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
74  * by pers_lock.
75  * pers_lock does extra service to protect accesses to
76  * mddev->thread when the mutex cannot be held.
77  */
78 static LIST_HEAD(pers_list);
79 static DEFINE_SPINLOCK(pers_lock);
80 
81 static struct kobj_type md_ktype;
82 
83 struct md_cluster_operations *md_cluster_ops;
84 EXPORT_SYMBOL(md_cluster_ops);
85 static struct module *md_cluster_mod;
86 
87 static DECLARE_WAIT_QUEUE_HEAD(resync_wait);
88 static struct workqueue_struct *md_wq;
89 static struct workqueue_struct *md_misc_wq;
90 static struct workqueue_struct *md_rdev_misc_wq;
91 
92 static int remove_and_add_spares(struct mddev *mddev,
93 				 struct md_rdev *this);
94 static void mddev_detach(struct mddev *mddev);
95 
96 /*
97  * Default number of read corrections we'll attempt on an rdev
98  * before ejecting it from the array. We divide the read error
99  * count by 2 for every hour elapsed between read errors.
100  */
101 #define MD_DEFAULT_MAX_CORRECTED_READ_ERRORS 20
102 /* Default safemode delay: 200 msec */
103 #define DEFAULT_SAFEMODE_DELAY ((200 * HZ)/1000 +1)
104 /*
105  * Current RAID-1,4,5 parallel reconstruction 'guaranteed speed limit'
106  * is 1000 KB/sec, so the extra system load does not show up that much.
107  * Increase it if you want to have more _guaranteed_ speed. Note that
108  * the RAID driver will use the maximum available bandwidth if the IO
109  * subsystem is idle. There is also an 'absolute maximum' reconstruction
110  * speed limit - in case reconstruction slows down your system despite
111  * idle IO detection.
112  *
113  * you can change it via /proc/sys/dev/raid/speed_limit_min and _max.
114  * or /sys/block/mdX/md/sync_speed_{min,max}
115  */
116 
117 static int sysctl_speed_limit_min = 1000;
118 static int sysctl_speed_limit_max = 200000;
speed_min(struct mddev * mddev)119 static inline int speed_min(struct mddev *mddev)
120 {
121 	return mddev->sync_speed_min ?
122 		mddev->sync_speed_min : sysctl_speed_limit_min;
123 }
124 
speed_max(struct mddev * mddev)125 static inline int speed_max(struct mddev *mddev)
126 {
127 	return mddev->sync_speed_max ?
128 		mddev->sync_speed_max : sysctl_speed_limit_max;
129 }
130 
rdev_uninit_serial(struct md_rdev * rdev)131 static void rdev_uninit_serial(struct md_rdev *rdev)
132 {
133 	if (!test_and_clear_bit(CollisionCheck, &rdev->flags))
134 		return;
135 
136 	kvfree(rdev->serial);
137 	rdev->serial = NULL;
138 }
139 
rdevs_uninit_serial(struct mddev * mddev)140 static void rdevs_uninit_serial(struct mddev *mddev)
141 {
142 	struct md_rdev *rdev;
143 
144 	rdev_for_each(rdev, mddev)
145 		rdev_uninit_serial(rdev);
146 }
147 
rdev_init_serial(struct md_rdev * rdev)148 static int rdev_init_serial(struct md_rdev *rdev)
149 {
150 	/* serial_nums equals with BARRIER_BUCKETS_NR */
151 	int i, serial_nums = 1 << ((PAGE_SHIFT - ilog2(sizeof(atomic_t))));
152 	struct serial_in_rdev *serial = NULL;
153 
154 	if (test_bit(CollisionCheck, &rdev->flags))
155 		return 0;
156 
157 	serial = kvmalloc(sizeof(struct serial_in_rdev) * serial_nums,
158 			  GFP_KERNEL);
159 	if (!serial)
160 		return -ENOMEM;
161 
162 	for (i = 0; i < serial_nums; i++) {
163 		struct serial_in_rdev *serial_tmp = &serial[i];
164 
165 		spin_lock_init(&serial_tmp->serial_lock);
166 		serial_tmp->serial_rb = RB_ROOT_CACHED;
167 		init_waitqueue_head(&serial_tmp->serial_io_wait);
168 	}
169 
170 	rdev->serial = serial;
171 	set_bit(CollisionCheck, &rdev->flags);
172 
173 	return 0;
174 }
175 
rdevs_init_serial(struct mddev * mddev)176 static int rdevs_init_serial(struct mddev *mddev)
177 {
178 	struct md_rdev *rdev;
179 	int ret = 0;
180 
181 	rdev_for_each(rdev, mddev) {
182 		ret = rdev_init_serial(rdev);
183 		if (ret)
184 			break;
185 	}
186 
187 	/* Free all resources if pool is not existed */
188 	if (ret && !mddev->serial_info_pool)
189 		rdevs_uninit_serial(mddev);
190 
191 	return ret;
192 }
193 
194 /*
195  * rdev needs to enable serial stuffs if it meets the conditions:
196  * 1. it is multi-queue device flaged with writemostly.
197  * 2. the write-behind mode is enabled.
198  */
rdev_need_serial(struct md_rdev * rdev)199 static int rdev_need_serial(struct md_rdev *rdev)
200 {
201 	return (rdev && rdev->mddev->bitmap_info.max_write_behind > 0 &&
202 		rdev->bdev->bd_disk->queue->nr_hw_queues != 1 &&
203 		test_bit(WriteMostly, &rdev->flags));
204 }
205 
206 /*
207  * Init resource for rdev(s), then create serial_info_pool if:
208  * 1. rdev is the first device which return true from rdev_enable_serial.
209  * 2. rdev is NULL, means we want to enable serialization for all rdevs.
210  */
mddev_create_serial_pool(struct mddev * mddev,struct md_rdev * rdev,bool is_suspend)211 void mddev_create_serial_pool(struct mddev *mddev, struct md_rdev *rdev,
212 			      bool is_suspend)
213 {
214 	int ret = 0;
215 
216 	if (rdev && !rdev_need_serial(rdev) &&
217 	    !test_bit(CollisionCheck, &rdev->flags))
218 		return;
219 
220 	if (!is_suspend)
221 		mddev_suspend(mddev);
222 
223 	if (!rdev)
224 		ret = rdevs_init_serial(mddev);
225 	else
226 		ret = rdev_init_serial(rdev);
227 	if (ret)
228 		goto abort;
229 
230 	if (mddev->serial_info_pool == NULL) {
231 		/*
232 		 * already in memalloc noio context by
233 		 * mddev_suspend()
234 		 */
235 		mddev->serial_info_pool =
236 			mempool_create_kmalloc_pool(NR_SERIAL_INFOS,
237 						sizeof(struct serial_info));
238 		if (!mddev->serial_info_pool) {
239 			rdevs_uninit_serial(mddev);
240 			pr_err("can't alloc memory pool for serialization\n");
241 		}
242 	}
243 
244 abort:
245 	if (!is_suspend)
246 		mddev_resume(mddev);
247 }
248 
249 /*
250  * Free resource from rdev(s), and destroy serial_info_pool under conditions:
251  * 1. rdev is the last device flaged with CollisionCheck.
252  * 2. when bitmap is destroyed while policy is not enabled.
253  * 3. for disable policy, the pool is destroyed only when no rdev needs it.
254  */
mddev_destroy_serial_pool(struct mddev * mddev,struct md_rdev * rdev,bool is_suspend)255 void mddev_destroy_serial_pool(struct mddev *mddev, struct md_rdev *rdev,
256 			       bool is_suspend)
257 {
258 	if (rdev && !test_bit(CollisionCheck, &rdev->flags))
259 		return;
260 
261 	if (mddev->serial_info_pool) {
262 		struct md_rdev *temp;
263 		int num = 0; /* used to track if other rdevs need the pool */
264 
265 		if (!is_suspend)
266 			mddev_suspend(mddev);
267 		rdev_for_each(temp, mddev) {
268 			if (!rdev) {
269 				if (!mddev->serialize_policy ||
270 				    !rdev_need_serial(temp))
271 					rdev_uninit_serial(temp);
272 				else
273 					num++;
274 			} else if (temp != rdev &&
275 				   test_bit(CollisionCheck, &temp->flags))
276 				num++;
277 		}
278 
279 		if (rdev)
280 			rdev_uninit_serial(rdev);
281 
282 		if (num)
283 			pr_info("The mempool could be used by other devices\n");
284 		else {
285 			mempool_destroy(mddev->serial_info_pool);
286 			mddev->serial_info_pool = NULL;
287 		}
288 		if (!is_suspend)
289 			mddev_resume(mddev);
290 	}
291 }
292 
293 static struct ctl_table_header *raid_table_header;
294 
295 static struct ctl_table raid_table[] = {
296 	{
297 		.procname	= "speed_limit_min",
298 		.data		= &sysctl_speed_limit_min,
299 		.maxlen		= sizeof(int),
300 		.mode		= S_IRUGO|S_IWUSR,
301 		.proc_handler	= proc_dointvec,
302 	},
303 	{
304 		.procname	= "speed_limit_max",
305 		.data		= &sysctl_speed_limit_max,
306 		.maxlen		= sizeof(int),
307 		.mode		= S_IRUGO|S_IWUSR,
308 		.proc_handler	= proc_dointvec,
309 	},
310 	{ }
311 };
312 
313 static struct ctl_table raid_dir_table[] = {
314 	{
315 		.procname	= "raid",
316 		.maxlen		= 0,
317 		.mode		= S_IRUGO|S_IXUGO,
318 		.child		= raid_table,
319 	},
320 	{ }
321 };
322 
323 static struct ctl_table raid_root_table[] = {
324 	{
325 		.procname	= "dev",
326 		.maxlen		= 0,
327 		.mode		= 0555,
328 		.child		= raid_dir_table,
329 	},
330 	{  }
331 };
332 
333 static int start_readonly;
334 
335 /*
336  * The original mechanism for creating an md device is to create
337  * a device node in /dev and to open it.  This causes races with device-close.
338  * The preferred method is to write to the "new_array" module parameter.
339  * This can avoid races.
340  * Setting create_on_open to false disables the original mechanism
341  * so all the races disappear.
342  */
343 static bool create_on_open = true;
344 
345 /*
346  * We have a system wide 'event count' that is incremented
347  * on any 'interesting' event, and readers of /proc/mdstat
348  * can use 'poll' or 'select' to find out when the event
349  * count increases.
350  *
351  * Events are:
352  *  start array, stop array, error, add device, remove device,
353  *  start build, activate spare
354  */
355 static DECLARE_WAIT_QUEUE_HEAD(md_event_waiters);
356 static atomic_t md_event_count;
md_new_event(void)357 void md_new_event(void)
358 {
359 	atomic_inc(&md_event_count);
360 	wake_up(&md_event_waiters);
361 }
362 EXPORT_SYMBOL_GPL(md_new_event);
363 
364 /*
365  * Enables to iterate over all existing md arrays
366  * all_mddevs_lock protects this list.
367  */
368 static LIST_HEAD(all_mddevs);
369 static DEFINE_SPINLOCK(all_mddevs_lock);
370 
371 /*
372  * iterates through all used mddevs in the system.
373  * We take care to grab the all_mddevs_lock whenever navigating
374  * the list, and to always hold a refcount when unlocked.
375  * Any code which breaks out of this loop while own
376  * a reference to the current mddev and must mddev_put it.
377  */
378 #define for_each_mddev(_mddev,_tmp)					\
379 									\
380 	for (({ spin_lock(&all_mddevs_lock);				\
381 		_tmp = all_mddevs.next;					\
382 		_mddev = NULL;});					\
383 	     ({ if (_tmp != &all_mddevs)				\
384 			mddev_get(list_entry(_tmp, struct mddev, all_mddevs));\
385 		spin_unlock(&all_mddevs_lock);				\
386 		if (_mddev) mddev_put(_mddev);				\
387 		_mddev = list_entry(_tmp, struct mddev, all_mddevs);	\
388 		_tmp != &all_mddevs;});					\
389 	     ({ spin_lock(&all_mddevs_lock);				\
390 		_tmp = _tmp->next;})					\
391 		)
392 
393 /* Rather than calling directly into the personality make_request function,
394  * IO requests come here first so that we can check if the device is
395  * being suspended pending a reconfiguration.
396  * We hold a refcount over the call to ->make_request.  By the time that
397  * call has finished, the bio has been linked into some internal structure
398  * and so is visible to ->quiesce(), so we don't need the refcount any more.
399  */
is_suspended(struct mddev * mddev,struct bio * bio)400 static bool is_suspended(struct mddev *mddev, struct bio *bio)
401 {
402 	if (mddev->suspended)
403 		return true;
404 	if (bio_data_dir(bio) != WRITE)
405 		return false;
406 	if (mddev->suspend_lo >= mddev->suspend_hi)
407 		return false;
408 	if (bio->bi_iter.bi_sector >= mddev->suspend_hi)
409 		return false;
410 	if (bio_end_sector(bio) < mddev->suspend_lo)
411 		return false;
412 	return true;
413 }
414 
md_handle_request(struct mddev * mddev,struct bio * bio)415 void md_handle_request(struct mddev *mddev, struct bio *bio)
416 {
417 check_suspended:
418 	rcu_read_lock();
419 	if (is_suspended(mddev, bio)) {
420 		DEFINE_WAIT(__wait);
421 		/* Bail out if REQ_NOWAIT is set for the bio */
422 		if (bio->bi_opf & REQ_NOWAIT) {
423 			rcu_read_unlock();
424 			bio_wouldblock_error(bio);
425 			return;
426 		}
427 		for (;;) {
428 			prepare_to_wait(&mddev->sb_wait, &__wait,
429 					TASK_UNINTERRUPTIBLE);
430 			if (!is_suspended(mddev, bio))
431 				break;
432 			rcu_read_unlock();
433 			schedule();
434 			rcu_read_lock();
435 		}
436 		finish_wait(&mddev->sb_wait, &__wait);
437 	}
438 	atomic_inc(&mddev->active_io);
439 	rcu_read_unlock();
440 
441 	if (!mddev->pers->make_request(mddev, bio)) {
442 		atomic_dec(&mddev->active_io);
443 		wake_up(&mddev->sb_wait);
444 		goto check_suspended;
445 	}
446 
447 	if (atomic_dec_and_test(&mddev->active_io) && mddev->suspended)
448 		wake_up(&mddev->sb_wait);
449 }
450 EXPORT_SYMBOL(md_handle_request);
451 
md_submit_bio(struct bio * bio)452 static void md_submit_bio(struct bio *bio)
453 {
454 	const int rw = bio_data_dir(bio);
455 	struct mddev *mddev = bio->bi_bdev->bd_disk->private_data;
456 
457 	if (mddev == NULL || mddev->pers == NULL) {
458 		bio_io_error(bio);
459 		return;
460 	}
461 
462 	if (unlikely(test_bit(MD_BROKEN, &mddev->flags)) && (rw == WRITE)) {
463 		bio_io_error(bio);
464 		return;
465 	}
466 
467 	blk_queue_split(&bio);
468 
469 	if (mddev->ro == 1 && unlikely(rw == WRITE)) {
470 		if (bio_sectors(bio) != 0)
471 			bio->bi_status = BLK_STS_IOERR;
472 		bio_endio(bio);
473 		return;
474 	}
475 
476 	/* bio could be mergeable after passing to underlayer */
477 	bio->bi_opf &= ~REQ_NOMERGE;
478 
479 	md_handle_request(mddev, bio);
480 }
481 
482 /* mddev_suspend makes sure no new requests are submitted
483  * to the device, and that any requests that have been submitted
484  * are completely handled.
485  * Once mddev_detach() is called and completes, the module will be
486  * completely unused.
487  */
mddev_suspend(struct mddev * mddev)488 void mddev_suspend(struct mddev *mddev)
489 {
490 	WARN_ON_ONCE(mddev->thread && current == mddev->thread->tsk);
491 	lockdep_assert_held(&mddev->reconfig_mutex);
492 	if (mddev->suspended++)
493 		return;
494 	synchronize_rcu();
495 	wake_up(&mddev->sb_wait);
496 	set_bit(MD_ALLOW_SB_UPDATE, &mddev->flags);
497 	smp_mb__after_atomic();
498 	wait_event(mddev->sb_wait, atomic_read(&mddev->active_io) == 0);
499 	mddev->pers->quiesce(mddev, 1);
500 	clear_bit_unlock(MD_ALLOW_SB_UPDATE, &mddev->flags);
501 	wait_event(mddev->sb_wait, !test_bit(MD_UPDATING_SB, &mddev->flags));
502 
503 	del_timer_sync(&mddev->safemode_timer);
504 	/* restrict memory reclaim I/O during raid array is suspend */
505 	mddev->noio_flag = memalloc_noio_save();
506 }
507 EXPORT_SYMBOL_GPL(mddev_suspend);
508 
mddev_resume(struct mddev * mddev)509 void mddev_resume(struct mddev *mddev)
510 {
511 	/* entred the memalloc scope from mddev_suspend() */
512 	memalloc_noio_restore(mddev->noio_flag);
513 	lockdep_assert_held(&mddev->reconfig_mutex);
514 	if (--mddev->suspended)
515 		return;
516 	wake_up(&mddev->sb_wait);
517 	mddev->pers->quiesce(mddev, 0);
518 
519 	set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
520 	md_wakeup_thread(mddev->thread);
521 	md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
522 }
523 EXPORT_SYMBOL_GPL(mddev_resume);
524 
525 /*
526  * Generic flush handling for md
527  */
528 
md_end_flush(struct bio * bio)529 static void md_end_flush(struct bio *bio)
530 {
531 	struct md_rdev *rdev = bio->bi_private;
532 	struct mddev *mddev = rdev->mddev;
533 
534 	rdev_dec_pending(rdev, mddev);
535 
536 	if (atomic_dec_and_test(&mddev->flush_pending)) {
537 		/* The pre-request flush has finished */
538 		queue_work(md_wq, &mddev->flush_work);
539 	}
540 	bio_put(bio);
541 }
542 
543 static void md_submit_flush_data(struct work_struct *ws);
544 
submit_flushes(struct work_struct * ws)545 static void submit_flushes(struct work_struct *ws)
546 {
547 	struct mddev *mddev = container_of(ws, struct mddev, flush_work);
548 	struct md_rdev *rdev;
549 
550 	mddev->start_flush = ktime_get_boottime();
551 	INIT_WORK(&mddev->flush_work, md_submit_flush_data);
552 	atomic_set(&mddev->flush_pending, 1);
553 	rcu_read_lock();
554 	rdev_for_each_rcu(rdev, mddev)
555 		if (rdev->raid_disk >= 0 &&
556 		    !test_bit(Faulty, &rdev->flags)) {
557 			/* Take two references, one is dropped
558 			 * when request finishes, one after
559 			 * we reclaim rcu_read_lock
560 			 */
561 			struct bio *bi;
562 			atomic_inc(&rdev->nr_pending);
563 			atomic_inc(&rdev->nr_pending);
564 			rcu_read_unlock();
565 			bi = bio_alloc_bioset(rdev->bdev, 0,
566 					      REQ_OP_WRITE | REQ_PREFLUSH,
567 					      GFP_NOIO, &mddev->bio_set);
568 			bi->bi_end_io = md_end_flush;
569 			bi->bi_private = rdev;
570 			atomic_inc(&mddev->flush_pending);
571 			submit_bio(bi);
572 			rcu_read_lock();
573 			rdev_dec_pending(rdev, mddev);
574 		}
575 	rcu_read_unlock();
576 	if (atomic_dec_and_test(&mddev->flush_pending))
577 		queue_work(md_wq, &mddev->flush_work);
578 }
579 
md_submit_flush_data(struct work_struct * ws)580 static void md_submit_flush_data(struct work_struct *ws)
581 {
582 	struct mddev *mddev = container_of(ws, struct mddev, flush_work);
583 	struct bio *bio = mddev->flush_bio;
584 
585 	/*
586 	 * must reset flush_bio before calling into md_handle_request to avoid a
587 	 * deadlock, because other bios passed md_handle_request suspend check
588 	 * could wait for this and below md_handle_request could wait for those
589 	 * bios because of suspend check
590 	 */
591 	spin_lock_irq(&mddev->lock);
592 	mddev->prev_flush_start = mddev->start_flush;
593 	mddev->flush_bio = NULL;
594 	spin_unlock_irq(&mddev->lock);
595 	wake_up(&mddev->sb_wait);
596 
597 	if (bio->bi_iter.bi_size == 0) {
598 		/* an empty barrier - all done */
599 		bio_endio(bio);
600 	} else {
601 		bio->bi_opf &= ~REQ_PREFLUSH;
602 		md_handle_request(mddev, bio);
603 	}
604 }
605 
606 /*
607  * Manages consolidation of flushes and submitting any flushes needed for
608  * a bio with REQ_PREFLUSH.  Returns true if the bio is finished or is
609  * being finished in another context.  Returns false if the flushing is
610  * complete but still needs the I/O portion of the bio to be processed.
611  */
md_flush_request(struct mddev * mddev,struct bio * bio)612 bool md_flush_request(struct mddev *mddev, struct bio *bio)
613 {
614 	ktime_t req_start = ktime_get_boottime();
615 	spin_lock_irq(&mddev->lock);
616 	/* flush requests wait until ongoing flush completes,
617 	 * hence coalescing all the pending requests.
618 	 */
619 	wait_event_lock_irq(mddev->sb_wait,
620 			    !mddev->flush_bio ||
621 			    ktime_before(req_start, mddev->prev_flush_start),
622 			    mddev->lock);
623 	/* new request after previous flush is completed */
624 	if (ktime_after(req_start, mddev->prev_flush_start)) {
625 		WARN_ON(mddev->flush_bio);
626 		mddev->flush_bio = bio;
627 		bio = NULL;
628 	}
629 	spin_unlock_irq(&mddev->lock);
630 
631 	if (!bio) {
632 		INIT_WORK(&mddev->flush_work, submit_flushes);
633 		queue_work(md_wq, &mddev->flush_work);
634 	} else {
635 		/* flush was performed for some other bio while we waited. */
636 		if (bio->bi_iter.bi_size == 0)
637 			/* an empty barrier - all done */
638 			bio_endio(bio);
639 		else {
640 			bio->bi_opf &= ~REQ_PREFLUSH;
641 			return false;
642 		}
643 	}
644 	return true;
645 }
646 EXPORT_SYMBOL(md_flush_request);
647 
mddev_get(struct mddev * mddev)648 static inline struct mddev *mddev_get(struct mddev *mddev)
649 {
650 	atomic_inc(&mddev->active);
651 	return mddev;
652 }
653 
654 static void mddev_delayed_delete(struct work_struct *ws);
655 
mddev_put(struct mddev * mddev)656 static void mddev_put(struct mddev *mddev)
657 {
658 	if (!atomic_dec_and_lock(&mddev->active, &all_mddevs_lock))
659 		return;
660 	if (!mddev->raid_disks && list_empty(&mddev->disks) &&
661 	    mddev->ctime == 0 && !mddev->hold_active) {
662 		/* Array is not configured at all, and not held active,
663 		 * so destroy it */
664 		list_del_init(&mddev->all_mddevs);
665 
666 		/*
667 		 * Call queue_work inside the spinlock so that
668 		 * flush_workqueue() after mddev_find will succeed in waiting
669 		 * for the work to be done.
670 		 */
671 		INIT_WORK(&mddev->del_work, mddev_delayed_delete);
672 		queue_work(md_misc_wq, &mddev->del_work);
673 	}
674 	spin_unlock(&all_mddevs_lock);
675 }
676 
677 static void md_safemode_timeout(struct timer_list *t);
678 
mddev_init(struct mddev * mddev)679 void mddev_init(struct mddev *mddev)
680 {
681 	kobject_init(&mddev->kobj, &md_ktype);
682 	mutex_init(&mddev->open_mutex);
683 	mutex_init(&mddev->reconfig_mutex);
684 	mutex_init(&mddev->bitmap_info.mutex);
685 	INIT_LIST_HEAD(&mddev->disks);
686 	INIT_LIST_HEAD(&mddev->all_mddevs);
687 	timer_setup(&mddev->safemode_timer, md_safemode_timeout, 0);
688 	atomic_set(&mddev->active, 1);
689 	atomic_set(&mddev->openers, 0);
690 	atomic_set(&mddev->active_io, 0);
691 	spin_lock_init(&mddev->lock);
692 	atomic_set(&mddev->flush_pending, 0);
693 	init_waitqueue_head(&mddev->sb_wait);
694 	init_waitqueue_head(&mddev->recovery_wait);
695 	mddev->reshape_position = MaxSector;
696 	mddev->reshape_backwards = 0;
697 	mddev->last_sync_action = "none";
698 	mddev->resync_min = 0;
699 	mddev->resync_max = MaxSector;
700 	mddev->level = LEVEL_NONE;
701 }
702 EXPORT_SYMBOL_GPL(mddev_init);
703 
mddev_find_locked(dev_t unit)704 static struct mddev *mddev_find_locked(dev_t unit)
705 {
706 	struct mddev *mddev;
707 
708 	list_for_each_entry(mddev, &all_mddevs, all_mddevs)
709 		if (mddev->unit == unit)
710 			return mddev;
711 
712 	return NULL;
713 }
714 
715 /* find an unused unit number */
mddev_alloc_unit(void)716 static dev_t mddev_alloc_unit(void)
717 {
718 	static int next_minor = 512;
719 	int start = next_minor;
720 	bool is_free = 0;
721 	dev_t dev = 0;
722 
723 	while (!is_free) {
724 		dev = MKDEV(MD_MAJOR, next_minor);
725 		next_minor++;
726 		if (next_minor > MINORMASK)
727 			next_minor = 0;
728 		if (next_minor == start)
729 			return 0;		/* Oh dear, all in use. */
730 		is_free = !mddev_find_locked(dev);
731 	}
732 
733 	return dev;
734 }
735 
mddev_find(dev_t unit)736 static struct mddev *mddev_find(dev_t unit)
737 {
738 	struct mddev *mddev;
739 
740 	if (MAJOR(unit) != MD_MAJOR)
741 		unit &= ~((1 << MdpMinorShift) - 1);
742 
743 	spin_lock(&all_mddevs_lock);
744 	mddev = mddev_find_locked(unit);
745 	if (mddev)
746 		mddev_get(mddev);
747 	spin_unlock(&all_mddevs_lock);
748 
749 	return mddev;
750 }
751 
mddev_alloc(dev_t unit)752 static struct mddev *mddev_alloc(dev_t unit)
753 {
754 	struct mddev *new;
755 	int error;
756 
757 	if (unit && MAJOR(unit) != MD_MAJOR)
758 		unit &= ~((1 << MdpMinorShift) - 1);
759 
760 	new = kzalloc(sizeof(*new), GFP_KERNEL);
761 	if (!new)
762 		return ERR_PTR(-ENOMEM);
763 	mddev_init(new);
764 
765 	spin_lock(&all_mddevs_lock);
766 	if (unit) {
767 		error = -EEXIST;
768 		if (mddev_find_locked(unit))
769 			goto out_free_new;
770 		new->unit = unit;
771 		if (MAJOR(unit) == MD_MAJOR)
772 			new->md_minor = MINOR(unit);
773 		else
774 			new->md_minor = MINOR(unit) >> MdpMinorShift;
775 		new->hold_active = UNTIL_IOCTL;
776 	} else {
777 		error = -ENODEV;
778 		new->unit = mddev_alloc_unit();
779 		if (!new->unit)
780 			goto out_free_new;
781 		new->md_minor = MINOR(new->unit);
782 		new->hold_active = UNTIL_STOP;
783 	}
784 
785 	list_add(&new->all_mddevs, &all_mddevs);
786 	spin_unlock(&all_mddevs_lock);
787 	return new;
788 out_free_new:
789 	spin_unlock(&all_mddevs_lock);
790 	kfree(new);
791 	return ERR_PTR(error);
792 }
793 
794 static const struct attribute_group md_redundancy_group;
795 
mddev_unlock(struct mddev * mddev)796 void mddev_unlock(struct mddev *mddev)
797 {
798 	if (mddev->to_remove) {
799 		/* These cannot be removed under reconfig_mutex as
800 		 * an access to the files will try to take reconfig_mutex
801 		 * while holding the file unremovable, which leads to
802 		 * a deadlock.
803 		 * So hold set sysfs_active while the remove in happeing,
804 		 * and anything else which might set ->to_remove or my
805 		 * otherwise change the sysfs namespace will fail with
806 		 * -EBUSY if sysfs_active is still set.
807 		 * We set sysfs_active under reconfig_mutex and elsewhere
808 		 * test it under the same mutex to ensure its correct value
809 		 * is seen.
810 		 */
811 		const struct attribute_group *to_remove = mddev->to_remove;
812 		mddev->to_remove = NULL;
813 		mddev->sysfs_active = 1;
814 		mutex_unlock(&mddev->reconfig_mutex);
815 
816 		if (mddev->kobj.sd) {
817 			if (to_remove != &md_redundancy_group)
818 				sysfs_remove_group(&mddev->kobj, to_remove);
819 			if (mddev->pers == NULL ||
820 			    mddev->pers->sync_request == NULL) {
821 				sysfs_remove_group(&mddev->kobj, &md_redundancy_group);
822 				if (mddev->sysfs_action)
823 					sysfs_put(mddev->sysfs_action);
824 				if (mddev->sysfs_completed)
825 					sysfs_put(mddev->sysfs_completed);
826 				if (mddev->sysfs_degraded)
827 					sysfs_put(mddev->sysfs_degraded);
828 				mddev->sysfs_action = NULL;
829 				mddev->sysfs_completed = NULL;
830 				mddev->sysfs_degraded = NULL;
831 			}
832 		}
833 		mddev->sysfs_active = 0;
834 	} else
835 		mutex_unlock(&mddev->reconfig_mutex);
836 
837 	/* As we've dropped the mutex we need a spinlock to
838 	 * make sure the thread doesn't disappear
839 	 */
840 	spin_lock(&pers_lock);
841 	md_wakeup_thread(mddev->thread);
842 	wake_up(&mddev->sb_wait);
843 	spin_unlock(&pers_lock);
844 }
845 EXPORT_SYMBOL_GPL(mddev_unlock);
846 
md_find_rdev_nr_rcu(struct mddev * mddev,int nr)847 struct md_rdev *md_find_rdev_nr_rcu(struct mddev *mddev, int nr)
848 {
849 	struct md_rdev *rdev;
850 
851 	rdev_for_each_rcu(rdev, mddev)
852 		if (rdev->desc_nr == nr)
853 			return rdev;
854 
855 	return NULL;
856 }
857 EXPORT_SYMBOL_GPL(md_find_rdev_nr_rcu);
858 
find_rdev(struct mddev * mddev,dev_t dev)859 static struct md_rdev *find_rdev(struct mddev *mddev, dev_t dev)
860 {
861 	struct md_rdev *rdev;
862 
863 	rdev_for_each(rdev, mddev)
864 		if (rdev->bdev->bd_dev == dev)
865 			return rdev;
866 
867 	return NULL;
868 }
869 
md_find_rdev_rcu(struct mddev * mddev,dev_t dev)870 struct md_rdev *md_find_rdev_rcu(struct mddev *mddev, dev_t dev)
871 {
872 	struct md_rdev *rdev;
873 
874 	rdev_for_each_rcu(rdev, mddev)
875 		if (rdev->bdev->bd_dev == dev)
876 			return rdev;
877 
878 	return NULL;
879 }
880 EXPORT_SYMBOL_GPL(md_find_rdev_rcu);
881 
find_pers(int level,char * clevel)882 static struct md_personality *find_pers(int level, char *clevel)
883 {
884 	struct md_personality *pers;
885 	list_for_each_entry(pers, &pers_list, list) {
886 		if (level != LEVEL_NONE && pers->level == level)
887 			return pers;
888 		if (strcmp(pers->name, clevel)==0)
889 			return pers;
890 	}
891 	return NULL;
892 }
893 
894 /* return the offset of the super block in 512byte sectors */
calc_dev_sboffset(struct md_rdev * rdev)895 static inline sector_t calc_dev_sboffset(struct md_rdev *rdev)
896 {
897 	return MD_NEW_SIZE_SECTORS(bdev_nr_sectors(rdev->bdev));
898 }
899 
alloc_disk_sb(struct md_rdev * rdev)900 static int alloc_disk_sb(struct md_rdev *rdev)
901 {
902 	rdev->sb_page = alloc_page(GFP_KERNEL);
903 	if (!rdev->sb_page)
904 		return -ENOMEM;
905 	return 0;
906 }
907 
md_rdev_clear(struct md_rdev * rdev)908 void md_rdev_clear(struct md_rdev *rdev)
909 {
910 	if (rdev->sb_page) {
911 		put_page(rdev->sb_page);
912 		rdev->sb_loaded = 0;
913 		rdev->sb_page = NULL;
914 		rdev->sb_start = 0;
915 		rdev->sectors = 0;
916 	}
917 	if (rdev->bb_page) {
918 		put_page(rdev->bb_page);
919 		rdev->bb_page = NULL;
920 	}
921 	badblocks_exit(&rdev->badblocks);
922 }
923 EXPORT_SYMBOL_GPL(md_rdev_clear);
924 
super_written(struct bio * bio)925 static void super_written(struct bio *bio)
926 {
927 	struct md_rdev *rdev = bio->bi_private;
928 	struct mddev *mddev = rdev->mddev;
929 
930 	if (bio->bi_status) {
931 		pr_err("md: %s gets error=%d\n", __func__,
932 		       blk_status_to_errno(bio->bi_status));
933 		md_error(mddev, rdev);
934 		if (!test_bit(Faulty, &rdev->flags)
935 		    && (bio->bi_opf & MD_FAILFAST)) {
936 			set_bit(MD_SB_NEED_REWRITE, &mddev->sb_flags);
937 			set_bit(LastDev, &rdev->flags);
938 		}
939 	} else
940 		clear_bit(LastDev, &rdev->flags);
941 
942 	if (atomic_dec_and_test(&mddev->pending_writes))
943 		wake_up(&mddev->sb_wait);
944 	rdev_dec_pending(rdev, mddev);
945 	bio_put(bio);
946 }
947 
md_super_write(struct mddev * mddev,struct md_rdev * rdev,sector_t sector,int size,struct page * page)948 void md_super_write(struct mddev *mddev, struct md_rdev *rdev,
949 		   sector_t sector, int size, struct page *page)
950 {
951 	/* write first size bytes of page to sector of rdev
952 	 * Increment mddev->pending_writes before returning
953 	 * and decrement it on completion, waking up sb_wait
954 	 * if zero is reached.
955 	 * If an error occurred, call md_error
956 	 */
957 	struct bio *bio;
958 
959 	if (!page)
960 		return;
961 
962 	if (test_bit(Faulty, &rdev->flags))
963 		return;
964 
965 	bio = bio_alloc_bioset(rdev->meta_bdev ? rdev->meta_bdev : rdev->bdev,
966 			       1,
967 			       REQ_OP_WRITE | REQ_SYNC | REQ_PREFLUSH | REQ_FUA,
968 			       GFP_NOIO, &mddev->sync_set);
969 
970 	atomic_inc(&rdev->nr_pending);
971 
972 	bio->bi_iter.bi_sector = sector;
973 	bio_add_page(bio, page, size, 0);
974 	bio->bi_private = rdev;
975 	bio->bi_end_io = super_written;
976 
977 	if (test_bit(MD_FAILFAST_SUPPORTED, &mddev->flags) &&
978 	    test_bit(FailFast, &rdev->flags) &&
979 	    !test_bit(LastDev, &rdev->flags))
980 		bio->bi_opf |= MD_FAILFAST;
981 
982 	atomic_inc(&mddev->pending_writes);
983 	submit_bio(bio);
984 }
985 
md_super_wait(struct mddev * mddev)986 int md_super_wait(struct mddev *mddev)
987 {
988 	/* wait for all superblock writes that were scheduled to complete */
989 	wait_event(mddev->sb_wait, atomic_read(&mddev->pending_writes)==0);
990 	if (test_and_clear_bit(MD_SB_NEED_REWRITE, &mddev->sb_flags))
991 		return -EAGAIN;
992 	return 0;
993 }
994 
sync_page_io(struct md_rdev * rdev,sector_t sector,int size,struct page * page,int op,int op_flags,bool metadata_op)995 int sync_page_io(struct md_rdev *rdev, sector_t sector, int size,
996 		 struct page *page, int op, int op_flags, bool metadata_op)
997 {
998 	struct bio bio;
999 	struct bio_vec bvec;
1000 
1001 	if (metadata_op && rdev->meta_bdev)
1002 		bio_init(&bio, rdev->meta_bdev, &bvec, 1, op | op_flags);
1003 	else
1004 		bio_init(&bio, rdev->bdev, &bvec, 1, op | op_flags);
1005 
1006 	if (metadata_op)
1007 		bio.bi_iter.bi_sector = sector + rdev->sb_start;
1008 	else if (rdev->mddev->reshape_position != MaxSector &&
1009 		 (rdev->mddev->reshape_backwards ==
1010 		  (sector >= rdev->mddev->reshape_position)))
1011 		bio.bi_iter.bi_sector = sector + rdev->new_data_offset;
1012 	else
1013 		bio.bi_iter.bi_sector = sector + rdev->data_offset;
1014 	bio_add_page(&bio, page, size, 0);
1015 
1016 	submit_bio_wait(&bio);
1017 
1018 	return !bio.bi_status;
1019 }
1020 EXPORT_SYMBOL_GPL(sync_page_io);
1021 
read_disk_sb(struct md_rdev * rdev,int size)1022 static int read_disk_sb(struct md_rdev *rdev, int size)
1023 {
1024 	if (rdev->sb_loaded)
1025 		return 0;
1026 
1027 	if (!sync_page_io(rdev, 0, size, rdev->sb_page, REQ_OP_READ, 0, true))
1028 		goto fail;
1029 	rdev->sb_loaded = 1;
1030 	return 0;
1031 
1032 fail:
1033 	pr_err("md: disabled device %pg, could not read superblock.\n",
1034 	       rdev->bdev);
1035 	return -EINVAL;
1036 }
1037 
md_uuid_equal(mdp_super_t * sb1,mdp_super_t * sb2)1038 static int md_uuid_equal(mdp_super_t *sb1, mdp_super_t *sb2)
1039 {
1040 	return	sb1->set_uuid0 == sb2->set_uuid0 &&
1041 		sb1->set_uuid1 == sb2->set_uuid1 &&
1042 		sb1->set_uuid2 == sb2->set_uuid2 &&
1043 		sb1->set_uuid3 == sb2->set_uuid3;
1044 }
1045 
md_sb_equal(mdp_super_t * sb1,mdp_super_t * sb2)1046 static int md_sb_equal(mdp_super_t *sb1, mdp_super_t *sb2)
1047 {
1048 	int ret;
1049 	mdp_super_t *tmp1, *tmp2;
1050 
1051 	tmp1 = kmalloc(sizeof(*tmp1),GFP_KERNEL);
1052 	tmp2 = kmalloc(sizeof(*tmp2),GFP_KERNEL);
1053 
1054 	if (!tmp1 || !tmp2) {
1055 		ret = 0;
1056 		goto abort;
1057 	}
1058 
1059 	*tmp1 = *sb1;
1060 	*tmp2 = *sb2;
1061 
1062 	/*
1063 	 * nr_disks is not constant
1064 	 */
1065 	tmp1->nr_disks = 0;
1066 	tmp2->nr_disks = 0;
1067 
1068 	ret = (memcmp(tmp1, tmp2, MD_SB_GENERIC_CONSTANT_WORDS * 4) == 0);
1069 abort:
1070 	kfree(tmp1);
1071 	kfree(tmp2);
1072 	return ret;
1073 }
1074 
md_csum_fold(u32 csum)1075 static u32 md_csum_fold(u32 csum)
1076 {
1077 	csum = (csum & 0xffff) + (csum >> 16);
1078 	return (csum & 0xffff) + (csum >> 16);
1079 }
1080 
calc_sb_csum(mdp_super_t * sb)1081 static unsigned int calc_sb_csum(mdp_super_t *sb)
1082 {
1083 	u64 newcsum = 0;
1084 	u32 *sb32 = (u32*)sb;
1085 	int i;
1086 	unsigned int disk_csum, csum;
1087 
1088 	disk_csum = sb->sb_csum;
1089 	sb->sb_csum = 0;
1090 
1091 	for (i = 0; i < MD_SB_BYTES/4 ; i++)
1092 		newcsum += sb32[i];
1093 	csum = (newcsum & 0xffffffff) + (newcsum>>32);
1094 
1095 #ifdef CONFIG_ALPHA
1096 	/* This used to use csum_partial, which was wrong for several
1097 	 * reasons including that different results are returned on
1098 	 * different architectures.  It isn't critical that we get exactly
1099 	 * the same return value as before (we always csum_fold before
1100 	 * testing, and that removes any differences).  However as we
1101 	 * know that csum_partial always returned a 16bit value on
1102 	 * alphas, do a fold to maximise conformity to previous behaviour.
1103 	 */
1104 	sb->sb_csum = md_csum_fold(disk_csum);
1105 #else
1106 	sb->sb_csum = disk_csum;
1107 #endif
1108 	return csum;
1109 }
1110 
1111 /*
1112  * Handle superblock details.
1113  * We want to be able to handle multiple superblock formats
1114  * so we have a common interface to them all, and an array of
1115  * different handlers.
1116  * We rely on user-space to write the initial superblock, and support
1117  * reading and updating of superblocks.
1118  * Interface methods are:
1119  *   int load_super(struct md_rdev *dev, struct md_rdev *refdev, int minor_version)
1120  *      loads and validates a superblock on dev.
1121  *      if refdev != NULL, compare superblocks on both devices
1122  *    Return:
1123  *      0 - dev has a superblock that is compatible with refdev
1124  *      1 - dev has a superblock that is compatible and newer than refdev
1125  *          so dev should be used as the refdev in future
1126  *     -EINVAL superblock incompatible or invalid
1127  *     -othererror e.g. -EIO
1128  *
1129  *   int validate_super(struct mddev *mddev, struct md_rdev *dev)
1130  *      Verify that dev is acceptable into mddev.
1131  *       The first time, mddev->raid_disks will be 0, and data from
1132  *       dev should be merged in.  Subsequent calls check that dev
1133  *       is new enough.  Return 0 or -EINVAL
1134  *
1135  *   void sync_super(struct mddev *mddev, struct md_rdev *dev)
1136  *     Update the superblock for rdev with data in mddev
1137  *     This does not write to disc.
1138  *
1139  */
1140 
1141 struct super_type  {
1142 	char		    *name;
1143 	struct module	    *owner;
1144 	int		    (*load_super)(struct md_rdev *rdev,
1145 					  struct md_rdev *refdev,
1146 					  int minor_version);
1147 	int		    (*validate_super)(struct mddev *mddev,
1148 					      struct md_rdev *rdev);
1149 	void		    (*sync_super)(struct mddev *mddev,
1150 					  struct md_rdev *rdev);
1151 	unsigned long long  (*rdev_size_change)(struct md_rdev *rdev,
1152 						sector_t num_sectors);
1153 	int		    (*allow_new_offset)(struct md_rdev *rdev,
1154 						unsigned long long new_offset);
1155 };
1156 
1157 /*
1158  * Check that the given mddev has no bitmap.
1159  *
1160  * This function is called from the run method of all personalities that do not
1161  * support bitmaps. It prints an error message and returns non-zero if mddev
1162  * has a bitmap. Otherwise, it returns 0.
1163  *
1164  */
md_check_no_bitmap(struct mddev * mddev)1165 int md_check_no_bitmap(struct mddev *mddev)
1166 {
1167 	if (!mddev->bitmap_info.file && !mddev->bitmap_info.offset)
1168 		return 0;
1169 	pr_warn("%s: bitmaps are not supported for %s\n",
1170 		mdname(mddev), mddev->pers->name);
1171 	return 1;
1172 }
1173 EXPORT_SYMBOL(md_check_no_bitmap);
1174 
1175 /*
1176  * load_super for 0.90.0
1177  */
super_90_load(struct md_rdev * rdev,struct md_rdev * refdev,int minor_version)1178 static int super_90_load(struct md_rdev *rdev, struct md_rdev *refdev, int minor_version)
1179 {
1180 	mdp_super_t *sb;
1181 	int ret;
1182 	bool spare_disk = true;
1183 
1184 	/*
1185 	 * Calculate the position of the superblock (512byte sectors),
1186 	 * it's at the end of the disk.
1187 	 *
1188 	 * It also happens to be a multiple of 4Kb.
1189 	 */
1190 	rdev->sb_start = calc_dev_sboffset(rdev);
1191 
1192 	ret = read_disk_sb(rdev, MD_SB_BYTES);
1193 	if (ret)
1194 		return ret;
1195 
1196 	ret = -EINVAL;
1197 
1198 	sb = page_address(rdev->sb_page);
1199 
1200 	if (sb->md_magic != MD_SB_MAGIC) {
1201 		pr_warn("md: invalid raid superblock magic on %pg\n",
1202 			rdev->bdev);
1203 		goto abort;
1204 	}
1205 
1206 	if (sb->major_version != 0 ||
1207 	    sb->minor_version < 90 ||
1208 	    sb->minor_version > 91) {
1209 		pr_warn("Bad version number %d.%d on %pg\n",
1210 			sb->major_version, sb->minor_version, rdev->bdev);
1211 		goto abort;
1212 	}
1213 
1214 	if (sb->raid_disks <= 0)
1215 		goto abort;
1216 
1217 	if (md_csum_fold(calc_sb_csum(sb)) != md_csum_fold(sb->sb_csum)) {
1218 		pr_warn("md: invalid superblock checksum on %pg\n", rdev->bdev);
1219 		goto abort;
1220 	}
1221 
1222 	rdev->preferred_minor = sb->md_minor;
1223 	rdev->data_offset = 0;
1224 	rdev->new_data_offset = 0;
1225 	rdev->sb_size = MD_SB_BYTES;
1226 	rdev->badblocks.shift = -1;
1227 
1228 	if (sb->level == LEVEL_MULTIPATH)
1229 		rdev->desc_nr = -1;
1230 	else
1231 		rdev->desc_nr = sb->this_disk.number;
1232 
1233 	/* not spare disk, or LEVEL_MULTIPATH */
1234 	if (sb->level == LEVEL_MULTIPATH ||
1235 		(rdev->desc_nr >= 0 &&
1236 		 rdev->desc_nr < MD_SB_DISKS &&
1237 		 sb->disks[rdev->desc_nr].state &
1238 		 ((1<<MD_DISK_SYNC) | (1 << MD_DISK_ACTIVE))))
1239 		spare_disk = false;
1240 
1241 	if (!refdev) {
1242 		if (!spare_disk)
1243 			ret = 1;
1244 		else
1245 			ret = 0;
1246 	} else {
1247 		__u64 ev1, ev2;
1248 		mdp_super_t *refsb = page_address(refdev->sb_page);
1249 		if (!md_uuid_equal(refsb, sb)) {
1250 			pr_warn("md: %pg has different UUID to %pg\n",
1251 				rdev->bdev, refdev->bdev);
1252 			goto abort;
1253 		}
1254 		if (!md_sb_equal(refsb, sb)) {
1255 			pr_warn("md: %pg has same UUID but different superblock to %pg\n",
1256 				rdev->bdev, refdev->bdev);
1257 			goto abort;
1258 		}
1259 		ev1 = md_event(sb);
1260 		ev2 = md_event(refsb);
1261 
1262 		if (!spare_disk && ev1 > ev2)
1263 			ret = 1;
1264 		else
1265 			ret = 0;
1266 	}
1267 	rdev->sectors = rdev->sb_start;
1268 	/* Limit to 4TB as metadata cannot record more than that.
