sb->disks[rdev2->desc_nr];
1504 		nr_disks++;
1505 		d->number = rdev2->desc_nr;
1506 		d->major = MAJOR(rdev2->bdev->bd_dev);
1507 		d->minor = MINOR(rdev2->bdev->bd_dev);
1508 		if (is_active)
1509 			d->raid_disk = rdev2->raid_disk;
1510 		else
1511 			d->raid_disk = rdev2->desc_nr; /* compatibility */
1512 		if (test_bit(Faulty, &rdev2->flags))
1513 			d->state = (1<<MD_DISK_FAULTY);
1514 		else if (is_active) {
1515 			d->state = (1<<MD_DISK_ACTIVE);
1516 			if (test_bit(In_sync, &rdev2->flags))
1517 				d->state |= (1<<MD_DISK_SYNC);
1518 			active++;
1519 			working++;
1520 		} else {
1521 			d->state = 0;
1522 			spare++;
1523 			working++;
1524 		}
1525 		if (test_bit(WriteMostly, &rdev2->flags))
1526 			d->state |= (1<<MD_DISK_WRITEMOSTLY);
1527 		if (test_bit(FailFast, &rdev2->flags))
1528 			d->state |= (1<<MD_DISK_FAILFAST);
1529 	}
1530 	/* now set the "removed" and "faulty" bits on any missing devices */
1531 	for (i=0 ; i < mddev->raid_disks ; i++) {
1532 		mdp_disk_t *d = &sb->disks[i];
1533 		if (d->state == 0 && d->number == 0) {
1534 			d->number = i;
1535 			d->raid_disk = i;
1536 			d->state = (1<<MD_DISK_REMOVED);
1537 			d->state |= (1<<MD_DISK_FAULTY);
1538 			failed++;
1539 		}
1540 	}
1541 	sb->nr_disks = nr_disks;
1542 	sb->active_disks = active;
1543 	sb->working_disks = working;
1544 	sb->failed_disks = failed;
1545 	sb->spare_disks = spare;
1546 
1547 	sb->this_disk = sb->disks[rdev->desc_nr];
1548 	sb->sb_csum = calc_sb_csum(sb);
1549 }
1550 
1551 /*
1552  * rdev_size_change for 0.90.0
1553  */
1554 static unsigned long long
super_90_rdev_size_change(struct md_rdev * rdev,sector_t num_sectors)1555 super_90_rdev_size_change(struct md_rdev *rdev, sector_t num_sectors)
1556 {
1557 	if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
1558 		return 0; /* component must fit device */
1559 	if (rdev->mddev->bitmap_info.offset)
1560 		return 0; /* can't move bitmap */
1561 	rdev->sb_start = calc_dev_sboffset(rdev);
1562 	if (!num_sectors || num_sectors > rdev->sb_start)
1563 		num_sectors = rdev->sb_start;
1564 	/* Limit to 4TB as metadata cannot record more than that.
1565 	 * 4TB == 2^32 KB, or 2*2^32 sectors.
1566 	 */
1567 	if ((u64)num_sectors >= (2ULL << 32) && rdev->mddev->level >= 1)
1568 		num_sectors = (sector_t)(2ULL << 32) - 2;
1569 	do {
1570 		md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
1571 		       rdev->sb_page);
1572 	} while (md_super_wait(rdev->mddev) < 0);
1573 	return num_sectors;
1574 }
1575 
1576 static int
super_90_allow_new_offset(struct md_rdev * rdev,unsigned long long new_offset)1577 super_90_allow_new_offset(struct md_rdev *rdev, unsigned long long new_offset)
1578 {
1579 	/* non-zero offset changes not possible with v0.90 */
1580 	return new_offset == 0;
1581 }
1582 
1583 /*
1584  * version 1 superblock
1585  */
1586 
calc_sb_1_csum(struct mdp_superblock_1 * sb)1587 static __le32 calc_sb_1_csum(struct mdp_superblock_1 *sb)
1588 {
1589 	__le32 disk_csum;
1590 	u32 csum;
1591 	unsigned long long newcsum;
1592 	int size = 256 + le32_to_cpu(sb->max_dev)*2;
1593 	__le32 *isuper = (__le32*)sb;
1594 
1595 	disk_csum = sb->sb_csum;
1596 	sb->sb_csum = 0;
1597 	newcsum = 0;
1598 	for (; size >= 4; size -= 4)
1599 		newcsum += le32_to_cpu(*isuper++);
1600 
1601 	if (size == 2)
1602 		newcsum += le16_to_cpu(*(__le16*) isuper);
1603 
1604 	csum = (newcsum & 0xffffffff) + (newcsum >> 32);
1605 	sb->sb_csum = disk_csum;
1606 	return cpu_to_le32(csum);
1607 }
1608 
super_1_load(struct md_rdev * rdev,struct md_rdev * refdev,int minor_version)1609 static int super_1_load(struct md_rdev *rdev, struct md_rdev *refdev, int minor_version)
1610 {
1611 	struct mdp_superblock_1 *sb;
1612 	int ret;
1613 	sector_t sb_start;
1614 	sector_t sectors;
1615 	int bmask;
1616 	bool spare_disk = true;
1617 
1618 	/*
1619 	 * Calculate the position of the superblock in 512byte sectors.
1620 	 * It is always aligned to a 4K boundary and
1621 	 * depeding on minor_version, it can be:
1622 	 * 0: At least 8K, but less than 12K, from end of device
1623 	 * 1: At start of device
1624 	 * 2: 4K from start of device.
1625 	 */
1626 	switch(minor_version) {
1627 	case 0:
1628 		sb_start = bdev_nr_sectors(rdev->bdev) - 8 * 2;
1629 		sb_start &= ~(sector_t)(4*2-1);
1630 		break;
1631 	case 1:
1632 		sb_start = 0;
1633 		break;
1634 	case 2:
1635 		sb_start = 8;
1636 		break;
1637 	default:
1638 		return -EINVAL;
1639 	}
1640 	rdev->sb_start = sb_start;
1641 
1642 	/* superblock is rarely larger than 1K, but it can be larger,
1643 	 * and it is safe to read 4k, so we do that
1644 	 */
1645 	ret = read_disk_sb(rdev, 4096);
1646 	if (ret) return ret;
1647 
1648 	sb = page_address(rdev->sb_page);
1649 
1650 	if (sb->magic != cpu_to_le32(MD_SB_MAGIC) ||
1651 	    sb->major_version != cpu_to_le32(1) ||
1652 	    le32_to_cpu(sb->max_dev) > (4096-256)/2 ||
1653 	    le64_to_cpu(sb->super_offset) != rdev->sb_start ||
1654 	    (le32_to_cpu(sb->feature_map) & ~MD_FEATURE_ALL) != 0)
1655 		return -EINVAL;
1656 
1657 	if (calc_sb_1_csum(sb) != sb->sb_csum) {
1658 		pr_warn("md: invalid superblock checksum on %pg\n",
1659 			rdev->bdev);
1660 		return -EINVAL;
1661 	}
1662 	if (le64_to_cpu(sb->data_size) < 10) {
1663 		pr_warn("md: data_size too small on %pg\n",
1664 			rdev->bdev);
1665 		return -EINVAL;
1666 	}
1667 	if (sb->pad0 ||
1668 	    sb->pad3[0] ||
1669 	    memcmp(sb->pad3, sb->pad3+1, sizeof(sb->pad3) - sizeof(sb->pad3[1])))
1670 		/* Some padding is non-zero, might be a new feature */
1671 		return -EINVAL;
1672 
1673 	rdev->preferred_minor = 0xffff;
1674 	rdev->data_offset = le64_to_cpu(sb->data_offset);
1675 	rdev->new_data_offset = rdev->data_offset;
1676 	if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE) &&
1677 	    (le32_to_cpu(sb->feature_map) & MD_FEATURE_NEW_OFFSET))
1678 		rdev->new_data_offset += (s32)le32_to_cpu(sb->new_offset);
1679 	atomic_set(&rdev->corrected_errors, le32_to_cpu(sb->cnt_corrected_read));
1680 
1681 	rdev->sb_size = le32_to_cpu(sb->max_dev) * 2 + 256;
1682 	bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
1683 	if (rdev->sb_size & bmask)
1684 		rdev->sb_size = (rdev->sb_size | bmask) + 1;
1685 
1686 	if (minor_version
1687 	    && rdev->data_offset < sb_start + (rdev->sb_size/512))
1688 		return -EINVAL;
1689 	if (minor_version
1690 	    && rdev->new_data_offset < sb_start + (rdev->sb_size/512))
1691 		return -EINVAL;
1692 
1693 	if (sb->level == cpu_to_le32(LEVEL_MULTIPATH))
1694 		rdev->desc_nr = -1;
1695 	else
1696 		rdev->desc_nr = le32_to_cpu(sb->dev_number);
1697 
1698 	if (!rdev->bb_page) {
1699 		rdev->bb_page = alloc_page(GFP_KERNEL);
1700 		if (!rdev->bb_page)
1701 			return -ENOMEM;
1702 	}
1703 	if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BAD_BLOCKS) &&
1704 	    rdev->badblocks.count == 0) {
1705 		/* need to load the bad block list.
1706 		 * Currently we limit it to one page.
1707 		 */
1708 		s32 offset;
1709 		sector_t bb_sector;
1710 		__le64 *bbp;
1711 		int i;
1712 		int sectors = le16_to_cpu(sb->bblog_size);
1713 		if (sectors > (PAGE_SIZE / 512))
1714 			return -EINVAL;
1715 		offset = le32_to_cpu(sb->bblog_offset);
1716 		if (offset == 0)
1717 			return -EINVAL;
1718 		bb_sector = (long long)offset;
1719 		if (!sync_page_io(rdev, bb_sector, sectors << 9,
1720 				  rdev->bb_page, REQ_OP_READ, true))
1721 			return -EIO;
1722 		bbp = (__le64 *)page_address(rdev->bb_page);
1723 		rdev->badblocks.shift = sb->bblog_shift;
1724 		for (i = 0 ; i < (sectors << (9-3)) ; i++, bbp++) {
1725 			u64 bb = le64_to_cpu(*bbp);
1726 			int count = bb & (0x3ff);
1727 			u64 sector = bb >> 10;
1728 			sector <<= sb->bblog_shift;
1729 			count <<= sb->bblog_shift;
1730 			if (bb + 1 == 0)
1731 				break;
1732 			if (badblocks_set(&rdev->badblocks, sector, count, 1))
1733 				return -EINVAL;
1734 		}
1735 	} else if (sb->bblog_offset != 0)
1736 		rdev->badblocks.shift = 0;
1737 
1738 	if ((le32_to_cpu(sb->feature_map) &
1739 	    (MD_FEATURE_PPL | MD_FEATURE_MULTIPLE_PPLS))) {
1740 		rdev->ppl.offset = (__s16)le16_to_cpu(sb->ppl.offset);
1741 		rdev->ppl.size = le16_to_cpu(sb->ppl.size);
1742 		rdev->ppl.sector = rdev->sb_start + rdev->ppl.offset;
1743 	}
1744 
1745 	if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RAID0_LAYOUT) &&
1746 	    sb->level != 0)
1747 		return -EINVAL;
1748 
1749 	/* not spare disk, or LEVEL_MULTIPATH */
1750 	if (sb->level == cpu_to_le32(LEVEL_MULTIPATH) ||
1751 		(rdev->desc_nr >= 0 &&
1752 		rdev->desc_nr < le32_to_cpu(sb->max_dev) &&
1753 		(le16_to_cpu(sb->dev_roles[rdev->desc_nr]) < MD_DISK_ROLE_MAX ||
1754 		 le16_to_cpu(sb->dev_roles[rdev->desc_nr]) == MD_DISK_ROLE_JOURNAL)))
1755 		spare_disk = false;
1756 
1757 	if (!refdev) {
1758 		if (!spare_disk)
1759 			ret = 1;
1760 		else
1761 			ret = 0;
1762 	} else {
1763 		__u64 ev1, ev2;
1764 		struct mdp_superblock_1 *refsb = page_address(refdev->sb_page);
1765 
1766 		if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 ||
1767 		    sb->level != refsb->level ||
1768 		    sb->layout != refsb->layout ||
1769 		    sb->chunksize != refsb->chunksize) {
1770 			pr_warn("md: %pg has strangely different superblock to %pg\n",
1771 				rdev->bdev,
1772 				refdev->bdev);
1773 			return -EINVAL;
1774 		}
1775 		ev1 = le64_to_cpu(sb->events);
1776 		ev2 = le64_to_cpu(refsb->events);
1777 
1778 		if (!spare_disk && ev1 > ev2)
1779 			ret = 1;
1780 		else
1781 			ret = 0;
1782 	}
1783 	if (minor_version)
1784 		sectors = bdev_nr_sectors(rdev->bdev) - rdev->data_offset;
1785 	else
1786 		sectors = rdev->sb_start;
1787 	if (sectors < le64_to_cpu(sb->data_size))
1788 		return -EINVAL;
1789 	rdev->sectors = le64_to_cpu(sb->data_size);
1790 	return ret;
1791 }
1792 
super_1_validate(struct mddev * mddev,struct md_rdev * freshest,struct md_rdev * rdev)1793 static int super_1_validate(struct mddev *mddev, struct md_rdev *freshest, struct md_rdev *rdev)
1794 {
1795 	struct mdp_superblock_1 *sb = page_address(rdev->sb_page);
1796 	__u64 ev1 = le64_to_cpu(sb->events);
1797 
1798 	rdev->raid_disk = -1;
1799 	clear_bit(Faulty, &rdev->flags);
1800 	clear_bit(In_sync, &rdev->flags);
1801 	clear_bit(Bitmap_sync, &rdev->flags);
1802 	clear_bit(WriteMostly, &rdev->flags);
1803 
1804 	if (mddev->raid_disks == 0) {
1805 		mddev->major_version = 1;
1806 		mddev->patch_version = 0;
1807 		mddev->external = 0;
1808 		mddev->chunk_sectors = le32_to_cpu(sb->chunksize);
1809 		mddev->ctime = le64_to_cpu(sb->ctime);
1810 		mddev->utime = le64_to_cpu(sb->utime);
1811 		mddev->level = le32_to_cpu(sb->level);
1812 		mddev->clevel[0] = 0;
1813 		mddev->layout = le32_to_cpu(sb->layout);
1814 		mddev->raid_disks = le32_to_cpu(sb->raid_disks);
1815 		mddev->dev_sectors = le64_to_cpu(sb->size);
1816 		mddev->events = ev1;
1817 		mddev->bitmap_info.offset = 0;
1818 		mddev->bitmap_info.space = 0;
1819 		/* Default location for bitmap is 1K after superblock
1820 		 * using 3K - total of 4K
1821 		 */
1822 		mddev->bitmap_info.default_offset = 1024 >> 9;
1823 		mddev->bitmap_info.default_space = (4096-1024) >> 9;
1824 		mddev->reshape_backwards = 0;
1825 
1826 		mddev->recovery_cp = le64_to_cpu(sb->resync_offset);
1827 		memcpy(mddev->uuid, sb->set_uuid, 16);
1828 
1829 		mddev->max_disks =  (4096-256)/2;
1830 
1831 		if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) &&
1832 		    mddev->bitmap_info.file == NULL) {
1833 			mddev->bitmap_info.offset =
1834 				(__s32)le32_to_cpu(sb->bitmap_offset);
1835 			/* Metadata doesn't record how much space is available.
1836 			 * For 1.0, we assume we can use up to the superblock
1837 			 * if before, else to 4K beyond superblock.
1838 			 * For others, assume no change is possible.
1839 			 */
1840 			if (mddev->minor_version > 0)
1841 				mddev->bitmap_info.space = 0;
1842 			else if (mddev->bitmap_info.offset > 0)
1843 				mddev->bitmap_info.space =
1844 					8 - mddev->bitmap_info.offset;
1845 			else
1846 				mddev->bitmap_info.space =
1847 					-mddev->bitmap_info.offset;
1848 		}
1849 
1850 		if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) {
1851 			mddev->reshape_position = le64_to_cpu(sb->reshape_position);
1852 			mddev->delta_disks = le32_to_cpu(sb->delta_disks);
1853 			mddev->new_level = le32_to_cpu(sb->new_level);
1854 			mddev->new_layout = le32_to_cpu(sb->new_layout);
1855 			mddev->new_chunk_sectors = le32_to_cpu(sb->new_chunk);
1856 			if (mddev->delta_disks < 0 ||
1857 			    (mddev->delta_disks == 0 &&
1858 			     (le32_to_cpu(sb->feature_map)
1859 			      & MD_FEATURE_RESHAPE_BACKWARDS)))
1860 				mddev->reshape_backwards = 1;
1861 		} else {
1862 			mddev->reshape_position = MaxSector;
1863 			mddev->delta_disks = 0;
1864 			mddev->new_level = mddev->level;
1865 			mddev->new_layout = mddev->layout;
1866 			mddev->new_chunk_sectors = mddev->chunk_sectors;
1867 		}
1868 
1869 		if (mddev->level == 0 &&
1870 		    !(le32_to_cpu(sb->feature_map) & MD_FEATURE_RAID0_LAYOUT))
1871 			mddev->layout = -1;
1872 
1873 		if (le32_to_cpu(sb->feature_map) & MD_FEATURE_JOURNAL)
1874 			set_bit(MD_HAS_JOURNAL, &mddev->flags);
1875 
1876 		if (le32_to_cpu(sb->feature_map) &
1877 		    (MD_FEATURE_PPL | MD_FEATURE_MULTIPLE_PPLS)) {
1878 			if (le32_to_cpu(sb->feature_map) &
1879 			    (MD_FEATURE_BITMAP_OFFSET | MD_FEATURE_JOURNAL))
1880 				return -EINVAL;
1881 			if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_PPL) &&
1882 			    (le32_to_cpu(sb->feature_map) &
1883 					    MD_FEATURE_MULTIPLE_PPLS))
1884 				return -EINVAL;
1885 			set_bit(MD_HAS_PPL, &mddev->flags);
1886 		}
1887 	} else if (mddev->pers == NULL) {
1888 		/* Insist of good event counter while assembling, except for
1889 		 * spares (which don't need an event count).
1890 		 * Similar to mdadm, we allow event counter difference of 1
1891 		 * from the freshest device.
1892 		 */
1893 		if (rdev->desc_nr >= 0 &&
1894 		    rdev->desc_nr < le32_to_cpu(sb->max_dev) &&
1895 		    (le16_to_cpu(sb->dev_roles[rdev->desc_nr]) < MD_DISK_ROLE_MAX ||
1896 		     le16_to_cpu(sb->dev_roles[rdev->desc_nr]) == MD_DISK_ROLE_JOURNAL))
1897 			if (ev1 + 1 < mddev->events)
1898 				return -EINVAL;
1899 	} else if (mddev->bitmap) {
1900 		/* If adding to array with a bitmap, then we can accept an
1901 		 * older device, but not too old.
1902 		 */
1903 		if (ev1 < mddev->bitmap->events_cleared)
1904 			return 0;
1905 		if (ev1 < mddev->events)
1906 			set_bit(Bitmap_sync, &rdev->flags);
1907 	} else {
1908 		if (ev1 < mddev->events)
1909 			/* just a hot-add of a new device, leave raid_disk at -1 */
1910 			return 0;
1911 	}
1912 	if (mddev->level != LEVEL_MULTIPATH) {
1913 		int role;
1914 		if (rdev->desc_nr < 0 ||
1915 		    rdev->desc_nr >= le32_to_cpu(sb->max_dev)) {
1916 			role = MD_DISK_ROLE_SPARE;
1917 			rdev->desc_nr = -1;
1918 		} else if (mddev->pers == NULL && freshest && ev1 < mddev->events) {
1919 			/*
1920 			 * If we are assembling, and our event counter is smaller than the
1921 			 * highest event counter, we cannot trust our superblock about the role.
1922 			 * It could happen that our rdev was marked as Faulty, and all other
1923 			 * superblocks were updated with +1 event counter.
1924 			 * Then, before the next superblock update, which typically happens when
1925 			 * remove_and_add_spares() removes the device from the array, there was
1926 			 * a crash or reboot.
1927 			 * If we allow current rdev without consulting the freshest superblock,
1928 			 * we could cause data corruption.
1929 			 * Note that in this case our event counter is smaller by 1 than the
1930 			 * highest, otherwise, this rdev would not be allowed into array;
1931 			 * both kernel and mdadm allow event counter difference of 1.
1932 			 */
1933 			struct mdp_superblock_1 *freshest_sb = page_address(freshest->sb_page);
1934 			u32 freshest_max_dev = le32_to_cpu(freshest_sb->max_dev);
1935 
1936 			if (rdev->desc_nr >= freshest_max_dev) {
1937 				/* this is unexpected, better not proceed */
1938 				pr_warn("md: %s: rdev[%pg]: desc_nr(%d) >= freshest(%pg)->sb->max_dev(%u)\n",
1939 						mdname(mddev), rdev->bdev, rdev->desc_nr,
1940 						freshest->bdev, freshest_max_dev);
1941 				return -EUCLEAN;
1942 			}
1943 
1944 			role = le16_to_cpu(freshest_sb->dev_roles[rdev->desc_nr]);
1945 			pr_debug("md: %s: rdev[%pg]: role=%d(0x%x) according to freshest %pg\n",
1946 				     mdname(mddev), rdev->bdev, role, role, freshest->bdev);
1947 		} else {
1948 			role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
1949 		}
1950 		switch(role) {
1951 		case MD_DISK_ROLE_SPARE: /* spare */
1952 			break;
1953 		case MD_DISK_ROLE_FAULTY: /* faulty */
1954 			set_bit(Faulty, &rdev->flags);
1955 			break;
1956 		case MD_DISK_ROLE_JOURNAL: /* journal device */
1957 			if (!(le32_to_cpu(sb->feature_map) & MD_FEATURE_JOURNAL)) {
1958 				/* journal device without journal feature */
1959 				pr_warn("md: journal device provided without journal feature, ignoring the device\n");
1960 				return -EINVAL;
1961 			}
1962 			set_bit(Journal, &rdev->flags);
1963 			rdev->journal_tail = le64_to_cpu(sb->journal_tail);
1964 			rdev->raid_disk = 0;
1965 			break;
1966 		default:
1967 			rdev->saved_raid_disk = role;
1968 			if ((le32_to_cpu(sb->feature_map) &
1969 			     MD_FEATURE_RECOVERY_OFFSET)) {
1970 				rdev->recovery_offset = le64_to_cpu(sb->recovery_offset);
1971 				if (!(le32_to_cpu(sb->feature_map) &
1972 				      MD_FEATURE_RECOVERY_BITMAP))
1973 					rdev->saved_raid_disk = -1;
1974 			} else {
1975 				/*
1976 				 * If the array is FROZEN, then the device can't
1977 				 * be in_sync with rest of array.
1978 				 */
1979 				if (!test_bit(MD_RECOVERY_FROZEN,
1980 					      &mddev->recovery))
1981 					set_bit(In_sync, &rdev->flags);
1982 			}
1983 			rdev->raid_disk = role;
1984 			break;
1985 		}
1986 		if (sb->devflags & WriteMostly1)
1987 			set_bit(WriteMostly, &rdev->flags);
1988 		if (sb->devflags & FailFast1)
1989 			set_bit(FailFast, &rdev->flags);
1990 		if (le32_to_cpu(sb->feature_map) & MD_FEATURE_REPLACEMENT)
1991 			set_bit(Replacement, &rdev->flags);
1992 	} else /* MULTIPATH are always insync */
1993 		set_bit(In_sync, &rdev->flags);
1994 
1995 	return 0;
1996 }
1997 
super_1_sync(struct mddev * mddev,struct md_rdev * rdev)1998 static void super_1_sync(struct mddev *mddev, struct md_rdev *rdev)
1999 {
2000 	struct mdp_superblock_1 *sb;
2001 	struct md_rdev *rdev2;
2002 	int max_dev, i;
2003 	/* make rdev->sb match mddev and rdev data. */
2004 
2005 	sb = page_address(rdev->sb_page);
2006 
2007 	sb->feature_map = 0;
2008 	sb->pad0 = 0;
2009 	sb->recovery_offset = cpu_to_le64(0);
2010 	memset(sb->pad3, 0, sizeof(sb->pad3));
2011 
2012 	sb->utime = cpu_to_le64((__u64)mddev->utime);
2013 	sb->events = cpu_to_le64(mddev->events);
2014 	if (mddev->in_sync)
2015 		sb->resync_offset = cpu_to_le64(mddev->recovery_cp);
2016 	else if (test_bit(MD_JOURNAL_CLEAN, &mddev->flags))
2017 		sb->resync_offset = cpu_to_le64(MaxSector);
2018 	else
2019 		sb->resync_offset = cpu_to_le64(0);
2020 
2021 	sb->cnt_corrected_read = cpu_to_le32(atomic_read(&rdev->corrected_errors));
2022 
2023 	sb->raid_disks = cpu_to_le32(mddev->raid_disks);
2024 	sb->size = cpu_to_le64(mddev->dev_sectors);
2025 	sb->chunksize = cpu_to_le32(mddev->chunk_sectors);
2026 	sb->level = cpu_to_le32(mddev->level);
2027 	sb->layout = cpu_to_le32(mddev->layout);
2028 	if (test_bit(FailFast, &rdev->flags))
2029 		sb->devflags |= FailFast1;
2030 	else
2031 		sb->devflags &= ~FailFast1;
2032 
2033 	if (test_bit(WriteMostly, &rdev->flags))
2034 		sb->devflags |= WriteMostly1;
2035 	else
2036 		sb->devflags &= ~WriteMostly1;
2037 	sb->data_offset = cpu_to_le64(rdev->data_offset);
2038 	sb->data_size = cpu_to_le64(rdev->sectors);
2039 
2040 	if (mddev->bitmap && mddev->bitmap_info.file == NULL) {
2041 		sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_info.offset);
2042 		sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET);
2043 	}
2044 
2045 	if (rdev->raid_disk >= 0 && !test_bit(Journal, &rdev->flags) &&
2046 	    !test_bit(In_sync, &rdev->flags)) {
2047 		sb->feature_map |=
2048 			cpu_to_le32(MD_FEATURE_RECOVERY_OFFSET);
2049 		sb->recovery_offset =
2050 			cpu_to_le64(rdev->recovery_offset);
2051 		if (rdev->saved_raid_disk >= 0 && mddev->bitmap)
2052 			sb->feature_map |=
2053 				cpu_to_le32(MD_FEATURE_RECOVERY_BITMAP);
2054 	}
2055 	/* Note: recovery_offset and journal_tail share space  */
2056 	if (test_bit(Journal, &rdev->flags))
2057 		sb->journal_tail = cpu_to_le64(rdev->journal_tail);
2058 	if (test_bit(Replacement, &rdev->flags))
2059 		sb->feature_map |=
2060 			cpu_to_le32(MD_FEATURE_REPLACEMENT);
2061 
2062 	if (mddev->reshape_position != MaxSector) {
2063 		sb->feature_map |= cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE);
2064 		sb->reshape_position = cpu_to_le64(mddev->reshape_position);
2065 		sb->new_layout = cpu_to_le32(mddev->new_layout);
2066 		sb->delta_disks = cpu_to_le32(mddev->delta_disks);
2067 		sb->new_level = cpu_to_le32(mddev->new_level);
2068 		sb->new_chunk = cpu_to_le32(mddev->new_chunk_sectors);
2069 		if (mddev->delta_disks == 0 &&
2070 		    mddev->reshape_backwards)
2071 			sb->feature_map
2072 				|= cpu_to_le32(MD_FEATURE_RESHAPE_BACKWARDS);
2073 		if (rdev->new_data_offset != rdev->data_offset) {
2074 			sb->feature_map
2075 				|= cpu_to_le32(MD_FEATURE_NEW_OFFSET);
2076 			sb->new_offset = cpu_to_le32((__u32)(rdev->new_data_offset
2077 							     - rdev->data_offset));
2078 		}
2079 	}
2080 
2081 	if (mddev_is_clustered(mddev))
2082 		sb->feature_map |= cpu_to_le32(MD_FEATURE_CLUSTERED);
2083 
2084 	if (rdev->badblocks.count == 0)
2085 		/* Nothing to do for bad blocks*/ ;
2086 	else if (sb->bblog_offset == 0)
2087 		/* Cannot record bad blocks on this device */
2088 		md_error(mddev, rdev);
2089 	else {
2090 		struct badblocks *bb = &rdev->badblocks;
2091 		__le64 *bbp = (__le64 *)page_address(rdev->bb_page);
2092 		u64 *p = bb->page;
2093 		sb->feature_map |= cpu_to_le32(MD_FEATURE_BAD_BLOCKS);
2094 		if (bb->changed) {
2095 			unsigned seq;
2096 
2097 retry:
2098 			seq = read_seqbegin(&bb->lock);
2099 
2100 			memset(bbp, 0xff, PAGE_SIZE);
2101 
2102 			for (i = 0 ; i < bb->count ; i++) {
2103 				u64 internal_bb = p[i];
2104 				u64 store_bb = ((BB_OFFSET(internal_bb) << 10)
2105 						| BB_LEN(internal_bb));
2106 				bbp[i] = cpu_to_le64(store_bb);
2107 			}
2108 			bb->changed = 0;
2109 			if (read_seqretry(&bb->lock, seq))
2110 				goto retry;
2111 
2112 			bb->sector = (rdev->sb_start +
2113 				      (int)le32_to_cpu(sb->bblog_offset));
2114 			bb->size = le16_to_cpu(sb->bblog_size);
2115 		}
2116 	}
2117 
2118 	max_dev = 0;
2119 	rdev_for_each(rdev2, mddev)
2120 		if (rdev2->desc_nr+1 > max_dev)
2121 			max_dev = rdev2->desc_nr+1;
2122 
2123 	if (max_dev > le32_to_cpu(sb->max_dev)) {
2124 		int bmask;
2125 		sb->max_dev = cpu_to_le32(max_dev);
2126 		rdev->sb_size = max_dev * 2 + 256;
2127 		bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
2128 		if (rdev->sb_size & bmask)
2129 			rdev->sb_size = (rdev->sb_size | bmask) + 1;
2130 	} else
2131 		max_dev = le32_to_cpu(sb->max_dev);
2132 
2133 	for (i=0; i<max_dev;i++)
2134 		sb->dev_roles[i] = cpu_to_le16(MD_DISK_ROLE_SPARE);
2135 
2136 	if (test_bit(MD_HAS_JOURNAL, &mddev->flags))
2137 		sb->feature_map |= cpu_to_le32(MD_FEATURE_JOURNAL);
2138 
2139 	if (test_bit(MD_HAS_PPL, &mddev->flags)) {
2140 		if (test_bit(MD_HAS_MULTIPLE_PPLS, &mddev->flags))
2141 			sb->feature_map |=
2142 			    cpu_to_le32(MD_FEATURE_MULTIPLE_PPLS);
2143 		else
2144 			sb->feature_map |= cpu_to_le32(MD_FEATURE_PPL);
2145 		sb->ppl.offset = cpu_to_le16(rdev->ppl.offset);
2146 		sb->ppl.size = cpu_to_le16(rdev->ppl.size);
2147 	}
2148 
2149 	rdev_for_each(rdev2, mddev) {
2150 		i = rdev2->desc_nr;
2151 		if (test_bit(Faulty, &rdev2->flags))
2152 			sb->dev_roles[i] = cpu_to_le16(MD_DISK_ROLE_FAULTY);
2153 		else if (test_bit(In_sync, &rdev2->flags))
2154 			sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
2155 		else if (test_bit(Journal, &rdev2->flags))
2156 			sb->dev_roles[i] = cpu_to_le16(MD_DISK_ROLE_JOURNAL);
2157 		else if (rdev2->raid_disk >= 0)
2158 			sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
2159 		else
2160 			sb->dev_roles[i] = cpu_to_le16(MD_DISK_ROLE_SPARE);
2161 	}
2162 
2163 	sb->sb_csum = calc_sb_1_csum(sb);
2164 }
2165 
super_1_choose_bm_space(sector_t dev_size)2166 static sector_t super_1_choose_bm_space(sector_t dev_size)
2167 {
2168 	sector_t bm_space;
2169 
2170 	/* if the device is bigger than 8Gig, save 64k for bitmap
2171 	 * usage, if bigger than 200Gig, save 128k
2172 	 */
2173 	if (dev_size < 64*2)
2174 		bm_space = 0;
2175 	else if (dev_size - 64*2 >= 200*1024*1024*2)
2176 		bm_space = 128*2;
2177 	else if (dev_size - 4*2 > 8*1024*1024*2)
2178 		bm_space = 64*2;
2179 	else
2180 		bm_space = 4*2;
2181 	return bm_space;
2182 }
2183 
2184 static unsigned long long
super_1_rdev_size_change(struct md_rdev * rdev,sector_t num_sectors)2185 super_1_rdev_size_change(struct md_rdev *rdev, sector_t num_sectors)
2186 {
2187 	struct mdp_superblock_1 *sb;
2188 	sector_t max_sectors;
2189 	if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
2190 		return 0; /* component must fit device */
2191 	if (rdev->data_offset != rdev->new_data_offset)
2192 		return 0; /* too confusing */
2193 	if (rdev->sb_start < rdev->data_offset) {
2194 		/* minor versions 1 and 2; superblock before data */
2195 		max_sectors = bdev_nr_sectors(rdev->bdev) - rdev->data_offset;
2196 		if (!num_sectors || num_sectors > max_sectors)
2197 			num_sectors = max_sectors;
2198 	} else if (rdev->mddev->bitmap_info.offset) {
2199 		/* minor version 0 with bitmap we can't move */
2200 		return 0;
2201 	} else {
2202 		/* minor version 0; superblock after data */
2203 		sector_t sb_start, bm_space;
2204 		sector_t dev_size = bdev_nr_sectors(rdev->bdev);
2205 
2206 		/* 8K is for superblock */
2207 		sb_start = dev_size - 8*2;
2208 		sb_start &= ~(sector_t)(4*2 - 1);
2209 
2210 		bm_space = super_1_choose_bm_space(dev_size);
2211 
2212 		/* Space that can be used to store date needs to decrease
2213 		 * superblock bitmap space and bad block space(4K)
2214 		 */
2215 		max_sectors = sb_start - bm_space - 4*2;
2216 
2217 		if (!num_sectors || num_sectors > max_sectors)
2218 			num_sectors = max_sectors;
2219 		rdev->sb_start = sb_start;
2220 	}
2221 	sb = page_address(rdev->sb_page);
2222 	sb->data_size = cpu_to_le64(num_sectors);
2223 	sb->super_offset = cpu_to_le64(rdev->sb_start);
2224 	sb->sb_csum = calc_sb_1_csum(sb);
2225 	do {
2226 		md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
2227 			       rdev->sb_page);
2228 	} while (md_super_wait(rdev->mddev) < 0);
2229 	return num_sectors;
2230 
2231 }
2232 
2233 static int
super_1_allow_new_offset(struct md_rdev * rdev,unsigned long long new_offset)2234 super_1_allow_new_offset(struct md_rdev *rdev,
2235 			 unsigned long long new_offset)
2236 {
2237 	/* All necessary checks on new >= old have been done */
2238 	struct bitmap *bitmap;
2239 	if (new_offset >= rdev->data_offset)
2240 		return 1;
2241 
2242 	/* with 1.0 metadata, there is no metadata to tread on
2243 	 * so we can always move back */
2244 	if (rdev->mddev->minor_version == 0)
2245 		return 1;
2246 
2247 	/* otherwise we must be sure not to step on
2248 	 * any metadata, so stay:
2249 	 * 36K beyond start of superblock
2250 	 * beyond end of badblocks
2251 	 * beyond write-intent bitmap
2252 	 */
2253 	if (rdev->sb_start + (32+4)*2 > new_offset)
2254 		return 0;
2255 	bitmap = rdev->mddev->bitmap;
2256 	if (bitmap && !rdev->mddev->bitmap_info.file &&
2257 	    rdev->sb_start + rdev->mddev->bitmap_info.offset +
2258 	    bitmap->storage.file_pages * (PAGE_SIZE>>9) > new_offset)
2259 		return 0;
2260 	if (rdev->badblocks.sector + rdev->badblocks.size > new_offset)
2261 		return 0;
2262 
2263 	return 1;
2264 }
2265 
2266 static struct super_type super_types[] = {
2267 	[0] = {
2268 		.name	= "0.90.0",
2269 		.owner	= THIS_MODULE,
2270 		.load_super	    = super_90_load,
2271 		.validate_super	    = super_90_validate,
2272 		.sync_super	    = super_90_sync,
2273 		.rdev_size_change   = super_90_rdev_size_change,
2274 		.allow_new_offset   = super_90_allow_new_offset,
2275 	},
2276 	[1] = {
2277 		.name	= "md-1",
2278 		.owner	= THIS_MODULE,
2279 		.load_super	    = super_1_load,
2280 		.validate_super	    = super_1_validate,
2281 		.sync_super	    = super_1_sync,
2282 		.rdev_size_change   = super_1_rdev_size_change,
2283 		.allow_new_offset   = super_1_allow_new_offset,
2284 	},
2285 };
2286 
sync_super(struct mddev * mddev,struct md_rdev * rdev)2287 static void sync_super(struct mddev *mddev, struct md_rdev *rdev)
2288 {
2289 	if (mddev->sync_super) {
2290 		mddev->sync_super(mddev, rdev);
2291 		return;
2292 	}
2293 
2294 	BUG_ON(mddev->major_version >= ARRAY_SIZE(super_types));
2295 
2296 	super_types[mddev->major_version].sync_super(mddev, rdev);
2297 }
2298 
match_mddev_units(struct mddev * mddev1,struct mddev * mddev2)2299 static int match_mddev_units(struct mddev *mddev1, struct mddev *mddev2)
2300 {
2301 	struct md_rdev *rdev, *rdev2;
2302 
2303 	rcu_read_lock();
2304 	rdev_for_each_rcu(rdev, mddev1) {
2305 		if (test_bit(Faulty, &rdev->flags) ||
2306 		    test_bit(Journal, &rdev->flags) ||
2307 		    rdev->raid_disk == -1)
2308 			continue;
2309 		rdev_for_each_rcu(rdev2, mddev2) {
2310 			if (test_bit(Faulty, &rdev2->flags) ||
2311 			    test_bit(Journal, &rdev2->flags) ||
2312 			    rdev2->raid_disk == -1)
2313 				continue;
2314 			if (rdev->bdev->bd_disk == rdev2->bdev->bd_disk) {
2315 				rcu_read_unlock();
2316 				return 1;
2317 			}
2318 		}
2319 	}
2320 	rcu_read_unlock();
2321 	return 0;
2322 }
2323 
2324 static LIST_HEAD(pending_raid_disks);
2325 
2326 /*
2327  * Try to register data integrity profile for an mddev
2328  *
2329  * This is called when an array is started and after a disk has been kicked
2330  * from the array. It only succeeds if all working and active component devices
2331  * are integrity capable with matching profiles.
2332  */
md_integrity_register(struct mddev * mddev)2333 int md_integrity_register(struct mddev *mddev)
2334 {
2335 	struct md_rdev *rdev, *reference = NULL;
2336 
2337 	if (list_empty(&mddev->disks))
2338 		return 0; /* nothing to do */
2339 	if (!mddev->gendisk || blk_get_integrity(mddev->gendisk))
2340 		return 0; /* shouldn't register, or already is */
2341 	rdev_for_each(rdev, mddev) {
2342 		/* skip spares and non-functional disks */
2343 		if (test_bit(Faulty, &rdev->flags))
2344 			continue;
2345 		if (rdev->raid_disk < 0)
2346 			continue;
2347 		if (!reference) {
2348 			/* Use the first rdev as the reference */
2349 			reference = rdev;
2350 			continue;
2351 		}
2352 		/* does this rdev's profile match the reference profile? */
2353 		if (blk_integrity_compare(reference->bdev->bd_disk,
2354 				rdev->bdev->bd_disk) < 0)
2355 			return -EINVAL;
2356 	}
2357 	if (!reference || !bdev_get_integrity(reference->bdev))
2358 		return 0;
2359 	/*
2360 	 * All component devices are integrity capable and have matching
2361 	 * profiles, register the common profile for the md device.
2362 	 */
2363 	blk_integrity_register(mddev->gendisk,
2364 			       bdev_get_integrity(reference->bdev));
2365 
2366 	pr_debug("md: data integrity enabled on %s\n", mdname(mddev));
2367 	if (bioset_integrity_create(&mddev->bio_set, BIO_POOL_SIZE) ||
2368 	    (mddev->level != 1 && mddev->level != 10 &&
2369 	     bioset_integrity_create(&mddev->io_clone_set, BIO_POOL_SIZE))) {
2370 		/*
2371 		 * No need to handle the failure of bioset_integrity_create,
2372 		 * because the function is called by md_run() -> pers->run(),
2373 		 * md_run calls bioset_exit -> bioset_integrity_free in case
2374 		 * of failure case.