1269 	 * (not needed for Linear and RAID0 as metadata doesn't
1270 	 * record this size)
1271 	 */
1272 	if ((u64)rdev->sectors >= (2ULL << 32) && sb->level >= 1)
1273 		rdev->sectors = (sector_t)(2ULL << 32) - 2;
1274 
1275 	if (rdev->sectors < ((sector_t)sb->size) * 2 && sb->level >= 1)
1276 		/* "this cannot possibly happen" ... */
1277 		ret = -EINVAL;
1278 
1279  abort:
1280 	return ret;
1281 }
1282 
1283 /*
1284  * validate_super for 0.90.0
1285  */
super_90_validate(struct mddev * mddev,struct md_rdev * rdev)1286 static int super_90_validate(struct mddev *mddev, struct md_rdev *rdev)
1287 {
1288 	mdp_disk_t *desc;
1289 	mdp_super_t *sb = page_address(rdev->sb_page);
1290 	__u64 ev1 = md_event(sb);
1291 
1292 	rdev->raid_disk = -1;
1293 	clear_bit(Faulty, &rdev->flags);
1294 	clear_bit(In_sync, &rdev->flags);
1295 	clear_bit(Bitmap_sync, &rdev->flags);
1296 	clear_bit(WriteMostly, &rdev->flags);
1297 
1298 	if (mddev->raid_disks == 0) {
1299 		mddev->major_version = 0;
1300 		mddev->minor_version = sb->minor_version;
1301 		mddev->patch_version = sb->patch_version;
1302 		mddev->external = 0;
1303 		mddev->chunk_sectors = sb->chunk_size >> 9;
1304 		mddev->ctime = sb->ctime;
1305 		mddev->utime = sb->utime;
1306 		mddev->level = sb->level;
1307 		mddev->clevel[0] = 0;
1308 		mddev->layout = sb->layout;
1309 		mddev->raid_disks = sb->raid_disks;
1310 		mddev->dev_sectors = ((sector_t)sb->size) * 2;
1311 		mddev->events = ev1;
1312 		mddev->bitmap_info.offset = 0;
1313 		mddev->bitmap_info.space = 0;
1314 		/* bitmap can use 60 K after the 4K superblocks */
1315 		mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9;
1316 		mddev->bitmap_info.default_space = 64*2 - (MD_SB_BYTES >> 9);
1317 		mddev->reshape_backwards = 0;
1318 
1319 		if (mddev->minor_version >= 91) {
1320 			mddev->reshape_position = sb->reshape_position;
1321 			mddev->delta_disks = sb->delta_disks;
1322 			mddev->new_level = sb->new_level;
1323 			mddev->new_layout = sb->new_layout;
1324 			mddev->new_chunk_sectors = sb->new_chunk >> 9;
1325 			if (mddev->delta_disks < 0)
1326 				mddev->reshape_backwards = 1;
1327 		} else {
1328 			mddev->reshape_position = MaxSector;
1329 			mddev->delta_disks = 0;
1330 			mddev->new_level = mddev->level;
1331 			mddev->new_layout = mddev->layout;
1332 			mddev->new_chunk_sectors = mddev->chunk_sectors;
1333 		}
1334 		if (mddev->level == 0)
1335 			mddev->layout = -1;
1336 
1337 		if (sb->state & (1<<MD_SB_CLEAN))
1338 			mddev->recovery_cp = MaxSector;
1339 		else {
1340 			if (sb->events_hi == sb->cp_events_hi &&
1341 				sb->events_lo == sb->cp_events_lo) {
1342 				mddev->recovery_cp = sb->recovery_cp;
1343 			} else
1344 				mddev->recovery_cp = 0;
1345 		}
1346 
1347 		memcpy(mddev->uuid+0, &sb->set_uuid0, 4);
1348 		memcpy(mddev->uuid+4, &sb->set_uuid1, 4);
1349 		memcpy(mddev->uuid+8, &sb->set_uuid2, 4);
1350 		memcpy(mddev->uuid+12,&sb->set_uuid3, 4);
1351 
1352 		mddev->max_disks = MD_SB_DISKS;
1353 
1354 		if (sb->state & (1<<MD_SB_BITMAP_PRESENT) &&
1355 		    mddev->bitmap_info.file == NULL) {
1356 			mddev->bitmap_info.offset =
1357 				mddev->bitmap_info.default_offset;
1358 			mddev->bitmap_info.space =
1359 				mddev->bitmap_info.default_space;
1360 		}
1361 
1362 	} else if (mddev->pers == NULL) {
1363 		/* Insist on good event counter while assembling, except
1364 		 * for spares (which don't need an event count) */
1365 		++ev1;
1366 		if (sb->disks[rdev->desc_nr].state & (
1367 			    (1<<MD_DISK_SYNC) | (1 << MD_DISK_ACTIVE)))
1368 			if (ev1 < mddev->events)
1369 				return -EINVAL;
1370 	} else if (mddev->bitmap) {
1371 		/* if adding to array with a bitmap, then we can accept an
1372 		 * older device ... but not too old.
1373 		 */
1374 		if (ev1 < mddev->bitmap->events_cleared)
1375 			return 0;
1376 		if (ev1 < mddev->events)
1377 			set_bit(Bitmap_sync, &rdev->flags);
1378 	} else {
1379 		if (ev1 < mddev->events)
1380 			/* just a hot-add of a new device, leave raid_disk at -1 */
1381 			return 0;
1382 	}
1383 
1384 	if (mddev->level != LEVEL_MULTIPATH) {
1385 		desc = sb->disks + rdev->desc_nr;
1386 
1387 		if (desc->state & (1<<MD_DISK_FAULTY))
1388 			set_bit(Faulty, &rdev->flags);
1389 		else if (desc->state & (1<<MD_DISK_SYNC) /* &&
1390 			    desc->raid_disk < mddev->raid_disks */) {
1391 			set_bit(In_sync, &rdev->flags);
1392 			rdev->raid_disk = desc->raid_disk;
1393 			rdev->saved_raid_disk = desc->raid_disk;
1394 		} else if (desc->state & (1<<MD_DISK_ACTIVE)) {
1395 			/* active but not in sync implies recovery up to
1396 			 * reshape position.  We don't know exactly where
1397 			 * that is, so set to zero for now */
1398 			if (mddev->minor_version >= 91) {
1399 				rdev->recovery_offset = 0;
1400 				rdev->raid_disk = desc->raid_disk;
1401 			}
1402 		}
1403 		if (desc->state & (1<<MD_DISK_WRITEMOSTLY))
1404 			set_bit(WriteMostly, &rdev->flags);
1405 		if (desc->state & (1<<MD_DISK_FAILFAST))
1406 			set_bit(FailFast, &rdev->flags);
1407 	} else /* MULTIPATH are always insync */
1408 		set_bit(In_sync, &rdev->flags);
1409 	return 0;
1410 }
1411 
1412 /*
1413  * sync_super for 0.90.0
1414  */
super_90_sync(struct mddev * mddev,struct md_rdev * rdev)1415 static void super_90_sync(struct mddev *mddev, struct md_rdev *rdev)
1416 {
1417 	mdp_super_t *sb;
1418 	struct md_rdev *rdev2;
1419 	int next_spare = mddev->raid_disks;
1420 
1421 	/* make rdev->sb match mddev data..
1422 	 *
1423 	 * 1/ zero out disks
1424 	 * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
1425 	 * 3/ any empty disks < next_spare become removed
1426 	 *
1427 	 * disks[0] gets initialised to REMOVED because
1428 	 * we cannot be sure from other fields if it has
1429 	 * been initialised or not.
1430 	 */
1431 	int i;
1432 	int active=0, working=0,failed=0,spare=0,nr_disks=0;
1433 
1434 	rdev->sb_size = MD_SB_BYTES;
1435 
1436 	sb = page_address(rdev->sb_page);
1437 
1438 	memset(sb, 0, sizeof(*sb));
1439 
1440 	sb->md_magic = MD_SB_MAGIC;
1441 	sb->major_version = mddev->major_version;
1442 	sb->patch_version = mddev->patch_version;
1443 	sb->gvalid_words  = 0; /* ignored */
1444 	memcpy(&sb->set_uuid0, mddev->uuid+0, 4);
1445 	memcpy(&sb->set_uuid1, mddev->uuid+4, 4);
1446 	memcpy(&sb->set_uuid2, mddev->uuid+8, 4);
1447 	memcpy(&sb->set_uuid3, mddev->uuid+12,4);
1448 
1449 	sb->ctime = clamp_t(time64_t, mddev->ctime, 0, U32_MAX);
1450 	sb->level = mddev->level;
1451 	sb->size = mddev->dev_sectors / 2;
1452 	sb->raid_disks = mddev->raid_disks;
1453 	sb->md_minor = mddev->md_minor;
1454 	sb->not_persistent = 0;
1455 	sb->utime = clamp_t(time64_t, mddev->utime, 0, U32_MAX);
1456 	sb->state = 0;
1457 	sb->events_hi = (mddev->events>>32);
1458 	sb->events_lo = (u32)mddev->events;
1459 
1460 	if (mddev->reshape_position == MaxSector)
1461 		sb->minor_version = 90;
1462 	else {
1463 		sb->minor_version = 91;
1464 		sb->reshape_position = mddev->reshape_position;
1465 		sb->new_level = mddev->new_level;
1466 		sb->delta_disks = mddev->delta_disks;
1467 		sb->new_layout = mddev->new_layout;
1468 		sb->new_chunk = mddev->new_chunk_sectors << 9;
1469 	}
1470 	mddev->minor_version = sb->minor_version;
1471 	if (mddev->in_sync)
1472 	{
1473 		sb->recovery_cp = mddev->recovery_cp;
1474 		sb->cp_events_hi = (mddev->events>>32);
1475 		sb->cp_events_lo = (u32)mddev->events;
1476 		if (mddev->recovery_cp == MaxSector)
1477 			sb->state = (1<< MD_SB_CLEAN);
1478 	} else
1479 		sb->recovery_cp = 0;
1480 
1481 	sb->layout = mddev->layout;
1482 	sb->chunk_size = mddev->chunk_sectors << 9;
1483 
1484 	if (mddev->bitmap && mddev->bitmap_info.file == NULL)
1485 		sb->state |= (1<<MD_SB_BITMAP_PRESENT);
1486 
1487 	sb->disks[0].state = (1<<MD_DISK_REMOVED);
1488 	rdev_for_each(rdev2, mddev) {
1489 		mdp_disk_t *d;
1490 		int desc_nr;
1491 		int is_active = test_bit(In_sync, &rdev2->flags);
1492 
1493 		if (rdev2->raid_disk >= 0 &&
1494 		    sb->minor_version >= 91)
1495 			/* we have nowhere to store the recovery_offset,
1496 			 * but if it is not below the reshape_position,
1497 			 * we can piggy-back on that.
1498 			 */
1499 			is_active = 1;
1500 		if (rdev2->raid_disk < 0 ||
1501 		    test_bit(Faulty, &rdev2->flags))
1502 			is_active = 0;
1503 		if (is_active)
1504 			desc_nr = rdev2->raid_disk;
1505 		else
1506 			desc_nr = next_spare++;
1507 		rdev2->desc_nr = desc_nr;
1508 		d = &sb->disks[rdev2->desc_nr];
1509 		nr_disks++;
1510 		d->number = rdev2->desc_nr;
1511 		d->major = MAJOR(rdev2->bdev->bd_dev);
1512 		d->minor = MINOR(rdev2->bdev->bd_dev);
1513 		if (is_active)
1514 			d->raid_disk = rdev2->raid_disk;
1515 		else
1516 			d->raid_disk = rdev2->desc_nr; /* compatibility */
1517 		if (test_bit(Faulty, &rdev2->flags))
1518 			d->state = (1<<MD_DISK_FAULTY);
1519 		else if (is_active) {
1520 			d->state = (1<<MD_DISK_ACTIVE);
1521 			if (test_bit(In_sync, &rdev2->flags))
1522 				d->state |= (1<<MD_DISK_SYNC);
1523 			active++;
1524 			working++;
1525 		} else {
1526 			d->state = 0;
1527 			spare++;
1528 			working++;
1529 		}
1530 		if (test_bit(WriteMostly, &rdev2->flags))
1531 			d->state |= (1<<MD_DISK_WRITEMOSTLY);
1532 		if (test_bit(FailFast, &rdev2->flags))
1533 			d->state |= (1<<MD_DISK_FAILFAST);
1534 	}
1535 	/* now set the "removed" and "faulty" bits on any missing devices */
1536 	for (i=0 ; i < mddev->raid_disks ; i++) {
1537 		mdp_disk_t *d = &sb->disks[i];
1538 		if (d->state == 0 && d->number == 0) {
1539 			d->number = i;
1540 			d->raid_disk = i;
1541 			d->state = (1<<MD_DISK_REMOVED);
1542 			d->state |= (1<<MD_DISK_FAULTY);
1543 			failed++;
1544 		}
1545 	}
1546 	sb->nr_disks = nr_disks;
1547 	sb->active_disks = active;
1548 	sb->working_disks = working;
1549 	sb->failed_disks = failed;
1550 	sb->spare_disks = spare;
1551 
1552 	sb->this_disk = sb->disks[rdev->desc_nr];
1553 	sb->sb_csum = calc_sb_csum(sb);
1554 }
1555 
1556 /*
1557  * rdev_size_change for 0.90.0
1558  */
1559 static unsigned long long
super_90_rdev_size_change(struct md_rdev * rdev,sector_t num_sectors)1560 super_90_rdev_size_change(struct md_rdev *rdev, sector_t num_sectors)
1561 {
1562 	if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
1563 		return 0; /* component must fit device */
1564 	if (rdev->mddev->bitmap_info.offset)
1565 		return 0; /* can't move bitmap */
1566 	rdev->sb_start = calc_dev_sboffset(rdev);
1567 	if (!num_sectors || num_sectors > rdev->sb_start)
1568 		num_sectors = rdev->sb_start;
1569 	/* Limit to 4TB as metadata cannot record more than that.
1570 	 * 4TB == 2^32 KB, or 2*2^32 sectors.
1571 	 */
1572 	if ((u64)num_sectors >= (2ULL << 32) && rdev->mddev->level >= 1)
1573 		num_sectors = (sector_t)(2ULL << 32) - 2;
1574 	do {
1575 		md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
1576 		       rdev->sb_page);
1577 	} while (md_super_wait(rdev->mddev) < 0);
1578 	return num_sectors;
1579 }
1580 
1581 static int
super_90_allow_new_offset(struct md_rdev * rdev,unsigned long long new_offset)1582 super_90_allow_new_offset(struct md_rdev *rdev, unsigned long long new_offset)
1583 {
1584 	/* non-zero offset changes not possible with v0.90 */
1585 	return new_offset == 0;
1586 }
1587 
1588 /*
1589  * version 1 superblock
1590  */
1591 
calc_sb_1_csum(struct mdp_superblock_1 * sb)1592 static __le32 calc_sb_1_csum(struct mdp_superblock_1 *sb)
1593 {
1594 	__le32 disk_csum;
1595 	u32 csum;
1596 	unsigned long long newcsum;
1597 	int size = 256 + le32_to_cpu(sb->max_dev)*2;
1598 	__le32 *isuper = (__le32*)sb;
1599 
1600 	disk_csum = sb->sb_csum;
1601 	sb->sb_csum = 0;
1602 	newcsum = 0;
1603 	for (; size >= 4; size -= 4)
1604 		newcsum += le32_to_cpu(*isuper++);
1605 
1606 	if (size == 2)
1607 		newcsum += le16_to_cpu(*(__le16*) isuper);
1608 
1609 	csum = (newcsum & 0xffffffff) + (newcsum >> 32);
1610 	sb->sb_csum = disk_csum;
1611 	return cpu_to_le32(csum);
1612 }
1613 
super_1_load(struct md_rdev * rdev,struct md_rdev * refdev,int minor_version)1614 static int super_1_load(struct md_rdev *rdev, struct md_rdev *refdev, int minor_version)
1615 {
1616 	struct mdp_superblock_1 *sb;
1617 	int ret;
1618 	sector_t sb_start;
1619 	sector_t sectors;
1620 	int bmask;
1621 	bool spare_disk = true;
1622 
1623 	/*
1624 	 * Calculate the position of the superblock in 512byte sectors.
1625 	 * It is always aligned to a 4K boundary and
1626 	 * depeding on minor_version, it can be:
1627 	 * 0: At least 8K, but less than 12K, from end of device
1628 	 * 1: At start of device
1629 	 * 2: 4K from start of device.
1630 	 */
1631 	switch(minor_version) {
1632 	case 0:
1633 		sb_start = bdev_nr_sectors(rdev->bdev) - 8 * 2;
1634 		sb_start &= ~(sector_t)(4*2-1);
1635 		break;
1636 	case 1:
1637 		sb_start = 0;
1638 		break;
1639 	case 2:
1640 		sb_start = 8;
1641 		break;
1642 	default:
1643 		return -EINVAL;
1644 	}
1645 	rdev->sb_start = sb_start;
1646 
1647 	/* superblock is rarely larger than 1K, but it can be larger,
1648 	 * and it is safe to read 4k, so we do that
1649 	 */
1650 	ret = read_disk_sb(rdev, 4096);
1651 	if (ret) return ret;
1652 
1653 	sb = page_address(rdev->sb_page);
1654 
1655 	if (sb->magic != cpu_to_le32(MD_SB_MAGIC) ||
1656 	    sb->major_version != cpu_to_le32(1) ||
1657 	    le32_to_cpu(sb->max_dev) > (4096-256)/2 ||
1658 	    le64_to_cpu(sb->super_offset) != rdev->sb_start ||
1659 	    (le32_to_cpu(sb->feature_map) & ~MD_FEATURE_ALL) != 0)
1660 		return -EINVAL;
1661 
1662 	if (calc_sb_1_csum(sb) != sb->sb_csum) {
1663 		pr_warn("md: invalid superblock checksum on %pg\n",
1664 			rdev->bdev);
1665 		return -EINVAL;
1666 	}
1667 	if (le64_to_cpu(sb->data_size) < 10) {
1668 		pr_warn("md: data_size too small on %pg\n",
1669 			rdev->bdev);
1670 		return -EINVAL;
1671 	}
1672 	if (sb->pad0 ||
1673 	    sb->pad3[0] ||
1674 	    memcmp(sb->pad3, sb->pad3+1, sizeof(sb->pad3) - sizeof(sb->pad3[1])))
1675 		/* Some padding is non-zero, might be a new feature */
1676 		return -EINVAL;
1677 
1678 	rdev->preferred_minor = 0xffff;
1679 	rdev->data_offset = le64_to_cpu(sb->data_offset);
1680 	rdev->new_data_offset = rdev->data_offset;
1681 	if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE) &&
1682 	    (le32_to_cpu(sb->feature_map) & MD_FEATURE_NEW_OFFSET))
1683 		rdev->new_data_offset += (s32)le32_to_cpu(sb->new_offset);
1684 	atomic_set(&rdev->corrected_errors, le32_to_cpu(sb->cnt_corrected_read));
1685 
1686 	rdev->sb_size = le32_to_cpu(sb->max_dev) * 2 + 256;
1687 	bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
1688 	if (rdev->sb_size & bmask)
1689 		rdev->sb_size = (rdev->sb_size | bmask) + 1;
1690 
1691 	if (minor_version
1692 	    && rdev->data_offset < sb_start + (rdev->sb_size/512))
1693 		return -EINVAL;
1694 	if (minor_version
1695 	    && rdev->new_data_offset < sb_start + (rdev->sb_size/512))
1696 		return -EINVAL;
1697 
1698 	if (sb->level == cpu_to_le32(LEVEL_MULTIPATH))
1699 		rdev->desc_nr = -1;
1700 	else
1701 		rdev->desc_nr = le32_to_cpu(sb->dev_number);
1702 
1703 	if (!rdev->bb_page) {
1704 		rdev->bb_page = alloc_page(GFP_KERNEL);
1705 		if (!rdev->bb_page)
1706 			return -ENOMEM;
1707 	}
1708 	if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BAD_BLOCKS) &&
1709 	    rdev->badblocks.count == 0) {
1710 		/* need to load the bad block list.
1711 		 * Currently we limit it to one page.
1712 		 */
1713 		s32 offset;
1714 		sector_t bb_sector;
1715 		__le64 *bbp;
1716 		int i;
1717 		int sectors = le16_to_cpu(sb->bblog_size);
1718 		if (sectors > (PAGE_SIZE / 512))
1719 			return -EINVAL;
1720 		offset = le32_to_cpu(sb->bblog_offset);
1721 		if (offset == 0)
1722 			return -EINVAL;
1723 		bb_sector = (long long)offset;
1724 		if (!sync_page_io(rdev, bb_sector, sectors << 9,
1725 				  rdev->bb_page, REQ_OP_READ, 0, true))
1726 			return -EIO;
1727 		bbp = (__le64 *)page_address(rdev->bb_page);
1728 		rdev->badblocks.shift = sb->bblog_shift;
1729 		for (i = 0 ; i < (sectors << (9-3)) ; i++, bbp++) {
1730 			u64 bb = le64_to_cpu(*bbp);
1731 			int count = bb & (0x3ff);
1732 			u64 sector = bb >> 10;
1733 			sector <<= sb->bblog_shift;
1734 			count <<= sb->bblog_shift;
1735 			if (bb + 1 == 0)
1736 				break;
1737 			if (badblocks_set(&rdev->badblocks, sector, count, 1))
1738 				return -EINVAL;
1739 		}
1740 	} else if (sb->bblog_offset != 0)
1741 		rdev->badblocks.shift = 0;
1742 
1743 	if ((le32_to_cpu(sb->feature_map) &
1744 	    (MD_FEATURE_PPL | MD_FEATURE_MULTIPLE_PPLS))) {
1745 		rdev->ppl.offset = (__s16)le16_to_cpu(sb->ppl.offset);
1746 		rdev->ppl.size = le16_to_cpu(sb->ppl.size);
1747 		rdev->ppl.sector = rdev->sb_start + rdev->ppl.offset;
1748 	}
1749 
1750 	if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RAID0_LAYOUT) &&
1751 	    sb->level != 0)
1752 		return -EINVAL;
1753 
1754 	/* not spare disk, or LEVEL_MULTIPATH */
1755 	if (sb->level == cpu_to_le32(LEVEL_MULTIPATH) ||
1756 		(rdev->desc_nr >= 0 &&
1757 		rdev->desc_nr < le32_to_cpu(sb->max_dev) &&
1758 		(le16_to_cpu(sb->dev_roles[rdev->desc_nr]) < MD_DISK_ROLE_MAX ||
1759 		 le16_to_cpu(sb->dev_roles[rdev->desc_nr]) == MD_DISK_ROLE_JOURNAL)))
1760 		spare_disk = false;
1761 
1762 	if (!refdev) {
1763 		if (!spare_disk)
1764 			ret = 1;
1765 		else
1766 			ret = 0;
1767 	} else {
1768 		__u64 ev1, ev2;
1769 		struct mdp_superblock_1 *refsb = page_address(refdev->sb_page);
1770 
1771 		if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 ||
1772 		    sb->level != refsb->level ||
1773 		    sb->layout != refsb->layout ||
1774 		    sb->chunksize != refsb->chunksize) {
1775 			pr_warn("md: %pg has strangely different superblock to %pg\n",
1776 				rdev->bdev,
1777 				refdev->bdev);
1778 			return -EINVAL;
1779 		}
1780 		ev1 = le64_to_cpu(sb->events);
1781 		ev2 = le64_to_cpu(refsb->events);
1782 
1783 		if (!spare_disk && ev1 > ev2)
1784 			ret = 1;
1785 		else
1786 			ret = 0;
1787 	}
1788 	if (minor_version)
1789 		sectors = bdev_nr_sectors(rdev->bdev) - rdev->data_offset;
1790 	else
1791 		sectors = rdev->sb_start;
1792 	if (sectors < le64_to_cpu(sb->data_size))
1793 		return -EINVAL;
1794 	rdev->sectors = le64_to_cpu(sb->data_size);
1795 	return ret;
1796 }
1797 
super_1_validate(struct mddev * mddev,struct md_rdev * rdev)1798 static int super_1_validate(struct mddev *mddev, struct md_rdev *rdev)
1799 {
1800 	struct mdp_superblock_1 *sb = page_address(rdev->sb_page);
1801 	__u64 ev1 = le64_to_cpu(sb->events);
1802 
1803 	rdev->raid_disk = -1;
1804 	clear_bit(Faulty, &rdev->flags);
1805 	clear_bit(In_sync, &rdev->flags);
1806 	clear_bit(Bitmap_sync, &rdev->flags);
1807 	clear_bit(WriteMostly, &rdev->flags);
1808 
1809 	if (mddev->raid_disks == 0) {
1810 		mddev->major_version = 1;
1811 		mddev->patch_version = 0;
1812 		mddev->external = 0;
1813 		mddev->chunk_sectors = le32_to_cpu(sb->chunksize);
1814 		mddev->ctime = le64_to_cpu(sb->ctime);
1815 		mddev->utime = le64_to_cpu(sb->utime);
1816 		mddev->level = le32_to_cpu(sb->level);
1817 		mddev->clevel[0] = 0;
1818 		mddev->layout = le32_to_cpu(sb->layout);
1819 		mddev->raid_disks = le32_to_cpu(sb->raid_disks);
1820 		mddev->dev_sectors = le64_to_cpu(sb->size);
1821 		mddev->events = ev1;
1822 		mddev->bitmap_info.offset = 0;
1823 		mddev->bitmap_info.space = 0;
1824 		/* Default location for bitmap is 1K after superblock
1825 		 * using 3K - total of 4K
1826 		 */
1827 		mddev->bitmap_info.default_offset = 1024 >> 9;
1828 		mddev->bitmap_info.default_space = (4096-1024) >> 9;
1829 		mddev->reshape_backwards = 0;
1830 
1831 		mddev->recovery_cp = le64_to_cpu(sb->resync_offset);
1832 		memcpy(mddev->uuid, sb->set_uuid, 16);
1833 
1834 		mddev->max_disks =  (4096-256)/2;
1835 
1836 		if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) &&
1837 		    mddev->bitmap_info.file == NULL) {
1838 			mddev->bitmap_info.offset =
1839 				(__s32)le32_to_cpu(sb->bitmap_offset);
1840 			/* Metadata doesn't record how much space is available.
1841 			 * For 1.0, we assume we can use up to the superblock
1842 			 * if before, else to 4K beyond superblock.
1843 			 * For others, assume no change is possible.
1844 			 */
1845 			if (mddev->minor_version > 0)
1846 				mddev->bitmap_info.space = 0;
1847 			else if (mddev->bitmap_info.offset > 0)
1848 				mddev->bitmap_info.space =
1849 					8 - mddev->bitmap_info.offset;
1850 			else
1851 				mddev->bitmap_info.space =
1852 					-mddev->bitmap_info.offset;
1853 		}
1854 
1855 		if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) {
1856 			mddev->reshape_position = le64_to_cpu(sb->reshape_position);
1857 			mddev->delta_disks = le32_to_cpu(sb->delta_disks);
1858 			mddev->new_level = le32_to_cpu(sb->new_level);
1859 			mddev->new_layout = le32_to_cpu(sb->new_layout);
1860 			mddev->new_chunk_sectors = le32_to_cpu(sb->new_chunk);
1861 			if (mddev->delta_disks < 0 ||
1862 			    (mddev->delta_disks == 0 &&
1863 			     (le32_to_cpu(sb->feature_map)
1864 			      & MD_FEATURE_RESHAPE_BACKWARDS)))
1865 				mddev->reshape_backwards = 1;
1866 		} else {
1867 			mddev->reshape_position = MaxSector;
1868 			mddev->delta_disks = 0;
1869 			mddev->new_level = mddev->level;
1870 			mddev->new_layout = mddev->layout;
1871 			mddev->new_chunk_sectors = mddev->chunk_sectors;
1872 		}
1873 
1874 		if (mddev->level == 0 &&
1875 		    !(le32_to_cpu(sb->feature_map) & MD_FEATURE_RAID0_LAYOUT))
1876 			mddev->layout = -1;
1877 
1878 		if (le32_to_cpu(sb->feature_map) & MD_FEATURE_JOURNAL)
1879 			set_bit(MD_HAS_JOURNAL, &mddev->flags);
1880 
1881 		if (le32_to_cpu(sb->feature_map) &
1882 		    (MD_FEATURE_PPL | MD_FEATURE_MULTIPLE_PPLS)) {
1883 			if (le32_to_cpu(sb->feature_map) &
1884 			    (MD_FEATURE_BITMAP_OFFSET | MD_FEATURE_JOURNAL))
1885 				return -EINVAL;
1886 			if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_PPL) &&
1887 			    (le32_to_cpu(sb->feature_map) &
1888 					    MD_FEATURE_MULTIPLE_PPLS))
1889 				return -EINVAL;
1890 			set_bit(MD_HAS_PPL, &mddev->flags);
1891 		}
1892 	} else if (mddev->pers == NULL) {
1893 		/* Insist of good event counter while assembling, except for
1894 		 * spares (which don't need an event count) */
1895 		++ev1;
1896 		if (rdev->desc_nr >= 0 &&
1897 		    rdev->desc_nr < le32_to_cpu(sb->max_dev) &&
1898 		    (le16_to_cpu(sb->dev_roles[rdev->desc_nr]) < MD_DISK_ROLE_MAX ||
1899 		     le16_to_cpu(sb->dev_roles[rdev->desc_nr]) == MD_DISK_ROLE_JOURNAL))
1900 			if (ev1 < mddev->events)
1901 				return -EINVAL;
1902 	} else if (mddev->bitmap) {
1903 		/* If adding to array with a bitmap, then we can accept an
1904 		 * older device, but not too old.
1905 		 */
1906 		if (ev1 < mddev->bitmap->events_cleared)
1907 			return 0;
1908 		if (ev1 < mddev->events)
1909 			set_bit(Bitmap_sync, &rdev->flags);
1910 	} else {
1911 		if (ev1 < mddev->events)
1912 			/* just a hot-add of a new device, leave raid_disk at -1 */
1913 			return 0;
1914 	}
1915 	if (mddev->level != LEVEL_MULTIPATH) {
1916 		int role;
1917 		if (rdev->desc_nr < 0 ||
1918 		    rdev->desc_nr >= le32_to_cpu(sb->max_dev)) {
1919 			role = MD_DISK_ROLE_SPARE;
1920 			rdev->desc_nr = -1;
1921 		} else
1922 			role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
1923 		switch(role) {
1924 		case MD_DISK_ROLE_SPARE: /* spare */
1925 			break;
1926 		case MD_DISK_ROLE_FAULTY: /* faulty */
1927 			set_bit(Faulty, &rdev->flags);
1928 			break;
1929 		case MD_DISK_ROLE_JOURNAL: /* journal device */
1930 			if (!(le32_to_cpu(sb->feature_map) & MD_FEATURE_JOURNAL)) {
1931 				/* journal device without journal feature */
1932 				pr_warn("md: journal device provided without journal feature, ignoring the device\n");
1933 				return -EINVAL;
1934 			}
1935 			set_bit(Journal, &rdev->flags);
1936 			rdev->journal_tail = le64_to_cpu(sb->journal_tail);
1937 			rdev->raid_disk = 0;
1938 			break;
1939 		default:
1940 			rdev->saved_raid_disk = role;
1941 			if ((le32_to_cpu(sb->feature_map) &
1942 			     MD_FEATURE_RECOVERY_OFFSET)) {
1943 				rdev->recovery_offset = le64_to_cpu(sb->recovery_offset);
1944 				if (!(le32_to_cpu(sb->feature_map) &
1945 				      MD_FEATURE_RECOVERY_BITMAP))
1946 					rdev->saved_raid_disk = -1;
1947 			} else {
1948 				/*
1949 				 * If the array is FROZEN, then the device can't
1950 				 * be in_sync with rest of array.
1951 				 */
1952 				if (!test_bit(MD_RECOVERY_FROZEN,
1953 					      &mddev->recovery))
1954 					set_bit(In_sync, &rdev->flags);
1955 			}
1956 			rdev->raid_disk = role;
1957 			break;
1958 		}
1959 		if (sb->devflags & WriteMostly1)
1960 			set_bit(WriteMostly, &rdev->flags);
1961 		if (sb->devflags & FailFast1)
1962 			set_bit(FailFast, &rdev->flags);
1963 		if (le32_to_cpu(sb->feature_map) & MD_FEATURE_REPLACEMENT)
1964 			set_bit(Replacement, &rdev->flags);
1965 	} else /* MULTIPATH are always insync */
1966 		set_bit(In_sync, &rdev->flags);
1967 
1968 	return 0;
1969 }
1970 
super_1_sync(struct mddev * mddev,struct md_rdev * rdev)1971 static void super_1_sync(struct mddev *mddev, struct md_rdev *rdev)
1972 {
1973 	struct mdp_superblock_1 *sb;
1974 	struct md_rdev *rdev2;
1975 	int max_dev, i;
1976 	/* make rdev->sb match mddev and rdev data. */
1977 
1978 	sb = page_address(rdev->sb_page);
1979 
1980 	sb->feature_map = 0;
1981 	sb->pad0 = 0;
1982 	sb->recovery_offset = cpu_to_le64(0);
1983 	memset(sb->pad3, 0, sizeof(sb->pad3));
1984 
1985 	sb->utime = cpu_to_le64((__u64)mddev->utime);
1986 	sb->events = cpu_to_le64(mddev->events);
1987 	if (mddev->in_sync)
1988 		sb->resync_offset = cpu_to_le64(mddev->recovery_cp);
1989 	else if (test_bit(MD_JOURNAL_CLEAN, &mddev->flags))
1990 		sb->resync_offset = cpu_to_le64(MaxSector);
1991 	else
1992 		sb->resync_offset = cpu_to_le64(0);
1993 
1994 	sb->cnt_corrected_read = cpu_to_le32(atomic_read(&rdev->corrected_errors));
1995 
1996 	sb->raid_disks = cpu_to_le32(mddev->raid_disks);
1997 	sb->size = cpu_to_le64(mddev->dev_sectors);
1998 	sb->chunksize = cpu_to_le32(mddev->chunk_sectors);
1999 	sb->level = cpu_to_le32(mddev->level);
2000 	sb->layout = cpu_to_le32(mddev->layout);
2001 	if (test_bit(FailFast, &rdev->flags))
2002 		sb->devflags |= FailFast1;
2003 	else
2004 		sb->devflags &= ~FailFast1;
2005 
2006 	if (test_bit(WriteMostly, &rdev->flags))
2007 		sb->devflags |= WriteMostly1;
2008 	else
2009 		sb->devflags &= ~WriteMostly1;
2010 	sb->data_offset = cpu_to_le64(rdev->data_offset);
2011 	sb->data_size = cpu_to_le64(rdev->sectors);
2012 
2013 	if (mddev->bitmap && mddev->bitmap_info.file == NULL) {
2014 		sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_info.offset);
2015 		sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET);
2016 	}
2017 
2018 	if (rdev->raid_disk >= 0 && !test_bit(Journal, &rdev->flags) &&
2019 	    !test_bit(In_sync, &rdev->flags)) {
2020 		sb->feature_map |=
2021 			cpu_to_le32(MD_FEATURE_RECOVERY_OFFSET);
2022 		sb->recovery_offset =
2023 			cpu_to_le64(rdev->recovery_offset);
2024 		if (rdev->saved_raid_disk >= 0 && mddev->bitmap)
2025 			sb->feature_map |=
2026 				cpu_to_le32(MD_FEATURE_RECOVERY_BITMAP);
2027 	}
2028 	/* Note: recovery_offset and journal_tail share space  */
2029 	if (test_bit(Journal, &rdev->flags))
2030 		sb->journal_tail = cpu_to_le64(rdev->journal_tail);
2031 	if (test_bit(Replacement, &rdev->flags))
2032 		sb->feature_map |=
2033 			cpu_to_le32(MD_FEATURE_REPLACEMENT);
2034 
2035 	if (mddev->reshape_position != MaxSector) {
2036 		sb->feature_map |= cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE);
2037 		sb->reshape_position = cpu_to_le64(mddev->reshape_position);
2038 		sb->new_layout = cpu_to_le32(mddev->new_layout);
2039 		sb->delta_disks = cpu_to_le32(mddev->delta_disks);
2040 		sb->new_level = cpu_to_le32(mddev->new_level);
2041 		sb->new_chunk = cpu_to_le32(mddev->new_chunk_sectors);
2042 		if (mddev->delta_disks == 0 &&
2043 		    mddev->reshape_backwards)
2044 			sb->feature_map
2045 				|= cpu_to_le32(MD_FEATURE_RESHAPE_BACKWARDS);
2046 		if (rdev->new_data_offset != rdev->data_offset) {
2047 			sb->feature_map
2048 				|= cpu_to_le32(MD_FEATURE_NEW_OFFSET);
2049 			sb->new_offset = cpu_to_le32((__u32)(rdev->new_data_offset
2050 							     - rdev->data_offset));
2051 		}
2052 	}
2053 
2054 	if (mddev_is_clustered(mddev))
2055 		sb->feature_map |= cpu_to_le32(MD_FEATURE_CLUSTERED);
2056 
2057 	if (rdev->badblocks.count == 0)
2058 		/* Nothing to do for bad blocks*/ ;
2059 	else if (sb->bblog_offset == 0)
2060 		/* Cannot record bad blocks on this device */
2061 		md_error(mddev, rdev);
2062 	else {
2063 		struct badblocks *bb = &rdev->badblocks;
2064 		__le64 *bbp = (__le64 *)page_address(rdev->bb_page);
2065 		u64 *p = bb->page;
2066 		sb->feature_map |= cpu_to_le32(MD_FEATURE_BAD_BLOCKS);
2067 		if (bb->changed) {
2068 			unsigned seq;
2069 
2070 retry:
2071 			seq = read_seqbegin(&bb->lock);
2072 
2073 			memset(bbp, 0xff, PAGE_SIZE);
2074 
2075 			for (i = 0 ; i < bb->count ; i++) {
2076 				u64 internal_bb = p[i];
2077 				u64 store_bb = ((BB_OFFSET(internal_bb) << 10)
2078 						| BB_LEN(internal_bb));
2079 				bbp[i] = cpu_to_le64(store_bb);
2080 			}
2081 			bb->changed = 0;
2082 			if (read_seqretry(&bb->lock, seq))
2083 				goto retry;
2084 
2085 			bb->sector = (rdev->sb_start +
2086 				      (int)le32_to_cpu(sb->bblog_offset));
2087 			bb->size = le16_to_cpu(sb->bblog_size);
2088 		}
2089 	}
2090 
2091 	max_dev = 0;
2092 	rdev_for_each(rdev2, mddev)
2093 		if (rdev2->desc_nr+1 > max_dev)
2094 			max_dev = rdev2->desc_nr+1;
2095 
2096 	if (max_dev > le32_to_cpu(sb->max_dev)) {
2097 		int bmask;
2098 		sb->max_dev = cpu_to_le32(max_dev);
2099 		rdev->sb_size = max_dev * 2 + 256;
2100 		bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
2101 		if (rdev->sb_size & bmask)
2102 			rdev->sb_size = (rdev->sb_size | bmask) + 1;
2103 	} else
2104 		max_dev = le32_to_cpu(sb->max_dev);
2105 
2106 	for (i=0; i<max_dev;i++)
2107 		sb->dev_roles[i] = cpu_to_le16(MD_DISK_ROLE_SPARE);
2108 
2109 	if (test_bit(MD_HAS_JOURNAL, &mddev->flags))
2110 		sb->feature_map |= cpu_to_le32(MD_FEATURE_JOURNAL);
2111 
2112 	if (test_bit(MD_HAS_PPL, &mddev->flags)) {
2113 		if (test_bit(MD_HAS_MULTIPLE_PPLS, &mddev->flags))
2114 			sb->feature_map |=
2115 			    cpu_to_le32(MD_FEATURE_MULTIPLE_PPLS);
2116 		else
2117 			sb->feature_map |= cpu_to_le32(MD_FEATURE_PPL);
2118 		sb->ppl.offset = cpu_to_le16(rdev->ppl.offset);
2119 		sb->ppl.size = cpu_to_le16(rdev->ppl.size);
2120 	}
2121 
2122 	rdev_for_each(rdev2, mddev) {
2123 		i = rdev2->desc_nr;
2124 		if (test_bit(Faulty, &rdev2->flags))
2125 			sb->dev_roles[i] = cpu_to_le16(MD_DISK_ROLE_FAULTY);
2126 		else if (test_bit(In_sync, &rdev2->flags))
2127 			sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
2128 		else if (test_bit(Journal, &rdev2->flags))
2129 			sb->dev_roles[i] = cpu_to_le16(MD_DISK_ROLE_JOURNAL);
2130 		else if (rdev2->raid_disk >= 0)
2131 			sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
2132 		else
2133 			sb->dev_roles[i] = cpu_to_le16(MD_DISK_ROLE_SPARE);
2134 	}
2135 
2136 	sb->sb_csum = calc_sb_1_csum(sb);
2137 }
2138 
super_1_choose_bm_space(sector_t dev_size)2139 static sector_t super_1_choose_bm_space(sector_t dev_size)
2140 {
2141 	sector_t bm_space;
2142 
2143 	/* if the device is bigger than 8Gig, save 64k for bitmap
2144 	 * usage, if bigger than 200Gig, save 128k
2145 	 */
2146 	if (dev_size < 64*2)
2147 		bm_space = 0;
2148 	else if (dev_size - 64*2 >= 200*1024*1024*2)
2149 		bm_space = 128*2;
2150 	else if (dev_size - 4*2 > 8*1024*1024*2)
2151 		bm_space = 64*2;
2152 	else
2153 		bm_space = 4*2;
2154 	return bm_space;
2155 }
2156 
2157 static unsigned long long
super_1_rdev_size_change(struct md_rdev * rdev,sector_t num_sectors)2158 super_1_rdev_size_change(struct md_rdev *rdev, sector_t num_sectors)
2159 {
2160 	struct mdp_superblock_1 *sb;
2161 	sector_t max_sectors;
2162 	if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
2163 		return 0; /* component must fit device */
2164 	if (rdev->data_offset != rdev->new_data_offset)
2165 		return 0; /* too confusing */
2166 	if (rdev->sb_start < rdev->data_offset) {
2167 		/* minor versions 1 and 2; superblock before data */
2168 		max_sectors = bdev_nr_sectors(rdev->bdev) - rdev->data_offset;
2169 		if (!num_sectors || num_sectors > max_sectors)
2170 			num_sectors = max_sectors;
2171 	} else if (rdev->mddev->bitmap_info.offset) {
2172 		/* minor version 0 with bitmap we can't move */
2173 		return 0;
2174 	} else {
2175 		/* minor version 0; superblock after data */
2176 		sector_t sb_start, bm_space;
2177 		sector_t dev_size = bdev_nr_sectors(rdev->bdev);
2178 
2179 		/* 8K is for superblock */
2180 		sb_start = dev_size - 8*2;
2181 		sb_start &= ~(sector_t)(4*2 - 1);
2182 
2183 		bm_space = super_1_choose_bm_space(dev_size);
2184 
2185 		/* Space that can be used to store date needs to decrease
2186 		 * superblock bitmap space and bad block space(4K)
2187 		 */
2188 		max_sectors = sb_start - bm_space - 4*2;
2189 
2190 		if (!num_sectors || num_sectors > max_sectors)
2191 			num_sectors = max_sectors;
2192 		rdev->sb_start = sb_start;
2193 	}
2194 	sb = page_address(rdev->sb_page);
2195 	sb->data_size = cpu_to_le64(num_sectors);
2196 	sb->super_offset = cpu_to_le64(rdev->sb_start);
2197 	sb->sb_csum = calc_sb_1_csum(sb);
2198 	do {
2199 		md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
2200 			       rdev->sb_page);
2201 	} while (md_super_wait(rdev->mddev) < 0);
2202 	return num_sectors;
2203 
2204 }
2205 
2206 static int
super_1_allow_new_offset(struct md_rdev * rdev,unsigned long long new_offset)2207 super_1_allow_new_offset(struct md_rdev *rdev,
2208 			 unsigned long long new_offset)
2209 {
2210 	/* All necessary checks on new >= old have been done */
2211 	struct bitmap *bitmap;
2212 	if (new_offset >= rdev->data_offset)
2213 		return 1;
2214 
2215 	/* with 1.0 metadata, there is no metadata to tread on
2216 	 * so we can always move back */
2217 	if (rdev->mddev->minor_version == 0)
2218 		return 1;
2219 
2220 	/* otherwise we must be sure not to step on
2221 	 * any metadata, so stay:
2222 	 * 36K beyond start of superblock
2223 	 * beyond end of badblocks
2224 	 * beyond write-intent bitmap
2225 	 */
2226 	if (rdev->sb_start + (32+4)*2 > new_offset)
2227 		return 0;
2228 	bitmap = rdev->mddev->bitmap;
2229 	if (bitmap && !rdev->mddev->bitmap_info.file &&
2230 	    rdev->sb_start + rdev->mddev->bitmap_info.offset +
2231 	    bitmap->storage.file_pages * (PAGE_SIZE>>9) > new_offset)
2232 		return 0;
2233 	if (rdev->badblocks.sector + rdev->badblocks.size > new_offset)
2234 		return 0;
2235 
2236 	return 1;
2237 }
2238 
2239 static struct super_type super_types[] = {
2240 	[0] = {
2241 		.name	= "0.90.0",
2242 		.owner	= THIS_MODULE,
2243 		.load_super	    = super_90_load,
2244 		.validate_super	    = super_90_validate,
2245 		.sync_super	    = super_90_sync,
2246 		.rdev_size_change   = super_90_rdev_size_change,
2247 		.allow_new_offset   = super_90_allow_new_offset,
2248 	},
2249 	[1] = {
2250 		.name	= "md-1",
2251 		.owner	= THIS_MODULE,
2252 		.load_super	    = super_1_load,
2253 		.validate_super	    = super_1_validate,
2254 		.sync_super	    = super_1_sync,
2255 		.rdev_size_change   = super_1_rdev_size_change,
2256 		.allow_new_offset   = super_1_allow_new_offset,
2257 	},
2258 };
2259 
sync_super(struct mddev * mddev,struct md_rdev * rdev)2260 static void sync_super(struct mddev *mddev, struct md_rdev *rdev)
2261 {
2262 	if (mddev->sync_super) {
2263 		mddev->sync_super(mddev, rdev);
2264 		return;
2265 	}
2266 
2267 	BUG_ON(mddev->major_version >= ARRAY_SIZE(super_types));
2268 
2269 	super_types[mddev->major_version].sync_super(mddev, rdev);
2270 }
2271 
match_mddev_units(struct mddev * mddev1,struct mddev * mddev2)2272 static int match_mddev_units(struct mddev *mddev1, struct mddev *mddev2)
2273 {
2274 	struct md_rdev *rdev, *rdev2;
2275 
2276 	rcu_read_lock();
2277 	rdev_for_each_rcu(rdev, mddev1) {
2278 		if (test_bit(Faulty, &rdev->flags) ||
2279 		    test_bit(Journal, &rdev->flags) ||
2280 		    rdev->raid_disk == -1)
2281 			continue;
2282 		rdev_for_each_rcu(rdev2, mddev2) {
2283 			if (test_bit(Faulty, &rdev2->flags) ||
2284 			    test_bit(Journal, &rdev2->flags) ||
2285 			    rdev2->raid_disk == -1)
2286 				continue;
2287 			if (rdev->bdev->bd_disk == rdev2->bdev->bd_disk) {
2288 				rcu_read_unlock();
2289 				return 1;
2290 			}
2291 		}
2292 	}
2293 	rcu_read_unlock();
2294 	return 0;
2295 }
2296 
2297 static LIST_HEAD(pending_raid_disks);
2298 
2299 /*
2300  * Try to register data integrity profile for an mddev
2301  *
2302  * This is called when an array is started and after a disk has been kicked
2303  * from the array. It only succeeds if all working and active component devices
2304  * are integrity capable with matching profiles.
2305  */
md_integrity_register(struct mddev * mddev)2306 int md_integrity_register(struct mddev *mddev)
2307 {
2308 	struct md_rdev *rdev, *reference = NULL;
2309 
2310 	if (list_empty(&mddev->disks))
2311 		return 0; /* nothing to do */
2312 	if (!mddev->gendisk || blk_get_integrity(mddev->gendisk))
2313 		return 0; /* shouldn't register, or already is */
2314 	rdev_for_each(rdev, mddev) {
2315 		/* skip spares and non-functional disks */
2316 		if (test_bit(Faulty, &rdev->flags))
2317 			continue;
2318 		if (rdev->raid_disk < 0)
2319 			continue;
2320 		if (!reference) {
2321 			/* Use the first rdev as the reference */
2322 			reference = rdev;
2323 			continue;
2324 		}
2325 		/* does this rdev's profile match the reference profile? */
2326 		if (blk_integrity_compare(reference->bdev->bd_disk,
2327 				rdev->bdev->bd_disk) < 0)
2328 			return -EINVAL;
2329 	}
2330 	if (!reference || !bdev_get_integrity(reference->bdev))
2331 		return 0;
2332 	/*
2333 	 * All component devices are integrity capable and have matching
2334 	 * profiles, register the common profile for the md device.
2335 	 */
2336 	blk_integrity_register(mddev->gendisk,
2337 			       bdev_get_integrity(reference->bdev));
2338 
2339 	pr_debug("md: data integrity enabled on %s\n", mdname(mddev));
2340 	if (bioset_integrity_create(&mddev->bio_set, BIO_POOL_SIZE) ||
2341 	    (mddev->level != 1 && mddev->level != 10 &&
2342 	     bioset_integrity_create(&mddev->io_acct_set, BIO_POOL_SIZE))) {
2343 		/*
2344 		 * No need to handle the failure of bioset_integrity_create,
2345 		 * because the function is called by md_run() -> pers->run(),
2346 		 * md_run calls bioset_exit -> bioset_integrity_free in case
2347 		 * of failure case.
2348 		 */
2349 		pr_err("md: failed to create integrity pool for %s\n",
2350 		       mdname(mddev));
2351 		return -EINVAL;
2352 	}
2353 	return 0;
2354 }
2355 EXPORT_SYMBOL(md_integrity_register);
2356 
2357 /*
2358  * Attempt to add an rdev, but only if it is consistent with the current
2359  * integrity profile
2360  */
md_integrity_add_rdev(struct md_rdev * rdev,struct mddev * mddev)2361 int md_integrity_add_rdev(struct md_rdev *rdev, struct mddev *mddev)
2362 {
2363 	struct blk_integrity *bi_mddev;
2364 
2365 	if (!mddev->gendisk)
2366 		return 0;
2367 
2368 	bi_mddev = blk_get_integrity(mddev->gendisk);
2369 
2370 	if (!bi_mddev) /* nothing to do */
2371 		return 0;
2372 
2373 	if (blk_integrity_compare(mddev->gendisk, rdev->bdev->bd_disk) != 0) {
2374 		pr_err("%s: incompatible integrity profile for %pg\n",
2375 		       mdname(mddev), rdev->bdev);
2376 		return -ENXIO;
2377 	}
2378 
2379 	return 0;
2380 }
2381 EXPORT_SYMBOL(md_integrity_add_rdev);
2382 
rdev_read_only(struct md_rdev * rdev)2383 static bool rdev_read_only(struct md_rdev *rdev)
2384 {
2385 	return bdev_read_only(rdev->bdev) ||
2386 		(rdev->meta_bdev && bdev_read_only(rdev->meta_bdev));
2387 }
2388 
bind_rdev_to_array(struct md_rdev * rdev,struct mddev * mddev)2389 static int bind_rdev_to_array(struct md_rdev *rdev, struct mddev *mddev)
2390 {
2391 	char b[BDEVNAME_SIZE];
2392 	int err;
2393 
2394 	/* prevent duplicates */
2395 	if (find_rdev(mddev, rdev->bdev->bd_dev))
2396 		return -EEXIST;
2397 
2398 	if (rdev_read_only(rdev) && mddev->pers)
2399 		return -EROFS;
2400 
2401 	/* make sure rdev->sectors exceeds mddev->dev_sectors */
2402 	if (!test_bit(Journal, &rdev->flags) &&
2403 	    rdev->sectors &&
2404 	    (mddev->dev_sectors == 0 || rdev->sectors < mddev->dev_sectors)) {
2405 		if (mddev->pers) {
2406 			/* Cannot change size, so fail
2407 			 * If mddev->level <= 0, then we don't care
2408 			 * about aligning sizes (e.g. linear)
2409 			 */
2410 			if (mddev->level > 0)
2411 				return -ENOSPC;
2412 		} else
2413 			mddev->dev_sectors = rdev->sectors;
2414 	}
2415 
2416 	/* Verify rdev->desc_nr is unique.