2375 		 */
2376 		pr_err("md: failed to create integrity pool for %s\n",
2377 		       mdname(mddev));
2378 		return -EINVAL;
2379 	}
2380 	return 0;
2381 }
2382 EXPORT_SYMBOL(md_integrity_register);
2383 
2384 /*
2385  * Attempt to add an rdev, but only if it is consistent with the current
2386  * integrity profile
2387  */
md_integrity_add_rdev(struct md_rdev * rdev,struct mddev * mddev)2388 int md_integrity_add_rdev(struct md_rdev *rdev, struct mddev *mddev)
2389 {
2390 	struct blk_integrity *bi_mddev;
2391 
2392 	if (!mddev->gendisk)
2393 		return 0;
2394 
2395 	bi_mddev = blk_get_integrity(mddev->gendisk);
2396 
2397 	if (!bi_mddev) /* nothing to do */
2398 		return 0;
2399 
2400 	if (blk_integrity_compare(mddev->gendisk, rdev->bdev->bd_disk) != 0) {
2401 		pr_err("%s: incompatible integrity profile for %pg\n",
2402 		       mdname(mddev), rdev->bdev);
2403 		return -ENXIO;
2404 	}
2405 
2406 	return 0;
2407 }
2408 EXPORT_SYMBOL(md_integrity_add_rdev);
2409 
rdev_read_only(struct md_rdev * rdev)2410 static bool rdev_read_only(struct md_rdev *rdev)
2411 {
2412 	return bdev_read_only(rdev->bdev) ||
2413 		(rdev->meta_bdev && bdev_read_only(rdev->meta_bdev));
2414 }
2415 
bind_rdev_to_array(struct md_rdev * rdev,struct mddev * mddev)2416 static int bind_rdev_to_array(struct md_rdev *rdev, struct mddev *mddev)
2417 {
2418 	char b[BDEVNAME_SIZE];
2419 	int err;
2420 
2421 	/* prevent duplicates */
2422 	if (find_rdev(mddev, rdev->bdev->bd_dev))
2423 		return -EEXIST;
2424 
2425 	if (rdev_read_only(rdev) && mddev->pers)
2426 		return -EROFS;
2427 
2428 	/* make sure rdev->sectors exceeds mddev->dev_sectors */
2429 	if (!test_bit(Journal, &rdev->flags) &&
2430 	    rdev->sectors &&
2431 	    (mddev->dev_sectors == 0 || rdev->sectors < mddev->dev_sectors)) {
2432 		if (mddev->pers) {
2433 			/* Cannot change size, so fail
2434 			 * If mddev->level <= 0, then we don't care
2435 			 * about aligning sizes (e.g. linear)
2436 			 */
2437 			if (mddev->level > 0)
2438 				return -ENOSPC;
2439 		} else
2440 			mddev->dev_sectors = rdev->sectors;
2441 	}
2442 
2443 	/* Verify rdev->desc_nr is unique.
2444 	 * If it is -1, assign a free number, else
2445 	 * check number is not in use
2446 	 */
2447 	rcu_read_lock();
2448 	if (rdev->desc_nr < 0) {
2449 		int choice = 0;
2450 		if (mddev->pers)
2451 			choice = mddev->raid_disks;
2452 		while (md_find_rdev_nr_rcu(mddev, choice))
2453 			choice++;
2454 		rdev->desc_nr = choice;
2455 	} else {
2456 		if (md_find_rdev_nr_rcu(mddev, rdev->desc_nr)) {
2457 			rcu_read_unlock();
2458 			return -EBUSY;
2459 		}
2460 	}
2461 	rcu_read_unlock();
2462 	if (!test_bit(Journal, &rdev->flags) &&
2463 	    mddev->max_disks && rdev->desc_nr >= mddev->max_disks) {
2464 		pr_warn("md: %s: array is limited to %d devices\n",
2465 			mdname(mddev), mddev->max_disks);
2466 		return -EBUSY;
2467 	}
2468 	snprintf(b, sizeof(b), "%pg", rdev->bdev);
2469 	strreplace(b, '/', '!');
2470 
2471 	rdev->mddev = mddev;
2472 	pr_debug("md: bind<%s>\n", b);
2473 
2474 	if (mddev->raid_disks)
2475 		mddev_create_serial_pool(mddev, rdev, false);
2476 
2477 	if ((err = kobject_add(&rdev->kobj, &mddev->kobj, "dev-%s", b)))
2478 		goto fail;
2479 
2480 	/* failure here is OK */
2481 	err = sysfs_create_link(&rdev->kobj, bdev_kobj(rdev->bdev), "block");
2482 	rdev->sysfs_state = sysfs_get_dirent_safe(rdev->kobj.sd, "state");
2483 	rdev->sysfs_unack_badblocks =
2484 		sysfs_get_dirent_safe(rdev->kobj.sd, "unacknowledged_bad_blocks");
2485 	rdev->sysfs_badblocks =
2486 		sysfs_get_dirent_safe(rdev->kobj.sd, "bad_blocks");
2487 
2488 	list_add_rcu(&rdev->same_set, &mddev->disks);
2489 	bd_link_disk_holder(rdev->bdev, mddev->gendisk);
2490 
2491 	/* May as well allow recovery to be retried once */
2492 	mddev->recovery_disabled++;
2493 
2494 	return 0;
2495 
2496  fail:
2497 	pr_warn("md: failed to register dev-%s for %s\n",
2498 		b, mdname(mddev));
2499 	mddev_destroy_serial_pool(mddev, rdev, false);
2500 	return err;
2501 }
2502 
2503 void md_autodetect_dev(dev_t dev);
2504 
2505 /* just for claiming the bdev */
2506 static struct md_rdev claim_rdev;
2507 
export_rdev(struct md_rdev * rdev,struct mddev * mddev)2508 static void export_rdev(struct md_rdev *rdev, struct mddev *mddev)
2509 {
2510 	pr_debug("md: export_rdev(%pg)\n", rdev->bdev);
2511 	md_rdev_clear(rdev);
2512 #ifndef MODULE
2513 	if (test_bit(AutoDetected, &rdev->flags))
2514 		md_autodetect_dev(rdev->bdev->bd_dev);
2515 #endif
2516 	blkdev_put(rdev->bdev,
2517 		   test_bit(Holder, &rdev->flags) ? rdev : &claim_rdev);
2518 	rdev->bdev = NULL;
2519 	kobject_put(&rdev->kobj);
2520 }
2521 
md_kick_rdev_from_array(struct md_rdev * rdev)2522 static void md_kick_rdev_from_array(struct md_rdev *rdev)
2523 {
2524 	struct mddev *mddev = rdev->mddev;
2525 
2526 	bd_unlink_disk_holder(rdev->bdev, rdev->mddev->gendisk);
2527 	list_del_rcu(&rdev->same_set);
2528 	pr_debug("md: unbind<%pg>\n", rdev->bdev);
2529 	mddev_destroy_serial_pool(rdev->mddev, rdev, false);
2530 	rdev->mddev = NULL;
2531 	sysfs_remove_link(&rdev->kobj, "block");
2532 	sysfs_put(rdev->sysfs_state);
2533 	sysfs_put(rdev->sysfs_unack_badblocks);
2534 	sysfs_put(rdev->sysfs_badblocks);
2535 	rdev->sysfs_state = NULL;
2536 	rdev->sysfs_unack_badblocks = NULL;
2537 	rdev->sysfs_badblocks = NULL;
2538 	rdev->badblocks.count = 0;
2539 
2540 	synchronize_rcu();
2541 
2542 	/*
2543 	 * kobject_del() will wait for all in progress writers to be done, where
2544 	 * reconfig_mutex is held, hence it can't be called under
2545 	 * reconfig_mutex and it's delayed to mddev_unlock().
2546 	 */
2547 	list_add(&rdev->same_set, &mddev->deleting);
2548 }
2549 
export_array(struct mddev * mddev)2550 static void export_array(struct mddev *mddev)
2551 {
2552 	struct md_rdev *rdev;
2553 
2554 	while (!list_empty(&mddev->disks)) {
2555 		rdev = list_first_entry(&mddev->disks, struct md_rdev,
2556 					same_set);
2557 		md_kick_rdev_from_array(rdev);
2558 	}
2559 	mddev->raid_disks = 0;
2560 	mddev->major_version = 0;
2561 }
2562 
set_in_sync(struct mddev * mddev)2563 static bool set_in_sync(struct mddev *mddev)
2564 {
2565 	lockdep_assert_held(&mddev->lock);
2566 	if (!mddev->in_sync) {
2567 		mddev->sync_checkers++;
2568 		spin_unlock(&mddev->lock);
2569 		percpu_ref_switch_to_atomic_sync(&mddev->writes_pending);
2570 		spin_lock(&mddev->lock);
2571 		if (!mddev->in_sync &&
2572 		    percpu_ref_is_zero(&mddev->writes_pending)) {
2573 			mddev->in_sync = 1;
2574 			/*
2575 			 * Ensure ->in_sync is visible before we clear
2576 			 * ->sync_checkers.
2577 			 */
2578 			smp_mb();
2579 			set_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags);
2580 			sysfs_notify_dirent_safe(mddev->sysfs_state);
2581 		}
2582 		if (--mddev->sync_checkers == 0)
2583 			percpu_ref_switch_to_percpu(&mddev->writes_pending);
2584 	}
2585 	if (mddev->safemode == 1)
2586 		mddev->safemode = 0;
2587 	return mddev->in_sync;
2588 }
2589 
sync_sbs(struct mddev * mddev,int nospares)2590 static void sync_sbs(struct mddev *mddev, int nospares)
2591 {
2592 	/* Update each superblock (in-memory image), but
2593 	 * if we are allowed to, skip spares which already
2594 	 * have the right event counter, or have one earlier
2595 	 * (which would mean they aren't being marked as dirty
2596 	 * with the rest of the array)
2597 	 */
2598 	struct md_rdev *rdev;
2599 	rdev_for_each(rdev, mddev) {
2600 		if (rdev->sb_events == mddev->events ||
2601 		    (nospares &&
2602 		     rdev->raid_disk < 0 &&
2603 		     rdev->sb_events+1 == mddev->events)) {
2604 			/* Don't update this superblock */
2605 			rdev->sb_loaded = 2;
2606 		} else {
2607 			sync_super(mddev, rdev);
2608 			rdev->sb_loaded = 1;
2609 		}
2610 	}
2611 }
2612 
does_sb_need_changing(struct mddev * mddev)2613 static bool does_sb_need_changing(struct mddev *mddev)
2614 {
2615 	struct md_rdev *rdev = NULL, *iter;
2616 	struct mdp_superblock_1 *sb;
2617 	int role;
2618 
2619 	/* Find a good rdev */
2620 	rdev_for_each(iter, mddev)
2621 		if ((iter->raid_disk >= 0) && !test_bit(Faulty, &iter->flags)) {
2622 			rdev = iter;
2623 			break;
2624 		}
2625 
2626 	/* No good device found. */
2627 	if (!rdev)
2628 		return false;
2629 
2630 	sb = page_address(rdev->sb_page);
2631 	/* Check if a device has become faulty or a spare become active */
2632 	rdev_for_each(rdev, mddev) {
2633 		role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
2634 		/* Device activated? */
2635 		if (role == MD_DISK_ROLE_SPARE && rdev->raid_disk >= 0 &&
2636 		    !test_bit(Faulty, &rdev->flags))
2637 			return true;
2638 		/* Device turned faulty? */
2639 		if (test_bit(Faulty, &rdev->flags) && (role < MD_DISK_ROLE_MAX))
2640 			return true;
2641 	}
2642 
2643 	/* Check if any mddev parameters have changed */
2644 	if ((mddev->dev_sectors != le64_to_cpu(sb->size)) ||
2645 	    (mddev->reshape_position != le64_to_cpu(sb->reshape_position)) ||
2646 	    (mddev->layout != le32_to_cpu(sb->layout)) ||
2647 	    (mddev->raid_disks != le32_to_cpu(sb->raid_disks)) ||
2648 	    (mddev->chunk_sectors != le32_to_cpu(sb->chunksize)))
2649 		return true;
2650 
2651 	return false;
2652 }
2653 
md_update_sb(struct mddev * mddev,int force_change)2654 void md_update_sb(struct mddev *mddev, int force_change)
2655 {
2656 	struct md_rdev *rdev;
2657 	int sync_req;
2658 	int nospares = 0;
2659 	int any_badblocks_changed = 0;
2660 	int ret = -1;
2661 
2662 	if (!md_is_rdwr(mddev)) {
2663 		if (force_change)
2664 			set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
2665 		return;
2666 	}
2667 
2668 repeat:
2669 	if (mddev_is_clustered(mddev)) {
2670 		if (test_and_clear_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags))
2671 			force_change = 1;
2672 		if (test_and_clear_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags))
2673 			nospares = 1;
2674 		ret = md_cluster_ops->metadata_update_start(mddev);
2675 		/* Has someone else has updated the sb */
2676 		if (!does_sb_need_changing(mddev)) {
2677 			if (ret == 0)
2678 				md_cluster_ops->metadata_update_cancel(mddev);
2679 			bit_clear_unless(&mddev->sb_flags, BIT(MD_SB_CHANGE_PENDING),
2680 							 BIT(MD_SB_CHANGE_DEVS) |
2681 							 BIT(MD_SB_CHANGE_CLEAN));
2682 			return;
2683 		}
2684 	}
2685 
2686 	/*
2687 	 * First make sure individual recovery_offsets are correct
2688 	 * curr_resync_completed can only be used during recovery.
2689 	 * During reshape/resync it might use array-addresses rather
2690 	 * that device addresses.
2691 	 */
2692 	rdev_for_each(rdev, mddev) {
2693 		if (rdev->raid_disk >= 0 &&
2694 		    mddev->delta_disks >= 0 &&
2695 		    test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
2696 		    test_bit(MD_RECOVERY_RECOVER, &mddev->recovery) &&
2697 		    !test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
2698 		    !test_bit(Journal, &rdev->flags) &&
2699 		    !test_bit(In_sync, &rdev->flags) &&
2700 		    mddev->curr_resync_completed > rdev->recovery_offset)
2701 				rdev->recovery_offset = mddev->curr_resync_completed;
2702 
2703 	}
2704 	if (!mddev->persistent) {
2705 		clear_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags);
2706 		clear_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
2707 		if (!mddev->external) {
2708 			clear_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags);
2709 			rdev_for_each(rdev, mddev) {
2710 				if (rdev->badblocks.changed) {
2711 					rdev->badblocks.changed = 0;
2712 					ack_all_badblocks(&rdev->badblocks);
2713 					md_error(mddev, rdev);
2714 				}
2715 				clear_bit(Blocked, &rdev->flags);
2716 				clear_bit(BlockedBadBlocks, &rdev->flags);
2717 				wake_up(&rdev->blocked_wait);
2718 			}
2719 		}
2720 		wake_up(&mddev->sb_wait);
2721 		return;
2722 	}
2723 
2724 	spin_lock(&mddev->lock);
2725 
2726 	mddev->utime = ktime_get_real_seconds();
2727 
2728 	if (test_and_clear_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags))
2729 		force_change = 1;
2730 	if (test_and_clear_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags))
2731 		/* just a clean<-> dirty transition, possibly leave spares alone,
2732 		 * though if events isn't the right even/odd, we will have to do
2733 		 * spares after all
2734 		 */
2735 		nospares = 1;
2736 	if (force_change)
2737 		nospares = 0;
2738 	if (mddev->degraded)
2739 		/* If the array is degraded, then skipping spares is both
2740 		 * dangerous and fairly pointless.
2741 		 * Dangerous because a device that was removed from the array
2742 		 * might have a event_count that still looks up-to-date,
2743 		 * so it can be re-added without a resync.
2744 		 * Pointless because if there are any spares to skip,
2745 		 * then a recovery will happen and soon that array won't
2746 		 * be degraded any more and the spare can go back to sleep then.
2747 		 */
2748 		nospares = 0;
2749 
2750 	sync_req = mddev->in_sync;
2751 
2752 	/* If this is just a dirty<->clean transition, and the array is clean
2753 	 * and 'events' is odd, we can roll back to the previous clean state */
2754 	if (nospares
2755 	    && (mddev->in_sync && mddev->recovery_cp == MaxSector)
2756 	    && mddev->can_decrease_events
2757 	    && mddev->events != 1) {
2758 		mddev->events--;
2759 		mddev->can_decrease_events = 0;
2760 	} else {
2761 		/* otherwise we have to go forward and ... */
2762 		mddev->events ++;
2763 		mddev->can_decrease_events = nospares;
2764 	}
2765 
2766 	/*
2767 	 * This 64-bit counter should never wrap.
2768 	 * Either we are in around ~1 trillion A.C., assuming
2769 	 * 1 reboot per second, or we have a bug...
2770 	 */
2771 	WARN_ON(mddev->events == 0);
2772 
2773 	rdev_for_each(rdev, mddev) {
2774 		if (rdev->badblocks.changed)
2775 			any_badblocks_changed++;
2776 		if (test_bit(Faulty, &rdev->flags))
2777 			set_bit(FaultRecorded, &rdev->flags);
2778 	}
2779 
2780 	sync_sbs(mddev, nospares);
2781 	spin_unlock(&mddev->lock);
2782 
2783 	pr_debug("md: updating %s RAID superblock on device (in sync %d)\n",
2784 		 mdname(mddev), mddev->in_sync);
2785 
2786 	if (mddev->queue)
2787 		blk_add_trace_msg(mddev->queue, "md md_update_sb");
2788 rewrite:
2789 	md_bitmap_update_sb(mddev->bitmap);
2790 	rdev_for_each(rdev, mddev) {
2791 		if (rdev->sb_loaded != 1)
2792 			continue; /* no noise on spare devices */
2793 
2794 		if (!test_bit(Faulty, &rdev->flags)) {
2795 			md_super_write(mddev,rdev,
2796 				       rdev->sb_start, rdev->sb_size,
2797 				       rdev->sb_page);
2798 			pr_debug("md: (write) %pg's sb offset: %llu\n",
2799 				 rdev->bdev,
2800 				 (unsigned long long)rdev->sb_start);
2801 			rdev->sb_events = mddev->events;
2802 			if (rdev->badblocks.size) {
2803 				md_super_write(mddev, rdev,
2804 					       rdev->badblocks.sector,
2805 					       rdev->badblocks.size << 9,
2806 					       rdev->bb_page);
2807 				rdev->badblocks.size = 0;
2808 			}
2809 
2810 		} else
2811 			pr_debug("md: %pg (skipping faulty)\n",
2812 				 rdev->bdev);
2813 
2814 		if (mddev->level == LEVEL_MULTIPATH)
2815 			/* only need to write one superblock... */
2816 			break;
2817 	}
2818 	if (md_super_wait(mddev) < 0)
2819 		goto rewrite;
2820 	/* if there was a failure, MD_SB_CHANGE_DEVS was set, and we re-write super */
2821 
2822 	if (mddev_is_clustered(mddev) && ret == 0)
2823 		md_cluster_ops->metadata_update_finish(mddev);
2824 
2825 	if (mddev->in_sync != sync_req ||
2826 	    !bit_clear_unless(&mddev->sb_flags, BIT(MD_SB_CHANGE_PENDING),
2827 			       BIT(MD_SB_CHANGE_DEVS) | BIT(MD_SB_CHANGE_CLEAN)))
2828 		/* have to write it out again */
2829 		goto repeat;
2830 	wake_up(&mddev->sb_wait);
2831 	if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
2832 		sysfs_notify_dirent_safe(mddev->sysfs_completed);
2833 
2834 	rdev_for_each(rdev, mddev) {
2835 		if (test_and_clear_bit(FaultRecorded, &rdev->flags))
2836 			clear_bit(Blocked, &rdev->flags);
2837 
2838 		if (any_badblocks_changed)
2839 			ack_all_badblocks(&rdev->badblocks);
2840 		clear_bit(BlockedBadBlocks, &rdev->flags);
2841 		wake_up(&rdev->blocked_wait);
2842 	}
2843 }
2844 EXPORT_SYMBOL(md_update_sb);
2845 
add_bound_rdev(struct md_rdev * rdev)2846 static int add_bound_rdev(struct md_rdev *rdev)
2847 {
2848 	struct mddev *mddev = rdev->mddev;
2849 	int err = 0;
2850 	bool add_journal = test_bit(Journal, &rdev->flags);
2851 
2852 	if (!mddev->pers->hot_remove_disk || add_journal) {
2853 		/* If there is hot_add_disk but no hot_remove_disk
2854 		 * then added disks for geometry changes,
2855 		 * and should be added immediately.
2856 		 */
2857 		super_types[mddev->major_version].
2858 			validate_super(mddev, NULL/*freshest*/, rdev);
2859 		if (add_journal)
2860 			mddev_suspend(mddev);
2861 		err = mddev->pers->hot_add_disk(mddev, rdev);
2862 		if (add_journal)
2863 			mddev_resume(mddev);
2864 		if (err) {
2865 			md_kick_rdev_from_array(rdev);
2866 			return err;
2867 		}
2868 	}
2869 	sysfs_notify_dirent_safe(rdev->sysfs_state);
2870 
2871 	set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
2872 	if (mddev->degraded)
2873 		set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
2874 	set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2875 	md_new_event();
2876 	md_wakeup_thread(mddev->thread);
2877 	return 0;
2878 }
2879 
2880 /* words written to sysfs files may, or may not, be \n terminated.
2881  * We want to accept with case. For this we use cmd_match.
2882  */
cmd_match(const char * cmd,const char * str)2883 static int cmd_match(const char *cmd, const char *str)
2884 {
2885 	/* See if cmd, written into a sysfs file, matches
2886 	 * str.  They must either be the same, or cmd can
2887 	 * have a trailing newline
2888 	 */
2889 	while (*cmd && *str && *cmd == *str) {
2890 		cmd++;
2891 		str++;
2892 	}
2893 	if (*cmd == '\n')
2894 		cmd++;
2895 	if (*str || *cmd)
2896 		return 0;
2897 	return 1;
2898 }
2899 
2900 struct rdev_sysfs_entry {
2901 	struct attribute attr;
2902 	ssize_t (*show)(struct md_rdev *, char *);
2903 	ssize_t (*store)(struct md_rdev *, const char *, size_t);
2904 };
2905 
2906 static ssize_t
state_show(struct md_rdev * rdev,char * page)2907 state_show(struct md_rdev *rdev, char *page)
2908 {
2909 	char *sep = ",";
2910 	size_t len = 0;
2911 	unsigned long flags = READ_ONCE(rdev->flags);
2912 
2913 	if (test_bit(Faulty, &flags) ||
2914 	    (!test_bit(ExternalBbl, &flags) &&
2915 	    rdev->badblocks.unacked_exist))
2916 		len += sprintf(page+len, "faulty%s", sep);
2917 	if (test_bit(In_sync, &flags))
2918 		len += sprintf(page+len, "in_sync%s", sep);
2919 	if (test_bit(Journal, &flags))
2920 		len += sprintf(page+len, "journal%s", sep);
2921 	if (test_bit(WriteMostly, &flags))
2922 		len += sprintf(page+len, "write_mostly%s", sep);
2923 	if (test_bit(Blocked, &flags) ||
2924 	    (rdev->badblocks.unacked_exist
2925 	     && !test_bit(Faulty, &flags)))
2926 		len += sprintf(page+len, "blocked%s", sep);
2927 	if (!test_bit(Faulty, &flags) &&
2928 	    !test_bit(Journal, &flags) &&
2929 	    !test_bit(In_sync, &flags))
2930 		len += sprintf(page+len, "spare%s", sep);
2931 	if (test_bit(WriteErrorSeen, &flags))
2932 		len += sprintf(page+len, "write_error%s", sep);
2933 	if (test_bit(WantReplacement, &flags))
2934 		len += sprintf(page+len, "want_replacement%s", sep);
2935 	if (test_bit(Replacement, &flags))
2936 		len += sprintf(page+len, "replacement%s", sep);
2937 	if (test_bit(ExternalBbl, &flags))
2938 		len += sprintf(page+len, "external_bbl%s", sep);
2939 	if (test_bit(FailFast, &flags))
2940 		len += sprintf(page+len, "failfast%s", sep);
2941 
2942 	if (len)
2943 		len -= strlen(sep);
2944 
2945 	return len+sprintf(page+len, "\n");
2946 }
2947 
2948 static ssize_t
state_store(struct md_rdev * rdev,const char * buf,size_t len)2949 state_store(struct md_rdev *rdev, const char *buf, size_t len)
2950 {
2951 	/* can write
2952 	 *  faulty  - simulates an error
2953 	 *  remove  - disconnects the device
2954 	 *  writemostly - sets write_mostly
2955 	 *  -writemostly - clears write_mostly
2956 	 *  blocked - sets the Blocked flags
2957 	 *  -blocked - clears the Blocked and possibly simulates an error
2958 	 *  insync - sets Insync providing device isn't active
2959 	 *  -insync - clear Insync for a device with a slot assigned,
2960 	 *            so that it gets rebuilt based on bitmap
2961 	 *  write_error - sets WriteErrorSeen
2962 	 *  -write_error - clears WriteErrorSeen
2963 	 *  {,-}failfast - set/clear FailFast
2964 	 */
2965 
2966 	struct mddev *mddev = rdev->mddev;
2967 	int err = -EINVAL;
2968 	bool need_update_sb = false;
2969 
2970 	if (cmd_match(buf, "faulty") && rdev->mddev->pers) {
2971 		md_error(rdev->mddev, rdev);
2972 
2973 		if (test_bit(MD_BROKEN, &rdev->mddev->flags))
2974 			err = -EBUSY;
2975 		else
2976 			err = 0;
2977 	} else if (cmd_match(buf, "remove")) {
2978 		if (rdev->mddev->pers) {
2979 			clear_bit(Blocked, &rdev->flags);
2980 			remove_and_add_spares(rdev->mddev, rdev);
2981 		}
2982 		if (rdev->raid_disk >= 0)
2983 			err = -EBUSY;
2984 		else {
2985 			err = 0;
2986 			if (mddev_is_clustered(mddev))
2987 				err = md_cluster_ops->remove_disk(mddev, rdev);
2988 
2989 			if (err == 0) {
2990 				md_kick_rdev_from_array(rdev);
2991 				if (mddev->pers) {
2992 					set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
2993 					md_wakeup_thread(mddev->thread);
2994 				}
2995 				md_new_event();
2996 			}
2997 		}
2998 	} else if (cmd_match(buf, "writemostly")) {
2999 		set_bit(WriteMostly, &rdev->flags);
3000 		mddev_create_serial_pool(rdev->mddev, rdev, false);
3001 		need_update_sb = true;
3002 		err = 0;
3003 	} else if (cmd_match(buf, "-writemostly")) {
3004 		mddev_destroy_serial_pool(rdev->mddev, rdev, false);
3005 		clear_bit(WriteMostly, &rdev->flags);
3006 		need_update_sb = true;
3007 		err = 0;
3008 	} else if (cmd_match(buf, "blocked")) {
3009 		set_bit(Blocked, &rdev->flags);
3010 		err = 0;
3011 	} else if (cmd_match(buf, "-blocked")) {
3012 		if (!test_bit(Faulty, &rdev->flags) &&
3013 		    !test_bit(ExternalBbl, &rdev->flags) &&
3014 		    rdev->badblocks.unacked_exist) {
3015 			/* metadata handler doesn't understand badblocks,
3016 			 * so we need to fail the device
3017 			 */
3018 			md_error(rdev->mddev, rdev);
3019 		}
3020 		clear_bit(Blocked, &rdev->flags);
3021 		clear_bit(BlockedBadBlocks, &rdev->flags);
3022 		wake_up(&rdev->blocked_wait);
3023 		set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
3024 		md_wakeup_thread(rdev->mddev->thread);
3025 
3026 		err = 0;
3027 	} else if (cmd_match(buf, "insync") && rdev->raid_disk == -1) {
3028 		set_bit(In_sync, &rdev->flags);
3029 		err = 0;
3030 	} else if (cmd_match(buf, "failfast")) {
3031 		set_bit(FailFast, &rdev->flags);
3032 		need_update_sb = true;
3033 		err = 0;
3034 	} else if (cmd_match(buf, "-failfast")) {
3035 		clear_bit(FailFast, &rdev->flags);
3036 		need_update_sb = true;
3037 		err = 0;
3038 	} else if (cmd_match(buf, "-insync") && rdev->raid_disk >= 0 &&
3039 		   !test_bit(Journal, &rdev->flags)) {
3040 		if (rdev->mddev->pers == NULL) {
3041 			clear_bit(In_sync, &rdev->flags);
3042 			rdev->saved_raid_disk = rdev->raid_disk;
3043 			rdev->raid_disk = -1;
3044 			err = 0;
3045 		}
3046 	} else if (cmd_match(buf, "write_error")) {
3047 		set_bit(WriteErrorSeen, &rdev->flags);
3048 		err = 0;
3049 	} else if (cmd_match(buf, "-write_error")) {
3050 		clear_bit(WriteErrorSeen, &rdev->flags);
3051 		err = 0;
3052 	} else if (cmd_match(buf, "want_replacement")) {
3053 		/* Any non-spare device that is not a replacement can
3054 		 * become want_replacement at any time, but we then need to
3055 		 * check if recovery is needed.
3056 		 */
3057 		if (rdev->raid_disk >= 0 &&
3058 		    !test_bit(Journal, &rdev->flags) &&
3059 		    !test_bit(Replacement, &rdev->flags))
3060 			set_bit(WantReplacement, &rdev->flags);
3061 		set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
3062 		md_wakeup_thread(rdev->mddev->thread);
3063 		err = 0;
3064 	} else if (cmd_match(buf, "-want_replacement")) {
3065 		/* Clearing 'want_replacement' is always allowed.
3066 		 * Once replacements starts it is too late though.
3067 		 */
3068 		err = 0;
3069 		clear_bit(WantReplacement, &rdev->flags);
3070 	} else if (cmd_match(buf, "replacement")) {
3071 		/* Can only set a device as a replacement when array has not
3072 		 * yet been started.  Once running, replacement is automatic
3073 		 * from spares, or by assigning 'slot'.
3074 		 */
3075 		if (rdev->mddev->pers)
3076 			err = -EBUSY;
3077 		else {
3078 			set_bit(Replacement, &rdev->flags);
3079 			err = 0;
3080 		}
3081 	} else if (cmd_match(buf, "-replacement")) {
3082 		/* Similarly, can only clear Replacement before start */
3083 		if (rdev->mddev->pers)
3084 			err = -EBUSY;
3085 		else {
3086 			clear_bit(Replacement, &rdev->flags);
3087 			err = 0;
3088 		}
3089 	} else if (cmd_match(buf, "re-add")) {
3090 		if (!rdev->mddev->pers)
3091 			err = -EINVAL;
3092 		else if (test_bit(Faulty, &rdev->flags) && (rdev->raid_disk == -1) &&
3093 				rdev->saved_raid_disk >= 0) {
3094 			/* clear_bit is performed _after_ all the devices
3095 			 * have their local Faulty bit cleared. If any writes
3096 			 * happen in the meantime in the local node, they
3097 			 * will land in the local bitmap, which will be synced
3098 			 * by this node eventually
3099 			 */
3100 			if (!mddev_is_clustered(rdev->mddev) ||
3101 			    (err = md_cluster_ops->gather_bitmaps(rdev)) == 0) {
3102 				clear_bit(Faulty, &rdev->flags);
3103 				err = add_bound_rdev(rdev);
3104 			}
3105 		} else
3106 			err = -EBUSY;
3107 	} else if (cmd_match(buf, "external_bbl") && (rdev->mddev->external)) {
3108 		set_bit(ExternalBbl, &rdev->flags);
3109 		rdev->badblocks.shift = 0;
3110 		err = 0;
3111 	} else if (cmd_match(buf, "-external_bbl") && (rdev->mddev->external)) {
3112 		clear_bit(ExternalBbl, &rdev->flags);
3113 		err = 0;
3114 	}
3115 	if (need_update_sb)
3116 		md_update_sb(mddev, 1);
3117 	if (!err)
3118 		sysfs_notify_dirent_safe(rdev->sysfs_state);
3119 	return err ? err : len;
3120 }
3121 static struct rdev_sysfs_entry rdev_state =
3122 __ATTR_PREALLOC(state, S_IRUGO|S_IWUSR, state_show, state_store);
3123 
3124 static ssize_t
errors_show(struct md_rdev * rdev,char * page)3125 errors_show(struct md_rdev *rdev, char *page)
3126 {
3127 	return sprintf(page, "%d\n", atomic_read(&rdev->corrected_errors));
3128 }
3129 
3130 static ssize_t
errors_store(struct md_rdev * rdev,const char * buf,size_t len)3131 errors_store(struct md_rdev *rdev, const char *buf, size_t len)
3132 {
3133 	unsigned int n;
3134 	int rv;
3135 
3136 	rv = kstrtouint(buf, 10, &n);
3137 	if (rv < 0)
3138 		return rv;
3139 	atomic_set(&rdev->corrected_errors, n);
3140 	return len;
3141 }
3142 static struct rdev_sysfs_entry rdev_errors =
3143 __ATTR(errors, S_IRUGO|S_IWUSR, errors_show, errors_store);
3144 
3145 static ssize_t
slot_show(struct md_rdev * rdev,char * page)3146 slot_show(struct md_rdev *rdev, char *page)
3147 {
3148 	if (test_bit(Journal, &rdev->flags))
3149 		return sprintf(page, "journal\n");
3150 	else if (rdev->raid_disk < 0)
3151 		return sprintf(page, "none\n");
3152 	else
3153 		return sprintf(page, "%d\n", rdev->raid_disk);
3154 }
3155 
3156 static ssize_t
slot_store(struct md_rdev * rdev,const char * buf,size_t len)3157 slot_store(struct md_rdev *rdev, const char *buf, size_t len)
3158 {
3159 	int slot;
3160 	int err;
3161 
3162 	if (test_bit(Journal, &rdev->flags))
3163 		return -EBUSY;
3164 	if (strncmp(buf, "none", 4)==0)
3165 		slot = -1;
3166 	else {
3167 		err = kstrtouint(buf, 10, (unsigned int *)&slot);
3168 		if (err < 0)
3169 			return err;
3170 		if (slot < 0)
3171 			/* overflow */
3172 			return -ENOSPC;
3173 	}
3174 	if (rdev->mddev->pers && slot == -1) {
3175 		/* Setting 'slot' on an active array requires also
3176 		 * updating the 'rd%d' link, and communicating
3177 		 * with the personality with ->hot_*_disk.
3178 		 * For now we only support removing
3179 		 * failed/spare devices.  This normally happens automatically,
3180 		 * but not when the metadata is externally managed.
3181 		 */
3182 		if (rdev->raid_disk == -1)
3183 			return -EEXIST;
3184 		/* personality does all needed checks */
3185 		if (rdev->mddev->pers->hot_remove_disk == NULL)
3186 			return -EINVAL;
3187 		clear_bit(Blocked, &rdev->flags);
3188 		remove_and_add_spares(rdev->mddev, rdev);
3189 		if (rdev->raid_disk >= 0)
3190 			return -EBUSY;
3191 		set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
3192 		md_wakeup_thread(rdev->mddev->thread);
3193 	} else if (rdev->mddev->pers) {
3194 		/* Activating a spare .. or possibly reactivating
3195 		 * if we ever get bitmaps working here.
3196 		 */
3197 		int err;
3198 
3199 		if (rdev->raid_disk != -1)
3200 			return -EBUSY;
3201 
3202 		if (test_bit(MD_RECOVERY_RUNNING, &rdev->mddev->recovery))
3203 			return -EBUSY;
3204 
3205 		if (rdev->mddev->pers->hot_add_disk == NULL)
3206 			return -EINVAL;
3207 
3208 		if (slot >= rdev->mddev->raid_disks &&
3209 		    slot >= rdev->mddev->raid_disks + rdev->mddev->delta_disks)
3210 			return -ENOSPC;
3211 
3212 		rdev->raid_disk = slot;
3213 		if (test_bit(In_sync, &rdev->flags))
3214 			rdev->saved_raid_disk = slot;
3215 		else
3216 			rdev->saved_raid_disk = -1;
3217 		clear_bit(In_sync, &rdev->flags);
3218 		clear_bit(Bitmap_sync, &rdev->flags);
3219 		err = rdev->mddev->pers->hot_add_disk(rdev->mddev, rdev);
3220 		if (err) {
3221 			rdev->raid_disk = -1;
3222 			return err;
3223 		} else
3224 			sysfs_notify_dirent_safe(rdev->sysfs_state);
3225 		/* failure here is OK */;
3226 		sysfs_link_rdev(rdev->mddev, rdev);
3227 		/* don't wakeup anyone, leave that to userspace. */
3228 	} else {
3229 		if (slot >= rdev->mddev->raid_disks &&
3230 		    slot >= rdev->mddev->raid_disks + rdev->mddev->delta_disks)
3231 			return -ENOSPC;
3232 		rdev->raid_disk = slot;
3233 		/* assume it is working */
3234 		clear_bit(Faulty, &rdev->flags);
3235 		clear_bit(WriteMostly, &rdev->flags);
3236 		set_bit(In_sync, &rdev->flags);
3237 		sysfs_notify_dirent_safe(rdev->sysfs_state);
3238 	}
3239 	return len;
3240 }
3241 
3242 static struct rdev_sysfs_entry rdev_slot =
3243 __ATTR(slot, S_IRUGO|S_IWUSR, slot_show, slot_store);
3244 
3245 static ssize_t
offset_show(struct md_rdev * rdev,char * page)3246 offset_show(struct md_rdev *rdev, char *page)
3247 {
3248 	return sprintf(page, "%llu\n", (unsigned long long)rdev->data_offset);
3249 }
3250 
3251 static ssize_t
offset_store(struct md_rdev * rdev,const char * buf,size_t len)3252 offset_store(struct md_rdev *rdev, const char *buf, size_t len)
3253 {
3254 	unsigned long long offset;
3255 	if (kstrtoull(buf, 10, &offset) < 0)
3256 		return -EINVAL;
3257 	if (rdev->mddev->pers && rdev->raid_disk >= 0)
3258 		return -EBUSY;
3259 	if (rdev->sectors && rdev->mddev->external)
3260 		/* Must set offset before size, so overlap checks
3261 		 * can be sane */
3262 		return -EBUSY;
3263 	rdev->data_offset = offset;
3264 	rdev->new_data_offset = offset;
3265 	return len;
3266 }
3267 
3268 static struct rdev_sysfs_entry rdev_offset =
3269 __ATTR(offset, S_IRUGO|S_IWUSR, offset_show, offset_store);
3270 
new_offset_show(struct md_rdev * rdev,char * page)3271 static ssize_t new_offset_show(struct md_rdev *rdev, char *page)
3272 {
3273 	return sprintf(page, "%llu\n",
3274 		       (unsigned long long)rdev->new_data_offset);
3275 }
3276 
new_offset_store(struct md_rdev * rdev,const char * buf,size_t len)3277 static ssize_t new_offset_store(struct md_rdev *rdev,
3278 				const char *buf, size_t len)
3279 {
3280 	unsigned long long new_offset;
3281 	struct mddev *mddev = rdev->mddev;
3282 
3283 	if (kstrtoull(buf, 10, &new_offset) < 0)
3284 		return -EINVAL;
3285 
3286 	if (mddev->sync_thread ||
3287 	    test_bit(MD_RECOVERY_RUNNING,&mddev->recovery))
3288 		return -EBUSY;
3289 	if (new_offset == rdev->data_offset)
3290 		/* reset is always permitted */
3291 		;
3292 	else if (new_offset > rdev->data_offset) {
3293 		/* must not push array size beyond rdev_sectors */
3294 		if (new_offset - rdev->data_offset
3295 		    + mddev->dev_sectors > rdev->sectors)
3296 				return -E2BIG;
3297 	}
3298 	/* Metadata worries about other space details. */
3299 
3300 	/* decreasing the offset is inconsistent with a backwards
3301 	 * reshape.
3302 	 */
3303 	if (new_offset < rdev->data_offset &&
3304 	    mddev->reshape_backwards)
3305 		return -EINVAL;
3306 	/* Increasing offset is inconsistent with forwards
3307 	 * reshape.  reshape_direction should be set to
3308 	 * 'backwards' first.
3309 	 */
3310 	if (new_offset > rdev->data_offset &&
3311 	    !mddev->reshape_backwards)
3312 		return -EINVAL;
3313 
3314 	if (mddev->pers && mddev->persistent &&
3315 	    !super_types[mddev->major_version]
3316 	    .allow_new_offset(rdev, new_offset))
3317 		return -E2BIG;
3318 	rdev->new_data_offset = new_offset;
3319 	if (new_offset > rdev->data_offset)
3320 		mddev->reshape_backwards = 1;
3321 	else if (new_offset < rdev->data_offset)
3322 		mddev->reshape_backwards = 0;
3323 
3324 	return len;
3325 }
3326 static struct rdev_sysfs_entry rdev_new_offset =
3327 __ATTR(new_offset, S_IRUGO|S_IWUSR, new_offset_show, new_offset_store);
3328 
3329 static ssize_t
rdev_size_show(struct md_rdev * rdev,char * page)3330 rdev_size_show(struct md_rdev *rdev, char *page)
3331 {
3332 	return sprintf(page, "%llu\n", (unsigned long long)rdev->sectors / 2);
3333 }
3334 
md_rdevs_overlap(struct md_rdev * a,struct md_rdev * b)3335 static int md_rdevs_overlap(struct md_rdev *a, struct md_rdev *b)
3336 {
3337 	/* check if two start/length pairs overlap */
3338 	if (a->data_offset + a->sectors <= b->data_offset)
3339 		return false;
3340 	if (b->data_offset + b->sectors <= a->data_offset)
3341 		return false;
3342 	return true;
3343 }
3344 
md_rdev_overlaps(struct md_rdev * rdev)3345 static bool md_rdev_overlaps(struct md_rdev *rdev)
3346 {
3347 	struct mddev *mddev;
3348 	struct md_rdev *rdev2;
3349 
3350 	spin_lock(&all_mddevs_lock);
3351 	list_for_each_entry(mddev, &all_mddevs, all_mddevs) {
3352 		if (test_bit(MD_DELETED, &mddev->flags))
3353 			continue;
3354 		rdev_for_each(rdev2, mddev) {
3355 			if (rdev != rdev2 && rdev->bdev == rdev2->bdev &&
3356 			    md_rdevs_overlap(rdev, rdev2)) {
3357 				spin_unlock(&all_mddevs_lock);
3358 				return true;
3359 			}
3360 		}
3361 	}
3362 	spin_unlock(&all_mddevs_lock);
3363 	return false;
3364 }
3365 
strict_blocks_to_sectors(const char * buf,sector_t * sectors)3366 static int strict_blocks_to_sectors(const char *buf, sector_t *sectors)
3367 {
3368 	unsigned long long blocks;
3369 	sector_t new;
3370 
3371 	if (kstrtoull(buf, 10, &blocks) < 0)
3372 		return -EINVAL;
3373 
3374 	if (blocks & 1ULL << (8 * sizeof(blocks) - 1))
3375 		return -EINVAL; /* sector conversion overflow */
3376 
3377 	new = blocks * 2;
3378 	if (new != blocks * 2)
3379 		return -EINVAL; /* unsigned long long to sector_t overflow */
3380 
3381 	*sectors = new;
3382 	return 0;
3383 }
3384 
3385 static ssize_t
rdev_size_store(struct md_rdev * rdev,const char * buf,size_t len)3386 rdev_size_store(struct md_rdev *rdev, const char *buf, size_t len)
3387 {
3388 	struct mddev *my_mddev = rdev->mddev;
3389 	sector_t oldsectors = rdev->sectors;
3390 	sector_t sectors;
3391 
3392 	if (test_bit(Journal, &rdev->flags))
3393 		return -EBUSY;
3394 	if (strict_blocks_to_sectors(buf, &sectors) < 0)
3395 		return -EINVAL;
3396 	if (rdev->data_offset != rdev->new_data_offset)
3397 		return -EINVAL; /* too confusing */
3398 	if (my_mddev->pers && rdev->raid_disk >= 0) {
3399 		if (my_mddev->persistent) {
3400 			sectors = super_types[my_mddev->major_version].
3401 				rdev_size_change(rdev, sectors);
3402 			if (!sectors)
3403 				return -EBUSY;
3404 		} else if (!sectors)
3405 			sectors = bdev_nr_sectors(rdev->bdev) -
3406 				rdev->data_offset;
3407 		if (!my_mddev->pers->resize)
3408 			/* Cannot change size for RAID0 or Linear etc */
3409 			return -EINVAL;
3410 	}
3411 	if (sectors < my_mddev->dev_sectors)
3412 		return -EINVAL; /* component must fit device */
3413 
3414 	rdev->sectors = sectors;
3415 
3416 	/*
3417 	 * Check that all other rdevs with the same bdev do not overlap.  This
3418 	 * check does not provide a hard guarantee, it just helps avoid
3419 	 * dangerous mistakes.
3420 	 */
3421 	if (sectors > oldsectors && my_mddev->external &&
3422 	    md_rdev_overlaps(rdev)) {
3423 		/*
3424 		 * Someone else could have slipped in a size change here, but
3425 		 * doing so is just silly.  We put oldsectors back because we
3426 		 * know it is safe, and trust userspace not to race with itself.
3427 		 */
3428 		rdev->sectors = oldsectors;
3429 		return -EBUSY;
3430 	}
3431 	return len;
3432 }
3433 
3434 static struct rdev_sysfs_entry rdev_size =
3435 __ATTR(size, S_IRUGO|S_IWUSR, rdev_size_show, rdev_size_store);
3436 
recovery_start_show(struct md_rdev * rdev,char * page)3437 static ssize_t recovery_start_show(struct md_rdev *rdev, char *page)
3438 {
3439 	unsigned long long recovery_start = rdev->recovery_offset;
3440 
3441 	if (test_bit(In_sync, &rdev->flags) ||
3442 	    recovery_start == MaxSector)
3443 		return sprintf(page, "none\n");
3444 
3445 	return sprintf(page, "%llu\n", recovery_start);
3446 }
3447 
recovery_start_store(struct md_rdev * rdev,const char * buf,size_t len)3448 static ssize_t recovery_start_store(struct md_rdev *rdev, const char *buf, size_t len)
3449 {
3450 	unsigned long long recovery_start;
3451 
3452 	if (cmd_match(buf, "none"))
3453 		recovery_start = MaxSector;
3454 	else if (kstrtoull(buf, 10, &recovery_start))
3455 		return -EINVAL;
3456 
3457 	if (rdev->mddev->pers &&
3458 	    rdev->raid_disk >= 0)
3459 		return -EBUSY;
3460 
3461 	rdev->recovery_offset = recovery_start;
3462 	if (recovery_start == MaxSector)
3463 		set_bit(In_sync, &rdev->flags);
3464 	else
3465 		clear_bit(In_sync, &rdev->flags);
3466 	return len;
3467 }
3468 
3469 static struct rdev_sysfs_entry rdev_recovery_start =
3470 __ATTR(recovery_start, S_IRUGO|S_IWUSR, recovery_start_show, recovery_start_store);
3471 
3472 /* sysfs access to bad-blocks list.
3473  * We present two files.
3474  * 'bad-blocks' lists sector numbers and lengths of ranges that
3475  *    are recorded as bad.  The list is truncated to fit within
3476  *    the one-page limit of sysfs.
3477  *    Writing "sector length" to this file adds an acknowledged
3478  *    bad block list.
3479  * 'unacknowledged-bad-blocks' lists bad blocks that have not yet
3480  *    been acknowledged.  Writing to this file adds bad blocks
3481  *    without acknowledging them.  This is largely for testing.
3482  */
bb_show(struct md_rdev * rdev,char * page)3483 static ssize_t bb_show(struct md_rdev *rdev, char *page)
3484 {
3485 	return badblocks_show(&rdev->badblocks, page, 0);
3486 }
bb_store(struct md_rdev * rdev,const char * page,size_t len)3487 static ssize_t bb_store(struct md_rdev *rdev, const char *page, size_t len)
3488 {
3489 	int rv = badblocks_store(&rdev->badblocks, page, len, 0);
3490 	/* Maybe that ack was all we needed */
3491 	if (test_and_clear_bit(BlockedBadBlocks, &rdev->flags))
3492 		wake_up(&rdev->blocked_wait);
3493 	return rv;
3494 }
3495 static struct rdev_sysfs_entry rdev_bad_blocks =
3496 __ATTR(bad_blocks, S_IRUGO|S_IWUSR, bb_show, bb_store);
3497 
ubb_show(struct md_rdev * rdev,char * page)3498 static ssize_t ubb_show(struct md_rdev *rdev, char *page)
3499 {
3500 	return badblocks_show(&rdev->badblocks, page, 1);
3501 }
ubb_store(struct md_rdev * rdev,const char * page,size_t len)3502 static ssize_t ubb_store(struct md_rdev *rdev, const char *page, size_t len)
3503 {
3504 	return badblocks_store(&rdev->badblocks, page, len, 1);
3505 }
3506 static struct rdev_sysfs_entry rdev_unack_bad_blocks =
3507 __ATTR(unacknowledged_bad_blocks, S_IRUGO|S_IWUSR, ubb_show, ubb_store);
3508 
3509 static ssize_t
ppl_sector_show(struct md_rdev * rdev,char * page)3510 ppl_sector_show(struct md_rdev *rdev, char *page)
3511 {
3512 	return sprintf(page, "%llu\n", (unsigned long long)rdev->ppl.sector);
3513 }
3514 
3515 static ssize_t
ppl_sector_store(struct md_rdev * rdev,const char * buf,size_t len)3516 ppl_sector_store(struct md_rdev *rdev, const char *buf, size_t len)
3517 {
3518 	unsigned long long sector;
3519 
3520 	if (kstrtoull(buf, 10, &sector) < 0)
3521 		return -EINVAL;
3522 	if (sector != (sector_t)sector)
3523 		return -EINVAL;
3524 
3525 	if (rdev->mddev->pers && test_bit(MD_HAS_PPL, &rdev->mddev->flags) &&
3526 	    rdev->raid_disk >= 0)
3527 		return -EBUSY;
3528 
3529 	if (rdev->mddev->persistent) {
3530 		if (rdev->mddev->major_version == 0)
3531 			return -EINVAL;
3532 		if ((sector > rdev->sb_start &&
3533 		     sector - rdev->sb_start > S16_MAX) ||
3534 		    (sector < rdev->sb_start &&
3535 		     rdev->sb_start - sector > -S16_MIN))
3536 			return -EINVAL;
3537 		rdev->ppl.offset = sector - rdev->sb_start;
3538 	} else if (!rdev->mddev->external) {
3539 		return -EBUSY;
3540 	}
3541 	rdev->ppl.sector = sector;
3542 	return len;
3543 }
3544 
3545 static struct rdev_sysfs_entry rdev_ppl_sector =
3546 __ATTR(ppl_sector, S_IRUGO|S_IWUSR, ppl_sector_show, ppl_sector_store);
3547 
3548 static ssize_t
ppl_size_show(struct md_rdev * rdev,char * page)3549 ppl_size_show(struct md_rdev *rdev, char *page)
3550 {
3551 	return sprintf(page, "%u\n", rdev->ppl.size);
3552 }
3553 
3554 static ssize_t
ppl_size_store(struct md_rdev * rdev,const char * buf,size_t len)3555 ppl_size_store(struct md_rdev *rdev, const char *buf, size_t len)
3556 {
3557 	unsigned int size;
3558 
3559 	if (kstrtouint(buf, 10, &size) < 0)
3560 		return -EINVAL;
3561 
3562 	if (rdev->mddev->pers && test_bit(MD_HAS_PPL, &rdev->mddev->flags) &&
3563 	    rdev->raid_disk >= 0)
3564 		return -EBUSY;
3565 
3566 	if (rdev->mddev->persistent) {
3567 		if (rdev->mddev->major_version == 0)
3568 			return -EINVAL;
3569 		if (size > U16_MAX)
3570 			return -EINVAL;
3571 	} else if (!rdev->mddev->external) {
3572 		return -EBUSY;
3573 	}
3574 	rdev->ppl.size = size;
3575 	return len;
3576 }
3577 
3578 static struct rdev_sysfs_entry rdev_ppl_size =
3579 __ATTR(ppl_size, S_IRUGO|S_IWUSR, ppl_size_show, ppl_size_store);
3580 
3581 static struct attribute *rdev_default_attrs[] = {
3582 	&rdev_state.attr,
3583 	&rdev_errors.attr,
3584 	&rdev_slot.attr,
3585 	&rdev_offset.attr,
3586 	&rdev_new_offset.attr,
3587 	&rdev_size.attr,
3588 	&rdev_recovery_start.attr,
3589 	&rdev_bad_blocks.attr,
3590 	&rdev_unack_bad_blocks.attr,
3591 	&rdev_ppl_sector.attr,
3592 	&rdev_ppl_size.attr,
3593 	NULL,
3594 };
3595 ATTRIBUTE_GROUPS(rdev_default);
3596 static ssize_t
rdev_attr_show(struct kobject * kobj,struct attribute * attr,char * page)3597 rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
3598 {
3599 	struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
3600 	struct md_rdev *rdev = container_of(kobj, struct md_rdev, kobj);
3601 
3602 	if (!entry->show)
3603 		return -EIO;
3604 	if (!rdev->mddev)
3605 		return -ENODEV;
3606 	return entry->show(rdev, page);
3607 }
3608 
3609 static ssize_t
rdev_attr_store(struct kobject * kobj,struct attribute * attr,const char * page,size_t length)3610 rdev_attr_store(struct kobject *kobj, struct attribute *attr,
3611 	      const char *page, size_t length)
3612 {
3613 	struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
3614 	struct md_rdev *rdev = container_of(kobj, struct md_rdev, kobj);
3615 	struct kernfs_node *kn = NULL;
3616 	ssize_t rv;
3617 	struct mddev *mddev = rdev->mddev;
3618 
3619 	if (!entry->store)
3620 		return -EIO;
3621 	if (!capable(CAP_SYS_ADMIN))
3622 		return -EACCES;
3623 
3624 	if (entry->store == state_store && cmd_match(page, "remove"))
3625 		kn = sysfs_break_active_protection(kobj, attr);
3626 
3627 	rv = mddev ? mddev_lock(mddev) : -ENODEV;
3628 	if (!rv) {
3629 		if (rdev->mddev == NULL)
3630 			rv = -ENODEV;
3631 		else
3632 			rv = entry->store(rdev, page, length);
3633 		mddev_unlock(mddev);
3634 	}
3635 
3636 	if (kn)
3637 		sysfs_unbreak_active_protection(kn);
3638 
3639 	return rv;
3640 }
3641 
rdev_free(struct kobject * ko)3642 static void rdev_free(struct kobject *ko)
3643 {
3644 	struct md_rdev *rdev = container_of(ko, struct md_rdev, kobj);
3645 	kfree(rdev);
3646 }
3647 static const struct sysfs_ops rdev_sysfs_ops = {
3648 	.show		= rdev_attr_show,
3649 	.store		= rdev_attr_store,
3650 };
3651 static const struct kobj_type rdev_ktype = {
3652 	.release	= rdev_free,
3653 	.sysfs_ops	= &rdev_sysfs_ops,
3654 	.default_groups	= rdev_default_groups,
3655 };
3656 
md_rdev_init(struct md_rdev * rdev)3657 int md_rdev_init(struct md_rdev *rdev)
3658 {
3659 	rdev->desc_nr = -1;
3660 	rdev->saved_raid_disk = -1;
3661 	rdev->raid_disk = -1;
3662 	rdev->flags = 0;
3663 	rdev->data_offset = 0;
3664 	rdev->new_data_offset = 0;
3665 	rdev->sb_events = 0;
3666 	rdev->last_read_error = 0;
3667 	rdev->sb_loaded = 0;
3668 	rdev->bb_page = NULL;
3669 	atomic_set(&rdev->nr_pending, 0);
3670 	atomic_set(&rdev->read_errors, 0);
3671 	atomic_set(&rdev->corrected_errors, 0);
3672 
3673 	INIT_LIST_HEAD(&rdev->same_set);
3674 	init_waitqueue_head(&rdev->blocked_wait);
3675 
3676 	/* Add space to store bad block list.
3677 	 * This reserves the space even on arrays where it cannot
3678 	 * be used - I wonder if that matters
3679 	 */
3680 	return badblocks_init(&rdev->badblocks, 0);
3681 }
3682 EXPORT_SYMBOL_GPL(md_rdev_init);
3683 
3684 /*
3685  * Import a device. If 'super_format' >= 0, then sanity check the superblock
3686  *
3687  * mark the device faulty if:
3688  *
3689  *   - the device is nonexistent (zero size)
3690  *   - the device has no valid superblock
3691  *
3692  * a faulty rdev _never_ has rdev->sb set.
3693  */
md_import_device(dev_t newdev,int super_format,int super_minor)3694 static struct md_rdev *md_import_device(dev_t newdev, int super_format, int super_minor)
3695 {
3696 	struct md_rdev *rdev;
3697 	struct md_rdev *holder;
3698 	sector_t size;
3699 	int err;
3700 
3701 	rdev = kzalloc(sizeof(*rdev), GFP_KERNEL);
3702 	if (!rdev)
3703 		return ERR_PTR(-ENOMEM);
3704 
3705 	err = md_rdev_init(rdev);
3706 	if (err)
3707 		goto out_free_rdev;
3708 	err = alloc_disk_sb(rdev);
3709 	if (err)
3710 		goto out_clear_rdev;
3711 
3712 	if (super_format == -2) {
3713 		holder = &claim_rdev;
3714 	} else {
3715 		holder = rdev;
3716 		set_bit(Holder, &rdev->flags);
3717 	}
3718 
3719 	rdev->bdev = blkdev_get_by_dev(newdev, BLK_OPEN_READ | BLK_OPEN_WRITE,
3720 				       holder, NULL);
3721 	if (IS_ERR(rdev->bdev)) {
3722 		pr_warn("md: could not open device unknown-block(%u,%u).\n",
3723 			MAJOR(newdev), MINOR(newdev));
3724 		err = PTR_ERR(rdev->bdev);
3725 		goto out_clear_rdev;
3726 	}
3727 
3728 	kobject_init(&rdev->kobj, &rdev_ktype);
3729 
3730 	size = bdev_nr_bytes(rdev->bdev) >> BLOCK_SIZE_BITS;
3731 	if (!size) {
3732 		pr_warn("md: %pg has zero or unknown size, marking faulty!\n",
3733 			rdev->bdev);
3734 		err = -EINVAL;
3735 		goto out_blkdev_put;
3736 	}
3737 
3738 	if (super_format >= 0) {
3739 		err = super_types[super_format].
3740 			load_super(rdev, NULL, super_minor);
3741 		if (err == -EINVAL) {
3742 			pr_warn("md: %pg does not have a valid v%d.%d superblock, not importing!\n",
3743 				rdev->bdev,
3744 				super_format, super_minor);
3745 			goto out_blkdev_put;
3746 		}
3747 		if (err < 0) {
3748 			pr_warn("md: could not read %pg's sb, not importing!\n",
3749 				rdev->bdev);
3750 			goto out_blkdev_put;
3751 		}
3752 	}
3753 
3754 	return rdev;
3755 
3756 out_blkdev_put:
3757 	blkdev_put(rdev->bdev, holder);
3758 out_clear_rdev:
3759 	md_rdev_clear(rdev);
3760 out_free_rdev:
3761 	kfree(rdev);
3762 	return ERR_PTR(err);
3763 }
3764 
3765 /*
3766  * Check a full RAID array for plausibility
3767  */
3768 
analyze_sbs(struct mddev * mddev)3769 static int analyze_sbs(struct mddev *mddev)
3770 {
3771 	int i;
3772 	struct md_rdev *rdev, *freshest, *tmp;
3773 
3774 	freshest = NULL;
3775 	rdev_for_each_safe(rdev, tmp, mddev)
3776 		switch (super_types[mddev->major_version].
3777 			load_super(rdev, freshest, mddev->minor_version)) {
3778 		case 1:
3779 			freshest = rdev;
3780 			break;
3781 		case 0:
3782 			break;
3783 		default:
3784 			pr_warn("md: fatal superblock inconsistency in %pg -- removing from array\n",
3785 				rdev->bdev);
3786 			md_kick_rdev_from_array(rdev);
3787 		}
3788 
3789 	/* Cannot find a valid fresh disk */
3790 	if (!freshest) {
3791 		pr_warn("md: cannot find a valid disk\n");
3792 		return -EINVAL;
3793 	}
3794 
3795 	super_types[mddev->major_version].
3796 		validate_super(mddev, NULL/*freshest*/, freshest);
3797 
3798 	i = 0;
3799 	rdev_for_each_safe(rdev, tmp, mddev) {
3800 		if (mddev->max_disks &&
3801 		    (rdev->desc_nr >= mddev->max_disks ||
3802 		     i > mddev->max_disks)) {
3803 			pr_warn("md: %s: %pg: only %d devices permitted\n",
3804 				mdname(mddev), rdev->bdev,
3805 				mddev->max_disks);
3806 			md_kick_rdev_from_array(rdev);
3807 			continue;
3808 		}
3809 		if (rdev != freshest) {
3810 			if (super_types[mddev->major_version].
3811 			    validate_super(mddev, freshest, rdev)) {
3812 				pr_warn("md: kicking non-fresh %pg from array!\n",
3813 					rdev->bdev);
3814 				md_kick_rdev_from_array(rdev);
3815 				continue;
3816 			}
3817 		}
3818 		if (mddev->level == LEVEL_MULTIPATH) {
3819 			rdev->desc_nr = i++;
3820 			rdev->raid_disk = rdev->desc_nr;
3821 			set_bit(In_sync, &rdev->flags);
3822 		} else if (rdev->raid_disk >=
3823 			    (mddev->raid_disks - min(0, mddev->delta_disks)) &&
3824 			   !test_bit(Journal, &rdev->flags)) {
3825 			rdev->raid_disk = -1;
3826 			clear_bit(In_sync, &rdev->flags);
3827 		}
3828 	}
3829 
3830 	return 0;
3831 }
3832 
3833 /* Read a fixed-point number.
3834  * Numbers in sysfs attributes should be in "standard" units where
3835  * possible, so time should be in seconds.
3836  * However we internally use a a much smaller unit such as
3837  * milliseconds or jiffies.
3838  * This function takes a decimal number with a possible fractional
3839  * component, and produces an integer which is the result of
3840  * multiplying that number by 10^'scale'.
3841  * all without any floating-point arithmetic.
3842  */
strict_strtoul_scaled(const char * cp,unsigned long * res,int scale)3843 int strict_strtoul_scaled(const char *cp, unsigned long *res, int scale)
3844 {
3845 	unsigned long result = 0;
3846 	long decimals = -1;
3847 	while (isdigit(*cp) || (*cp == '.' && decimals < 0)) {
3848 		if (*cp == '.')
3849 			decimals = 0;
3850 		else if (decimals < scale) {
3851 			unsigned int value;
3852 			value = *cp - '0';
3853 			result = result * 10 + value;
3854 			if (decimals >= 0)
3855 				decimals++;
3856 		}
3857 		cp++;
3858 	}
3859 	if (*cp == '\n')
3860 		cp++;
3861 	if (*cp)
3862 		return -EINVAL;
3863 	if (decimals < 0)
3864 		decimals = 0;
3865 	*res = result * int_pow(10, scale - decimals);
3866 	return 0;
3867 }
3868 
3869 static ssize_t
safe_delay_show(struct mddev * mddev,char * page)3870 safe_delay_show(struct mddev *mddev, char *page)
3871 {
3872 	unsigned int msec = ((unsigned long)mddev->safemode_delay*1000)/HZ;
3873 
3874 	return sprintf(page, "%u.%03u\n", msec/1000, msec%1000);
3875 }
3876 static ssize_t
safe_delay_store(struct mddev * mddev,const char * cbuf,size_t len)3877 safe_delay_store(struct mddev *mddev, const char *cbuf, size_t len)
3878 {
3879 	unsigned long msec;
3880 
3881 	if (mddev_is_clustered(mddev)) {
3882 		pr_warn("md: Safemode is disabled for clustered mode\n");
3883 		return -EINVAL;
3884 	}
3885 
3886 	if (strict_strtoul_scaled(cbuf, &msec, 3) < 0 || msec > UINT_MAX / HZ)
3887 		return -EINVAL;
3888 	if (msec == 0)
3889 		mddev->safemode_delay = 0;
3890 	else {
3891 		unsigned long old_delay = mddev->safemode_delay;
3892 		unsigned long new_delay = (msec*HZ)/1000;
3893 
3894 		if (new_delay == 0)
3895 			new_delay = 1;
3896 		mddev->safemode_delay = new_delay;
3897 		if (new_delay < old_delay || old_delay == 0)
3898 			mod_timer(&mddev->safemode_timer, jiffies+1);
3899 	}
3900 	return len;
3901 }
3902 static struct md_sysfs_entry md_safe_delay =
3903 __ATTR(safe_mode_delay, S_IRUGO|S_IWUSR,safe_delay_show, safe_delay_store);
3904 
3905 static ssize_t
level_show(struct mddev * mddev,char * page)3906 level_show(struct mddev *mddev, char *page)
3907 {
3908 	struct md_personality *p;
3909 	int ret;
3910 	spin_lock(&mddev->lock);
3911 	p = mddev->pers;
3912 	if (p)
3913 		ret = sprintf(page, "%s\n", p->name);
3914 	else if (mddev->clevel[0])
3915 		ret = sprintf(page, "%s\n", mddev->clevel);
3916 	else if (mddev->level != LEVEL_NONE)
3917 		ret = sprintf(page, "%d\n", mddev->level);
3918 	else
3919 		ret = 0;
3920 	spin_unlock(&mddev->lock);
3921 	return ret;
3922 }
3923 
3924 static ssize_t
level_store(struct mddev * mddev,const char * buf,size_t len)3925 level_store(struct mddev *mddev, const char *buf, size_t len)
3926 {
3927 	char clevel[16];
3928 	ssize_t rv;
3929 	size_t slen = len;
3930 	struct md_personality *pers, *oldpers;
3931 	long level;
3932 	void *priv, *oldpriv;
3933 	struct md_rdev *rdev;
3934 
3935 	if (slen == 0 || slen >= sizeof(clevel))
3936 		return -EINVAL;
3937 
3938 	rv = mddev_lock(mddev);
3939 	if (rv)
3940 		return rv;
3941 
3942 	if (mddev->pers == NULL) {
3943 		strncpy(mddev->clevel, buf, slen);
3944 		if (mddev->clevel[slen-1] == '\n')
3945 			slen--;
3946 		mddev->clevel[slen] = 0;
3947 		mddev->level = LEVEL_NONE;
3948 		rv = len;
3949 		goto out_unlock;
3950 	}
3951 	rv = -EROFS;
3952 	if (!md_is_rdwr(mddev))
3953 		goto out_unlock;
3954 
3955 	/* request to change the personality.  Need to ensure:
3956 	 *  - array is not engaged in resync/recovery/reshape
3957 	 *  - old personality can be suspended
3958 	 *  - new personality will access other array.
3959 	 */
3960 
3961 	rv = -EBUSY;
3962 	if (mddev->sync_thread ||
3963 	    test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
3964 	    mddev->reshape_position != MaxSector ||
3965 	    mddev->sysfs_active)
3966 		goto out_unlock;
3967 
3968 	rv = -EINVAL;
3969 	if (!mddev->pers->quiesce) {
3970 		pr_warn("md: %s: %s does not support online personality change\n",
3971 			mdname(mddev), mddev->pers->name);
3972 		goto out_unlock;
3973 	}
3974 
3975 	/* Now find the new personality */
3976 	strncpy(clevel, buf, slen);
3977 	if (clevel[slen-1] == '\n')
3978 		slen--;
3979 	clevel[slen] = 0;
3980 	if (kstrtol(clevel, 10, &level))
3981 		level = LEVEL_NONE;
3982 
3983 	if (request_module("md-%s", clevel) != 0)
3984 		request_module("md-level-%s", clevel);
3985 	spin_lock(&pers_lock);
3986 	pers = find_pers(level, clevel);
3987 	if (!pers || !try_module_get(pers->owner)) {
3988 		spin_unlock(&pers_lock);
3989 		pr_warn("md: personality %s not loaded\n", clevel);
3990 		rv = -EINVAL;
3991 		goto out_unlock;
3992 	}
3993 	spin_unlock(&pers_lock);
3994 
3995 	if (pers == mddev->pers) {
3996 		/* Nothing to do! */
3997 		module_put(pers->owner);
3998 		rv = len;
3999 		goto out_unlock;
4000 	}
4001 	if (!pers->takeover) {
4002 		module_put(pers->owner);
4003 		pr_warn("md: %s: %s does not support personality takeover\n",
4004 			mdname(mddev), clevel);
4005 		rv = -EINVAL;
4006 		goto out_unlock;
4007 	}
4008 
4009 	rdev_for_each(rdev, mddev)
4010 		rdev->new_raid_disk = rdev->raid_disk;
4011 
4012 	/* ->takeover must set new_* and/or delta_disks
4013 	 * if it succeeds, and may set them when it fails.
4014 	 */
4015 	priv = pers->takeover(mddev);
4016 	if (IS_ERR(priv)) {
4017 		mddev->new_level = mddev->level;
4018 		mddev->new_layout = mddev->layout;
4019 		mddev->new_chunk_sectors = mddev->chunk_sectors;
4020 		mddev->raid_disks -= mddev->delta_disks;
4021 		mddev->delta_disks = 0;
4022 		mddev->reshape_backwards = 0;
4023 		module_put(pers->owner);
4024 		pr_warn("md: %s: %s would not accept array\n",
4025 			mdname(mddev), clevel);
4026 		rv = PTR_ERR(priv);
4027 		goto out_unlock;
4028 	}
4029 
4030 	/* Looks like we have a winner */
4031 	mddev_suspend(mddev);
4032 	mddev_detach(mddev);
4033 
4034 	spin_lock(&mddev->lock);
4035 	oldpers = mddev->pers;
4036 	oldpriv = mddev->private;
4037 	mddev->pers = pers;
4038 	mddev->private = priv;
4039 	strscpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
4040 	mddev->level = mddev->new_level;
4041 	mddev->layout = mddev->new_layout;
4042 	mddev->chunk_sectors = mddev->new_chunk_sectors;
4043 	mddev->delta_disks = 0;
4044 	mddev->reshape_backwards = 0;
4045 	mddev->degraded = 0;
4046 	spin_unlock(&mddev->lock);
4047 
4048 	if (oldpers->sync_request == NULL &&
4049 	    mddev->external) {
4050 		/* We are converting from a no-redundancy array
4051 		 * to a redundancy array and metadata is managed
4052 		 * externally so we need to be sure that writes
4053 		 * won't block due to a need to transition
4054 		 *      clean->dirty
4055 		 * until external management is started.
4056 		 */
4057 		mddev->in_sync = 0;
4058 		mddev->safemode_delay = 0;
4059 		mddev->safemode = 0;
4060 	}
4061 
4062 	oldpers->free(mddev, oldpriv);
4063 
4064 	if (oldpers->sync_request == NULL &&
4065 	    pers->sync_request != NULL) {
4066 		/* need to add the md_redundancy_group */
4067 		if (sysfs_create_group(&mddev->kobj, &md_redundancy_group))
4068 			pr_warn("md: cannot register extra attributes for %s\n",
4069 				mdname(mddev));
4070 		mddev->sysfs_action = sysfs_get_dirent(mddev->kobj.sd, "sync_action");
4071 		mddev->sysfs_completed = sysfs_get_dirent_safe(mddev->kobj.sd, "sync_completed");
4072 		mddev->sysfs_degraded = sysfs_get_dirent_safe(mddev->kobj.sd, "degraded");
4073 	}
4074 	if (oldpers->sync_request != NULL &&
4075 	    pers->sync_request == NULL) {
4076 		/* need to remove the md_redundancy_group */
4077 		if (mddev->to_remove == NULL)
4078 			mddev->to_remove = &md_redundancy_group;
4079 	}
4080 
4081 	module_put(oldpers->owner);
4082 
4083 	rdev_for_each(rdev, mddev) {
4084 		if (rdev->raid_disk < 0)
4085 			continue;
4086 		if (rdev->new_raid_disk >= mddev->raid_disks)
4087 			rdev->new_raid_disk = -1;
4088 		if (rdev->new_raid_disk == rdev->raid_disk)
4089 			continue;
4090 		sysfs_unlink_rdev(mddev, rdev);
4091 	}
4092 	rdev_for_each(rdev, mddev) {
4093 		if (rdev->raid_disk < 0)
4094 			continue;
4095 		if (rdev->new_raid_disk == rdev->raid_disk)
4096 			continue;
4097 		rdev->raid_disk = rdev->new_raid_disk;
4098 		if (rdev->raid_disk < 0)
4099 			clear_bit(In_sync, &rdev->flags);
4100 		else {
4101 			if (sysfs_link_rdev(mddev, rdev))
4102 				pr_warn("md: cannot register rd%d for %s after level change\n",
4103 					rdev->raid_disk, mdname(mddev));
4104 		}
4105 	}
4106 
4107 	if (pers->sync_request == NULL) {
4108 		/* this is now an array without redundancy, so
4109 		 * it must always be in_sync
4110 		 */
4111 		mddev->in_sync = 1;
4112 		del_timer_sync(&mddev->safemode_timer);
4113 	}
4114 	blk_set_stacking_limits(&mddev->queue->limits);
4115 	pers->run(mddev);
4116 	set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
4117 	mddev_resume(mddev);
4118 	if (!mddev->thread)
4119 		md_update_sb(mddev, 1);
4120 	sysfs_notify_dirent_safe(mddev->sysfs_level);
4121 	md_new_event();
4122 	rv = len;
4123 out_unlock:
4124 	mddev_unlock(mddev);
4125 	return rv;
4126 }
4127 
4128 static struct md_sysfs_entry md_level =
4129 __ATTR(level, S_IRUGO|S_IWUSR, level_show, level_store);
4130 
4131 static ssize_t
layout_show(struct mddev * mddev,char * page)4132 layout_show(struct mddev *mddev, char *page)
4133 {
4134 	/* just a number, not meaningful for all levels */
4135 	if (mddev->reshape_position != MaxSector &&
4136 	    mddev->layout != mddev->new_layout)
4137 		return sprintf(page, "%d (%d)\n",
4138 			       mddev->new_layout, mddev->layout);
4139 	return sprintf(page, "%d\n", mddev->layout);
4140 }
4141 
4142 static ssize_t
layout_store(struct mddev * mddev,const char * buf,size_t len)4143 layout_store(struct mddev *mddev, const char *buf, size_t len)
4144 {
4145 	unsigned int n;
4146 	int err;
4147 
4148 	err = kstrtouint(buf, 10, &n);
4149 	if (err < 0)
4150 		return err;
4151 	err = mddev_lock(mddev);
4152 	if (err)
4153 		return err;
4154 
4155 	if (mddev->pers) {
4156 		if (mddev->pers->check_reshape == NULL)
4157 			err = -EBUSY;
4158 		else if (!md_is_rdwr(mddev))
4159 			err = -EROFS;
4160 		else {
4161 			mddev->new_layout = n;
4162 			err = mddev->pers->check_reshape(mddev);
4163 			if (err)
4164 				mddev->new_layout = mddev->layout;
4165 		}
4166 	} else {
4167 		mddev->new_layout = n;
4168 		if (mddev->reshape_position == MaxSector)
4169 			mddev->layout = n;
4170 	}
4171 	mddev_unlock(mddev);
4172 	return err ?: len;
4173 }
4174 static struct md_sysfs_entry md_layout =
4175 __ATTR(layout, S_IRUGO|S_IWUSR, layout_show, layout_store);
4176 
4177 static ssize_t
raid_disks_show(struct mddev * mddev,char * page)4178 raid_disks_show(struct mddev *mddev, char *page)
4179 {
4180 	if (mddev->raid_disks == 0)
4181 		return 0;
4182 	if (mddev->reshape_position != MaxSector &&
4183 	    mddev->delta_disks != 0)
4184 		return sprintf(page, "%d (%d)\n", mddev->raid_disks,
4185 			       mddev->raid_disks - mddev->delta_disks);
4186 	return sprintf(page, "%d\n", mddev->raid_disks);
4187 }
4188 
4189 static int update_raid_disks(struct mddev *mddev, int raid_disks);
4190 
4191 static ssize_t
raid_disks_store(struct mddev * mddev,const char * buf,size_t len)4192 raid_disks_store(struct mddev *mddev, const char *buf, size_t len)
4193 {
4194 	unsigned int n;
4195 	int err;
4196 
4197 	err = kstrtouint(buf, 10, &n);
4198 	if (err < 0)
4199 		return err;
4200 
4201 	err = mddev_lock(mddev);
4202 	if (err)
4203 		return err;
4204 	if (mddev->pers)
4205 		err = update_raid_disks(mddev, n);
4206 	else if (mddev->reshape_position != MaxSector) {
4207 		struct md_rdev *rdev;
4208 		int olddisks = mddev->raid_disks - mddev->delta_disks;
4209 
4210 		err = -EINVAL;
4211 		rdev_for_each(rdev, mddev) {
4212 			if (olddisks < n &&
4213 			    rdev->data_offset < rdev->new_data_offset)
4214 				goto out_unlock;
4215 			if (olddisks > n &&
4216 			    rdev->data_offset > rdev->new_data_offset)
4217 				goto out_unlock;
4218 		}
4219 		err = 0;
4220 		mddev->delta_disks = n - olddisks;
4221 		mddev->raid_disks = n;
4222 		mddev->reshape_backwards = (mddev->delta_disks < 0);
4223 	} else
4224 		mddev->raid_disks = n;
4225 out_unlock:
4226 	mddev_unlock(mddev);
4227 	return err ? err : len;
4228 }
4229 static struct md_sysfs_entry md_raid_disks =
4230 __ATTR(raid_disks, S_IRUGO|S_IWUSR, raid_disks_show, raid_disks_store);
4231 
4232 static ssize_t
uuid_show(struct mddev * mddev,char * page)4233 uuid_show(struct mddev *mddev, char *page)
4234 {
4235 	return sprintf(page, "%pU\n", mddev->uuid);
4236 }
4237 static struct md_sysfs_entry md_uuid =
4238 __ATTR(uuid, S_IRUGO, uuid_show, NULL);
4239 
4240 static ssize_t
chunk_size_show(struct mddev * mddev,char * page)4241 chunk_size_show(struct mddev *mddev, char *page)
4242 {
4243 	if (mddev->reshape_position != MaxSector &&
4244 	    mddev->chunk_sectors != mddev->new_chunk_sectors)
4245 		return sprintf(page, "%d (%d)\n",
4246 			       mddev->new_chunk_sectors << 9,
4247 			       mddev->chunk_sectors << 9);
4248 	return sprintf(page, "%d\n", mddev->chunk_sectors << 9);
4249 }
4250 
4251 static ssize_t
chunk_size_store(struct mddev * mddev,const char * buf,size_t len)4252 chunk_size_store(struct mddev *mddev, const char *buf, size_t len)
4253 {
4254 	unsigned long n;
4255 	int err;
4256 
4257 	err = kstrtoul(buf, 10, &n);
4258 	if (err < 0)
4259 		return err;
4260 
4261 	err = mddev_lock(mddev);
4262 	if (err)
4263 		return err;
4264 	if (mddev->pers) {
4265 		if (mddev->pers->check_reshape == NULL)
4266 			err = -EBUSY;
4267 		else if (!md_is_rdwr(mddev))
4268 			err = -EROFS;
4269 		else {
4270 			mddev->new_chunk_sectors = n >> 9;
4271 			err = mddev->pers->check_reshape(mddev);
4272 			if (err)
4273 				mddev->new_chunk_sectors = mddev->chunk_sectors;
4274 		}
4275 	} else {
4276 		mddev->new_chunk_sectors = n >> 9;
4277 		if (mddev->reshape_position == MaxSector)
4278 			mddev->chunk_sectors = n >> 9;
4279 	}
4280 	mddev_unlock(mddev);
4281 	return err ?: len;
4282 }
4283 static struct md_sysfs_entry md_chunk_size =
4284 __ATTR(chunk_size, S_IRUGO|S_IWUSR, chunk_size_show, chunk_size_store);
4285 
4286 static ssize_t
resync_start_show(struct mddev * mddev,char * page)4287 resync_start_show(struct mddev *mddev, char *page)
4288 {
4289 	if (mddev->recovery_cp == MaxSector)
4290 		return sprintf(page, "none\n");
4291 	return sprintf(page, "%llu\n", (unsigned long long)mddev->recovery_cp);
4292 }
4293 
4294 static ssize_t
resync_start_store(struct mddev * mddev,const char * buf,size_t len)4295 resync_start_store(struct mddev *mddev, const char *buf, size_t len)
4296 {
4297 	unsigned long long n;
4298 	int err;
4299 
4300 	if (cmd_match(buf, "none"))
4301 		n = MaxSector;
4302 	else {
4303 		err = kstrtoull(buf, 10, &n);
4304 		if (err < 0)
4305 			return err;
4306 		if (n != (sector_t)n)
4307 			return -EINVAL;
4308 	}
4309 
4310 	err = mddev_lock(mddev);
4311 	if (err)
4312 		return err;
4313 	if (mddev->pers && !test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
4314 		err = -EBUSY;
4315 
4316 	if (!err) {
4317 		mddev->recovery_cp = n;
4318 		if (mddev->pers)
4319 			set_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags);
4320 	}
4321 	mddev_unlock(mddev);
4322 	return err ?: len;
4323 }
4324 static struct md_sysfs_entry md_resync_start =
4325 __ATTR_PREALLOC(resync_start, S_IRUGO|S_IWUSR,
4326 		resync_start_show, resync_start_store);
4327 
4328 /*
4329  * The array state can be:
4330  *
4331  * clear
4332  *     No devices, no size, no level
4333  *     Equivalent to STOP_ARRAY ioctl
4334  * inactive
4335  *     May have some settings, but array is not active
4336  *        all IO results in error
4337  *     When written, doesn't tear down array, but just stops it
4338  * suspended (not supported yet)
4339  *     All IO requests will block. The array can be reconfigured.
4340  *     Writing this, if accepted, will block until array is quiescent
4341  * readonly
4342  *     no resync can happen.  no superblocks get written.
4343  *     write requests fail
4344  * read-auto
4345  *     like readonly, but behaves like 'clean' on a write request.
4346  *
4347  * clean - no pending writes, but otherwise active.
4348  *     When written to inactive array, starts without resync
4349  *     If a write request arrives then
4350  *       if metadata is known, mark 'dirty' and switch to 'active'.
4351  *       if not known, block and switch to write-pending
4352  *     If written to an active array that has pending writes, then fails.
4353  * active
4354  *     fully active: IO and resync can be happening.
4355  *     When written to inactive array, starts with resync
4356  *
4357  * write-pending
4358  *     clean, but writes are blocked waiting for 'active' to be written.
4359  *
4360  * active-idle
4361  *     like active, but no writes have been seen for a while (100msec).
4362  *
4363  * broken
4364 *     Array is failed. It's useful because mounted-arrays aren't stopped
4365 *     when array is failed, so this state will at least alert the user that
4366 *     something is wrong.