2417 	 * If it is -1, assign a free number, else
2418 	 * check number is not in use
2419 	 */
2420 	rcu_read_lock();
2421 	if (rdev->desc_nr < 0) {
2422 		int choice = 0;
2423 		if (mddev->pers)
2424 			choice = mddev->raid_disks;
2425 		while (md_find_rdev_nr_rcu(mddev, choice))
2426 			choice++;
2427 		rdev->desc_nr = choice;
2428 	} else {
2429 		if (md_find_rdev_nr_rcu(mddev, rdev->desc_nr)) {
2430 			rcu_read_unlock();
2431 			return -EBUSY;
2432 		}
2433 	}
2434 	rcu_read_unlock();
2435 	if (!test_bit(Journal, &rdev->flags) &&
2436 	    mddev->max_disks && rdev->desc_nr >= mddev->max_disks) {
2437 		pr_warn("md: %s: array is limited to %d devices\n",
2438 			mdname(mddev), mddev->max_disks);
2439 		return -EBUSY;
2440 	}
2441 	bdevname(rdev->bdev,b);
2442 	strreplace(b, '/', '!');
2443 
2444 	rdev->mddev = mddev;
2445 	pr_debug("md: bind<%s>\n", b);
2446 
2447 	if (mddev->raid_disks)
2448 		mddev_create_serial_pool(mddev, rdev, false);
2449 
2450 	if ((err = kobject_add(&rdev->kobj, &mddev->kobj, "dev-%s", b)))
2451 		goto fail;
2452 
2453 	/* failure here is OK */
2454 	err = sysfs_create_link(&rdev->kobj, bdev_kobj(rdev->bdev), "block");
2455 	rdev->sysfs_state = sysfs_get_dirent_safe(rdev->kobj.sd, "state");
2456 	rdev->sysfs_unack_badblocks =
2457 		sysfs_get_dirent_safe(rdev->kobj.sd, "unacknowledged_bad_blocks");
2458 	rdev->sysfs_badblocks =
2459 		sysfs_get_dirent_safe(rdev->kobj.sd, "bad_blocks");
2460 
2461 	list_add_rcu(&rdev->same_set, &mddev->disks);
2462 	bd_link_disk_holder(rdev->bdev, mddev->gendisk);
2463 
2464 	/* May as well allow recovery to be retried once */
2465 	mddev->recovery_disabled++;
2466 
2467 	return 0;
2468 
2469  fail:
2470 	pr_warn("md: failed to register dev-%s for %s\n",
2471 		b, mdname(mddev));
2472 	return err;
2473 }
2474 
rdev_delayed_delete(struct work_struct * ws)2475 static void rdev_delayed_delete(struct work_struct *ws)
2476 {
2477 	struct md_rdev *rdev = container_of(ws, struct md_rdev, del_work);
2478 	kobject_del(&rdev->kobj);
2479 	kobject_put(&rdev->kobj);
2480 }
2481 
unbind_rdev_from_array(struct md_rdev * rdev)2482 static void unbind_rdev_from_array(struct md_rdev *rdev)
2483 {
2484 	bd_unlink_disk_holder(rdev->bdev, rdev->mddev->gendisk);
2485 	list_del_rcu(&rdev->same_set);
2486 	pr_debug("md: unbind<%pg>\n", rdev->bdev);
2487 	mddev_destroy_serial_pool(rdev->mddev, rdev, false);
2488 	rdev->mddev = NULL;
2489 	sysfs_remove_link(&rdev->kobj, "block");
2490 	sysfs_put(rdev->sysfs_state);
2491 	sysfs_put(rdev->sysfs_unack_badblocks);
2492 	sysfs_put(rdev->sysfs_badblocks);
2493 	rdev->sysfs_state = NULL;
2494 	rdev->sysfs_unack_badblocks = NULL;
2495 	rdev->sysfs_badblocks = NULL;
2496 	rdev->badblocks.count = 0;
2497 	/* We need to delay this, otherwise we can deadlock when
2498 	 * writing to 'remove' to "dev/state".  We also need
2499 	 * to delay it due to rcu usage.
2500 	 */
2501 	synchronize_rcu();
2502 	INIT_WORK(&rdev->del_work, rdev_delayed_delete);
2503 	kobject_get(&rdev->kobj);
2504 	queue_work(md_rdev_misc_wq, &rdev->del_work);
2505 }
2506 
2507 /*
2508  * prevent the device from being mounted, repartitioned or
2509  * otherwise reused by a RAID array (or any other kernel
2510  * subsystem), by bd_claiming the device.
2511  */
lock_rdev(struct md_rdev * rdev,dev_t dev,int shared)2512 static int lock_rdev(struct md_rdev *rdev, dev_t dev, int shared)
2513 {
2514 	int err = 0;
2515 	struct block_device *bdev;
2516 
2517 	bdev = blkdev_get_by_dev(dev, FMODE_READ|FMODE_WRITE|FMODE_EXCL,
2518 				 shared ? (struct md_rdev *)lock_rdev : rdev);
2519 	if (IS_ERR(bdev)) {
2520 		pr_warn("md: could not open device unknown-block(%u,%u).\n",
2521 			MAJOR(dev), MINOR(dev));
2522 		return PTR_ERR(bdev);
2523 	}
2524 	rdev->bdev = bdev;
2525 	return err;
2526 }
2527 
unlock_rdev(struct md_rdev * rdev)2528 static void unlock_rdev(struct md_rdev *rdev)
2529 {
2530 	struct block_device *bdev = rdev->bdev;
2531 	rdev->bdev = NULL;
2532 	blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
2533 }
2534 
2535 void md_autodetect_dev(dev_t dev);
2536 
export_rdev(struct md_rdev * rdev)2537 static void export_rdev(struct md_rdev *rdev)
2538 {
2539 	pr_debug("md: export_rdev(%pg)\n", rdev->bdev);
2540 	md_rdev_clear(rdev);
2541 #ifndef MODULE
2542 	if (test_bit(AutoDetected, &rdev->flags))
2543 		md_autodetect_dev(rdev->bdev->bd_dev);
2544 #endif
2545 	unlock_rdev(rdev);
2546 	kobject_put(&rdev->kobj);
2547 }
2548 
md_kick_rdev_from_array(struct md_rdev * rdev)2549 void md_kick_rdev_from_array(struct md_rdev *rdev)
2550 {
2551 	unbind_rdev_from_array(rdev);
2552 	export_rdev(rdev);
2553 }
2554 EXPORT_SYMBOL_GPL(md_kick_rdev_from_array);
2555 
export_array(struct mddev * mddev)2556 static void export_array(struct mddev *mddev)
2557 {
2558 	struct md_rdev *rdev;
2559 
2560 	while (!list_empty(&mddev->disks)) {
2561 		rdev = list_first_entry(&mddev->disks, struct md_rdev,
2562 					same_set);
2563 		md_kick_rdev_from_array(rdev);
2564 	}
2565 	mddev->raid_disks = 0;
2566 	mddev->major_version = 0;
2567 }
2568 
set_in_sync(struct mddev * mddev)2569 static bool set_in_sync(struct mddev *mddev)
2570 {
2571 	lockdep_assert_held(&mddev->lock);
2572 	if (!mddev->in_sync) {
2573 		mddev->sync_checkers++;
2574 		spin_unlock(&mddev->lock);
2575 		percpu_ref_switch_to_atomic_sync(&mddev->writes_pending);
2576 		spin_lock(&mddev->lock);
2577 		if (!mddev->in_sync &&
2578 		    percpu_ref_is_zero(&mddev->writes_pending)) {
2579 			mddev->in_sync = 1;
2580 			/*
2581 			 * Ensure ->in_sync is visible before we clear
2582 			 * ->sync_checkers.
2583 			 */
2584 			smp_mb();
2585 			set_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags);
2586 			sysfs_notify_dirent_safe(mddev->sysfs_state);
2587 		}
2588 		if (--mddev->sync_checkers == 0)
2589 			percpu_ref_switch_to_percpu(&mddev->writes_pending);
2590 	}
2591 	if (mddev->safemode == 1)
2592 		mddev->safemode = 0;
2593 	return mddev->in_sync;
2594 }
2595 
sync_sbs(struct mddev * mddev,int nospares)2596 static void sync_sbs(struct mddev *mddev, int nospares)
2597 {
2598 	/* Update each superblock (in-memory image), but
2599 	 * if we are allowed to, skip spares which already
2600 	 * have the right event counter, or have one earlier
2601 	 * (which would mean they aren't being marked as dirty
2602 	 * with the rest of the array)
2603 	 */
2604 	struct md_rdev *rdev;
2605 	rdev_for_each(rdev, mddev) {
2606 		if (rdev->sb_events == mddev->events ||
2607 		    (nospares &&
2608 		     rdev->raid_disk < 0 &&
2609 		     rdev->sb_events+1 == mddev->events)) {
2610 			/* Don't update this superblock */
2611 			rdev->sb_loaded = 2;
2612 		} else {
2613 			sync_super(mddev, rdev);
2614 			rdev->sb_loaded = 1;
2615 		}
2616 	}
2617 }
2618 
does_sb_need_changing(struct mddev * mddev)2619 static bool does_sb_need_changing(struct mddev *mddev)
2620 {
2621 	struct md_rdev *rdev = NULL, *iter;
2622 	struct mdp_superblock_1 *sb;
2623 	int role;
2624 
2625 	/* Find a good rdev */
2626 	rdev_for_each(iter, mddev)
2627 		if ((iter->raid_disk >= 0) && !test_bit(Faulty, &iter->flags)) {
2628 			rdev = iter;
2629 			break;
2630 		}
2631 
2632 	/* No good device found. */
2633 	if (!rdev)
2634 		return false;
2635 
2636 	sb = page_address(rdev->sb_page);
2637 	/* Check if a device has become faulty or a spare become active */
2638 	rdev_for_each(rdev, mddev) {
2639 		role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
2640 		/* Device activated? */
2641 		if (role == MD_DISK_ROLE_SPARE && rdev->raid_disk >= 0 &&
2642 		    !test_bit(Faulty, &rdev->flags))
2643 			return true;
2644 		/* Device turned faulty? */
2645 		if (test_bit(Faulty, &rdev->flags) && (role < MD_DISK_ROLE_MAX))
2646 			return true;
2647 	}
2648 
2649 	/* Check if any mddev parameters have changed */
2650 	if ((mddev->dev_sectors != le64_to_cpu(sb->size)) ||
2651 	    (mddev->reshape_position != le64_to_cpu(sb->reshape_position)) ||
2652 	    (mddev->layout != le32_to_cpu(sb->layout)) ||
2653 	    (mddev->raid_disks != le32_to_cpu(sb->raid_disks)) ||
2654 	    (mddev->chunk_sectors != le32_to_cpu(sb->chunksize)))
2655 		return true;
2656 
2657 	return false;
2658 }
2659 
md_update_sb(struct mddev * mddev,int force_change)2660 void md_update_sb(struct mddev *mddev, int force_change)
2661 {
2662 	struct md_rdev *rdev;
2663 	int sync_req;
2664 	int nospares = 0;
2665 	int any_badblocks_changed = 0;
2666 	int ret = -1;
2667 
2668 	if (mddev->ro) {
2669 		if (force_change)
2670 			set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
2671 		return;
2672 	}
2673 
2674 repeat:
2675 	if (mddev_is_clustered(mddev)) {
2676 		if (test_and_clear_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags))
2677 			force_change = 1;
2678 		if (test_and_clear_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags))
2679 			nospares = 1;
2680 		ret = md_cluster_ops->metadata_update_start(mddev);
2681 		/* Has someone else has updated the sb */
2682 		if (!does_sb_need_changing(mddev)) {
2683 			if (ret == 0)
2684 				md_cluster_ops->metadata_update_cancel(mddev);
2685 			bit_clear_unless(&mddev->sb_flags, BIT(MD_SB_CHANGE_PENDING),
2686 							 BIT(MD_SB_CHANGE_DEVS) |
2687 							 BIT(MD_SB_CHANGE_CLEAN));
2688 			return;
2689 		}
2690 	}
2691 
2692 	/*
2693 	 * First make sure individual recovery_offsets are correct
2694 	 * curr_resync_completed can only be used during recovery.
2695 	 * During reshape/resync it might use array-addresses rather
2696 	 * that device addresses.
2697 	 */
2698 	rdev_for_each(rdev, mddev) {
2699 		if (rdev->raid_disk >= 0 &&
2700 		    mddev->delta_disks >= 0 &&
2701 		    test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
2702 		    test_bit(MD_RECOVERY_RECOVER, &mddev->recovery) &&
2703 		    !test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
2704 		    !test_bit(Journal, &rdev->flags) &&
2705 		    !test_bit(In_sync, &rdev->flags) &&
2706 		    mddev->curr_resync_completed > rdev->recovery_offset)
2707 				rdev->recovery_offset = mddev->curr_resync_completed;
2708 
2709 	}
2710 	if (!mddev->persistent) {
2711 		clear_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags);
2712 		clear_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
2713 		if (!mddev->external) {
2714 			clear_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags);
2715 			rdev_for_each(rdev, mddev) {
2716 				if (rdev->badblocks.changed) {
2717 					rdev->badblocks.changed = 0;
2718 					ack_all_badblocks(&rdev->badblocks);
2719 					md_error(mddev, rdev);
2720 				}
2721 				clear_bit(Blocked, &rdev->flags);
2722 				clear_bit(BlockedBadBlocks, &rdev->flags);
2723 				wake_up(&rdev->blocked_wait);
2724 			}
2725 		}
2726 		wake_up(&mddev->sb_wait);
2727 		return;
2728 	}
2729 
2730 	spin_lock(&mddev->lock);
2731 
2732 	mddev->utime = ktime_get_real_seconds();
2733 
2734 	if (test_and_clear_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags))
2735 		force_change = 1;
2736 	if (test_and_clear_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags))
2737 		/* just a clean<-> dirty transition, possibly leave spares alone,
2738 		 * though if events isn't the right even/odd, we will have to do
2739 		 * spares after all
2740 		 */
2741 		nospares = 1;
2742 	if (force_change)
2743 		nospares = 0;
2744 	if (mddev->degraded)
2745 		/* If the array is degraded, then skipping spares is both
2746 		 * dangerous and fairly pointless.
2747 		 * Dangerous because a device that was removed from the array
2748 		 * might have a event_count that still looks up-to-date,
2749 		 * so it can be re-added without a resync.
2750 		 * Pointless because if there are any spares to skip,
2751 		 * then a recovery will happen and soon that array won't
2752 		 * be degraded any more and the spare can go back to sleep then.
2753 		 */
2754 		nospares = 0;
2755 
2756 	sync_req = mddev->in_sync;
2757 
2758 	/* If this is just a dirty<->clean transition, and the array is clean
2759 	 * and 'events' is odd, we can roll back to the previous clean state */
2760 	if (nospares
2761 	    && (mddev->in_sync && mddev->recovery_cp == MaxSector)
2762 	    && mddev->can_decrease_events
2763 	    && mddev->events != 1) {
2764 		mddev->events--;
2765 		mddev->can_decrease_events = 0;
2766 	} else {
2767 		/* otherwise we have to go forward and ... */
2768 		mddev->events ++;
2769 		mddev->can_decrease_events = nospares;
2770 	}
2771 
2772 	/*
2773 	 * This 64-bit counter should never wrap.
2774 	 * Either we are in around ~1 trillion A.C., assuming
2775 	 * 1 reboot per second, or we have a bug...
2776 	 */
2777 	WARN_ON(mddev->events == 0);
2778 
2779 	rdev_for_each(rdev, mddev) {
2780 		if (rdev->badblocks.changed)
2781 			any_badblocks_changed++;
2782 		if (test_bit(Faulty, &rdev->flags))
2783 			set_bit(FaultRecorded, &rdev->flags);
2784 	}
2785 
2786 	sync_sbs(mddev, nospares);
2787 	spin_unlock(&mddev->lock);
2788 
2789 	pr_debug("md: updating %s RAID superblock on device (in sync %d)\n",
2790 		 mdname(mddev), mddev->in_sync);
2791 
2792 	if (mddev->queue)
2793 		blk_add_trace_msg(mddev->queue, "md md_update_sb");
2794 rewrite:
2795 	md_bitmap_update_sb(mddev->bitmap);
2796 	rdev_for_each(rdev, mddev) {
2797 		if (rdev->sb_loaded != 1)
2798 			continue; /* no noise on spare devices */
2799 
2800 		if (!test_bit(Faulty, &rdev->flags)) {
2801 			md_super_write(mddev,rdev,
2802 				       rdev->sb_start, rdev->sb_size,
2803 				       rdev->sb_page);
2804 			pr_debug("md: (write) %pg's sb offset: %llu\n",
2805 				 rdev->bdev,
2806 				 (unsigned long long)rdev->sb_start);
2807 			rdev->sb_events = mddev->events;
2808 			if (rdev->badblocks.size) {
2809 				md_super_write(mddev, rdev,
2810 					       rdev->badblocks.sector,
2811 					       rdev->badblocks.size << 9,
2812 					       rdev->bb_page);
2813 				rdev->badblocks.size = 0;
2814 			}
2815 
2816 		} else
2817 			pr_debug("md: %pg (skipping faulty)\n",
2818 				 rdev->bdev);
2819 
2820 		if (mddev->level == LEVEL_MULTIPATH)
2821 			/* only need to write one superblock... */
2822 			break;
2823 	}
2824 	if (md_super_wait(mddev) < 0)
2825 		goto rewrite;
2826 	/* if there was a failure, MD_SB_CHANGE_DEVS was set, and we re-write super */
2827 
2828 	if (mddev_is_clustered(mddev) && ret == 0)
2829 		md_cluster_ops->metadata_update_finish(mddev);
2830 
2831 	if (mddev->in_sync != sync_req ||
2832 	    !bit_clear_unless(&mddev->sb_flags, BIT(MD_SB_CHANGE_PENDING),
2833 			       BIT(MD_SB_CHANGE_DEVS) | BIT(MD_SB_CHANGE_CLEAN)))
2834 		/* have to write it out again */
2835 		goto repeat;
2836 	wake_up(&mddev->sb_wait);
2837 	if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
2838 		sysfs_notify_dirent_safe(mddev->sysfs_completed);
2839 
2840 	rdev_for_each(rdev, mddev) {
2841 		if (test_and_clear_bit(FaultRecorded, &rdev->flags))
2842 			clear_bit(Blocked, &rdev->flags);
2843 
2844 		if (any_badblocks_changed)
2845 			ack_all_badblocks(&rdev->badblocks);
2846 		clear_bit(BlockedBadBlocks, &rdev->flags);
2847 		wake_up(&rdev->blocked_wait);
2848 	}
2849 }
2850 EXPORT_SYMBOL(md_update_sb);
2851 
add_bound_rdev(struct md_rdev * rdev)2852 static int add_bound_rdev(struct md_rdev *rdev)
2853 {
2854 	struct mddev *mddev = rdev->mddev;
2855 	int err = 0;
2856 	bool add_journal = test_bit(Journal, &rdev->flags);
2857 
2858 	if (!mddev->pers->hot_remove_disk || add_journal) {
2859 		/* If there is hot_add_disk but no hot_remove_disk
2860 		 * then added disks for geometry changes,
2861 		 * and should be added immediately.
2862 		 */
2863 		super_types[mddev->major_version].
2864 			validate_super(mddev, rdev);
2865 		if (add_journal)
2866 			mddev_suspend(mddev);
2867 		err = mddev->pers->hot_add_disk(mddev, rdev);
2868 		if (add_journal)
2869 			mddev_resume(mddev);
2870 		if (err) {
2871 			md_kick_rdev_from_array(rdev);
2872 			return err;
2873 		}
2874 	}
2875 	sysfs_notify_dirent_safe(rdev->sysfs_state);
2876 
2877 	set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
2878 	if (mddev->degraded)
2879 		set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
2880 	set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2881 	md_new_event();
2882 	md_wakeup_thread(mddev->thread);
2883 	return 0;
2884 }
2885 
2886 /* words written to sysfs files may, or may not, be \n terminated.
2887  * We want to accept with case. For this we use cmd_match.
2888  */
cmd_match(const char * cmd,const char * str)2889 static int cmd_match(const char *cmd, const char *str)
2890 {
2891 	/* See if cmd, written into a sysfs file, matches
2892 	 * str.  They must either be the same, or cmd can
2893 	 * have a trailing newline
2894 	 */
2895 	while (*cmd && *str && *cmd == *str) {
2896 		cmd++;
2897 		str++;
2898 	}
2899 	if (*cmd == '\n')
2900 		cmd++;
2901 	if (*str || *cmd)
2902 		return 0;
2903 	return 1;
2904 }
2905 
2906 struct rdev_sysfs_entry {
2907 	struct attribute attr;
2908 	ssize_t (*show)(struct md_rdev *, char *);
2909 	ssize_t (*store)(struct md_rdev *, const char *, size_t);
2910 };
2911 
2912 static ssize_t
state_show(struct md_rdev * rdev,char * page)2913 state_show(struct md_rdev *rdev, char *page)
2914 {
2915 	char *sep = ",";
2916 	size_t len = 0;
2917 	unsigned long flags = READ_ONCE(rdev->flags);
2918 
2919 	if (test_bit(Faulty, &flags) ||
2920 	    (!test_bit(ExternalBbl, &flags) &&
2921 	    rdev->badblocks.unacked_exist))
2922 		len += sprintf(page+len, "faulty%s", sep);
2923 	if (test_bit(In_sync, &flags))
2924 		len += sprintf(page+len, "in_sync%s", sep);
2925 	if (test_bit(Journal, &flags))
2926 		len += sprintf(page+len, "journal%s", sep);
2927 	if (test_bit(WriteMostly, &flags))
2928 		len += sprintf(page+len, "write_mostly%s", sep);
2929 	if (test_bit(Blocked, &flags) ||
2930 	    (rdev->badblocks.unacked_exist
2931 	     && !test_bit(Faulty, &flags)))
2932 		len += sprintf(page+len, "blocked%s", sep);
2933 	if (!test_bit(Faulty, &flags) &&
2934 	    !test_bit(Journal, &flags) &&
2935 	    !test_bit(In_sync, &flags))
2936 		len += sprintf(page+len, "spare%s", sep);
2937 	if (test_bit(WriteErrorSeen, &flags))
2938 		len += sprintf(page+len, "write_error%s", sep);
2939 	if (test_bit(WantReplacement, &flags))
2940 		len += sprintf(page+len, "want_replacement%s", sep);
2941 	if (test_bit(Replacement, &flags))
2942 		len += sprintf(page+len, "replacement%s", sep);
2943 	if (test_bit(ExternalBbl, &flags))
2944 		len += sprintf(page+len, "external_bbl%s", sep);
2945 	if (test_bit(FailFast, &flags))
2946 		len += sprintf(page+len, "failfast%s", sep);
2947 
2948 	if (len)
2949 		len -= strlen(sep);
2950 
2951 	return len+sprintf(page+len, "\n");
2952 }
2953 
2954 static ssize_t
state_store(struct md_rdev * rdev,const char * buf,size_t len)2955 state_store(struct md_rdev *rdev, const char *buf, size_t len)
2956 {
2957 	/* can write
2958 	 *  faulty  - simulates an error
2959 	 *  remove  - disconnects the device
2960 	 *  writemostly - sets write_mostly
2961 	 *  -writemostly - clears write_mostly
2962 	 *  blocked - sets the Blocked flags
2963 	 *  -blocked - clears the Blocked and possibly simulates an error
2964 	 *  insync - sets Insync providing device isn't active
2965 	 *  -insync - clear Insync for a device with a slot assigned,
2966 	 *            so that it gets rebuilt based on bitmap
2967 	 *  write_error - sets WriteErrorSeen
2968 	 *  -write_error - clears WriteErrorSeen
2969 	 *  {,-}failfast - set/clear FailFast
2970 	 */
2971 
2972 	struct mddev *mddev = rdev->mddev;
2973 	int err = -EINVAL;
2974 	bool need_update_sb = false;
2975 
2976 	if (cmd_match(buf, "faulty") && rdev->mddev->pers) {
2977 		md_error(rdev->mddev, rdev);
2978 
2979 		if (test_bit(MD_BROKEN, &rdev->mddev->flags))
2980 			err = -EBUSY;
2981 		else
2982 			err = 0;
2983 	} else if (cmd_match(buf, "remove")) {
2984 		if (rdev->mddev->pers) {
2985 			clear_bit(Blocked, &rdev->flags);
2986 			remove_and_add_spares(rdev->mddev, rdev);
2987 		}
2988 		if (rdev->raid_disk >= 0)
2989 			err = -EBUSY;
2990 		else {
2991 			err = 0;
2992 			if (mddev_is_clustered(mddev))
2993 				err = md_cluster_ops->remove_disk(mddev, rdev);
2994 
2995 			if (err == 0) {
2996 				md_kick_rdev_from_array(rdev);
2997 				if (mddev->pers) {
2998 					set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
2999 					md_wakeup_thread(mddev->thread);
3000 				}
3001 				md_new_event();
3002 			}
3003 		}
3004 	} else if (cmd_match(buf, "writemostly")) {
3005 		set_bit(WriteMostly, &rdev->flags);
3006 		mddev_create_serial_pool(rdev->mddev, rdev, false);
3007 		need_update_sb = true;
3008 		err = 0;
3009 	} else if (cmd_match(buf, "-writemostly")) {
3010 		mddev_destroy_serial_pool(rdev->mddev, rdev, false);
3011 		clear_bit(WriteMostly, &rdev->flags);
3012 		need_update_sb = true;
3013 		err = 0;
3014 	} else if (cmd_match(buf, "blocked")) {
3015 		set_bit(Blocked, &rdev->flags);
3016 		err = 0;
3017 	} else if (cmd_match(buf, "-blocked")) {
3018 		if (!test_bit(Faulty, &rdev->flags) &&
3019 		    !test_bit(ExternalBbl, &rdev->flags) &&
3020 		    rdev->badblocks.unacked_exist) {
3021 			/* metadata handler doesn't understand badblocks,
3022 			 * so we need to fail the device
3023 			 */
3024 			md_error(rdev->mddev, rdev);
3025 		}
3026 		clear_bit(Blocked, &rdev->flags);
3027 		clear_bit(BlockedBadBlocks, &rdev->flags);
3028 		wake_up(&rdev->blocked_wait);
3029 		set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
3030 		md_wakeup_thread(rdev->mddev->thread);
3031 
3032 		err = 0;
3033 	} else if (cmd_match(buf, "insync") && rdev->raid_disk == -1) {
3034 		set_bit(In_sync, &rdev->flags);
3035 		err = 0;
3036 	} else if (cmd_match(buf, "failfast")) {
3037 		set_bit(FailFast, &rdev->flags);
3038 		need_update_sb = true;
3039 		err = 0;
3040 	} else if (cmd_match(buf, "-failfast")) {
3041 		clear_bit(FailFast, &rdev->flags);
3042 		need_update_sb = true;
3043 		err = 0;
3044 	} else if (cmd_match(buf, "-insync") && rdev->raid_disk >= 0 &&
3045 		   !test_bit(Journal, &rdev->flags)) {
3046 		if (rdev->mddev->pers == NULL) {
3047 			clear_bit(In_sync, &rdev->flags);
3048 			rdev->saved_raid_disk = rdev->raid_disk;
3049 			rdev->raid_disk = -1;
3050 			err = 0;
3051 		}
3052 	} else if (cmd_match(buf, "write_error")) {
3053 		set_bit(WriteErrorSeen, &rdev->flags);
3054 		err = 0;
3055 	} else if (cmd_match(buf, "-write_error")) {
3056 		clear_bit(WriteErrorSeen, &rdev->flags);
3057 		err = 0;
3058 	} else if (cmd_match(buf, "want_replacement")) {
3059 		/* Any non-spare device that is not a replacement can
3060 		 * become want_replacement at any time, but we then need to
3061 		 * check if recovery is needed.
3062 		 */
3063 		if (rdev->raid_disk >= 0 &&
3064 		    !test_bit(Journal, &rdev->flags) &&
3065 		    !test_bit(Replacement, &rdev->flags))
3066 			set_bit(WantReplacement, &rdev->flags);
3067 		set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
3068 		md_wakeup_thread(rdev->mddev->thread);
3069 		err = 0;
3070 	} else if (cmd_match(buf, "-want_replacement")) {
3071 		/* Clearing 'want_replacement' is always allowed.
3072 		 * Once replacements starts it is too late though.
3073 		 */
3074 		err = 0;
3075 		clear_bit(WantReplacement, &rdev->flags);
3076 	} else if (cmd_match(buf, "replacement")) {
3077 		/* Can only set a device as a replacement when array has not
3078 		 * yet been started.  Once running, replacement is automatic
3079 		 * from spares, or by assigning 'slot'.
3080 		 */
3081 		if (rdev->mddev->pers)
3082 			err = -EBUSY;
3083 		else {
3084 			set_bit(Replacement, &rdev->flags);
3085 			err = 0;
3086 		}
3087 	} else if (cmd_match(buf, "-replacement")) {
3088 		/* Similarly, can only clear Replacement before start */
3089 		if (rdev->mddev->pers)
3090 			err = -EBUSY;
3091 		else {
3092 			clear_bit(Replacement, &rdev->flags);
3093 			err = 0;
3094 		}
3095 	} else if (cmd_match(buf, "re-add")) {
3096 		if (!rdev->mddev->pers)
3097 			err = -EINVAL;
3098 		else if (test_bit(Faulty, &rdev->flags) && (rdev->raid_disk == -1) &&
3099 				rdev->saved_raid_disk >= 0) {
3100 			/* clear_bit is performed _after_ all the devices
3101 			 * have their local Faulty bit cleared. If any writes
3102 			 * happen in the meantime in the local node, they
3103 			 * will land in the local bitmap, which will be synced
3104 			 * by this node eventually
3105 			 */
3106 			if (!mddev_is_clustered(rdev->mddev) ||
3107 			    (err = md_cluster_ops->gather_bitmaps(rdev)) == 0) {
3108 				clear_bit(Faulty, &rdev->flags);
3109 				err = add_bound_rdev(rdev);
3110 			}
3111 		} else
3112 			err = -EBUSY;
3113 	} else if (cmd_match(buf, "external_bbl") && (rdev->mddev->external)) {
3114 		set_bit(ExternalBbl, &rdev->flags);
3115 		rdev->badblocks.shift = 0;
3116 		err = 0;
3117 	} else if (cmd_match(buf, "-external_bbl") && (rdev->mddev->external)) {
3118 		clear_bit(ExternalBbl, &rdev->flags);
3119 		err = 0;
3120 	}
3121 	if (need_update_sb)
3122 		md_update_sb(mddev, 1);
3123 	if (!err)
3124 		sysfs_notify_dirent_safe(rdev->sysfs_state);
3125 	return err ? err : len;
3126 }
3127 static struct rdev_sysfs_entry rdev_state =
3128 __ATTR_PREALLOC(state, S_IRUGO|S_IWUSR, state_show, state_store);
3129 
3130 static ssize_t
errors_show(struct md_rdev * rdev,char * page)3131 errors_show(struct md_rdev *rdev, char *page)
3132 {
3133 	return sprintf(page, "%d\n", atomic_read(&rdev->corrected_errors));
3134 }
3135 
3136 static ssize_t
errors_store(struct md_rdev * rdev,const char * buf,size_t len)3137 errors_store(struct md_rdev *rdev, const char *buf, size_t len)
3138 {
3139 	unsigned int n;
3140 	int rv;
3141 
3142 	rv = kstrtouint(buf, 10, &n);
3143 	if (rv < 0)
3144 		return rv;
3145 	atomic_set(&rdev->corrected_errors, n);
3146 	return len;
3147 }
3148 static struct rdev_sysfs_entry rdev_errors =
3149 __ATTR(errors, S_IRUGO|S_IWUSR, errors_show, errors_store);
3150 
3151 static ssize_t
slot_show(struct md_rdev * rdev,char * page)3152 slot_show(struct md_rdev *rdev, char *page)
3153 {
3154 	if (test_bit(Journal, &rdev->flags))
3155 		return sprintf(page, "journal\n");
3156 	else if (rdev->raid_disk < 0)
3157 		return sprintf(page, "none\n");
3158 	else
3159 		return sprintf(page, "%d\n", rdev->raid_disk);
3160 }
3161 
3162 static ssize_t
slot_store(struct md_rdev * rdev,const char * buf,size_t len)3163 slot_store(struct md_rdev *rdev, const char *buf, size_t len)
3164 {
3165 	int slot;
3166 	int err;
3167 
3168 	if (test_bit(Journal, &rdev->flags))
3169 		return -EBUSY;
3170 	if (strncmp(buf, "none", 4)==0)
3171 		slot = -1;
3172 	else {
3173 		err = kstrtouint(buf, 10, (unsigned int *)&slot);
3174 		if (err < 0)
3175 			return err;
3176 	}
3177 	if (rdev->mddev->pers && slot == -1) {
3178 		/* Setting 'slot' on an active array requires also
3179 		 * updating the 'rd%d' link, and communicating
3180 		 * with the personality with ->hot_*_disk.
3181 		 * For now we only support removing
3182 		 * failed/spare devices.  This normally happens automatically,
3183 		 * but not when the metadata is externally managed.
3184 		 */
3185 		if (rdev->raid_disk == -1)
3186 			return -EEXIST;
3187 		/* personality does all needed checks */
3188 		if (rdev->mddev->pers->hot_remove_disk == NULL)
3189 			return -EINVAL;
3190 		clear_bit(Blocked, &rdev->flags);
3191 		remove_and_add_spares(rdev->mddev, rdev);
3192 		if (rdev->raid_disk >= 0)
3193 			return -EBUSY;
3194 		set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
3195 		md_wakeup_thread(rdev->mddev->thread);
3196 	} else if (rdev->mddev->pers) {
3197 		/* Activating a spare .. or possibly reactivating
3198 		 * if we ever get bitmaps working here.
3199 		 */
3200 		int err;
3201 
3202 		if (rdev->raid_disk != -1)
3203 			return -EBUSY;
3204 
3205 		if (test_bit(MD_RECOVERY_RUNNING, &rdev->mddev->recovery))
3206 			return -EBUSY;
3207 
3208 		if (rdev->mddev->pers->hot_add_disk == NULL)
3209 			return -EINVAL;
3210 
3211 		if (slot >= rdev->mddev->raid_disks &&
3212 		    slot >= rdev->mddev->raid_disks + rdev->mddev->delta_disks)
3213 			return -ENOSPC;
3214 
3215 		rdev->raid_disk = slot;
3216 		if (test_bit(In_sync, &rdev->flags))
3217 			rdev->saved_raid_disk = slot;
3218 		else
3219 			rdev->saved_raid_disk = -1;
3220 		clear_bit(In_sync, &rdev->flags);
3221 		clear_bit(Bitmap_sync, &rdev->flags);
3222 		err = rdev->mddev->pers->hot_add_disk(rdev->mddev, rdev);
3223 		if (err) {
3224 			rdev->raid_disk = -1;
3225 			return err;
3226 		} else
3227 			sysfs_notify_dirent_safe(rdev->sysfs_state);
3228 		/* failure here is OK */;
3229 		sysfs_link_rdev(rdev->mddev, rdev);
3230 		/* don't wakeup anyone, leave that to userspace. */
3231 	} else {
3232 		if (slot >= rdev->mddev->raid_disks &&
3233 		    slot >= rdev->mddev->raid_disks + rdev->mddev->delta_disks)
3234 			return -ENOSPC;
3235 		rdev->raid_disk = slot;
3236 		/* assume it is working */
3237 		clear_bit(Faulty, &rdev->flags);
3238 		clear_bit(WriteMostly, &rdev->flags);
3239 		set_bit(In_sync, &rdev->flags);
3240 		sysfs_notify_dirent_safe(rdev->sysfs_state);
3241 	}
3242 	return len;
3243 }
3244 
3245 static struct rdev_sysfs_entry rdev_slot =
3246 __ATTR(slot, S_IRUGO|S_IWUSR, slot_show, slot_store);
3247 
3248 static ssize_t
offset_show(struct md_rdev * rdev,char * page)3249 offset_show(struct md_rdev *rdev, char *page)
3250 {
3251 	return sprintf(page, "%llu\n", (unsigned long long)rdev->data_offset);
3252 }
3253 
3254 static ssize_t
offset_store(struct md_rdev * rdev,const char * buf,size_t len)3255 offset_store(struct md_rdev *rdev, const char *buf, size_t len)
3256 {
3257 	unsigned long long offset;
3258 	if (kstrtoull(buf, 10, &offset) < 0)
3259 		return -EINVAL;
3260 	if (rdev->mddev->pers && rdev->raid_disk >= 0)
3261 		return -EBUSY;
3262 	if (rdev->sectors && rdev->mddev->external)
3263 		/* Must set offset before size, so overlap checks
3264 		 * can be sane */
3265 		return -EBUSY;
3266 	rdev->data_offset = offset;
3267 	rdev->new_data_offset = offset;
3268 	return len;
3269 }
3270 
3271 static struct rdev_sysfs_entry rdev_offset =
3272 __ATTR(offset, S_IRUGO|S_IWUSR, offset_show, offset_store);
3273 
new_offset_show(struct md_rdev * rdev,char * page)3274 static ssize_t new_offset_show(struct md_rdev *rdev, char *page)
3275 {
3276 	return sprintf(page, "%llu\n",
3277 		       (unsigned long long)rdev->new_data_offset);
3278 }
3279 
new_offset_store(struct md_rdev * rdev,const char * buf,size_t len)3280 static ssize_t new_offset_store(struct md_rdev *rdev,
3281 				const char *buf, size_t len)
3282 {
3283 	unsigned long long new_offset;
3284 	struct mddev *mddev = rdev->mddev;
3285 
3286 	if (kstrtoull(buf, 10, &new_offset) < 0)
3287 		return -EINVAL;
3288 
3289 	if (mddev->sync_thread ||
3290 	    test_bit(MD_RECOVERY_RUNNING,&mddev->recovery))
3291 		return -EBUSY;
3292 	if (new_offset == rdev->data_offset)
3293 		/* reset is always permitted */
3294 		;
3295 	else if (new_offset > rdev->data_offset) {
3296 		/* must not push array size beyond rdev_sectors */
3297 		if (new_offset - rdev->data_offset
3298 		    + mddev->dev_sectors > rdev->sectors)
3299 				return -E2BIG;
3300 	}
3301 	/* Metadata worries about other space details. */
3302 
3303 	/* decreasing the offset is inconsistent with a backwards
3304 	 * reshape.
3305 	 */
3306 	if (new_offset < rdev->data_offset &&
3307 	    mddev->reshape_backwards)
3308 		return -EINVAL;
3309 	/* Increasing offset is inconsistent with forwards
3310 	 * reshape.  reshape_direction should be set to
3311 	 * 'backwards' first.
3312 	 */
3313 	if (new_offset > rdev->data_offset &&
3314 	    !mddev->reshape_backwards)
3315 		return -EINVAL;
3316 
3317 	if (mddev->pers && mddev->persistent &&
3318 	    !super_types[mddev->major_version]
3319 	    .allow_new_offset(rdev, new_offset))
3320 		return -E2BIG;
3321 	rdev->new_data_offset = new_offset;
3322 	if (new_offset > rdev->data_offset)
3323 		mddev->reshape_backwards = 1;
3324 	else if (new_offset < rdev->data_offset)
3325 		mddev->reshape_backwards = 0;
3326 
3327 	return len;
3328 }
3329 static struct rdev_sysfs_entry rdev_new_offset =
3330 __ATTR(new_offset, S_IRUGO|S_IWUSR, new_offset_show, new_offset_store);
3331 
3332 static ssize_t
rdev_size_show(struct md_rdev * rdev,char * page)3333 rdev_size_show(struct md_rdev *rdev, char *page)
3334 {
3335 	return sprintf(page, "%llu\n", (unsigned long long)rdev->sectors / 2);
3336 }
3337 
overlaps(sector_t s1,sector_t l1,sector_t s2,sector_t l2)3338 static int overlaps(sector_t s1, sector_t l1, sector_t s2, sector_t l2)
3339 {
3340 	/* check if two start/length pairs overlap */
3341 	if (s1+l1 <= s2)
3342 		return 0;
3343 	if (s2+l2 <= s1)
3344 		return 0;
3345 	return 1;
3346 }
3347 
strict_blocks_to_sectors(const char * buf,sector_t * sectors)3348 static int strict_blocks_to_sectors(const char *buf, sector_t *sectors)
3349 {
3350 	unsigned long long blocks;
3351 	sector_t new;
3352 
3353 	if (kstrtoull(buf, 10, &blocks) < 0)
3354 		return -EINVAL;
3355 
3356 	if (blocks & 1ULL << (8 * sizeof(blocks) - 1))
3357 		return -EINVAL; /* sector conversion overflow */
3358 
3359 	new = blocks * 2;
3360 	if (new != blocks * 2)
3361 		return -EINVAL; /* unsigned long long to sector_t overflow */
3362 
3363 	*sectors = new;
3364 	return 0;
3365 }
3366 
3367 static ssize_t
rdev_size_store(struct md_rdev * rdev,const char * buf,size_t len)3368 rdev_size_store(struct md_rdev *rdev, const char *buf, size_t len)
3369 {
3370 	struct mddev *my_mddev = rdev->mddev;
3371 	sector_t oldsectors = rdev->sectors;
3372 	sector_t sectors;
3373 
3374 	if (test_bit(Journal, &rdev->flags))
3375 		return -EBUSY;
3376 	if (strict_blocks_to_sectors(buf, &sectors) < 0)
3377 		return -EINVAL;
3378 	if (rdev->data_offset != rdev->new_data_offset)
3379 		return -EINVAL; /* too confusing */
3380 	if (my_mddev->pers && rdev->raid_disk >= 0) {
3381 		if (my_mddev->persistent) {
3382 			sectors = super_types[my_mddev->major_version].
3383 				rdev_size_change(rdev, sectors);
3384 			if (!sectors)
3385 				return -EBUSY;
3386 		} else if (!sectors)
3387 			sectors = bdev_nr_sectors(rdev->bdev) -
3388 				rdev->data_offset;
3389 		if (!my_mddev->pers->resize)
3390 			/* Cannot change size for RAID0 or Linear etc */
3391 			return -EINVAL;
3392 	}
3393 	if (sectors < my_mddev->dev_sectors)
3394 		return -EINVAL; /* component must fit device */
3395 
3396 	rdev->sectors = sectors;
3397 	if (sectors > oldsectors && my_mddev->external) {
3398 		/* Need to check that all other rdevs with the same
3399 		 * ->bdev do not overlap.  'rcu' is sufficient to walk
3400 		 * the rdev lists safely.
3401 		 * This check does not provide a hard guarantee, it
3402 		 * just helps avoid dangerous mistakes.
3403 		 */
3404 		struct mddev *mddev;
3405 		int overlap = 0;
3406 		struct list_head *tmp;
3407 
3408 		rcu_read_lock();
3409 		for_each_mddev(mddev, tmp) {
3410 			struct md_rdev *rdev2;
3411 
3412 			rdev_for_each(rdev2, mddev)
3413 				if (rdev->bdev == rdev2->bdev &&
3414 				    rdev != rdev2 &&
3415 				    overlaps(rdev->data_offset, rdev->sectors,
3416 					     rdev2->data_offset,
3417 					     rdev2->sectors)) {
3418 					overlap = 1;
3419 					break;
3420 				}
3421 			if (overlap) {
3422 				mddev_put(mddev);
3423 				break;
3424 			}
3425 		}
3426 		rcu_read_unlock();
3427 		if (overlap) {
3428 			/* Someone else could have slipped in a size
3429 			 * change here, but doing so is just silly.
3430 			 * We put oldsectors back because we *know* it is
3431 			 * safe, and trust userspace not to race with
3432 			 * itself
3433 			 */
3434 			rdev->sectors = oldsectors;
3435 			return -EBUSY;
3436 		}
3437 	}
3438 	return len;
3439 }
3440 
3441 static struct rdev_sysfs_entry rdev_size =
3442 __ATTR(size, S_IRUGO|S_IWUSR, rdev_size_show, rdev_size_store);
3443 
recovery_start_show(struct md_rdev * rdev,char * page)3444 static ssize_t recovery_start_show(struct md_rdev *rdev, char *page)
3445 {
3446 	unsigned long long recovery_start = rdev->recovery_offset;
3447 
3448 	if (test_bit(In_sync, &rdev->flags) ||
3449 	    recovery_start == MaxSector)
3450 		return sprintf(page, "none\n");
3451 
3452 	return sprintf(page, "%llu\n", recovery_start);
3453 }
3454 
recovery_start_store(struct md_rdev * rdev,const char * buf,size_t len)3455 static ssize_t recovery_start_store(struct md_rdev *rdev, const char *buf, size_t len)
3456 {
3457 	unsigned long long recovery_start;
3458 
3459 	if (cmd_match(buf, "none"))
3460 		recovery_start = MaxSector;
3461 	else if (kstrtoull(buf, 10, &recovery_start))
3462 		return -EINVAL;
3463 
3464 	if (rdev->mddev->pers &&
3465 	    rdev->raid_disk >= 0)
3466 		return -EBUSY;
3467 
3468 	rdev->recovery_offset = recovery_start;
3469 	if (recovery_start == MaxSector)
3470 		set_bit(In_sync, &rdev->flags);
3471 	else
3472 		clear_bit(In_sync, &rdev->flags);
3473 	return len;
3474 }
3475 
3476 static struct rdev_sysfs_entry rdev_recovery_start =
3477 __ATTR(recovery_start, S_IRUGO|S_IWUSR, recovery_start_show, recovery_start_store);
3478 
3479 /* sysfs access to bad-blocks list.
3480  * We present two files.
3481  * 'bad-blocks' lists sector numbers and lengths of ranges that
3482  *    are recorded as bad.  The list is truncated to fit within
3483  *    the one-page limit of sysfs.
3484  *    Writing "sector length" to this file adds an acknowledged
3485  *    bad block list.
3486  * 'unacknowledged-bad-blocks' lists bad blocks that have not yet
3487  *    been acknowledged.  Writing to this file adds bad blocks
3488  *    without acknowledging them.  This is largely for testing.