4367  */
4368 enum array_state { clear, inactive, suspended, readonly, read_auto, clean, active,
4369 		   write_pending, active_idle, broken, bad_word};
4370 static char *array_states[] = {
4371 	"clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active",
4372 	"write-pending", "active-idle", "broken", NULL };
4373 
match_word(const char * word,char ** list)4374 static int match_word(const char *word, char **list)
4375 {
4376 	int n;
4377 	for (n=0; list[n]; n++)
4378 		if (cmd_match(word, list[n]))
4379 			break;
4380 	return n;
4381 }
4382 
4383 static ssize_t
array_state_show(struct mddev * mddev,char * page)4384 array_state_show(struct mddev *mddev, char *page)
4385 {
4386 	enum array_state st = inactive;
4387 
4388 	if (mddev->pers && !test_bit(MD_NOT_READY, &mddev->flags)) {
4389 		switch(mddev->ro) {
4390 		case MD_RDONLY:
4391 			st = readonly;
4392 			break;
4393 		case MD_AUTO_READ:
4394 			st = read_auto;
4395 			break;
4396 		case MD_RDWR:
4397 			spin_lock(&mddev->lock);
4398 			if (test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags))
4399 				st = write_pending;
4400 			else if (mddev->in_sync)
4401 				st = clean;
4402 			else if (mddev->safemode)
4403 				st = active_idle;
4404 			else
4405 				st = active;
4406 			spin_unlock(&mddev->lock);
4407 		}
4408 
4409 		if (test_bit(MD_BROKEN, &mddev->flags) && st == clean)
4410 			st = broken;
4411 	} else {
4412 		if (list_empty(&mddev->disks) &&
4413 		    mddev->raid_disks == 0 &&
4414 		    mddev->dev_sectors == 0)
4415 			st = clear;
4416 		else
4417 			st = inactive;
4418 	}
4419 	return sprintf(page, "%s\n", array_states[st]);
4420 }
4421 
4422 static int do_md_stop(struct mddev *mddev, int ro, struct block_device *bdev);
4423 static int md_set_readonly(struct mddev *mddev, struct block_device *bdev);
4424 static int restart_array(struct mddev *mddev);
4425 
4426 static ssize_t
array_state_store(struct mddev * mddev,const char * buf,size_t len)4427 array_state_store(struct mddev *mddev, const char *buf, size_t len)
4428 {
4429 	int err = 0;
4430 	enum array_state st = match_word(buf, array_states);
4431 
4432 	if (mddev->pers && (st == active || st == clean) &&
4433 	    mddev->ro != MD_RDONLY) {
4434 		/* don't take reconfig_mutex when toggling between
4435 		 * clean and active
4436 		 */
4437 		spin_lock(&mddev->lock);
4438 		if (st == active) {
4439 			restart_array(mddev);
4440 			clear_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags);
4441 			md_wakeup_thread(mddev->thread);
4442 			wake_up(&mddev->sb_wait);
4443 		} else /* st == clean */ {
4444 			restart_array(mddev);
4445 			if (!set_in_sync(mddev))
4446 				err = -EBUSY;
4447 		}
4448 		if (!err)
4449 			sysfs_notify_dirent_safe(mddev->sysfs_state);
4450 		spin_unlock(&mddev->lock);
4451 		return err ?: len;
4452 	}
4453 	err = mddev_lock(mddev);
4454 	if (err)
4455 		return err;
4456 	err = -EINVAL;
4457 	switch(st) {
4458 	case bad_word:
4459 		break;
4460 	case clear:
4461 		/* stopping an active array */
4462 		err = do_md_stop(mddev, 0, NULL);
4463 		break;
4464 	case inactive:
4465 		/* stopping an active array */
4466 		if (mddev->pers)
4467 			err = do_md_stop(mddev, 2, NULL);
4468 		else
4469 			err = 0; /* already inactive */
4470 		break;
4471 	case suspended:
4472 		break; /* not supported yet */
4473 	case readonly:
4474 		if (mddev->pers)
4475 			err = md_set_readonly(mddev, NULL);
4476 		else {
4477 			mddev->ro = MD_RDONLY;
4478 			set_disk_ro(mddev->gendisk, 1);
4479 			err = do_md_run(mddev);
4480 		}
4481 		break;
4482 	case read_auto:
4483 		if (mddev->pers) {
4484 			if (md_is_rdwr(mddev))
4485 				err = md_set_readonly(mddev, NULL);
4486 			else if (mddev->ro == MD_RDONLY)
4487 				err = restart_array(mddev);
4488 			if (err == 0) {
4489 				mddev->ro = MD_AUTO_READ;
4490 				set_disk_ro(mddev->gendisk, 0);
4491 			}
4492 		} else {
4493 			mddev->ro = MD_AUTO_READ;
4494 			err = do_md_run(mddev);
4495 		}
4496 		break;
4497 	case clean:
4498 		if (mddev->pers) {
4499 			err = restart_array(mddev);
4500 			if (err)
4501 				break;
4502 			spin_lock(&mddev->lock);
4503 			if (!set_in_sync(mddev))
4504 				err = -EBUSY;
4505 			spin_unlock(&mddev->lock);
4506 		} else
4507 			err = -EINVAL;
4508 		break;
4509 	case active:
4510 		if (mddev->pers) {
4511 			err = restart_array(mddev);
4512 			if (err)
4513 				break;
4514 			clear_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags);
4515 			wake_up(&mddev->sb_wait);
4516 			err = 0;
4517 		} else {
4518 			mddev->ro = MD_RDWR;
4519 			set_disk_ro(mddev->gendisk, 0);
4520 			err = do_md_run(mddev);
4521 		}
4522 		break;
4523 	case write_pending:
4524 	case active_idle:
4525 	case broken:
4526 		/* these cannot be set */
4527 		break;
4528 	}
4529 
4530 	if (!err) {
4531 		if (mddev->hold_active == UNTIL_IOCTL)
4532 			mddev->hold_active = 0;
4533 		sysfs_notify_dirent_safe(mddev->sysfs_state);
4534 	}
4535 	mddev_unlock(mddev);
4536 	return err ?: len;
4537 }
4538 static struct md_sysfs_entry md_array_state =
4539 __ATTR_PREALLOC(array_state, S_IRUGO|S_IWUSR, array_state_show, array_state_store);
4540 
4541 static ssize_t
max_corrected_read_errors_show(struct mddev * mddev,char * page)4542 max_corrected_read_errors_show(struct mddev *mddev, char *page) {
4543 	return sprintf(page, "%d\n",
4544 		       atomic_read(&mddev->max_corr_read_errors));
4545 }
4546 
4547 static ssize_t
max_corrected_read_errors_store(struct mddev * mddev,const char * buf,size_t len)4548 max_corrected_read_errors_store(struct mddev *mddev, const char *buf, size_t len)
4549 {
4550 	unsigned int n;
4551 	int rv;
4552 
4553 	rv = kstrtouint(buf, 10, &n);
4554 	if (rv < 0)
4555 		return rv;
4556 	if (n > INT_MAX)
4557 		return -EINVAL;
4558 	atomic_set(&mddev->max_corr_read_errors, n);
4559 	return len;
4560 }
4561 
4562 static struct md_sysfs_entry max_corr_read_errors =
4563 __ATTR(max_read_errors, S_IRUGO|S_IWUSR, max_corrected_read_errors_show,
4564 	max_corrected_read_errors_store);
4565 
4566 static ssize_t
null_show(struct mddev * mddev,char * page)4567 null_show(struct mddev *mddev, char *page)
4568 {
4569 	return -EINVAL;
4570 }
4571 
4572 static ssize_t
new_dev_store(struct mddev * mddev,const char * buf,size_t len)4573 new_dev_store(struct mddev *mddev, const char *buf, size_t len)
4574 {
4575 	/* buf must be %d:%d\n? giving major and minor numbers */
4576 	/* The new device is added to the array.
4577 	 * If the array has a persistent superblock, we read the
4578 	 * superblock to initialise info and check validity.
4579 	 * Otherwise, only checking done is that in bind_rdev_to_array,
4580 	 * which mainly checks size.
4581 	 */
4582 	char *e;
4583 	int major = simple_strtoul(buf, &e, 10);
4584 	int minor;
4585 	dev_t dev;
4586 	struct md_rdev *rdev;
4587 	int err;
4588 
4589 	if (!*buf || *e != ':' || !e[1] || e[1] == '\n')
4590 		return -EINVAL;
4591 	minor = simple_strtoul(e+1, &e, 10);
4592 	if (*e && *e != '\n')
4593 		return -EINVAL;
4594 	dev = MKDEV(major, minor);
4595 	if (major != MAJOR(dev) ||
4596 	    minor != MINOR(dev))
4597 		return -EOVERFLOW;
4598 
4599 	err = mddev_lock(mddev);
4600 	if (err)
4601 		return err;
4602 	if (mddev->persistent) {
4603 		rdev = md_import_device(dev, mddev->major_version,
4604 					mddev->minor_version);
4605 		if (!IS_ERR(rdev) && !list_empty(&mddev->disks)) {
4606 			struct md_rdev *rdev0
4607 				= list_entry(mddev->disks.next,
4608 					     struct md_rdev, same_set);
4609 			err = super_types[mddev->major_version]
4610 				.load_super(rdev, rdev0, mddev->minor_version);
4611 			if (err < 0)
4612 				goto out;
4613 		}
4614 	} else if (mddev->external)
4615 		rdev = md_import_device(dev, -2, -1);
4616 	else
4617 		rdev = md_import_device(dev, -1, -1);
4618 
4619 	if (IS_ERR(rdev)) {
4620 		mddev_unlock(mddev);
4621 		return PTR_ERR(rdev);
4622 	}
4623 	err = bind_rdev_to_array(rdev, mddev);
4624  out:
4625 	if (err)
4626 		export_rdev(rdev, mddev);
4627 	mddev_unlock(mddev);
4628 	if (!err)
4629 		md_new_event();
4630 	return err ? err : len;
4631 }
4632 
4633 static struct md_sysfs_entry md_new_device =
4634 __ATTR(new_dev, S_IWUSR, null_show, new_dev_store);
4635 
4636 static ssize_t
bitmap_store(struct mddev * mddev,const char * buf,size_t len)4637 bitmap_store(struct mddev *mddev, const char *buf, size_t len)
4638 {
4639 	char *end;
4640 	unsigned long chunk, end_chunk;
4641 	int err;
4642 
4643 	err = mddev_lock(mddev);
4644 	if (err)
4645 		return err;
4646 	if (!mddev->bitmap)
4647 		goto out;
4648 	/* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */
4649 	while (*buf) {
4650 		chunk = end_chunk = simple_strtoul(buf, &end, 0);
4651 		if (buf == end) break;
4652 		if (*end == '-') { /* range */
4653 			buf = end + 1;
4654 			end_chunk = simple_strtoul(buf, &end, 0);
4655 			if (buf == end) break;
4656 		}
4657 		if (*end && !isspace(*end)) break;
4658 		md_bitmap_dirty_bits(mddev->bitmap, chunk, end_chunk);
4659 		buf = skip_spaces(end);
4660 	}
4661 	md_bitmap_unplug(mddev->bitmap); /* flush the bits to disk */
4662 out:
4663 	mddev_unlock(mddev);
4664 	return len;
4665 }
4666 
4667 static struct md_sysfs_entry md_bitmap =
4668 __ATTR(bitmap_set_bits, S_IWUSR, null_show, bitmap_store);
4669 
4670 static ssize_t
size_show(struct mddev * mddev,char * page)4671 size_show(struct mddev *mddev, char *page)
4672 {
4673 	return sprintf(page, "%llu\n",
4674 		(unsigned long long)mddev->dev_sectors / 2);
4675 }
4676 
4677 static int update_size(struct mddev *mddev, sector_t num_sectors);
4678 
4679 static ssize_t
size_store(struct mddev * mddev,const char * buf,size_t len)4680 size_store(struct mddev *mddev, const char *buf, size_t len)
4681 {
4682 	/* If array is inactive, we can reduce the component size, but
4683 	 * not increase it (except from 0).
4684 	 * If array is active, we can try an on-line resize
4685 	 */
4686 	sector_t sectors;
4687 	int err = strict_blocks_to_sectors(buf, &sectors);
4688 
4689 	if (err < 0)
4690 		return err;
4691 	err = mddev_lock(mddev);
4692 	if (err)
4693 		return err;
4694 	if (mddev->pers) {
4695 		err = update_size(mddev, sectors);
4696 		if (err == 0)
4697 			md_update_sb(mddev, 1);
4698 	} else {
4699 		if (mddev->dev_sectors == 0 ||
4700 		    mddev->dev_sectors > sectors)
4701 			mddev->dev_sectors = sectors;
4702 		else
4703 			err = -ENOSPC;
4704 	}
4705 	mddev_unlock(mddev);
4706 	return err ? err : len;
4707 }
4708 
4709 static struct md_sysfs_entry md_size =
4710 __ATTR(component_size, S_IRUGO|S_IWUSR, size_show, size_store);
4711 
4712 /* Metadata version.
4713  * This is one of
4714  *   'none' for arrays with no metadata (good luck...)
4715  *   'external' for arrays with externally managed metadata,
4716  * or N.M for internally known formats
4717  */
4718 static ssize_t
metadata_show(struct mddev * mddev,char * page)4719 metadata_show(struct mddev *mddev, char *page)
4720 {
4721 	if (mddev->persistent)
4722 		return sprintf(page, "%d.%d\n",
4723 			       mddev->major_version, mddev->minor_version);
4724 	else if (mddev->external)
4725 		return sprintf(page, "external:%s\n", mddev->metadata_type);
4726 	else
4727 		return sprintf(page, "none\n");
4728 }
4729 
4730 static ssize_t
metadata_store(struct mddev * mddev,const char * buf,size_t len)4731 metadata_store(struct mddev *mddev, const char *buf, size_t len)
4732 {
4733 	int major, minor;
4734 	char *e;
4735 	int err;
4736 	/* Changing the details of 'external' metadata is
4737 	 * always permitted.  Otherwise there must be
4738 	 * no devices attached to the array.
4739 	 */
4740 
4741 	err = mddev_lock(mddev);
4742 	if (err)
4743 		return err;
4744 	err = -EBUSY;
4745 	if (mddev->external && strncmp(buf, "external:", 9) == 0)
4746 		;
4747 	else if (!list_empty(&mddev->disks))
4748 		goto out_unlock;
4749 
4750 	err = 0;
4751 	if (cmd_match(buf, "none")) {
4752 		mddev->persistent = 0;
4753 		mddev->external = 0;
4754 		mddev->major_version = 0;
4755 		mddev->minor_version = 90;
4756 		goto out_unlock;
4757 	}
4758 	if (strncmp(buf, "external:", 9) == 0) {
4759 		size_t namelen = len-9;
4760 		if (namelen >= sizeof(mddev->metadata_type))
4761 			namelen = sizeof(mddev->metadata_type)-1;
4762 		strncpy(mddev->metadata_type, buf+9, namelen);
4763 		mddev->metadata_type[namelen] = 0;
4764 		if (namelen && mddev->metadata_type[namelen-1] == '\n')
4765 			mddev->metadata_type[--namelen] = 0;
4766 		mddev->persistent = 0;
4767 		mddev->external = 1;
4768 		mddev->major_version = 0;
4769 		mddev->minor_version = 90;
4770 		goto out_unlock;
4771 	}
4772 	major = simple_strtoul(buf, &e, 10);
4773 	err = -EINVAL;
4774 	if (e==buf || *e != '.')
4775 		goto out_unlock;
4776 	buf = e+1;
4777 	minor = simple_strtoul(buf, &e, 10);
4778 	if (e==buf || (*e && *e != '\n') )
4779 		goto out_unlock;
4780 	err = -ENOENT;
4781 	if (major >= ARRAY_SIZE(super_types) || super_types[major].name == NULL)
4782 		goto out_unlock;
4783 	mddev->major_version = major;
4784 	mddev->minor_version = minor;
4785 	mddev->persistent = 1;
4786 	mddev->external = 0;
4787 	err = 0;
4788 out_unlock:
4789 	mddev_unlock(mddev);
4790 	return err ?: len;
4791 }
4792 
4793 static struct md_sysfs_entry md_metadata =
4794 __ATTR_PREALLOC(metadata_version, S_IRUGO|S_IWUSR, metadata_show, metadata_store);
4795 
4796 static ssize_t
action_show(struct mddev * mddev,char * page)4797 action_show(struct mddev *mddev, char *page)
4798 {
4799 	char *type = "idle";
4800 	unsigned long recovery = mddev->recovery;
4801 	if (test_bit(MD_RECOVERY_FROZEN, &recovery))
4802 		type = "frozen";
4803 	else if (test_bit(MD_RECOVERY_RUNNING, &recovery) ||
4804 	    (md_is_rdwr(mddev) && test_bit(MD_RECOVERY_NEEDED, &recovery))) {
4805 		if (test_bit(MD_RECOVERY_RESHAPE, &recovery))
4806 			type = "reshape";
4807 		else if (test_bit(MD_RECOVERY_SYNC, &recovery)) {
4808 			if (!test_bit(MD_RECOVERY_REQUESTED, &recovery))
4809 				type = "resync";
4810 			else if (test_bit(MD_RECOVERY_CHECK, &recovery))
4811 				type = "check";
4812 			else
4813 				type = "repair";
4814 		} else if (test_bit(MD_RECOVERY_RECOVER, &recovery))
4815 			type = "recover";
4816 		else if (mddev->reshape_position != MaxSector)
4817 			type = "reshape";
4818 	}
4819 	return sprintf(page, "%s\n", type);
4820 }
4821 
stop_sync_thread(struct mddev * mddev)4822 static void stop_sync_thread(struct mddev *mddev)
4823 {
4824 	if (!test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4825 		return;
4826 
4827 	if (mddev_lock(mddev))
4828 		return;
4829 
4830 	/*
4831 	 * Check again in case MD_RECOVERY_RUNNING is cleared before lock is
4832 	 * held.
4833 	 */
4834 	if (!test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) {
4835 		mddev_unlock(mddev);
4836 		return;
4837 	}
4838 
4839 	if (work_pending(&mddev->sync_work))
4840 		flush_workqueue(md_misc_wq);
4841 
4842 	set_bit(MD_RECOVERY_INTR, &mddev->recovery);
4843 	/*
4844 	 * Thread might be blocked waiting for metadata update which will now
4845 	 * never happen
4846 	 */
4847 	md_wakeup_thread_directly(mddev->sync_thread);
4848 
4849 	mddev_unlock(mddev);
4850 }
4851 
idle_sync_thread(struct mddev * mddev)4852 static void idle_sync_thread(struct mddev *mddev)
4853 {
4854 	int sync_seq = atomic_read(&mddev->sync_seq);
4855 
4856 	mutex_lock(&mddev->sync_mutex);
4857 	clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4858 	stop_sync_thread(mddev);
4859 
4860 	wait_event(resync_wait, sync_seq != atomic_read(&mddev->sync_seq) ||
4861 			!test_bit(MD_RECOVERY_RUNNING, &mddev->recovery));
4862 
4863 	mutex_unlock(&mddev->sync_mutex);
4864 }
4865 
frozen_sync_thread(struct mddev * mddev)4866 static void frozen_sync_thread(struct mddev *mddev)
4867 {
4868 	mutex_lock(&mddev->sync_mutex);
4869 	set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4870 	stop_sync_thread(mddev);
4871 
4872 	wait_event(resync_wait, mddev->sync_thread == NULL &&
4873 			!test_bit(MD_RECOVERY_RUNNING, &mddev->recovery));
4874 
4875 	mutex_unlock(&mddev->sync_mutex);
4876 }
4877 
4878 static ssize_t
action_store(struct mddev * mddev,const char * page,size_t len)4879 action_store(struct mddev *mddev, const char *page, size_t len)
4880 {
4881 	if (!mddev->pers || !mddev->pers->sync_request)
4882 		return -EINVAL;
4883 
4884 
4885 	if (cmd_match(page, "idle"))
4886 		idle_sync_thread(mddev);
4887 	else if (cmd_match(page, "frozen"))
4888 		frozen_sync_thread(mddev);
4889 	else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4890 		return -EBUSY;
4891 	else if (cmd_match(page, "resync"))
4892 		clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4893 	else if (cmd_match(page, "recover")) {
4894 		clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4895 		set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
4896 	} else if (cmd_match(page, "reshape")) {
4897 		int err;
4898 		if (mddev->pers->start_reshape == NULL)
4899 			return -EINVAL;
4900 		err = mddev_lock(mddev);
4901 		if (!err) {
4902 			if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) {
4903 				err =  -EBUSY;
4904 			} else if (mddev->reshape_position == MaxSector ||
4905 				   mddev->pers->check_reshape == NULL ||
4906 				   mddev->pers->check_reshape(mddev)) {
4907 				clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4908 				err = mddev->pers->start_reshape(mddev);
4909 			} else {
4910 				/*
4911 				 * If reshape is still in progress, and
4912 				 * md_check_recovery() can continue to reshape,
4913 				 * don't restart reshape because data can be
4914 				 * corrupted for raid456.
4915 				 */
4916 				clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4917 			}
4918 			mddev_unlock(mddev);
4919 		}
4920 		if (err)
4921 			return err;
4922 		sysfs_notify_dirent_safe(mddev->sysfs_degraded);
4923 	} else {
4924 		if (cmd_match(page, "check"))
4925 			set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
4926 		else if (!cmd_match(page, "repair"))
4927 			return -EINVAL;
4928 		clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4929 		set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
4930 		set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
4931 	}
4932 	if (mddev->ro == MD_AUTO_READ) {
4933 		/* A write to sync_action is enough to justify
4934 		 * canceling read-auto mode
4935 		 */
4936 		mddev->ro = MD_RDWR;
4937 		md_wakeup_thread(mddev->sync_thread);
4938 	}
4939 	set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4940 	md_wakeup_thread(mddev->thread);
4941 	sysfs_notify_dirent_safe(mddev->sysfs_action);
4942 	return len;
4943 }
4944 
4945 static struct md_sysfs_entry md_scan_mode =
4946 __ATTR_PREALLOC(sync_action, S_IRUGO|S_IWUSR, action_show, action_store);
4947 
4948 static ssize_t
last_sync_action_show(struct mddev * mddev,char * page)4949 last_sync_action_show(struct mddev *mddev, char *page)
4950 {
4951 	return sprintf(page, "%s\n", mddev->last_sync_action);
4952 }
4953 
4954 static struct md_sysfs_entry md_last_scan_mode = __ATTR_RO(last_sync_action);
4955 
4956 static ssize_t
mismatch_cnt_show(struct mddev * mddev,char * page)4957 mismatch_cnt_show(struct mddev *mddev, char *page)
4958 {
4959 	return sprintf(page, "%llu\n",
4960 		       (unsigned long long)
4961 		       atomic64_read(&mddev->resync_mismatches));
4962 }
4963 
4964 static struct md_sysfs_entry md_mismatches = __ATTR_RO(mismatch_cnt);
4965 
4966 static ssize_t
sync_min_show(struct mddev * mddev,char * page)4967 sync_min_show(struct mddev *mddev, char *page)
4968 {
4969 	return sprintf(page, "%d (%s)\n", speed_min(mddev),
4970 		       mddev->sync_speed_min ? "local": "system");
4971 }
4972 
4973 static ssize_t
sync_min_store(struct mddev * mddev,const char * buf,size_t len)4974 sync_min_store(struct mddev *mddev, const char *buf, size_t len)
4975 {
4976 	unsigned int min;
4977 	int rv;
4978 
4979 	if (strncmp(buf, "system", 6)==0) {
4980 		min = 0;
4981 	} else {
4982 		rv = kstrtouint(buf, 10, &min);
4983 		if (rv < 0)
4984 			return rv;
4985 		if (min == 0)
4986 			return -EINVAL;
4987 	}
4988 	mddev->sync_speed_min = min;
4989 	return len;
4990 }
4991 
4992 static struct md_sysfs_entry md_sync_min =
4993 __ATTR(sync_speed_min, S_IRUGO|S_IWUSR, sync_min_show, sync_min_store);
4994 
4995 static ssize_t
sync_max_show(struct mddev * mddev,char * page)4996 sync_max_show(struct mddev *mddev, char *page)
4997 {
4998 	return sprintf(page, "%d (%s)\n", speed_max(mddev),
4999 		       mddev->sync_speed_max ? "local": "system");
5000 }
5001 
5002 static ssize_t
sync_max_store(struct mddev * mddev,const char * buf,size_t len)5003 sync_max_store(struct mddev *mddev, const char *buf, size_t len)
5004 {
5005 	unsigned int max;
5006 	int rv;
5007 
5008 	if (strncmp(buf, "system", 6)==0) {
5009 		max = 0;
5010 	} else {
5011 		rv = kstrtouint(buf, 10, &max);
5012 		if (rv < 0)
5013 			return rv;
5014 		if (max == 0)
5015 			return -EINVAL;
5016 	}
5017 	mddev->sync_speed_max = max;
5018 	return len;
5019 }
5020 
5021 static struct md_sysfs_entry md_sync_max =
5022 __ATTR(sync_speed_max, S_IRUGO|S_IWUSR, sync_max_show, sync_max_store);
5023 
5024 static ssize_t
degraded_show(struct mddev * mddev,char * page)5025 degraded_show(struct mddev *mddev, char *page)
5026 {
5027 	return sprintf(page, "%d\n", mddev->degraded);
5028 }
5029 static struct md_sysfs_entry md_degraded = __ATTR_RO(degraded);
5030 
5031 static ssize_t
sync_force_parallel_show(struct mddev * mddev,char * page)5032 sync_force_parallel_show(struct mddev *mddev, char *page)
5033 {
5034 	return sprintf(page, "%d\n", mddev->parallel_resync);
5035 }
5036 
5037 static ssize_t
sync_force_parallel_store(struct mddev * mddev,const char * buf,size_t len)5038 sync_force_parallel_store(struct mddev *mddev, const char *buf, size_t len)
5039 {
5040 	long n;
5041 
5042 	if (kstrtol(buf, 10, &n))
5043 		return -EINVAL;
5044 
5045 	if (n != 0 && n != 1)
5046 		return -EINVAL;
5047 
5048 	mddev->parallel_resync = n;
5049 
5050 	if (mddev->sync_thread)
5051 		wake_up(&resync_wait);
5052 
5053 	return len;
5054 }
5055 
5056 /* force parallel resync, even with shared block devices */
5057 static struct md_sysfs_entry md_sync_force_parallel =
5058 __ATTR(sync_force_parallel, S_IRUGO|S_IWUSR,
5059        sync_force_parallel_show, sync_force_parallel_store);
5060 
5061 static ssize_t
sync_speed_show(struct mddev * mddev,char * page)5062 sync_speed_show(struct mddev *mddev, char *page)
5063 {
5064 	unsigned long resync, dt, db;
5065 	if (mddev->curr_resync == MD_RESYNC_NONE)
5066 		return sprintf(page, "none\n");
5067 	resync = mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active);
5068 	dt = (jiffies - mddev->resync_mark) / HZ;
5069 	if (!dt) dt++;
5070 	db = resync - mddev->resync_mark_cnt;
5071 	return sprintf(page, "%lu\n", db/dt/2); /* K/sec */
5072 }
5073 
5074 static struct md_sysfs_entry md_sync_speed = __ATTR_RO(sync_speed);
5075 
5076 static ssize_t
sync_completed_show(struct mddev * mddev,char * page)5077 sync_completed_show(struct mddev *mddev, char *page)
5078 {
5079 	unsigned long long max_sectors, resync;
5080 
5081 	if (!test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
5082 		return sprintf(page, "none\n");
5083 
5084 	if (mddev->curr_resync == MD_RESYNC_YIELDED ||
5085 	    mddev->curr_resync == MD_RESYNC_DELAYED)
5086 		return sprintf(page, "delayed\n");
5087 
5088 	if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ||
5089 	    test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
5090 		max_sectors = mddev->resync_max_sectors;
5091 	else
5092 		max_sectors = mddev->dev_sectors;
5093 
5094 	resync = mddev->curr_resync_completed;
5095 	return sprintf(page, "%llu / %llu\n", resync, max_sectors);
5096 }
5097 
5098 static struct md_sysfs_entry md_sync_completed =
5099 	__ATTR_PREALLOC(sync_completed, S_IRUGO, sync_completed_show, NULL);
5100 
5101 static ssize_t
min_sync_show(struct mddev * mddev,char * page)5102 min_sync_show(struct mddev *mddev, char *page)
5103 {
5104 	return sprintf(page, "%llu\n",
5105 		       (unsigned long long)mddev->resync_min);
5106 }
5107 static ssize_t
min_sync_store(struct mddev * mddev,const char * buf,size_t len)5108 min_sync_store(struct mddev *mddev, const char *buf, size_t len)
5109 {
5110 	unsigned long long min;
5111 	int err;
5112 
5113 	if (kstrtoull(buf, 10, &min))
5114 		return -EINVAL;
5115 
5116 	spin_lock(&mddev->lock);
5117 	err = -EINVAL;
5118 	if (min > mddev->resync_max)
5119 		goto out_unlock;
5120 
5121 	err = -EBUSY;
5122 	if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
5123 		goto out_unlock;
5124 
5125 	/* Round down to multiple of 4K for safety */
5126 	mddev->resync_min = round_down(min, 8);
5127 	err = 0;
5128 
5129 out_unlock:
5130 	spin_unlock(&mddev->lock);
5131 	return err ?: len;
5132 }
5133 
5134 static struct md_sysfs_entry md_min_sync =
5135 __ATTR(sync_min, S_IRUGO|S_IWUSR, min_sync_show, min_sync_store);
5136 
5137 static ssize_t
max_sync_show(struct mddev * mddev,char * page)5138 max_sync_show(struct mddev *mddev, char *page)
5139 {
5140 	if (mddev->resync_max == MaxSector)
5141 		return sprintf(page, "max\n");
5142 	else
5143 		return sprintf(page, "%llu\n",
5144 			       (unsigned long long)mddev->resync_max);
5145 }
5146 static ssize_t
max_sync_store(struct mddev * mddev,const char * buf,size_t len)5147 max_sync_store(struct mddev *mddev, const char *buf, size_t len)
5148 {
5149 	int err;
5150 	spin_lock(&mddev->lock);
5151 	if (strncmp(buf, "max", 3) == 0)
5152 		mddev->resync_max = MaxSector;
5153 	else {
5154 		unsigned long long max;
5155 		int chunk;
5156 
5157 		err = -EINVAL;
5158 		if (kstrtoull(buf, 10, &max))
5159 			goto out_unlock;
5160 		if (max < mddev->resync_min)
5161 			goto out_unlock;
5162 
5163 		err = -EBUSY;
5164 		if (max < mddev->resync_max && md_is_rdwr(mddev) &&
5165 		    test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
5166 			goto out_unlock;
5167 
5168 		/* Must be a multiple of chunk_size */
5169 		chunk = mddev->chunk_sectors;
5170 		if (chunk) {
5171 			sector_t temp = max;
5172 
5173 			err = -EINVAL;
5174 			if (sector_div(temp, chunk))
5175 				goto out_unlock;
5176 		}
5177 		mddev->resync_max = max;
5178 	}
5179 	wake_up(&mddev->recovery_wait);
5180 	err = 0;
5181 out_unlock:
5182 	spin_unlock(&mddev->lock);
5183 	return err ?: len;
5184 }
5185 
5186 static struct md_sysfs_entry md_max_sync =
5187 __ATTR(sync_max, S_IRUGO|S_IWUSR, max_sync_show, max_sync_store);
5188 
5189 static ssize_t
suspend_lo_show(struct mddev * mddev,char * page)5190 suspend_lo_show(struct mddev *mddev, char *page)
5191 {
5192 	return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_lo);
5193 }
5194 
5195 static ssize_t
suspend_lo_store(struct mddev * mddev,const char * buf,size_t len)5196 suspend_lo_store(struct mddev *mddev, const char *buf, size_t len)
5197 {
5198 	unsigned long long new;
5199 	int err;
5200 
5201 	err = kstrtoull(buf, 10, &new);
5202 	if (err < 0)
5203 		return err;
5204 	if (new != (sector_t)new)
5205 		return -EINVAL;
5206 
5207 	err = mddev_lock(mddev);
5208 	if (err)
5209 		return err;
5210 	err = -EINVAL;
5211 	if (mddev->pers == NULL ||
5212 	    mddev->pers->quiesce == NULL)
5213 		goto unlock;
5214 	mddev_suspend(mddev);
5215 	mddev->suspend_lo = new;
5216 	mddev_resume(mddev);
5217 
5218 	err = 0;
5219 unlock:
5220 	mddev_unlock(mddev);
5221 	return err ?: len;
5222 }
5223 static struct md_sysfs_entry md_suspend_lo =
5224 __ATTR(suspend_lo, S_IRUGO|S_IWUSR, suspend_lo_show, suspend_lo_store);
5225 
5226 static ssize_t
suspend_hi_show(struct mddev * mddev,char * page)5227 suspend_hi_show(struct mddev *mddev, char *page)
5228 {
5229 	return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_hi);
5230 }
5231 
5232 static ssize_t
suspend_hi_store(struct mddev * mddev,const char * buf,size_t len)5233 suspend_hi_store(struct mddev *mddev, const char *buf, size_t len)
5234 {
5235 	unsigned long long new;
5236 	int err;
5237 
5238 	err = kstrtoull(buf, 10, &new);
5239 	if (err < 0)
5240 		return err;
5241 	if (new != (sector_t)new)
5242 		return -EINVAL;
5243 
5244 	err = mddev_lock(mddev);
5245 	if (err)
5246 		return err;
5247 	err = -EINVAL;
5248 	if (mddev->pers == NULL)
5249 		goto unlock;
5250 
5251 	mddev_suspend(mddev);
5252 	mddev->suspend_hi = new;
5253 	mddev_resume(mddev);
5254 
5255 	err = 0;
5256 unlock:
5257 	mddev_unlock(mddev);
5258 	return err ?: len;
5259 }
5260 static struct md_sysfs_entry md_suspend_hi =
5261 __ATTR(suspend_hi, S_IRUGO|S_IWUSR, suspend_hi_show, suspend_hi_store);
5262 
5263 static ssize_t
reshape_position_show(struct mddev * mddev,char * page)5264 reshape_position_show(struct mddev *mddev, char *page)
5265 {
5266 	if (mddev->reshape_position != MaxSector)
5267 		return sprintf(page, "%llu\n",
5268 			       (unsigned long long)mddev->reshape_position);
5269 	strcpy(page, "none\n");
5270 	return 5;
5271 }
5272 
5273 static ssize_t
reshape_position_store(struct mddev * mddev,const char * buf,size_t len)5274 reshape_position_store(struct mddev *mddev, const char *buf, size_t len)
5275 {
5276 	struct md_rdev *rdev;
5277 	unsigned long long new;
5278 	int err;
5279 
5280 	err = kstrtoull(buf, 10, &new);
5281 	if (err < 0)
5282 		return err;
5283 	if (new != (sector_t)new)
5284 		return -EINVAL;
5285 	err = mddev_lock(mddev);
5286 	if (err)
5287 		return err;
5288 	err = -EBUSY;
5289 	if (mddev->pers)
5290 		goto unlock;
5291 	mddev->reshape_position = new;
5292 	mddev->delta_disks = 0;
5293 	mddev->reshape_backwards = 0;
5294 	mddev->new_level = mddev->level;
5295 	mddev->new_layout = mddev->layout;
5296 	mddev->new_chunk_sectors = mddev->chunk_sectors;
5297 	rdev_for_each(rdev, mddev)
5298 		rdev->new_data_offset = rdev->data_offset;
5299 	err = 0;
5300 unlock:
5301 	mddev_unlock(mddev);
5302 	return err ?: len;
5303 }
5304 
5305 static struct md_sysfs_entry md_reshape_position =
5306 __ATTR(reshape_position, S_IRUGO|S_IWUSR, reshape_position_show,
5307        reshape_position_store);
5308 
5309 static ssize_t
reshape_direction_show(struct mddev * mddev,char * page)5310 reshape_direction_show(struct mddev *mddev, char *page)
5311 {
5312 	return sprintf(page, "%s\n",
5313 		       mddev->reshape_backwards ? "backwards" : "forwards");
5314 }
5315 
5316 static ssize_t
reshape_direction_store(struct mddev * mddev,const char * buf,size_t len)5317 reshape_direction_store(struct mddev *mddev, const char *buf, size_t len)
5318 {
5319 	int backwards = 0;
5320 	int err;
5321 
5322 	if (cmd_match(buf, "forwards"))
5323 		backwards = 0;
5324 	else if (cmd_match(buf, "backwards"))
5325 		backwards = 1;
5326 	else
5327 		return -EINVAL;
5328 	if (mddev->reshape_backwards == backwards)
5329 		return len;
5330 
5331 	err = mddev_lock(mddev);
5332 	if (err)
5333 		return err;
5334 	/* check if we are allowed to change */
5335 	if (mddev->delta_disks)
5336 		err = -EBUSY;
5337 	else if (mddev->persistent &&
5338 	    mddev->major_version == 0)
5339 		err =  -EINVAL;
5340 	else
5341 		mddev->reshape_backwards = backwards;
5342 	mddev_unlock(mddev);
5343 	return err ?: len;
5344 }
5345 
5346 static struct md_sysfs_entry md_reshape_direction =
5347 __ATTR(reshape_direction, S_IRUGO|S_IWUSR, reshape_direction_show,
5348        reshape_direction_store);
5349 
5350 static ssize_t
array_size_show(struct mddev * mddev,char * page)5351 array_size_show(struct mddev *mddev, char *page)
5352 {
5353 	if (mddev->external_size)
5354 		return sprintf(page, "%llu\n",
5355 			       (unsigned long long)mddev->array_sectors/2);
5356 	else
5357 		return sprintf(page, "default\n");
5358 }
5359 
5360 static ssize_t
array_size_store(struct mddev * mddev,const char * buf,size_t len)5361 array_size_store(struct mddev *mddev, const char *buf, size_t len)
5362 {
5363 	sector_t sectors;
5364 	int err;
5365 
5366 	err = mddev_lock(mddev);
5367 	if (err)
5368 		return err;
5369 
5370 	/* cluster raid doesn't support change array_sectors */
5371 	if (mddev_is_clustered(mddev)) {
5372 		mddev_unlock(mddev);
5373 		return -EINVAL;
5374 	}
5375 
5376 	if (strncmp(buf, "default", 7) == 0) {
5377 		if (mddev->pers)
5378 			sectors = mddev->pers->size(mddev, 0, 0);
5379 		else
5380 			sectors = mddev->array_sectors;
5381 
5382 		mddev->external_size = 0;
5383 	} else {
5384 		if (strict_blocks_to_sectors(buf, &sectors) < 0)
5385 			err = -EINVAL;
5386 		else if (mddev->pers && mddev->pers->size(mddev, 0, 0) < sectors)
5387 			err = -E2BIG;
5388 		else
5389 			mddev->external_size = 1;
5390 	}
5391 
5392 	if (!err) {
5393 		mddev->array_sectors = sectors;
5394 		if (mddev->pers)
5395 			set_capacity_and_notify(mddev->gendisk,
5396 						mddev->array_sectors);
5397 	}
5398 	mddev_unlock(mddev);
5399 	return err ?: len;
5400 }
5401 
5402 static struct md_sysfs_entry md_array_size =
5403 __ATTR(array_size, S_IRUGO|S_IWUSR, array_size_show,
5404        array_size_store);
5405 
5406 static ssize_t
consistency_policy_show(struct mddev * mddev,char * page)5407 consistency_policy_show(struct mddev *mddev, char *page)
5408 {
5409 	int ret;
5410 
5411 	if (test_bit(MD_HAS_JOURNAL, &mddev->flags)) {
5412 		ret = sprintf(page, "journal\n");
5413 	} else if (test_bit(MD_HAS_PPL, &mddev->flags)) {
5414 		ret = sprintf(page, "ppl\n");
5415 	} else if (mddev->bitmap) {
5416 		ret = sprintf(page, "bitmap\n");
5417 	} else if (mddev->pers) {
5418 		if (mddev->pers->sync_request)
5419 			ret = sprintf(page, "resync\n");
5420 		else
5421 			ret = sprintf(page, "none\n");
5422 	} else {
5423 		ret = sprintf(page, "unknown\n");
5424 	}
5425 
5426 	return ret;
5427 }
5428 
5429 static ssize_t
consistency_policy_store(struct mddev * mddev,const char * buf,size_t len)5430 consistency_policy_store(struct mddev *mddev, const char *buf, size_t len)
5431 {
5432 	int err = 0;
5433 
5434 	if (mddev->pers) {
5435 		if (mddev->pers->change_consistency_policy)
5436 			err = mddev->pers->change_consistency_policy(mddev, buf);
5437 		else
5438 			err = -EBUSY;
5439 	} else if (mddev->external && strncmp(buf, "ppl", 3) == 0) {
5440 		set_bit(MD_HAS_PPL, &mddev->flags);
5441 	} else {
5442 		err = -EINVAL;
5443 	}
5444 
5445 	return err ? err : len;
5446 }
5447 
5448 static struct md_sysfs_entry md_consistency_policy =
5449 __ATTR(consistency_policy, S_IRUGO | S_IWUSR, consistency_policy_show,
5450        consistency_policy_store);
5451 
fail_last_dev_show(struct mddev * mddev,char * page)5452 static ssize_t fail_last_dev_show(struct mddev *mddev, char *page)
5453 {
5454 	return sprintf(page, "%d\n", mddev->fail_last_dev);
5455 }
5456 
5457 /*
5458  * Setting fail_last_dev to true to allow last device to be forcibly removed
5459  * from RAID1/RAID10.
5460  */
5461 static ssize_t
fail_last_dev_store(struct mddev * mddev,const char * buf,size_t len)5462 fail_last_dev_store(struct mddev *mddev, const char *buf, size_t len)
5463 {
5464 	int ret;
5465 	bool value;
5466 
5467 	ret = kstrtobool(buf, &value);
5468 	if (ret)
5469 		return ret;
5470 
5471 	if (value != mddev->fail_last_dev)
5472 		mddev->fail_last_dev = value;
5473 
5474 	return len;
5475 }
5476 static struct md_sysfs_entry md_fail_last_dev =
5477 __ATTR(fail_last_dev, S_IRUGO | S_IWUSR, fail_last_dev_show,
5478        fail_last_dev_store);
5479 
serialize_policy_show(struct mddev * mddev,char * page)5480 static ssize_t serialize_policy_show(struct mddev *mddev, char *page)
5481 {
5482 	if (mddev->pers == NULL || (mddev->pers->level != 1))
5483 		return sprintf(page, "n/a\n");
5484 	else
5485 		return sprintf(page, "%d\n", mddev->serialize_policy);
5486 }
5487 
5488 /*
5489  * Setting serialize_policy to true to enforce write IO is not reordered
5490  * for raid1.
5491  */
5492 static ssize_t
serialize_policy_store(struct mddev * mddev,const char * buf,size_t len)5493 serialize_policy_store(struct mddev *mddev, const char *buf, size_t len)
5494 {
5495 	int err;
5496 	bool value;
5497 
5498 	err = kstrtobool(buf, &value);
5499 	if (err)
5500 		return err;
5501 
5502 	if (value == mddev->serialize_policy)
5503 		return len;
5504 
5505 	err = mddev_lock(mddev);
5506 	if (err)
5507 		return err;
5508 	if (mddev->pers == NULL || (mddev->pers->level != 1)) {
5509 		pr_err("md: serialize_policy is only effective for raid1\n");
5510 		err = -EINVAL;
5511 		goto unlock;
5512 	}
5513 
5514 	mddev_suspend(mddev);
5515 	if (value)
5516 		mddev_create_serial_pool(mddev, NULL, true);
5517 	else
5518 		mddev_destroy_serial_pool(mddev, NULL, true);
5519 	mddev->serialize_policy = value;
5520 	mddev_resume(mddev);
5521 unlock:
5522 	mddev_unlock(mddev);
5523 	return err ?: len;
5524 }
5525 
5526 static struct md_sysfs_entry md_serialize_policy =
5527 __ATTR(serialize_policy, S_IRUGO | S_IWUSR, serialize_policy_show,
5528        serialize_policy_store);
5529 
5530 
5531 static struct attribute *md_default_attrs[] = {
5532 	&md_level.attr,
5533 	&md_layout.attr,
5534 	&md_raid_disks.attr,
5535 	&md_uuid.attr,
5536 	&md_chunk_size.attr,
5537 	&md_size.attr,
5538 	&md_resync_start.attr,
5539 	&md_metadata.attr,
5540 	&md_new_device.attr,
5541 	&md_safe_delay.attr,
5542 	&md_array_state.attr,
5543 	&md_reshape_position.attr,
5544 	&md_reshape_direction.attr,
5545 	&md_array_size.attr,
5546 	&max_corr_read_errors.attr,
5547 	&md_consistency_policy.attr,
5548 	&md_fail_last_dev.attr,
5549 	&md_serialize_policy.attr,
5550 	NULL,
5551 };
5552 
5553 static const struct attribute_group md_default_group = {
5554 	.attrs = md_default_attrs,
5555 };
5556 
5557 static struct attribute *md_redundancy_attrs[] = {
5558 	&md_scan_mode.attr,
5559 	&md_last_scan_mode.attr,
5560 	&md_mismatches.attr,
5561 	&md_sync_min.attr,
5562 	&md_sync_max.attr,
5563 	&md_sync_speed.attr,
5564 	&md_sync_force_parallel.attr,
5565 	&md_sync_completed.attr,
5566 	&md_min_sync.attr,
5567 	&md_max_sync.attr,
5568 	&md_suspend_lo.attr,
5569 	&md_suspend_hi.attr,
5570 	&md_bitmap.attr,
5571 	&md_degraded.attr,
5572 	NULL,
5573 };
5574 static const struct attribute_group md_redundancy_group = {
5575 	.name = NULL,
5576 	.attrs = md_redundancy_attrs,
5577 };
5578 
5579 static const struct attribute_group *md_attr_groups[] = {
5580 	&md_default_group,
5581 	&md_bitmap_group,
5582 	NULL,
5583 };
5584 
5585 static ssize_t
md_attr_show(struct kobject * kobj,struct attribute * attr,char * page)5586 md_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
5587 {
5588 	struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
5589 	struct mddev *mddev = container_of(kobj, struct mddev, kobj);
5590 	ssize_t rv;
5591 
5592 	if (!entry->show)
5593 		return -EIO;
5594 	spin_lock(&all_mddevs_lock);
5595 	if (!mddev_get(mddev)) {
5596 		spin_unlock(&all_mddevs_lock);
5597 		return -EBUSY;
5598 	}
5599 	spin_unlock(&all_mddevs_lock);
5600 
5601 	rv = entry->show(mddev, page);
5602 	mddev_put(mddev);
5603 	return rv;
5604 }
5605 
5606 static ssize_t
md_attr_store(struct kobject * kobj,struct attribute * attr,const char * page,size_t length)5607 md_attr_store(struct kobject *kobj, struct attribute *attr,
5608 	      const char *page, size_t length)
5609 {
5610 	struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
5611 	struct mddev *mddev = container_of(kobj, struct mddev, kobj);
5612 	ssize_t rv;
5613 
5614 	if (!entry->store)
5615 		return -EIO;
5616 	if (!capable(CAP_SYS_ADMIN))
5617 		return -EACCES;
5618 	spin_lock(&all_mddevs_lock);
5619 	if (!mddev_get(mddev)) {
5620 		spin_unlock(&all_mddevs_lock);
5621 		return -EBUSY;
5622 	}
5623 	spin_unlock(&all_mddevs_lock);
5624 	rv = entry->store(mddev, page, length);
5625 	mddev_put(mddev);
5626 	return rv;
5627 }
5628 
md_kobj_release(struct kobject * ko)5629 static void md_kobj_release(struct kobject *ko)
5630 {
5631 	struct mddev *mddev = container_of(ko, struct mddev, kobj);
5632 
5633 	if (mddev->sysfs_state)
5634 		sysfs_put(mddev->sysfs_state);
5635 	if (mddev->sysfs_level)
5636 		sysfs_put(mddev->sysfs_level);
5637 
5638 	del_gendisk(mddev->gendisk);
5639 	put_disk(mddev->gendisk);
5640 }
5641 
5642 static const struct sysfs_ops md_sysfs_ops = {
5643 	.show	= md_attr_show,
5644 	.store	= md_attr_store,
5645 };
5646 static const struct kobj_type md_ktype = {
5647 	.release	= md_kobj_release,
5648 	.sysfs_ops	= &md_sysfs_ops,
5649 	.default_groups	= md_attr_groups,
5650 };
5651 
5652 int mdp_major = 0;
5653 
mddev_delayed_delete(struct work_struct * ws)5654 static void mddev_delayed_delete(struct work_struct *ws)
5655 {
5656 	struct mddev *mddev = container_of(ws, struct mddev, del_work);
5657 
5658 	kobject_put(&mddev->kobj);
5659 }
5660 
no_op(struct percpu_ref * r)5661 static void no_op(struct percpu_ref *r) {}
5662 
mddev_init_writes_pending(struct mddev * mddev)5663 int mddev_init_writes_pending(struct mddev *mddev)
5664 {
5665 	if (mddev->writes_pending.percpu_count_ptr)
5666 		return 0;
5667 	if (percpu_ref_init(&mddev->writes_pending, no_op,
5668 			    PERCPU_REF_ALLOW_REINIT, GFP_KERNEL) < 0)
5669 		return -ENOMEM;
5670 	/* We want to start with the refcount at zero */
5671 	percpu_ref_put(&mddev->writes_pending);
5672 	return 0;
5673 }
5674 EXPORT_SYMBOL_GPL(mddev_init_writes_pending);
5675 
md_alloc(dev_t dev,char * name)5676 struct mddev *md_alloc(dev_t dev, char *name)
5677 {
5678 	/*
5679 	 * If dev is zero, name is the name of a device to allocate with
5680 	 * an arbitrary minor number.  It will be "md_???"
5681 	 * If dev is non-zero it must be a device number with a MAJOR of
5682 	 * MD_MAJOR or mdp_major.  In this case, if "name" is NULL, then
5683 	 * the device is being created by opening a node in /dev.
5684 	 * If "name" is not NULL, the device is being created by
5685 	 * writing to /sys/module/md_mod/parameters/new_array.
5686 	 */
5687 	static DEFINE_MUTEX(disks_mutex);
5688 	struct mddev *mddev;
5689 	struct gendisk *disk;
5690 	int partitioned;
5691 	int shift;
5692 	int unit;
5693 	int error ;
5694 
5695 	/*
5696 	 * Wait for any previous instance of this device to be completely
5697 	 * removed (mddev_delayed_delete).
5698 	 */
5699 	flush_workqueue(md_misc_wq);
5700 
5701 	mutex_lock(&disks_mutex);
5702 	mddev = mddev_alloc(dev);
5703 	if (IS_ERR(mddev)) {
5704 		error = PTR_ERR(mddev);
5705 		goto out_unlock;
5706 	}
5707 
5708 	partitioned = (MAJOR(mddev->unit) != MD_MAJOR);
5709 	shift = partitioned ? MdpMinorShift : 0;
5710 	unit = MINOR(mddev->unit) >> shift;
5711 
5712 	if (name && !dev) {
5713 		/* Need to ensure that 'name' is not a duplicate.
5714 		 */
5715 		struct mddev *mddev2;
5716 		spin_lock(&all_mddevs_lock);
5717 
5718 		list_for_each_entry(mddev2, &all_mddevs, all_mddevs)
5719 			if (mddev2->gendisk &&
5720 			    strcmp(mddev2->gendisk->disk_name, name) == 0) {
5721 				spin_unlock(&all_mddevs_lock);
5722 				error = -EEXIST;
5723 				goto out_free_mddev;
5724 			}
5725 		spin_unlock(&all_mddevs_lock);
5726 	}
5727 	if (name && dev)
5728 		/*
5729 		 * Creating /dev/mdNNN via "newarray", so adjust hold_active.
5730 		 */
5731 		mddev->hold_active = UNTIL_STOP;
5732 
5733 	error = -ENOMEM;
5734 	disk = blk_alloc_disk(NUMA_NO_NODE);
5735 	if (!disk)
5736 		goto out_free_mddev;
5737 
5738 	disk->major = MAJOR(mddev->unit);
5739 	disk->first_minor = unit << shift;
5740 	disk->minors = 1 << shift;
5741 	if (name)
5742 		strcpy(disk->disk_name, name);
5743 	else if (partitioned)
5744 		sprintf(disk->disk_name, "md_d%d", unit);
5745 	else
5746 		sprintf(disk->disk_name, "md%d", unit);
5747 	disk->fops = &md_fops;
5748 	disk->private_data = mddev;
5749 
5750 	mddev->queue = disk->queue;
5751 	blk_set_stacking_limits(&mddev->queue->limits);
5752 	blk_queue_write_cache(mddev->queue, true, true);
5753 	disk->events |= DISK_EVENT_MEDIA_CHANGE;
5754 	mddev->gendisk = disk;
5755 	error = add_disk(disk);
5756 	if (error)
5757 		goto out_put_disk;
5758 
5759 	kobject_init(&mddev->kobj, &md_ktype);
5760 	error = kobject_add(&mddev->kobj, &disk_to_dev(disk)->kobj, "%s", "md");
5761 	if (error) {
5762 		/*
5763 		 * The disk is already live at this point.  Clear the hold flag
5764 		 * and let mddev_put take care of the deletion, as it isn't any
5765 		 * different from a normal close on last release now.
5766 		 */
5767 		mddev->hold_active = 0;
5768 		mutex_unlock(&disks_mutex);
5769 		mddev_put(mddev);
5770 		return ERR_PTR(error);
5771 	}
5772 
5773 	kobject_uevent(&mddev->kobj, KOBJ_ADD);
5774 	mddev->sysfs_state = sysfs_get_dirent_safe(mddev->kobj.sd, "array_state");
5775 	mddev->sysfs_level = sysfs_get_dirent_safe(mddev->kobj.sd, "level");
5776 	mutex_unlock(&disks_mutex);
5777 	return mddev;
5778 
5779 out_put_disk:
5780 	put_disk(disk);
5781 out_free_mddev:
5782 	mddev_free(mddev);
5783 out_unlock:
5784 	mutex_unlock(&disks_mutex);
5785 	return ERR_PTR(error);
5786 }
5787 
md_alloc_and_put(dev_t dev,char * name)5788 static int md_alloc_and_put(dev_t dev, char *name)
5789 {
5790 	struct mddev *mddev = md_alloc(dev, name);
5791 
5792 	if (IS_ERR(mddev))
5793 		return PTR_ERR(mddev);
5794 	mddev_put(mddev);
5795 	return 0;
5796 }
5797 
md_probe(dev_t dev)5798 static void md_probe(dev_t dev)
5799 {
5800 	if (MAJOR(dev) == MD_MAJOR && MINOR(dev) >= 512)
5801 		return;
5802 	if (create_on_open)
5803 		md_alloc_and_put(dev, NULL);
5804 }
5805 
add_named_array(const char * val,const struct kernel_param * kp)5806 static int add_named_array(const char *val, const struct kernel_param *kp)
5807 {
5808 	/*
5809 	 * val must be "md_*" or "mdNNN".
5810 	 * For "md_*" we allocate an array with a large free minor number, and
5811 	 * set the name to val.  val must not already be an active name.
5812 	 * For "mdNNN" we allocate an array with the minor number NNN
5813 	 * which must not already be in use.
5814 	 */
5815 	int len = strlen(val);
5816 	char buf[DISK_NAME_LEN];
5817 	unsigned long devnum;
5818 
5819 	while (len && val[len-1] == '\n')
5820 		len--;
5821 	if (len >= DISK_NAME_LEN)
5822 		return -E2BIG;
5823 	strscpy(buf, val, len+1);
5824 	if (strncmp(buf, "md_", 3) == 0)
5825 		return md_alloc_and_put(0, buf);
5826 	if (strncmp(buf, "md", 2) == 0 &&
5827 	    isdigit(buf[2]) &&
5828 	    kstrtoul(buf+2, 10, &devnum) == 0 &&
5829 	    devnum <= MINORMASK)
5830 		return md_alloc_and_put(MKDEV(MD_MAJOR, devnum), NULL);
5831 
5832 	return -EINVAL;
5833 }
5834 
md_safemode_timeout(struct timer_list * t)5835 static void md_safemode_timeout(struct timer_list *t)
5836 {
5837 	struct mddev *mddev = from_timer(mddev, t, safemode_timer);
5838 
5839 	mddev->safemode = 1;
5840 	if (mddev->external)
5841 		sysfs_notify_dirent_safe(mddev->sysfs_state);
5842 
5843 	md_wakeup_thread(mddev->thread);
5844 }
5845 
5846 static int start_dirty_degraded;
active_io_release(struct percpu_ref * ref)5847 static void active_io_release(struct percpu_ref *ref)
5848 {
5849 	struct mddev *mddev = container_of(ref, struct mddev, active_io);
5850 
5851 	wake_up(&mddev->sb_wait);
5852 }
5853 
md_run(struct mddev * mddev)5854 int md_run(struct mddev *mddev)
5855 {
5856 	int err;
5857 	struct md_rdev *rdev;
5858 	struct md_personality *pers;
5859 	bool nowait = true;
5860 
5861 	if (list_empty(&mddev->disks))
5862 		/* cannot run an array with no devices.. */
5863 		return -EINVAL;
5864 
5865 	if (mddev->pers)
5866 		return -EBUSY;
5867 	/* Cannot run until previous stop completes properly */
5868 	if (mddev->sysfs_active)
5869 		return -EBUSY;
5870 
5871 	/*
5872 	 * Analyze all RAID superblock(s)
5873 	 */
5874 	if (!mddev->raid_disks) {
5875 		if (!mddev->persistent)
5876 			return -EINVAL;
5877 		err = analyze_sbs(mddev);
5878 		if (err)
5879 			return -EINVAL;
5880 	}
5881 
5882 	if (mddev->level != LEVEL_NONE)
5883 		request_module("md-level-%d", mddev->level);
5884 	else if (mddev->clevel[0])
5885 		request_module("md-%s", mddev->clevel);
5886 
5887 	/*
5888 	 * Drop all container device buffers, from now on
5889 	 * the only valid external interface is through the md
5890 	 * device.
5891 	 */
5892 	mddev->has_superblocks = false;
5893 	rdev_for_each(rdev, mddev) {
5894 		if (test_bit(Faulty, &rdev->flags))
5895 			continue;
5896 		sync_blockdev(rdev->bdev);
5897 		invalidate_bdev(rdev->bdev);
5898 		if (mddev->ro != MD_RDONLY && rdev_read_only(rdev)) {
5899 			mddev->ro = MD_RDONLY;
5900 			if (mddev->gendisk)
5901 				set_disk_ro(mddev->gendisk, 1);
5902 		}
5903 
5904 		if (rdev->sb_page)
5905 			mddev->has_superblocks = true;
5906 
5907 		/* perform some consistency tests on the device.
5908 		 * We don't want the data to overlap the metadata,
5909 		 * Internal Bitmap issues have been handled elsewhere.
5910 		 */
5911 		if (rdev->meta_bdev) {
5912 			/* Nothing to check */;
5913 		} else if (rdev->data_offset < rdev->sb_start) {
5914 			if (mddev->dev_sectors &&
5915 			    rdev->data_offset + mddev->dev_sectors
5916 			    > rdev->sb_start) {
5917 				pr_warn("md: %s: data overlaps metadata\n",
5918 					mdname(mddev));
5919 				return -EINVAL;
5920 			}
5921 		} else {
5922 			if (rdev->sb_start + rdev->sb_size/512
5923 			    > rdev->data_offset) {
5924 				pr_warn("md: %s: metadata overlaps data\n",
5925 					mdname(mddev));
5926 				return -EINVAL;
5927 			}
5928 		}
5929 		sysfs_notify_dirent_safe(rdev->sysfs_state);
5930 		nowait = nowait && bdev_nowait(rdev->bdev);
5931 	}
5932 
5933 	err = percpu_ref_init(&mddev->active_io, active_io_release,
5934 				PERCPU_REF_ALLOW_REINIT, GFP_KERNEL);
5935 	if (err)
5936 		return err;
5937 
5938 	if (!bioset_initialized(&mddev->bio_set)) {
5939 		err = bioset_init(&mddev->bio_set, BIO_POOL_SIZE, 0, BIOSET_NEED_BVECS);
5940 		if (err)
5941 			goto exit_active_io;
5942 	}
5943 	if (!bioset_initialized(&mddev->sync_set)) {
5944 		err = bioset_init(&mddev->sync_set, BIO_POOL_SIZE, 0, BIOSET_NEED_BVECS);
5945 		if (err)
5946 			goto exit_bio_set;
5947 	}
5948 
5949 	if (!bioset_initialized(&mddev->io_clone_set)) {
5950 		err = bioset_init(&mddev->io_clone_set, BIO_POOL_SIZE,
5951 				  offsetof(struct md_io_clone, bio_clone), 0);
5952 		if (err)
5953 			goto exit_sync_set;
5954 	}
5955 
5956 	spin_lock(&pers_lock);
5957 	pers = find_pers(mddev->level, mddev->clevel);
5958 	if (!pers || !try_module_get(pers->owner)) {
5959 		spin_unlock(&pers_lock);
5960 		if (mddev->level != LEVEL_NONE)
5961 			pr_warn("md: personality for level %d is not loaded!\n",
5962 				mddev->level);
5963 		else
5964 			pr_warn("md: personality for level %s is not loaded!\n",
5965 				mddev->clevel);
5966 		err = -EINVAL;
5967 		goto abort;
5968 	}
5969 	spin_unlock(&pers_lock);
5970 	if (mddev->level != pers->level) {
5971 		mddev->level = pers->level;
5972 		mddev->new_level = pers->level;
5973 	}
5974 	strscpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
5975 
5976 	if (mddev->reshape_position != MaxSector &&
5977 	    pers->start_reshape == NULL) {
5978 		/* This personality cannot handle reshaping... */
5979 		module_put(pers->owner);
5980 		err = -EINVAL;
5981 		goto abort;
5982 	}
5983 
5984 	if (pers->sync_request) {
5985 		/* Warn if this is a potentially silly
5986 		 * configuration.
5987 		 */
5988 		struct md_rdev *rdev2;
5989 		int warned = 0;
5990 
5991 		rdev_for_each(rdev, mddev)
5992 			rdev_for_each(rdev2, mddev) {
5993 				if (rdev < rdev2 &&
5994 				    rdev->bdev->bd_disk ==
5995 				    rdev2->bdev->bd_disk) {
5996 					pr_warn("%s: WARNING: %pg appears to be on the same physical disk as %pg.\n",
5997 						mdname(mddev),
5998 						rdev->bdev,
5999 						rdev2->bdev);
6000 					warned = 1;
6001 				}
6002 			}
6003 
6004 		if (warned)
6005 			pr_warn("True protection against single-disk failure might be compromised.\n");
6006 	}
6007 
6008 	mddev->recovery = 0;
6009 	/* may be over-ridden by personality */
6010 	mddev->resync_max_sectors = mddev->dev_sectors;
6011 
6012 	mddev->ok_start_degraded = start_dirty_degraded;
6013 
6014 	if (start_readonly && md_is_rdwr(mddev))
6015 		mddev->ro = MD_AUTO_READ; /* read-only, but switch on first write */
6016 
6017 	err = pers->run(mddev);
6018 	if (err)
6019 		pr_warn("md: pers->run() failed ...\n");
6020 	else if (pers->size(mddev, 0, 0) < mddev->array_sectors) {
6021 		WARN_ONCE(!mddev->external_size,
6022 			  "%s: default size too small, but 'external_size' not in effect?\n",
6023 			  __func__);
6024 		pr_warn("md: invalid array_size %llu > default size %llu\n",
6025 			(unsigned long long)mddev->array_sectors / 2,
6026 			(unsigned long long)pers->size(mddev, 0, 0) / 2);
6027 		err = -EINVAL;
6028 	}
6029 	if (err == 0 && pers->sync_request &&
6030 	    (mddev->bitmap_info.file || mddev->bitmap_info.offset)) {
6031 		struct bitmap *bitmap;
6032 
6033 		bitmap = md_bitmap_create(mddev, -1);
6034 		if (IS_ERR(bitmap)) {
6035 			err = PTR_ERR(bitmap);
6036 			pr_warn("%s: failed to create bitmap (%d)\n",
6037 				mdname(mddev), err);
6038 		} else
6039 			mddev->bitmap = bitmap;
6040 
6041 	}
6042 	if (err)
6043 		goto bitmap_abort;
6044 
6045 	if (mddev->bitmap_info.max_write_behind > 0) {
6046 		bool create_pool = false;
6047 
6048 		rdev_for_each(rdev, mddev) {
6049 			if (test_bit(WriteMostly, &rdev->flags) &&
6050 			    rdev_init_serial(rdev))
6051 				create_pool = true;
6052 		}
6053 		if (create_pool && mddev->serial_info_pool == NULL) {
6054 			mddev->serial_info_pool =
6055 				mempool_create_kmalloc_pool(NR_SERIAL_INFOS,
6056 						    sizeof(struct serial_info));
6057 			if (!mddev->serial_info_pool) {
6058 				err = -ENOMEM;
6059 				goto bitmap_abort;
6060 			}
6061 		}
6062 	}
6063 
6064 	if (mddev->queue) {
6065 		bool nonrot = true;
6066 
6067 		rdev_for_each(rdev, mddev) {
6068 			if (rdev->raid_disk >= 0 && !bdev_nonrot(rdev->bdev)) {
6069 				nonrot = false;
6070 				break;
6071 			}
6072 		}
6073 		if (mddev->degraded)
6074 			nonrot = false;
6075 		if (nonrot)
6076 			blk_queue_flag_set(QUEUE_FLAG_NONROT, mddev->queue);
6077 		else
6078 			blk_queue_flag_clear(QUEUE_FLAG_NONROT, mddev->queue);
6079 		blk_queue_flag_set(QUEUE_FLAG_IO_STAT, mddev->queue);
6080 
6081 		/* Set the NOWAIT flags if all underlying devices support it */
6082 		if (nowait)
6083 			blk_queue_flag_set(QUEUE_FLAG_NOWAIT, mddev->queue);
6084 	}
6085 	if (pers->sync_request) {
6086 		if (mddev->kobj.sd &&
6087 		    sysfs_create_group(&mddev->kobj, &md_redundancy_group))
6088 			pr_warn("md: cannot register extra attributes for %s\n",
6089 				mdname(mddev));
6090 		mddev->sysfs_action = sysfs_get_dirent_safe(mddev->kobj.sd, "sync_action");
6091 		mddev->sysfs_completed = sysfs_get_dirent_safe(mddev->kobj.sd, "sync_completed");
6092 		mddev->sysfs_degraded = sysfs_get_dirent_safe(mddev->kobj.sd, "degraded");
6093 	} else if (mddev->ro == MD_AUTO_READ)
6094 		mddev->ro = MD_RDWR;
6095 
6096 	atomic_set(&mddev->max_corr_read_errors,
6097 		   MD_DEFAULT_MAX_CORRECTED_READ_ERRORS);
6098 	mddev->safemode = 0;
6099 	if (mddev_is_clustered(mddev))
6100 		mddev->safemode_delay = 0;
6101 	else
6102 		mddev->safemode_delay = DEFAULT_SAFEMODE_DELAY;
6103 	mddev->in_sync = 1;
6104 	smp_wmb();
6105 	spin_lock(&mddev->lock);
6106 	mddev->pers = pers;
6107 	spin_unlock(&mddev->lock);
6108 	rdev_for_each(rdev, mddev)
6109 		if (rdev->raid_disk >= 0)
6110 			sysfs_link_rdev(mddev, rdev); /* failure here is OK */
6111 
6112 	if (mddev->degraded && md_is_rdwr(mddev))
6113 		/* This ensures that recovering status is reported immediately
6114 		 * via sysfs - until a lack of spares is confirmed.
6115 		 */
6116 		set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
6117 	set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6118 
6119 	if (mddev->sb_flags)
6120 		md_update_sb(mddev, 0);
6121 
6122 	md_new_event();
6123 	return 0;
6124 
6125 bitmap_abort:
6126 	mddev_detach(mddev);
6127 	if (mddev->private)
6128 		pers->free(mddev, mddev->private);
6129 	mddev->private = NULL;
6130 	module_put(pers->owner);
6131 	md_bitmap_destroy(mddev);
6132 abort:
6133 	bioset_exit(&mddev->io_clone_set);
6134 exit_sync_set:
6135 	bioset_exit(&mddev->sync_set);
6136 exit_bio_set:
6137 	bioset_exit(&mddev->bio_set);
6138 exit_active_io:
6139 	percpu_ref_exit(&mddev->active_io);
6140 	return err;
6141 }
6142 EXPORT_SYMBOL_GPL(md_run);
6143 
do_md_run(struct mddev * mddev)6144 int do_md_run(struct mddev *mddev)
6145 {
6146 	int err;
6147 
6148 	set_bit(MD_NOT_READY, &mddev->flags);
6149 	err = md_run(mddev);
6150 	if (err)
6151 		goto out;
6152 	err = md_bitmap_load(mddev);
6153 	if (err) {
6154 		md_bitmap_destroy(mddev);
6155 		goto out;
6156 	}
6157 
6158 	if (mddev_is_clustered(mddev))
6159 		md_allow_write(mddev);
6160 
6161 	/* run start up tasks that require md_thread */
6162 	md_start(mddev);
6163 
6164 	md_wakeup_thread(mddev->thread);
6165 	md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
6166 
6167 	set_capacity_and_notify(mddev->gendisk, mddev->array_sectors);
6168 	clear_bit(MD_NOT_READY, &mddev->flags);
6169 	mddev->changed = 1;
6170 	kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
6171 	sysfs_notify_dirent_safe(mddev->sysfs_state);
6172 	sysfs_notify_dirent_safe(mddev->sysfs_action);
6173 	sysfs_notify_dirent_safe(mddev->sysfs_degraded);
6174 out:
6175 	clear_bit(MD_NOT_READY, &mddev->flags);
6176 	return err;
6177 }
6178 
md_start(struct mddev * mddev)6179 int md_start(struct mddev *mddev)
6180 {
6181 	int ret = 0;
6182 
6183 	if (mddev->pers->start) {
6184 		set_bit(MD_RECOVERY_WAIT, &mddev->recovery);
6185 		md_wakeup_thread(mddev->thread);
6186 		ret = mddev->pers->start(mddev);
6187 		clear_bit(MD_RECOVERY_WAIT, &mddev->recovery);
6188 		md_wakeup_thread(mddev->sync_thread);
6189 	}
6190 	return ret;
6191 }
6192 EXPORT_SYMBOL_GPL(md_start);
6193 
restart_array(struct mddev * mddev)6194 static int restart_array(struct mddev *mddev)
6195 {
6196 	struct gendisk *disk = mddev->gendisk;
6197 	struct md_rdev *rdev;
6198 	bool has_journal = false;
6199 	bool has_readonly = false;
6200 
6201 	/* Complain if it has no devices */
6202 	if (list_empty(&mddev->disks))
6203 		return -ENXIO;
6204 	if (!mddev->pers)
6205 		return -EINVAL;
6206 	if (md_is_rdwr(mddev))
6207 		return -EBUSY;
6208 
6209 	rcu_read_lock();
6210 	rdev_for_each_rcu(rdev, mddev) {
6211 		if (test_bit(Journal, &rdev->flags) &&
6212 		    !test_bit(Faulty, &rdev->flags))
6213 			has_journal = true;
6214 		if (rdev_read_only(rdev))
6215 			has_readonly = true;
6216 	}
6217 	rcu_read_unlock();
6218 	if (test_bit(MD_HAS_JOURNAL, &mddev->flags) && !has_journal)
6219 		/* Don't restart rw with journal missing/faulty */
6220 			return -EINVAL;
6221 	if (has_readonly)
6222 		return -EROFS;
6223 
6224 	mddev->safemode = 0;
6225 	mddev->ro = MD_RDWR;
6226 	set_disk_ro(disk, 0);
6227 	pr_debug("md: %s switched to read-write mode.\n", mdname(mddev));
6228 	/* Kick recovery or resync if necessary */
6229 	set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6230 	md_wakeup_thread(mddev->thread);
6231 	md_wakeup_thread(mddev->sync_thread);
6232 	sysfs_notify_dirent_safe(mddev->sysfs_state);
6233 	return 0;
6234 }
6235 
md_clean(struct mddev * mddev)6236 static void md_clean(struct mddev *mddev)
6237 {
6238 	mddev->array_sectors = 0;
6239 	mddev->external_size = 0;
6240 	mddev->dev_sectors = 0;
6241 	mddev->raid_disks = 0;
6242 	mddev->recovery_cp = 0;
6243 	mddev->resync_min = 0;
6244 	mddev->resync_max = MaxSector;
6245 	mddev->reshape_position = MaxSector;
6246 	/* we still need mddev->external in export_rdev, do not clear it yet */
6247 	mddev->persistent = 0;
6248 	mddev->level = LEVEL_NONE;
6249 	mddev->clevel[0] = 0;
6250 	/*
6251 	 * Don't clear MD_CLOSING, or mddev can be opened again.
6252 	 * 'hold_active != 0' means mddev is still in the creation
6253 	 * process and will be used later.
6254 	 */
6255 	if (mddev->hold_active)
6256 		mddev->flags = 0;
6257 	else
6258 		mddev->flags &= BIT_ULL_MASK(MD_CLOSING);
6259 	mddev->sb_flags = 0;
6260 	mddev->ro = MD_RDWR;
6261 	mddev->metadata_type[0] = 0;
6262 	mddev->chunk_sectors = 0;
6263 	mddev->ctime = mddev->utime = 0;
6264 	mddev->layout = 0;
6265 	mddev->max_disks = 0;
6266 	mddev->events = 0;
6267 	mddev->can_decrease_events = 0;
6268 	mddev->delta_disks = 0;
6269 	mddev->reshape_backwards = 0;
6270 	mddev->new_level = LEVEL_NONE;
6271 	mddev->new_layout = 0;
6272 	mddev->new_chunk_sectors = 0;
6273 	mddev->curr_resync = MD_RESYNC_NONE;
6274 	atomic64_set(&mddev->resync_mismatches, 0);
6275 	mddev->suspend_lo = mddev->suspend_hi = 0;
6276 	mddev->sync_speed_min = mddev->sync_speed_max = 0;
6277 	mddev->recovery = 0;
6278 	mddev->in_sync = 0;
6279 	mddev->changed = 0;
6280 	mddev->degraded = 0;
6281 	mddev->safemode = 0;
6282 	mddev->private = NULL;
6283 	mddev->cluster_info = NULL;
6284 	mddev->bitmap_info.offset = 0;
6285 	mddev->bitmap_info.default_offset = 0;
6286 	mddev->bitmap_info.default_space = 0;
6287 	mddev->bitmap_info.chunksize = 0;
6288 	mddev->bitmap_info.daemon_sleep = 0;
6289 	mddev->bitmap_info.max_write_behind = 0;
6290 	mddev->bitmap_info.nodes = 0;
6291 }
6292 
__md_stop_writes(struct mddev * mddev)6293 static void __md_stop_writes(struct mddev *mddev)
6294 {
6295 	set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
6296 	if (work_pending(&mddev->sync_work))
6297 		flush_workqueue(md_misc_wq);
6298 	if (mddev->sync_thread) {
6299 		set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6300 		md_reap_sync_thread(mddev);
6301 	}
6302 
6303 	del_timer_sync(&mddev->safemode_timer);
6304 
6305 	if (mddev->pers && mddev->pers->quiesce) {
6306 		mddev->pers->quiesce(mddev, 1);
6307 		mddev->pers->quiesce(mddev, 0);
6308 	}
6309 	md_bitmap_flush(mddev);
6310 
6311 	if (md_is_rdwr(mddev) &&
6312 	    ((!mddev->in_sync && !mddev_is_clustered(mddev)) ||
6313 	     mddev->sb_flags)) {
6314 		/* mark array as shutdown cleanly */
6315 		if (!mddev_is_clustered(mddev))
6316 			mddev->in_sync = 1;
6317 		md_update_sb(mddev, 1);
6318 	}
6319 	/* disable policy to guarantee rdevs free resources for serialization */
6320 	mddev->serialize_policy = 0;
6321 	mddev_destroy_serial_pool(mddev, NULL, true);
6322 }
6323 
md_stop_writes(struct mddev * mddev)6324 void md_stop_writes(struct mddev *mddev)
6325 {
6326 	mddev_lock_nointr(mddev);
6327 	__md_stop_writes(mddev);
6328 	mddev_unlock(mddev);
6329 }
6330 EXPORT_SYMBOL_GPL(md_stop_writes);
6331 
mddev_detach(struct mddev * mddev)6332 static void mddev_detach(struct mddev *mddev)
6333 {
6334 	md_bitmap_wait_behind_writes(mddev);
6335 	if (mddev->pers && mddev->pers->quiesce && !is_md_suspended(mddev)) {
6336 		mddev->pers->quiesce(mddev, 1);
6337 		mddev->pers->quiesce(mddev, 0);
6338 	}
6339 	md_unregister_thread(mddev, &mddev->thread);
6340 	if (mddev->queue)
6341 		blk_sync_queue(mddev->queue); /* the unplug fn references 'conf'*/
6342 }
6343 
__md_stop(struct mddev * mddev)6344 static void __md_stop(struct mddev *mddev)
6345 {
6346 	struct md_personality *pers = mddev->pers;
6347 	md_bitmap_destroy(mddev);
6348 	mddev_detach(mddev);
6349 	/* Ensure ->event_work is done */
6350 	if (mddev->event_work.func)
6351 		flush_workqueue(md_misc_wq);
6352 	spin_lock(&mddev->lock);
6353 	mddev->pers = NULL;
6354 	spin_unlock(&mddev->lock);
6355 	if (mddev->private)
6356 		pers->free(mddev, mddev->private);
6357 	mddev->private = NULL;
6358 	if (pers->sync_request && mddev->to_remove == NULL)
6359 		mddev->to_remove = &md_redundancy_group;
6360 	module_put(pers->owner);
6361 	clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
6362 
6363 	percpu_ref_exit(&mddev->active_io);
6364 	bioset_exit(&mddev->bio_set);
6365 	bioset_exit(&mddev->sync_set);
6366 	bioset_exit(&mddev->io_clone_set);
6367 }
6368 
md_stop(struct mddev * mddev)6369 void md_stop(struct mddev *mddev)
6370 {
6371 	lockdep_assert_held(&mddev->reconfig_mutex);
6372 
6373 	/* stop the array and free an attached data structures.
6374 	 * This is called from dm-raid
6375 	 */
6376 	__md_stop_writes(mddev);
6377 	__md_stop(mddev);
6378 	percpu_ref_exit(&mddev->writes_pending);
6379 }
6380 
6381 EXPORT_SYMBOL_GPL(md_stop);
6382 
md_set_readonly(struct mddev * mddev,struct block_device * bdev)6383 static int md_set_readonly(struct mddev *mddev, struct block_device *bdev)
6384 {
6385 	int err = 0;
6386 	int did_freeze = 0;
6387 
6388 	if (mddev->external && test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags))
6389 		return -EBUSY;
6390 
6391 	if (!test_bit(MD_RECOVERY_FROZEN, &mddev->recovery)) {
6392 		did_freeze = 1;
6393 		set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
6394 		md_wakeup_thread(mddev->thread);
6395 	}
6396 	if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
6397 		set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6398 
6399 	/*
6400 	 * Thread might be blocked waiting for metadata update which will now
6401 	 * never happen
6402 	 */
6403 	md_wakeup_thread_directly(mddev->sync_thread);
6404 
6405 	mddev_unlock(mddev);
6406 	wait_event(resync_wait, !test_bit(MD_RECOVERY_RUNNING,
6407 					  &mddev->recovery));
6408 	wait_event(mddev->sb_wait,
6409 		   !test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags));
6410 	mddev_lock_nointr(mddev);
6411 
6412 	mutex_lock(&mddev->open_mutex);
6413 	if ((mddev->pers && atomic_read(&mddev->openers) > !!bdev) ||
6414 	    mddev->sync_thread ||
6415 	    test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) {
6416 		pr_warn("md: %s still in use.\n",mdname(mddev));
6417 		err = -EBUSY;
6418 		goto out;
6419 	}
6420 
6421 	if (mddev->pers) {
6422 		__md_stop_writes(mddev);
6423 
6424 		if (mddev->ro == MD_RDONLY) {
6425 			err  = -ENXIO;
6426 			goto out;
6427 		}
6428 
6429 		mddev->ro = MD_RDONLY;
6430 		set_disk_ro(mddev->gendisk, 1);
6431 	}
6432 
6433 out:
6434 	if ((mddev->pers && !err) || did_freeze) {
6435 		clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
6436 		set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6437 		md_wakeup_thread(mddev->thread);
6438 		sysfs_notify_dirent_safe(mddev->sysfs_state);
6439 	}
6440 
6441 	mutex_unlock(&mddev->open_mutex);
6442 	return err;
6443 }
6444 
6445 /* mode:
6446  *   0 - completely stop and dis-assemble array
6447  *   2 - stop but do not disassemble array
6448  */
do_md_stop(struct mddev * mddev,int mode,struct block_device * bdev)6449 static int do_md_stop(struct mddev *mddev, int mode,
6450 		      struct block_device *bdev)
6451 {
6452 	struct gendisk *disk = mddev->gendisk;
6453 	struct md_rdev *rdev;
6454 	int did_freeze = 0;
6455 
6456 	if (!test_bit(MD_RECOVERY_FROZEN, &mddev->recovery)) {
6457 		did_freeze = 1;
6458 		set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
6459 		md_wakeup_thread(mddev->thread);
6460 	}
6461 	if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
6462 		set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6463 
6464 	/*
6465 	 * Thread might be blocked waiting for metadata update which will now
6466 	 * never happen
6467 	 */
6468 	md_wakeup_thread_directly(mddev->sync_thread);
6469 
6470 	mddev_unlock(mddev);
6471 	wait_event(resync_wait, (mddev->sync_thread == NULL &&
6472 				 !test_bit(MD_RECOVERY_RUNNING,
6473 					   &mddev->recovery)));
6474 	mddev_lock_nointr(mddev);
6475 
6476 	mutex_lock(&mddev->open_mutex);
6477 	if ((mddev->pers && atomic_read(&mddev->openers) > !!bdev) ||
6478 	    mddev->sysfs_active ||
6479 	    mddev->sync_thread ||
6480 	    test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) {
6481 		pr_warn("md: %s still in use.\n",mdname(mddev));
6482 		mutex_unlock(&mddev->open_mutex);
6483 		if (did_freeze) {
6484 			clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
6485 			set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6486 			md_wakeup_thread(mddev->thread);
6487 		}
6488 		return -EBUSY;
6489 	}
6490 	if (mddev->pers) {
6491 		if (!md_is_rdwr(mddev))
6492 			set_disk_ro(disk, 0);
6493 
6494 		__md_stop_writes(mddev);
6495 		__md_stop(mddev);
6496 
6497 		/* tell userspace to handle 'inactive' */
6498 		sysfs_notify_dirent_safe(mddev->sysfs_state);
6499 
6500 		rdev_for_each(rdev, mddev)
6501 			if (rdev->raid_disk >= 0)
6502 				sysfs_unlink_rdev(mddev, rdev);
6503 
6504 		set_capacity_and_notify(disk, 0);
6505 		mutex_unlock(&mddev->open_mutex);
6506 		mddev->changed = 1;
6507 
6508 		if (!md_is_rdwr(mddev))
6509 			mddev->ro = MD_RDWR;
6510 	} else
6511 		mutex_unlock(&mddev->open_mutex);
6512 	/*
6513 	 * Free resources if final stop
6514 	 */
6515 	if (mode == 0) {
6516 		pr_info("md: %s stopped.\n", mdname(mddev));
6517 
6518 		if (mddev->bitmap_info.file) {
6519 			struct file *f = mddev->bitmap_info.file;
6520 			spin_lock(&mddev->lock);
6521 			mddev->bitmap_info.file = NULL;
6522 			spin_unlock(&mddev->lock);
6523 			fput(f);
6524 		}
6525 		mddev->bitmap_info.offset = 0;
6526 
6527 		export_array(mddev);
6528 
6529 		md_clean(mddev);
6530 		if (mddev->hold_active == UNTIL_STOP)
6531 			mddev->hold_active = 0;
6532 	}
6533 	md_new_event();
6534 	sysfs_notify_dirent_safe(mddev->sysfs_state);
6535 	return 0;
6536 }
6537 
6538 #ifndef MODULE
autorun_array(struct mddev * mddev)6539 static void autorun_array(struct mddev *mddev)
6540 {
6541 	struct md_rdev *rdev;
6542 	int err;
6543 
6544 	if (list_empty(&mddev->disks))
6545 		return;
6546 
6547 	pr_info("md: running: ");
6548 
6549 	rdev_for_each(rdev, mddev) {
6550 		pr_cont("<%pg>", rdev->bdev);
6551 	}
6552 	pr_cont("\n");
6553 
6554 	err = do_md_run(mddev);
6555 	if (err) {
6556 		pr_warn("md: do_md_run() returned %d\n", err);
6557 		do_md_stop(mddev, 0, NULL);
6558 	}
6559 }
6560 
6561 /*
6562  * lets try to run arrays based on all disks that have arrived
6563  * until now. (those are in pending_raid_disks)
6564  *
6565  * the method: pick the first pending disk, collect all disks with
6566  * the same UUID, remove all from the pending list and put them into
6567  * the 'same_array' list. Then order this list based on superblock
6568  * update time (freshest comes first), kick out 'old' disks and
6569  * compare superblocks. If everything's fine then run it.