3489  */
bb_show(struct md_rdev * rdev,char * page)3490 static ssize_t bb_show(struct md_rdev *rdev, char *page)
3491 {
3492 	return badblocks_show(&rdev->badblocks, page, 0);
3493 }
bb_store(struct md_rdev * rdev,const char * page,size_t len)3494 static ssize_t bb_store(struct md_rdev *rdev, const char *page, size_t len)
3495 {
3496 	int rv = badblocks_store(&rdev->badblocks, page, len, 0);
3497 	/* Maybe that ack was all we needed */
3498 	if (test_and_clear_bit(BlockedBadBlocks, &rdev->flags))
3499 		wake_up(&rdev->blocked_wait);
3500 	return rv;
3501 }
3502 static struct rdev_sysfs_entry rdev_bad_blocks =
3503 __ATTR(bad_blocks, S_IRUGO|S_IWUSR, bb_show, bb_store);
3504 
ubb_show(struct md_rdev * rdev,char * page)3505 static ssize_t ubb_show(struct md_rdev *rdev, char *page)
3506 {
3507 	return badblocks_show(&rdev->badblocks, page, 1);
3508 }
ubb_store(struct md_rdev * rdev,const char * page,size_t len)3509 static ssize_t ubb_store(struct md_rdev *rdev, const char *page, size_t len)
3510 {
3511 	return badblocks_store(&rdev->badblocks, page, len, 1);
3512 }
3513 static struct rdev_sysfs_entry rdev_unack_bad_blocks =
3514 __ATTR(unacknowledged_bad_blocks, S_IRUGO|S_IWUSR, ubb_show, ubb_store);
3515 
3516 static ssize_t
ppl_sector_show(struct md_rdev * rdev,char * page)3517 ppl_sector_show(struct md_rdev *rdev, char *page)
3518 {
3519 	return sprintf(page, "%llu\n", (unsigned long long)rdev->ppl.sector);
3520 }
3521 
3522 static ssize_t
ppl_sector_store(struct md_rdev * rdev,const char * buf,size_t len)3523 ppl_sector_store(struct md_rdev *rdev, const char *buf, size_t len)
3524 {
3525 	unsigned long long sector;
3526 
3527 	if (kstrtoull(buf, 10, &sector) < 0)
3528 		return -EINVAL;
3529 	if (sector != (sector_t)sector)
3530 		return -EINVAL;
3531 
3532 	if (rdev->mddev->pers && test_bit(MD_HAS_PPL, &rdev->mddev->flags) &&
3533 	    rdev->raid_disk >= 0)
3534 		return -EBUSY;
3535 
3536 	if (rdev->mddev->persistent) {
3537 		if (rdev->mddev->major_version == 0)
3538 			return -EINVAL;
3539 		if ((sector > rdev->sb_start &&
3540 		     sector - rdev->sb_start > S16_MAX) ||
3541 		    (sector < rdev->sb_start &&
3542 		     rdev->sb_start - sector > -S16_MIN))
3543 			return -EINVAL;
3544 		rdev->ppl.offset = sector - rdev->sb_start;
3545 	} else if (!rdev->mddev->external) {
3546 		return -EBUSY;
3547 	}
3548 	rdev->ppl.sector = sector;
3549 	return len;
3550 }
3551 
3552 static struct rdev_sysfs_entry rdev_ppl_sector =
3553 __ATTR(ppl_sector, S_IRUGO|S_IWUSR, ppl_sector_show, ppl_sector_store);
3554 
3555 static ssize_t
ppl_size_show(struct md_rdev * rdev,char * page)3556 ppl_size_show(struct md_rdev *rdev, char *page)
3557 {
3558 	return sprintf(page, "%u\n", rdev->ppl.size);
3559 }
3560 
3561 static ssize_t
ppl_size_store(struct md_rdev * rdev,const char * buf,size_t len)3562 ppl_size_store(struct md_rdev *rdev, const char *buf, size_t len)
3563 {
3564 	unsigned int size;
3565 
3566 	if (kstrtouint(buf, 10, &size) < 0)
3567 		return -EINVAL;
3568 
3569 	if (rdev->mddev->pers && test_bit(MD_HAS_PPL, &rdev->mddev->flags) &&
3570 	    rdev->raid_disk >= 0)
3571 		return -EBUSY;
3572 
3573 	if (rdev->mddev->persistent) {
3574 		if (rdev->mddev->major_version == 0)
3575 			return -EINVAL;
3576 		if (size > U16_MAX)
3577 			return -EINVAL;
3578 	} else if (!rdev->mddev->external) {
3579 		return -EBUSY;
3580 	}
3581 	rdev->ppl.size = size;
3582 	return len;
3583 }
3584 
3585 static struct rdev_sysfs_entry rdev_ppl_size =
3586 __ATTR(ppl_size, S_IRUGO|S_IWUSR, ppl_size_show, ppl_size_store);
3587 
3588 static struct attribute *rdev_default_attrs[] = {
3589 	&rdev_state.attr,
3590 	&rdev_errors.attr,
3591 	&rdev_slot.attr,
3592 	&rdev_offset.attr,
3593 	&rdev_new_offset.attr,
3594 	&rdev_size.attr,
3595 	&rdev_recovery_start.attr,
3596 	&rdev_bad_blocks.attr,
3597 	&rdev_unack_bad_blocks.attr,
3598 	&rdev_ppl_sector.attr,
3599 	&rdev_ppl_size.attr,
3600 	NULL,
3601 };
3602 ATTRIBUTE_GROUPS(rdev_default);
3603 static ssize_t
rdev_attr_show(struct kobject * kobj,struct attribute * attr,char * page)3604 rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
3605 {
3606 	struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
3607 	struct md_rdev *rdev = container_of(kobj, struct md_rdev, kobj);
3608 
3609 	if (!entry->show)
3610 		return -EIO;
3611 	if (!rdev->mddev)
3612 		return -ENODEV;
3613 	return entry->show(rdev, page);
3614 }
3615 
3616 static ssize_t
rdev_attr_store(struct kobject * kobj,struct attribute * attr,const char * page,size_t length)3617 rdev_attr_store(struct kobject *kobj, struct attribute *attr,
3618 	      const char *page, size_t length)
3619 {
3620 	struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
3621 	struct md_rdev *rdev = container_of(kobj, struct md_rdev, kobj);
3622 	ssize_t rv;
3623 	struct mddev *mddev = rdev->mddev;
3624 
3625 	if (!entry->store)
3626 		return -EIO;
3627 	if (!capable(CAP_SYS_ADMIN))
3628 		return -EACCES;
3629 	rv = mddev ? mddev_lock(mddev) : -ENODEV;
3630 	if (!rv) {
3631 		if (rdev->mddev == NULL)
3632 			rv = -ENODEV;
3633 		else
3634 			rv = entry->store(rdev, page, length);
3635 		mddev_unlock(mddev);
3636 	}
3637 	return rv;
3638 }
3639 
rdev_free(struct kobject * ko)3640 static void rdev_free(struct kobject *ko)
3641 {
3642 	struct md_rdev *rdev = container_of(ko, struct md_rdev, kobj);
3643 	kfree(rdev);
3644 }
3645 static const struct sysfs_ops rdev_sysfs_ops = {
3646 	.show		= rdev_attr_show,
3647 	.store		= rdev_attr_store,
3648 };
3649 static struct kobj_type rdev_ktype = {
3650 	.release	= rdev_free,
3651 	.sysfs_ops	= &rdev_sysfs_ops,
3652 	.default_groups	= rdev_default_groups,
3653 };
3654 
md_rdev_init(struct md_rdev * rdev)3655 int md_rdev_init(struct md_rdev *rdev)
3656 {
3657 	rdev->desc_nr = -1;
3658 	rdev->saved_raid_disk = -1;
3659 	rdev->raid_disk = -1;
3660 	rdev->flags = 0;
3661 	rdev->data_offset = 0;
3662 	rdev->new_data_offset = 0;
3663 	rdev->sb_events = 0;
3664 	rdev->last_read_error = 0;
3665 	rdev->sb_loaded = 0;
3666 	rdev->bb_page = NULL;
3667 	atomic_set(&rdev->nr_pending, 0);
3668 	atomic_set(&rdev->read_errors, 0);
3669 	atomic_set(&rdev->corrected_errors, 0);
3670 
3671 	INIT_LIST_HEAD(&rdev->same_set);
3672 	init_waitqueue_head(&rdev->blocked_wait);
3673 
3674 	/* Add space to store bad block list.
3675 	 * This reserves the space even on arrays where it cannot
3676 	 * be used - I wonder if that matters
3677 	 */
3678 	return badblocks_init(&rdev->badblocks, 0);
3679 }
3680 EXPORT_SYMBOL_GPL(md_rdev_init);
3681 /*
3682  * Import a device. If 'super_format' >= 0, then sanity check the superblock
3683  *
3684  * mark the device faulty if:
3685  *
3686  *   - the device is nonexistent (zero size)
3687  *   - the device has no valid superblock
3688  *
3689  * a faulty rdev _never_ has rdev->sb set.
3690  */
md_import_device(dev_t newdev,int super_format,int super_minor)3691 static struct md_rdev *md_import_device(dev_t newdev, int super_format, int super_minor)
3692 {
3693 	int err;
3694 	struct md_rdev *rdev;
3695 	sector_t size;
3696 
3697 	rdev = kzalloc(sizeof(*rdev), GFP_KERNEL);
3698 	if (!rdev)
3699 		return ERR_PTR(-ENOMEM);
3700 
3701 	err = md_rdev_init(rdev);
3702 	if (err)
3703 		goto abort_free;
3704 	err = alloc_disk_sb(rdev);
3705 	if (err)
3706 		goto abort_free;
3707 
3708 	err = lock_rdev(rdev, newdev, super_format == -2);
3709 	if (err)
3710 		goto abort_free;
3711 
3712 	kobject_init(&rdev->kobj, &rdev_ktype);
3713 
3714 	size = bdev_nr_bytes(rdev->bdev) >> BLOCK_SIZE_BITS;
3715 	if (!size) {
3716 		pr_warn("md: %pg has zero or unknown size, marking faulty!\n",
3717 			rdev->bdev);
3718 		err = -EINVAL;
3719 		goto abort_free;
3720 	}
3721 
3722 	if (super_format >= 0) {
3723 		err = super_types[super_format].
3724 			load_super(rdev, NULL, super_minor);
3725 		if (err == -EINVAL) {
3726 			pr_warn("md: %pg does not have a valid v%d.%d superblock, not importing!\n",
3727 				rdev->bdev,
3728 				super_format, super_minor);
3729 			goto abort_free;
3730 		}
3731 		if (err < 0) {
3732 			pr_warn("md: could not read %pg's sb, not importing!\n",
3733 				rdev->bdev);
3734 			goto abort_free;
3735 		}
3736 	}
3737 
3738 	return rdev;
3739 
3740 abort_free:
3741 	if (rdev->bdev)
3742 		unlock_rdev(rdev);
3743 	md_rdev_clear(rdev);
3744 	kfree(rdev);
3745 	return ERR_PTR(err);
3746 }
3747 
3748 /*
3749  * Check a full RAID array for plausibility
3750  */
3751 
analyze_sbs(struct mddev * mddev)3752 static int analyze_sbs(struct mddev *mddev)
3753 {
3754 	int i;
3755 	struct md_rdev *rdev, *freshest, *tmp;
3756 
3757 	freshest = NULL;
3758 	rdev_for_each_safe(rdev, tmp, mddev)
3759 		switch (super_types[mddev->major_version].
3760 			load_super(rdev, freshest, mddev->minor_version)) {
3761 		case 1:
3762 			freshest = rdev;
3763 			break;
3764 		case 0:
3765 			break;
3766 		default:
3767 			pr_warn("md: fatal superblock inconsistency in %pg -- removing from array\n",
3768 				rdev->bdev);
3769 			md_kick_rdev_from_array(rdev);
3770 		}
3771 
3772 	/* Cannot find a valid fresh disk */
3773 	if (!freshest) {
3774 		pr_warn("md: cannot find a valid disk\n");
3775 		return -EINVAL;
3776 	}
3777 
3778 	super_types[mddev->major_version].
3779 		validate_super(mddev, freshest);
3780 
3781 	i = 0;
3782 	rdev_for_each_safe(rdev, tmp, mddev) {
3783 		if (mddev->max_disks &&
3784 		    (rdev->desc_nr >= mddev->max_disks ||
3785 		     i > mddev->max_disks)) {
3786 			pr_warn("md: %s: %pg: only %d devices permitted\n",
3787 				mdname(mddev), rdev->bdev,
3788 				mddev->max_disks);
3789 			md_kick_rdev_from_array(rdev);
3790 			continue;
3791 		}
3792 		if (rdev != freshest) {
3793 			if (super_types[mddev->major_version].
3794 			    validate_super(mddev, rdev)) {
3795 				pr_warn("md: kicking non-fresh %pg from array!\n",
3796 					rdev->bdev);
3797 				md_kick_rdev_from_array(rdev);
3798 				continue;
3799 			}
3800 		}
3801 		if (mddev->level == LEVEL_MULTIPATH) {
3802 			rdev->desc_nr = i++;
3803 			rdev->raid_disk = rdev->desc_nr;
3804 			set_bit(In_sync, &rdev->flags);
3805 		} else if (rdev->raid_disk >=
3806 			    (mddev->raid_disks - min(0, mddev->delta_disks)) &&
3807 			   !test_bit(Journal, &rdev->flags)) {
3808 			rdev->raid_disk = -1;
3809 			clear_bit(In_sync, &rdev->flags);
3810 		}
3811 	}
3812 
3813 	return 0;
3814 }
3815 
3816 /* Read a fixed-point number.
3817  * Numbers in sysfs attributes should be in "standard" units where
3818  * possible, so time should be in seconds.
3819  * However we internally use a a much smaller unit such as
3820  * milliseconds or jiffies.
3821  * This function takes a decimal number with a possible fractional
3822  * component, and produces an integer which is the result of
3823  * multiplying that number by 10^'scale'.
3824  * all without any floating-point arithmetic.
3825  */
strict_strtoul_scaled(const char * cp,unsigned long * res,int scale)3826 int strict_strtoul_scaled(const char *cp, unsigned long *res, int scale)
3827 {
3828 	unsigned long result = 0;
3829 	long decimals = -1;
3830 	while (isdigit(*cp) || (*cp == '.' && decimals < 0)) {
3831 		if (*cp == '.')
3832 			decimals = 0;
3833 		else if (decimals < scale) {
3834 			unsigned int value;
3835 			value = *cp - '0';
3836 			result = result * 10 + value;
3837 			if (decimals >= 0)
3838 				decimals++;
3839 		}
3840 		cp++;
3841 	}
3842 	if (*cp == '\n')
3843 		cp++;
3844 	if (*cp)
3845 		return -EINVAL;
3846 	if (decimals < 0)
3847 		decimals = 0;
3848 	*res = result * int_pow(10, scale - decimals);
3849 	return 0;
3850 }
3851 
3852 static ssize_t
safe_delay_show(struct mddev * mddev,char * page)3853 safe_delay_show(struct mddev *mddev, char *page)
3854 {
3855 	int msec = (mddev->safemode_delay*1000)/HZ;
3856 	return sprintf(page, "%d.%03d\n", msec/1000, msec%1000);
3857 }
3858 static ssize_t
safe_delay_store(struct mddev * mddev,const char * cbuf,size_t len)3859 safe_delay_store(struct mddev *mddev, const char *cbuf, size_t len)
3860 {
3861 	unsigned long msec;
3862 
3863 	if (mddev_is_clustered(mddev)) {
3864 		pr_warn("md: Safemode is disabled for clustered mode\n");
3865 		return -EINVAL;
3866 	}
3867 
3868 	if (strict_strtoul_scaled(cbuf, &msec, 3) < 0)
3869 		return -EINVAL;
3870 	if (msec == 0)
3871 		mddev->safemode_delay = 0;
3872 	else {
3873 		unsigned long old_delay = mddev->safemode_delay;
3874 		unsigned long new_delay = (msec*HZ)/1000;
3875 
3876 		if (new_delay == 0)
3877 			new_delay = 1;
3878 		mddev->safemode_delay = new_delay;
3879 		if (new_delay < old_delay || old_delay == 0)
3880 			mod_timer(&mddev->safemode_timer, jiffies+1);
3881 	}
3882 	return len;
3883 }
3884 static struct md_sysfs_entry md_safe_delay =
3885 __ATTR(safe_mode_delay, S_IRUGO|S_IWUSR,safe_delay_show, safe_delay_store);
3886 
3887 static ssize_t
level_show(struct mddev * mddev,char * page)3888 level_show(struct mddev *mddev, char *page)
3889 {
3890 	struct md_personality *p;
3891 	int ret;
3892 	spin_lock(&mddev->lock);
3893 	p = mddev->pers;
3894 	if (p)
3895 		ret = sprintf(page, "%s\n", p->name);
3896 	else if (mddev->clevel[0])
3897 		ret = sprintf(page, "%s\n", mddev->clevel);
3898 	else if (mddev->level != LEVEL_NONE)
3899 		ret = sprintf(page, "%d\n", mddev->level);
3900 	else
3901 		ret = 0;
3902 	spin_unlock(&mddev->lock);
3903 	return ret;
3904 }
3905 
3906 static ssize_t
level_store(struct mddev * mddev,const char * buf,size_t len)3907 level_store(struct mddev *mddev, const char *buf, size_t len)
3908 {
3909 	char clevel[16];
3910 	ssize_t rv;
3911 	size_t slen = len;
3912 	struct md_personality *pers, *oldpers;
3913 	long level;
3914 	void *priv, *oldpriv;
3915 	struct md_rdev *rdev;
3916 
3917 	if (slen == 0 || slen >= sizeof(clevel))
3918 		return -EINVAL;
3919 
3920 	rv = mddev_lock(mddev);
3921 	if (rv)
3922 		return rv;
3923 
3924 	if (mddev->pers == NULL) {
3925 		strncpy(mddev->clevel, buf, slen);
3926 		if (mddev->clevel[slen-1] == '\n')
3927 			slen--;
3928 		mddev->clevel[slen] = 0;
3929 		mddev->level = LEVEL_NONE;
3930 		rv = len;
3931 		goto out_unlock;
3932 	}
3933 	rv = -EROFS;
3934 	if (mddev->ro)
3935 		goto out_unlock;
3936 
3937 	/* request to change the personality.  Need to ensure:
3938 	 *  - array is not engaged in resync/recovery/reshape
3939 	 *  - old personality can be suspended
3940 	 *  - new personality will access other array.
3941 	 */
3942 
3943 	rv = -EBUSY;
3944 	if (mddev->sync_thread ||
3945 	    test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
3946 	    mddev->reshape_position != MaxSector ||
3947 	    mddev->sysfs_active)
3948 		goto out_unlock;
3949 
3950 	rv = -EINVAL;
3951 	if (!mddev->pers->quiesce) {
3952 		pr_warn("md: %s: %s does not support online personality change\n",
3953 			mdname(mddev), mddev->pers->name);
3954 		goto out_unlock;
3955 	}
3956 
3957 	/* Now find the new personality */
3958 	strncpy(clevel, buf, slen);
3959 	if (clevel[slen-1] == '\n')
3960 		slen--;
3961 	clevel[slen] = 0;
3962 	if (kstrtol(clevel, 10, &level))
3963 		level = LEVEL_NONE;
3964 
3965 	if (request_module("md-%s", clevel) != 0)
3966 		request_module("md-level-%s", clevel);
3967 	spin_lock(&pers_lock);
3968 	pers = find_pers(level, clevel);
3969 	if (!pers || !try_module_get(pers->owner)) {
3970 		spin_unlock(&pers_lock);
3971 		pr_warn("md: personality %s not loaded\n", clevel);
3972 		rv = -EINVAL;
3973 		goto out_unlock;
3974 	}
3975 	spin_unlock(&pers_lock);
3976 
3977 	if (pers == mddev->pers) {
3978 		/* Nothing to do! */
3979 		module_put(pers->owner);
3980 		rv = len;
3981 		goto out_unlock;
3982 	}
3983 	if (!pers->takeover) {
3984 		module_put(pers->owner);
3985 		pr_warn("md: %s: %s does not support personality takeover\n",
3986 			mdname(mddev), clevel);
3987 		rv = -EINVAL;
3988 		goto out_unlock;
3989 	}
3990 
3991 	rdev_for_each(rdev, mddev)
3992 		rdev->new_raid_disk = rdev->raid_disk;
3993 
3994 	/* ->takeover must set new_* and/or delta_disks
3995 	 * if it succeeds, and may set them when it fails.
3996 	 */
3997 	priv = pers->takeover(mddev);
3998 	if (IS_ERR(priv)) {
3999 		mddev->new_level = mddev->level;
4000 		mddev->new_layout = mddev->layout;
4001 		mddev->new_chunk_sectors = mddev->chunk_sectors;
4002 		mddev->raid_disks -= mddev->delta_disks;
4003 		mddev->delta_disks = 0;
4004 		mddev->reshape_backwards = 0;
4005 		module_put(pers->owner);
4006 		pr_warn("md: %s: %s would not accept array\n",
4007 			mdname(mddev), clevel);
4008 		rv = PTR_ERR(priv);
4009 		goto out_unlock;
4010 	}
4011 
4012 	/* Looks like we have a winner */
4013 	mddev_suspend(mddev);
4014 	mddev_detach(mddev);
4015 
4016 	spin_lock(&mddev->lock);
4017 	oldpers = mddev->pers;
4018 	oldpriv = mddev->private;
4019 	mddev->pers = pers;
4020 	mddev->private = priv;
4021 	strscpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
4022 	mddev->level = mddev->new_level;
4023 	mddev->layout = mddev->new_layout;
4024 	mddev->chunk_sectors = mddev->new_chunk_sectors;
4025 	mddev->delta_disks = 0;
4026 	mddev->reshape_backwards = 0;
4027 	mddev->degraded = 0;
4028 	spin_unlock(&mddev->lock);
4029 
4030 	if (oldpers->sync_request == NULL &&
4031 	    mddev->external) {
4032 		/* We are converting from a no-redundancy array
4033 		 * to a redundancy array and metadata is managed
4034 		 * externally so we need to be sure that writes
4035 		 * won't block due to a need to transition
4036 		 *      clean->dirty
4037 		 * until external management is started.
4038 		 */
4039 		mddev->in_sync = 0;
4040 		mddev->safemode_delay = 0;
4041 		mddev->safemode = 0;
4042 	}
4043 
4044 	oldpers->free(mddev, oldpriv);
4045 
4046 	if (oldpers->sync_request == NULL &&
4047 	    pers->sync_request != NULL) {
4048 		/* need to add the md_redundancy_group */
4049 		if (sysfs_create_group(&mddev->kobj, &md_redundancy_group))
4050 			pr_warn("md: cannot register extra attributes for %s\n",
4051 				mdname(mddev));
4052 		mddev->sysfs_action = sysfs_get_dirent(mddev->kobj.sd, "sync_action");
4053 		mddev->sysfs_completed = sysfs_get_dirent_safe(mddev->kobj.sd, "sync_completed");
4054 		mddev->sysfs_degraded = sysfs_get_dirent_safe(mddev->kobj.sd, "degraded");
4055 	}
4056 	if (oldpers->sync_request != NULL &&
4057 	    pers->sync_request == NULL) {
4058 		/* need to remove the md_redundancy_group */
4059 		if (mddev->to_remove == NULL)
4060 			mddev->to_remove = &md_redundancy_group;
4061 	}
4062 
4063 	module_put(oldpers->owner);
4064 
4065 	rdev_for_each(rdev, mddev) {
4066 		if (rdev->raid_disk < 0)
4067 			continue;
4068 		if (rdev->new_raid_disk >= mddev->raid_disks)
4069 			rdev->new_raid_disk = -1;
4070 		if (rdev->new_raid_disk == rdev->raid_disk)
4071 			continue;
4072 		sysfs_unlink_rdev(mddev, rdev);
4073 	}
4074 	rdev_for_each(rdev, mddev) {
4075 		if (rdev->raid_disk < 0)
4076 			continue;
4077 		if (rdev->new_raid_disk == rdev->raid_disk)
4078 			continue;
4079 		rdev->raid_disk = rdev->new_raid_disk;
4080 		if (rdev->raid_disk < 0)
4081 			clear_bit(In_sync, &rdev->flags);
4082 		else {
4083 			if (sysfs_link_rdev(mddev, rdev))
4084 				pr_warn("md: cannot register rd%d for %s after level change\n",
4085 					rdev->raid_disk, mdname(mddev));
4086 		}
4087 	}
4088 
4089 	if (pers->sync_request == NULL) {
4090 		/* this is now an array without redundancy, so
4091 		 * it must always be in_sync
4092 		 */
4093 		mddev->in_sync = 1;
4094 		del_timer_sync(&mddev->safemode_timer);
4095 	}
4096 	blk_set_stacking_limits(&mddev->queue->limits);
4097 	pers->run(mddev);
4098 	set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
4099 	mddev_resume(mddev);
4100 	if (!mddev->thread)
4101 		md_update_sb(mddev, 1);
4102 	sysfs_notify_dirent_safe(mddev->sysfs_level);
4103 	md_new_event();
4104 	rv = len;
4105 out_unlock:
4106 	mddev_unlock(mddev);
4107 	return rv;
4108 }
4109 
4110 static struct md_sysfs_entry md_level =
4111 __ATTR(level, S_IRUGO|S_IWUSR, level_show, level_store);
4112 
4113 static ssize_t
layout_show(struct mddev * mddev,char * page)4114 layout_show(struct mddev *mddev, char *page)
4115 {
4116 	/* just a number, not meaningful for all levels */
4117 	if (mddev->reshape_position != MaxSector &&
4118 	    mddev->layout != mddev->new_layout)
4119 		return sprintf(page, "%d (%d)\n",
4120 			       mddev->new_layout, mddev->layout);
4121 	return sprintf(page, "%d\n", mddev->layout);
4122 }
4123 
4124 static ssize_t
layout_store(struct mddev * mddev,const char * buf,size_t len)4125 layout_store(struct mddev *mddev, const char *buf, size_t len)
4126 {
4127 	unsigned int n;
4128 	int err;
4129 
4130 	err = kstrtouint(buf, 10, &n);
4131 	if (err < 0)
4132 		return err;
4133 	err = mddev_lock(mddev);
4134 	if (err)
4135 		return err;
4136 
4137 	if (mddev->pers) {
4138 		if (mddev->pers->check_reshape == NULL)
4139 			err = -EBUSY;
4140 		else if (mddev->ro)
4141 			err = -EROFS;
4142 		else {
4143 			mddev->new_layout = n;
4144 			err = mddev->pers->check_reshape(mddev);
4145 			if (err)
4146 				mddev->new_layout = mddev->layout;
4147 		}
4148 	} else {
4149 		mddev->new_layout = n;
4150 		if (mddev->reshape_position == MaxSector)
4151 			mddev->layout = n;
4152 	}
4153 	mddev_unlock(mddev);
4154 	return err ?: len;
4155 }
4156 static struct md_sysfs_entry md_layout =
4157 __ATTR(layout, S_IRUGO|S_IWUSR, layout_show, layout_store);
4158 
4159 static ssize_t
raid_disks_show(struct mddev * mddev,char * page)4160 raid_disks_show(struct mddev *mddev, char *page)
4161 {
4162 	if (mddev->raid_disks == 0)
4163 		return 0;
4164 	if (mddev->reshape_position != MaxSector &&
4165 	    mddev->delta_disks != 0)
4166 		return sprintf(page, "%d (%d)\n", mddev->raid_disks,
4167 			       mddev->raid_disks - mddev->delta_disks);
4168 	return sprintf(page, "%d\n", mddev->raid_disks);
4169 }
4170 
4171 static int update_raid_disks(struct mddev *mddev, int raid_disks);
4172 
4173 static ssize_t
raid_disks_store(struct mddev * mddev,const char * buf,size_t len)4174 raid_disks_store(struct mddev *mddev, const char *buf, size_t len)
4175 {
4176 	unsigned int n;
4177 	int err;
4178 
4179 	err = kstrtouint(buf, 10, &n);
4180 	if (err < 0)
4181 		return err;
4182 
4183 	err = mddev_lock(mddev);
4184 	if (err)
4185 		return err;
4186 	if (mddev->pers)
4187 		err = update_raid_disks(mddev, n);
4188 	else if (mddev->reshape_position != MaxSector) {
4189 		struct md_rdev *rdev;
4190 		int olddisks = mddev->raid_disks - mddev->delta_disks;
4191 
4192 		err = -EINVAL;
4193 		rdev_for_each(rdev, mddev) {
4194 			if (olddisks < n &&
4195 			    rdev->data_offset < rdev->new_data_offset)
4196 				goto out_unlock;
4197 			if (olddisks > n &&
4198 			    rdev->data_offset > rdev->new_data_offset)
4199 				goto out_unlock;
4200 		}
4201 		err = 0;
4202 		mddev->delta_disks = n - olddisks;
4203 		mddev->raid_disks = n;
4204 		mddev->reshape_backwards = (mddev->delta_disks < 0);
4205 	} else
4206 		mddev->raid_disks = n;
4207 out_unlock:
4208 	mddev_unlock(mddev);
4209 	return err ? err : len;
4210 }
4211 static struct md_sysfs_entry md_raid_disks =
4212 __ATTR(raid_disks, S_IRUGO|S_IWUSR, raid_disks_show, raid_disks_store);
4213 
4214 static ssize_t
uuid_show(struct mddev * mddev,char * page)4215 uuid_show(struct mddev *mddev, char *page)
4216 {
4217 	return sprintf(page, "%pU\n", mddev->uuid);
4218 }
4219 static struct md_sysfs_entry md_uuid =
4220 __ATTR(uuid, S_IRUGO, uuid_show, NULL);
4221 
4222 static ssize_t
chunk_size_show(struct mddev * mddev,char * page)4223 chunk_size_show(struct mddev *mddev, char *page)
4224 {
4225 	if (mddev->reshape_position != MaxSector &&
4226 	    mddev->chunk_sectors != mddev->new_chunk_sectors)
4227 		return sprintf(page, "%d (%d)\n",
4228 			       mddev->new_chunk_sectors << 9,
4229 			       mddev->chunk_sectors << 9);
4230 	return sprintf(page, "%d\n", mddev->chunk_sectors << 9);
4231 }
4232 
4233 static ssize_t
chunk_size_store(struct mddev * mddev,const char * buf,size_t len)4234 chunk_size_store(struct mddev *mddev, const char *buf, size_t len)
4235 {
4236 	unsigned long n;
4237 	int err;
4238 
4239 	err = kstrtoul(buf, 10, &n);
4240 	if (err < 0)
4241 		return err;
4242 
4243 	err = mddev_lock(mddev);
4244 	if (err)
4245 		return err;
4246 	if (mddev->pers) {
4247 		if (mddev->pers->check_reshape == NULL)
4248 			err = -EBUSY;
4249 		else if (mddev->ro)
4250 			err = -EROFS;
4251 		else {
4252 			mddev->new_chunk_sectors = n >> 9;
4253 			err = mddev->pers->check_reshape(mddev);
4254 			if (err)
4255 				mddev->new_chunk_sectors = mddev->chunk_sectors;
4256 		}
4257 	} else {
4258 		mddev->new_chunk_sectors = n >> 9;
4259 		if (mddev->reshape_position == MaxSector)
4260 			mddev->chunk_sectors = n >> 9;
4261 	}
4262 	mddev_unlock(mddev);
4263 	return err ?: len;
4264 }
4265 static struct md_sysfs_entry md_chunk_size =
4266 __ATTR(chunk_size, S_IRUGO|S_IWUSR, chunk_size_show, chunk_size_store);
4267 
4268 static ssize_t
resync_start_show(struct mddev * mddev,char * page)4269 resync_start_show(struct mddev *mddev, char *page)
4270 {
4271 	if (mddev->recovery_cp == MaxSector)
4272 		return sprintf(page, "none\n");
4273 	return sprintf(page, "%llu\n", (unsigned long long)mddev->recovery_cp);
4274 }
4275 
4276 static ssize_t
resync_start_store(struct mddev * mddev,const char * buf,size_t len)4277 resync_start_store(struct mddev *mddev, const char *buf, size_t len)
4278 {
4279 	unsigned long long n;
4280 	int err;
4281 
4282 	if (cmd_match(buf, "none"))
4283 		n = MaxSector;
4284 	else {
4285 		err = kstrtoull(buf, 10, &n);
4286 		if (err < 0)
4287 			return err;
4288 		if (n != (sector_t)n)
4289 			return -EINVAL;
4290 	}
4291 
4292 	err = mddev_lock(mddev);
4293 	if (err)
4294 		return err;
4295 	if (mddev->pers && !test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
4296 		err = -EBUSY;
4297 
4298 	if (!err) {
4299 		mddev->recovery_cp = n;
4300 		if (mddev->pers)
4301 			set_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags);
4302 	}
4303 	mddev_unlock(mddev);
4304 	return err ?: len;
4305 }
4306 static struct md_sysfs_entry md_resync_start =
4307 __ATTR_PREALLOC(resync_start, S_IRUGO|S_IWUSR,
4308 		resync_start_show, resync_start_store);
4309 
4310 /*
4311  * The array state can be:
4312  *
4313  * clear
4314  *     No devices, no size, no level
4315  *     Equivalent to STOP_ARRAY ioctl
4316  * inactive
4317  *     May have some settings, but array is not active
4318  *        all IO results in error
4319  *     When written, doesn't tear down array, but just stops it
4320  * suspended (not supported yet)
4321  *     All IO requests will block. The array can be reconfigured.
4322  *     Writing this, if accepted, will block until array is quiescent
4323  * readonly
4324  *     no resync can happen.  no superblocks get written.
4325  *     write requests fail
4326  * read-auto
4327  *     like readonly, but behaves like 'clean' on a write request.
4328  *
4329  * clean - no pending writes, but otherwise active.
4330  *     When written to inactive array, starts without resync
4331  *     If a write request arrives then
4332  *       if metadata is known, mark 'dirty' and switch to 'active'.
4333  *       if not known, block and switch to write-pending
4334  *     If written to an active array that has pending writes, then fails.
4335  * active
4336  *     fully active: IO and resync can be happening.
4337  *     When written to inactive array, starts with resync
4338  *
4339  * write-pending
4340  *     clean, but writes are blocked waiting for 'active' to be written.
4341  *
4342  * active-idle
4343  *     like active, but no writes have been seen for a while (100msec).
4344  *
4345  * broken
4346 *     Array is failed. It's useful because mounted-arrays aren't stopped
4347 *     when array is failed, so this state will at least alert the user that
4348 *     something is wrong.
4349  */
4350 enum array_state { clear, inactive, suspended, readonly, read_auto, clean, active,
4351 		   write_pending, active_idle, broken, bad_word};
4352 static char *array_states[] = {
4353 	"clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active",
4354 	"write-pending", "active-idle", "broken", NULL };
4355 
match_word(const char * word,char ** list)4356 static int match_word(const char *word, char **list)
4357 {
4358 	int n;
4359 	for (n=0; list[n]; n++)
4360 		if (cmd_match(word, list[n]))
4361 			break;
4362 	return n;
4363 }
4364 
4365 static ssize_t
array_state_show(struct mddev * mddev,char * page)4366 array_state_show(struct mddev *mddev, char *page)
4367 {
4368 	enum array_state st = inactive;
4369 
4370 	if (mddev->pers && !test_bit(MD_NOT_READY, &mddev->flags)) {
4371 		switch(mddev->ro) {
4372 		case 1:
4373 			st = readonly;
4374 			break;
4375 		case 2:
4376 			st = read_auto;
4377 			break;
4378 		case 0:
4379 			spin_lock(&mddev->lock);
4380 			if (test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags))
4381 				st = write_pending;
4382 			else if (mddev->in_sync)
4383 				st = clean;
4384 			else if (mddev->safemode)
4385 				st = active_idle;
4386 			else
4387 				st = active;
4388 			spin_unlock(&mddev->lock);
4389 		}
4390 
4391 		if (test_bit(MD_BROKEN, &mddev->flags) && st == clean)
4392 			st = broken;
4393 	} else {
4394 		if (list_empty(&mddev->disks) &&
4395 		    mddev->raid_disks == 0 &&
4396 		    mddev->dev_sectors == 0)
4397 			st = clear;
4398 		else
4399 			st = inactive;
4400 	}
4401 	return sprintf(page, "%s\n", array_states[st]);
4402 }
4403 
4404 static int do_md_stop(struct mddev *mddev, int ro, struct block_device *bdev);
4405 static int md_set_readonly(struct mddev *mddev, struct block_device *bdev);
4406 static int restart_array(struct mddev *mddev);
4407 
4408 static ssize_t
array_state_store(struct mddev * mddev,const char * buf,size_t len)4409 array_state_store(struct mddev *mddev, const char *buf, size_t len)
4410 {
4411 	int err = 0;
4412 	enum array_state st = match_word(buf, array_states);
4413 
4414 	if (mddev->pers && (st == active || st == clean) && mddev->ro != 1) {
4415 		/* don't take reconfig_mutex when toggling between
4416 		 * clean and active
4417 		 */
4418 		spin_lock(&mddev->lock);
4419 		if (st == active) {
4420 			restart_array(mddev);
4421 			clear_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags);
4422 			md_wakeup_thread(mddev->thread);
4423 			wake_up(&mddev->sb_wait);
4424 		} else /* st == clean */ {
4425 			restart_array(mddev);
4426 			if (!set_in_sync(mddev))
4427 				err = -EBUSY;
4428 		}
4429 		if (!err)
4430 			sysfs_notify_dirent_safe(mddev->sysfs_state);
4431 		spin_unlock(&mddev->lock);
4432 		return err ?: len;
4433 	}
4434 	err = mddev_lock(mddev);
4435 	if (err)
4436 		return err;
4437 	err = -EINVAL;
4438 	switch(st) {
4439 	case bad_word:
4440 		break;
4441 	case clear:
4442 		/* stopping an active array */
4443 		err = do_md_stop(mddev, 0, NULL);
4444 		break;
4445 	case inactive:
4446 		/* stopping an active array */
4447 		if (mddev->pers)
4448 			err = do_md_stop(mddev, 2, NULL);
4449 		else
4450 			err = 0; /* already inactive */
4451 		break;
4452 	case suspended:
4453 		break; /* not supported yet */
4454 	case readonly:
4455 		if (mddev->pers)
4456 			err = md_set_readonly(mddev, NULL);
4457 		else {
4458 			mddev->ro = 1;
4459 			set_disk_ro(mddev->gendisk, 1);
4460 			err = do_md_run(mddev);
4461 		}
4462 		break;
4463 	case read_auto:
4464 		if (mddev->pers) {
4465 			if (mddev->ro == 0)
4466 				err = md_set_readonly(mddev, NULL);
4467 			else if (mddev->ro == 1)
4468 				err = restart_array(mddev);
4469 			if (err == 0) {
4470 				mddev->ro = 2;
4471 				set_disk_ro(mddev->gendisk, 0);
4472 			}
4473 		} else {
4474 			mddev->ro = 2;
4475 			err = do_md_run(mddev);
4476 		}
4477 		break;
4478 	case clean:
4479 		if (mddev->pers) {
4480 			err = restart_array(mddev);
4481 			if (err)
4482 				break;
4483 			spin_lock(&mddev->lock);
4484 			if (!set_in_sync(mddev))
4485 				err = -EBUSY;
4486 			spin_unlock(&mddev->lock);
4487 		} else
4488 			err = -EINVAL;
4489 		break;
4490 	case active:
4491 		if (mddev->pers) {
4492 			err = restart_array(mddev);
4493 			if (err)
4494 				break;
4495 			clear_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags);
4496 			wake_up(&mddev->sb_wait);
4497 			err = 0;
4498 		} else {
4499 			mddev->ro = 0;
4500 			set_disk_ro(mddev->gendisk, 0);
4501 			err = do_md_run(mddev);
4502 		}
4503 		break;
4504 	case write_pending:
4505 	case active_idle:
4506 	case broken:
4507 		/* these cannot be set */
4508 		break;
4509 	}
4510 
4511 	if (!err) {
4512 		if (mddev->hold_active == UNTIL_IOCTL)
4513 			mddev->hold_active = 0;
4514 		sysfs_notify_dirent_safe(mddev->sysfs_state);
4515 	}
4516 	mddev_unlock(mddev);
4517 	return err ?: len;
4518 }
4519 static struct md_sysfs_entry md_array_state =
4520 __ATTR_PREALLOC(array_state, S_IRUGO|S_IWUSR, array_state_show, array_state_store);
4521 
4522 static ssize_t
max_corrected_read_errors_show(struct mddev * mddev,char * page)4523 max_corrected_read_errors_show(struct mddev *mddev, char *page) {
4524 	return sprintf(page, "%d\n",
4525 		       atomic_read(&mddev->max_corr_read_errors));
4526 }
4527 
4528 static ssize_t
max_corrected_read_errors_store(struct mddev * mddev,const char * buf,size_t len)4529 max_corrected_read_errors_store(struct mddev *mddev, const char *buf, size_t len)
4530 {
4531 	unsigned int n;
4532 	int rv;
4533 
4534 	rv = kstrtouint(buf, 10, &n);
4535 	if (rv < 0)
4536 		return rv;
4537 	atomic_set(&mddev->max_corr_read_errors, n);
4538 	return len;
4539 }
4540 
4541 static struct md_sysfs_entry max_corr_read_errors =
4542 __ATTR(max_read_errors, S_IRUGO|S_IWUSR, max_corrected_read_errors_show,
4543 	max_corrected_read_errors_store);
4544 
4545 static ssize_t
null_show(struct mddev * mddev,char * page)4546 null_show(struct mddev *mddev, char *page)
4547 {
4548 	return -EINVAL;
4549 }
4550 
4551 /* need to ensure rdev_delayed_delete() has completed */
flush_rdev_wq(struct mddev * mddev)4552 static void flush_rdev_wq(struct mddev *mddev)
4553 {
4554 	struct md_rdev *rdev;
4555 
4556 	rcu_read_lock();
4557 	rdev_for_each_rcu(rdev, mddev)
4558 		if (work_pending(&rdev->del_work)) {
4559 			flush_workqueue(md_rdev_misc_wq);
4560 			break;
4561 		}
4562 	rcu_read_unlock();
4563 }
4564 
4565 static ssize_t
new_dev_store(struct mddev * mddev,const char * buf,size_t len)4566 new_dev_store(struct mddev *mddev, const char *buf, size_t len)
4567 {
4568 	/* buf must be %d:%d\n? giving major and minor numbers */
4569 	/* The new device is added to the array.
4570 	 * If the array has a persistent superblock, we read the
4571 	 * superblock to initialise info and check validity.
4572 	 * Otherwise, only checking done is that in bind_rdev_to_array,
4573 	 * which mainly checks size.
4574 	 */
4575 	char *e;
4576 	int major = simple_strtoul(buf, &e, 10);
4577 	int minor;
4578 	dev_t dev;
4579 	struct md_rdev *rdev;
4580 	int err;
4581 
4582 	if (!*buf || *e != ':' || !e[1] || e[1] == '\n')
4583 		return -EINVAL;
4584 	minor = simple_strtoul(e+1, &e, 10);
4585 	if (*e && *e != '\n')
4586 		return -EINVAL;
4587 	dev = MKDEV(major, minor);
4588 	if (major != MAJOR(dev) ||
4589 	    minor != MINOR(dev))
4590 		return -EOVERFLOW;
4591 
4592 	flush_rdev_wq(mddev);
4593 	err = mddev_lock(mddev);
4594 	if (err)
4595 		return err;
4596 	if (mddev->persistent) {
4597 		rdev = md_import_device(dev, mddev->major_version,
4598 					mddev->minor_version);
4599 		if (!IS_ERR(rdev) && !list_empty(&mddev->disks)) {
4600 			struct md_rdev *rdev0
4601 				= list_entry(mddev->disks.next,
4602 					     struct md_rdev, same_set);
4603 			err = super_types[mddev->major_version]
4604 				.load_super(rdev, rdev0, mddev->minor_version);
4605 			if (err < 0)
4606 				goto out;
4607 		}
4608 	} else if (mddev->external)
4609 		rdev = md_import_device(dev, -2, -1);
4610 	else
4611 		rdev = md_import_device(dev, -1, -1);
4612 
4613 	if (IS_ERR(rdev)) {
4614 		mddev_unlock(mddev);
4615 		return PTR_ERR(rdev);
4616 	}
4617 	err = bind_rdev_to_array(rdev, mddev);
4618  out:
4619 	if (err)
4620 		export_rdev(rdev);
4621 	mddev_unlock(mddev);
4622 	if (!err)
4623 		md_new_event();
4624 	return err ? err : len;
4625 }
4626 
4627 static struct md_sysfs_entry md_new_device =
4628 __ATTR(new_dev, S_IWUSR, null_show, new_dev_store);
4629 
4630 static ssize_t
bitmap_store(struct mddev * mddev,const char * buf,size_t len)4631 bitmap_store(struct mddev *mddev, const char *buf, size_t len)
4632 {
4633 	char *end;
4634 	unsigned long chunk, end_chunk;
4635 	int err;
4636 
4637 	err = mddev_lock(mddev);
4638 	if (err)
4639 		return err;
4640 	if (!mddev->bitmap)
4641 		goto out;
4642 	/* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */
4643 	while (*buf) {
4644 		chunk = end_chunk = simple_strtoul(buf, &end, 0);
4645 		if (buf == end) break;
4646 		if (*end == '-') { /* range */
4647 			buf = end + 1;
4648 			end_chunk = simple_strtoul(buf, &end, 0);
4649 			if (buf == end) break;
4650 		}
4651 		if (*end && !isspace(*end)) break;
4652 		md_bitmap_dirty_bits(mddev->bitmap, chunk, end_chunk);
4653 		buf = skip_spaces(end);
4654 	}
4655 	md_bitmap_unplug(mddev->bitmap); /* flush the bits to disk */
4656 out:
4657 	mddev_unlock(mddev);
4658 	return len;
4659 }
4660 
4661 static struct md_sysfs_entry md_bitmap =
4662 __ATTR(bitmap_set_bits, S_IWUSR, null_show, bitmap_store);
4663 
4664 static ssize_t
size_show(struct mddev * mddev,char * page)4665 size_show(struct mddev *mddev, char *page)
4666 {
4667 	return sprintf(page, "%llu\n",
4668 		(unsigned long long)mddev->dev_sectors / 2);
4669 }
4670 
4671 static int update_size(struct mddev *mddev, sector_t num_sectors);
4672 
4673 static ssize_t
size_store(struct mddev * mddev,const char * buf,size_t len)4674 size_store(struct mddev *mddev, const char *buf, size_t len)
4675 {
4676 	/* If array is inactive, we can reduce the component size, but
4677 	 * not increase it (except from 0).
4678 	 * If array is active, we can try an on-line resize
4679 	 */
4680 	sector_t sectors;
4681 	int err = strict_blocks_to_sectors(buf, &sectors);
4682 
4683 	if (err < 0)
4684 		return err;
4685 	err = mddev_lock(mddev);
4686 	if (err)
4687 		return err;
4688 	if (mddev->pers) {
4689 		err = update_size(mddev, sectors);
4690 		if (err == 0)
4691 			md_update_sb(mddev, 1);
4692 	} else {
4693 		if (mddev->dev_sectors == 0 ||
4694 		    mddev->dev_sectors > sectors)
4695 			mddev->dev_sectors = sectors;
4696 		else
4697 			err = -ENOSPC;
4698 	}
4699 	mddev_unlock(mddev);
4700 	return err ? err : len;
4701 }
4702 
4703 static struct md_sysfs_entry md_size =
4704 __ATTR(component_size, S_IRUGO|S_IWUSR, size_show, size_store);
4705 
4706 /* Metadata version.
4707  * This is one of
4708  *   'none' for arrays with no metadata (good luck...)
4709  *   'external' for arrays with externally managed metadata,
4710  * or N.M for internally known formats
4711  */
4712 static ssize_t
metadata_show(struct mddev * mddev,char * page)4713 metadata_show(struct mddev *mddev, char *page)
4714 {
4715 	if (mddev->persistent)
4716 		return sprintf(page, "%d.%d\n",
4717 			       mddev->major_version, mddev->minor_version);
4718 	else if (mddev->external)
4719 		return sprintf(page, "external:%s\n", mddev->metadata_type);
4720 	else
4721 		return sprintf(page, "none\n");
4722 }
4723 
4724 static ssize_t
metadata_store(struct mddev * mddev,const char * buf,size_t len)4725 metadata_store(struct mddev *mddev, const char *buf, size_t len)
4726 {
4727 	int major, minor;
4728 	char *e;
4729 	int err;
4730 	/* Changing the details of 'external' metadata is
4731 	 * always permitted.  Otherwise there must be
4732 	 * no devices attached to the array.
4733 	 */
4734 
4735 	err = mddev_lock(mddev);
4736 	if (err)
4737 		return err;
4738 	err = -EBUSY;
4739 	if (mddev->external && strncmp(buf, "external:", 9) == 0)
4740 		;
4741 	else if (!list_empty(&mddev->disks))
4742 		goto out_unlock;
4743 
4744 	err = 0;
4745 	if (cmd_match(buf, "none")) {
4746 		mddev->persistent = 0;
4747 		mddev->external = 0;
4748 		mddev->major_version = 0;
4749 		mddev->minor_version = 90;
4750 		goto out_unlock;
4751 	}
4752 	if (strncmp(buf, "external:", 9) == 0) {
4753 		size_t namelen = len-9;
4754 		if (namelen >= sizeof(mddev->metadata_type))
4755 			namelen = sizeof(mddev->metadata_type)-1;
4756 		strncpy(mddev->metadata_type, buf+9, namelen);
4757 		mddev->metadata_type[namelen] = 0;
4758 		if (namelen && mddev->metadata_type[namelen-1] == '\n')
4759 			mddev->metadata_type[--namelen] = 0;
4760 		mddev->persistent = 0;
4761 		mddev->external = 1;
4762 		mddev->major_version = 0;
4763 		mddev->minor_version = 90;
4764 		goto out_unlock;
4765 	}
4766 	major = simple_strtoul(buf, &e, 10);
4767 	err = -EINVAL;
4768 	if (e==buf || *e != '.')