6570  *
6571  * If "unit" is allocated, then bump its reference count
6572  */
autorun_devices(int part)6573 static void autorun_devices(int part)
6574 {
6575 	struct md_rdev *rdev0, *rdev, *tmp;
6576 	struct mddev *mddev;
6577 
6578 	pr_info("md: autorun ...\n");
6579 	while (!list_empty(&pending_raid_disks)) {
6580 		int unit;
6581 		dev_t dev;
6582 		LIST_HEAD(candidates);
6583 		rdev0 = list_entry(pending_raid_disks.next,
6584 					 struct md_rdev, same_set);
6585 
6586 		pr_debug("md: considering %pg ...\n", rdev0->bdev);
6587 		INIT_LIST_HEAD(&candidates);
6588 		rdev_for_each_list(rdev, tmp, &pending_raid_disks)
6589 			if (super_90_load(rdev, rdev0, 0) >= 0) {
6590 				pr_debug("md:  adding %pg ...\n",
6591 					 rdev->bdev);
6592 				list_move(&rdev->same_set, &candidates);
6593 			}
6594 		/*
6595 		 * now we have a set of devices, with all of them having
6596 		 * mostly sane superblocks. It's time to allocate the
6597 		 * mddev.
6598 		 */
6599 		if (part) {
6600 			dev = MKDEV(mdp_major,
6601 				    rdev0->preferred_minor << MdpMinorShift);
6602 			unit = MINOR(dev) >> MdpMinorShift;
6603 		} else {
6604 			dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
6605 			unit = MINOR(dev);
6606 		}
6607 		if (rdev0->preferred_minor != unit) {
6608 			pr_warn("md: unit number in %pg is bad: %d\n",
6609 				rdev0->bdev, rdev0->preferred_minor);
6610 			break;
6611 		}
6612 
6613 		mddev = md_alloc(dev, NULL);
6614 		if (IS_ERR(mddev))
6615 			break;
6616 
6617 		if (mddev_lock(mddev))
6618 			pr_warn("md: %s locked, cannot run\n", mdname(mddev));
6619 		else if (mddev->raid_disks || mddev->major_version
6620 			 || !list_empty(&mddev->disks)) {
6621 			pr_warn("md: %s already running, cannot run %pg\n",
6622 				mdname(mddev), rdev0->bdev);
6623 			mddev_unlock(mddev);
6624 		} else {
6625 			pr_debug("md: created %s\n", mdname(mddev));
6626 			mddev->persistent = 1;
6627 			rdev_for_each_list(rdev, tmp, &candidates) {
6628 				list_del_init(&rdev->same_set);
6629 				if (bind_rdev_to_array(rdev, mddev))
6630 					export_rdev(rdev, mddev);
6631 			}
6632 			autorun_array(mddev);
6633 			mddev_unlock(mddev);
6634 		}
6635 		/* on success, candidates will be empty, on error
6636 		 * it won't...
6637 		 */
6638 		rdev_for_each_list(rdev, tmp, &candidates) {
6639 			list_del_init(&rdev->same_set);
6640 			export_rdev(rdev, mddev);
6641 		}
6642 		mddev_put(mddev);
6643 	}
6644 	pr_info("md: ... autorun DONE.\n");
6645 }
6646 #endif /* !MODULE */
6647 
get_version(void __user * arg)6648 static int get_version(void __user *arg)
6649 {
6650 	mdu_version_t ver;
6651 
6652 	ver.major = MD_MAJOR_VERSION;
6653 	ver.minor = MD_MINOR_VERSION;
6654 	ver.patchlevel = MD_PATCHLEVEL_VERSION;
6655 
6656 	if (copy_to_user(arg, &ver, sizeof(ver)))
6657 		return -EFAULT;
6658 
6659 	return 0;
6660 }
6661 
get_array_info(struct mddev * mddev,void __user * arg)6662 static int get_array_info(struct mddev *mddev, void __user *arg)
6663 {
6664 	mdu_array_info_t info;
6665 	int nr,working,insync,failed,spare;
6666 	struct md_rdev *rdev;
6667 
6668 	nr = working = insync = failed = spare = 0;
6669 	rcu_read_lock();
6670 	rdev_for_each_rcu(rdev, mddev) {
6671 		nr++;
6672 		if (test_bit(Faulty, &rdev->flags))
6673 			failed++;
6674 		else {
6675 			working++;
6676 			if (test_bit(In_sync, &rdev->flags))
6677 				insync++;
6678 			else if (test_bit(Journal, &rdev->flags))
6679 				/* TODO: add journal count to md_u.h */
6680 				;
6681 			else
6682 				spare++;
6683 		}
6684 	}
6685 	rcu_read_unlock();
6686 
6687 	info.major_version = mddev->major_version;
6688 	info.minor_version = mddev->minor_version;
6689 	info.patch_version = MD_PATCHLEVEL_VERSION;
6690 	info.ctime         = clamp_t(time64_t, mddev->ctime, 0, U32_MAX);
6691 	info.level         = mddev->level;
6692 	info.size          = mddev->dev_sectors / 2;
6693 	if (info.size != mddev->dev_sectors / 2) /* overflow */
6694 		info.size = -1;
6695 	info.nr_disks      = nr;
6696 	info.raid_disks    = mddev->raid_disks;
6697 	info.md_minor      = mddev->md_minor;
6698 	info.not_persistent= !mddev->persistent;
6699 
6700 	info.utime         = clamp_t(time64_t, mddev->utime, 0, U32_MAX);
6701 	info.state         = 0;
6702 	if (mddev->in_sync)
6703 		info.state = (1<<MD_SB_CLEAN);
6704 	if (mddev->bitmap && mddev->bitmap_info.offset)
6705 		info.state |= (1<<MD_SB_BITMAP_PRESENT);
6706 	if (mddev_is_clustered(mddev))
6707 		info.state |= (1<<MD_SB_CLUSTERED);
6708 	info.active_disks  = insync;
6709 	info.working_disks = working;
6710 	info.failed_disks  = failed;
6711 	info.spare_disks   = spare;
6712 
6713 	info.layout        = mddev->layout;
6714 	info.chunk_size    = mddev->chunk_sectors << 9;
6715 
6716 	if (copy_to_user(arg, &info, sizeof(info)))
6717 		return -EFAULT;
6718 
6719 	return 0;
6720 }
6721 
get_bitmap_file(struct mddev * mddev,void __user * arg)6722 static int get_bitmap_file(struct mddev *mddev, void __user * arg)
6723 {
6724 	mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */
6725 	char *ptr;
6726 	int err;
6727 
6728 	file = kzalloc(sizeof(*file), GFP_NOIO);
6729 	if (!file)
6730 		return -ENOMEM;
6731 
6732 	err = 0;
6733 	spin_lock(&mddev->lock);
6734 	/* bitmap enabled */
6735 	if (mddev->bitmap_info.file) {
6736 		ptr = file_path(mddev->bitmap_info.file, file->pathname,
6737 				sizeof(file->pathname));
6738 		if (IS_ERR(ptr))
6739 			err = PTR_ERR(ptr);
6740 		else
6741 			memmove(file->pathname, ptr,
6742 				sizeof(file->pathname)-(ptr-file->pathname));
6743 	}
6744 	spin_unlock(&mddev->lock);
6745 
6746 	if (err == 0 &&
6747 	    copy_to_user(arg, file, sizeof(*file)))
6748 		err = -EFAULT;
6749 
6750 	kfree(file);
6751 	return err;
6752 }
6753 
get_disk_info(struct mddev * mddev,void __user * arg)6754 static int get_disk_info(struct mddev *mddev, void __user * arg)
6755 {
6756 	mdu_disk_info_t info;
6757 	struct md_rdev *rdev;
6758 
6759 	if (copy_from_user(&info, arg, sizeof(info)))
6760 		return -EFAULT;
6761 
6762 	rcu_read_lock();
6763 	rdev = md_find_rdev_nr_rcu(mddev, info.number);
6764 	if (rdev) {
6765 		info.major = MAJOR(rdev->bdev->bd_dev);
6766 		info.minor = MINOR(rdev->bdev->bd_dev);
6767 		info.raid_disk = rdev->raid_disk;
6768 		info.state = 0;
6769 		if (test_bit(Faulty, &rdev->flags))
6770 			info.state |= (1<<MD_DISK_FAULTY);
6771 		else if (test_bit(In_sync, &rdev->flags)) {
6772 			info.state |= (1<<MD_DISK_ACTIVE);
6773 			info.state |= (1<<MD_DISK_SYNC);
6774 		}
6775 		if (test_bit(Journal, &rdev->flags))
6776 			info.state |= (1<<MD_DISK_JOURNAL);
6777 		if (test_bit(WriteMostly, &rdev->flags))
6778 			info.state |= (1<<MD_DISK_WRITEMOSTLY);
6779 		if (test_bit(FailFast, &rdev->flags))
6780 			info.state |= (1<<MD_DISK_FAILFAST);
6781 	} else {
6782 		info.major = info.minor = 0;
6783 		info.raid_disk = -1;
6784 		info.state = (1<<MD_DISK_REMOVED);
6785 	}
6786 	rcu_read_unlock();
6787 
6788 	if (copy_to_user(arg, &info, sizeof(info)))
6789 		return -EFAULT;
6790 
6791 	return 0;
6792 }
6793 
md_add_new_disk(struct mddev * mddev,struct mdu_disk_info_s * info)6794 int md_add_new_disk(struct mddev *mddev, struct mdu_disk_info_s *info)
6795 {
6796 	struct md_rdev *rdev;
6797 	dev_t dev = MKDEV(info->major,info->minor);
6798 
6799 	if (mddev_is_clustered(mddev) &&
6800 		!(info->state & ((1 << MD_DISK_CLUSTER_ADD) | (1 << MD_DISK_CANDIDATE)))) {
6801 		pr_warn("%s: Cannot add to clustered mddev.\n",
6802 			mdname(mddev));
6803 		return -EINVAL;
6804 	}
6805 
6806 	if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
6807 		return -EOVERFLOW;
6808 
6809 	if (!mddev->raid_disks) {
6810 		int err;
6811 		/* expecting a device which has a superblock */
6812 		rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
6813 		if (IS_ERR(rdev)) {
6814 			pr_warn("md: md_import_device returned %ld\n",
6815 				PTR_ERR(rdev));
6816 			return PTR_ERR(rdev);
6817 		}
6818 		if (!list_empty(&mddev->disks)) {
6819 			struct md_rdev *rdev0
6820 				= list_entry(mddev->disks.next,
6821 					     struct md_rdev, same_set);
6822 			err = super_types[mddev->major_version]
6823 				.load_super(rdev, rdev0, mddev->minor_version);
6824 			if (err < 0) {
6825 				pr_warn("md: %pg has different UUID to %pg\n",
6826 					rdev->bdev,
6827 					rdev0->bdev);
6828 				export_rdev(rdev, mddev);
6829 				return -EINVAL;
6830 			}
6831 		}
6832 		err = bind_rdev_to_array(rdev, mddev);
6833 		if (err)
6834 			export_rdev(rdev, mddev);
6835 		return err;
6836 	}
6837 
6838 	/*
6839 	 * md_add_new_disk can be used once the array is assembled
6840 	 * to add "hot spares".  They must already have a superblock
6841 	 * written
6842 	 */
6843 	if (mddev->pers) {
6844 		int err;
6845 		if (!mddev->pers->hot_add_disk) {
6846 			pr_warn("%s: personality does not support diskops!\n",
6847 				mdname(mddev));
6848 			return -EINVAL;
6849 		}
6850 		if (mddev->persistent)
6851 			rdev = md_import_device(dev, mddev->major_version,
6852 						mddev->minor_version);
6853 		else
6854 			rdev = md_import_device(dev, -1, -1);
6855 		if (IS_ERR(rdev)) {
6856 			pr_warn("md: md_import_device returned %ld\n",
6857 				PTR_ERR(rdev));
6858 			return PTR_ERR(rdev);
6859 		}
6860 		/* set saved_raid_disk if appropriate */
6861 		if (!mddev->persistent) {
6862 			if (info->state & (1<<MD_DISK_SYNC)  &&
6863 			    info->raid_disk < mddev->raid_disks) {
6864 				rdev->raid_disk = info->raid_disk;
6865 				clear_bit(Bitmap_sync, &rdev->flags);
6866 			} else
6867 				rdev->raid_disk = -1;
6868 			rdev->saved_raid_disk = rdev->raid_disk;
6869 		} else
6870 			super_types[mddev->major_version].
6871 				validate_super(mddev, NULL/*freshest*/, rdev);
6872 		if ((info->state & (1<<MD_DISK_SYNC)) &&
6873 		     rdev->raid_disk != info->raid_disk) {
6874 			/* This was a hot-add request, but events doesn't
6875 			 * match, so reject it.
6876 			 */
6877 			export_rdev(rdev, mddev);
6878 			return -EINVAL;
6879 		}
6880 
6881 		clear_bit(In_sync, &rdev->flags); /* just to be sure */
6882 		if (info->state & (1<<MD_DISK_WRITEMOSTLY))
6883 			set_bit(WriteMostly, &rdev->flags);
6884 		else
6885 			clear_bit(WriteMostly, &rdev->flags);
6886 		if (info->state & (1<<MD_DISK_FAILFAST))
6887 			set_bit(FailFast, &rdev->flags);
6888 		else
6889 			clear_bit(FailFast, &rdev->flags);
6890 
6891 		if (info->state & (1<<MD_DISK_JOURNAL)) {
6892 			struct md_rdev *rdev2;
6893 			bool has_journal = false;
6894 
6895 			/* make sure no existing journal disk */
6896 			rdev_for_each(rdev2, mddev) {
6897 				if (test_bit(Journal, &rdev2->flags)) {
6898 					has_journal = true;
6899 					break;
6900 				}
6901 			}
6902 			if (has_journal || mddev->bitmap) {
6903 				export_rdev(rdev, mddev);
6904 				return -EBUSY;
6905 			}
6906 			set_bit(Journal, &rdev->flags);
6907 		}
6908 		/*
6909 		 * check whether the device shows up in other nodes
6910 		 */
6911 		if (mddev_is_clustered(mddev)) {
6912 			if (info->state & (1 << MD_DISK_CANDIDATE))
6913 				set_bit(Candidate, &rdev->flags);
6914 			else if (info->state & (1 << MD_DISK_CLUSTER_ADD)) {
6915 				/* --add initiated by this node */
6916 				err = md_cluster_ops->add_new_disk(mddev, rdev);
6917 				if (err) {
6918 					export_rdev(rdev, mddev);
6919 					return err;
6920 				}
6921 			}
6922 		}
6923 
6924 		rdev->raid_disk = -1;
6925 		err = bind_rdev_to_array(rdev, mddev);
6926 
6927 		if (err)
6928 			export_rdev(rdev, mddev);
6929 
6930 		if (mddev_is_clustered(mddev)) {
6931 			if (info->state & (1 << MD_DISK_CANDIDATE)) {
6932 				if (!err) {
6933 					err = md_cluster_ops->new_disk_ack(mddev,
6934 						err == 0);
6935 					if (err)
6936 						md_kick_rdev_from_array(rdev);
6937 				}
6938 			} else {
6939 				if (err)
6940 					md_cluster_ops->add_new_disk_cancel(mddev);
6941 				else
6942 					err = add_bound_rdev(rdev);
6943 			}
6944 
6945 		} else if (!err)
6946 			err = add_bound_rdev(rdev);
6947 
6948 		return err;
6949 	}
6950 
6951 	/* otherwise, md_add_new_disk is only allowed
6952 	 * for major_version==0 superblocks
6953 	 */
6954 	if (mddev->major_version != 0) {
6955 		pr_warn("%s: ADD_NEW_DISK not supported\n", mdname(mddev));
6956 		return -EINVAL;
6957 	}
6958 
6959 	if (!(info->state & (1<<MD_DISK_FAULTY))) {
6960 		int err;
6961 		rdev = md_import_device(dev, -1, 0);
6962 		if (IS_ERR(rdev)) {
6963 			pr_warn("md: error, md_import_device() returned %ld\n",
6964 				PTR_ERR(rdev));
6965 			return PTR_ERR(rdev);
6966 		}
6967 		rdev->desc_nr = info->number;
6968 		if (info->raid_disk < mddev->raid_disks)
6969 			rdev->raid_disk = info->raid_disk;
6970 		else
6971 			rdev->raid_disk = -1;
6972 
6973 		if (rdev->raid_disk < mddev->raid_disks)
6974 			if (info->state & (1<<MD_DISK_SYNC))
6975 				set_bit(In_sync, &rdev->flags);
6976 
6977 		if (info->state & (1<<MD_DISK_WRITEMOSTLY))
6978 			set_bit(WriteMostly, &rdev->flags);
6979 		if (info->state & (1<<MD_DISK_FAILFAST))
6980 			set_bit(FailFast, &rdev->flags);
6981 
6982 		if (!mddev->persistent) {
6983 			pr_debug("md: nonpersistent superblock ...\n");
6984 			rdev->sb_start = bdev_nr_sectors(rdev->bdev);
6985 		} else
6986 			rdev->sb_start = calc_dev_sboffset(rdev);
6987 		rdev->sectors = rdev->sb_start;
6988 
6989 		err = bind_rdev_to_array(rdev, mddev);
6990 		if (err) {
6991 			export_rdev(rdev, mddev);
6992 			return err;
6993 		}
6994 	}
6995 
6996 	return 0;
6997 }
6998 
hot_remove_disk(struct mddev * mddev,dev_t dev)6999 static int hot_remove_disk(struct mddev *mddev, dev_t dev)
7000 {
7001 	struct md_rdev *rdev;
7002 
7003 	if (!mddev->pers)
7004 		return -ENODEV;
7005 
7006 	rdev = find_rdev(mddev, dev);
7007 	if (!rdev)
7008 		return -ENXIO;
7009 
7010 	if (rdev->raid_disk < 0)
7011 		goto kick_rdev;
7012 
7013 	clear_bit(Blocked, &rdev->flags);
7014 	remove_and_add_spares(mddev, rdev);
7015 
7016 	if (rdev->raid_disk >= 0)
7017 		goto busy;
7018 
7019 kick_rdev:
7020 	if (mddev_is_clustered(mddev)) {
7021 		if (md_cluster_ops->remove_disk(mddev, rdev))
7022 			goto busy;
7023 	}
7024 
7025 	md_kick_rdev_from_array(rdev);
7026 	set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
7027 	if (mddev->thread)
7028 		md_wakeup_thread(mddev->thread);
7029 	else
7030 		md_update_sb(mddev, 1);
7031 	md_new_event();
7032 
7033 	return 0;
7034 busy:
7035 	pr_debug("md: cannot remove active disk %pg from %s ...\n",
7036 		 rdev->bdev, mdname(mddev));
7037 	return -EBUSY;
7038 }
7039 
hot_add_disk(struct mddev * mddev,dev_t dev)7040 static int hot_add_disk(struct mddev *mddev, dev_t dev)
7041 {
7042 	int err;
7043 	struct md_rdev *rdev;
7044 
7045 	if (!mddev->pers)
7046 		return -ENODEV;
7047 
7048 	if (mddev->major_version != 0) {
7049 		pr_warn("%s: HOT_ADD may only be used with version-0 superblocks.\n",
7050 			mdname(mddev));
7051 		return -EINVAL;
7052 	}
7053 	if (!mddev->pers->hot_add_disk) {
7054 		pr_warn("%s: personality does not support diskops!\n",
7055 			mdname(mddev));
7056 		return -EINVAL;
7057 	}
7058 
7059 	rdev = md_import_device(dev, -1, 0);
7060 	if (IS_ERR(rdev)) {
7061 		pr_warn("md: error, md_import_device() returned %ld\n",
7062 			PTR_ERR(rdev));
7063 		return -EINVAL;
7064 	}
7065 
7066 	if (mddev->persistent)
7067 		rdev->sb_start = calc_dev_sboffset(rdev);
7068 	else
7069 		rdev->sb_start = bdev_nr_sectors(rdev->bdev);
7070 
7071 	rdev->sectors = rdev->sb_start;
7072 
7073 	if (test_bit(Faulty, &rdev->flags)) {
7074 		pr_warn("md: can not hot-add faulty %pg disk to %s!\n",
7075 			rdev->bdev, mdname(mddev));
7076 		err = -EINVAL;
7077 		goto abort_export;
7078 	}
7079 
7080 	clear_bit(In_sync, &rdev->flags);
7081 	rdev->desc_nr = -1;
7082 	rdev->saved_raid_disk = -1;
7083 	err = bind_rdev_to_array(rdev, mddev);
7084 	if (err)
7085 		goto abort_export;
7086 
7087 	/*
7088 	 * The rest should better be atomic, we can have disk failures
7089 	 * noticed in interrupt contexts ...
7090 	 */
7091 
7092 	rdev->raid_disk = -1;
7093 
7094 	set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
7095 	if (!mddev->thread)
7096 		md_update_sb(mddev, 1);
7097 	/*
7098 	 * If the new disk does not support REQ_NOWAIT,
7099 	 * disable on the whole MD.
7100 	 */
7101 	if (!bdev_nowait(rdev->bdev)) {
7102 		pr_info("%s: Disabling nowait because %pg does not support nowait\n",
7103 			mdname(mddev), rdev->bdev);
7104 		blk_queue_flag_clear(QUEUE_FLAG_NOWAIT, mddev->queue);
7105 	}
7106 	/*
7107 	 * Kick recovery, maybe this spare has to be added to the
7108 	 * array immediately.
7109 	 */
7110 	set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7111 	md_wakeup_thread(mddev->thread);
7112 	md_new_event();
7113 	return 0;
7114 
7115 abort_export:
7116 	export_rdev(rdev, mddev);
7117 	return err;
7118 }
7119 
set_bitmap_file(struct mddev * mddev,int fd)7120 static int set_bitmap_file(struct mddev *mddev, int fd)
7121 {
7122 	int err = 0;
7123 
7124 	if (mddev->pers) {
7125 		if (!mddev->pers->quiesce || !mddev->thread)
7126 			return -EBUSY;
7127 		if (mddev->recovery || mddev->sync_thread)
7128 			return -EBUSY;
7129 		/* we should be able to change the bitmap.. */
7130 	}
7131 
7132 	if (fd >= 0) {
7133 		struct inode *inode;
7134 		struct file *f;
7135 
7136 		if (mddev->bitmap || mddev->bitmap_info.file)
7137 			return -EEXIST; /* cannot add when bitmap is present */
7138 
7139 		if (!IS_ENABLED(CONFIG_MD_BITMAP_FILE)) {
7140 			pr_warn("%s: bitmap files not supported by this kernel\n",
7141 				mdname(mddev));
7142 			return -EINVAL;
7143 		}
7144 		pr_warn("%s: using deprecated bitmap file support\n",
7145 			mdname(mddev));
7146 
7147 		f = fget(fd);
7148 
7149 		if (f == NULL) {
7150 			pr_warn("%s: error: failed to get bitmap file\n",
7151 				mdname(mddev));
7152 			return -EBADF;
7153 		}
7154 
7155 		inode = f->f_mapping->host;
7156 		if (!S_ISREG(inode->i_mode)) {
7157 			pr_warn("%s: error: bitmap file must be a regular file\n",
7158 				mdname(mddev));
7159 			err = -EBADF;
7160 		} else if (!(f->f_mode & FMODE_WRITE)) {
7161 			pr_warn("%s: error: bitmap file must open for write\n",
7162 				mdname(mddev));
7163 			err = -EBADF;
7164 		} else if (atomic_read(&inode->i_writecount) != 1) {
7165 			pr_warn("%s: error: bitmap file is already in use\n",
7166 				mdname(mddev));
7167 			err = -EBUSY;
7168 		}
7169 		if (err) {
7170 			fput(f);
7171 			return err;
7172 		}
7173 		mddev->bitmap_info.file = f;
7174 		mddev->bitmap_info.offset = 0; /* file overrides offset */
7175 	} else if (mddev->bitmap == NULL)
7176 		return -ENOENT; /* cannot remove what isn't there */
7177 	err = 0;
7178 	if (mddev->pers) {
7179 		if (fd >= 0) {
7180 			struct bitmap *bitmap;
7181 
7182 			bitmap = md_bitmap_create(mddev, -1);
7183 			mddev_suspend(mddev);
7184 			if (!IS_ERR(bitmap)) {
7185 				mddev->bitmap = bitmap;
7186 				err = md_bitmap_load(mddev);
7187 			} else
7188 				err = PTR_ERR(bitmap);
7189 			if (err) {
7190 				md_bitmap_destroy(mddev);
7191 				fd = -1;
7192 			}
7193 			mddev_resume(mddev);
7194 		} else if (fd < 0) {
7195 			mddev_suspend(mddev);
7196 			md_bitmap_destroy(mddev);
7197 			mddev_resume(mddev);
7198 		}
7199 	}
7200 	if (fd < 0) {
7201 		struct file *f = mddev->bitmap_info.file;
7202 		if (f) {
7203 			spin_lock(&mddev->lock);
7204 			mddev->bitmap_info.file = NULL;
7205 			spin_unlock(&mddev->lock);
7206 			fput(f);
7207 		}
7208 	}
7209 
7210 	return err;
7211 }
7212 
7213 /*
7214  * md_set_array_info is used two different ways
7215  * The original usage is when creating a new array.
7216  * In this usage, raid_disks is > 0 and it together with
7217  *  level, size, not_persistent,layout,chunksize determine the
7218  *  shape of the array.
7219  *  This will always create an array with a type-0.90.0 superblock.
7220  * The newer usage is when assembling an array.
7221  *  In this case raid_disks will be 0, and the major_version field is
7222  *  use to determine which style super-blocks are to be found on the devices.
7223  *  The minor and patch _version numbers are also kept incase the
7224  *  super_block handler wishes to interpret them.
7225  */
md_set_array_info(struct mddev * mddev,struct mdu_array_info_s * info)7226 int md_set_array_info(struct mddev *mddev, struct mdu_array_info_s *info)
7227 {
7228 	if (info->raid_disks == 0) {
7229 		/* just setting version number for superblock loading */
7230 		if (info->major_version < 0 ||
7231 		    info->major_version >= ARRAY_SIZE(super_types) ||
7232 		    super_types[info->major_version].name == NULL) {
7233 			/* maybe try to auto-load a module? */
7234 			pr_warn("md: superblock version %d not known\n",
7235 				info->major_version);
7236 			return -EINVAL;
7237 		}
7238 		mddev->major_version = info->major_version;
7239 		mddev->minor_version = info->minor_version;
7240 		mddev->patch_version = info->patch_version;
7241 		mddev->persistent = !info->not_persistent;
7242 		/* ensure mddev_put doesn't delete this now that there
7243 		 * is some minimal configuration.
7244 		 */
7245 		mddev->ctime         = ktime_get_real_seconds();
7246 		return 0;
7247 	}
7248 	mddev->major_version = MD_MAJOR_VERSION;
7249 	mddev->minor_version = MD_MINOR_VERSION;
7250 	mddev->patch_version = MD_PATCHLEVEL_VERSION;
7251 	mddev->ctime         = ktime_get_real_seconds();
7252 
7253 	mddev->level         = info->level;
7254 	mddev->clevel[0]     = 0;
7255 	mddev->dev_sectors   = 2 * (sector_t)info->size;
7256 	mddev->raid_disks    = info->raid_disks;
7257 	/* don't set md_minor, it is determined by which /dev/md* was
7258 	 * openned
7259 	 */
7260 	if (info->state & (1<<MD_SB_CLEAN))
7261 		mddev->recovery_cp = MaxSector;
7262 	else
7263 		mddev->recovery_cp = 0;
7264 	mddev->persistent    = ! info->not_persistent;
7265 	mddev->external	     = 0;
7266 
7267 	mddev->layout        = info->layout;
7268 	if (mddev->level == 0)
7269 		/* Cannot trust RAID0 layout info here */
7270 		mddev->layout = -1;
7271 	mddev->chunk_sectors = info->chunk_size >> 9;
7272 
7273 	if (mddev->persistent) {
7274 		mddev->max_disks = MD_SB_DISKS;
7275 		mddev->flags = 0;
7276 		mddev->sb_flags = 0;
7277 	}
7278 	set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
7279 
7280 	mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9;
7281 	mddev->bitmap_info.default_space = 64*2 - (MD_SB_BYTES >> 9);
7282 	mddev->bitmap_info.offset = 0;
7283 
7284 	mddev->reshape_position = MaxSector;
7285 
7286 	/*
7287 	 * Generate a 128 bit UUID
7288 	 */
7289 	get_random_bytes(mddev->uuid, 16);
7290 
7291 	mddev->new_level = mddev->level;
7292 	mddev->new_chunk_sectors = mddev->chunk_sectors;
7293 	mddev->new_layout = mddev->layout;
7294 	mddev->delta_disks = 0;
7295 	mddev->reshape_backwards = 0;
7296 
7297 	return 0;
7298 }
7299 
md_set_array_sectors(struct mddev * mddev,sector_t array_sectors)7300 void md_set_array_sectors(struct mddev *mddev, sector_t array_sectors)
7301 {
7302 	lockdep_assert_held(&mddev->reconfig_mutex);
7303 
7304 	if (mddev->external_size)
7305 		return;
7306 
7307 	mddev->array_sectors = array_sectors;
7308 }
7309 EXPORT_SYMBOL(md_set_array_sectors);
7310 
update_size(struct mddev * mddev,sector_t num_sectors)7311 static int update_size(struct mddev *mddev, sector_t num_sectors)
7312 {
7313 	struct md_rdev *rdev;
7314 	int rv;
7315 	int fit = (num_sectors == 0);
7316 	sector_t old_dev_sectors = mddev->dev_sectors;
7317 
7318 	if (mddev->pers->resize == NULL)
7319 		return -EINVAL;
7320 	/* The "num_sectors" is the number of sectors of each device that
7321 	 * is used.  This can only make sense for arrays with redundancy.
7322 	 * linear and raid0 always use whatever space is available. We can only
7323 	 * consider changing this number if no resync or reconstruction is
7324 	 * happening, and if the new size is acceptable. It must fit before the
7325 	 * sb_start or, if that is <data_offset, it must fit before the size
7326 	 * of each device.  If num_sectors is zero, we find the largest size
7327 	 * that fits.
7328 	 */
7329 	if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
7330 	    mddev->sync_thread)
7331 		return -EBUSY;
7332 	if (!md_is_rdwr(mddev))
7333 		return -EROFS;
7334 
7335 	rdev_for_each(rdev, mddev) {
7336 		sector_t avail = rdev->sectors;
7337 
7338 		if (fit && (num_sectors == 0 || num_sectors > avail))
7339 			num_sectors = avail;
7340 		if (avail < num_sectors)
7341 			return -ENOSPC;
7342 	}
7343 	rv = mddev->pers->resize(mddev, num_sectors);
7344 	if (!rv) {
7345 		if (mddev_is_clustered(mddev))
7346 			md_cluster_ops->update_size(mddev, old_dev_sectors);
7347 		else if (mddev->queue) {
7348 			set_capacity_and_notify(mddev->gendisk,
7349 						mddev->array_sectors);
7350 		}
7351 	}
7352 	return rv;
7353 }
7354 
update_raid_disks(struct mddev * mddev,int raid_disks)7355 static int update_raid_disks(struct mddev *mddev, int raid_disks)
7356 {
7357 	int rv;
7358 	struct md_rdev *rdev;
7359 	/* change the number of raid disks */
7360 	if (mddev->pers->check_reshape == NULL)
7361 		return -EINVAL;
7362 	if (!md_is_rdwr(mddev))
7363 		return -EROFS;
7364 	if (raid_disks <= 0 ||
7365 	    (mddev->max_disks && raid_disks >= mddev->max_disks))
7366 		return -EINVAL;
7367 	if (mddev->sync_thread ||
7368 	    test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
7369 	    test_bit(MD_RESYNCING_REMOTE, &mddev->recovery) ||
7370 	    mddev->reshape_position != MaxSector)
7371 		return -EBUSY;
7372 
7373 	rdev_for_each(rdev, mddev) {
7374 		if (mddev->raid_disks < raid_disks &&
7375 		    rdev->data_offset < rdev->new_data_offset)
7376 			return -EINVAL;
7377 		if (mddev->raid_disks > raid_disks &&
7378 		    rdev->data_offset > rdev->new_data_offset)
7379 			return -EINVAL;
7380 	}
7381 
7382 	mddev->delta_disks = raid_disks - mddev->raid_disks;
7383 	if (mddev->delta_disks < 0)
7384 		mddev->reshape_backwards = 1;
7385 	else if (mddev->delta_disks > 0)
7386 		mddev->reshape_backwards = 0;
7387 
7388 	rv = mddev->pers->check_reshape(mddev);
7389 	if (rv < 0) {
7390 		mddev->delta_disks = 0;
7391 		mddev->reshape_backwards = 0;
7392 	}
7393 	return rv;
7394 }
7395 
7396 /*
7397  * update_array_info is used to change the configuration of an
7398  * on-line array.
7399  * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
7400  * fields in the info are checked against the array.
7401  * Any differences that cannot be handled will cause an error.
7402  * Normally, only one change can be managed at a time.
7403  */
update_array_info(struct mddev * mddev,mdu_array_info_t * info)7404 static int update_array_info(struct mddev *mddev, mdu_array_info_t *info)
7405 {
7406 	int rv = 0;
7407 	int cnt = 0;
7408 	int state = 0;
7409 
7410 	/* calculate expected state,ignoring low bits */
7411 	if (mddev->bitmap && mddev->bitmap_info.offset)
7412 		state |= (1 << MD_SB_BITMAP_PRESENT);
7413 
7414 	if (mddev->major_version != info->major_version ||
7415 	    mddev->minor_version != info->minor_version ||
7416 /*	    mddev->patch_version != info->patch_version || */
7417 	    mddev->ctime         != info->ctime         ||
7418 	    mddev->level         != info->level         ||
7419 /*	    mddev->layout        != info->layout        || */
7420 	    mddev->persistent	 != !info->not_persistent ||
7421 	    mddev->chunk_sectors != info->chunk_size >> 9 ||
7422 	    /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
7423 	    ((state^info->state) & 0xfffffe00)
7424 		)
7425 		return -EINVAL;
7426 	/* Check there is only one change */
7427 	if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
7428 		cnt++;
7429 	if (mddev->raid_disks != info->raid_disks)
7430 		cnt++;
7431 	if (mddev->layout != info->layout)
7432 		cnt++;
7433 	if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT))
7434 		cnt++;
7435 	if (cnt == 0)
7436 		return 0;
7437 	if (cnt > 1)
7438 		return -EINVAL;
7439 
7440 	if (mddev->layout != info->layout) {
7441 		/* Change layout
7442 		 * we don't need to do anything at the md level, the
7443 		 * personality will take care of it all.
7444 		 */
7445 		if (mddev->pers->check_reshape == NULL)
7446 			return -EINVAL;
7447 		else {
7448 			mddev->new_layout = info->layout;
7449 			rv = mddev->pers->check_reshape(mddev);
7450 			if (rv)
7451 				mddev->new_layout = mddev->layout;
7452 			return rv;
7453 		}
7454 	}
7455 	if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
7456 		rv = update_size(mddev, (sector_t)info->size * 2);
7457 
7458 	if (mddev->raid_disks    != info->raid_disks)
7459 		rv = update_raid_disks(mddev, info->raid_disks);
7460 
7461 	if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) {
7462 		if (mddev->pers->quiesce == NULL || mddev->thread == NULL) {
7463 			rv = -EINVAL;
7464 			goto err;
7465 		}
7466 		if (mddev->recovery || mddev->sync_thread) {
7467 			rv = -EBUSY;
7468 			goto err;
7469 		}
7470 		if (info->state & (1<<MD_SB_BITMAP_PRESENT)) {
7471 			struct bitmap *bitmap;
7472 			/* add the bitmap */
7473 			if (mddev->bitmap) {
7474 				rv = -EEXIST;
7475 				goto err;
7476 			}
7477 			if (mddev->bitmap_info.default_offset == 0) {
7478 				rv = -EINVAL;
7479 				goto err;
7480 			}
7481 			mddev->bitmap_info.offset =
7482 				mddev->bitmap_info.default_offset;
7483 			mddev->bitmap_info.space =
7484 				mddev->bitmap_info.default_space;
7485 			bitmap = md_bitmap_create(mddev, -1);
7486 			mddev_suspend(mddev);
7487 			if (!IS_ERR(bitmap)) {
7488 				mddev->bitmap = bitmap;
7489 				rv = md_bitmap_load(mddev);
7490 			} else
7491 				rv = PTR_ERR(bitmap);
7492 			if (rv)
7493 				md_bitmap_destroy(mddev);
7494 			mddev_resume(mddev);
7495 		} else {
7496 			/* remove the bitmap */
7497 			if (!mddev->bitmap) {
7498 				rv = -ENOENT;
7499 				goto err;
7500 			}
7501 			if (mddev->bitmap->storage.file) {
7502 				rv = -EINVAL;
7503 				goto err;
7504 			}
7505 			if (mddev->bitmap_info.nodes) {
7506 				/* hold PW on all the bitmap lock */
7507 				if (md_cluster_ops->lock_all_bitmaps(mddev) <= 0) {
7508 					pr_warn("md: can't change bitmap to none since the array is in use by more than one node\n");
7509 					rv = -EPERM;
7510 					md_cluster_ops->unlock_all_bitmaps(mddev);
7511 					goto err;
7512 				}
7513 
7514 				mddev->bitmap_info.nodes = 0;
7515 				md_cluster_ops->leave(mddev);
7516 				module_put(md_cluster_mod);
7517 				mddev->safemode_delay = DEFAULT_SAFEMODE_DELAY;
7518 			}
7519 			mddev_suspend(mddev);
7520 			md_bitmap_destroy(mddev);
7521 			mddev_resume(mddev);
7522 			mddev->bitmap_info.offset = 0;
7523 		}
7524 	}
7525 	md_update_sb(mddev, 1);
7526 	return rv;
7527 err:
7528 	return rv;
7529 }
7530 
set_disk_faulty(struct mddev * mddev,dev_t dev)7531 static int set_disk_faulty(struct mddev *mddev, dev_t dev)
7532 {
7533 	struct md_rdev *rdev;
7534 	int err = 0;
7535 
7536 	if (mddev->pers == NULL)
7537 		return -ENODEV;
7538 
7539 	rcu_read_lock();
7540 	rdev = md_find_rdev_rcu(mddev, dev);
7541 	if (!rdev)
7542 		err =  -ENODEV;
7543 	else {
7544 		md_error(mddev, rdev);
7545 		if (test_bit(MD_BROKEN, &mddev->flags))
7546 			err = -EBUSY;
7547 	}
7548 	rcu_read_unlock();
7549 	return err;
7550 }
7551 
7552 /*
7553  * We have a problem here : there is no easy way to give a CHS
7554  * virtual geometry. We currently pretend that we have a 2 heads
7555  * 4 sectors (with a BIG number of cylinders...). This drives
7556  * dosfs just mad... ;-)
7557  */
md_getgeo(struct block_device * bdev,struct hd_geometry * geo)7558 static int md_getgeo(struct block_device *bdev, struct hd_geometry *geo)
7559 {
7560 	struct mddev *mddev = bdev->bd_disk->private_data;
7561 
7562 	geo->heads = 2;
7563 	geo->sectors = 4;
7564 	geo->cylinders = mddev->array_sectors / 8;
7565 	return 0;
7566 }
7567 
md_ioctl_valid(unsigned int cmd)7568 static inline bool md_ioctl_valid(unsigned int cmd)
7569 {
7570 	switch (cmd) {
7571 	case ADD_NEW_DISK:
7572 	case GET_ARRAY_INFO:
7573 	case GET_BITMAP_FILE:
7574 	case GET_DISK_INFO:
7575 	case HOT_ADD_DISK:
7576 	case HOT_REMOVE_DISK:
7577 	case RAID_VERSION:
7578 	case RESTART_ARRAY_RW:
7579 	case RUN_ARRAY:
7580 	case SET_ARRAY_INFO:
7581 	case SET_BITMAP_FILE:
7582 	case SET_DISK_FAULTY:
7583 	case STOP_ARRAY:
7584 	case STOP_ARRAY_RO:
7585 	case CLUSTERED_DISK_NACK:
7586 		return true;
7587 	default:
7588 		return false;
7589 	}
7590 }
7591 
__md_set_array_info(struct mddev * mddev,void __user * argp)7592 static int __md_set_array_info(struct mddev *mddev, void __user *argp)
7593 {
7594 	mdu_array_info_t info;
7595 	int err;
7596 
7597 	if (!argp)
7598 		memset(&info, 0, sizeof(info));
7599 	else if (copy_from_user(&info, argp, sizeof(info)))
7600 		return -EFAULT;
7601 
7602 	if (mddev->pers) {
7603 		err = update_array_info(mddev, &info);
7604 		if (err)
7605 			pr_warn("md: couldn't update array info. %d\n", err);
7606 		return err;
7607 	}
7608 
7609 	if (!list_empty(&mddev->disks)) {
7610 		pr_warn("md: array %s already has disks!\n", mdname(mddev));
7611 		return -EBUSY;
7612 	}
7613 
7614 	if (mddev->raid_disks) {
7615 		pr_warn("md: array %s already initialised!\n", mdname(mddev));
7616 		return -EBUSY;
7617 	}
7618 
7619 	err = md_set_array_info(mddev, &info);
7620 	if (err)
7621 		pr_warn("md: couldn't set array info. %d\n", err);
7622 
7623 	return err;
7624 }
7625 
md_ioctl(struct block_device * bdev,blk_mode_t mode,unsigned int cmd,unsigned long arg)7626 static int md_ioctl(struct block_device *bdev, blk_mode_t mode,
7627 			unsigned int cmd, unsigned long arg)
7628 {
7629 	int err = 0;
7630 	void __user *argp = (void __user *)arg;
7631 	struct mddev *mddev = NULL;
7632 
7633 	if (!md_ioctl_valid(cmd))
7634 		return -ENOTTY;
7635 
7636 	switch (cmd) {
7637 	case RAID_VERSION:
7638 	case GET_ARRAY_INFO:
7639 	case GET_DISK_INFO:
7640 		break;
7641 	default:
7642 		if (!capable(CAP_SYS_ADMIN))
7643 			return -EACCES;
7644 	}
7645 
7646 	/*
7647 	 * Commands dealing with the RAID driver but not any
7648 	 * particular array:
7649 	 */
7650 	switch (cmd) {
7651 	case RAID_VERSION:
7652 		err = get_version(argp);
7653 		goto out;
7654 	default:;
7655 	}
7656 
7657 	/*
7658 	 * Commands creating/starting a new array:
7659 	 */
7660 
7661 	mddev = bdev->bd_disk->private_data;
7662 
7663 	/* Some actions do not requires the mutex */
7664 	switch (cmd) {
7665 	case GET_ARRAY_INFO:
7666 		if (!mddev->raid_disks && !mddev->external)
7667 			err = -ENODEV;
7668 		else
7669 			err = get_array_info(mddev, argp);
7670 		goto out;
7671 
7672 	case GET_DISK_INFO:
7673 		if (!mddev->raid_disks && !mddev->external)
7674 			err = -ENODEV;
7675 		else
7676 			err = get_disk_info(mddev, argp);
7677 		goto out;
7678 
7679 	case SET_DISK_FAULTY:
7680 		err = set_disk_faulty(mddev, new_decode_dev(arg));
7681 		goto out;
7682 
7683 	case GET_BITMAP_FILE:
7684 		err = get_bitmap_file(mddev, argp);
7685 		goto out;
7686 
7687 	}
7688 
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
md_compat_ioctl(struct block_device * bdev,blk_mode_t mode,unsigned int cmd,unsigned long arg)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 
md_set_read_only(struct block_device * bdev,bool ro)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 
md_open(struct gendisk * disk,blk_mode_t mode)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 
md_release(struct gendisk * disk)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 
md_check_events(struct gendisk * disk,unsigned int clearing)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 
md_free_disk(struct gendisk * disk)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 
md_thread(void * arg)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 
md_wakeup_thread_directly(struct md_thread __rcu * thread)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 
md_wakeup_thread(struct md_thread __rcu * thread)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 
md_register_thread(void (* run)(struct md_thread *),struct mddev * mddev,const char * name)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 
md_unregister_thread(struct mddev * mddev,struct md_thread __rcu ** threadp)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 
md_error(struct mddev * mddev,struct md_rdev * rdev)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 
status_unused(struct seq_file * seq)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 
status_personalities(struct seq_file * seq)8124 static void status_personalities(struct seq_file *seq)
8125 {
8126 	struct md_personality *pers;
8127 
8128 	seq_puts(seq, "Personalities : ");
8129 	spin_lock(&pers_lock);
8130 	list_for_each_entry(pers, &pers_list, list)
8131 		seq_printf(seq, "[%s] ", pers->name);
8132 
8133 	spin_unlock(&pers_lock);
8134 	seq_puts(seq, "\n");
8135 }
8136 
status_resync(struct seq_file * seq,struct mddev * mddev)8137 static int status_resync(struct seq_file *seq, struct mddev *mddev)
8138 {
8139 	sector_t max_sectors, resync, res;
8140 	unsigned long dt, db = 0;
8141 	sector_t rt, curr_mark_cnt, resync_mark_cnt;
8142 	int scale, recovery_active;
8143 	unsigned int per_milli;
8144 
8145 	if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ||
8146 	    test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
8147 		max_sectors = mddev->resync_max_sectors;
8148 	else
8149 		max_sectors = mddev->dev_sectors;
8150 
8151 	resync = mddev->curr_resync;
8152 	if (resync < MD_RESYNC_ACTIVE) {
8153 		if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
8154 			/* Still cleaning up */
8155 			resync = max_sectors;
8156 	} else if (resync > max_sectors) {
8157 		resync = max_sectors;
8158 	} else {
8159 		res = atomic_read(&mddev->recovery_active);
8160 		/*
8161 		 * Resync has started, but the subtraction has overflowed or
8162 		 * yielded one of the special values. Force it to active to
8163 		 * ensure the status reports an active resync.
8164 		 */
8165 		if (resync < res || resync - res < MD_RESYNC_ACTIVE)
8166 			resync = MD_RESYNC_ACTIVE;
8167 		else
8168 			resync -= res;
8169 	}
8170 
8171 	if (resync == MD_RESYNC_NONE) {
8172 		if (test_bit(MD_RESYNCING_REMOTE, &mddev->recovery)) {
8173 			struct md_rdev *rdev;
8174 
8175 			rdev_for_each(rdev, mddev)
8176 				if (rdev->raid_disk >= 0 &&
8177 				    !test_bit(Faulty, &rdev->flags) &&
8178 				    rdev->recovery_offset != MaxSector &&
8179 				    rdev->recovery_offset) {
8180 					seq_printf(seq, "\trecover=REMOTE");
8181 					return 1;
8182 				}
8183 			if (mddev->reshape_position != MaxSector)
8184 				seq_printf(seq, "\treshape=REMOTE");
8185 			else
8186 				seq_printf(seq, "\tresync=REMOTE");
8187 			return 1;
8188 		}
8189 		if (mddev->recovery_cp < MaxSector) {
8190 			seq_printf(seq, "\tresync=PENDING");
8191 			return 1;
8192 		}
8193 		return 0;
8194 	}
8195 	if (resync < MD_RESYNC_ACTIVE) {
8196 		seq_printf(seq, "\tresync=DELAYED");
8197 		return 1;
8198 	}
8199 
8200 	WARN_ON(max_sectors == 0);
8201 	/* Pick 'scale' such that (resync>>scale)*1000 will fit
8202 	 * in a sector_t, and (max_sectors>>scale) will fit in a
8203 	 * u32, as those are the requirements for sector_div.
8204 	 * Thus 'scale' must be at least 10
8205 	 */
8206 	scale = 10;
8207 	if (sizeof(sector_t) > sizeof(unsigned long)) {
8208 		while ( max_sectors/2 > (1ULL<<(scale+32)))
8209 			scale++;
8210 	}
8211 	res = (resync>>scale)*1000;
8212 	sector_div(res, (u32)((max_sectors>>scale)+1));
8213 
8214 	per_milli = res;
8215 	{
8216 		int i, x = per_milli/50, y = 20-x;
8217 		seq_printf(seq, "[");
8218 		for (i = 0; i < x; i++)
8219 			seq_printf(seq, "=");
8220 		seq_printf(seq, ">");
8221 		for (i = 0; i < y; i++)
8222 			seq_printf(seq, ".");
8223 		seq_printf(seq, "] ");
8224 	}
8225 	seq_printf(seq, " %s =%3u.%u%% (%llu/%llu)",
8226 		   (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)?
8227 		    "reshape" :
8228 		    (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)?
8229 		     "check" :
8230 		     (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
8231 		      "resync" : "recovery"))),
8232 		   per_milli/10, per_milli % 10,
8233 		   (unsigned long long) resync/2,
8234 		   (unsigned long long) max_sectors/2);
8235 
8236 	/*
8237 	 * dt: time from mark until now
8238 	 * db: blocks written from mark until now
8239 	 * rt: remaining time
8240 	 *
8241 	 * rt is a sector_t, which is always 64bit now. We are keeping
8242 	 * the original algorithm, but it is not really necessary.
8243 	 *
8244 	 * Original algorithm:
8245 	 *   So we divide before multiply in case it is 32bit and close
8246 	 *   to the limit.
8247 	 *   We scale the divisor (db) by 32 to avoid losing precision
8248 	 *   near the end of resync when the number of remaining sectors
8249 	 *   is close to 'db'.
8250 	 *   We then divide rt by 32 after multiplying by db to compensate.
8251 	 *   The '+1' avoids division by zero if db is very small.
8252 	 */
8253 	dt = ((jiffies - mddev->resync_mark) / HZ);
8254 	if (!dt) dt++;
8255 
8256 	curr_mark_cnt = mddev->curr_mark_cnt;
8257 	recovery_active = atomic_read(&mddev->recovery_active);
8258 	resync_mark_cnt = mddev->resync_mark_cnt;
8259 
8260 	if (curr_mark_cnt >= (recovery_active + resync_mark_cnt))
8261 		db = curr_mark_cnt - (recovery_active + resync_mark_cnt);
8262 
8263 	rt = max_sectors - resync;    /* number of remaining sectors */
8264 	rt = div64_u64(rt, db/32+1);
8265 	rt *= dt;
8266 	rt >>= 5;
8267 
8268 	seq_printf(seq, " finish=%lu.%lumin", (unsigned long)rt / 60,
8269 		   ((unsigned long)rt % 60)/6);
8270 
8271 	seq_printf(seq, " speed=%ldK/sec", db/2/dt);
8272 	return 1;
8273 }
8274 
md_seq_start(struct seq_file * seq,loff_t * pos)8275 static void *md_seq_start(struct seq_file *seq, loff_t *pos)
8276 	__acquires(&all_mddevs_lock)
8277 {
8278 	seq->poll_event = atomic_read(&md_event_count);
8279 	spin_lock(&all_mddevs_lock);
8280 
8281 	return seq_list_start_head(&all_mddevs, *pos);
8282 }
8283 
md_seq_next(struct seq_file * seq,void * v,loff_t * pos)8284 static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
8285 {
8286 	return seq_list_next(v, &all_mddevs, pos);
8287 }
8288 
md_seq_stop(struct seq_file * seq,void * v)8289 static void md_seq_stop(struct seq_file *seq, void *v)
8290 	__releases(&all_mddevs_lock)
8291 {
8292 	spin_unlock(&all_mddevs_lock);
8293 }
8294 
md_bitmap_status(struct seq_file * seq,struct mddev * mddev)8295 static void md_bitmap_status(struct seq_file *seq, struct mddev *mddev)
8296 {
8297 	struct md_bitmap_stats stats;
8298 	unsigned long used_pages;
8299 	unsigned long chunk_kb;
8300 	int err;
8301 
8302 	err = md_bitmap_get_stats(mddev->bitmap, &stats);
8303 	if (err)
8304 		return;
8305 
8306 	chunk_kb = mddev->bitmap_info.chunksize >> 10;
8307 	used_pages = stats.pages - stats.missing_pages;
8308 
8309 	seq_printf(seq, "bitmap: %lu/%lu pages [%luKB], %lu%s chunk",
8310 		   used_pages, stats.pages, used_pages << (PAGE_SHIFT - 10),
8311 		   chunk_kb ? chunk_kb : mddev->bitmap_info.chunksize,
8312 		   chunk_kb ? "KB" : "B");
8313 
8314 	if (stats.file) {
8315 		seq_puts(seq, ", file: ");
8316 		seq_file_path(seq, stats.file, " \t\n");
8317 	}
8318 
8319 	seq_putc(seq, '\n');
8320 }
8321 
md_seq_show(struct seq_file * seq,void * v)8322 static int md_seq_show(struct seq_file *seq, void *v)
8323 {
8324 	struct mddev *mddev;
8325 	sector_t sectors;
8326 	struct md_rdev *rdev;
8327 
8328 	if (v == &all_mddevs) {
8329 		status_personalities(seq);
8330 		if (list_empty(&all_mddevs))
8331 			status_unused(seq);
8332 		return 0;
8333 	}
8334 
8335 	mddev = list_entry(v, struct mddev, all_mddevs);
8336 	if (!mddev_get(mddev))
8337 		return 0;
8338 
8339 	spin_unlock(&all_mddevs_lock);
8340 
8341 	/* prevent bitmap to be freed after checking */
8342 	mutex_lock(&mddev->bitmap_info.mutex);
8343 
8344 	spin_lock(&mddev->lock);
8345 	if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
8346 		seq_printf(seq, "%s : %sactive", mdname(mddev),
8347 						mddev->pers ? "" : "in");
8348 		if (mddev->pers) {
8349 			if (mddev->ro == MD_RDONLY)
8350 				seq_printf(seq, " (read-only)");
8351 			if (mddev->ro == MD_AUTO_READ)
8352 				seq_printf(seq, " (auto-read-only)");
8353 			seq_printf(seq, " %s", mddev->pers->name);
8354 		}
8355 
8356 		sectors = 0;
8357 		rcu_read_lock();
8358 		rdev_for_each_rcu(rdev, mddev) {
8359 			seq_printf(seq, " %pg[%d]", rdev->bdev, rdev->desc_nr);
8360 
8361 			if (test_bit(WriteMostly, &rdev->flags))
8362 				seq_printf(seq, "(W)");
8363 			if (test_bit(Journal, &rdev->flags))
8364 				seq_printf(seq, "(J)");
8365 			if (test_bit(Faulty, &rdev->flags)) {
8366 				seq_printf(seq, "(F)");
8367 				continue;
8368 			}
8369 			if (rdev->raid_disk < 0)
8370 				seq_printf(seq, "(S)"); /* spare */
8371 			if (test_bit(Replacement, &rdev->flags))
8372 				seq_printf(seq, "(R)");
8373 			sectors += rdev->sectors;
8374 		}
8375 		rcu_read_unlock();
8376 
8377 		if (!list_empty(&mddev->disks)) {
8378 			if (mddev->pers)
8379 				seq_printf(seq, "\n      %llu blocks",
8380 					   (unsigned long long)
8381 					   mddev->array_sectors / 2);
8382 			else
8383 				seq_printf(seq, "\n      %llu blocks",
8384 					   (unsigned long long)sectors / 2);
8385 		}
8386 		if (mddev->persistent) {
8387 			if (mddev->major_version != 0 ||
8388 			    mddev->minor_version != 90) {
8389 				seq_printf(seq," super %d.%d",
8390 					   mddev->major_version,
8391 					   mddev->minor_version);
8392 			}
8393 		} else if (mddev->external)
8394 			seq_printf(seq, " super external:%s",
8395 				   mddev->metadata_type);
8396 		else
8397 			seq_printf(seq, " super non-persistent");
8398 
8399 		if (mddev->pers) {
8400 			mddev->pers->status(seq, mddev);
8401 			seq_printf(seq, "\n      ");
8402 			if (mddev->pers->sync_request) {
8403 				if (status_resync(seq, mddev))
8404 					seq_printf(seq, "\n      ");
8405 			}
8406 		} else
8407 			seq_printf(seq, "\n       ");
8408 
8409 		md_bitmap_status(seq, mddev);
8410 
8411 		seq_printf(seq, "\n");
8412 	}
8413 	spin_unlock(&mddev->lock);
8414 	mutex_unlock(&mddev->bitmap_info.mutex);
8415 	spin_lock(&all_mddevs_lock);
8416 
8417 	if (mddev == list_last_entry(&all_mddevs, struct mddev, all_mddevs))
8418 		status_unused(seq);
8419 
8420 	if (atomic_dec_and_test(&mddev->active))
8421 		__mddev_put(mddev);
8422 
8423 	return 0;
8424 }
8425 
8426 static const struct seq_operations md_seq_ops = {
8427 	.start  = md_seq_start,
8428 	.next   = md_seq_next,
8429 	.stop   = md_seq_stop,
8430 	.show   = md_seq_show,
8431 };
8432 
md_seq_open(struct inode * inode,struct file * file)8433 static int md_seq_open(struct inode *inode, struct file *file)
8434 {
8435 	struct seq_file *seq;
8436 	int error;
8437 
8438 	error = seq_open(file, &md_seq_ops);
8439 	if (error)
8440 		return error;
8441 
8442 	seq = file->private_data;
8443 	seq->poll_event = atomic_read(&md_event_count);
8444 	return error;
8445 }
8446 
8447 static int md_unloading;
mdstat_poll(struct file * filp,poll_table * wait)8448 static __poll_t mdstat_poll(struct file *filp, poll_table *wait)
8449 {
8450 	struct seq_file *seq = filp->private_data;
8451 	__poll_t mask;
8452 
8453 	if (md_unloading)
8454 		return EPOLLIN|EPOLLRDNORM|EPOLLERR|EPOLLPRI;
8455 	poll_wait(filp, &md_event_waiters, wait);
8456 
8457 	/* always allow read */
8458 	mask = EPOLLIN | EPOLLRDNORM;
8459 
8460 	if (seq->poll_event != atomic_read(&md_event_count))
8461 		mask |= EPOLLERR | EPOLLPRI;
8462 	return mask;
8463 }
8464 
8465 static const struct proc_ops mdstat_proc_ops = {
8466 	.proc_open	= md_seq_open,
8467 	.proc_read	= seq_read,
8468 	.proc_lseek	= seq_lseek,
8469 	.proc_release	= seq_release,
8470 	.proc_poll	= mdstat_poll,
8471 };
8472 
register_md_personality(struct md_personality * p)8473 int register_md_personality(struct md_personality *p)
8474 {
8475 	pr_debug("md: %s personality registered for level %d\n",
8476 		 p->name, p->level);
8477 	spin_lock(&pers_lock);
8478 	list_add_tail(&p->list, &pers_list);
8479 	spin_unlock(&pers_lock);
8480 	return 0;
8481 }
8482 EXPORT_SYMBOL(register_md_personality);
8483 
unregister_md_personality(struct md_personality * p)8484 int unregister_md_personality(struct md_personality *p)
8485 {
8486 	pr_debug("md: %s personality unregistered\n", p->name);
8487 	spin_lock(&pers_lock);
8488 	list_del_init(&p->list);
8489 	spin_unlock(&pers_lock);
8490 	return 0;
8491 }
8492 EXPORT_SYMBOL(unregister_md_personality);
8493 
register_md_cluster_operations(struct md_cluster_operations * ops,struct module * module)8494 int register_md_cluster_operations(struct md_cluster_operations *ops,
8495 				   struct module *module)
8496 {
8497 	int ret = 0;
8498 	spin_lock(&pers_lock);
8499 	if (md_cluster_ops != NULL)
8500 		ret = -EALREADY;
8501 	else {
8502 		md_cluster_ops = ops;
8503 		md_cluster_mod = module;
8504 	}
8505 	spin_unlock(&pers_lock);
8506 	return ret;
8507 }
8508 EXPORT_SYMBOL(register_md_cluster_operations);
8509 
unregister_md_cluster_operations(void)8510 int unregister_md_cluster_operations(void)
8511 {
8512 	spin_lock(&pers_lock);
8513 	md_cluster_ops = NULL;
8514 	spin_unlock(&pers_lock);
8515 	return 0;
8516 }
8517 EXPORT_SYMBOL(unregister_md_cluster_operations);
8518 
md_setup_cluster(struct mddev * mddev,int nodes)8519 int md_setup_cluster(struct mddev *mddev, int nodes)
8520 {
8521 	int ret;
8522 	if (!md_cluster_ops)
8523 		request_module("md-cluster");
8524 	spin_lock(&pers_lock);
8525 	/* ensure module won't be unloaded */
8526 	if (!md_cluster_ops || !try_module_get(md_cluster_mod)) {
8527 		pr_warn("can't find md-cluster module or get its reference.\n");
8528 		spin_unlock(&pers_lock);
8529 		return -ENOENT;
8530 	}
8531 	spin_unlock(&pers_lock);
8532 
8533 	ret = md_cluster_ops->join(mddev, nodes);
8534 	if (!ret)
8535 		mddev->safemode_delay = 0;
8536 	return ret;
8537 }
8538 
md_cluster_stop(struct mddev * mddev)8539 void md_cluster_stop(struct mddev *mddev)
8540 {
8541 	if (!md_cluster_ops)
8542 		return;
8543 	md_cluster_ops->leave(mddev);
8544 	module_put(md_cluster_mod);
8545 }
8546 
is_mddev_idle(struct mddev * mddev,int init)8547 static int is_mddev_idle(struct mddev *mddev, int init)
8548 {
8549 	struct md_rdev *rdev;
8550 	int idle;
8551 	int curr_events;
8552 
8553 	idle = 1;
8554 	rcu_read_lock();
8555 	rdev_for_each_rcu(rdev, mddev) {
8556 		struct gendisk *disk = rdev->bdev->bd_disk;
8557 		curr_events = (int)part_stat_read_accum(disk->part0, sectors) -
8558 			      atomic_read(&disk->sync_io);
8559 		/* sync IO will cause sync_io to increase before the disk_stats
8560 		 * as sync_io is counted when a request starts, and
8561 		 * disk_stats is counted when it completes.
8562 		 * So resync activity will cause curr_events to be smaller than
8563 		 * when there was no such activity.
8564 		 * non-sync IO will cause disk_stat to increase without
8565 		 * increasing sync_io so curr_events will (eventually)
8566 		 * be larger than it was before.  Once it becomes
8567 		 * substantially larger, the test below will cause
8568 		 * the array to appear non-idle, and resync will slow
8569 		 * down.
8570 		 * If there is a lot of outstanding resync activity when
8571 		 * we set last_event to curr_events, then all that activity
8572 		 * completing might cause the array to appear non-idle
8573 		 * and resync will be slowed down even though there might
8574 		 * not have been non-resync activity.  This will only
8575 		 * happen once though.  'last_events' will soon reflect
8576 		 * the state where there is little or no outstanding
8577 		 * resync requests, and further resync activity will
8578 		 * always make curr_events less than last_events.
8579 		 *
8580 		 */
8581 		if (init || curr_events - rdev->last_events > 64) {
8582 			rdev->last_events = curr_events;
8583 			idle = 0;
8584 		}
8585 	}
8586 	rcu_read_unlock();
8587 	return idle;
8588 }
8589 
md_done_sync(struct mddev * mddev,int blocks,int ok)8590 void md_done_sync(struct mddev *mddev, int blocks, int ok)
8591 {
8592 	/* another "blocks" (512byte) blocks have been synced */
8593 	atomic_sub(blocks, &mddev->recovery_active);
8594 	wake_up(&mddev->recovery_wait);
8595 	if (!ok) {
8596 		set_bit(MD_RECOVERY_INTR, &mddev->recovery);
8597 		set_bit(MD_RECOVERY_ERROR, &mddev->recovery);
8598 		md_wakeup_thread(mddev->thread);
8599 		// stop recovery, signal do_sync ....
8600 	}
8601 }
8602 EXPORT_SYMBOL(md_done_sync);
8603 
8604 /* md_write_start(mddev, bi)
8605  * If we need to update some array metadata (e.g. 'active' flag
8606  * in superblock) before writing, schedule a superblock update
8607  * and wait for it to complete.
8608  * A return value of 'false' means that the write wasn't recorded
8609  * and cannot proceed as the array is being suspend.
8610  */
md_write_start(struct mddev * mddev,struct bio * bi)8611 bool md_write_start(struct mddev *mddev, struct bio *bi)
8612 {
8613 	int did_change = 0;
8614 
8615 	if (bio_data_dir(bi) != WRITE)
8616 		return true;
8617 
8618 	BUG_ON(mddev->ro == MD_RDONLY);
8619 	if (mddev->ro == MD_AUTO_READ) {
8620 		/* need to switch to read/write */
8621 		mddev->ro = MD_RDWR;
8622 		set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
8623 		md_wakeup_thread(mddev->thread);
8624 		md_wakeup_thread(mddev->sync_thread);
8625 		did_change = 1;
8626 	}
8627 	rcu_read_lock();
8628 	percpu_ref_get(&mddev->writes_pending);
8629 	smp_mb(); /* Match smp_mb in set_in_sync() */
8630 	if (mddev->safemode == 1)
8631 		mddev->safemode = 0;
8632 	/* sync_checkers is always 0 when writes_pending is in per-cpu mode */
8633 	if (mddev->in_sync || mddev->sync_checkers) {
8634 		spin_lock(&mddev->lock);
8635 		if (mddev->in_sync) {
8636 			mddev->in_sync = 0;
8637 			set_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags);
8638 			set_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags);
8639 			md_wakeup_thread(mddev->thread);
8640 			did_change = 1;
8641 		}
8642 		spin_unlock(&mddev->lock);
8643 	}
8644 	rcu_read_unlock();
8645 	if (did_change)
8646 		sysfs_notify_dirent_safe(mddev->sysfs_state);
8647 	if (!mddev->has_superblocks)
8648 		return true;
8649 	wait_event(mddev->sb_wait,
8650 		   !test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags) ||
8651 		   is_md_suspended(mddev));
8652 	if (test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags)) {
8653 		percpu_ref_put(&mddev->writes_pending);
8654 		return false;
8655 	}
8656 	return true;
8657 }
8658 EXPORT_SYMBOL(md_write_start);
8659 
8660 /* md_write_inc can only be called when md_write_start() has
8661  * already been called at least once of the current request.
8662  * It increments the counter and is useful when a single request
8663  * is split into several parts.  Each part causes an increment and
8664  * so needs a matching md_write_end().
8665  * Unlike md_write_start(), it is safe to call md_write_inc() inside
8666  * a spinlocked region.
8667  */
md_write_inc(struct mddev * mddev,struct bio * bi)8668 void md_write_inc(struct mddev *mddev, struct bio *bi)
8669 {
8670 	if (bio_data_dir(bi) != WRITE)
8671 		return;
8672 	WARN_ON_ONCE(mddev->in_sync || !md_is_rdwr(mddev));
8673 	percpu_ref_get(&mddev->writes_pending);
8674 }
8675 EXPORT_SYMBOL(md_write_inc);
8676 
md_write_end(struct mddev * mddev)8677 void md_write_end(struct mddev *mddev)
8678 {
8679 	percpu_ref_put(&mddev->writes_pending);
8680 
8681 	if (mddev->safemode == 2)
8682 		md_wakeup_thread(mddev->thread);
8683 	else if (mddev->safemode_delay)
8684 		/* The roundup() ensures this only performs locking once
8685 		 * every ->safemode_delay jiffies
8686 		 */
8687 		mod_timer(&mddev->safemode_timer,
8688 			  roundup(jiffies, mddev->safemode_delay) +
8689 			  mddev->safemode_delay);
8690 }
8691 
8692 EXPORT_SYMBOL(md_write_end);
8693 
8694 /* This is used by raid0 and raid10 */
md_submit_discard_bio(struct mddev * mddev,struct md_rdev * rdev,struct bio * bio,sector_t start,sector_t size)8695 void md_submit_discard_bio(struct mddev *mddev, struct md_rdev *rdev,
8696 			struct bio *bio, sector_t start, sector_t size)
8697 {
8698 	struct bio *discard_bio = NULL;
8699 
8700 	if (__blkdev_issue_discard(rdev->bdev, start, size, GFP_NOIO,
8701 			&discard_bio) || !discard_bio)
8702 		return;
8703 
8704 	bio_chain(discard_bio, bio);
8705 	bio_clone_blkg_association(discard_bio, bio);
8706 	if (mddev->gendisk)
8707 		trace_block_bio_remap(discard_bio,
8708 				disk_devt(mddev->gendisk),
8709 				bio->bi_iter.bi_sector);
8710 	submit_bio_noacct(discard_bio);
8711 }
8712 EXPORT_SYMBOL_GPL(md_submit_discard_bio);
8713 
md_bitmap_start(struct mddev * mddev,struct md_io_clone * md_io_clone)8714 static void md_bitmap_start(struct mddev *mddev,
8715 			    struct md_io_clone *md_io_clone)
8716 {
8717 	if (mddev->pers->bitmap_sector)
8718 		mddev->pers->bitmap_sector(mddev, &md_io_clone->offset,
8719 					   &md_io_clone->sectors);
8720 
8721 	md_bitmap_startwrite(mddev->bitmap, md_io_clone->offset,
8722 			     md_io_clone->sectors);
8723 }
8724 
md_bitmap_end(struct mddev * mddev,struct md_io_clone * md_io_clone)8725 static void md_bitmap_end(struct mddev *mddev, struct md_io_clone *md_io_clone)
8726 {
8727 	md_bitmap_endwrite(mddev->bitmap, md_io_clone->offset,
8728 			   md_io_clone->sectors);
8729 }
8730 
md_end_clone_io(struct bio * bio)8731 static void md_end_clone_io(struct bio *bio)
8732 {
8733 	struct md_io_clone *md_io_clone = bio->bi_private;
8734 	struct bio *orig_bio = md_io_clone->orig_bio;
8735 	struct mddev *mddev = md_io_clone->mddev;
8736 
8737 	if (bio_data_dir(orig_bio) == WRITE && mddev->bitmap)
8738 		md_bitmap_end(mddev, md_io_clone);
8739 
8740 	if (bio->bi_status && !orig_bio->bi_status)
8741 		orig_bio->bi_status = bio->bi_status;
8742 
8743 	if (md_io_clone->start_time)
8744 		bio_end_io_acct(orig_bio, md_io_clone->start_time);
8745 
8746 	bio_put(bio);
8747 	bio_endio(orig_bio);
8748 	percpu_ref_put(&mddev->active_io);
8749 }
8750 
md_clone_bio(struct mddev * mddev,struct bio ** bio)8751 static void md_clone_bio(struct mddev *mddev, struct bio **bio)
8752 {
8753 	struct block_device *bdev = (*bio)->bi_bdev;
8754 	struct md_io_clone *md_io_clone;
8755 	struct bio *clone =
8756 		bio_alloc_clone(bdev, *bio, GFP_NOIO, &mddev->io_clone_set);
8757 
8758 	md_io_clone = container_of(clone, struct md_io_clone, bio_clone);
8759 	md_io_clone->orig_bio = *bio;
8760 	md_io_clone->mddev = mddev;
8761 	if (blk_queue_io_stat(bdev->bd_disk->queue))
8762 		md_io_clone->start_time = bio_start_io_acct(*bio);
8763 
8764 	if (bio_data_dir(*bio) == WRITE && mddev->bitmap) {
8765 		md_io_clone->offset = (*bio)->bi_iter.bi_sector;
8766 		md_io_clone->sectors = bio_sectors(*bio);
8767 		md_bitmap_start(mddev, md_io_clone);
8768 	}
8769 
8770 	clone->bi_end_io = md_end_clone_io;
8771 	clone->bi_private = md_io_clone;
8772 	*bio = clone;
8773 }
8774 
md_account_bio(struct mddev * mddev,struct bio ** bio)8775 void md_account_bio(struct mddev *mddev, struct bio **bio)
8776 {
8777 	percpu_ref_get(&mddev->active_io);
8778 	md_clone_bio(mddev, bio);
8779 }
8780 EXPORT_SYMBOL_GPL(md_account_bio);
8781 
8782 /* md_allow_write(mddev)
8783  * Calling this ensures that the array is marked 'active' so that writes
8784  * may proceed without blocking.  It is important to call this before
8785  * attempting a GFP_KERNEL allocation while holding the mddev lock.
8786  * Must be called with mddev_lock held.
8787  */
md_allow_write(struct mddev * mddev)8788 void md_allow_write(struct mddev *mddev)
8789 {
8790 	if (!mddev->pers)
8791 		return;
8792 	if (!md_is_rdwr(mddev))
8793 		return;
8794 	if (!mddev->pers->sync_request)
8795 		return;
8796 
8797 	spin_lock(&mddev->lock);
8798 	if (mddev->in_sync) {
8799 		mddev->in_sync = 0;
8800 		set_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags);
8801 		set_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags);
8802 		if (mddev->safemode_delay &&
8803 		    mddev->safemode == 0)
8804 			mddev->safemode = 1;
8805 		spin_unlock(&mddev->lock);
8806 		md_update_sb(mddev, 0);
8807 		sysfs_notify_dirent_safe(mddev->sysfs_state);
8808 		/* wait for the dirty state to be recorded in the metadata */
8809 		wait_event(mddev->sb_wait,
8810 			   !test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags));
8811 	} else
8812 		spin_unlock(&mddev->lock);
8813 }
8814 EXPORT_SYMBOL_GPL(md_allow_write);
8815 
8816 #define SYNC_MARKS	10
8817 #define	SYNC_MARK_STEP	(3*HZ)
8818 #define UPDATE_FREQUENCY (5*60*HZ)
md_do_sync(struct md_thread * thread)8819 void md_do_sync(struct md_thread *thread)
8820 {
8821 	struct mddev *mddev = thread->mddev;
8822 	struct mddev *mddev2;
8823 	unsigned int currspeed = 0, window;
8824 	sector_t max_sectors,j, io_sectors, recovery_done;
8825 	unsigned long mark[SYNC_MARKS];
8826 	unsigned long update_time;
8827 	sector_t mark_cnt[SYNC_MARKS];
8828 	int last_mark,m;
8829 	sector_t last_check;
8830 	int skipped = 0;
8831 	struct md_rdev *rdev;
8832 	char *desc, *action = NULL;
8833 	struct blk_plug plug;
8834 	int ret;
8835 
8836 	/* just incase thread restarts... */
8837 	if (test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
8838 	    test_bit(MD_RECOVERY_WAIT, &mddev->recovery))
8839 		return;
8840 	if (!md_is_rdwr(mddev)) {/* never try to sync a read-only array */
8841 		set_bit(MD_RECOVERY_INTR, &mddev->recovery);
8842 		return;
8843 	}
8844 
8845 	if (mddev_is_clustered(mddev)) {
8846 		ret = md_cluster_ops->resync_start(mddev);
8847 		if (ret)
8848 			goto skip;
8849 
8850 		set_bit(MD_CLUSTER_RESYNC_LOCKED, &mddev->flags);
8851 		if (!(test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ||
8852 			test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) ||
8853 			test_bit(MD_RECOVERY_RECOVER, &mddev->recovery))
8854 		     && ((unsigned long long)mddev->curr_resync_completed
8855 			 < (unsigned long long)mddev->resync_max_sectors))
8856 			goto skip;
8857 	}
8858 
8859 	if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
8860 		if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)) {
8861 			desc = "data-check";
8862 			action = "check";
8863 		} else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
8864 			desc = "requested-resync";
8865 			action = "repair";
8866 		} else
8867 			desc = "resync";
8868 	} else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
8869 		desc = "reshape";
8870 	else
8871 		desc = "recovery";
8872 
8873 	mddev->last_sync_action = action ?: desc;
8874 
8875 	/*
8876 	 * Before starting a resync we must have set curr_resync to
8877 	 * 2, and then checked that every "conflicting" array has curr_resync
8878 	 * less than ours.  When we find one that is the same or higher
8879 	 * we wait on resync_wait.  To avoid deadlock, we reduce curr_resync
8880 	 * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
8881 	 * This will mean we have to start checking from the beginning again.
8882 	 *
8883 	 */
8884 
8885 	do {
8886 		int mddev2_minor = -1;
8887 		mddev->curr_resync = MD_RESYNC_DELAYED;
8888 
8889 	try_again:
8890 		if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
8891 			goto skip;
8892 		spin_lock(&all_mddevs_lock);
8893 		list_for_each_entry(mddev2, &all_mddevs, all_mddevs) {
8894 			if (test_bit(MD_DELETED, &mddev2->flags))
8895 				continue;
8896 			if (mddev2 == mddev)
8897 				continue;
8898 			if (!mddev->parallel_resync
8899 			&&  mddev2->curr_resync
8900 			&&  match_mddev_units(mddev, mddev2)) {
8901 				DEFINE_WAIT(wq);
8902 				if (mddev < mddev2 &&
8903 				    mddev->curr_resync == MD_RESYNC_DELAYED) {
8904 					/* arbitrarily yield */
8905 					mddev->curr_resync = MD_RESYNC_YIELDED;
8906 					wake_up(&resync_wait);
8907 				}
8908 				if (mddev > mddev2 &&
8909 				    mddev->curr_resync == MD_RESYNC_YIELDED)
8910 					/* no need to wait here, we can wait the next
8911 					 * time 'round when curr_resync == 2
8912 					 */
8913 					continue;
8914 				/* We need to wait 'interruptible' so as not to
8915 				 * contribute to the load average, and not to
8916 				 * be caught by 'softlockup'
8917 				 */
8918 				prepare_to_wait(&resync_wait, &wq, TASK_INTERRUPTIBLE);
8919 				if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
8920 				    mddev2->curr_resync >= mddev->curr_resync) {
8921 					if (mddev2_minor != mddev2->md_minor) {
8922 						mddev2_minor = mddev2->md_minor;
8923 						pr_info("md: delaying %s of %s until %s has finished (they share one or more physical units)\n",
8924 							desc, mdname(mddev),
8925 							mdname(mddev2));
8926 					}
8927 					spin_unlock(&all_mddevs_lock);
8928 
8929 					if (signal_pending(current))
8930 						flush_signals(current);
8931 					schedule();
8932 					finish_wait(&resync_wait, &wq);
8933 					goto try_again;
8934 				}
8935 				finish_wait(&resync_wait, &wq);
8936 			}
8937 		}
8938 		spin_unlock(&all_mddevs_lock);
8939 	} while (mddev->curr_resync < MD_RESYNC_DELAYED);
8940 
8941 	j = 0;
8942 	if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
8943 		/* resync follows the size requested by the personality,
8944 		 * which defaults to physical size, but can be virtual size
8945 		 */
8946 		max_sectors = mddev->resync_max_sectors;
8947 		atomic64_set(&mddev->resync_mismatches, 0);
8948 		/* we don't use the checkpoint if there's a bitmap */
8949 		if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
8950 			j = mddev->resync_min;
8951 		else if (!mddev->bitmap)
8952 			j = mddev->recovery_cp;
8953 
8954 	} else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) {
8955 		max_sectors = mddev->resync_max_sectors;
8956 		/*
8957 		 * If the original node aborts reshaping then we continue the
8958 		 * reshaping, so set j again to avoid restart reshape from the
8959 		 * first beginning
8960 		 */
8961 		if (mddev_is_clustered(mddev) &&
8962 		    mddev->reshape_position != MaxSector)
8963 			j = mddev->reshape_position;
8964 	} else {
8965 		/* recovery follows the physical size of devices */
8966 		max_sectors = mddev->dev_sectors;
8967 		j = MaxSector;
8968 		rcu_read_lock();
8969 		rdev_for_each_rcu(rdev, mddev)
8970 			if (rdev->raid_disk >= 0 &&
8971 			    !test_bit(Journal, &rdev->flags) &&
8972 			    !test_bit(Faulty, &rdev->flags) &&
8973 			    !test_bit(In_sync, &rdev->flags) &&
8974 			    rdev->recovery_offset < j)
8975 				j = rdev->recovery_offset;
8976 		rcu_read_unlock();
8977 
8978 		/* If there is a bitmap, we need to make sure all
8979 		 * writes that started before we added a spare
8980 		 * complete before we start doing a recovery.
8981 		 * Otherwise the write might complete and (via
8982 		 * bitmap_endwrite) set a bit in the bitmap after the
8983 		 * recovery has checked that bit and skipped that
8984 		 * region.
8985 		 */
8986 		if (mddev->bitmap) {
8987 			mddev->pers->quiesce(mddev, 1);
8988 			mddev->pers->quiesce(mddev, 0);
8989 		}
8990 	}
8991 
8992 	pr_info("md: %s of RAID array %s\n", desc, mdname(mddev));
8993 	pr_debug("md: minimum _guaranteed_  speed: %d KB/sec/disk.\n", speed_min(mddev));
8994 	pr_debug("md: using maximum available idle IO bandwidth (but not more than %d KB/sec) for %s.\n",
8995 		 speed_max(mddev), desc);
8996 
8997 	is_mddev_idle(mddev, 1); /* this initializes IO event counters */
8998 
8999 	io_sectors = 0;
9000 	for (m = 0; m < SYNC_MARKS; m++) {
9001 		mark[m] = jiffies;
9002 		mark_cnt[m] = io_sectors;
9003 	}
9004 	last_mark = 0;
9005 	mddev->resync_mark = mark[last_mark];
9006 	mddev->resync_mark_cnt = mark_cnt[last_mark];
9007 
9008 	/*
9009 	 * Tune reconstruction:
9010 	 */
9011 	window = 32 * (PAGE_SIZE / 512);
9012 	pr_debug("md: using %dk window, over a total of %lluk.\n",
9013 		 window/2, (unsigned long long)max_sectors/2);
9014 
9015 	atomic_set(&mddev->recovery_active, 0);
9016 	last_check = 0;
9017 
9018 	if (j >= MD_RESYNC_ACTIVE) {
9019 		pr_debug("md: resuming %s of %s from checkpoint.\n",
9020 			 desc, mdname(mddev));
9021 		mddev->curr_resync = j;
9022 	} else
9023 		mddev->curr_resync = MD_RESYNC_ACTIVE; /* no longer delayed */
9024 	mddev->curr_resync_completed = j;
9025 	sysfs_notify_dirent_safe(mddev->sysfs_completed);
9026 	md_new_event();
9027 	update_time = jiffies;
9028 
9029 	blk_start_plug(&plug);
9030 	while (j < max_sectors) {
9031 		sector_t sectors;
9032 
9033 		skipped = 0;
9034 
9035 		if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
9036 		    ((mddev->curr_resync > mddev->curr_resync_completed &&
9037 		      (mddev->curr_resync - mddev->curr_resync_completed)
9038 		      > (max_sectors >> 4)) ||
9039 		     time_after_eq(jiffies, update_time + UPDATE_FREQUENCY) ||
9040 		     (j - mddev->curr_resync_completed)*2
9041 		     >= mddev->resync_max - mddev->curr_resync_completed ||
9042 		     mddev->curr_resync_completed > mddev->resync_max
9043 			    )) {
9044 			/* time to update curr_resync_completed */
9045 			wait_event(mddev->recovery_wait,
9046 				   atomic_read(&mddev->recovery_active) == 0);
9047 			mddev->curr_resync_completed = j;
9048 			if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) &&
9049 			    j > mddev->recovery_cp)
9050 				mddev->recovery_cp = j;
9051 			update_time = jiffies;
9052 			set_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags);
9053 			sysfs_notify_dirent_safe(mddev->sysfs_completed);
9054 		}
9055 
9056 		while (j >= mddev->resync_max &&
9057 		       !test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
9058 			/* As this condition is controlled by user-space,
9059 			 * we can block indefinitely, so use '_interruptible'
9060 			 * to avoid triggering warnings.