4769 		goto out_unlock;
4770 	buf = e+1;
4771 	minor = simple_strtoul(buf, &e, 10);
4772 	if (e==buf || (*e && *e != '\n') )
4773 		goto out_unlock;
4774 	err = -ENOENT;
4775 	if (major >= ARRAY_SIZE(super_types) || super_types[major].name == NULL)
4776 		goto out_unlock;
4777 	mddev->major_version = major;
4778 	mddev->minor_version = minor;
4779 	mddev->persistent = 1;
4780 	mddev->external = 0;
4781 	err = 0;
4782 out_unlock:
4783 	mddev_unlock(mddev);
4784 	return err ?: len;
4785 }
4786 
4787 static struct md_sysfs_entry md_metadata =
4788 __ATTR_PREALLOC(metadata_version, S_IRUGO|S_IWUSR, metadata_show, metadata_store);
4789 
4790 static ssize_t
action_show(struct mddev * mddev,char * page)4791 action_show(struct mddev *mddev, char *page)
4792 {
4793 	char *type = "idle";
4794 	unsigned long recovery = mddev->recovery;
4795 	if (test_bit(MD_RECOVERY_FROZEN, &recovery))
4796 		type = "frozen";
4797 	else if (test_bit(MD_RECOVERY_RUNNING, &recovery) ||
4798 	    (!mddev->ro && test_bit(MD_RECOVERY_NEEDED, &recovery))) {
4799 		if (test_bit(MD_RECOVERY_RESHAPE, &recovery))
4800 			type = "reshape";
4801 		else if (test_bit(MD_RECOVERY_SYNC, &recovery)) {
4802 			if (!test_bit(MD_RECOVERY_REQUESTED, &recovery))
4803 				type = "resync";
4804 			else if (test_bit(MD_RECOVERY_CHECK, &recovery))
4805 				type = "check";
4806 			else
4807 				type = "repair";
4808 		} else if (test_bit(MD_RECOVERY_RECOVER, &recovery))
4809 			type = "recover";
4810 		else if (mddev->reshape_position != MaxSector)
4811 			type = "reshape";
4812 	}
4813 	return sprintf(page, "%s\n", type);
4814 }
4815 
4816 static ssize_t
action_store(struct mddev * mddev,const char * page,size_t len)4817 action_store(struct mddev *mddev, const char *page, size_t len)
4818 {
4819 	if (!mddev->pers || !mddev->pers->sync_request)
4820 		return -EINVAL;
4821 
4822 
4823 	if (cmd_match(page, "idle") || cmd_match(page, "frozen")) {
4824 		if (cmd_match(page, "frozen"))
4825 			set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4826 		else
4827 			clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4828 		if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
4829 		    mddev_lock(mddev) == 0) {
4830 			if (work_pending(&mddev->del_work))
4831 				flush_workqueue(md_misc_wq);
4832 			if (mddev->sync_thread) {
4833 				set_bit(MD_RECOVERY_INTR, &mddev->recovery);
4834 				md_reap_sync_thread(mddev);
4835 			}
4836 			mddev_unlock(mddev);
4837 		}
4838 	} else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4839 		return -EBUSY;
4840 	else if (cmd_match(page, "resync"))
4841 		clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4842 	else if (cmd_match(page, "recover")) {
4843 		clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4844 		set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
4845 	} else if (cmd_match(page, "reshape")) {
4846 		int err;
4847 		if (mddev->pers->start_reshape == NULL)
4848 			return -EINVAL;
4849 		err = mddev_lock(mddev);
4850 		if (!err) {
4851 			if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4852 				err =  -EBUSY;
4853 			else {
4854 				clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4855 				err = mddev->pers->start_reshape(mddev);
4856 			}
4857 			mddev_unlock(mddev);
4858 		}
4859 		if (err)
4860 			return err;
4861 		sysfs_notify_dirent_safe(mddev->sysfs_degraded);
4862 	} else {
4863 		if (cmd_match(page, "check"))
4864 			set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
4865 		else if (!cmd_match(page, "repair"))
4866 			return -EINVAL;
4867 		clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4868 		set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
4869 		set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
4870 	}
4871 	if (mddev->ro == 2) {
4872 		/* A write to sync_action is enough to justify
4873 		 * canceling read-auto mode
4874 		 */
4875 		mddev->ro = 0;
4876 		md_wakeup_thread(mddev->sync_thread);
4877 	}
4878 	set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4879 	md_wakeup_thread(mddev->thread);
4880 	sysfs_notify_dirent_safe(mddev->sysfs_action);
4881 	return len;
4882 }
4883 
4884 static struct md_sysfs_entry md_scan_mode =
4885 __ATTR_PREALLOC(sync_action, S_IRUGO|S_IWUSR, action_show, action_store);
4886 
4887 static ssize_t
last_sync_action_show(struct mddev * mddev,char * page)4888 last_sync_action_show(struct mddev *mddev, char *page)
4889 {
4890 	return sprintf(page, "%s\n", mddev->last_sync_action);
4891 }
4892 
4893 static struct md_sysfs_entry md_last_scan_mode = __ATTR_RO(last_sync_action);
4894 
4895 static ssize_t
mismatch_cnt_show(struct mddev * mddev,char * page)4896 mismatch_cnt_show(struct mddev *mddev, char *page)
4897 {
4898 	return sprintf(page, "%llu\n",
4899 		       (unsigned long long)
4900 		       atomic64_read(&mddev->resync_mismatches));
4901 }
4902 
4903 static struct md_sysfs_entry md_mismatches = __ATTR_RO(mismatch_cnt);
4904 
4905 static ssize_t
sync_min_show(struct mddev * mddev,char * page)4906 sync_min_show(struct mddev *mddev, char *page)
4907 {
4908 	return sprintf(page, "%d (%s)\n", speed_min(mddev),
4909 		       mddev->sync_speed_min ? "local": "system");
4910 }
4911 
4912 static ssize_t
sync_min_store(struct mddev * mddev,const char * buf,size_t len)4913 sync_min_store(struct mddev *mddev, const char *buf, size_t len)
4914 {
4915 	unsigned int min;
4916 	int rv;
4917 
4918 	if (strncmp(buf, "system", 6)==0) {
4919 		min = 0;
4920 	} else {
4921 		rv = kstrtouint(buf, 10, &min);
4922 		if (rv < 0)
4923 			return rv;
4924 		if (min == 0)
4925 			return -EINVAL;
4926 	}
4927 	mddev->sync_speed_min = min;
4928 	return len;
4929 }
4930 
4931 static struct md_sysfs_entry md_sync_min =
4932 __ATTR(sync_speed_min, S_IRUGO|S_IWUSR, sync_min_show, sync_min_store);
4933 
4934 static ssize_t
sync_max_show(struct mddev * mddev,char * page)4935 sync_max_show(struct mddev *mddev, char *page)
4936 {
4937 	return sprintf(page, "%d (%s)\n", speed_max(mddev),
4938 		       mddev->sync_speed_max ? "local": "system");
4939 }
4940 
4941 static ssize_t
sync_max_store(struct mddev * mddev,const char * buf,size_t len)4942 sync_max_store(struct mddev *mddev, const char *buf, size_t len)
4943 {
4944 	unsigned int max;
4945 	int rv;
4946 
4947 	if (strncmp(buf, "system", 6)==0) {
4948 		max = 0;
4949 	} else {
4950 		rv = kstrtouint(buf, 10, &max);
4951 		if (rv < 0)
4952 			return rv;
4953 		if (max == 0)
4954 			return -EINVAL;
4955 	}
4956 	mddev->sync_speed_max = max;
4957 	return len;
4958 }
4959 
4960 static struct md_sysfs_entry md_sync_max =
4961 __ATTR(sync_speed_max, S_IRUGO|S_IWUSR, sync_max_show, sync_max_store);
4962 
4963 static ssize_t
degraded_show(struct mddev * mddev,char * page)4964 degraded_show(struct mddev *mddev, char *page)
4965 {
4966 	return sprintf(page, "%d\n", mddev->degraded);
4967 }
4968 static struct md_sysfs_entry md_degraded = __ATTR_RO(degraded);
4969 
4970 static ssize_t
sync_force_parallel_show(struct mddev * mddev,char * page)4971 sync_force_parallel_show(struct mddev *mddev, char *page)
4972 {
4973 	return sprintf(page, "%d\n", mddev->parallel_resync);
4974 }
4975 
4976 static ssize_t
sync_force_parallel_store(struct mddev * mddev,const char * buf,size_t len)4977 sync_force_parallel_store(struct mddev *mddev, const char *buf, size_t len)
4978 {
4979 	long n;
4980 
4981 	if (kstrtol(buf, 10, &n))
4982 		return -EINVAL;
4983 
4984 	if (n != 0 && n != 1)
4985 		return -EINVAL;
4986 
4987 	mddev->parallel_resync = n;
4988 
4989 	if (mddev->sync_thread)
4990 		wake_up(&resync_wait);
4991 
4992 	return len;
4993 }
4994 
4995 /* force parallel resync, even with shared block devices */
4996 static struct md_sysfs_entry md_sync_force_parallel =
4997 __ATTR(sync_force_parallel, S_IRUGO|S_IWUSR,
4998        sync_force_parallel_show, sync_force_parallel_store);
4999 
5000 static ssize_t
sync_speed_show(struct mddev * mddev,char * page)5001 sync_speed_show(struct mddev *mddev, char *page)
5002 {
5003 	unsigned long resync, dt, db;
5004 	if (mddev->curr_resync == 0)
5005 		return sprintf(page, "none\n");
5006 	resync = mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active);
5007 	dt = (jiffies - mddev->resync_mark) / HZ;
5008 	if (!dt) dt++;
5009 	db = resync - mddev->resync_mark_cnt;
5010 	return sprintf(page, "%lu\n", db/dt/2); /* K/sec */
5011 }
5012 
5013 static struct md_sysfs_entry md_sync_speed = __ATTR_RO(sync_speed);
5014 
5015 static ssize_t
sync_completed_show(struct mddev * mddev,char * page)5016 sync_completed_show(struct mddev *mddev, char *page)
5017 {
5018 	unsigned long long max_sectors, resync;
5019 
5020 	if (!test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
5021 		return sprintf(page, "none\n");
5022 
5023 	if (mddev->curr_resync == 1 ||
5024 	    mddev->curr_resync == 2)
5025 		return sprintf(page, "delayed\n");
5026 
5027 	if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ||
5028 	    test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
5029 		max_sectors = mddev->resync_max_sectors;
5030 	else
5031 		max_sectors = mddev->dev_sectors;
5032 
5033 	resync = mddev->curr_resync_completed;
5034 	return sprintf(page, "%llu / %llu\n", resync, max_sectors);
5035 }
5036 
5037 static struct md_sysfs_entry md_sync_completed =
5038 	__ATTR_PREALLOC(sync_completed, S_IRUGO, sync_completed_show, NULL);
5039 
5040 static ssize_t
min_sync_show(struct mddev * mddev,char * page)5041 min_sync_show(struct mddev *mddev, char *page)
5042 {
5043 	return sprintf(page, "%llu\n",
5044 		       (unsigned long long)mddev->resync_min);
5045 }
5046 static ssize_t
min_sync_store(struct mddev * mddev,const char * buf,size_t len)5047 min_sync_store(struct mddev *mddev, const char *buf, size_t len)
5048 {
5049 	unsigned long long min;
5050 	int err;
5051 
5052 	if (kstrtoull(buf, 10, &min))
5053 		return -EINVAL;
5054 
5055 	spin_lock(&mddev->lock);
5056 	err = -EINVAL;
5057 	if (min > mddev->resync_max)
5058 		goto out_unlock;
5059 
5060 	err = -EBUSY;
5061 	if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
5062 		goto out_unlock;
5063 
5064 	/* Round down to multiple of 4K for safety */
5065 	mddev->resync_min = round_down(min, 8);
5066 	err = 0;
5067 
5068 out_unlock:
5069 	spin_unlock(&mddev->lock);
5070 	return err ?: len;
5071 }
5072 
5073 static struct md_sysfs_entry md_min_sync =
5074 __ATTR(sync_min, S_IRUGO|S_IWUSR, min_sync_show, min_sync_store);
5075 
5076 static ssize_t
max_sync_show(struct mddev * mddev,char * page)5077 max_sync_show(struct mddev *mddev, char *page)
5078 {
5079 	if (mddev->resync_max == MaxSector)
5080 		return sprintf(page, "max\n");
5081 	else
5082 		return sprintf(page, "%llu\n",
5083 			       (unsigned long long)mddev->resync_max);
5084 }
5085 static ssize_t
max_sync_store(struct mddev * mddev,const char * buf,size_t len)5086 max_sync_store(struct mddev *mddev, const char *buf, size_t len)
5087 {
5088 	int err;
5089 	spin_lock(&mddev->lock);
5090 	if (strncmp(buf, "max", 3) == 0)
5091 		mddev->resync_max = MaxSector;
5092 	else {
5093 		unsigned long long max;
5094 		int chunk;
5095 
5096 		err = -EINVAL;
5097 		if (kstrtoull(buf, 10, &max))
5098 			goto out_unlock;
5099 		if (max < mddev->resync_min)
5100 			goto out_unlock;
5101 
5102 		err = -EBUSY;
5103 		if (max < mddev->resync_max &&
5104 		    mddev->ro == 0 &&
5105 		    test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
5106 			goto out_unlock;
5107 
5108 		/* Must be a multiple of chunk_size */
5109 		chunk = mddev->chunk_sectors;
5110 		if (chunk) {
5111 			sector_t temp = max;
5112 
5113 			err = -EINVAL;
5114 			if (sector_div(temp, chunk))
5115 				goto out_unlock;
5116 		}
5117 		mddev->resync_max = max;
5118 	}
5119 	wake_up(&mddev->recovery_wait);
5120 	err = 0;
5121 out_unlock:
5122 	spin_unlock(&mddev->lock);
5123 	return err ?: len;
5124 }
5125 
5126 static struct md_sysfs_entry md_max_sync =
5127 __ATTR(sync_max, S_IRUGO|S_IWUSR, max_sync_show, max_sync_store);
5128 
5129 static ssize_t
suspend_lo_show(struct mddev * mddev,char * page)5130 suspend_lo_show(struct mddev *mddev, char *page)
5131 {
5132 	return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_lo);
5133 }
5134 
5135 static ssize_t
suspend_lo_store(struct mddev * mddev,const char * buf,size_t len)5136 suspend_lo_store(struct mddev *mddev, const char *buf, size_t len)
5137 {
5138 	unsigned long long new;
5139 	int err;
5140 
5141 	err = kstrtoull(buf, 10, &new);
5142 	if (err < 0)
5143 		return err;
5144 	if (new != (sector_t)new)
5145 		return -EINVAL;
5146 
5147 	err = mddev_lock(mddev);
5148 	if (err)
5149 		return err;
5150 	err = -EINVAL;
5151 	if (mddev->pers == NULL ||
5152 	    mddev->pers->quiesce == NULL)
5153 		goto unlock;
5154 	mddev_suspend(mddev);
5155 	mddev->suspend_lo = new;
5156 	mddev_resume(mddev);
5157 
5158 	err = 0;
5159 unlock:
5160 	mddev_unlock(mddev);
5161 	return err ?: len;
5162 }
5163 static struct md_sysfs_entry md_suspend_lo =
5164 __ATTR(suspend_lo, S_IRUGO|S_IWUSR, suspend_lo_show, suspend_lo_store);
5165 
5166 static ssize_t
suspend_hi_show(struct mddev * mddev,char * page)5167 suspend_hi_show(struct mddev *mddev, char *page)
5168 {
5169 	return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_hi);
5170 }
5171 
5172 static ssize_t
suspend_hi_store(struct mddev * mddev,const char * buf,size_t len)5173 suspend_hi_store(struct mddev *mddev, const char *buf, size_t len)
5174 {
5175 	unsigned long long new;
5176 	int err;
5177 
5178 	err = kstrtoull(buf, 10, &new);
5179 	if (err < 0)
5180 		return err;
5181 	if (new != (sector_t)new)
5182 		return -EINVAL;
5183 
5184 	err = mddev_lock(mddev);
5185 	if (err)
5186 		return err;
5187 	err = -EINVAL;
5188 	if (mddev->pers == NULL)
5189 		goto unlock;
5190 
5191 	mddev_suspend(mddev);
5192 	mddev->suspend_hi = new;
5193 	mddev_resume(mddev);
5194 
5195 	err = 0;
5196 unlock:
5197 	mddev_unlock(mddev);
5198 	return err ?: len;
5199 }
5200 static struct md_sysfs_entry md_suspend_hi =
5201 __ATTR(suspend_hi, S_IRUGO|S_IWUSR, suspend_hi_show, suspend_hi_store);
5202 
5203 static ssize_t
reshape_position_show(struct mddev * mddev,char * page)5204 reshape_position_show(struct mddev *mddev, char *page)
5205 {
5206 	if (mddev->reshape_position != MaxSector)
5207 		return sprintf(page, "%llu\n",
5208 			       (unsigned long long)mddev->reshape_position);
5209 	strcpy(page, "none\n");
5210 	return 5;
5211 }
5212 
5213 static ssize_t
reshape_position_store(struct mddev * mddev,const char * buf,size_t len)5214 reshape_position_store(struct mddev *mddev, const char *buf, size_t len)
5215 {
5216 	struct md_rdev *rdev;
5217 	unsigned long long new;
5218 	int err;
5219 
5220 	err = kstrtoull(buf, 10, &new);
5221 	if (err < 0)
5222 		return err;
5223 	if (new != (sector_t)new)
5224 		return -EINVAL;
5225 	err = mddev_lock(mddev);
5226 	if (err)
5227 		return err;
5228 	err = -EBUSY;
5229 	if (mddev->pers)
5230 		goto unlock;
5231 	mddev->reshape_position = new;
5232 	mddev->delta_disks = 0;
5233 	mddev->reshape_backwards = 0;
5234 	mddev->new_level = mddev->level;
5235 	mddev->new_layout = mddev->layout;
5236 	mddev->new_chunk_sectors = mddev->chunk_sectors;
5237 	rdev_for_each(rdev, mddev)
5238 		rdev->new_data_offset = rdev->data_offset;
5239 	err = 0;
5240 unlock:
5241 	mddev_unlock(mddev);
5242 	return err ?: len;
5243 }
5244 
5245 static struct md_sysfs_entry md_reshape_position =
5246 __ATTR(reshape_position, S_IRUGO|S_IWUSR, reshape_position_show,
5247        reshape_position_store);
5248 
5249 static ssize_t
reshape_direction_show(struct mddev * mddev,char * page)5250 reshape_direction_show(struct mddev *mddev, char *page)
5251 {
5252 	return sprintf(page, "%s\n",
5253 		       mddev->reshape_backwards ? "backwards" : "forwards");
5254 }
5255 
5256 static ssize_t
reshape_direction_store(struct mddev * mddev,const char * buf,size_t len)5257 reshape_direction_store(struct mddev *mddev, const char *buf, size_t len)
5258 {
5259 	int backwards = 0;
5260 	int err;
5261 
5262 	if (cmd_match(buf, "forwards"))
5263 		backwards = 0;
5264 	else if (cmd_match(buf, "backwards"))
5265 		backwards = 1;
5266 	else
5267 		return -EINVAL;
5268 	if (mddev->reshape_backwards == backwards)
5269 		return len;
5270 
5271 	err = mddev_lock(mddev);
5272 	if (err)
5273 		return err;
5274 	/* check if we are allowed to change */
5275 	if (mddev->delta_disks)
5276 		err = -EBUSY;
5277 	else if (mddev->persistent &&
5278 	    mddev->major_version == 0)
5279 		err =  -EINVAL;
5280 	else
5281 		mddev->reshape_backwards = backwards;
5282 	mddev_unlock(mddev);
5283 	return err ?: len;
5284 }
5285 
5286 static struct md_sysfs_entry md_reshape_direction =
5287 __ATTR(reshape_direction, S_IRUGO|S_IWUSR, reshape_direction_show,
5288        reshape_direction_store);
5289 
5290 static ssize_t
array_size_show(struct mddev * mddev,char * page)5291 array_size_show(struct mddev *mddev, char *page)
5292 {
5293 	if (mddev->external_size)
5294 		return sprintf(page, "%llu\n",
5295 			       (unsigned long long)mddev->array_sectors/2);
5296 	else
5297 		return sprintf(page, "default\n");
5298 }
5299 
5300 static ssize_t
array_size_store(struct mddev * mddev,const char * buf,size_t len)5301 array_size_store(struct mddev *mddev, const char *buf, size_t len)
5302 {
5303 	sector_t sectors;
5304 	int err;
5305 
5306 	err = mddev_lock(mddev);
5307 	if (err)
5308 		return err;
5309 
5310 	/* cluster raid doesn't support change array_sectors */
5311 	if (mddev_is_clustered(mddev)) {
5312 		mddev_unlock(mddev);
5313 		return -EINVAL;
5314 	}
5315 
5316 	if (strncmp(buf, "default", 7) == 0) {
5317 		if (mddev->pers)
5318 			sectors = mddev->pers->size(mddev, 0, 0);
5319 		else
5320 			sectors = mddev->array_sectors;
5321 
5322 		mddev->external_size = 0;
5323 	} else {
5324 		if (strict_blocks_to_sectors(buf, &sectors) < 0)
5325 			err = -EINVAL;
5326 		else if (mddev->pers && mddev->pers->size(mddev, 0, 0) < sectors)
5327 			err = -E2BIG;
5328 		else
5329 			mddev->external_size = 1;
5330 	}
5331 
5332 	if (!err) {
5333 		mddev->array_sectors = sectors;
5334 		if (mddev->pers)
5335 			set_capacity_and_notify(mddev->gendisk,
5336 						mddev->array_sectors);
5337 	}
5338 	mddev_unlock(mddev);
5339 	return err ?: len;
5340 }
5341 
5342 static struct md_sysfs_entry md_array_size =
5343 __ATTR(array_size, S_IRUGO|S_IWUSR, array_size_show,
5344        array_size_store);
5345 
5346 static ssize_t
consistency_policy_show(struct mddev * mddev,char * page)5347 consistency_policy_show(struct mddev *mddev, char *page)
5348 {
5349 	int ret;
5350 
5351 	if (test_bit(MD_HAS_JOURNAL, &mddev->flags)) {
5352 		ret = sprintf(page, "journal\n");
5353 	} else if (test_bit(MD_HAS_PPL, &mddev->flags)) {
5354 		ret = sprintf(page, "ppl\n");
5355 	} else if (mddev->bitmap) {
5356 		ret = sprintf(page, "bitmap\n");
5357 	} else if (mddev->pers) {
5358 		if (mddev->pers->sync_request)
5359 			ret = sprintf(page, "resync\n");
5360 		else
5361 			ret = sprintf(page, "none\n");
5362 	} else {
5363 		ret = sprintf(page, "unknown\n");
5364 	}
5365 
5366 	return ret;
5367 }
5368 
5369 static ssize_t
consistency_policy_store(struct mddev * mddev,const char * buf,size_t len)5370 consistency_policy_store(struct mddev *mddev, const char *buf, size_t len)
5371 {
5372 	int err = 0;
5373 
5374 	if (mddev->pers) {
5375 		if (mddev->pers->change_consistency_policy)
5376 			err = mddev->pers->change_consistency_policy(mddev, buf);
5377 		else
5378 			err = -EBUSY;
5379 	} else if (mddev->external && strncmp(buf, "ppl", 3) == 0) {
5380 		set_bit(MD_HAS_PPL, &mddev->flags);
5381 	} else {
5382 		err = -EINVAL;
5383 	}
5384 
5385 	return err ? err : len;
5386 }
5387 
5388 static struct md_sysfs_entry md_consistency_policy =
5389 __ATTR(consistency_policy, S_IRUGO | S_IWUSR, consistency_policy_show,
5390        consistency_policy_store);
5391 
fail_last_dev_show(struct mddev * mddev,char * page)5392 static ssize_t fail_last_dev_show(struct mddev *mddev, char *page)
5393 {
5394 	return sprintf(page, "%d\n", mddev->fail_last_dev);
5395 }
5396 
5397 /*
5398  * Setting fail_last_dev to true to allow last device to be forcibly removed
5399  * from RAID1/RAID10.
5400  */
5401 static ssize_t
fail_last_dev_store(struct mddev * mddev,const char * buf,size_t len)5402 fail_last_dev_store(struct mddev *mddev, const char *buf, size_t len)
5403 {
5404 	int ret;
5405 	bool value;
5406 
5407 	ret = kstrtobool(buf, &value);
5408 	if (ret)
5409 		return ret;
5410 
5411 	if (value != mddev->fail_last_dev)
5412 		mddev->fail_last_dev = value;
5413 
5414 	return len;
5415 }
5416 static struct md_sysfs_entry md_fail_last_dev =
5417 __ATTR(fail_last_dev, S_IRUGO | S_IWUSR, fail_last_dev_show,
5418        fail_last_dev_store);
5419 
serialize_policy_show(struct mddev * mddev,char * page)5420 static ssize_t serialize_policy_show(struct mddev *mddev, char *page)
5421 {
5422 	if (mddev->pers == NULL || (mddev->pers->level != 1))
5423 		return sprintf(page, "n/a\n");
5424 	else
5425 		return sprintf(page, "%d\n", mddev->serialize_policy);
5426 }
5427 
5428 /*
5429  * Setting serialize_policy to true to enforce write IO is not reordered
5430  * for raid1.
5431  */
5432 static ssize_t
serialize_policy_store(struct mddev * mddev,const char * buf,size_t len)5433 serialize_policy_store(struct mddev *mddev, const char *buf, size_t len)
5434 {
5435 	int err;
5436 	bool value;
5437 
5438 	err = kstrtobool(buf, &value);
5439 	if (err)
5440 		return err;
5441 
5442 	if (value == mddev->serialize_policy)
5443 		return len;
5444 
5445 	err = mddev_lock(mddev);
5446 	if (err)
5447 		return err;
5448 	if (mddev->pers == NULL || (mddev->pers->level != 1)) {
5449 		pr_err("md: serialize_policy is only effective for raid1\n");
5450 		err = -EINVAL;
5451 		goto unlock;
5452 	}
5453 
5454 	mddev_suspend(mddev);
5455 	if (value)
5456 		mddev_create_serial_pool(mddev, NULL, true);
5457 	else
5458 		mddev_destroy_serial_pool(mddev, NULL, true);
5459 	mddev->serialize_policy = value;
5460 	mddev_resume(mddev);
5461 unlock:
5462 	mddev_unlock(mddev);
5463 	return err ?: len;
5464 }
5465 
5466 static struct md_sysfs_entry md_serialize_policy =
5467 __ATTR(serialize_policy, S_IRUGO | S_IWUSR, serialize_policy_show,
5468        serialize_policy_store);
5469 
5470 
5471 static struct attribute *md_default_attrs[] = {
5472 	&md_level.attr,
5473 	&md_layout.attr,
5474 	&md_raid_disks.attr,
5475 	&md_uuid.attr,
5476 	&md_chunk_size.attr,
5477 	&md_size.attr,
5478 	&md_resync_start.attr,
5479 	&md_metadata.attr,
5480 	&md_new_device.attr,
5481 	&md_safe_delay.attr,
5482 	&md_array_state.attr,
5483 	&md_reshape_position.attr,
5484 	&md_reshape_direction.attr,
5485 	&md_array_size.attr,
5486 	&max_corr_read_errors.attr,
5487 	&md_consistency_policy.attr,
5488 	&md_fail_last_dev.attr,
5489 	&md_serialize_policy.attr,
5490 	NULL,
5491 };
5492 
5493 static const struct attribute_group md_default_group = {
5494 	.attrs = md_default_attrs,
5495 };
5496 
5497 static struct attribute *md_redundancy_attrs[] = {
5498 	&md_scan_mode.attr,
5499 	&md_last_scan_mode.attr,
5500 	&md_mismatches.attr,
5501 	&md_sync_min.attr,
5502 	&md_sync_max.attr,
5503 	&md_sync_speed.attr,
5504 	&md_sync_force_parallel.attr,
5505 	&md_sync_completed.attr,
5506 	&md_min_sync.attr,
5507 	&md_max_sync.attr,
5508 	&md_suspend_lo.attr,
5509 	&md_suspend_hi.attr,
5510 	&md_bitmap.attr,
5511 	&md_degraded.attr,
5512 	NULL,
5513 };
5514 static const struct attribute_group md_redundancy_group = {
5515 	.name = NULL,
5516 	.attrs = md_redundancy_attrs,
5517 };
5518 
5519 static const struct attribute_group *md_attr_groups[] = {
5520 	&md_default_group,
5521 	&md_bitmap_group,
5522 	NULL,
5523 };
5524 
5525 static ssize_t
md_attr_show(struct kobject * kobj,struct attribute * attr,char * page)5526 md_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
5527 {
5528 	struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
5529 	struct mddev *mddev = container_of(kobj, struct mddev, kobj);
5530 	ssize_t rv;
5531 
5532 	if (!entry->show)
5533 		return -EIO;
5534 	spin_lock(&all_mddevs_lock);
5535 	if (list_empty(&mddev->all_mddevs)) {
5536 		spin_unlock(&all_mddevs_lock);
5537 		return -EBUSY;
5538 	}
5539 	mddev_get(mddev);
5540 	spin_unlock(&all_mddevs_lock);
5541 
5542 	rv = entry->show(mddev, page);
5543 	mddev_put(mddev);
5544 	return rv;
5545 }
5546 
5547 static ssize_t
md_attr_store(struct kobject * kobj,struct attribute * attr,const char * page,size_t length)5548 md_attr_store(struct kobject *kobj, struct attribute *attr,
5549 	      const char *page, size_t length)
5550 {
5551 	struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
5552 	struct mddev *mddev = container_of(kobj, struct mddev, kobj);
5553 	ssize_t rv;
5554 
5555 	if (!entry->store)
5556 		return -EIO;
5557 	if (!capable(CAP_SYS_ADMIN))
5558 		return -EACCES;
5559 	spin_lock(&all_mddevs_lock);
5560 	if (list_empty(&mddev->all_mddevs)) {
5561 		spin_unlock(&all_mddevs_lock);
5562 		return -EBUSY;
5563 	}
5564 	mddev_get(mddev);
5565 	spin_unlock(&all_mddevs_lock);
5566 	rv = entry->store(mddev, page, length);
5567 	mddev_put(mddev);
5568 	return rv;
5569 }
5570 
md_free(struct kobject * ko)5571 static void md_free(struct kobject *ko)
5572 {
5573 	struct mddev *mddev = container_of(ko, struct mddev, kobj);
5574 
5575 	if (mddev->sysfs_state)
5576 		sysfs_put(mddev->sysfs_state);
5577 	if (mddev->sysfs_level)
5578 		sysfs_put(mddev->sysfs_level);
5579 
5580 	if (mddev->gendisk) {
5581 		del_gendisk(mddev->gendisk);
5582 		blk_cleanup_disk(mddev->gendisk);
5583 	}
5584 	percpu_ref_exit(&mddev->writes_pending);
5585 
5586 	bioset_exit(&mddev->bio_set);
5587 	bioset_exit(&mddev->sync_set);
5588 	kfree(mddev);
5589 }
5590 
5591 static const struct sysfs_ops md_sysfs_ops = {
5592 	.show	= md_attr_show,
5593 	.store	= md_attr_store,
5594 };
5595 static struct kobj_type md_ktype = {
5596 	.release	= md_free,
5597 	.sysfs_ops	= &md_sysfs_ops,
5598 	.default_groups	= md_attr_groups,
5599 };
5600 
5601 int mdp_major = 0;
5602 
mddev_delayed_delete(struct work_struct * ws)5603 static void mddev_delayed_delete(struct work_struct *ws)
5604 {
5605 	struct mddev *mddev = container_of(ws, struct mddev, del_work);
5606 
5607 	kobject_del(&mddev->kobj);
5608 	kobject_put(&mddev->kobj);
5609 }
5610 
no_op(struct percpu_ref * r)5611 static void no_op(struct percpu_ref *r) {}
5612 
mddev_init_writes_pending(struct mddev * mddev)5613 int mddev_init_writes_pending(struct mddev *mddev)
5614 {
5615 	if (mddev->writes_pending.percpu_count_ptr)
5616 		return 0;
5617 	if (percpu_ref_init(&mddev->writes_pending, no_op,
5618 			    PERCPU_REF_ALLOW_REINIT, GFP_KERNEL) < 0)
5619 		return -ENOMEM;
5620 	/* We want to start with the refcount at zero */
5621 	percpu_ref_put(&mddev->writes_pending);
5622 	return 0;
5623 }
5624 EXPORT_SYMBOL_GPL(mddev_init_writes_pending);
5625 
md_alloc(dev_t dev,char * name)5626 static int md_alloc(dev_t dev, char *name)
5627 {
5628 	/*
5629 	 * If dev is zero, name is the name of a device to allocate with
5630 	 * an arbitrary minor number.  It will be "md_???"
5631 	 * If dev is non-zero it must be a device number with a MAJOR of
5632 	 * MD_MAJOR or mdp_major.  In this case, if "name" is NULL, then
5633 	 * the device is being created by opening a node in /dev.
5634 	 * If "name" is not NULL, the device is being created by
5635 	 * writing to /sys/module/md_mod/parameters/new_array.
5636 	 */
5637 	static DEFINE_MUTEX(disks_mutex);
5638 	struct mddev *mddev;
5639 	struct gendisk *disk;
5640 	int partitioned;
5641 	int shift;
5642 	int unit;
5643 	int error ;
5644 
5645 	/*
5646 	 * Wait for any previous instance of this device to be completely
5647 	 * removed (mddev_delayed_delete).
5648 	 */
5649 	flush_workqueue(md_misc_wq);
5650 	flush_workqueue(md_rdev_misc_wq);
5651 
5652 	mutex_lock(&disks_mutex);
5653 	mddev = mddev_alloc(dev);
5654 	if (IS_ERR(mddev)) {
5655 		mutex_unlock(&disks_mutex);
5656 		return PTR_ERR(mddev);
5657 	}
5658 
5659 	partitioned = (MAJOR(mddev->unit) != MD_MAJOR);
5660 	shift = partitioned ? MdpMinorShift : 0;
5661 	unit = MINOR(mddev->unit) >> shift;
5662 
5663 	if (name && !dev) {
5664 		/* Need to ensure that 'name' is not a duplicate.
5665 		 */
5666 		struct mddev *mddev2;
5667 		spin_lock(&all_mddevs_lock);
5668 
5669 		list_for_each_entry(mddev2, &all_mddevs, all_mddevs)
5670 			if (mddev2->gendisk &&
5671 			    strcmp(mddev2->gendisk->disk_name, name) == 0) {
5672 				spin_unlock(&all_mddevs_lock);
5673 				error = -EEXIST;
5674 				goto out_unlock_disks_mutex;
5675 			}
5676 		spin_unlock(&all_mddevs_lock);
5677 	}
5678 	if (name && dev)
5679 		/*
5680 		 * Creating /dev/mdNNN via "newarray", so adjust hold_active.
5681 		 */
5682 		mddev->hold_active = UNTIL_STOP;
5683 
5684 	error = -ENOMEM;
5685 	disk = blk_alloc_disk(NUMA_NO_NODE);
5686 	if (!disk)
5687 		goto out_unlock_disks_mutex;
5688 
5689 	disk->major = MAJOR(mddev->unit);
5690 	disk->first_minor = unit << shift;
5691 	disk->minors = 1 << shift;
5692 	if (name)
5693 		strcpy(disk->disk_name, name);
5694 	else if (partitioned)
5695 		sprintf(disk->disk_name, "md_d%d", unit);
5696 	else
5697 		sprintf(disk->disk_name, "md%d", unit);
5698 	disk->fops = &md_fops;
5699 	disk->private_data = mddev;
5700 
5701 	mddev->queue = disk->queue;
5702 	blk_set_stacking_limits(&mddev->queue->limits);
5703 	blk_queue_write_cache(mddev->queue, true, true);
5704 	disk->events |= DISK_EVENT_MEDIA_CHANGE;
5705 	mddev->gendisk = disk;
5706 	error = add_disk(disk);
5707 	if (error)
5708 		goto out_cleanup_disk;
5709 
5710 	error = kobject_add(&mddev->kobj, &disk_to_dev(disk)->kobj, "%s", "md");
5711 	if (error)
5712 		goto out_del_gendisk;
5713 
5714 	kobject_uevent(&mddev->kobj, KOBJ_ADD);
5715 	mddev->sysfs_state = sysfs_get_dirent_safe(mddev->kobj.sd, "array_state");
5716 	mddev->sysfs_level = sysfs_get_dirent_safe(mddev->kobj.sd, "level");
5717 	goto out_unlock_disks_mutex;
5718 
5719 out_del_gendisk:
5720 	del_gendisk(disk);
5721 out_cleanup_disk:
5722 	blk_cleanup_disk(disk);
5723 out_unlock_disks_mutex:
5724 	mutex_unlock(&disks_mutex);
5725 	mddev_put(mddev);
5726 	return error;
5727 }
5728 
md_probe(dev_t dev)5729 static void md_probe(dev_t dev)
5730 {
5731 	if (MAJOR(dev) == MD_MAJOR && MINOR(dev) >= 512)
5732 		return;
5733 	if (create_on_open)
5734 		md_alloc(dev, NULL);
5735 }
5736 
add_named_array(const char * val,const struct kernel_param * kp)5737 static int add_named_array(const char *val, const struct kernel_param *kp)
5738 {
5739 	/*
5740 	 * val must be "md_*" or "mdNNN".
5741 	 * For "md_*" we allocate an array with a large free minor number, and
5742 	 * set the name to val.  val must not already be an active name.
5743 	 * For "mdNNN" we allocate an array with the minor number NNN
5744 	 * which must not already be in use.
5745 	 */
5746 	int len = strlen(val);
5747 	char buf[DISK_NAME_LEN];
5748 	unsigned long devnum;
5749 
5750 	while (len && val[len-1] == '\n')
5751 		len--;
5752 	if (len >= DISK_NAME_LEN)
5753 		return -E2BIG;
5754 	strscpy(buf, val, len+1);
5755 	if (strncmp(buf, "md_", 3) == 0)
5756 		return md_alloc(0, buf);
5757 	if (strncmp(buf, "md", 2) == 0 &&
5758 	    isdigit(buf[2]) &&
5759 	    kstrtoul(buf+2, 10, &devnum) == 0 &&
5760 	    devnum <= MINORMASK)
5761 		return md_alloc(MKDEV(MD_MAJOR, devnum), NULL);
5762 
5763 	return -EINVAL;
5764 }
5765 
md_safemode_timeout(struct timer_list * t)5766 static void md_safemode_timeout(struct timer_list *t)
5767 {
5768 	struct mddev *mddev = from_timer(mddev, t, safemode_timer);
5769 
5770 	mddev->safemode = 1;
5771 	if (mddev->external)
5772 		sysfs_notify_dirent_safe(mddev->sysfs_state);
5773 
5774 	md_wakeup_thread(mddev->thread);
5775 }
5776 
5777 static int start_dirty_degraded;
5778 
md_run(struct mddev * mddev)5779 int md_run(struct mddev *mddev)
5780 {
5781 	int err;
5782 	struct md_rdev *rdev;
5783 	struct md_personality *pers;
5784 	bool nowait = true;
5785 
5786 	if (list_empty(&mddev->disks))
5787 		/* cannot run an array with no devices.. */
5788 		return -EINVAL;
5789 
5790 	if (mddev->pers)
5791 		return -EBUSY;
5792 	/* Cannot run until previous stop completes properly */
5793 	if (mddev->sysfs_active)
5794 		return -EBUSY;
5795 
5796 	/*
5797 	 * Analyze all RAID superblock(s)
5798 	 */
5799 	if (!mddev->raid_disks) {
5800 		if (!mddev->persistent)
5801 			return -EINVAL;
5802 		err = analyze_sbs(mddev);
5803 		if (err)
5804 			return -EINVAL;
5805 	}
5806 
5807 	if (mddev->level != LEVEL_NONE)
5808 		request_module("md-level-%d", mddev->level);
5809 	else if (mddev->clevel[0])
5810 		request_module("md-%s", mddev->clevel);
5811 
5812 	/*
5813 	 * Drop all container device buffers, from now on
5814 	 * the only valid external interface is through the md
5815 	 * device.
5816 	 */
5817 	mddev->has_superblocks = false;
5818 	rdev_for_each(rdev, mddev) {
5819 		if (test_bit(Faulty, &rdev->flags))
5820 			continue;
5821 		sync_blockdev(rdev->bdev);
5822 		invalidate_bdev(rdev->bdev);
5823 		if (mddev->ro != 1 && rdev_read_only(rdev)) {
5824 			mddev->ro = 1;
5825 			if (mddev->gendisk)
5826 				set_disk_ro(mddev->gendisk, 1);
5827 		}
5828 
5829 		if (rdev->sb_page)
5830 			mddev->has_superblocks = true;
5831 
5832 		/* perform some consistency tests on the device.
5833 		 * We don't want the data to overlap the metadata,
5834 		 * Internal Bitmap issues have been handled elsewhere.
5835 		 */
5836 		if (rdev->meta_bdev) {
5837 			/* Nothing to check */;
5838 		} else if (rdev->data_offset < rdev->sb_start) {
5839 			if (mddev->dev_sectors &&
5840 			    rdev->data_offset + mddev->dev_sectors
5841 			    > rdev->sb_start) {
5842 				pr_warn("md: %s: data overlaps metadata\n",
5843 					mdname(mddev));
5844 				return -EINVAL;
5845 			}
5846 		} else {
5847 			if (rdev->sb_start + rdev->sb_size/512
5848 			    > rdev->data_offset) {
5849 				pr_warn("md: %s: metadata overlaps data\n",
5850 					mdname(mddev));
5851 				return -EINVAL;
5852 			}
5853 		}
5854 		sysfs_notify_dirent_safe(rdev->sysfs_state);
5855 		nowait = nowait && blk_queue_nowait(bdev_get_queue(rdev->bdev));
5856 	}
5857 
5858 	if (!bioset_initialized(&mddev->bio_set)) {
5859 		err = bioset_init(&mddev->bio_set, BIO_POOL_SIZE, 0, BIOSET_NEED_BVECS);
5860 		if (err)
5861 			return err;
5862 	}
5863 	if (!bioset_initialized(&mddev->sync_set)) {
5864 		err = bioset_init(&mddev->sync_set, BIO_POOL_SIZE, 0, BIOSET_NEED_BVECS);
5865 		if (err)
5866 			goto exit_bio_set;
5867 	}
5868 
5869 	spin_lock(&pers_lock);
5870 	pers = find_pers(mddev->level, mddev->clevel);
5871 	if (!pers || !try_module_get(pers->owner)) {
5872 		spin_unlock(&pers_lock);
5873 		if (mddev->level != LEVEL_NONE)
5874 			pr_warn("md: personality for level %d is not loaded!\n",
5875 				mddev->level);
5876 		else
5877 			pr_warn("md: personality for level %s is not loaded!\n",
5878 				mddev->clevel);
5879 		err = -EINVAL;
5880 		goto abort;
5881 	}
5882 	spin_unlock(&pers_lock);
5883 	if (mddev->level != pers->level) {
5884 		mddev->level = pers->level;
5885 		mddev->new_level = pers->level;
5886 	}
5887 	strscpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
5888 
5889 	if (mddev->reshape_position != MaxSector &&
5890 	    pers->start_reshape == NULL) {
5891 		/* This personality cannot handle reshaping... */
5892 		module_put(pers->owner);
5893 		err = -EINVAL;
5894 		goto abort;
5895 	}
5896 
5897 	if (pers->sync_request) {
5898 		/* Warn if this is a potentially silly
5899 		 * configuration.
5900 		 */
5901 		struct md_rdev *rdev2;
5902 		int warned = 0;
5903 
5904 		rdev_for_each(rdev, mddev)
5905 			rdev_for_each(rdev2, mddev) {
5906 				if (rdev < rdev2 &&
5907 				    rdev->bdev->bd_disk ==
5908 				    rdev2->bdev->bd_disk) {
5909 					pr_warn("%s: WARNING: %pg appears to be on the same physical disk as %pg.\n",
5910 						mdname(mddev),
5911 						rdev->bdev,
5912 						rdev2->bdev);
5913 					warned = 1;
5914 				}
5915 			}
5916 
5917 		if (warned)
5918 			pr_warn("True protection against single-disk failure might be compromised.\n");
5919 	}
5920 
5921 	mddev->recovery = 0;
5922 	/* may be over-ridden by personality */
5923 	mddev->resync_max_sectors = mddev->dev_sectors;
5924 
5925 	mddev->ok_start_degraded = start_dirty_degraded;
5926 
5927 	if (start_readonly && mddev->ro == 0)
5928 		mddev->ro = 2; /* read-only, but switch on first write */
5929 
5930 	err = pers->run(mddev);
5931 	if (err)
5932 		pr_warn("md: pers->run() failed ...\n");
5933 	else if (pers->size(mddev, 0, 0) < mddev->array_sectors) {
5934 		WARN_ONCE(!mddev->external_size,
5935 			  "%s: default size too small, but 'external_size' not in effect?\n",
5936 			  __func__);
5937 		pr_warn("md: invalid array_size %llu > default size %llu\n",
5938 			(unsigned long long)mddev->array_sectors / 2,
5939 			(unsigned long long)pers->size(mddev, 0, 0) / 2);
5940 		err = -EINVAL;
5941 	}
5942 	if (err == 0 && pers->sync_request &&
5943 	    (mddev->bitmap_info.file || mddev->bitmap_info.offset)) {
5944 		struct bitmap *bitmap;
5945 
5946 		bitmap = md_bitmap_create(mddev, -1);
5947 		if (IS_ERR(bitmap)) {
5948 			err = PTR_ERR(bitmap);
5949 			pr_warn("%s: failed to create bitmap (%d)\n",
5950 				mdname(mddev), err);
5951 		} else
5952 			mddev->bitmap = bitmap;
5953 
5954 	}
5955 	if (err)
5956 		goto bitmap_abort;
5957 
5958 	if (mddev->bitmap_info.max_write_behind > 0) {
5959 		bool create_pool = false;
5960 
5961 		rdev_for_each(rdev, mddev) {
5962 			if (test_bit(WriteMostly, &rdev->flags) &&
5963 			    rdev_init_serial(rdev))
5964 				create_pool = true;
5965 		}
5966 		if (create_pool && mddev->serial_info_pool == NULL) {
5967 			mddev->serial_info_pool =
5968 				mempool_create_kmalloc_pool(NR_SERIAL_INFOS,
5969 						    sizeof(struct serial_info));
5970 			if (!mddev->serial_info_pool) {
5971 				err = -ENOMEM;
5972 				goto bitmap_abort;
5973 			}
5974 		}
5975 	}
5976 
5977 	if (mddev->queue) {
5978 		bool nonrot = true;
5979 
5980 		rdev_for_each(rdev, mddev) {
5981 			if (rdev->raid_disk >= 0 && !bdev_nonrot(rdev->bdev)) {
5982 				nonrot = false;
5983 				break;
5984 			}
5985 		}
5986 		if (mddev->degraded)
5987 			nonrot = false;
5988 		if (nonrot)
5989 			blk_queue_flag_set(QUEUE_FLAG_NONROT, mddev->queue);
5990 		else
5991 			blk_queue_flag_clear(QUEUE_FLAG_NONROT, mddev->queue);
5992 		blk_queue_flag_set(QUEUE_FLAG_IO_STAT, mddev->queue);
5993 
5994 		/* Set the NOWAIT flags if all underlying devices support it */
5995 		if (nowait)
5996 			blk_queue_flag_set(QUEUE_FLAG_NOWAIT, mddev->queue);
5997 	}
5998 	if (pers->sync_request) {
5999 		if (mddev->kobj.sd &&
6000 		    sysfs_create_group(&mddev->kobj, &md_redundancy_group))
6001 			pr_warn("md: cannot register extra attributes for %s\n",
6002 				mdname(mddev));
6003 		mddev->sysfs_action = sysfs_get_dirent_safe(mddev->kobj.sd, "sync_action");
6004 		mddev->sysfs_completed = sysfs_get_dirent_safe(mddev->kobj.sd, "sync_completed");
6005 		mddev->sysfs_degraded = sysfs_get_dirent_safe(mddev->kobj.sd, "degraded");
6006 	} else if (mddev->ro == 2) /* auto-readonly not meaningful */
6007 		mddev->ro = 0;
6008 
6009 	atomic_set(&mddev->max_corr_read_errors,
6010 		   MD_DEFAULT_MAX_CORRECTED_READ_ERRORS);
6011 	mddev->safemode = 0;
6012 	if (mddev_is_clustered(mddev))
6013 		mddev->safemode_delay = 0;
6014 	else
6015 		mddev->safemode_delay = DEFAULT_SAFEMODE_DELAY;
6016 	mddev->in_sync = 1;
6017 	smp_wmb();
6018 	spin_lock(&mddev->lock);
6019 	mddev->pers = pers;
6020 	spin_unlock(&mddev->lock);
6021 	rdev_for_each(rdev, mddev)
6022 		if (rdev->raid_disk >= 0)
6023 			sysfs_link_rdev(mddev, rdev); /* failure here is OK */
6024 
6025 	if (mddev->degraded && !mddev->ro)
6026 		/* This ensures that recovering status is reported immediately
6027 		 * via sysfs - until a lack of spares is confirmed.
6028 		 */
6029 		set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
6030 	set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6031 
6032 	if (mddev->sb_flags)
6033 		md_update_sb(mddev, 0);
6034 
6035 	md_new_event();
6036 	return 0;
6037 
6038 bitmap_abort:
6039 	mddev_detach(mddev);
6040 	if (mddev->private)
6041 		pers->free(mddev, mddev->private);
6042 	mddev->private = NULL;
6043 	module_put(pers->owner);
6044 	md_bitmap_destroy(mddev);
6045 abort:
6046 	bioset_exit(&mddev->sync_set);
6047 exit_bio_set:
6048 	bioset_exit(&mddev->bio_set);
6049 	return err;
6050 }
6051 EXPORT_SYMBOL_GPL(md_run);
6052 
do_md_run(struct mddev * mddev)6053 int do_md_run(struct mddev *mddev)
6054 {
6055 	int err;
6056 
6057 	set_bit(MD_NOT_READY, &mddev->flags);
6058 	err = md_run(mddev);
6059 	if (err)
6060 		goto out;
6061 	err = md_bitmap_load(mddev);
6062 	if (err) {
6063 		md_bitmap_destroy(mddev);
6064 		goto out;
6065 	}
6066 
6067 	if (mddev_is_clustered(mddev))
6068 		md_allow_write(mddev);
6069 
6070 	/* run start up tasks that require md_thread */
6071 	md_start(mddev);
6072 
6073 	md_wakeup_thread(mddev->thread);
6074 	md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
6075 
6076 	set_capacity_and_notify(mddev->gendisk, mddev->array_sectors);
6077 	clear_bit(MD_NOT_READY, &mddev->flags);
6078 	mddev->changed = 1;
6079 	kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
6080 	sysfs_notify_dirent_safe(mddev->sysfs_state);
6081 	sysfs_notify_dirent_safe(mddev->sysfs_action);
6082 	sysfs_notify_dirent_safe(mddev->sysfs_degraded);
6083 out:
6084 	clear_bit(MD_NOT_READY, &mddev->flags);
6085 	return err;
6086 }
6087 
md_start(struct mddev * mddev)6088 int md_start(struct mddev *mddev)
6089 {
6090 	int ret = 0;
6091 
6092 	if (mddev->pers->start) {
6093 		set_bit(MD_RECOVERY_WAIT, &mddev->recovery);
6094 		md_wakeup_thread(mddev->thread);
6095 		ret = mddev->pers->start(mddev);
6096 		clear_bit(MD_RECOVERY_WAIT, &mddev->recovery);
6097 		md_wakeup_thread(mddev->sync_thread);
6098 	}
6099 	return ret;
6100 }
6101 EXPORT_SYMBOL_GPL(md_start);
6102 
restart_array(struct mddev * mddev)6103 static int restart_array(struct mddev *mddev)
6104 {
6105 	struct gendisk *disk = mddev->gendisk;
6106 	struct md_rdev *rdev;
6107 	bool has_journal = false;
6108 	bool has_readonly = false;
6109 
6110 	/* Complain if it has no devices */
6111 	if (list_empty(&mddev->disks))
6112 		return -ENXIO;
6113 	if (!mddev->pers)
6114 		return -EINVAL;
6115 	if (!mddev->ro)
6116 		return -EBUSY;
6117 
6118 	rcu_read_lock();
6119 	rdev_for_each_rcu(rdev, mddev) {
6120 		if (test_bit(Journal, &rdev->flags) &&
6121 		    !test_bit(Faulty, &rdev->flags))
6122 			has_journal = true;
6123 		if (rdev_read_only(rdev))
6124 			has_readonly = true;
6125 	}
6126 	rcu_read_unlock();
6127 	if (test_bit(MD_HAS_JOURNAL, &mddev->flags) && !has_journal)
6128 		/* Don't restart rw with journal missing/faulty */
6129 			return -EINVAL;
6130 	if (has_readonly)
6131 		return -EROFS;
6132 
6133 	mddev->safemode = 0;
6134 	mddev->ro = 0;
6135 	set_disk_ro(disk, 0);
6136 	pr_debug("md: %s switched to read-write mode.\n", mdname(mddev));
6137 	/* Kick recovery or resync if necessary */
6138 	set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6139 	md_wakeup_thread(mddev->thread);
6140 	md_wakeup_thread(mddev->sync_thread);
6141 	sysfs_notify_dirent_safe(mddev->sysfs_state);
6142 	return 0;
6143 }
6144 
md_clean(struct mddev * mddev)6145 static void md_clean(struct mddev *mddev)
6146 {
6147 	mddev->array_sectors = 0;
6148 	mddev->external_size = 0;
6149 	mddev->dev_sectors = 0;
6150 	mddev->raid_disks = 0;
6151 	mddev->recovery_cp = 0;
6152 	mddev->resync_min = 0;
6153 	mddev->resync_max = MaxSector;
6154 	mddev->reshape_position = MaxSector;
6155 	mddev->external = 0;
6156 	mddev->persistent = 0;
6157 	mddev->level = LEVEL_NONE;
6158 	mddev->clevel[0] = 0;
6159 	mddev->flags = 0;
6160 	mddev->sb_flags = 0;
6161 	mddev->ro = 0;
6162 	mddev->metadata_type[0] = 0;
6163 	mddev->chunk_sectors = 0;
6164 	mddev->ctime = mddev->utime = 0;
6165 	mddev->layout = 0;
6166 	mddev->max_disks = 0;
6167 	mddev->events = 0;
6168 	mddev->can_decrease_events = 0;
6169 	mddev->delta_disks = 0;
6170 	mddev->reshape_backwards = 0;
6171 	mddev->new_level = LEVEL_NONE;
6172 	mddev->new_layout = 0;
6173 	mddev->new_chunk_sectors = 0;
6174 	mddev->curr_resync = 0;
6175 	atomic64_set(&mddev->resync_mismatches, 0);
6176 	mddev->suspend_lo = mddev->suspend_hi = 0;
6177 	mddev->sync_speed_min = mddev->sync_speed_max = 0;
6178 	mddev->recovery = 0;
6179 	mddev->in_sync = 0;
6180 	mddev->changed = 0;
6181 	mddev->degraded = 0;
6182 	mddev->safemode = 0;
6183 	mddev->private = NULL;
6184 	mddev->cluster_info = NULL;
6185 	mddev->bitmap_info.offset = 0;
6186 	mddev->bitmap_info.default_offset = 0;
6187 	mddev->bitmap_info.default_space = 0;
6188 	mddev->bitmap_info.chunksize = 0;
6189 	mddev->bitmap_info.daemon_sleep = 0;
6190 	mddev->bitmap_info.max_write_behind = 0;
6191 	mddev->bitmap_info.nodes = 0;
6192 }
6193 
__md_stop_writes(struct mddev * mddev)6194 static void __md_stop_writes(struct mddev *mddev)
6195 {
6196 	set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
6197 	if (work_pending(&mddev->del_work))
6198 		flush_workqueue(md_misc_wq);
6199 	if (mddev->sync_thread) {
6200 		set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6201 		md_reap_sync_thread(mddev);
6202 	}
6203 
6204 	del_timer_sync(&mddev->safemode_timer);
6205 
6206 	if (mddev->pers && mddev->pers->quiesce) {
6207 		mddev->pers->quiesce(mddev, 1);
6208 		mddev->pers->quiesce(mddev, 0);
6209 	}
6210 	md_bitmap_flush(mddev);
6211 
6212 	if (mddev->ro == 0 &&
6213 	    ((!mddev->in_sync && !mddev_is_clustered(mddev)) ||
6214 	     mddev->sb_flags)) {
6215 		/* mark array as shutdown cleanly */
6216 		if (!mddev_is_clustered(mddev))
6217 			mddev->in_sync = 1;
6218 		md_update_sb(mddev, 1);
6219 	}
6220 	/* disable policy to guarantee rdevs free resources for serialization */
6221 	mddev->serialize_policy = 0;
6222 	mddev_destroy_serial_pool(mddev, NULL, true);
6223 }
6224 
md_stop_writes(struct mddev * mddev)6225 void md_stop_writes(struct mddev *mddev)
6226 {
6227 	mddev_lock_nointr(mddev);
6228 	__md_stop_writes(mddev);
6229 	mddev_unlock(mddev);
6230 }
6231 EXPORT_SYMBOL_GPL(md_stop_writes);
6232 
mddev_detach(struct mddev * mddev)6233 static void mddev_detach(struct mddev *mddev)
6234 {
6235 	md_bitmap_wait_behind_writes(mddev);
6236 	if (mddev->pers && mddev->pers->quiesce && !mddev->suspended) {
6237 		mddev->pers->quiesce(mddev, 1);
6238 		mddev->pers->quiesce(mddev, 0);
6239 	}
6240 	md_unregister_thread(&mddev->thread);
6241 	if (mddev->queue)
6242 		blk_sync_queue(mddev->queue); /* the unplug fn references 'conf'*/
6243 }
6244 
__md_stop(struct mddev * mddev)6245 static void __md_stop(struct mddev *mddev)
6246 {
6247 	struct md_personality *pers = mddev->pers;
6248 	md_bitmap_destroy(mddev);
6249 	mddev_detach(mddev);
6250 	/* Ensure ->event_work is done */
6251 	if (mddev->event_work.func)
6252 		flush_workqueue(md_misc_wq);
6253 	spin_lock(&mddev->lock);
6254 	mddev->pers = NULL;
6255 	spin_unlock(&mddev->lock);
6256 	if (mddev->private)
6257 		pers->free(mddev, mddev->private);
6258 	mddev->private = NULL;
6259 	if (pers->sync_request && mddev->to_remove == NULL)
6260 		mddev->to_remove = &md_redundancy_group;
6261 	module_put(pers->owner);
6262 	clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
6263 }
6264 
md_stop(struct mddev * mddev)6265 void md_stop(struct mddev *mddev)
6266 {
6267 	/* stop the array and free an attached data structures.
6268 	 * This is called from dm-raid
6269 	 */
6270 	__md_stop_writes(mddev);
6271 	__md_stop(mddev);
6272 	bioset_exit(&mddev->bio_set);
6273 	bioset_exit(&mddev->sync_set);
6274 }
6275 
6276 EXPORT_SYMBOL_GPL(md_stop);
6277 
md_set_readonly(struct mddev * mddev,struct block_device * bdev)6278 static int md_set_readonly(struct mddev *mddev, struct block_device *bdev)
6279 {
6280 	int err = 0;
6281 	int did_freeze = 0;
6282 
6283 	if (!test_bit(MD_RECOVERY_FROZEN, &mddev->recovery)) {
6284 		did_freeze = 1;
6285 		set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
6286 		md_wakeup_thread(mddev->thread);
6287 	}
6288 	if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
6289 		set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6290 	if (mddev->sync_thread)
6291 		/* Thread might be blocked waiting for metadata update
6292 		 * which will now never happen */
6293 		wake_up_process(mddev->sync_thread->tsk);
6294 
6295 	if (mddev->external && test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags))
6296 		return -EBUSY;
6297 	mddev_unlock(mddev);
6298 	wait_event(resync_wait, !test_bit(MD_RECOVERY_RUNNING,
6299 					  &mddev->recovery));
6300 	wait_event(mddev->sb_wait,
6301 		   !test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags));
6302 	mddev_lock_nointr(mddev);
6303 
6304 	mutex_lock(&mddev->open_mutex);
6305 	if ((mddev->pers && atomic_read(&mddev->openers) > !!bdev) ||
6306 	    mddev->sync_thread ||
6307 	    test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) {
6308 		pr_warn("md: %s still in use.\n",mdname(mddev));
6309 		if (did_freeze) {
6310 			clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
6311 			set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6312 			md_wakeup_thread(mddev->thread);
6313 		}
6314 		err = -EBUSY;
6315 		goto out;
6316 	}
6317 	if (mddev->pers) {
6318 		__md_stop_writes(mddev);
6319 
6320 		err  = -ENXIO;
6321 		if (mddev->ro==1)
6322 			goto out;
6323 		mddev->ro = 1;
6324 		set_disk_ro(mddev->gendisk, 1);
6325 		clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
6326 		set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6327 		md_wakeup_thread(mddev->thread);
6328 		sysfs_notify_dirent_safe(mddev->sysfs_state);
6329 		err = 0;
6330 	}
6331 out:
6332 	mutex_unlock(&mddev->open_mutex);
6333 	return err;
6334 }
6335 
6336 /* mode:
6337  *   0 - completely stop and dis-assemble array
6338  *   2 - stop but do not disassemble array
6339  */
do_md_stop(struct mddev * mddev,int mode,struct block_device * bdev)6340 static int do_md_stop(struct mddev *mddev, int mode,
6341 		      struct block_device *bdev)
6342 {
6343 	struct gendisk *disk = mddev->gendisk;
6344 	struct md_rdev *rdev;
6345 	int did_freeze = 0;
6346 
6347 	if (!test_bit(MD_RECOVERY_FROZEN, &mddev->recovery)) {
6348 		did_freeze = 1;
6349 		set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
6350 		md_wakeup_thread(mddev->thread);
6351 	}
6352 	if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
6353 		set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6354 	if (mddev->sync_thread)
6355 		/* Thread might be blocked waiting for metadata update
6356 		 * which will now never happen */
6357 		wake_up_process(mddev->sync_thread->tsk);
6358 
6359 	mddev_unlock(mddev);
6360 	wait_event(resync_wait, (mddev->sync_thread == NULL &&
6361 				 !test_bit(MD_RECOVERY_RUNNING,
6362 					   &mddev->recovery)));
6363 	mddev_lock_nointr(mddev);
6364 
6365 	mutex_lock(&mddev->open_mutex);
6366 	if ((mddev->pers && atomic_read(&mddev->openers) > !!bdev) ||
6367 	    mddev->sysfs_active ||
6368 	    mddev->sync_thread ||
6369 	    test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) {
6370 		pr_warn("md: %s still in use.\n",mdname(mddev));
6371 		mutex_unlock(&mddev->open_mutex);
6372 		if (did_freeze) {
6373 			clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
6374 			set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6375 			md_wakeup_thread(mddev->thread);
6376 		}
6377 		return -EBUSY;
6378 	}
6379 	if (mddev->pers) {
6380 		if (mddev->ro)
6381 			set_disk_ro(disk, 0);
6382 
6383 		__md_stop_writes(mddev);
6384 		__md_stop(mddev);
6385 
6386 		/* tell userspace to handle 'inactive' */
6387 		sysfs_notify_dirent_safe(mddev->sysfs_state);
6388 
6389 		rdev_for_each(rdev, mddev)
6390 			if (rdev->raid_disk >= 0)
6391 				sysfs_unlink_rdev(mddev, rdev);
6392 
6393 		set_capacity_and_notify(disk, 0);
6394 		mutex_unlock(&mddev->open_mutex);
6395 		mddev->changed = 1;
6396 
6397 		if (mddev->ro)
6398 			mddev->ro = 0;
6399 	} else
6400 		mutex_unlock(&mddev->open_mutex);
6401 	/*
6402 	 * Free resources if final stop
6403 	 */
6404 	if (mode == 0) {
6405 		pr_info("md: %s stopped.\n", mdname(mddev));
6406 
6407 		if (mddev->bitmap_info.file) {
6408 			struct file *f = mddev->bitmap_info.file;
6409 			spin_lock(&mddev->lock);
6410 			mddev->bitmap_info.file = NULL;
6411 			spin_unlock(&mddev->lock);
6412 			fput(f);
6413 		}
6414 		mddev->bitmap_info.offset = 0;
6415 
6416 		export_array(mddev);
6417 
6418 		md_clean(mddev);
6419 		if (mddev->hold_active == UNTIL_STOP)
6420 			mddev->hold_active = 0;
6421 	}
6422 	md_new_event();
6423 	sysfs_notify_dirent_safe(mddev->sysfs_state);
6424 	return 0;
6425 }
6426 
6427 #ifndef MODULE
autorun_array(struct mddev * mddev)6428 static void autorun_array(struct mddev *mddev)
6429 {
6430 	struct md_rdev *rdev;
6431 	int err;
6432 
6433 	if (list_empty(&mddev->disks))
6434 		return;
6435 
6436 	pr_info("md: running: ");
6437 
6438 	rdev_for_each(rdev, mddev) {
6439 		pr_cont("<%pg>", rdev->bdev);
6440 	}
6441 	pr_cont("\n");
6442 
6443 	err = do_md_run(mddev);
6444 	if (err) {
6445 		pr_warn("md: do_md_run() returned %d\n", err);
6446 		do_md_stop(mddev, 0, NULL);
6447 	}
6448 }
6449 
6450 /*
6451  * lets try to run arrays based on all disks that have arrived
6452  * until now. (those are in pending_raid_disks)
6453  *
6454  * the method: pick the first pending disk, collect all disks with
6455  * the same UUID, remove all from the pending list and put them into
6456  * the 'same_array' list. Then order this list based on superblock
6457  * update time (freshest comes first), kick out 'old' disks and
6458  * compare superblocks. If everything's fine then run it.
6459  *
6460  * If "unit" is allocated, then bump its reference count
6461  */
autorun_devices(int part)6462 static void autorun_devices(int part)
6463 {
6464 	struct md_rdev *rdev0, *rdev, *tmp;
6465 	struct mddev *mddev;
6466 
6467 	pr_info("md: autorun ...\n");
6468 	while (!list_empty(&pending_raid_disks)) {
6469 		int unit;
6470 		dev_t dev;
6471 		LIST_HEAD(candidates);
6472 		rdev0 = list_entry(pending_raid_disks.next,
6473 					 struct md_rdev, same_set);
6474 
6475 		pr_debug("md: considering %pg ...\n", rdev0->bdev);
6476 		INIT_LIST_HEAD(&candidates);
6477 		rdev_for_each_list(rdev, tmp, &pending_raid_disks)
6478 			if (super_90_load(rdev, rdev0, 0) >= 0) {
6479 				pr_debug("md:  adding %pg ...\n",
6480 					 rdev->bdev);
6481 				list_move(&rdev->same_set, &candidates);
6482 			}
6483 		/*
6484 		 * now we have a set of devices, with all of them having
6485 		 * mostly sane superblocks. It's time to allocate the
6486 		 * mddev.
6487 		 */
6488 		if (part) {
6489 			dev = MKDEV(mdp_major,
6490 				    rdev0->preferred_minor << MdpMinorShift);
6491 			unit = MINOR(dev) >> MdpMinorShift;
6492 		} else {
6493 			dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
6494 			unit = MINOR(dev);
6495 		}
6496 		if (rdev0->preferred_minor != unit) {
6497 			pr_warn("md: unit number in %pg is bad: %d\n",
6498 				rdev0->bdev, rdev0->preferred_minor);
6499 			break;
6500 		}
6501 
6502 		md_probe(dev);
6503 		mddev = mddev_find(dev);
6504 		if (!mddev)
6505 			break;
6506 
6507 		if (mddev_lock(mddev))
6508 			pr_warn("md: %s locked, cannot run\n", mdname(mddev));
6509 		else if (mddev->raid_disks || mddev->major_version
6510 			 || !list_empty(&mddev->disks)) {
6511 			pr_warn("md: %s already running, cannot run %pg\n",
6512 				mdname(mddev), rdev0->bdev);
6513 			mddev_unlock(mddev);
6514 		} else {
6515 			pr_debug("md: created %s\n", mdname(mddev));
6516 			mddev->persistent = 1;
6517 			rdev_for_each_list(rdev, tmp, &candidates) {
6518 				list_del_init(&rdev->same_set);
6519 				if (bind_rdev_to_array(rdev, mddev))
6520 					export_rdev(rdev);
6521 			}
6522 			autorun_array(mddev);
6523 			mddev_unlock(mddev);
6524 		}
6525 		/* on success, candidates will be empty, on error
6526 		 * it won't...
6527 		 */
6528 		rdev_for_each_list(rdev, tmp, &candidates) {
6529 			list_del_init(&rdev->same_set);
6530 			export_rdev(rdev);
6531 		}
6532 		mddev_put(mddev);
6533 	}
6534 	pr_info("md: ... autorun DONE.\n");
6535 }
6536 #endif /* !MODULE */
6537 
get_version(void __user * arg)6538 static int get_version(void __user *arg)
6539 {
6540 	mdu_version_t ver;
6541 
6542 	ver.major = MD_MAJOR_VERSION;
6543 	ver.minor = MD_MINOR_VERSION;
6544 	ver.patchlevel = MD_PATCHLEVEL_VERSION;
6545 
6546 	if (copy_to_user(arg, &ver, sizeof(ver)))
6547 		return -EFAULT;
6548 
6549 	return 0;
6550 }
6551 
get_array_info(struct mddev * mddev,void __user * arg)6552 static int get_array_info(struct mddev *mddev, void __user *arg)
6553 {
6554 	mdu_array_info_t info;
6555 	int nr,working,insync,failed,spare;
6556 	struct md_rdev *rdev;
6557 
6558 	nr = working = insync = failed = spare = 0;
6559 	rcu_read_lock();
6560 	rdev_for_each_rcu(rdev, mddev) {
6561 		nr++;
6562 		if (test_bit(Faulty, &rdev->flags))
6563 			failed++;
6564 		else {
6565 			working++;
6566 			if (test_bit(In_sync, &rdev->flags))
6567 				insync++;
6568 			else if (test_bit(Journal, &rdev->flags))
6569 				/* TODO: add journal count to md_u.h */
6570 				;
6571 			else
6572 				spare++;
6573 		}
6574 	}
6575 	rcu_read_unlock();
6576 
6577 	info.major_version = mddev->major_version;
6578 	info.minor_version = mddev->minor_version;
6579 	info.patch_version = MD_PATCHLEVEL_VERSION;
6580 	info.ctime         = clamp_t(time64_t, mddev->ctime, 0, U32_MAX);
6581 	info.level         = mddev->level;
6582 	info.size          = mddev->dev_sectors / 2;
6583 	if (info.size != mddev->dev_sectors / 2) /* overflow */
6584 		info.size = -1;
6585 	info.nr_disks      = nr;
6586 	info.raid_disks    = mddev->raid_disks;
6587 	info.md_minor      = mddev->md_minor;
6588 	info.not_persistent= !mddev->persistent;
6589 
6590 	info.utime         = clamp_t(time64_t, mddev->utime, 0, U32_MAX);
6591 	info.state         = 0;
6592 	if (mddev->in_sync)
6593 		info.state = (1<<MD_SB_CLEAN);
6594 	if (mddev->bitmap && mddev->bitmap_info.offset)
6595 		info.state |= (1<<MD_SB_BITMAP_PRESENT);
6596 	if (mddev_is_clustered(mddev))
6597 		info.state |= (1<<MD_SB_CLUSTERED);
6598 	info.active_disks  = insync;
6599 	info.working_disks = working;
6600 	info.failed_disks  = failed;
6601 	info.spare_disks   = spare;
6602 
6603 	info.layout        = mddev->layout;
6604 	info.chunk_size    = mddev->chunk_sectors << 9;
6605 
6606 	if (copy_to_user(arg, &info, sizeof(info)))
6607 		return -EFAULT;
6608 
6609 	return 0;
6610 }
6611 
get_bitmap_file(struct mddev * mddev,void __user * arg)6612 static int get_bitmap_file(struct mddev *mddev, void __user * arg)
6613 {
6614 	mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */
6615 	char *ptr;
6616 	int err;
6617 
6618 	file = kzalloc(sizeof(*file), GFP_NOIO);
6619 	if (!file)
6620 		return -ENOMEM;
6621 
6622 	err = 0;
6623 	spin_lock(&mddev->lock);
6624 	/* bitmap enabled */
6625 	if (mddev->bitmap_info.file) {
6626 		ptr = file_path(mddev->bitmap_info.file, file->pathname,
6627 				sizeof(file->pathname));
6628 		if (IS_ERR(ptr))
6629 			err = PTR_ERR(ptr);
6630 		else
6631 			memmove(file->pathname, ptr,
6632 				sizeof(file->pathname)-(ptr-file->pathname));
6633 	}
6634 	spin_unlock(&mddev->lock);
6635 
6636 	if (err == 0 &&
6637 	    copy_to_user(arg, file, sizeof(*file)))
6638 		err = -EFAULT;
6639 
6640 	kfree(file);
6641 	return err;
6642 }
6643 
get_disk_info(struct mddev * mddev,void __user * arg)6644 static int get_disk_info(struct mddev *mddev, void __user * arg)
6645 {
6646 	mdu_disk_info_t info;
6647 	struct md_rdev *rdev;
6648 
6649 	if (copy_from_user(&info, arg, sizeof(info)))
6650 		return -EFAULT;
6651 
6652 	rcu_read_lock();
6653 	rdev = md_find_rdev_nr_rcu(mddev, info.number);
6654 	if (rdev) {
6655 		info.major = MAJOR(rdev->bdev->bd_dev);
6656 		info.minor = MINOR(rdev->bdev->bd_dev);
6657 		info.raid_disk = rdev->raid_disk;
6658 		info.state = 0;
6659 		if (test_bit(Faulty, &rdev->flags))
6660 			info.state |= (1<<MD_DISK_FAULTY);
6661 		else if (test_bit(In_sync, &rdev->flags)) {
6662 			info.state |= (1<<MD_DISK_ACTIVE);
6663 			info.state |= (1<<MD_DISK_SYNC);
6664 		}
6665 		if (test_bit(Journal, &rdev->flags))
6666 			info.state |= (1<<MD_DISK_JOURNAL);
6667 		if (test_bit(WriteMostly, &rdev->flags))
6668 			info.state |= (1<<MD_DISK_WRITEMOSTLY);
6669 		if (test_bit(FailFast, &rdev->flags))
6670 			info.state |= (1<<MD_DISK_FAILFAST);
6671 	} else {
6672 		info.major = info.minor = 0;
6673 		info.raid_disk = -1;
6674 		info.state = (1<<MD_DISK_REMOVED);
6675 	}
6676 	rcu_read_unlock();
6677 
6678 	if (copy_to_user(arg, &info, sizeof(info)))
6679 		return -EFAULT;
6680 
6681 	return 0;
6682 }
6683 
md_add_new_disk(struct mddev * mddev,struct mdu_disk_info_s * info)6684 int md_add_new_disk(struct mddev *mddev, struct mdu_disk_info_s *info)
6685 {
6686 	struct md_rdev *rdev;
6687 	dev_t dev = MKDEV(info->major,info->minor);
6688 
6689 	if (mddev_is_clustered(mddev) &&
6690 		!(info->state & ((1 << MD_DISK_CLUSTER_ADD) | (1 << MD_DISK_CANDIDATE)))) {
6691 		pr_warn("%s: Cannot add to clustered mddev.\n",
6692 			mdname(mddev));
6693 		return -EINVAL;
6694 	}
6695 
6696 	if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
6697 		return -EOVERFLOW;
6698 
6699 	if (!mddev->raid_disks) {
6700 		int err;
6701 		/* expecting a device which has a superblock */
6702 		rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
6703 		if (IS_ERR(rdev)) {
6704 			pr_warn("md: md_import_device returned %ld\n",
6705 				PTR_ERR(rdev));
6706 			return PTR_ERR(rdev);
6707 		}
6708 		if (!list_empty(&mddev->disks)) {
6709 			struct md_rdev *rdev0
6710 				= list_entry(mddev->disks.next,
6711 					     struct md_rdev, same_set);
6712 			err = super_types[mddev->major_version]
6713 				.load_super(rdev, rdev0, mddev->minor_version);
6714 			if (err < 0) {
6715 				pr_warn("md: %pg has different UUID to %pg\n",
6716 					rdev->bdev,
6717 					rdev0->bdev);
6718 				export_rdev(rdev);
6719 				return -EINVAL;
6720 			}
6721 		}
6722 		err = bind_rdev_to_array(rdev, mddev);
6723 		if (err)
6724 			export_rdev(rdev);
6725 		return err;
6726 	}
6727 
6728 	/*
6729 	 * md_add_new_disk can be used once the array is assembled
6730 	 * to add "hot spares".  They must already have a superblock
6731 	 * written
6732 	 */
6733 	if (mddev->pers) {
6734 		int err;
6735 		if (!mddev->pers->hot_add_disk) {
6736 			pr_warn("%s: personality does not support diskops!\n",
6737 				mdname(mddev));
6738 			return -EINVAL;
6739 		}
6740 		if (mddev->persistent)
6741 			rdev = md_import_device(dev, mddev->major_version,
6742 						mddev->minor_version);
6743 		else
6744 			rdev = md_import_device(dev, -1, -1);
6745 		if (IS_ERR(rdev)) {
6746 			pr_warn("md: md_import_device returned %ld\n",
6747 				PTR_ERR(rdev));
6748 			return PTR_ERR(rdev);
6749 		}
6750 		/* set saved_raid_disk if appropriate */
6751 		if (!mddev->persistent) {
6752 			if (info->state & (1<<MD_DISK_SYNC)  &&
6753 			    info->raid_disk < mddev->raid_disks) {
6754 				rdev->raid_disk = info->raid_disk;
6755 				set_bit(In_sync, &rdev->flags);
6756 				clear_bit(Bitmap_sync, &rdev->flags);
6757 			} else
6758 				rdev->raid_disk = -1;
6759 			rdev->saved_raid_disk = rdev->raid_disk;
6760 		} else
6761 			super_types[mddev->major_version].
6762 				validate_super(mddev, rdev);
6763 		if ((info->state & (1<<MD_DISK_SYNC)) &&
6764 		     rdev->raid_disk != info->raid_disk) {
6765 			/* This was a hot-add request, but events doesn't
6766 			 * match, so reject it.
6767 			 */
6768 			export_rdev(rdev);
6769 			return -EINVAL;
6770 		}
6771 
6772 		clear_bit(In_sync, &rdev->flags); /* just to be sure */
6773 		if (info->state & (1<<MD_DISK_WRITEMOSTLY))
6774 			set_bit(WriteMostly, &rdev->flags);
6775 		else
6776 			clear_bit(WriteMostly, &rdev->flags);
6777 		if (info->state & (1<<MD_DISK_FAILFAST))
6778 			set_bit(FailFast, &rdev->flags);
6779 		else
6780 			clear_bit(FailFast, &rdev->flags);
6781 
6782 		if (info->state & (1<<MD_DISK_JOURNAL)) {
6783 			struct md_rdev *rdev2;
6784 			bool has_journal = false;
6785 
6786 			/* make sure no existing journal disk */
6787 			rdev_for_each(rdev2, mddev) {
6788 				if (test_bit(Journal, &rdev2->flags)) {
6789 					has_journal = true;
6790 					break;
6791 				}
6792 			}
6793 			if (has_journal || mddev->bitmap) {
6794 				export_rdev(rdev);
6795 				return -EBUSY;
6796 			}
6797 			set_bit(Journal, &rdev->flags);
6798 		}
6799 		/*
6800 		 * check whether the device shows up in other nodes
6801 		 */
6802 		if (mddev_is_clustered(mddev)) {
6803 			if (info->state & (1 << MD_DISK_CANDIDATE))
6804 				set_bit(Candidate, &rdev->flags);
6805 			else if (info->state & (1 << MD_DISK_CLUSTER_ADD)) {
6806 				/* --add initiated by this node */
6807 				err = md_cluster_ops->add_new_disk(mddev, rdev);
6808 				if (err) {
6809 					export_rdev(rdev);
6810 					return err;
6811 				}
6812 			}
6813 		}
6814 
6815 		rdev->raid_disk = -1;
6816 		err = bind_rdev_to_array(rdev, mddev);
6817 
6818 		if (err)
6819 			export_rdev(rdev);
6820 
6821 		if (mddev_is_clustered(mddev)) {
6822 			if (info->state & (1 << MD_DISK_CANDIDATE)) {
6823 				if (!err) {
6824 					err = md_cluster_ops->new_disk_ack(mddev,
6825 						err == 0);
6826 					if (err)
6827 						md_kick_rdev_from_array(rdev);
6828 				}
6829 			} else {
6830 				if (err)
6831 					md_cluster_ops->add_new_disk_cancel(mddev);
6832 				else
6833 					err = add_bound_rdev(rdev);
6834 			}
6835 
6836 		} else if (!err)
6837 			err = add_bound_rdev(rdev);
6838 
6839 		return err;
6840 	}
6841 
6842 	/* otherwise, md_add_new_disk is only allowed
6843 	 * for major_version==0 superblocks
6844 	 */
6845 	if (mddev->major_version != 0) {
6846 		pr_warn("%s: ADD_NEW_DISK not supported\n", mdname(mddev));
6847 		return -EINVAL;
6848 	}
6849 
6850 	if (!(info->state & (1<<MD_DISK_FAULTY))) {
6851 		int err;
6852 		rdev = md_import_device(dev, -1, 0);
6853 		if (IS_ERR(rdev)) {
6854 			pr_warn("md: error, md_import_device() returned %ld\n",
6855 				PTR_ERR(rdev));
6856 			return PTR_ERR(rdev);
6857 		}
6858 		rdev->desc_nr = info->number;
6859 		if (info->raid_disk < mddev->raid_disks)
6860 			rdev->raid_disk = info->raid_disk;
6861 		else
6862 			rdev->raid_disk = -1;
6863 
6864 		if (rdev->raid_disk < mddev->raid_disks)
6865 			if (info->state & (1<<MD_DISK_SYNC))
6866 				set_bit(In_sync, &rdev->flags);
6867 
6868 		if (info->state & (1<<MD_DISK_WRITEMOSTLY))
6869 			set_bit(WriteMostly, &rdev->flags);
6870 		if (info->state & (1<<MD_DISK_FAILFAST))
6871 			set_bit(FailFast, &rdev->flags);
6872 
6873 		if (!mddev->persistent) {
6874 			pr_debug("md: nonpersistent superblock ...\n");
6875 			rdev->sb_start = bdev_nr_sectors(rdev->bdev);
6876 		} else
6877 			rdev->sb_start = calc_dev_sboffset(rdev);
6878 		rdev->sectors = rdev->sb_start;
6879 
6880 		err = bind_rdev_to_array(rdev, mddev);
6881 		if (err) {
6882 			export_rdev(rdev);
6883 			return err;
6884 		}
6885 	}
6886 
6887 	return 0;
6888 }
6889 
hot_remove_disk(struct mddev * mddev,dev_t dev)6890 static int hot_remove_disk(struct mddev *mddev, dev_t dev)
6891 {
6892 	struct md_rdev *rdev;
6893 
6894 	if (!mddev->pers)
6895 		return -ENODEV;
6896 
6897 	rdev = find_rdev(mddev, dev);
6898 	if (!rdev)
6899 		return -ENXIO;
6900 
6901 	if (rdev->raid_disk < 0)
6902 		goto kick_rdev;
6903 
6904 	clear_bit(Blocked, &rdev->flags);
6905 	remove_and_add_spares(mddev, rdev);
6906 
6907 	if (rdev->raid_disk >= 0)
6908 		goto busy;
6909 
6910 kick_rdev:
6911 	if (mddev_is_clustered(mddev)) {
6912 		if (md_cluster_ops->remove_disk(mddev, rdev))
6913 			goto busy;
6914 	}
6915 
6916 	md_kick_rdev_from_array(rdev);
6917 	set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
6918 	if (mddev->thread)
6919 		md_wakeup_thread(mddev->thread);
6920 	else
6921 		md_update_sb(mddev, 1);
6922 	md_new_event();
6923 
6924 	return 0;
6925 busy:
6926 	pr_debug("md: cannot remove active disk %pg from %s ...\n",
6927 		 rdev->bdev, mdname(mddev));
6928 	return -EBUSY;
6929 }
6930 
hot_add_disk(struct mddev * mddev,dev_t dev)6931 static int hot_add_disk(struct mddev *mddev, dev_t dev)
6932 {
6933 	int err;
6934 	struct md_rdev *rdev;
6935 
6936 	if (!mddev->pers)
6937 		return -ENODEV;
6938 
6939 	if (mddev->major_version != 0) {
6940 		pr_warn("%s: HOT_ADD may only be used with version-0 superblocks.\n",
6941 			mdname(mddev));
6942 		return -EINVAL;
6943 	}
6944 	if (!mddev->pers->hot_add_disk) {
6945 		pr_warn("%s: personality does not support diskops!\n",
6946 			mdname(mddev));
6947 		return -EINVAL;
6948 	}
6949 
6950 	rdev = md_import_device(dev, -1, 0);
6951 	if (IS_ERR(rdev)) {
6952 		pr_warn("md: error, md_import_device() returned %ld\n",
6953 			PTR_ERR(rdev));
6954 		return -EINVAL;
6955 	}
6956 
6957 	if (mddev->persistent)
6958 		rdev->sb_start = calc_dev_sboffset(rdev);
6959 	else
6960 		rdev->sb_start = bdev_nr_sectors(rdev->bdev);
6961 
6962 	rdev->sectors = rdev->sb_start;
6963 
6964 	if (test_bit(Faulty, &rdev->flags)) {
6965 		pr_warn("md: can not hot-add faulty %pg disk to %s!\n",
6966 			rdev->bdev, mdname(mddev));
6967 		err = -EINVAL;
6968 		goto abort_export;
6969 	}
6970 
6971 	clear_bit(In_sync, &rdev->flags);
6972 	rdev->desc_nr = -1;
6973 	rdev->saved_raid_disk = -1;
6974 	err = bind_rdev_to_array(rdev, mddev);
6975 	if (err)
6976 		goto abort_export;
6977 
6978 	/*
6979 	 * The rest should better be atomic, we can have disk failures
6980 	 * noticed in interrupt contexts ...
6981 	 */
6982 
6983 	rdev->raid_disk = -1;
6984 
6985 	set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
6986 	if (!mddev->thread)
6987 		md_update_sb(mddev, 1);
6988 	/*
6989 	 * If the new disk does not support REQ_NOWAIT,
6990 	 * disable on the whole MD.
6991 	 */
6992 	if (!blk_queue_nowait(bdev_get_queue(rdev->bdev))) {
6993 		pr_info("%s: Disabling nowait because %pg does not support nowait\n",
6994 			mdname(mddev), rdev->bdev);
6995 		blk_queue_flag_clear(QUEUE_FLAG_NOWAIT, mddev->queue);
6996 	}
6997 	/*
6998 	 * Kick recovery, maybe this spare has to be added to the
6999 	 * array immediately.
7000 	 */
7001 	set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7002 	md_wakeup_thread(mddev->thread);
7003 	md_new_event();
7004 	return 0;
7005 
7006 abort_export:
7007 	export_rdev(rdev);
7008 	return err;
7009 }
7010 
set_bitmap_file(struct mddev * mddev,int fd)7011 static int set_bitmap_file(struct mddev *mddev, int fd)
7012 {
7013 	int err = 0;
7014 
7015 	if (mddev->pers) {
7016 		if (!mddev->pers->quiesce || !mddev->thread)
7017 			return -EBUSY;
7018 		if (mddev->recovery || mddev->sync_thread)
7019 			return -EBUSY;
7020 		/* we should be able to change the bitmap.. */
7021 	}
7022 
7023 	if (fd >= 0) {
7024 		struct inode *inode;
7025 		struct file *f;
7026 
7027 		if (mddev->bitmap || mddev->bitmap_info.file)
7028 			return -EEXIST; /* cannot add when bitmap is present */
7029 		f = fget(fd);
7030 
7031 		if (f == NULL) {
7032 			pr_warn("%s: error: failed to get bitmap file\n",
7033 				mdname(mddev));
7034 			return -EBADF;
7035 		}
7036 
7037 		inode = f->f_mapping->host;
7038 		if (!S_ISREG(inode->i_mode)) {
7039 			pr_warn("%s: error: bitmap file must be a regular file\n",
7040 				mdname(mddev));
7041 			err = -EBADF;
7042 		} else if (!(f->f_mode & FMODE_WRITE)) {
7043 			pr_warn("%s: error: bitmap file must open for write\n",
7044 				mdname(mddev));
7045 			err = -EBADF;
7046 		} else if (atomic_read(&inode->i_writecount) != 1) {
7047 			pr_warn("%s: error: bitmap file is already in use\n",
7048 				mdname(mddev));
7049 			err = -EBUSY;
7050 		}
7051 		if (err) {
7052 			fput(f);
7053 			return err;
7054 		}
7055 		mddev->bitmap_info.file = f;
7056 		mddev->bitmap_info.offset = 0; /* file overrides offset */
7057 	} else if (mddev->bitmap == NULL)
7058 		return -ENOENT; /* cannot remove what isn't there */
7059 	err = 0;
7060 	if (mddev->pers) {
7061 		if (fd >= 0) {
7062 			struct bitmap *bitmap;
7063 
7064 			bitmap = md_bitmap_create(mddev, -1);
7065 			mddev_suspend(mddev);
7066 			if (!IS_ERR(bitmap)) {
7067 				mddev->bitmap = bitmap;
7068 				err = md_bitmap_load(mddev);
7069 			} else
7070 				err = PTR_ERR(bitmap);
7071 			if (err) {
7072 				md_bitmap_destroy(mddev);
7073 				fd = -1;
7074 			}
7075 			mddev_resume(mddev);
7076 		} else if (fd < 0) {
7077 			mddev_suspend(mddev);
7078 			md_bitmap_destroy(mddev);
7079 			mddev_resume(mddev);
7080 		}
7081 	}
7082 	if (fd < 0) {
7083 		struct file *f = mddev->bitmap_info.file;
7084 		if (f) {
7085 			spin_lock(&mddev->lock);
7086 			mddev->bitmap_info.file = NULL;
7087 			spin_unlock(&mddev->lock);
7088 			fput(f);
7089 		}
7090 	}
7091 
7092 	return err;
7093 }
7094 
7095 /*
7096  * md_set_array_info is used two different ways
7097  * The original usage is when creating a new array.
7098  * In this usage, raid_disks is > 0 and it together with
7099  *  level, size, not_persistent,layout,chunksize determine the
7100  *  shape of the array.
7101  *  This will always create an array with a type-0.90.0 superblock.
7102  * The newer usage is when assembling an array.
7103  *  In this case raid_disks will be 0, and the major_version field is
7104  *  use to determine which style super-blocks are to be found on the devices.
7105  *  The minor and patch _version numbers are also kept incase the
7106  *  super_block handler wishes to interpret them.
7107  */
md_set_array_info(struct mddev * mddev,struct mdu_array_info_s * info)7108 int md_set_array_info(struct mddev *mddev, struct mdu_array_info_s *info)
7109 {
7110 	if (info->raid_disks == 0) {
7111 		/* just setting version number for superblock loading */
7112 		if (info->major_version < 0 ||
7113 		    info->major_version >= ARRAY_SIZE(super_types) ||
7114 		    super_types[info->major_version].name == NULL) {
7115 			/* maybe try to auto-load a module? */
7116 			pr_warn("md: superblock version %d not known\n",
7117 				info->major_version);
7118 			return -EINVAL;
7119 		}
7120 		mddev->major_version = info->major_version;
7121 		mddev->minor_version = info->minor_version;
7122 		mddev->patch_version = info->patch_version;
7123 		mddev->persistent = !info->not_persistent;
7124 		/* ensure mddev_put doesn't delete this now that there
7125 		 * is some minimal configuration.
7126 		 */
7127 		mddev->ctime         = ktime_get_real_seconds();
7128 		return 0;
7129 	}
7130 	mddev->major_version = MD_MAJOR_VERSION;
7131 	mddev->minor_version = MD_MINOR_VERSION;
7132 	mddev->patch_version = MD_PATCHLEVEL_VERSION;
7133 	mddev->ctime         = ktime_get_real_seconds();
7134 
7135 	mddev->level         = info->level;
7136 	mddev->clevel[0]     = 0;
7137 	mddev->dev_sectors   = 2 * (sector_t)info->size;
7138 	mddev->raid_disks    = info->raid_disks;
7139 	/* don't set md_minor, it is determined by which /dev/md* was
7140 	 * openned
7141 	 */
7142 	if (info->state & (1<<MD_SB_CLEAN))
7143 		mddev->recovery_cp = MaxSector;
7144 	else
7145 		mddev->recovery_cp = 0;
7146 	mddev->persistent    = ! info->not_persistent;
7147 	mddev->external	     = 0;
7148 
7149 	mddev->layout        = info->layout;
7150 	if (mddev->level == 0)
7151 		/* Cannot trust RAID0 layout info here */
7152 		mddev->layout = -1;
7153 	mddev->chunk_sectors = info->chunk_size >> 9;
7154 
7155 	if (mddev->persistent) {
7156 		mddev->max_disks = MD_SB_DISKS;
7157 		mddev->flags = 0;
7158 		mddev->sb_flags = 0;
7159 	}
7160 	set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
7161 
7162 	mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9;
7163 	mddev->bitmap_info.default_space = 64*2 - (MD_SB_BYTES >> 9);
7164 	mddev->bitmap_info.offset = 0;
7165 
7166 	mddev->reshape_position = MaxSector;
7167 
7168 	/*
7169 	 * Generate a 128 bit UUID
7170 	 */
7171 	get_random_bytes(mddev->uuid, 16);
7172 
7173 	mddev->new_level = mddev->level;
7174 	mddev->new_chunk_sectors = mddev->chunk_sectors;
7175 	mddev->new_layout = mddev->layout;
7176 	mddev->delta_disks = 0;
7177 	mddev->reshape_backwards = 0;
7178 
7179 	return 0;
7180 }
7181 
md_set_array_sectors(struct mddev * mddev,sector_t array_sectors)7182 void md_set_array_sectors(struct mddev *mddev, sector_t array_sectors)
7183 {
7184 	lockdep_assert_held(&mddev->reconfig_mutex);
7185 
7186 	if (mddev->external_size)
7187 		return;
7188 
7189 	mddev->array_sectors = array_sectors;
7190 }
7191 EXPORT_SYMBOL(md_set_array_sectors);
7192 
update_size(struct mddev * mddev,sector_t num_sectors)7193 static int update_size(struct mddev *mddev, sector_t num_sectors)
7194 {
7195 	struct md_rdev *rdev;
7196 	int rv;
7197 	int fit = (num_sectors == 0);
7198 	sector_t old_dev_sectors = mddev->dev_sectors;
7199 
7200 	if (mddev->pers->resize == NULL)
7201 		return -EINVAL;
7202 	/* The "num_sectors" is the number of sectors of each device that
7203 	 * is used.  This can only make sense for arrays with redundancy.
7204 	 * linear and raid0 always use whatever space is available. We can only
7205 	 * consider changing this number if no resync or reconstruction is
7206 	 * happening, and if the new size is acceptable. It must fit before the
7207 	 * sb_start or, if that is <data_offset, it must fit before the size
7208 	 * of each device.  If num_sectors is zero, we find the largest size
7209 	 * that fits.
7210 	 */
7211 	if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
7212 	    mddev->sync_thread)
7213 		return -EBUSY;
7214 	if (mddev->ro)
7215 		return -EROFS;
7216 
7217 	rdev_for_each(rdev, mddev) {
7218 		sector_t avail = rdev->sectors;
7219 
7220 		if (fit && (num_sectors == 0 || num_sectors > avail))
7221 			num_sectors = avail;
7222 		if (avail < num_sectors)
7223 			return -ENOSPC;
7224 	}
7225 	rv = mddev->pers->resize(mddev, num_sectors);
7226 	if (!rv) {
7227 		if (mddev_is_clustered(mddev))
7228 			md_cluster_ops->update_size(mddev, old_dev_sectors);
7229 		else if (mddev->queue) {
7230 			set_capacity_and_notify(mddev->gendisk,
7231 						mddev->array_sectors);
7232 		}
7233 	}
7234 	return rv;
7235 }
7236 
update_raid_disks(struct mddev * mddev,int raid_disks)7237 static int update_raid_disks(struct mddev *mddev, int raid_disks)
7238 {
7239 	int rv;
7240 	struct md_rdev *rdev;
7241 	/* change the number of raid disks */
7242 	if (mddev->pers->check_reshape == NULL)
7243 		return -EINVAL;
7244 	if (mddev->ro)
7245 		return -EROFS;
7246 	if (raid_disks <= 0 ||
7247 	    (mddev->max_disks && raid_disks >= mddev->max_disks))
7248 		return -EINVAL;
7249 	if (mddev->sync_thread ||
7250 	    test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
7251 	    test_bit(MD_RESYNCING_REMOTE, &mddev->recovery) ||
7252 	    mddev->reshape_position != MaxSector)
7253 		return -EBUSY;
7254 
7255 	rdev_for_each(rdev, mddev) {
7256 		if (mddev->raid_disks < raid_disks &&
7257 		    rdev->data_offset < rdev->new_data_offset)
7258 			return -EINVAL;
7259 		if (mddev->raid_disks > raid_disks &&
7260 		    rdev->data_offset > rdev->new_data_offset)
7261 			return -EINVAL;
7262 	}
7263 
7264 	mddev->delta_disks = raid_disks - mddev->raid_disks;
7265 	if (mddev->delta_disks < 0)
7266 		mddev->reshape_backwards = 1;
7267 	else if (mddev->delta_disks > 0)
7268 		mddev->reshape_backwards = 0;
7269 
7270 	rv = mddev->pers->check_reshape(mddev);
7271 	if (rv < 0) {
7272 		mddev->delta_disks = 0;
7273 		mddev->reshape_backwards = 0;
7274 	}
7275 	return rv;
7276 }
7277 
7278 /*
7279  * update_array_info is used to change the configuration of an
7280  * on-line array.
7281  * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
7282  * fields in the info are checked against the array.
7283  * Any differences that cannot be handled will cause an error.
7284  * Normally, only one change can be managed at a time.
7285  */
update_array_info(struct mddev * mddev,mdu_array_info_t * info)7286 static int update_array_info(struct mddev *mddev, mdu_array_info_t *info)
7287 {
7288 	int rv = 0;
7289 	int cnt = 0;
7290 	int state = 0;
7291 
7292 	/* calculate expected state,ignoring low bits */
7293 	if (mddev->bitmap && mddev->bitmap_info.offset)
7294 		state |= (1 << MD_SB_BITMAP_PRESENT);
7295 
7296 	if (mddev->major_version != info->major_version ||
7297 	    mddev->minor_version != info->minor_version ||
7298 /*	    mddev->patch_version != info->patch_version || */
7299 	    mddev->ctime         != info->ctime         ||
7300 	    mddev->level         != info->level         ||
7301 /*	    mddev->layout        != info->layout        || */
7302 	    mddev->persistent	 != !info->not_persistent ||
7303 	    mddev->chunk_sectors != info->chunk_size >> 9 ||
7304 	    /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
7305 	    ((state^info->state) & 0xfffffe00)
7306 		)
7307 		return -EINVAL;
7308 	/* Check there is only one change */
7309 	if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
7310 		cnt++;
7311 	if (mddev->raid_disks != info->raid_disks)
7312 		cnt++;
7313 	if (mddev->layout != info->layout)
7314 		cnt++;
7315 	if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT))
7316 		cnt++;
7317 	if (cnt == 0)
7318 		return 0;
7319 	if (cnt > 1)
7320 		return -EINVAL;
7321 
7322 	if (mddev->layout != info->layout) {
7323 		/* Change layout
7324 		 * we don't need to do anything at the md level, the
7325 		 * personality will take care of it all.