9061 			 */
9062 			flush_signals(current); /* just in case */
9063 			wait_event_interruptible(mddev->recovery_wait,
9064 						 mddev->resync_max > j
9065 						 || test_bit(MD_RECOVERY_INTR,
9066 							     &mddev->recovery));
9067 		}
9068 
9069 		if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
9070 			break;
9071 
9072 		sectors = mddev->pers->sync_request(mddev, j, &skipped);
9073 		if (sectors == 0) {
9074 			set_bit(MD_RECOVERY_INTR, &mddev->recovery);
9075 			break;
9076 		}
9077 
9078 		if (!skipped) { /* actual IO requested */
9079 			io_sectors += sectors;
9080 			atomic_add(sectors, &mddev->recovery_active);
9081 		}
9082 
9083 		if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
9084 			break;
9085 
9086 		j += sectors;
9087 		if (j > max_sectors)
9088 			/* when skipping, extra large numbers can be returned. */
9089 			j = max_sectors;
9090 		if (j >= MD_RESYNC_ACTIVE)
9091 			mddev->curr_resync = j;
9092 		mddev->curr_mark_cnt = io_sectors;
9093 		if (last_check == 0)
9094 			/* this is the earliest that rebuild will be
9095 			 * visible in /proc/mdstat
9096 			 */
9097 			md_new_event();
9098 
9099 		if (last_check + window > io_sectors || j == max_sectors)
9100 			continue;
9101 
9102 		last_check = io_sectors;
9103 	repeat:
9104 		if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) {
9105 			/* step marks */
9106 			int next = (last_mark+1) % SYNC_MARKS;
9107 
9108 			mddev->resync_mark = mark[next];
9109 			mddev->resync_mark_cnt = mark_cnt[next];
9110 			mark[next] = jiffies;
9111 			mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active);
9112 			last_mark = next;
9113 		}
9114 
9115 		if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
9116 			break;
9117 
9118 		/*
9119 		 * this loop exits only if either when we are slower than
9120 		 * the 'hard' speed limit, or the system was IO-idle for
9121 		 * a jiffy.
9122 		 * the system might be non-idle CPU-wise, but we only care
9123 		 * about not overloading the IO subsystem. (things like an
9124 		 * e2fsck being done on the RAID array should execute fast)
9125 		 */
9126 		cond_resched();
9127 
9128 		recovery_done = io_sectors - atomic_read(&mddev->recovery_active);
9129 		currspeed = ((unsigned long)(recovery_done - mddev->resync_mark_cnt))/2
9130 			/((jiffies-mddev->resync_mark)/HZ +1) +1;
9131 
9132 		if (currspeed > speed_min(mddev)) {
9133 			if (currspeed > speed_max(mddev)) {
9134 				msleep(500);
9135 				goto repeat;
9136 			}
9137 			if (!is_mddev_idle(mddev, 0)) {
9138 				/*
9139 				 * Give other IO more of a chance.
9140 				 * The faster the devices, the less we wait.
9141 				 */
9142 				wait_event(mddev->recovery_wait,
9143 					   !atomic_read(&mddev->recovery_active));
9144 			}
9145 		}
9146 	}
9147 	pr_info("md: %s: %s %s.\n",mdname(mddev), desc,
9148 		test_bit(MD_RECOVERY_INTR, &mddev->recovery)
9149 		? "interrupted" : "done");
9150 	/*
9151 	 * this also signals 'finished resyncing' to md_stop
9152 	 */
9153 	blk_finish_plug(&plug);
9154 	wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
9155 
9156 	if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
9157 	    !test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
9158 	    mddev->curr_resync >= MD_RESYNC_ACTIVE) {
9159 		mddev->curr_resync_completed = mddev->curr_resync;
9160 		sysfs_notify_dirent_safe(mddev->sysfs_completed);
9161 	}
9162 	mddev->pers->sync_request(mddev, max_sectors, &skipped);
9163 
9164 	if (!test_bit(MD_RECOVERY_CHECK, &mddev->recovery) &&
9165 	    mddev->curr_resync > MD_RESYNC_ACTIVE) {
9166 		if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
9167 			if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
9168 				if (mddev->curr_resync >= mddev->recovery_cp) {
9169 					pr_debug("md: checkpointing %s of %s.\n",
9170 						 desc, mdname(mddev));
9171 					if (test_bit(MD_RECOVERY_ERROR,
9172 						&mddev->recovery))
9173 						mddev->recovery_cp =
9174 							mddev->curr_resync_completed;
9175 					else
9176 						mddev->recovery_cp =
9177 							mddev->curr_resync;
9178 				}
9179 			} else
9180 				mddev->recovery_cp = MaxSector;
9181 		} else {
9182 			if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery))
9183 				mddev->curr_resync = MaxSector;
9184 			if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
9185 			    test_bit(MD_RECOVERY_RECOVER, &mddev->recovery)) {
9186 				rcu_read_lock();
9187 				rdev_for_each_rcu(rdev, mddev)
9188 					if (rdev->raid_disk >= 0 &&
9189 					    mddev->delta_disks >= 0 &&
9190 					    !test_bit(Journal, &rdev->flags) &&
9191 					    !test_bit(Faulty, &rdev->flags) &&
9192 					    !test_bit(In_sync, &rdev->flags) &&
9193 					    rdev->recovery_offset < mddev->curr_resync)
9194 						rdev->recovery_offset = mddev->curr_resync;
9195 				rcu_read_unlock();
9196 			}
9197 		}
9198 	}
9199  skip:
9200 	/* set CHANGE_PENDING here since maybe another update is needed,
9201 	 * so other nodes are informed. It should be harmless for normal
9202 	 * raid */
9203 	set_mask_bits(&mddev->sb_flags, 0,
9204 		      BIT(MD_SB_CHANGE_PENDING) | BIT(MD_SB_CHANGE_DEVS));
9205 
9206 	if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
9207 			!test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
9208 			mddev->delta_disks > 0 &&
9209 			mddev->pers->finish_reshape &&
9210 			mddev->pers->size &&
9211 			mddev->queue) {
9212 		mddev_lock_nointr(mddev);
9213 		md_set_array_sectors(mddev, mddev->pers->size(mddev, 0, 0));
9214 		mddev_unlock(mddev);
9215 		if (!mddev_is_clustered(mddev))
9216 			set_capacity_and_notify(mddev->gendisk,
9217 						mddev->array_sectors);
9218 	}
9219 
9220 	spin_lock(&mddev->lock);
9221 	if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
9222 		/* We completed so min/max setting can be forgotten if used. */
9223 		if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
9224 			mddev->resync_min = 0;
9225 		mddev->resync_max = MaxSector;
9226 	} else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
9227 		mddev->resync_min = mddev->curr_resync_completed;
9228 	set_bit(MD_RECOVERY_DONE, &mddev->recovery);
9229 	mddev->curr_resync = MD_RESYNC_NONE;
9230 	spin_unlock(&mddev->lock);
9231 
9232 	wake_up(&resync_wait);
9233 	wake_up(&mddev->sb_wait);
9234 	md_wakeup_thread(mddev->thread);
9235 	return;
9236 }
9237 EXPORT_SYMBOL_GPL(md_do_sync);
9238 
remove_and_add_spares(struct mddev * mddev,struct md_rdev * this)9239 static int remove_and_add_spares(struct mddev *mddev,
9240 				 struct md_rdev *this)
9241 {
9242 	struct md_rdev *rdev;
9243 	int spares = 0;
9244 	int removed = 0;
9245 	bool remove_some = false;
9246 
9247 	if (this && test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
9248 		/* Mustn't remove devices when resync thread is running */
9249 		return 0;
9250 
9251 	rdev_for_each(rdev, mddev) {
9252 		if ((this == NULL || rdev == this) &&
9253 		    rdev->raid_disk >= 0 &&
9254 		    !test_bit(Blocked, &rdev->flags) &&
9255 		    test_bit(Faulty, &rdev->flags) &&
9256 		    atomic_read(&rdev->nr_pending)==0) {
9257 			/* Faulty non-Blocked devices with nr_pending == 0
9258 			 * never get nr_pending incremented,
9259 			 * never get Faulty cleared, and never get Blocked set.
9260 			 * So we can synchronize_rcu now rather than once per device
9261 			 */
9262 			remove_some = true;
9263 			set_bit(RemoveSynchronized, &rdev->flags);
9264 		}
9265 	}
9266 
9267 	if (remove_some)
9268 		synchronize_rcu();
9269 	rdev_for_each(rdev, mddev) {
9270 		if ((this == NULL || rdev == this) &&
9271 		    rdev->raid_disk >= 0 &&
9272 		    !test_bit(Blocked, &rdev->flags) &&
9273 		    ((test_bit(RemoveSynchronized, &rdev->flags) ||
9274 		     (!test_bit(In_sync, &rdev->flags) &&
9275 		      !test_bit(Journal, &rdev->flags))) &&
9276 		    atomic_read(&rdev->nr_pending)==0)) {
9277 			if (mddev->pers->hot_remove_disk(
9278 				    mddev, rdev) == 0) {
9279 				sysfs_unlink_rdev(mddev, rdev);
9280 				rdev->saved_raid_disk = rdev->raid_disk;
9281 				rdev->raid_disk = -1;
9282 				removed++;
9283 			}
9284 		}
9285 		if (remove_some && test_bit(RemoveSynchronized, &rdev->flags))
9286 			clear_bit(RemoveSynchronized, &rdev->flags);
9287 	}
9288 
9289 	if (removed && mddev->kobj.sd)
9290 		sysfs_notify_dirent_safe(mddev->sysfs_degraded);
9291 
9292 	if (this && removed)
9293 		goto no_add;
9294 
9295 	rdev_for_each(rdev, mddev) {
9296 		if (this && this != rdev)
9297 			continue;
9298 		if (test_bit(Candidate, &rdev->flags))
9299 			continue;
9300 		if (rdev->raid_disk >= 0 &&
9301 		    !test_bit(In_sync, &rdev->flags) &&
9302 		    !test_bit(Journal, &rdev->flags) &&
9303 		    !test_bit(Faulty, &rdev->flags))
9304 			spares++;
9305 		if (rdev->raid_disk >= 0)
9306 			continue;
9307 		if (test_bit(Faulty, &rdev->flags))
9308 			continue;
9309 		if (!test_bit(Journal, &rdev->flags)) {
9310 			if (!md_is_rdwr(mddev) &&
9311 			    !(rdev->saved_raid_disk >= 0 &&
9312 			      !test_bit(Bitmap_sync, &rdev->flags)))
9313 				continue;
9314 
9315 			rdev->recovery_offset = 0;
9316 		}
9317 		if (mddev->pers->hot_add_disk(mddev, rdev) == 0) {
9318 			/* failure here is OK */
9319 			sysfs_link_rdev(mddev, rdev);
9320 			if (!test_bit(Journal, &rdev->flags))
9321 				spares++;
9322 			md_new_event();
9323 			set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
9324 		}
9325 	}
9326 no_add:
9327 	if (removed)
9328 		set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
9329 	return spares;
9330 }
9331 
md_start_sync(struct work_struct * ws)9332 static void md_start_sync(struct work_struct *ws)
9333 {
9334 	struct mddev *mddev = container_of(ws, struct mddev, sync_work);
9335 
9336 	rcu_assign_pointer(mddev->sync_thread,
9337 			   md_register_thread(md_do_sync, mddev, "resync"));
9338 	if (!mddev->sync_thread) {
9339 		pr_warn("%s: could not start resync thread...\n",
9340 			mdname(mddev));
9341 		/* leave the spares where they are, it shouldn't hurt */
9342 		clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
9343 		clear_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
9344 		clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
9345 		clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
9346 		clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
9347 		wake_up(&resync_wait);
9348 		if (test_and_clear_bit(MD_RECOVERY_RECOVER,
9349 				       &mddev->recovery))
9350 			if (mddev->sysfs_action)
9351 				sysfs_notify_dirent_safe(mddev->sysfs_action);
9352 	} else
9353 		md_wakeup_thread(mddev->sync_thread);
9354 	sysfs_notify_dirent_safe(mddev->sysfs_action);
9355 	md_new_event();
9356 }
9357 
9358 /*
9359  * This routine is regularly called by all per-raid-array threads to
9360  * deal with generic issues like resync and super-block update.
9361  * Raid personalities that don't have a thread (linear/raid0) do not
9362  * need this as they never do any recovery or update the superblock.
9363  *
9364  * It does not do any resync itself, but rather "forks" off other threads
9365  * to do that as needed.
9366  * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
9367  * "->recovery" and create a thread at ->sync_thread.
9368  * When the thread finishes it sets MD_RECOVERY_DONE
9369  * and wakeups up this thread which will reap the thread and finish up.
9370  * This thread also removes any faulty devices (with nr_pending == 0).
9371  *
9372  * The overall approach is:
9373  *  1/ if the superblock needs updating, update it.
9374  *  2/ If a recovery thread is running, don't do anything else.
9375  *  3/ If recovery has finished, clean up, possibly marking spares active.
9376  *  4/ If there are any faulty devices, remove them.
9377  *  5/ If array is degraded, try to add spares devices
9378  *  6/ If array has spares or is not in-sync, start a resync thread.
9379  */
md_check_recovery(struct mddev * mddev)9380 void md_check_recovery(struct mddev *mddev)
9381 {
9382 	if (test_bit(MD_ALLOW_SB_UPDATE, &mddev->flags) && mddev->sb_flags) {
9383 		/* Write superblock - thread that called mddev_suspend()
9384 		 * holds reconfig_mutex for us.
9385 		 */
9386 		set_bit(MD_UPDATING_SB, &mddev->flags);
9387 		smp_mb__after_atomic();
9388 		if (test_bit(MD_ALLOW_SB_UPDATE, &mddev->flags))
9389 			md_update_sb(mddev, 0);
9390 		clear_bit_unlock(MD_UPDATING_SB, &mddev->flags);
9391 		wake_up(&mddev->sb_wait);
9392 	}
9393 
9394 	if (is_md_suspended(mddev))
9395 		return;
9396 
9397 	if (mddev->bitmap)
9398 		md_bitmap_daemon_work(mddev);
9399 
9400 	if (signal_pending(current)) {
9401 		if (mddev->pers->sync_request && !mddev->external) {
9402 			pr_debug("md: %s in immediate safe mode\n",
9403 				 mdname(mddev));
9404 			mddev->safemode = 2;
9405 		}
9406 		flush_signals(current);
9407 	}
9408 
9409 	if (!md_is_rdwr(mddev) &&
9410 	    !test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
9411 		return;
9412 	if ( ! (
9413 		(mddev->sb_flags & ~ (1<<MD_SB_CHANGE_PENDING)) ||
9414 		test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
9415 		test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
9416 		(mddev->external == 0 && mddev->safemode == 1) ||
9417 		(mddev->safemode == 2
9418 		 && !mddev->in_sync && mddev->recovery_cp == MaxSector)
9419 		))
9420 		return;
9421 
9422 	if (mddev_trylock(mddev)) {
9423 		int spares = 0;
9424 		bool try_set_sync = mddev->safemode != 0;
9425 
9426 		if (!mddev->external && mddev->safemode == 1)
9427 			mddev->safemode = 0;
9428 
9429 		if (!md_is_rdwr(mddev)) {
9430 			struct md_rdev *rdev;
9431 			if (!mddev->external && mddev->in_sync)
9432 				/* 'Blocked' flag not needed as failed devices
9433 				 * will be recorded if array switched to read/write.
9434 				 * Leaving it set will prevent the device
9435 				 * from being removed.
9436 				 */
9437 				rdev_for_each(rdev, mddev)
9438 					clear_bit(Blocked, &rdev->flags);
9439 			/* On a read-only array we can:
9440 			 * - remove failed devices
9441 			 * - add already-in_sync devices if the array itself
9442 			 *   is in-sync.
9443 			 * As we only add devices that are already in-sync,
9444 			 * we can activate the spares immediately.
9445 			 */
9446 			remove_and_add_spares(mddev, NULL);
9447 			/* There is no thread, but we need to call
9448 			 * ->spare_active and clear saved_raid_disk
9449 			 */
9450 			set_bit(MD_RECOVERY_INTR, &mddev->recovery);
9451 			md_reap_sync_thread(mddev);
9452 			clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
9453 			clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
9454 			clear_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags);
9455 			goto unlock;
9456 		}
9457 
9458 		if (mddev_is_clustered(mddev)) {
9459 			struct md_rdev *rdev, *tmp;
9460 			/* kick the device if another node issued a
9461 			 * remove disk.
9462 			 */
9463 			rdev_for_each_safe(rdev, tmp, mddev) {
9464 				if (test_and_clear_bit(ClusterRemove, &rdev->flags) &&
9465 						rdev->raid_disk < 0)
9466 					md_kick_rdev_from_array(rdev);
9467 			}
9468 		}
9469 
9470 		if (try_set_sync && !mddev->external && !mddev->in_sync) {
9471 			spin_lock(&mddev->lock);
9472 			set_in_sync(mddev);
9473 			spin_unlock(&mddev->lock);
9474 		}
9475 
9476 		if (mddev->sb_flags)
9477 			md_update_sb(mddev, 0);
9478 
9479 		/*
9480 		 * Never start a new sync thread if MD_RECOVERY_RUNNING is
9481 		 * still set.
9482 		 */
9483 		if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) {
9484 			if (!test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
9485 				/* resync/recovery still happening */
9486 				clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
9487 				goto unlock;
9488 			}
9489 
9490 			if (WARN_ON_ONCE(!mddev->sync_thread))
9491 				goto unlock;
9492 
9493 			md_reap_sync_thread(mddev);
9494 			goto unlock;
9495 		}
9496 
9497 		/* Set RUNNING before clearing NEEDED to avoid
9498 		 * any transients in the value of "sync_action".
9499 		 */
9500 		mddev->curr_resync_completed = 0;
9501 		spin_lock(&mddev->lock);
9502 		set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
9503 		spin_unlock(&mddev->lock);
9504 		/* Clear some bits that don't mean anything, but
9505 		 * might be left set
9506 		 */
9507 		clear_bit(MD_RECOVERY_INTR, &mddev->recovery);
9508 		clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
9509 
9510 		if (!test_and_clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
9511 		    test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
9512 			goto not_running;
9513 		/* no recovery is running.
9514 		 * remove any failed drives, then
9515 		 * add spares if possible.
9516 		 * Spares are also removed and re-added, to allow
9517 		 * the personality to fail the re-add.
9518 		 */
9519 
9520 		if (mddev->reshape_position != MaxSector) {
9521 			if (mddev->pers->check_reshape == NULL ||
9522 			    mddev->pers->check_reshape(mddev) != 0)
9523 				/* Cannot proceed */
9524 				goto not_running;
9525 			set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
9526 			clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
9527 		} else if ((spares = remove_and_add_spares(mddev, NULL))) {
9528 			clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
9529 			clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
9530 			clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
9531 			set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
9532 		} else if (mddev->recovery_cp < MaxSector) {
9533 			set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
9534 			clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
9535 		} else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
9536 			/* nothing to be done ... */
9537 			goto not_running;
9538 
9539 		if (mddev->pers->sync_request) {
9540 			if (spares) {
9541 				/* We are adding a device or devices to an array
9542 				 * which has the bitmap stored on all devices.
9543 				 * So make sure all bitmap pages get written
9544 				 */
9545 				md_bitmap_write_all(mddev->bitmap);
9546 			}
9547 			queue_work(md_misc_wq, &mddev->sync_work);
9548 			goto unlock;
9549 		}
9550 	not_running:
9551 		if (!mddev->sync_thread) {
9552 			clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
9553 			wake_up(&resync_wait);
9554 			if (test_and_clear_bit(MD_RECOVERY_RECOVER,
9555 					       &mddev->recovery))
9556 				if (mddev->sysfs_action)
9557 					sysfs_notify_dirent_safe(mddev->sysfs_action);
9558 		}
9559 	unlock:
9560 		wake_up(&mddev->sb_wait);
9561 		mddev_unlock(mddev);
9562 	}
9563 }
9564 EXPORT_SYMBOL(md_check_recovery);
9565 
md_reap_sync_thread(struct mddev * mddev)9566 void md_reap_sync_thread(struct mddev *mddev)
9567 {
9568 	struct md_rdev *rdev;
9569 	sector_t old_dev_sectors = mddev->dev_sectors;
9570 	bool is_reshaped = false;
9571 
9572 	/* resync has finished, collect result */
9573 	md_unregister_thread(mddev, &mddev->sync_thread);
9574 	atomic_inc(&mddev->sync_seq);
9575 
9576 	if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
9577 	    !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery) &&
9578 	    mddev->degraded != mddev->raid_disks) {
9579 		/* success...*/
9580 		/* activate any spares */
9581 		if (mddev->pers->spare_active(mddev)) {
9582 			sysfs_notify_dirent_safe(mddev->sysfs_degraded);
9583 			set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
9584 		}
9585 	}
9586 	if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
9587 	    mddev->pers->finish_reshape) {
9588 		mddev->pers->finish_reshape(mddev);
9589 		if (mddev_is_clustered(mddev))
9590 			is_reshaped = true;
9591 	}
9592 
9593 	/* If array is no-longer degraded, then any saved_raid_disk
9594 	 * information must be scrapped.
9595 	 */
9596 	if (!mddev->degraded)
9597 		rdev_for_each(rdev, mddev)
9598 			rdev->saved_raid_disk = -1;
9599 
9600 	md_update_sb(mddev, 1);
9601 	/* MD_SB_CHANGE_PENDING should be cleared by md_update_sb, so we can
9602 	 * call resync_finish here if MD_CLUSTER_RESYNC_LOCKED is set by
9603 	 * clustered raid */
9604 	if (test_and_clear_bit(MD_CLUSTER_RESYNC_LOCKED, &mddev->flags))
9605 		md_cluster_ops->resync_finish(mddev);
9606 	clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
9607 	clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
9608 	clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
9609 	clear_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
9610 	clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
9611 	clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
9612 	/*
9613 	 * We call md_cluster_ops->update_size here because sync_size could
9614 	 * be changed by md_update_sb, and MD_RECOVERY_RESHAPE is cleared,
9615 	 * so it is time to update size across cluster.
9616 	 */
9617 	if (mddev_is_clustered(mddev) && is_reshaped
9618 				      && !test_bit(MD_CLOSING, &mddev->flags))
9619 		md_cluster_ops->update_size(mddev, old_dev_sectors);
9620 	/* flag recovery needed just to double check */
9621 	set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
9622 	sysfs_notify_dirent_safe(mddev->sysfs_completed);
9623 	sysfs_notify_dirent_safe(mddev->sysfs_action);
9624 	md_new_event();
9625 	if (mddev->event_work.func)
9626 		queue_work(md_misc_wq, &mddev->event_work);
9627 	wake_up(&resync_wait);
9628 }
9629 EXPORT_SYMBOL(md_reap_sync_thread);
9630 
md_wait_for_blocked_rdev(struct md_rdev * rdev,struct mddev * mddev)9631 void md_wait_for_blocked_rdev(struct md_rdev *rdev, struct mddev *mddev)
9632 {
9633 	sysfs_notify_dirent_safe(rdev->sysfs_state);
9634 	wait_event_timeout(rdev->blocked_wait,
9635 			   !test_bit(Blocked, &rdev->flags) &&
9636 			   !test_bit(BlockedBadBlocks, &rdev->flags),
9637 			   msecs_to_jiffies(5000));
9638 	rdev_dec_pending(rdev, mddev);
9639 }
9640 EXPORT_SYMBOL(md_wait_for_blocked_rdev);
9641 
md_finish_reshape(struct mddev * mddev)9642 void md_finish_reshape(struct mddev *mddev)
9643 {
9644 	/* called be personality module when reshape completes. */
9645 	struct md_rdev *rdev;
9646 
9647 	rdev_for_each(rdev, mddev) {
9648 		if (rdev->data_offset > rdev->new_data_offset)
9649 			rdev->sectors += rdev->data_offset - rdev->new_data_offset;
9650 		else
9651 			rdev->sectors -= rdev->new_data_offset - rdev->data_offset;
9652 		rdev->data_offset = rdev->new_data_offset;
9653 	}
9654 }
9655 EXPORT_SYMBOL(md_finish_reshape);
9656 
9657 /* Bad block management */
9658 
9659 /* Returns 1 on success, 0 on failure */
rdev_set_badblocks(struct md_rdev * rdev,sector_t s,int sectors,int is_new)9660 int rdev_set_badblocks(struct md_rdev *rdev, sector_t s, int sectors,
9661 		       int is_new)
9662 {
9663 	struct mddev *mddev = rdev->mddev;
9664 	int rv;
9665 	if (is_new)
9666 		s += rdev->new_data_offset;
9667 	else
9668 		s += rdev->data_offset;
9669 	rv = badblocks_set(&rdev->badblocks, s, sectors, 0);
9670 	if (rv == 0) {
9671 		/* Make sure they get written out promptly */
9672 		if (test_bit(ExternalBbl, &rdev->flags))
9673 			sysfs_notify_dirent_safe(rdev->sysfs_unack_badblocks);
9674 		sysfs_notify_dirent_safe(rdev->sysfs_state);
9675 		set_mask_bits(&mddev->sb_flags, 0,
9676 			      BIT(MD_SB_CHANGE_CLEAN) | BIT(MD_SB_CHANGE_PENDING));
9677 		md_wakeup_thread(rdev->mddev->thread);
9678 		return 1;
9679 	} else
9680 		return 0;
9681 }
9682 EXPORT_SYMBOL_GPL(rdev_set_badblocks);
9683 
rdev_clear_badblocks(struct md_rdev * rdev,sector_t s,int sectors,int is_new)9684 int rdev_clear_badblocks(struct md_rdev *rdev, sector_t s, int sectors,
9685 			 int is_new)
9686 {
9687 	int rv;
9688 	if (is_new)
9689 		s += rdev->new_data_offset;
9690 	else
9691 		s += rdev->data_offset;
9692 	rv = badblocks_clear(&rdev->badblocks, s, sectors);
9693 	if ((rv == 0) && test_bit(ExternalBbl, &rdev->flags))
9694 		sysfs_notify_dirent_safe(rdev->sysfs_badblocks);
9695 	return rv;
9696 }
9697 EXPORT_SYMBOL_GPL(rdev_clear_badblocks);
9698 
md_notify_reboot(struct notifier_block * this,unsigned long code,void * x)9699 static int md_notify_reboot(struct notifier_block *this,
9700 			    unsigned long code, void *x)
9701 {
9702 	struct mddev *mddev, *n;
9703 	int need_delay = 0;
9704 
9705 	spin_lock(&all_mddevs_lock);
9706 	list_for_each_entry_safe(mddev, n, &all_mddevs, all_mddevs) {
9707 		if (!mddev_get(mddev))
9708 			continue;
9709 		spin_unlock(&all_mddevs_lock);
9710 		if (mddev_trylock(mddev)) {
9711 			if (mddev->pers)
9712 				__md_stop_writes(mddev);
9713 			if (mddev->persistent)
9714 				mddev->safemode = 2;
9715 			mddev_unlock(mddev);
9716 		}
9717 		need_delay = 1;
9718 		mddev_put(mddev);
9719 		spin_lock(&all_mddevs_lock);
9720 	}
9721 	spin_unlock(&all_mddevs_lock);
9722 
9723 	/*
9724 	 * certain more exotic SCSI devices are known to be
9725 	 * volatile wrt too early system reboots. While the
9726 	 * right place to handle this issue is the given
9727 	 * driver, we do want to have a safe RAID driver ...
9728 	 */
9729 	if (need_delay)
9730 		msleep(1000);
9731 
9732 	return NOTIFY_DONE;
9733 }
9734 
9735 static struct notifier_block md_notifier = {
9736 	.notifier_call	= md_notify_reboot,
9737 	.next		= NULL,
9738 	.priority	= INT_MAX, /* before any real devices */
9739 };
9740 
md_geninit(void)9741 static void md_geninit(void)
9742 {
9743 	pr_debug("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
9744 
9745 	proc_create("mdstat", S_IRUGO, NULL, &mdstat_proc_ops);
9746 }
9747 
md_init(void)9748 static int __init md_init(void)
9749 {
9750 	int ret = -ENOMEM;
9751 
9752 	md_wq = alloc_workqueue("md", WQ_MEM_RECLAIM, 0);
9753 	if (!md_wq)
9754 		goto err_wq;
9755 
9756 	md_misc_wq = alloc_workqueue("md_misc", 0, 0);
9757 	if (!md_misc_wq)
9758 		goto err_misc_wq;
9759 
9760 	md_bitmap_wq = alloc_workqueue("md_bitmap", WQ_MEM_RECLAIM | WQ_UNBOUND,
9761 				       0);
9762 	if (!md_bitmap_wq)
9763 		goto err_bitmap_wq;
9764 
9765 	ret = __register_blkdev(MD_MAJOR, "md", md_probe);
9766 	if (ret < 0)
9767 		goto err_md;
9768 
9769 	ret = __register_blkdev(0, "mdp", md_probe);
9770 	if (ret < 0)
9771 		goto err_mdp;
9772 	mdp_major = ret;
9773 
9774 	register_reboot_notifier(&md_notifier);
9775 	raid_table_header = register_sysctl("dev/raid", raid_table);
9776 
9777 	md_geninit();
9778 	return 0;
9779 
9780 err_mdp:
9781 	unregister_blkdev(MD_MAJOR, "md");
9782 err_md:
9783 	destroy_workqueue(md_bitmap_wq);
9784 err_bitmap_wq:
9785 	destroy_workqueue(md_misc_wq);
9786 err_misc_wq:
9787 	destroy_workqueue(md_wq);
9788 err_wq:
9789 	return ret;
9790 }
9791 
check_sb_changes(struct mddev * mddev,struct md_rdev * rdev)9792 static void check_sb_changes(struct mddev *mddev, struct md_rdev *rdev)
9793 {
9794 	struct mdp_superblock_1 *sb = page_address(rdev->sb_page);
9795 	struct md_rdev *rdev2, *tmp;
9796 	int role, ret;
9797 
9798 	/*
9799 	 * If size is changed in another node then we need to
9800 	 * do resize as well.
9801 	 */
9802 	if (mddev->dev_sectors != le64_to_cpu(sb->size)) {
9803 		ret = mddev->pers->resize(mddev, le64_to_cpu(sb->size));
9804 		if (ret)
9805 			pr_info("md-cluster: resize failed\n");
9806 		else
9807 			md_bitmap_update_sb(mddev->bitmap);
9808 	}
9809 
9810 	/* Check for change of roles in the active devices */
9811 	rdev_for_each_safe(rdev2, tmp, mddev) {
9812 		if (test_bit(Faulty, &rdev2->flags))
9813 			continue;
9814 
9815 		/* Check if the roles changed */
9816 		role = le16_to_cpu(sb->dev_roles[rdev2->desc_nr]);
9817 
9818 		if (test_bit(Candidate, &rdev2->flags)) {
9819 			if (role == MD_DISK_ROLE_FAULTY) {
9820 				pr_info("md: Removing Candidate device %pg because add failed\n",
9821 					rdev2->bdev);
9822 				md_kick_rdev_from_array(rdev2);
9823 				continue;
9824 			}
9825 			else
9826 				clear_bit(Candidate, &rdev2->flags);
9827 		}
9828 
9829 		if (role != rdev2->raid_disk) {
9830 			/*
9831 			 * got activated except reshape is happening.
9832 			 */
9833 			if (rdev2->raid_disk == -1 && role != MD_DISK_ROLE_SPARE &&
9834 			    !(le32_to_cpu(sb->feature_map) &
9835 			      MD_FEATURE_RESHAPE_ACTIVE)) {
9836 				rdev2->saved_raid_disk = role;
9837 				ret = remove_and_add_spares(mddev, rdev2);
9838 				pr_info("Activated spare: %pg\n",
9839 					rdev2->bdev);
9840 				/* wakeup mddev->thread here, so array could
9841 				 * perform resync with the new activated disk */
9842 				set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
9843 				md_wakeup_thread(mddev->thread);
9844 			}
9845 			/* device faulty
9846 			 * We just want to do the minimum to mark the disk
9847 			 * as faulty. The recovery is performed by the
9848 			 * one who initiated the error.
9849 			 */
9850 			if (role == MD_DISK_ROLE_FAULTY ||
9851 			    role == MD_DISK_ROLE_JOURNAL) {
9852 				md_error(mddev, rdev2);
9853 				clear_bit(Blocked, &rdev2->flags);
9854 			}
9855 		}
9856 	}
9857 
9858 	if (mddev->raid_disks != le32_to_cpu(sb->raid_disks)) {
9859 		ret = update_raid_disks(mddev, le32_to_cpu(sb->raid_disks));
9860 		if (ret)
9861 			pr_warn("md: updating array disks failed. %d\n", ret);
9862 	}
9863 
9864 	/*
9865 	 * Since mddev->delta_disks has already updated in update_raid_disks,
9866 	 * so it is time to check reshape.
9867 	 */
9868 	if (test_bit(MD_RESYNCING_REMOTE, &mddev->recovery) &&
9869 	    (le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) {
9870 		/*
9871 		 * reshape is happening in the remote node, we need to
9872 		 * update reshape_position and call start_reshape.
9873 		 */
9874 		mddev->reshape_position = le64_to_cpu(sb->reshape_position);
9875 		if (mddev->pers->update_reshape_pos)
9876 			mddev->pers->update_reshape_pos(mddev);
9877 		if (mddev->pers->start_reshape)
9878 			mddev->pers->start_reshape(mddev);
9879 	} else if (test_bit(MD_RESYNCING_REMOTE, &mddev->recovery) &&
9880 		   mddev->reshape_position != MaxSector &&
9881 		   !(le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) {
9882 		/* reshape is just done in another node. */
9883 		mddev->reshape_position = MaxSector;
9884 		if (mddev->pers->update_reshape_pos)
9885 			mddev->pers->update_reshape_pos(mddev);
9886 	}
9887 
9888 	/* Finally set the event to be up to date */
9889 	mddev->events = le64_to_cpu(sb->events);
9890 }
9891 
read_rdev(struct mddev * mddev,struct md_rdev * rdev)9892 static int read_rdev(struct mddev *mddev, struct md_rdev *rdev)
9893 {
9894 	int err;
9895 	struct page *swapout = rdev->sb_page;
9896 	struct mdp_superblock_1 *sb;
9897 
9898 	/* Store the sb page of the rdev in the swapout temporary
9899 	 * variable in case we err in the future
9900 	 */
9901 	rdev->sb_page = NULL;
9902 	err = alloc_disk_sb(rdev);
9903 	if (err == 0) {
9904 		ClearPageUptodate(rdev->sb_page);
9905 		rdev->sb_loaded = 0;
9906 		err = super_types[mddev->major_version].
9907 			load_super(rdev, NULL, mddev->minor_version);
9908 	}
9909 	if (err < 0) {
9910 		pr_warn("%s: %d Could not reload rdev(%d) err: %d. Restoring old values\n",
9911 				__func__, __LINE__, rdev->desc_nr, err);
9912 		if (rdev->sb_page)
9913 			put_page(rdev->sb_page);
9914 		rdev->sb_page = swapout;
9915 		rdev->sb_loaded = 1;
9916 		return err;
9917 	}
9918 
9919 	sb = page_address(rdev->sb_page);
9920 	/* Read the offset unconditionally, even if MD_FEATURE_RECOVERY_OFFSET
9921 	 * is not set
9922 	 */
9923 
9924 	if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RECOVERY_OFFSET))
9925 		rdev->recovery_offset = le64_to_cpu(sb->recovery_offset);
9926 
9927 	/* The other node finished recovery, call spare_active to set
9928 	 * device In_sync and mddev->degraded
9929 	 */
9930 	if (rdev->recovery_offset == MaxSector &&
9931 	    !test_bit(In_sync, &rdev->flags) &&
9932 	    mddev->pers->spare_active(mddev))
9933 		sysfs_notify_dirent_safe(mddev->sysfs_degraded);
9934 
9935 	put_page(swapout);
9936 	return 0;
9937 }
9938 
md_reload_sb(struct mddev * mddev,int nr)9939 void md_reload_sb(struct mddev *mddev, int nr)
9940 {
9941 	struct md_rdev *rdev = NULL, *iter;
9942 	int err;
9943 
9944 	/* Find the rdev */
9945 	rdev_for_each_rcu(iter, mddev) {
9946 		if (iter->desc_nr == nr) {
9947 			rdev = iter;
9948 			break;
9949 		}
9950 	}
9951 
9952 	if (!rdev) {
9953 		pr_warn("%s: %d Could not find rdev with nr %d\n", __func__, __LINE__, nr);
9954 		return;
9955 	}
9956 
9957 	err = read_rdev(mddev, rdev);
9958 	if (err < 0)
9959 		return;
9960 
9961 	check_sb_changes(mddev, rdev);
9962 
9963 	/* Read all rdev's to update recovery_offset */
9964 	rdev_for_each_rcu(rdev, mddev) {
9965 		if (!test_bit(Faulty, &rdev->flags))
9966 			read_rdev(mddev, rdev);
9967 	}
9968 }
9969 EXPORT_SYMBOL(md_reload_sb);
9970 
9971 #ifndef MODULE
9972 
9973 /*
9974  * Searches all registered partitions for autorun RAID arrays
9975  * at boot time.
9976  */
9977 
9978 static DEFINE_MUTEX(detected_devices_mutex);
9979 static LIST_HEAD(all_detected_devices);
9980 struct detected_devices_node {
9981 	struct list_head list;
9982 	dev_t dev;
9983 };
9984 
md_autodetect_dev(dev_t dev)9985 void md_autodetect_dev(dev_t dev)
9986 {
9987 	struct detected_devices_node *node_detected_dev;
9988 
9989 	node_detected_dev = kzalloc(sizeof(*node_detected_dev), GFP_KERNEL);
9990 	if (node_detected_dev) {
9991 		node_detected_dev->dev = dev;
9992 		mutex_lock(&detected_devices_mutex);
9993 		list_add_tail(&node_detected_dev->list, &all_detected_devices);
9994 		mutex_unlock(&detected_devices_mutex);
9995 	}
9996 }
9997 
md_autostart_arrays(int part)9998 void md_autostart_arrays(int part)
9999 {
10000 	struct md_rdev *rdev;
10001 	struct detected_devices_node *node_detected_dev;
10002 	dev_t dev;
10003 	int i_scanned, i_passed;
10004 
10005 	i_scanned = 0;
10006 	i_passed = 0;
10007 
10008 	pr_info("md: Autodetecting RAID arrays.\n");
10009 
10010 	mutex_lock(&detected_devices_mutex);
10011 	while (!list_empty(&all_detected_devices) && i_scanned < INT_MAX) {
10012 		i_scanned++;
10013 		node_detected_dev = list_entry(all_detected_devices.next,
10014 					struct detected_devices_node, list);
10015 		list_del(&node_detected_dev->list);
10016 		dev = node_detected_dev->dev;
10017 		kfree(node_detected_dev);
10018 		mutex_unlock(&detected_devices_mutex);
10019 		rdev = md_import_device(dev,0, 90);
10020 		mutex_lock(&detected_devices_mutex);
10021 		if (IS_ERR(rdev))
10022 			continue;
10023 
10024 		if (test_bit(Faulty, &rdev->flags))
10025 			continue;
10026 
10027 		set_bit(AutoDetected, &rdev->flags);
10028 		list_add(&rdev->same_set, &pending_raid_disks);
10029 		i_passed++;
10030 	}
10031 	mutex_unlock(&detected_devices_mutex);
10032 
10033 	pr_debug("md: Scanned %d and added %d devices.\n", i_scanned, i_passed);
10034 
10035 	autorun_devices(part);
10036 }
10037 
10038 #endif /* !MODULE */
10039 
md_exit(void)10040 static __exit void md_exit(void)
10041 {
10042 	struct mddev *mddev, *n;
10043 	int delay = 1;
10044 
10045 	unregister_blkdev(MD_MAJOR,"md");
10046 	unregister_blkdev(mdp_major, "mdp");
10047 	unregister_reboot_notifier(&md_notifier);
10048 	unregister_sysctl_table(raid_table_header);
10049 
10050 	/* We cannot unload the modules while some process is
10051 	 * waiting for us in select() or poll() - wake them up
10052 	 */
10053 	md_unloading = 1;
10054 	while (waitqueue_active(&md_event_waiters)) {
10055 		/* not safe to leave yet */
10056 		wake_up(&md_event_waiters);
10057 		msleep(delay);
10058 		delay += delay;
10059 	}
10060 	remove_proc_entry("mdstat", NULL);
10061 
10062 	spin_lock(&all_mddevs_lock);
10063 	list_for_each_entry_safe(mddev, n, &all_mddevs, all_mddevs) {
10064 		if (!mddev_get(mddev))
10065 			continue;
10066 		spin_unlock(&all_mddevs_lock);
10067 		export_array(mddev);
10068 		mddev->ctime = 0;
10069 		mddev->hold_active = 0;
10070 		/*
10071 		 * As the mddev is now fully clear, mddev_put will schedule
10072 		 * the mddev for destruction by a workqueue, and the
10073 		 * destroy_workqueue() below will wait for that to complete.
10074 		 */
10075 		mddev_put(mddev);
10076 		spin_lock(&all_mddevs_lock);
10077 	}
10078 	spin_unlock(&all_mddevs_lock);
10079 
10080 	destroy_workqueue(md_misc_wq);
10081 	destroy_workqueue(md_bitmap_wq);
10082 	destroy_workqueue(md_wq);
10083 }
10084 
10085 subsys_initcall(md_init);
module_exit(md_exit)10086 module_exit(md_exit)
10087 
10088 static int get_ro(char *buffer, const struct kernel_param *kp)
10089 {
10090 	return sprintf(buffer, "%d\n", start_readonly);
10091 }
set_ro(const char * val,const struct kernel_param * kp)10092 static int set_ro(const char *val, const struct kernel_param *kp)
10093 {
10094 	return kstrtouint(val, 10, (unsigned int *)&start_readonly);
10095 }
10096 
10097 module_param_call(start_ro, set_ro, get_ro, NULL, S_IRUSR|S_IWUSR);
10098 module_param(start_dirty_degraded, int, S_IRUGO|S_IWUSR);
10099 module_param_call(new_array, add_named_array, NULL, NULL, S_IWUSR);
10100 module_param(create_on_open, bool, S_IRUSR|S_IWUSR);
10101 
10102 MODULE_LICENSE("GPL");
10103 MODULE_DESCRIPTION("MD RAID framework");
10104 MODULE_ALIAS("md");
10105 MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR);
10106