7326 		 */
7327 		if (mddev->pers->check_reshape == NULL)
7328 			return -EINVAL;
7329 		else {
7330 			mddev->new_layout = info->layout;
7331 			rv = mddev->pers->check_reshape(mddev);
7332 			if (rv)
7333 				mddev->new_layout = mddev->layout;
7334 			return rv;
7335 		}
7336 	}
7337 	if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
7338 		rv = update_size(mddev, (sector_t)info->size * 2);
7339 
7340 	if (mddev->raid_disks    != info->raid_disks)
7341 		rv = update_raid_disks(mddev, info->raid_disks);
7342 
7343 	if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) {
7344 		if (mddev->pers->quiesce == NULL || mddev->thread == NULL) {
7345 			rv = -EINVAL;
7346 			goto err;
7347 		}
7348 		if (mddev->recovery || mddev->sync_thread) {
7349 			rv = -EBUSY;
7350 			goto err;
7351 		}
7352 		if (info->state & (1<<MD_SB_BITMAP_PRESENT)) {
7353 			struct bitmap *bitmap;
7354 			/* add the bitmap */
7355 			if (mddev->bitmap) {
7356 				rv = -EEXIST;
7357 				goto err;
7358 			}
7359 			if (mddev->bitmap_info.default_offset == 0) {
7360 				rv = -EINVAL;
7361 				goto err;
7362 			}
7363 			mddev->bitmap_info.offset =
7364 				mddev->bitmap_info.default_offset;
7365 			mddev->bitmap_info.space =
7366 				mddev->bitmap_info.default_space;
7367 			bitmap = md_bitmap_create(mddev, -1);
7368 			mddev_suspend(mddev);
7369 			if (!IS_ERR(bitmap)) {
7370 				mddev->bitmap = bitmap;
7371 				rv = md_bitmap_load(mddev);
7372 			} else
7373 				rv = PTR_ERR(bitmap);
7374 			if (rv)
7375 				md_bitmap_destroy(mddev);
7376 			mddev_resume(mddev);
7377 		} else {
7378 			/* remove the bitmap */
7379 			if (!mddev->bitmap) {
7380 				rv = -ENOENT;
7381 				goto err;
7382 			}
7383 			if (mddev->bitmap->storage.file) {
7384 				rv = -EINVAL;
7385 				goto err;
7386 			}
7387 			if (mddev->bitmap_info.nodes) {
7388 				/* hold PW on all the bitmap lock */
7389 				if (md_cluster_ops->lock_all_bitmaps(mddev) <= 0) {
7390 					pr_warn("md: can't change bitmap to none since the array is in use by more than one node\n");
7391 					rv = -EPERM;
7392 					md_cluster_ops->unlock_all_bitmaps(mddev);
7393 					goto err;
7394 				}
7395 
7396 				mddev->bitmap_info.nodes = 0;
7397 				md_cluster_ops->leave(mddev);
7398 				module_put(md_cluster_mod);
7399 				mddev->safemode_delay = DEFAULT_SAFEMODE_DELAY;
7400 			}
7401 			mddev_suspend(mddev);
7402 			md_bitmap_destroy(mddev);
7403 			mddev_resume(mddev);
7404 			mddev->bitmap_info.offset = 0;
7405 		}
7406 	}
7407 	md_update_sb(mddev, 1);
7408 	return rv;
7409 err:
7410 	return rv;
7411 }
7412 
set_disk_faulty(struct mddev * mddev,dev_t dev)7413 static int set_disk_faulty(struct mddev *mddev, dev_t dev)
7414 {
7415 	struct md_rdev *rdev;
7416 	int err = 0;
7417 
7418 	if (mddev->pers == NULL)
7419 		return -ENODEV;
7420 
7421 	rcu_read_lock();
7422 	rdev = md_find_rdev_rcu(mddev, dev);
7423 	if (!rdev)
7424 		err =  -ENODEV;
7425 	else {
7426 		md_error(mddev, rdev);
7427 		if (test_bit(MD_BROKEN, &mddev->flags))
7428 			err = -EBUSY;
7429 	}
7430 	rcu_read_unlock();
7431 	return err;
7432 }
7433 
7434 /*
7435  * We have a problem here : there is no easy way to give a CHS
7436  * virtual geometry. We currently pretend that we have a 2 heads
7437  * 4 sectors (with a BIG number of cylinders...). This drives
7438  * dosfs just mad... ;-)
7439  */
md_getgeo(struct block_device * bdev,struct hd_geometry * geo)7440 static int md_getgeo(struct block_device *bdev, struct hd_geometry *geo)
7441 {
7442 	struct mddev *mddev = bdev->bd_disk->private_data;
7443 
7444 	geo->heads = 2;
7445 	geo->sectors = 4;
7446 	geo->cylinders = mddev->array_sectors / 8;
7447 	return 0;
7448 }
7449 
md_ioctl_valid(unsigned int cmd)7450 static inline bool md_ioctl_valid(unsigned int cmd)
7451 {
7452 	switch (cmd) {
7453 	case ADD_NEW_DISK:
7454 	case GET_ARRAY_INFO:
7455 	case GET_BITMAP_FILE:
7456 	case GET_DISK_INFO:
7457 	case HOT_ADD_DISK:
7458 	case HOT_REMOVE_DISK:
7459 	case RAID_VERSION:
7460 	case RESTART_ARRAY_RW:
7461 	case RUN_ARRAY:
7462 	case SET_ARRAY_INFO:
7463 	case SET_BITMAP_FILE:
7464 	case SET_DISK_FAULTY:
7465 	case STOP_ARRAY:
7466 	case STOP_ARRAY_RO:
7467 	case CLUSTERED_DISK_NACK:
7468 		return true;
7469 	default:
7470 		return false;
7471 	}
7472 }
7473 
md_ioctl(struct block_device * bdev,fmode_t mode,unsigned int cmd,unsigned long arg)7474 static int md_ioctl(struct block_device *bdev, fmode_t mode,
7475 			unsigned int cmd, unsigned long arg)
7476 {
7477 	int err = 0;
7478 	void __user *argp = (void __user *)arg;
7479 	struct mddev *mddev = NULL;
7480 	bool did_set_md_closing = false;
7481 
7482 	if (!md_ioctl_valid(cmd))
7483 		return -ENOTTY;
7484 
7485 	switch (cmd) {
7486 	case RAID_VERSION:
7487 	case GET_ARRAY_INFO:
7488 	case GET_DISK_INFO:
7489 		break;
7490 	default:
7491 		if (!capable(CAP_SYS_ADMIN))
7492 			return -EACCES;
7493 	}
7494 
7495 	/*
7496 	 * Commands dealing with the RAID driver but not any
7497 	 * particular array:
7498 	 */
7499 	switch (cmd) {
7500 	case RAID_VERSION:
7501 		err = get_version(argp);
7502 		goto out;
7503 	default:;
7504 	}
7505 
7506 	/*
7507 	 * Commands creating/starting a new array:
7508 	 */
7509 
7510 	mddev = bdev->bd_disk->private_data;
7511 
7512 	if (!mddev) {
7513 		BUG();
7514 		goto out;
7515 	}
7516 
7517 	/* Some actions do not requires the mutex */
7518 	switch (cmd) {
7519 	case GET_ARRAY_INFO:
7520 		if (!mddev->raid_disks && !mddev->external)
7521 			err = -ENODEV;
7522 		else
7523 			err = get_array_info(mddev, argp);
7524 		goto out;
7525 
7526 	case GET_DISK_INFO:
7527 		if (!mddev->raid_disks && !mddev->external)
7528 			err = -ENODEV;
7529 		else
7530 			err = get_disk_info(mddev, argp);
7531 		goto out;
7532 
7533 	case SET_DISK_FAULTY:
7534 		err = set_disk_faulty(mddev, new_decode_dev(arg));
7535 		goto out;
7536 
7537 	case GET_BITMAP_FILE:
7538 		err = get_bitmap_file(mddev, argp);
7539 		goto out;
7540 
7541 	}
7542 
7543 	if (cmd == ADD_NEW_DISK || cmd == HOT_ADD_DISK)
7544 		flush_rdev_wq(mddev);
7545 
7546 	if (cmd == HOT_REMOVE_DISK)
7547 		/* need to ensure recovery thread has run */
7548 		wait_event_interruptible_timeout(mddev->sb_wait,
7549 						 !test_bit(MD_RECOVERY_NEEDED,
7550 							   &mddev->recovery),
7551 						 msecs_to_jiffies(5000));
7552 	if (cmd == STOP_ARRAY || cmd == STOP_ARRAY_RO) {
7553 		/* Need to flush page cache, and ensure no-one else opens
7554 		 * and writes
7555 		 */
7556 		mutex_lock(&mddev->open_mutex);
7557 		if (mddev->pers && atomic_read(&mddev->openers) > 1) {
7558 			mutex_unlock(&mddev->open_mutex);
7559 			err = -EBUSY;
7560 			goto out;
7561 		}
7562 		if (test_and_set_bit(MD_CLOSING, &mddev->flags)) {
7563 			mutex_unlock(&mddev->open_mutex);
7564 			err = -EBUSY;
7565 			goto out;
7566 		}
7567 		did_set_md_closing = true;
7568 		mutex_unlock(&mddev->open_mutex);
7569 		sync_blockdev(bdev);
7570 	}
7571 	err = mddev_lock(mddev);
7572 	if (err) {
7573 		pr_debug("md: ioctl lock interrupted, reason %d, cmd %d\n",
7574 			 err, cmd);
7575 		goto out;
7576 	}
7577 
7578 	if (cmd == SET_ARRAY_INFO) {
7579 		mdu_array_info_t info;
7580 		if (!arg)
7581 			memset(&info, 0, sizeof(info));
7582 		else if (copy_from_user(&info, argp, sizeof(info))) {
7583 			err = -EFAULT;
7584 			goto unlock;
7585 		}
7586 		if (mddev->pers) {
7587 			err = update_array_info(mddev, &info);
7588 			if (err) {
7589 				pr_warn("md: couldn't update array info. %d\n", err);
7590 				goto unlock;
7591 			}
7592 			goto unlock;
7593 		}
7594 		if (!list_empty(&mddev->disks)) {
7595 			pr_warn("md: array %s already has disks!\n", mdname(mddev));
7596 			err = -EBUSY;
7597 			goto unlock;
7598 		}
7599 		if (mddev->raid_disks) {
7600 			pr_warn("md: array %s already initialised!\n", mdname(mddev));
7601 			err = -EBUSY;
7602 			goto unlock;
7603 		}
7604 		err = md_set_array_info(mddev, &info);
7605 		if (err) {
7606 			pr_warn("md: couldn't set array info. %d\n", err);
7607 			goto unlock;
7608 		}
7609 		goto unlock;
7610 	}
7611 
7612 	/*
7613 	 * Commands querying/configuring an existing array:
7614 	 */
7615 	/* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
7616 	 * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */
7617 	if ((!mddev->raid_disks && !mddev->external)
7618 	    && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY
7619 	    && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE
7620 	    && cmd != GET_BITMAP_FILE) {
7621 		err = -ENODEV;
7622 		goto unlock;
7623 	}
7624 
7625 	/*
7626 	 * Commands even a read-only array can execute:
7627 	 */
7628 	switch (cmd) {
7629 	case RESTART_ARRAY_RW:
7630 		err = restart_array(mddev);
7631 		goto unlock;
7632 
7633 	case STOP_ARRAY:
7634 		err = do_md_stop(mddev, 0, bdev);
7635 		goto unlock;
7636 
7637 	case STOP_ARRAY_RO:
7638 		err = md_set_readonly(mddev, bdev);
7639 		goto unlock;
7640 
7641 	case HOT_REMOVE_DISK:
7642 		err = hot_remove_disk(mddev, new_decode_dev(arg));
7643 		goto unlock;
7644 
7645 	case ADD_NEW_DISK:
7646 		/* We can support ADD_NEW_DISK on read-only arrays
7647 		 * only if we are re-adding a preexisting device.
7648 		 * So require mddev->pers and MD_DISK_SYNC.
7649 		 */
7650 		if (mddev->pers) {
7651 			mdu_disk_info_t info;
7652 			if (copy_from_user(&info, argp, sizeof(info)))
7653 				err = -EFAULT;
7654 			else if (!(info.state & (1<<MD_DISK_SYNC)))
7655 				/* Need to clear read-only for this */
7656 				break;
7657 			else
7658 				err = md_add_new_disk(mddev, &info);
7659 			goto unlock;
7660 		}
7661 		break;
7662 	}
7663 
7664 	/*
7665 	 * The remaining ioctls are changing the state of the
7666 	 * superblock, so we do not allow them on read-only arrays.
7667 	 */
7668 	if (mddev->ro && mddev->pers) {
7669 		if (mddev->ro == 2) {
7670 			mddev->ro = 0;
7671 			sysfs_notify_dirent_safe(mddev->sysfs_state);
7672 			set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7673 			/* mddev_unlock will wake thread */
7674 			/* If a device failed while we were read-only, we
7675 			 * need to make sure the metadata is updated now.
7676 			 */
7677 			if (test_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags)) {
7678 				mddev_unlock(mddev);
7679 				wait_event(mddev->sb_wait,
7680 					   !test_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags) &&
7681 					   !test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags));
7682 				mddev_lock_nointr(mddev);
7683 			}
7684 		} else {
7685 			err = -EROFS;
7686 			goto unlock;
7687 		}
7688 	}
7689 
7690 	switch (cmd) {
7691 	case ADD_NEW_DISK:
7692 	{
7693 		mdu_disk_info_t info;
7694 		if (copy_from_user(&info, argp, sizeof(info)))
7695 			err = -EFAULT;
7696 		else
7697 			err = md_add_new_disk(mddev, &info);
7698 		goto unlock;
7699 	}
7700 
7701 	case CLUSTERED_DISK_NACK:
7702 		if (mddev_is_clustered(mddev))
7703 			md_cluster_ops->new_disk_ack(mddev, false);
7704 		else
7705 			err = -EINVAL;
7706 		goto unlock;
7707 
7708 	case HOT_ADD_DISK:
7709 		err = hot_add_disk(mddev, new_decode_dev(arg));
7710 		goto unlock;
7711 
7712 	case RUN_ARRAY:
7713 		err = do_md_run(mddev);
7714 		goto unlock;
7715 
7716 	case SET_BITMAP_FILE:
7717 		err = set_bitmap_file(mddev, (int)arg);
7718 		goto unlock;
7719 
7720 	default:
7721 		err = -EINVAL;
7722 		goto unlock;
7723 	}
7724 
7725 unlock:
7726 	if (mddev->hold_active == UNTIL_IOCTL &&
7727 	    err != -EINVAL)
7728 		mddev->hold_active = 0;
7729 	mddev_unlock(mddev);
7730 out:
7731 	if(did_set_md_closing)
7732 		clear_bit(MD_CLOSING, &mddev->flags);
7733 	return err;
7734 }
7735 #ifdef CONFIG_COMPAT
md_compat_ioctl(struct block_device * bdev,fmode_t mode,unsigned int cmd,unsigned long arg)7736 static int md_compat_ioctl(struct block_device *bdev, fmode_t mode,
7737 		    unsigned int cmd, unsigned long arg)
7738 {
7739 	switch (cmd) {
7740 	case HOT_REMOVE_DISK:
7741 	case HOT_ADD_DISK:
7742 	case SET_DISK_FAULTY:
7743 	case SET_BITMAP_FILE:
7744 		/* These take in integer arg, do not convert */
7745 		break;
7746 	default:
7747 		arg = (unsigned long)compat_ptr(arg);
7748 		break;
7749 	}
7750 
7751 	return md_ioctl(bdev, mode, cmd, arg);
7752 }
7753 #endif /* CONFIG_COMPAT */
7754 
md_set_read_only(struct block_device * bdev,bool ro)7755 static int md_set_read_only(struct block_device *bdev, bool ro)
7756 {
7757 	struct mddev *mddev = bdev->bd_disk->private_data;
7758 	int err;
7759 
7760 	err = mddev_lock(mddev);
7761 	if (err)
7762 		return err;
7763 
7764 	if (!mddev->raid_disks && !mddev->external) {
7765 		err = -ENODEV;
7766 		goto out_unlock;
7767 	}
7768 
7769 	/*
7770 	 * Transitioning to read-auto need only happen for arrays that call
7771 	 * md_write_start and which are not ready for writes yet.
7772 	 */
7773 	if (!ro && mddev->ro == 1 && mddev->pers) {
7774 		err = restart_array(mddev);
7775 		if (err)
7776 			goto out_unlock;
7777 		mddev->ro = 2;
7778 	}
7779 
7780 out_unlock:
7781 	mddev_unlock(mddev);
7782 	return err;
7783 }
7784 
md_open(struct block_device * bdev,fmode_t mode)7785 static int md_open(struct block_device *bdev, fmode_t mode)
7786 {
7787 	/*
7788 	 * Succeed if we can lock the mddev, which confirms that
7789 	 * it isn't being stopped right now.
7790 	 */
7791 	struct mddev *mddev = mddev_find(bdev->bd_dev);
7792 	int err;
7793 
7794 	if (!mddev)
7795 		return -ENODEV;
7796 
7797 	if (mddev->gendisk != bdev->bd_disk) {
7798 		/* we are racing with mddev_put which is discarding this
7799 		 * bd_disk.
7800 		 */
7801 		mddev_put(mddev);
7802 		/* Wait until bdev->bd_disk is definitely gone */
7803 		if (work_pending(&mddev->del_work))
7804 			flush_workqueue(md_misc_wq);
7805 		return -EBUSY;
7806 	}
7807 	BUG_ON(mddev != bdev->bd_disk->private_data);
7808 
7809 	if ((err = mutex_lock_interruptible(&mddev->open_mutex)))
7810 		goto out;
7811 
7812 	if (test_bit(MD_CLOSING, &mddev->flags)) {
7813 		mutex_unlock(&mddev->open_mutex);
7814 		err = -ENODEV;
7815 		goto out;
7816 	}
7817 
7818 	err = 0;
7819 	atomic_inc(&mddev->openers);
7820 	mutex_unlock(&mddev->open_mutex);
7821 
7822 	bdev_check_media_change(bdev);
7823  out:
7824 	if (err)
7825 		mddev_put(mddev);
7826 	return err;
7827 }
7828 
md_release(struct gendisk * disk,fmode_t mode)7829 static void md_release(struct gendisk *disk, fmode_t mode)
7830 {
7831 	struct mddev *mddev = disk->private_data;
7832 
7833 	BUG_ON(!mddev);
7834 	atomic_dec(&mddev->openers);
7835 	mddev_put(mddev);
7836 }
7837 
md_check_events(struct gendisk * disk,unsigned int clearing)7838 static unsigned int md_check_events(struct gendisk *disk, unsigned int clearing)
7839 {
7840 	struct mddev *mddev = disk->private_data;
7841 	unsigned int ret = 0;
7842 
7843 	if (mddev->changed)
7844 		ret = DISK_EVENT_MEDIA_CHANGE;
7845 	mddev->changed = 0;
7846 	return ret;
7847 }
7848 
7849 const struct block_device_operations md_fops =
7850 {
7851 	.owner		= THIS_MODULE,
7852 	.submit_bio	= md_submit_bio,
7853 	.open		= md_open,
7854 	.release	= md_release,
7855 	.ioctl		= md_ioctl,
7856 #ifdef CONFIG_COMPAT
7857 	.compat_ioctl	= md_compat_ioctl,
7858 #endif
7859 	.getgeo		= md_getgeo,
7860 	.check_events	= md_check_events,
7861 	.set_read_only	= md_set_read_only,
7862 };
7863 
md_thread(void * arg)7864 static int md_thread(void *arg)
7865 {
7866 	struct md_thread *thread = arg;
7867 
7868 	/*
7869 	 * md_thread is a 'system-thread', it's priority should be very
7870 	 * high. We avoid resource deadlocks individually in each
7871 	 * raid personality. (RAID5 does preallocation) We also use RR and
7872 	 * the very same RT priority as kswapd, thus we will never get
7873 	 * into a priority inversion deadlock.
7874 	 *
7875 	 * we definitely have to have equal or higher priority than
7876 	 * bdflush, otherwise bdflush will deadlock if there are too
7877 	 * many dirty RAID5 blocks.
7878 	 */
7879 
7880 	allow_signal(SIGKILL);
7881 	while (!kthread_should_stop()) {
7882 
7883 		/* We need to wait INTERRUPTIBLE so that
7884 		 * we don't add to the load-average.
7885 		 * That means we need to be sure no signals are
7886 		 * pending
7887 		 */
7888 		if (signal_pending(current))
7889 			flush_signals(current);
7890 
7891 		wait_event_interruptible_timeout
7892 			(thread->wqueue,
7893 			 test_bit(THREAD_WAKEUP, &thread->flags)
7894 			 || kthread_should_stop() || kthread_should_park(),
7895 			 thread->timeout);
7896 
7897 		clear_bit(THREAD_WAKEUP, &thread->flags);
7898 		if (kthread_should_park())
7899 			kthread_parkme();
7900 		if (!kthread_should_stop())
7901 			thread->run(thread);
7902 	}
7903 
7904 	return 0;
7905 }
7906 
md_wakeup_thread(struct md_thread * thread)7907 void md_wakeup_thread(struct md_thread *thread)
7908 {
7909 	if (thread) {
7910 		pr_debug("md: waking up MD thread %s.\n", thread->tsk->comm);
7911 		set_bit(THREAD_WAKEUP, &thread->flags);
7912 		wake_up(&thread->wqueue);
7913 	}
7914 }
7915 EXPORT_SYMBOL(md_wakeup_thread);
7916 
md_register_thread(void (* run)(struct md_thread *),struct mddev * mddev,const char * name)7917 struct md_thread *md_register_thread(void (*run) (struct md_thread *),
7918 		struct mddev *mddev, const char *name)
7919 {
7920 	struct md_thread *thread;
7921 
7922 	thread = kzalloc(sizeof(struct md_thread), GFP_KERNEL);
7923 	if (!thread)
7924 		return NULL;
7925 
7926 	init_waitqueue_head(&thread->wqueue);
7927 
7928 	thread->run = run;
7929 	thread->mddev = mddev;
7930 	thread->timeout = MAX_SCHEDULE_TIMEOUT;
7931 	thread->tsk = kthread_run(md_thread, thread,
7932 				  "%s_%s",
7933 				  mdname(thread->mddev),
7934 				  name);
7935 	if (IS_ERR(thread->tsk)) {
7936 		kfree(thread);
7937 		return NULL;
7938 	}
7939 	return thread;
7940 }
7941 EXPORT_SYMBOL(md_register_thread);
7942 
md_unregister_thread(struct md_thread ** threadp)7943 void md_unregister_thread(struct md_thread **threadp)
7944 {
7945 	struct md_thread *thread;
7946 
7947 	/*
7948 	 * Locking ensures that mddev_unlock does not wake_up a
7949 	 * non-existent thread
7950 	 */
7951 	spin_lock(&pers_lock);
7952 	thread = *threadp;
7953 	if (!thread) {
7954 		spin_unlock(&pers_lock);
7955 		return;
7956 	}
7957 	*threadp = NULL;
7958 	spin_unlock(&pers_lock);
7959 
7960 	pr_debug("interrupting MD-thread pid %d\n", task_pid_nr(thread->tsk));
7961 	kthread_stop(thread->tsk);
7962 	kfree(thread);
7963 }
7964 EXPORT_SYMBOL(md_unregister_thread);
7965 
md_error(struct mddev * mddev,struct md_rdev * rdev)7966 void md_error(struct mddev *mddev, struct md_rdev *rdev)
7967 {
7968 	if (!rdev || test_bit(Faulty, &rdev->flags))
7969 		return;
7970 
7971 	if (!mddev->pers || !mddev->pers->error_handler)
7972 		return;
7973 	mddev->pers->error_handler(mddev, rdev);
7974 
7975 	if (mddev->degraded && !test_bit(MD_BROKEN, &mddev->flags))
7976 		set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
7977 	sysfs_notify_dirent_safe(rdev->sysfs_state);
7978 	set_bit(MD_RECOVERY_INTR, &mddev->recovery);
7979 	if (!test_bit(MD_BROKEN, &mddev->flags)) {
7980 		set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7981 		md_wakeup_thread(mddev->thread);
7982 	}
7983 	if (mddev->event_work.func)
7984 		queue_work(md_misc_wq, &mddev->event_work);
7985 	md_new_event();
7986 }
7987 EXPORT_SYMBOL(md_error);
7988 
7989 /* seq_file implementation /proc/mdstat */
7990 
status_unused(struct seq_file * seq)7991 static void status_unused(struct seq_file *seq)
7992 {
7993 	int i = 0;
7994 	struct md_rdev *rdev;
7995 
7996 	seq_printf(seq, "unused devices: ");
7997 
7998 	list_for_each_entry(rdev, &pending_raid_disks, same_set) {
7999 		i++;
8000 		seq_printf(seq, "%pg ", rdev->bdev);
8001 	}
8002 	if (!i)
8003 		seq_printf(seq, "<none>");
8004 
8005 	seq_printf(seq, "\n");
8006 }
8007 
status_resync(struct seq_file * seq,struct mddev * mddev)8008 static int status_resync(struct seq_file *seq, struct mddev *mddev)
8009 {
8010 	sector_t max_sectors, resync, res;
8011 	unsigned long dt, db = 0;
8012 	sector_t rt, curr_mark_cnt, resync_mark_cnt;
8013 	int scale, recovery_active;
8014 	unsigned int per_milli;
8015 
8016 	if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ||
8017 	    test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
8018 		max_sectors = mddev->resync_max_sectors;
8019 	else
8020 		max_sectors = mddev->dev_sectors;
8021 
8022 	resync = mddev->curr_resync;
8023 	if (resync <= 3) {
8024 		if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
8025 			/* Still cleaning up */
8026 			resync = max_sectors;
8027 	} else if (resync > max_sectors)
8028 		resync = max_sectors;
8029 	else
8030 		resync -= atomic_read(&mddev->recovery_active);
8031 
8032 	if (resync == 0) {
8033 		if (test_bit(MD_RESYNCING_REMOTE, &mddev->recovery)) {
8034 			struct md_rdev *rdev;
8035 
8036 			rdev_for_each(rdev, mddev)
8037 				if (rdev->raid_disk >= 0 &&
8038 				    !test_bit(Faulty, &rdev->flags) &&
8039 				    rdev->recovery_offset != MaxSector &&
8040 				    rdev->recovery_offset) {
8041 					seq_printf(seq, "\trecover=REMOTE");
8042 					return 1;
8043 				}
8044 			if (mddev->reshape_position != MaxSector)
8045 				seq_printf(seq, "\treshape=REMOTE");
8046 			else
8047 				seq_printf(seq, "\tresync=REMOTE");
8048 			return 1;
8049 		}
8050 		if (mddev->recovery_cp < MaxSector) {
8051 			seq_printf(seq, "\tresync=PENDING");
8052 			return 1;
8053 		}
8054 		return 0;
8055 	}
8056 	if (resync < 3) {
8057 		seq_printf(seq, "\tresync=DELAYED");
8058 		return 1;
8059 	}
8060 
8061 	WARN_ON(max_sectors == 0);
8062 	/* Pick 'scale' such that (resync>>scale)*1000 will fit
8063 	 * in a sector_t, and (max_sectors>>scale) will fit in a
8064 	 * u32, as those are the requirements for sector_div.
8065 	 * Thus 'scale' must be at least 10
8066 	 */
8067 	scale = 10;
8068 	if (sizeof(sector_t) > sizeof(unsigned long)) {
8069 		while ( max_sectors/2 > (1ULL<<(scale+32)))
8070 			scale++;
8071 	}
8072 	res = (resync>>scale)*1000;
8073 	sector_div(res, (u32)((max_sectors>>scale)+1));
8074 
8075 	per_milli = res;
8076 	{
8077 		int i, x = per_milli/50, y = 20-x;
8078 		seq_printf(seq, "[");
8079 		for (i = 0; i < x; i++)
8080 			seq_printf(seq, "=");
8081 		seq_printf(seq, ">");
8082 		for (i = 0; i < y; i++)
8083 			seq_printf(seq, ".");
8084 		seq_printf(seq, "] ");
8085 	}
8086 	seq_printf(seq, " %s =%3u.%u%% (%llu/%llu)",
8087 		   (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)?
8088 		    "reshape" :
8089 		    (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)?
8090 		     "check" :
8091 		     (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
8092 		      "resync" : "recovery"))),
8093 		   per_milli/10, per_milli % 10,
8094 		   (unsigned long long) resync/2,
8095 		   (unsigned long long) max_sectors/2);
8096 
8097 	/*
8098 	 * dt: time from mark until now
8099 	 * db: blocks written from mark until now
8100 	 * rt: remaining time
8101 	 *
8102 	 * rt is a sector_t, which is always 64bit now. We are keeping
8103 	 * the original algorithm, but it is not really necessary.
8104 	 *
8105 	 * Original algorithm:
8106 	 *   So we divide before multiply in case it is 32bit and close
8107 	 *   to the limit.
8108 	 *   We scale the divisor (db) by 32 to avoid losing precision
8109 	 *   near the end of resync when the number of remaining sectors
8110 	 *   is close to 'db'.
8111 	 *   We then divide rt by 32 after multiplying by db to compensate.
8112 	 *   The '+1' avoids division by zero if db is very small.
8113 	 */
8114 	dt = ((jiffies - mddev->resync_mark) / HZ);
8115 	if (!dt) dt++;
8116 
8117 	curr_mark_cnt = mddev->curr_mark_cnt;
8118 	recovery_active = atomic_read(&mddev->recovery_active);
8119 	resync_mark_cnt = mddev->resync_mark_cnt;
8120 
8121 	if (curr_mark_cnt >= (recovery_active + resync_mark_cnt))
8122 		db = curr_mark_cnt - (recovery_active + resync_mark_cnt);
8123 
8124 	rt = max_sectors - resync;    /* number of remaining sectors */
8125 	rt = div64_u64(rt, db/32+1);
8126 	rt *= dt;
8127 	rt >>= 5;
8128 
8129 	seq_printf(seq, " finish=%lu.%lumin", (unsigned long)rt / 60,
8130 		   ((unsigned long)rt % 60)/6);
8131 
8132 	seq_printf(seq, " speed=%ldK/sec", db/2/dt);
8133 	return 1;
8134 }
8135 
md_seq_start(struct seq_file * seq,loff_t * pos)8136 static void *md_seq_start(struct seq_file *seq, loff_t *pos)
8137 {
8138 	struct list_head *tmp;
8139 	loff_t l = *pos;
8140 	struct mddev *mddev;
8141 
8142 	if (l == 0x10000) {
8143 		++*pos;
8144 		return (void *)2;
8145 	}
8146 	if (l > 0x10000)
8147 		return NULL;
8148 	if (!l--)
8149 		/* header */
8150 		return (void*)1;
8151 
8152 	spin_lock(&all_mddevs_lock);
8153 	list_for_each(tmp,&all_mddevs)
8154 		if (!l--) {
8155 			mddev = list_entry(tmp, struct mddev, all_mddevs);
8156 			mddev_get(mddev);
8157 			spin_unlock(&all_mddevs_lock);
8158 			return mddev;
8159 		}
8160 	spin_unlock(&all_mddevs_lock);
8161 	if (!l--)
8162 		return (void*)2;/* tail */
8163 	return NULL;
8164 }
8165 
md_seq_next(struct seq_file * seq,void * v,loff_t * pos)8166 static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
8167 {
8168 	struct list_head *tmp;
8169 	struct mddev *next_mddev, *mddev = v;
8170 
8171 	++*pos;
8172 	if (v == (void*)2)
8173 		return NULL;
8174 
8175 	spin_lock(&all_mddevs_lock);
8176 	if (v == (void*)1)
8177 		tmp = all_mddevs.next;
8178 	else
8179 		tmp = mddev->all_mddevs.next;
8180 	if (tmp != &all_mddevs)
8181 		next_mddev = mddev_get(list_entry(tmp,struct mddev,all_mddevs));
8182 	else {
8183 		next_mddev = (void*)2;
8184 		*pos = 0x10000;
8185 	}
8186 	spin_unlock(&all_mddevs_lock);
8187 
8188 	if (v != (void*)1)
8189 		mddev_put(mddev);
8190 	return next_mddev;
8191 
8192 }
8193 
md_seq_stop(struct seq_file * seq,void * v)8194 static void md_seq_stop(struct seq_file *seq, void *v)
8195 {
8196 	struct mddev *mddev = v;
8197 
8198 	if (mddev && v != (void*)1 && v != (void*)2)
8199 		mddev_put(mddev);
8200 }
8201 
md_seq_show(struct seq_file * seq,void * v)8202 static int md_seq_show(struct seq_file *seq, void *v)
8203 {
8204 	struct mddev *mddev = v;
8205 	sector_t sectors;
8206 	struct md_rdev *rdev;
8207 
8208 	if (v == (void*)1) {
8209 		struct md_personality *pers;
8210 		seq_printf(seq, "Personalities : ");
8211 		spin_lock(&pers_lock);
8212 		list_for_each_entry(pers, &pers_list, list)
8213 			seq_printf(seq, "[%s] ", pers->name);
8214 
8215 		spin_unlock(&pers_lock);
8216 		seq_printf(seq, "\n");
8217 		seq->poll_event = atomic_read(&md_event_count);
8218 		return 0;
8219 	}
8220 	if (v == (void*)2) {
8221 		status_unused(seq);
8222 		return 0;
8223 	}
8224 
8225 	spin_lock(&mddev->lock);
8226 	if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
8227 		seq_printf(seq, "%s : %sactive", mdname(mddev),
8228 						mddev->pers ? "" : "in");
8229 		if (mddev->pers) {
8230 			if (mddev->ro==1)
8231 				seq_printf(seq, " (read-only)");
8232 			if (mddev->ro==2)
8233 				seq_printf(seq, " (auto-read-only)");
8234 			seq_printf(seq, " %s", mddev->pers->name);
8235 		}
8236 
8237 		sectors = 0;
8238 		rcu_read_lock();
8239 		rdev_for_each_rcu(rdev, mddev) {
8240 			seq_printf(seq, " %pg[%d]", rdev->bdev, rdev->desc_nr);
8241 
8242 			if (test_bit(WriteMostly, &rdev->flags))
8243 				seq_printf(seq, "(W)");
8244 			if (test_bit(Journal, &rdev->flags))
8245 				seq_printf(seq, "(J)");
8246 			if (test_bit(Faulty, &rdev->flags)) {
8247 				seq_printf(seq, "(F)");
8248 				continue;
8249 			}
8250 			if (rdev->raid_disk < 0)
8251 				seq_printf(seq, "(S)"); /* spare */
8252 			if (test_bit(Replacement, &rdev->flags))
8253 				seq_printf(seq, "(R)");
8254 			sectors += rdev->sectors;
8255 		}
8256 		rcu_read_unlock();
8257 
8258 		if (!list_empty(&mddev->disks)) {
8259 			if (mddev->pers)
8260 				seq_printf(seq, "\n      %llu blocks",
8261 					   (unsigned long long)
8262 					   mddev->array_sectors / 2);
8263 			else
8264 				seq_printf(seq, "\n      %llu blocks",
8265 					   (unsigned long long)sectors / 2);
8266 		}
8267 		if (mddev->persistent) {
8268 			if (mddev->major_version != 0 ||
8269 			    mddev->minor_version != 90) {
8270 				seq_printf(seq," super %d.%d",
8271 					   mddev->major_version,
8272 					   mddev->minor_version);
8273 			}
8274 		} else if (mddev->external)
8275 			seq_printf(seq, " super external:%s",
8276 				   mddev->metadata_type);
8277 		else
8278 			seq_printf(seq, " super non-persistent");
8279 
8280 		if (mddev->pers) {
8281 			mddev->pers->status(seq, mddev);
8282 			seq_printf(seq, "\n      ");
8283 			if (mddev->pers->sync_request) {
8284 				if (status_resync(seq, mddev))
8285 					seq_printf(seq, "\n      ");
8286 			}
8287 		} else
8288 			seq_printf(seq, "\n       ");
8289 
8290 		md_bitmap_status(seq, mddev->bitmap);
8291 
8292 		seq_printf(seq, "\n");
8293 	}
8294 	spin_unlock(&mddev->lock);
8295 
8296 	return 0;
8297 }
8298 
8299 static const struct seq_operations md_seq_ops = {
8300 	.start  = md_seq_start,
8301 	.next   = md_seq_next,
8302 	.stop   = md_seq_stop,
8303 	.show   = md_seq_show,
8304 };
8305 
md_seq_open(struct inode * inode,struct file * file)8306 static int md_seq_open(struct inode *inode, struct file *file)
8307 {
8308 	struct seq_file *seq;
8309 	int error;
8310 
8311 	error = seq_open(file, &md_seq_ops);
8312 	if (error)
8313 		return error;
8314 
8315 	seq = file->private_data;
8316 	seq->poll_event = atomic_read(&md_event_count);
8317 	return error;
8318 }
8319 
8320 static int md_unloading;
mdstat_poll(struct file * filp,poll_table * wait)8321 static __poll_t mdstat_poll(struct file *filp, poll_table *wait)
8322 {
8323 	struct seq_file *seq = filp->private_data;
8324 	__poll_t mask;
8325 
8326 	if (md_unloading)
8327 		return EPOLLIN|EPOLLRDNORM|EPOLLERR|EPOLLPRI;
8328 	poll_wait(filp, &md_event_waiters, wait);
8329 
8330 	/* always allow read */
8331 	mask = EPOLLIN | EPOLLRDNORM;
8332 
8333 	if (seq->poll_event != atomic_read(&md_event_count))
8334 		mask |= EPOLLERR | EPOLLPRI;
8335 	return mask;
8336 }
8337 
8338 static const struct proc_ops mdstat_proc_ops = {
8339 	.proc_open	= md_seq_open,
8340 	.proc_read	= seq_read,
8341 	.proc_lseek	= seq_lseek,
8342 	.proc_release	= seq_release,
8343 	.proc_poll	= mdstat_poll,
8344 };
8345 
register_md_personality(struct md_personality * p)8346 int register_md_personality(struct md_personality *p)
8347 {
8348 	pr_debug("md: %s personality registered for level %d\n",
8349 		 p->name, p->level);
8350 	spin_lock(&pers_lock);
8351 	list_add_tail(&p->list, &pers_list);
8352 	spin_unlock(&pers_lock);
8353 	return 0;
8354 }
8355 EXPORT_SYMBOL(register_md_personality);
8356 
unregister_md_personality(struct md_personality * p)8357 int unregister_md_personality(struct md_personality *p)
8358 {
8359 	pr_debug("md: %s personality unregistered\n", p->name);
8360 	spin_lock(&pers_lock);
8361 	list_del_init(&p->list);
8362 	spin_unlock(&pers_lock);
8363 	return 0;
8364 }
8365 EXPORT_SYMBOL(unregister_md_personality);
8366 
register_md_cluster_operations(struct md_cluster_operations * ops,struct module * module)8367 int register_md_cluster_operations(struct md_cluster_operations *ops,
8368 				   struct module *module)
8369 {
8370 	int ret = 0;
8371 	spin_lock(&pers_lock);
8372 	if (md_cluster_ops != NULL)
8373 		ret = -EALREADY;
8374 	else {
8375 		md_cluster_ops = ops;
8376 		md_cluster_mod = module;
8377 	}
8378 	spin_unlock(&pers_lock);
8379 	return ret;
8380 }
8381 EXPORT_SYMBOL(register_md_cluster_operations);
8382 
unregister_md_cluster_operations(void)8383 int unregister_md_cluster_operations(void)
8384 {
8385 	spin_lock(&pers_lock);
8386 	md_cluster_ops = NULL;
8387 	spin_unlock(&pers_lock);
8388 	return 0;
8389 }
8390 EXPORT_SYMBOL(unregister_md_cluster_operations);
8391 
md_setup_cluster(struct mddev * mddev,int nodes)8392 int md_setup_cluster(struct mddev *mddev, int nodes)
8393 {
8394 	int ret;
8395 	if (!md_cluster_ops)
8396 		request_module("md-cluster");
8397 	spin_lock(&pers_lock);
8398 	/* ensure module won't be unloaded */
8399 	if (!md_cluster_ops || !try_module_get(md_cluster_mod)) {
8400 		pr_warn("can't find md-cluster module or get its reference.\n");
8401 		spin_unlock(&pers_lock);
8402 		return -ENOENT;
8403 	}
8404 	spin_unlock(&pers_lock);
8405 
8406 	ret = md_cluster_ops->join(mddev, nodes);
8407 	if (!ret)
8408 		mddev->safemode_delay = 0;
8409 	return ret;
8410 }
8411 
md_cluster_stop(struct mddev * mddev)8412 void md_cluster_stop(struct mddev *mddev)
8413 {
8414 	if (!md_cluster_ops)
8415 		return;
8416 	md_cluster_ops->leave(mddev);
8417 	module_put(md_cluster_mod);
8418 }
8419 
is_mddev_idle(struct mddev * mddev,int init)8420 static int is_mddev_idle(struct mddev *mddev, int init)
8421 {
8422 	struct md_rdev *rdev;
8423 	int idle;
8424 	int curr_events;
8425 
8426 	idle = 1;
8427 	rcu_read_lock();
8428 	rdev_for_each_rcu(rdev, mddev) {
8429 		struct gendisk *disk = rdev->bdev->bd_disk;
8430 		curr_events = (int)part_stat_read_accum(disk->part0, sectors) -
8431 			      atomic_read(&disk->sync_io);
8432 		/* sync IO will cause sync_io to increase before the disk_stats
8433 		 * as sync_io is counted when a request starts, and
8434 		 * disk_stats is counted when it completes.
8435 		 * So resync activity will cause curr_events to be smaller than
8436 		 * when there was no such activity.
8437 		 * non-sync IO will cause disk_stat to increase without
8438 		 * increasing sync_io so curr_events will (eventually)
8439 		 * be larger than it was before.  Once it becomes
8440 		 * substantially larger, the test below will cause
8441 		 * the array to appear non-idle, and resync will slow
8442 		 * down.
8443 		 * If there is a lot of outstanding resync activity when
8444 		 * we set last_event to curr_events, then all that activity
8445 		 * completing might cause the array to appear non-idle
8446 		 * and resync will be slowed down even though there might
8447 		 * not have been non-resync activity.  This will only
8448 		 * happen once though.  'last_events' will soon reflect
8449 		 * the state where there is little or no outstanding
8450 		 * resync requests, and further resync activity will
8451 		 * always make curr_events less than last_events.
8452 		 *
8453 		 */
8454 		if (init || curr_events - rdev->last_events > 64) {
8455 			rdev->last_events = curr_events;
8456 			idle = 0;
8457 		}
8458 	}
8459 	rcu_read_unlock();
8460 	return idle;
8461 }
8462 
md_done_sync(struct mddev * mddev,int blocks,int ok)8463 void md_done_sync(struct mddev *mddev, int blocks, int ok)
8464 {
8465 	/* another "blocks" (512byte) blocks have been synced */
8466 	atomic_sub(blocks, &mddev->recovery_active);
8467 	wake_up(&mddev->recovery_wait);
8468 	if (!ok) {
8469 		set_bit(MD_RECOVERY_INTR, &mddev->recovery);
8470 		set_bit(MD_RECOVERY_ERROR, &mddev->recovery);
8471 		md_wakeup_thread(mddev->thread);
8472 		// stop recovery, signal do_sync ....
8473 	}
8474 }
8475 EXPORT_SYMBOL(md_done_sync);
8476 
8477 /* md_write_start(mddev, bi)
8478  * If we need to update some array metadata (e.g. 'active' flag
8479  * in superblock) before writing, schedule a superblock update
8480  * and wait for it to complete.
8481  * A return value of 'false' means that the write wasn't recorded
8482  * and cannot proceed as the array is being suspend.
8483  */
md_write_start(struct mddev * mddev,struct bio * bi)8484 bool md_write_start(struct mddev *mddev, struct bio *bi)
8485 {
8486 	int did_change = 0;
8487 
8488 	if (bio_data_dir(bi) != WRITE)
8489 		return true;
8490 
8491 	BUG_ON(mddev->ro == 1);
8492 	if (mddev->ro == 2) {
8493 		/* need to switch to read/write */
8494 		mddev->ro = 0;
8495 		set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
8496 		md_wakeup_thread(mddev->thread);
8497 		md_wakeup_thread(mddev->sync_thread);
8498 		did_change = 1;
8499 	}
8500 	rcu_read_lock();
8501 	percpu_ref_get(&mddev->writes_pending);
8502 	smp_mb(); /* Match smp_mb in set_in_sync() */
8503 	if (mddev->safemode == 1)
8504 		mddev->safemode = 0;
8505 	/* sync_checkers is always 0 when writes_pending is in per-cpu mode */
8506 	if (mddev->in_sync || mddev->sync_checkers) {
8507 		spin_lock(&mddev->lock);
8508 		if (mddev->in_sync) {
8509 			mddev->in_sync = 0;
8510 			set_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags);
8511 			set_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags);
8512 			md_wakeup_thread(mddev->thread);
8513 			did_change = 1;
8514 		}
8515 		spin_unlock(&mddev->lock);
8516 	}
8517 	rcu_read_unlock();
8518 	if (did_change)
8519 		sysfs_notify_dirent_safe(mddev->sysfs_state);
8520 	if (!mddev->has_superblocks)
8521 		return true;
8522 	wait_event(mddev->sb_wait,
8523 		   !test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags) ||
8524 		   mddev->suspended);
8525 	if (test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags)) {
8526 		percpu_ref_put(&mddev->writes_pending);
8527 		return false;
8528 	}
8529 	return true;
8530 }
8531 EXPORT_SYMBOL(md_write_start);
8532 
8533 /* md_write_inc can only be called when md_write_start() has
8534  * already been called at least once of the current request.
8535  * It increments the counter and is useful when a single request
8536  * is split into several parts.  Each part causes an increment and
8537  * so needs a matching md_write_end().
8538  * Unlike md_write_start(), it is safe to call md_write_inc() inside
8539  * a spinlocked region.
8540  */
md_write_inc(struct mddev * mddev,struct bio * bi)8541 void md_write_inc(struct mddev *mddev, struct bio *bi)
8542 {
8543 	if (bio_data_dir(bi) != WRITE)
8544 		return;
8545 	WARN_ON_ONCE(mddev->in_sync || mddev->ro);
8546 	percpu_ref_get(&mddev->writes_pending);
8547 }
8548 EXPORT_SYMBOL(md_write_inc);
8549 
md_write_end(struct mddev * mddev)8550 void md_write_end(struct mddev *mddev)
8551 {
8552 	percpu_ref_put(&mddev->writes_pending);
8553 
8554 	if (mddev->safemode == 2)
8555 		md_wakeup_thread(mddev->thread);
8556 	else if (mddev->safemode_delay)
8557 		/* The roundup() ensures this only performs locking once
8558 		 * every ->safemode_delay jiffies
8559 		 */
8560 		mod_timer(&mddev->safemode_timer,
8561 			  roundup(jiffies, mddev->safemode_delay) +
8562 			  mddev->safemode_delay);
8563 }
8564 
8565 EXPORT_SYMBOL(md_write_end);
8566 
8567 /* 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)8568 void md_submit_discard_bio(struct mddev *mddev, struct md_rdev *rdev,
8569 			struct bio *bio, sector_t start, sector_t size)
8570 {
8571 	struct bio *discard_bio = NULL;
8572 
8573 	if (__blkdev_issue_discard(rdev->bdev, start, size, GFP_NOIO,
8574 			&discard_bio) || !discard_bio)
8575 		return;
8576 
8577 	bio_chain(discard_bio, bio);
8578 	bio_clone_blkg_association(discard_bio, bio);
8579 	if (mddev->gendisk)
8580 		trace_block_bio_remap(discard_bio,
8581 				disk_devt(mddev->gendisk),
8582 				bio->bi_iter.bi_sector);
8583 	submit_bio_noacct(discard_bio);
8584 }
8585 EXPORT_SYMBOL_GPL(md_submit_discard_bio);
8586 
acct_bioset_init(struct mddev * mddev)8587 int acct_bioset_init(struct mddev *mddev)
8588 {
8589 	int err = 0;
8590 
8591 	if (!bioset_initialized(&mddev->io_acct_set))
8592 		err = bioset_init(&mddev->io_acct_set, BIO_POOL_SIZE,
8593 			offsetof(struct md_io_acct, bio_clone), 0);
8594 	return err;
8595 }
8596 EXPORT_SYMBOL_GPL(acct_bioset_init);
8597 
acct_bioset_exit(struct mddev * mddev)8598 void acct_bioset_exit(struct mddev *mddev)
8599 {
8600 	bioset_exit(&mddev->io_acct_set);
8601 }
8602 EXPORT_SYMBOL_GPL(acct_bioset_exit);
8603 
md_end_io_acct(struct bio * bio)8604 static void md_end_io_acct(struct bio *bio)
8605 {
8606 	struct md_io_acct *md_io_acct = bio->bi_private;
8607 	struct bio *orig_bio = md_io_acct->orig_bio;
8608 
8609 	orig_bio->bi_status = bio->bi_status;
8610 
8611 	bio_end_io_acct(orig_bio, md_io_acct->start_time);
8612 	bio_put(bio);
8613 	bio_endio(orig_bio);
8614 }
8615 
8616 /*
8617  * Used by personalities that don't already clone the bio and thus can't
8618  * easily add the timestamp to their extended bio structure.
8619  */
md_account_bio(struct mddev * mddev,struct bio ** bio)8620 void md_account_bio(struct mddev *mddev, struct bio **bio)
8621 {
8622 	struct block_device *bdev = (*bio)->bi_bdev;
8623 	struct md_io_acct *md_io_acct;
8624 	struct bio *clone;
8625 
8626 	if (!blk_queue_io_stat(bdev->bd_disk->queue))
8627 		return;
8628 
8629 	clone = bio_alloc_clone(bdev, *bio, GFP_NOIO, &mddev->io_acct_set);
8630 	md_io_acct = container_of(clone, struct md_io_acct, bio_clone);
8631 	md_io_acct->orig_bio = *bio;
8632 	md_io_acct->start_time = bio_start_io_acct(*bio);
8633 
8634 	clone->bi_end_io = md_end_io_acct;
8635 	clone->bi_private = md_io_acct;
8636 	*bio = clone;
8637 }
8638 EXPORT_SYMBOL_GPL(md_account_bio);
8639 
8640 /* md_allow_write(mddev)
8641  * Calling this ensures that the array is marked 'active' so that writes
8642  * may proceed without blocking.  It is important to call this before
8643  * attempting a GFP_KERNEL allocation while holding the mddev lock.
8644  * Must be called with mddev_lock held.
8645  */
md_allow_write(struct mddev * mddev)8646 void md_allow_write(struct mddev *mddev)
8647 {
8648 	if (!mddev->pers)
8649 		return;
8650 	if (mddev->ro)
8651 		return;
8652 	if (!mddev->pers->sync_request)
8653 		return;
8654 
8655 	spin_lock(&mddev->lock);
8656 	if (mddev->in_sync) {
8657 		mddev->in_sync = 0;
8658 		set_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags);
8659 		set_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags);
8660 		if (mddev->safemode_delay &&
8661 		    mddev->safemode == 0)
8662 			mddev->safemode = 1;
8663 		spin_unlock(&mddev->lock);
8664 		md_update_sb(mddev, 0);
8665 		sysfs_notify_dirent_safe(mddev->sysfs_state);
8666 		/* wait for the dirty state to be recorded in the metadata */
8667 		wait_event(mddev->sb_wait,
8668 			   !test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags));
8669 	} else
8670 		spin_unlock(&mddev->lock);
8671 }
8672 EXPORT_SYMBOL_GPL(md_allow_write);
8673 
8674 #define SYNC_MARKS	10
8675 #define	SYNC_MARK_STEP	(3*HZ)
8676 #define UPDATE_FREQUENCY (5*60*HZ)
md_do_sync(struct md_thread * thread)8677 void md_do_sync(struct md_thread *thread)
8678 {
8679 	struct mddev *mddev = thread->mddev;
8680 	struct mddev *mddev2;
8681 	unsigned int currspeed = 0, window;
8682 	sector_t max_sectors,j, io_sectors, recovery_done;
8683 	unsigned long mark[SYNC_MARKS];
8684 	unsigned long update_time;
8685 	sector_t mark_cnt[SYNC_MARKS];
8686 	int last_mark,m;
8687 	struct list_head *tmp;
8688 	sector_t last_check;
8689 	int skipped = 0;
8690 	struct md_rdev *rdev;
8691 	char *desc, *action = NULL;
8692 	struct blk_plug plug;
8693 	int ret;
8694 
8695 	/* just incase thread restarts... */
8696 	if (test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
8697 	    test_bit(MD_RECOVERY_WAIT, &mddev->recovery))
8698 		return;
8699 	if (mddev->ro) {/* never try to sync a read-only array */
8700 		set_bit(MD_RECOVERY_INTR, &mddev->recovery);
8701 		return;
8702 	}
8703 
8704 	if (mddev_is_clustered(mddev)) {
8705 		ret = md_cluster_ops->resync_start(mddev);
8706 		if (ret)
8707 			goto skip;
8708 
8709 		set_bit(MD_CLUSTER_RESYNC_LOCKED, &mddev->flags);
8710 		if (!(test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ||
8711 			test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) ||
8712 			test_bit(MD_RECOVERY_RECOVER, &mddev->recovery))
8713 		     && ((unsigned long long)mddev->curr_resync_completed
8714 			 < (unsigned long long)mddev->resync_max_sectors))
8715 			goto skip;
8716 	}
8717 
8718 	if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
8719 		if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)) {
8720 			desc = "data-check";
8721 			action = "check";
8722 		} else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
8723 			desc = "requested-resync";
8724 			action = "repair";
8725 		} else
8726 			desc = "resync";
8727 	} else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
8728 		desc = "reshape";
8729 	else
8730 		desc = "recovery";
8731 
8732 	mddev->last_sync_action = action ?: desc;
8733 
8734 	/* we overload curr_resync somewhat here.
8735 	 * 0 == not engaged in resync at all
8736 	 * 2 == checking that there is no conflict with another sync
8737 	 * 1 == like 2, but have yielded to allow conflicting resync to
8738 	 *		commence
8739 	 * other == active in resync - this many blocks
8740 	 *
8741 	 * Before starting a resync we must have set curr_resync to
8742 	 * 2, and then checked that every "conflicting" array has curr_resync
8743 	 * less than ours.  When we find one that is the same or higher
8744 	 * we wait on resync_wait.  To avoid deadlock, we reduce curr_resync
8745 	 * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
8746 	 * This will mean we have to start checking from the beginning again.
8747 	 *
8748 	 */
8749 
8750 	do {
8751 		int mddev2_minor = -1;
8752 		mddev->curr_resync = 2;
8753 
8754 	try_again:
8755 		if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
8756 			goto skip;
8757 		for_each_mddev(mddev2, tmp) {
8758 			if (mddev2 == mddev)
8759 				continue;
8760 			if (!mddev->parallel_resync
8761 			&&  mddev2->curr_resync
8762 			&&  match_mddev_units(mddev, mddev2)) {
8763 				DEFINE_WAIT(wq);
8764 				if (mddev < mddev2 && mddev->curr_resync == 2) {
8765 					/* arbitrarily yield */
8766 					mddev->curr_resync = 1;
8767 					wake_up(&resync_wait);
8768 				}
8769 				if (mddev > mddev2 && mddev->curr_resync == 1)
8770 					/* no need to wait here, we can wait the next
8771 					 * time 'round when curr_resync == 2
8772 					 */
8773 					continue;
8774 				/* We need to wait 'interruptible' so as not to
8775 				 * contribute to the load average, and not to
8776 				 * be caught by 'softlockup'
8777 				 */
8778 				prepare_to_wait(&resync_wait, &wq, TASK_INTERRUPTIBLE);
8779 				if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
8780 				    mddev2->curr_resync >= mddev->curr_resync) {
8781 					if (mddev2_minor != mddev2->md_minor) {
8782 						mddev2_minor = mddev2->md_minor;
8783 						pr_info("md: delaying %s of %s until %s has finished (they share one or more physical units)\n",
8784 							desc, mdname(mddev),
8785 							mdname(mddev2));
8786 					}
8787 					mddev_put(mddev2);
8788 					if (signal_pending(current))
8789 						flush_signals(current);
8790 					schedule();
8791 					finish_wait(&resync_wait, &wq);
8792 					goto try_again;
8793 				}
8794 				finish_wait(&resync_wait, &wq);
8795 			}
8796 		}
8797 	} while (mddev->curr_resync < 2);
8798 
8799 	j = 0;
8800 	if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
8801 		/* resync follows the size requested by the personality,
8802 		 * which defaults to physical size, but can be virtual size
8803 		 */
8804 		max_sectors = mddev->resync_max_sectors;
8805 		atomic64_set(&mddev->resync_mismatches, 0);
8806 		/* we don't use the checkpoint if there's a bitmap */
8807 		if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
8808 			j = mddev->resync_min;
8809 		else if (!mddev->bitmap)
8810 			j = mddev->recovery_cp;
8811 
8812 	} else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) {
8813 		max_sectors = mddev->resync_max_sectors;
8814 		/*
8815 		 * If the original node aborts reshaping then we continue the
8816 		 * reshaping, so set j again to avoid restart reshape from the
8817 		 * first beginning
8818 		 */
8819 		if (mddev_is_clustered(mddev) &&
8820 		    mddev->reshape_position != MaxSector)
8821 			j = mddev->reshape_position;
8822 	} else {
8823 		/* recovery follows the physical size of devices */
8824 		max_sectors = mddev->dev_sectors;
8825 		j = MaxSector;
8826 		rcu_read_lock();
8827 		rdev_for_each_rcu(rdev, mddev)
8828 			if (rdev->raid_disk >= 0 &&
8829 			    !test_bit(Journal, &rdev->flags) &&
8830 			    !test_bit(Faulty, &rdev->flags) &&
8831 			    !test_bit(In_sync, &rdev->flags) &&
8832 			    rdev->recovery_offset < j)
8833 				j = rdev->recovery_offset;
8834 		rcu_read_unlock();
8835 
8836 		/* If there is a bitmap, we need to make sure all
8837 		 * writes that started before we added a spare
8838 		 * complete before we start doing a recovery.
8839 		 * Otherwise the write might complete and (via
8840 		 * bitmap_endwrite) set a bit in the bitmap after the
8841 		 * recovery has checked that bit and skipped that
8842 		 * region.
8843 		 */
8844 		if (mddev->bitmap) {
8845 			mddev->pers->quiesce(mddev, 1);
8846 			mddev->pers->quiesce(mddev, 0);
8847 		}
8848 	}
8849 
8850 	pr_info("md: %s of RAID array %s\n", desc, mdname(mddev));
8851 	pr_debug("md: minimum _guaranteed_  speed: %d KB/sec/disk.\n", speed_min(mddev));
8852 	pr_debug("md: using maximum available idle IO bandwidth (but not more than %d KB/sec) for %s.\n",
8853 		 speed_max(mddev), desc);
8854 
8855 	is_mddev_idle(mddev, 1); /* this initializes IO event counters */
8856 
8857 	io_sectors = 0;
8858 	for (m = 0; m < SYNC_MARKS; m++) {
8859 		mark[m] = jiffies;
8860 		mark_cnt[m] = io_sectors;
8861 	}
8862 	last_mark = 0;
8863 	mddev->resync_mark = mark[last_mark];
8864 	mddev->resync_mark_cnt = mark_cnt[last_mark];
8865 
8866 	/*
8867 	 * Tune reconstruction:
8868 	 */
8869 	window = 32 * (PAGE_SIZE / 512);
8870 	pr_debug("md: using %dk window, over a total of %lluk.\n",
8871 		 window/2, (unsigned long long)max_sectors/2);
8872 
8873 	atomic_set(&mddev->recovery_active, 0);
8874 	last_check = 0;
8875 
8876 	if (j>2) {
8877 		pr_debug("md: resuming %s of %s from checkpoint.\n",
8878 			 desc, mdname(mddev));
8879 		mddev->curr_resync = j;
8880 	} else
8881 		mddev->curr_resync = 3; /* no longer delayed */
8882 	mddev->curr_resync_completed = j;
8883 	sysfs_notify_dirent_safe(mddev->sysfs_completed);
8884 	md_new_event();
8885 	update_time = jiffies;
8886 
8887 	blk_start_plug(&plug);
8888 	while (j < max_sectors) {
8889 		sector_t sectors;
8890 
8891 		skipped = 0;
8892 
8893 		if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
8894 		    ((mddev->curr_resync > mddev->curr_resync_completed &&
8895 		      (mddev->curr_resync - mddev->curr_resync_completed)
8896 		      > (max_sectors >> 4)) ||
8897 		     time_after_eq(jiffies, update_time + UPDATE_FREQUENCY) ||
8898 		     (j - mddev->curr_resync_completed)*2
8899 		     >= mddev->resync_max - mddev->curr_resync_completed ||
8900 		     mddev->curr_resync_completed > mddev->resync_max
8901 			    )) {
8902 			/* time to update curr_resync_completed */
8903 			wait_event(mddev->recovery_wait,
8904 				   atomic_read(&mddev->recovery_active) == 0);
8905 			mddev->curr_resync_completed = j;
8906 			if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) &&
8907 			    j > mddev->recovery_cp)
8908 				mddev->recovery_cp = j;
8909 			update_time = jiffies;
8910 			set_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags);
8911 			sysfs_notify_dirent_safe(mddev->sysfs_completed);
8912 		}
8913 
8914 		while (j >= mddev->resync_max &&
8915 		       !test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
8916 			/* As this condition is controlled by user-space,
8917 			 * we can block indefinitely, so use '_interruptible'
8918 			 * to avoid triggering warnings.
8919 			 */
8920 			flush_signals(current); /* just in case */
8921 			wait_event_interruptible(mddev->recovery_wait,
8922 						 mddev->resync_max > j
8923 						 || test_bit(MD_RECOVERY_INTR,
8924 							     &mddev->recovery));
8925 		}
8926 
8927 		if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
8928 			break;
8929 
8930 		sectors = mddev->pers->sync_request(mddev, j, &skipped);
8931 		if (sectors == 0) {
8932 			set_bit(MD_RECOVERY_INTR, &mddev->recovery);
8933 			break;
8934 		}
8935 
8936 		if (!skipped) { /* actual IO requested */
8937 			io_sectors += sectors;
8938 			atomic_add(sectors, &mddev->recovery_active);
8939 		}
8940 
8941 		if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
8942 			break;
8943 
8944 		j += sectors;
8945 		if (j > max_sectors)
8946 			/* when skipping, extra large numbers can be returned. */
8947 			j = max_sectors;
8948 		if (j > 2)
8949 			mddev->curr_resync = j;
8950 		mddev->curr_mark_cnt = io_sectors;
8951 		if (last_check == 0)
8952 			/* this is the earliest that rebuild will be
8953 			 * visible in /proc/mdstat
8954 			 */
8955 			md_new_event();
8956 
8957 		if (last_check + window > io_sectors || j == max_sectors)
8958 			continue;
8959 
8960 		last_check = io_sectors;
8961 	repeat:
8962 		if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) {
8963 			/* step marks */
8964 			int next = (last_mark+1) % SYNC_MARKS;
8965 
8966 			mddev->resync_mark = mark[next];
8967 			mddev->resync_mark_cnt = mark_cnt[next];
8968 			mark[next] = jiffies;
8969 			mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active);
8970 			last_mark = next;
8971 		}
8972 
8973 		if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
8974 			break;
8975 
8976 		/*
8977 		 * this loop exits only if either when we are slower than
8978 		 * the 'hard' speed limit, or the system was IO-idle for
8979 		 * a jiffy.
8980 		 * the system might be non-idle CPU-wise, but we only care
8981 		 * about not overloading the IO subsystem. (things like an
8982 		 * e2fsck being done on the RAID array should execute fast)
8983 		 */
8984 		cond_resched();
8985 
8986 		recovery_done = io_sectors - atomic_read(&mddev->recovery_active);
8987 		currspeed = ((unsigned long)(recovery_done - mddev->resync_mark_cnt))/2
8988 			/((jiffies-mddev->resync_mark)/HZ +1) +1;
8989 
8990 		if (currspeed > speed_min(mddev)) {
8991 			if (currspeed > speed_max(mddev)) {
8992 				msleep(500);
8993 				goto repeat;
8994 			}
8995 			if (!is_mddev_idle(mddev, 0)) {
8996 				/*
8997 				 * Give other IO more of a chance.
8998 				 * The faster the devices, the less we wait.
8999 				 */
9000 				wait_event(mddev->recovery_wait,
9001 					   !atomic_read(&mddev->recovery_active));
9002 			}
9003 		}
9004 	}
9005 	pr_info("md: %s: %s %s.\n",mdname(mddev), desc,
9006 		test_bit(MD_RECOVERY_INTR, &mddev->recovery)
9007 		? "interrupted" : "done");
9008 	/*
9009 	 * this also signals 'finished resyncing' to md_stop
9010 	 */
9011 	blk_finish_plug(&plug);
9012 	wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
9013 
9014 	if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
9015 	    !test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
9016 	    mddev->curr_resync > 3) {
9017 		mddev->curr_resync_completed = mddev->curr_resync;
9018 		sysfs_notify_dirent_safe(mddev->sysfs_completed);
9019 	}
9020 	mddev->pers->sync_request(mddev, max_sectors, &skipped);
9021 
9022 	if (!test_bit(MD_RECOVERY_CHECK, &mddev->recovery) &&
9023 	    mddev->curr_resync > 3) {
9024 		if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
9025 			if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
9026 				if (mddev->curr_resync >= mddev->recovery_cp) {
9027 					pr_debug("md: checkpointing %s of %s.\n",
9028 						 desc, mdname(mddev));
9029 					if (test_bit(MD_RECOVERY_ERROR,
9030 						&mddev->recovery))
9031 						mddev->recovery_cp =
9032 							mddev->curr_resync_completed;
9033 					else
9034 						mddev->recovery_cp =
9035 							mddev->curr_resync;
9036 				}
9037 			} else
9038 				mddev->recovery_cp = MaxSector;
9039 		} else {
9040 			if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery))
9041 				mddev->curr_resync = MaxSector;
9042 			if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
9043 			    test_bit(MD_RECOVERY_RECOVER, &mddev->recovery)) {
9044 				rcu_read_lock();
9045 				rdev_for_each_rcu(rdev, mddev)
9046 					if (rdev->raid_disk >= 0 &&
9047 					    mddev->delta_disks >= 0 &&
9048 					    !test_bit(Journal, &rdev->flags) &&
9049 					    !test_bit(Faulty, &rdev->flags) &&
9050 					    !test_bit(In_sync, &rdev->flags) &&
9051 					    rdev->recovery_offset < mddev->curr_resync)
9052 						rdev->recovery_offset = mddev->curr_resync;
9053 				rcu_read_unlock();
9054 			}
9055 		}
9056 	}
9057  skip:
9058 	/* set CHANGE_PENDING here since maybe another update is needed,
9059 	 * so other nodes are informed. It should be harmless for normal
9060 	 * raid */
9061 	set_mask_bits(&mddev->sb_flags, 0,
9062 		      BIT(MD_SB_CHANGE_PENDING) | BIT(MD_SB_CHANGE_DEVS));
9063 
9064 	if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
9065 			!test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
9066 			mddev->delta_disks > 0 &&
9067 			mddev->pers->finish_reshape &&
9068 			mddev->pers->size &&
9069 			mddev->queue) {
9070 		mddev_lock_nointr(mddev);
9071 		md_set_array_sectors(mddev, mddev->pers->size(mddev, 0, 0));
9072 		mddev_unlock(mddev);
9073 		if (!mddev_is_clustered(mddev))
9074 			set_capacity_and_notify(mddev->gendisk,
9075 						mddev->array_sectors);
9076 	}
9077 
9078 	spin_lock(&mddev->lock);
9079 	if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
9080 		/* We completed so min/max setting can be forgotten if used. */
9081 		if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
9082 			mddev->resync_min = 0;
9083 		mddev->resync_max = MaxSector;
9084 	} else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
9085 		mddev->resync_min = mddev->curr_resync_completed;
9086 	set_bit(MD_RECOVERY_DONE, &mddev->recovery);
9087 	mddev->curr_resync = 0;
9088 	spin_unlock(&mddev->lock);
9089 
9090 	wake_up(&resync_wait);
9091 	md_wakeup_thread(mddev->thread);
9092 	return;
9093 }
9094 EXPORT_SYMBOL_GPL(md_do_sync);
9095 
remove_and_add_spares(struct mddev * mddev,struct md_rdev * this)9096 static int remove_and_add_spares(struct mddev *mddev,
9097 				 struct md_rdev *this)
9098 {
9099 	struct md_rdev *rdev;
9100 	int spares = 0;
9101 	int removed = 0;
9102 	bool remove_some = false;
9103 
9104 	if (this && test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
9105 		/* Mustn't remove devices when resync thread is running */
9106 		return 0;
9107 
9108 	rdev_for_each(rdev, mddev) {
9109 		if ((this == NULL || rdev == this) &&
9110 		    rdev->raid_disk >= 0 &&
9111 		    !test_bit(Blocked, &rdev->flags) &&
9112 		    test_bit(Faulty, &rdev->flags) &&
9113 		    atomic_read(&rdev->nr_pending)==0) {
9114 			/* Faulty non-Blocked devices with nr_pending == 0
9115 			 * never get nr_pending incremented,
9116 			 * never get Faulty cleared, and never get Blocked set.
9117 			 * So we can synchronize_rcu now rather than once per device
9118 			 */
9119 			remove_some = true;
9120 			set_bit(RemoveSynchronized, &rdev->flags);
9121 		}
9122 	}
9123 
9124 	if (remove_some)
9125 		synchronize_rcu();
9126 	rdev_for_each(rdev, mddev) {
9127 		if ((this == NULL || rdev == this) &&
9128 		    rdev->raid_disk >= 0 &&
9129 		    !test_bit(Blocked, &rdev->flags) &&
9130 		    ((test_bit(RemoveSynchronized, &rdev->flags) ||
9131 		     (!test_bit(In_sync, &rdev->flags) &&
9132 		      !test_bit(Journal, &rdev->flags))) &&
9133 		    atomic_read(&rdev->nr_pending)==0)) {
9134 			if (mddev->pers->hot_remove_disk(
9135 				    mddev, rdev) == 0) {
9136 				sysfs_unlink_rdev(mddev, rdev);
9137 				rdev->saved_raid_disk = rdev->raid_disk;
9138 				rdev->raid_disk = -1;
9139 				removed++;
9140 			}
9141 		}
9142 		if (remove_some && test_bit(RemoveSynchronized, &rdev->flags))
9143 			clear_bit(RemoveSynchronized, &rdev->flags);
9144 	}
9145 
9146 	if (removed && mddev->kobj.sd)
9147 		sysfs_notify_dirent_safe(mddev->sysfs_degraded);
9148 
9149 	if (this && removed)
9150 		goto no_add;
9151 
9152 	rdev_for_each(rdev, mddev) {
9153 		if (this && this != rdev)
9154 			continue;
9155 		if (test_bit(Candidate, &rdev->flags))
9156 			continue;
9157 		if (rdev->raid_disk >= 0 &&
9158 		    !test_bit(In_sync, &rdev->flags) &&
9159 		    !test_bit(Journal, &rdev->flags) &&
9160 		    !test_bit(Faulty, &rdev->flags))
9161 			spares++;
9162 		if (rdev->raid_disk >= 0)
9163 			continue;
9164 		if (test_bit(Faulty, &rdev->flags))
9165 			continue;
9166 		if (!test_bit(Journal, &rdev->flags)) {
9167 			if (mddev->ro &&
9168 			    ! (rdev->saved_raid_disk >= 0 &&
9169 			       !test_bit(Bitmap_sync, &rdev->flags)))
9170 				continue;
9171 
9172 			rdev->recovery_offset = 0;
9173 		}
9174 		if (mddev->pers->hot_add_disk(mddev, rdev) == 0) {
9175 			/* failure here is OK */
9176 			sysfs_link_rdev(mddev, rdev);
9177 			if (!test_bit(Journal, &rdev->flags))
9178 				spares++;
9179 			md_new_event();
9180 			set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
9181 		}
9182 	}
9183 no_add:
9184 	if (removed)
9185 		set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
9186 	return spares;
9187 }
9188 
md_start_sync(struct work_struct * ws)9189 static void md_start_sync(struct work_struct *ws)
9190 {
9191 	struct mddev *mddev = container_of(ws, struct mddev, del_work);
9192 
9193 	mddev->sync_thread = md_register_thread(md_do_sync,
9194 						mddev,
9195 						"resync");
9196 	if (!mddev->sync_thread) {
9197 		pr_warn("%s: could not start resync thread...\n",
9198 			mdname(mddev));
9199 		/* leave the spares where they are, it shouldn't hurt */
9200 		clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
9201 		clear_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
9202 		clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
9203 		clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
9204 		clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
9205 		wake_up(&resync_wait);
9206 		if (test_and_clear_bit(MD_RECOVERY_RECOVER,
9207 				       &mddev->recovery))
9208 			if (mddev->sysfs_action)
9209 				sysfs_notify_dirent_safe(mddev->sysfs_action);
9210 	} else
9211 		md_wakeup_thread(mddev->sync_thread);
9212 	sysfs_notify_dirent_safe(mddev->sysfs_action);
9213 	md_new_event();
9214 }
9215 
9216 /*
9217  * This routine is regularly called by all per-raid-array threads to
9218  * deal with generic issues like resync and super-block update.
9219  * Raid personalities that don't have a thread (linear/raid0) do not
9220  * need this as they never do any recovery or update the superblock.
9221  *
9222  * It does not do any resync itself, but rather "forks" off other threads
9223  * to do that as needed.
9224  * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
9225  * "->recovery" and create a thread at ->sync_thread.
9226  * When the thread finishes it sets MD_RECOVERY_DONE
9227  * and wakeups up this thread which will reap the thread and finish up.
9228  * This thread also removes any faulty devices (with nr_pending == 0).
9229  *
9230  * The overall approach is:
9231  *  1/ if the superblock needs updating, update it.
9232  *  2/ If a recovery thread is running, don't do anything else.
9233  *  3/ If recovery has finished, clean up, possibly marking spares active.
9234  *  4/ If there are any faulty devices, remove them.
9235  *  5/ If array is degraded, try to add spares devices
9236  *  6/ If array has spares or is not in-sync, start a resync thread.
9237  */
md_check_recovery(struct mddev * mddev)9238 void md_check_recovery(struct mddev *mddev)
9239 {
9240 	if (test_bit(MD_ALLOW_SB_UPDATE, &mddev->flags) && mddev->sb_flags) {
9241 		/* Write superblock - thread that called mddev_suspend()
9242 		 * holds reconfig_mutex for us.
9243 		 */
9244 		set_bit(MD_UPDATING_SB, &mddev->flags);
9245 		smp_mb__after_atomic();
9246 		if (test_bit(MD_ALLOW_SB_UPDATE, &mddev->flags))
9247 			md_update_sb(mddev, 0);
9248 		clear_bit_unlock(MD_UPDATING_SB, &mddev->flags);
9249 		wake_up(&mddev->sb_wait);
9250 	}
9251 
9252 	if (mddev->suspended)
9253 		return;
9254 
9255 	if (mddev->bitmap)
9256 		md_bitmap_daemon_work(mddev);
9257 
9258 	if (signal_pending(current)) {
9259 		if (mddev->pers->sync_request && !mddev->external) {
9260 			pr_debug("md: %s in immediate safe mode\n",
9261 				 mdname(mddev));
9262 			mddev->safemode = 2;
9263 		}
9264 		flush_signals(current);
9265 	}
9266 
9267 	if (mddev->ro && !test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
9268 		return;
9269 	if ( ! (
9270 		(mddev->sb_flags & ~ (1<<MD_SB_CHANGE_PENDING)) ||
9271 		test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
9272 		test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
9273 		(mddev->external == 0 && mddev->safemode == 1) ||
9274 		(mddev->safemode == 2
9275 		 && !mddev->in_sync && mddev->recovery_cp == MaxSector)
9276 		))
9277 		return;
9278 
9279 	if (mddev_trylock(mddev)) {
9280 		int spares = 0;
9281 		bool try_set_sync = mddev->safemode != 0;
9282 
9283 		if (!mddev->external && mddev->safemode == 1)
9284 			mddev->safemode = 0;
9285 
9286 		if (mddev->ro) {
9287 			struct md_rdev *rdev;
9288 			if (!mddev->external && mddev->in_sync)
9289 				/* 'Blocked' flag not needed as failed devices
9290 				 * will be recorded if array switched to read/write.
9291 				 * Leaving it set will prevent the device
9292 				 * from being removed.
9293 				 */
9294 				rdev_for_each(rdev, mddev)
9295 					clear_bit(Blocked, &rdev->flags);
9296 			/* On a read-only array we can:
9297 			 * - remove failed devices
9298 			 * - add already-in_sync devices if the array itself
9299 			 *   is in-sync.
9300 			 * As we only add devices that are already in-sync,
9301 			 * we can activate the spares immediately.
9302 			 */
9303 			remove_and_add_spares(mddev, NULL);
9304 			/* There is no thread, but we need to call
9305 			 * ->spare_active and clear saved_raid_disk
9306 			 */
9307 			set_bit(MD_RECOVERY_INTR, &mddev->recovery);
9308 			md_reap_sync_thread(mddev);
9309 			clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
9310 			clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
9311 			clear_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags);
9312 			goto unlock;
9313 		}
9314 
9315 		if (mddev_is_clustered(mddev)) {
9316 			struct md_rdev *rdev, *tmp;
9317 			/* kick the device if another node issued a
9318 			 * remove disk.
9319 			 */
9320 			rdev_for_each_safe(rdev, tmp, mddev) {
9321 				if (test_and_clear_bit(ClusterRemove, &rdev->flags) &&
9322 						rdev->raid_disk < 0)
9323 					md_kick_rdev_from_array(rdev);
9324 			}
9325 		}
9326 
9327 		if (try_set_sync && !mddev->external && !mddev->in_sync) {
9328 			spin_lock(&mddev->lock);
9329 			set_in_sync(mddev);
9330 			spin_unlock(&mddev->lock);
9331 		}
9332 
9333 		if (mddev->sb_flags)
9334 			md_update_sb(mddev, 0);
9335 
9336 		if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
9337 		    !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
9338 			/* resync/recovery still happening */
9339 			clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
9340 			goto unlock;
9341 		}
9342 		if (mddev->sync_thread) {
9343 			md_reap_sync_thread(mddev);
9344 			goto unlock;
9345 		}
9346 		/* Set RUNNING before clearing NEEDED to avoid
9347 		 * any transients in the value of "sync_action".
9348 		 */
9349 		mddev->curr_resync_completed = 0;
9350 		spin_lock(&mddev->lock);
9351 		set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
9352 		spin_unlock(&mddev->lock);
9353 		/* Clear some bits that don't mean anything, but
9354 		 * might be left set
9355 		 */
9356 		clear_bit(MD_RECOVERY_INTR, &mddev->recovery);
9357 		clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
9358 
9359 		if (!test_and_clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
9360 		    test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
9361 			goto not_running;
9362 		/* no recovery is running.
9363 		 * remove any failed drives, then
9364 		 * add spares if possible.
9365 		 * Spares are also removed and re-added, to allow
9366 		 * the personality to fail the re-add.
9367 		 */
9368 
9369 		if (mddev->reshape_position != MaxSector) {
9370 			if (mddev->pers->check_reshape == NULL ||
9371 			    mddev->pers->check_reshape(mddev) != 0)
9372 				/* Cannot proceed */
9373 				goto not_running;
9374 			set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
9375 			clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
9376 		} else if ((spares = remove_and_add_spares(mddev, NULL))) {
9377 			clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
9378 			clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
9379 			clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
9380 			set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
9381 		} else if (mddev->recovery_cp < MaxSector) {
9382 			set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
9383 			clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
9384 		} else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
9385 			/* nothing to be done ... */
9386 			goto not_running;
9387 
9388 		if (mddev->pers->sync_request) {
9389 			if (spares) {
9390 				/* We are adding a device or devices to an array
9391 				 * which has the bitmap stored on all devices.
9392 				 * So make sure all bitmap pages get written
9393 				 */
9394 				md_bitmap_write_all(mddev->bitmap);
9395 			}
9396 			INIT_WORK(&mddev->del_work, md_start_sync);
9397 			queue_work(md_misc_wq, &mddev->del_work);
9398 			goto unlock;
9399 		}
9400 	not_running:
9401 		if (!mddev->sync_thread) {
9402 			clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
9403 			wake_up(&resync_wait);
9404 			if (test_and_clear_bit(MD_RECOVERY_RECOVER,
9405 					       &mddev->recovery))
9406 				if (mddev->sysfs_action)
9407 					sysfs_notify_dirent_safe(mddev->sysfs_action);
9408 		}
9409 	unlock:
9410 		wake_up(&mddev->sb_wait);
9411 		mddev_unlock(mddev);
9412 	}
9413 }
9414 EXPORT_SYMBOL(md_check_recovery);
9415 
md_reap_sync_thread(struct mddev * mddev)9416 void md_reap_sync_thread(struct mddev *mddev)
9417 {
9418 	struct md_rdev *rdev;
9419 	sector_t old_dev_sectors = mddev->dev_sectors;
9420 	bool is_reshaped = false;
9421 
9422 	/* resync has finished, collect result */
9423 	md_unregister_thread(&mddev->sync_thread);
9424 	if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
9425 	    !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery) &&
9426 	    mddev->degraded != mddev->raid_disks) {
9427 		/* success...*/
9428 		/* activate any spares */
9429 		if (mddev->pers->spare_active(mddev)) {
9430 			sysfs_notify_dirent_safe(mddev->sysfs_degraded);
9431 			set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
9432 		}
9433 	}
9434 	if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
9435 	    mddev->pers->finish_reshape) {
9436 		mddev->pers->finish_reshape(mddev);
9437 		if (mddev_is_clustered(mddev))
9438 			is_reshaped = true;
9439 	}
9440 
9441 	/* If array is no-longer degraded, then any saved_raid_disk
9442 	 * information must be scrapped.
9443 	 */
9444 	if (!mddev->degraded)
9445 		rdev_for_each(rdev, mddev)
9446 			rdev->saved_raid_disk = -1;
9447 
9448 	md_update_sb(mddev, 1);
9449 	/* MD_SB_CHANGE_PENDING should be cleared by md_update_sb, so we can
9450 	 * call resync_finish here if MD_CLUSTER_RESYNC_LOCKED is set by
9451 	 * clustered raid */
9452 	if (test_and_clear_bit(MD_CLUSTER_RESYNC_LOCKED, &mddev->flags))
9453 		md_cluster_ops->resync_finish(mddev);
9454 	clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
9455 	clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
9456 	clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
9457 	clear_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
9458 	clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
9459 	clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
9460 	/*
9461 	 * We call md_cluster_ops->update_size here because sync_size could
9462 	 * be changed by md_update_sb, and MD_RECOVERY_RESHAPE is cleared,
9463 	 * so it is time to update size across cluster.
9464 	 */
9465 	if (mddev_is_clustered(mddev) && is_reshaped
9466 				      && !test_bit(MD_CLOSING, &mddev->flags))
9467 		md_cluster_ops->update_size(mddev, old_dev_sectors);
9468 	wake_up(&resync_wait);
9469 	/* flag recovery needed just to double check */
9470 	set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
9471 	sysfs_notify_dirent_safe(mddev->sysfs_completed);
9472 	sysfs_notify_dirent_safe(mddev->sysfs_action);
9473 	md_new_event();
9474 	if (mddev->event_work.func)
9475 		queue_work(md_misc_wq, &mddev->event_work);
9476 }
9477 EXPORT_SYMBOL(md_reap_sync_thread);
9478 
md_wait_for_blocked_rdev(struct md_rdev * rdev,struct mddev * mddev)9479 void md_wait_for_blocked_rdev(struct md_rdev *rdev, struct mddev *mddev)
9480 {
9481 	sysfs_notify_dirent_safe(rdev->sysfs_state);
9482 	wait_event_timeout(rdev->blocked_wait,
9483 			   !test_bit(Blocked, &rdev->flags) &&
9484 			   !test_bit(BlockedBadBlocks, &rdev->flags),
9485 			   msecs_to_jiffies(5000));
9486 	rdev_dec_pending(rdev, mddev);
9487 }
9488 EXPORT_SYMBOL(md_wait_for_blocked_rdev);
9489 
md_finish_reshape(struct mddev * mddev)9490 void md_finish_reshape(struct mddev *mddev)
9491 {
9492 	/* called be personality module when reshape completes. */
9493 	struct md_rdev *rdev;
9494 
9495 	rdev_for_each(rdev, mddev) {
9496 		if (rdev->data_offset > rdev->new_data_offset)
9497 			rdev->sectors += rdev->data_offset - rdev->new_data_offset;
9498 		else
9499 			rdev->sectors -= rdev->new_data_offset - rdev->data_offset;
9500 		rdev->data_offset = rdev->new_data_offset;
9501 	}
9502 }
9503 EXPORT_SYMBOL(md_finish_reshape);
9504 
9505 /* Bad block management */
9506 
9507 /* Returns 1 on success, 0 on failure */
rdev_set_badblocks(struct md_rdev * rdev,sector_t s,int sectors,int is_new)9508 int rdev_set_badblocks(struct md_rdev *rdev, sector_t s, int sectors,
9509 		       int is_new)
9510 {
9511 	struct mddev *mddev = rdev->mddev;
9512 	int rv;
9513 	if (is_new)
9514 		s += rdev->new_data_offset;
9515 	else
9516 		s += rdev->data_offset;
9517 	rv = badblocks_set(&rdev->badblocks, s, sectors, 0);
9518 	if (rv == 0) {
9519 		/* Make sure they get written out promptly */
9520 		if (test_bit(ExternalBbl, &rdev->flags))
9521 			sysfs_notify_dirent_safe(rdev->sysfs_unack_badblocks);
9522 		sysfs_notify_dirent_safe(rdev->sysfs_state);
9523 		set_mask_bits(&mddev->sb_flags, 0,
9524 			      BIT(MD_SB_CHANGE_CLEAN) | BIT(MD_SB_CHANGE_PENDING));
9525 		md_wakeup_thread(rdev->mddev->thread);
9526 		return 1;
9527 	} else
9528 		return 0;
9529 }
9530 EXPORT_SYMBOL_GPL(rdev_set_badblocks);
9531 
rdev_clear_badblocks(struct md_rdev * rdev,sector_t s,int sectors,int is_new)9532 int rdev_clear_badblocks(struct md_rdev *rdev, sector_t s, int sectors,
9533 			 int is_new)
9534 {
9535 	int rv;
9536 	if (is_new)
9537 		s += rdev->new_data_offset;
9538 	else
9539 		s += rdev->data_offset;
9540 	rv = badblocks_clear(&rdev->badblocks, s, sectors);
9541 	if ((rv == 0) && test_bit(ExternalBbl, &rdev->flags))
9542 		sysfs_notify_dirent_safe(rdev->sysfs_badblocks);
9543 	return rv;
9544 }
9545 EXPORT_SYMBOL_GPL(rdev_clear_badblocks);
9546 
md_notify_reboot(struct notifier_block * this,unsigned long code,void * x)9547 static int md_notify_reboot(struct notifier_block *this,
9548 			    unsigned long code, void *x)
9549 {
9550 	struct list_head *tmp;
9551 	struct mddev *mddev;
9552 	int need_delay = 0;
9553 
9554 	for_each_mddev(mddev, tmp) {
9555 		if (mddev_trylock(mddev)) {
9556 			if (mddev->pers)
9557 				__md_stop_writes(mddev);
9558 			if (mddev->persistent)
9559 				mddev->safemode = 2;
9560 			mddev_unlock(mddev);
9561 		}
9562 		need_delay = 1;
9563 	}
9564 	/*
9565 	 * certain more exotic SCSI devices are known to be
9566 	 * volatile wrt too early system reboots. While the
9567 	 * right place to handle this issue is the given
9568 	 * driver, we do want to have a safe RAID driver ...
9569 	 */
9570 	if (need_delay)
9571 		msleep(1000);
9572 
9573 	return NOTIFY_DONE;
9574 }
9575 
9576 static struct notifier_block md_notifier = {
9577 	.notifier_call	= md_notify_reboot,
9578 	.next		= NULL,
9579 	.priority	= INT_MAX, /* before any real devices */
9580 };
9581 
md_geninit(void)9582 static void md_geninit(void)
9583 {
9584 	pr_debug("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
9585 
9586 	proc_create("mdstat", S_IRUGO, NULL, &mdstat_proc_ops);
9587 }
9588 
md_init(void)9589 static int __init md_init(void)
9590 {
9591 	int ret = -ENOMEM;
9592 
9593 	md_wq = alloc_workqueue("md", WQ_MEM_RECLAIM, 0);
9594 	if (!md_wq)
9595 		goto err_wq;
9596 
9597 	md_misc_wq = alloc_workqueue("md_misc", 0, 0);
9598 	if (!md_misc_wq)
9599 		goto err_misc_wq;
9600 
9601 	md_rdev_misc_wq = alloc_workqueue("md_rdev_misc", 0, 0);
9602 	if (!md_rdev_misc_wq)
9603 		goto err_rdev_misc_wq;
9604 
9605 	ret = __register_blkdev(MD_MAJOR, "md", md_probe);
9606 	if (ret < 0)
9607 		goto err_md;
9608 
9609 	ret = __register_blkdev(0, "mdp", md_probe);
9610 	if (ret < 0)
9611 		goto err_mdp;
9612 	mdp_major = ret;
9613 
9614 	register_reboot_notifier(&md_notifier);
9615 	raid_table_header = register_sysctl_table(raid_root_table);
9616 
9617 	md_geninit();
9618 	return 0;
9619 
9620 err_mdp:
9621 	unregister_blkdev(MD_MAJOR, "md");
9622 err_md:
9623 	destroy_workqueue(md_rdev_misc_wq);
9624 err_rdev_misc_wq:
9625 	destroy_workqueue(md_misc_wq);
9626 err_misc_wq:
9627 	destroy_workqueue(md_wq);
9628 err_wq:
9629 	return ret;
9630 }
9631 
check_sb_changes(struct mddev * mddev,struct md_rdev * rdev)9632 static void check_sb_changes(struct mddev *mddev, struct md_rdev *rdev)
9633 {
9634 	struct mdp_superblock_1 *sb = page_address(rdev->sb_page);
9635 	struct md_rdev *rdev2, *tmp;
9636 	int role, ret;
9637 
9638 	/*
9639 	 * If size is changed in another node then we need to
9640 	 * do resize as well.
9641 	 */
9642 	if (mddev->dev_sectors != le64_to_cpu(sb->size)) {
9643 		ret = mddev->pers->resize(mddev, le64_to_cpu(sb->size));
9644 		if (ret)
9645 			pr_info("md-cluster: resize failed\n");
9646 		else
9647 			md_bitmap_update_sb(mddev->bitmap);
9648 	}
9649 
9650 	/* Check for change of roles in the active devices */
9651 	rdev_for_each_safe(rdev2, tmp, mddev) {
9652 		if (test_bit(Faulty, &rdev2->flags))
9653 			continue;
9654 
9655 		/* Check if the roles changed */
9656 		role = le16_to_cpu(sb->dev_roles[rdev2->desc_nr]);
9657 
9658 		if (test_bit(Candidate, &rdev2->flags)) {
9659 			if (role == MD_DISK_ROLE_FAULTY) {
9660 				pr_info("md: Removing Candidate device %pg because add failed\n",
9661 					rdev2->bdev);
9662 				md_kick_rdev_from_array(rdev2);
9663 				continue;
9664 			}
9665 			else
9666 				clear_bit(Candidate, &rdev2->flags);
9667 		}
9668 
9669 		if (role != rdev2->raid_disk) {
9670 			/*
9671 			 * got activated except reshape is happening.
9672 			 */
9673 			if (rdev2->raid_disk == -1 && role != MD_DISK_ROLE_SPARE &&
9674 			    !(le32_to_cpu(sb->feature_map) &
9675 			      MD_FEATURE_RESHAPE_ACTIVE)) {
9676 				rdev2->saved_raid_disk = role;
9677 				ret = remove_and_add_spares(mddev, rdev2);
9678 				pr_info("Activated spare: %pg\n",
9679 					rdev2->bdev);
9680 				/* wakeup mddev->thread here, so array could
9681 				 * perform resync with the new activated disk */
9682 				set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
9683 				md_wakeup_thread(mddev->thread);
9684 			}
9685 			/* device faulty
9686 			 * We just want to do the minimum to mark the disk
9687 			 * as faulty. The recovery is performed by the
9688 			 * one who initiated the error.
9689 			 */
9690 			if (role == MD_DISK_ROLE_FAULTY ||
9691 			    role == MD_DISK_ROLE_JOURNAL) {
9692 				md_error(mddev, rdev2);
9693 				clear_bit(Blocked, &rdev2->flags);
9694 			}
9695 		}
9696 	}
9697 
9698 	if (mddev->raid_disks != le32_to_cpu(sb->raid_disks)) {
9699 		ret = update_raid_disks(mddev, le32_to_cpu(sb->raid_disks));
9700 		if (ret)
9701 			pr_warn("md: updating array disks failed. %d\n", ret);
9702 	}
9703 
9704 	/*
9705 	 * Since mddev->delta_disks has already updated in update_raid_disks,
9706 	 * so it is time to check reshape.
9707 	 */
9708 	if (test_bit(MD_RESYNCING_REMOTE, &mddev->recovery) &&
9709 	    (le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) {
9710 		/*
9711 		 * reshape is happening in the remote node, we need to
9712 		 * update reshape_position and call start_reshape.
9713 		 */
9714 		mddev->reshape_position = le64_to_cpu(sb->reshape_position);
9715 		if (mddev->pers->update_reshape_pos)
9716 			mddev->pers->update_reshape_pos(mddev);
9717 		if (mddev->pers->start_reshape)
9718 			mddev->pers->start_reshape(mddev);
9719 	} else if (test_bit(MD_RESYNCING_REMOTE, &mddev->recovery) &&
9720 		   mddev->reshape_position != MaxSector &&
9721 		   !(le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) {
9722 		/* reshape is just done in another node. */
9723 		mddev->reshape_position = MaxSector;
9724 		if (mddev->pers->update_reshape_pos)
9725 			mddev->pers->update_reshape_pos(mddev);
9726 	}
9727 
9728 	/* Finally set the event to be up to date */
9729 	mddev->events = le64_to_cpu(sb->events);
9730 }
9731 
read_rdev(struct mddev * mddev,struct md_rdev * rdev)9732 static int read_rdev(struct mddev *mddev, struct md_rdev *rdev)
9733 {
9734 	int err;
9735 	struct page *swapout = rdev->sb_page;
9736 	struct mdp_superblock_1 *sb;
9737 
9738 	/* Store the sb page of the rdev in the swapout temporary
9739 	 * variable in case we err in the future
9740 	 */
9741 	rdev->sb_page = NULL;
9742 	err = alloc_disk_sb(rdev);
9743 	if (err == 0) {
9744 		ClearPageUptodate(rdev->sb_page);
9745 		rdev->sb_loaded = 0;
9746 		err = super_types[mddev->major_version].
9747 			load_super(rdev, NULL, mddev->minor_version);
9748 	}
9749 	if (err < 0) {
9750 		pr_warn("%s: %d Could not reload rdev(%d) err: %d. Restoring old values\n",
9751 				__func__, __LINE__, rdev->desc_nr, err);
9752 		if (rdev->sb_page)
9753 			put_page(rdev->sb_page);
9754 		rdev->sb_page = swapout;
9755 		rdev->sb_loaded = 1;
9756 		return err;
9757 	}
9758 
9759 	sb = page_address(rdev->sb_page);
9760 	/* Read the offset unconditionally, even if MD_FEATURE_RECOVERY_OFFSET
9761 	 * is not set
9762 	 */
9763 
9764 	if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RECOVERY_OFFSET))
9765 		rdev->recovery_offset = le64_to_cpu(sb->recovery_offset);
9766 
9767 	/* The other node finished recovery, call spare_active to set
9768 	 * device In_sync and mddev->degraded
9769 	 */
9770 	if (rdev->recovery_offset == MaxSector &&
9771 	    !test_bit(In_sync, &rdev->flags) &&
9772 	    mddev->pers->spare_active(mddev))
9773 		sysfs_notify_dirent_safe(mddev->sysfs_degraded);
9774 
9775 	put_page(swapout);
9776 	return 0;
9777 }
9778 
md_reload_sb(struct mddev * mddev,int nr)9779 void md_reload_sb(struct mddev *mddev, int nr)
9780 {
9781 	struct md_rdev *rdev = NULL, *iter;
9782 	int err;
9783 
9784 	/* Find the rdev */
9785 	rdev_for_each_rcu(iter, mddev) {
9786 		if (iter->desc_nr == nr) {
9787 			rdev = iter;
9788 			break;
9789 		}
9790 	}
9791 
9792 	if (!rdev) {
9793 		pr_warn("%s: %d Could not find rdev with nr %d\n", __func__, __LINE__, nr);
9794 		return;
9795 	}
9796 
9797 	err = read_rdev(mddev, rdev);
9798 	if (err < 0)
9799 		return;
9800 
9801 	check_sb_changes(mddev, rdev);
9802 
9803 	/* Read all rdev's to update recovery_offset */
9804 	rdev_for_each_rcu(rdev, mddev) {
9805 		if (!test_bit(Faulty, &rdev->flags))
9806 			read_rdev(mddev, rdev);
9807 	}
9808 }
9809 EXPORT_SYMBOL(md_reload_sb);
9810 
9811 #ifndef MODULE
9812 
9813 /*
9814  * Searches all registered partitions for autorun RAID arrays
9815  * at boot time.
9816  */
9817 
9818 static DEFINE_MUTEX(detected_devices_mutex);
9819 static LIST_HEAD(all_detected_devices);
9820 struct detected_devices_node {
9821 	struct list_head list;
9822 	dev_t dev;
9823 };
9824 
md_autodetect_dev(dev_t dev)9825 void md_autodetect_dev(dev_t dev)
9826 {
9827 	struct detected_devices_node *node_detected_dev;
9828 
9829 	node_detected_dev = kzalloc(sizeof(*node_detected_dev), GFP_KERNEL);
9830 	if (node_detected_dev) {
9831 		node_detected_dev->dev = dev;
9832 		mutex_lock(&detected_devices_mutex);
9833 		list_add_tail(&node_detected_dev->list, &all_detected_devices);
9834 		mutex_unlock(&detected_devices_mutex);
9835 	}
9836 }
9837 
md_autostart_arrays(int part)9838 void md_autostart_arrays(int part)
9839 {
9840 	struct md_rdev *rdev;
9841 	struct detected_devices_node *node_detected_dev;
9842 	dev_t dev;
9843 	int i_scanned, i_passed;
9844 
9845 	i_scanned = 0;
9846 	i_passed = 0;
9847 
9848 	pr_info("md: Autodetecting RAID arrays.\n");
9849 
9850 	mutex_lock(&detected_devices_mutex);
9851 	while (!list_empty(&all_detected_devices) && i_scanned < INT_MAX) {
9852 		i_scanned++;
9853 		node_detected_dev = list_entry(all_detected_devices.next,
9854 					struct detected_devices_node, list);
9855 		list_del(&node_detected_dev->list);
9856 		dev = node_detected_dev->dev;
9857 		kfree(node_detected_dev);
9858 		mutex_unlock(&detected_devices_mutex);
9859 		rdev = md_import_device(dev,0, 90);
9860 		mutex_lock(&detected_devices_mutex);
9861 		if (IS_ERR(rdev))
9862 			continue;
9863 
9864 		if (test_bit(Faulty, &rdev->flags))
9865 			continue;
9866 
9867 		set_bit(AutoDetected, &rdev->flags);
9868 		list_add(&rdev->same_set, &pending_raid_disks);
9869 		i_passed++;
9870 	}
9871 	mutex_unlock(&detected_devices_mutex);
9872 
9873 	pr_debug("md: Scanned %d and added %d devices.\n", i_scanned, i_passed);
9874 
9875 	autorun_devices(part);
9876 }
9877 
9878 #endif /* !MODULE */
9879 
md_exit(void)9880 static __exit void md_exit(void)
9881 {
9882 	struct mddev *mddev;
9883 	struct list_head *tmp;
9884 	int delay = 1;
9885 
9886 	unregister_blkdev(MD_MAJOR,"md");
9887 	unregister_blkdev(mdp_major, "mdp");
9888 	unregister_reboot_notifier(&md_notifier);
9889 	unregister_sysctl_table(raid_table_header);
9890 
9891 	/* We cannot unload the modules while some process is
9892 	 * waiting for us in select() or poll() - wake them up
9893 	 */
9894 	md_unloading = 1;
9895 	while (waitqueue_active(&md_event_waiters)) {
9896 		/* not safe to leave yet */
9897 		wake_up(&md_event_waiters);
9898 		msleep(delay);
9899 		delay += delay;
9900 	}
9901 	remove_proc_entry("mdstat", NULL);
9902 
9903 	for_each_mddev(mddev, tmp) {
9904 		export_array(mddev);
9905 		mddev->ctime = 0;
9906 		mddev->hold_active = 0;
9907 		/*
9908 		 * for_each_mddev() will call mddev_put() at the end of each
9909 		 * iteration.  As the mddev is now fully clear, this will
9910 		 * schedule the mddev for destruction by a workqueue, and the
9911 		 * destroy_workqueue() below will wait for that to complete.
9912 		 */
9913 	}
9914 	destroy_workqueue(md_rdev_misc_wq);
9915 	destroy_workqueue(md_misc_wq);
9916 	destroy_workqueue(md_wq);
9917 }
9918 
9919 subsys_initcall(md_init);
module_exit(md_exit)9920 module_exit(md_exit)
9921 
9922 static int get_ro(char *buffer, const struct kernel_param *kp)
9923 {
9924 	return sprintf(buffer, "%d\n", start_readonly);
9925 }
set_ro(const char * val,const struct kernel_param * kp)9926 static int set_ro(const char *val, const struct kernel_param *kp)
9927 {
9928 	return kstrtouint(val, 10, (unsigned int *)&start_readonly);
9929 }
9930 
9931 module_param_call(start_ro, set_ro, get_ro, NULL, S_IRUSR|S_IWUSR);
9932 module_param(start_dirty_degraded, int, S_IRUGO|S_IWUSR);
9933 module_param_call(new_array, add_named_array, NULL, NULL, S_IWUSR);
9934 module_param(create_on_open, bool, S_IRUSR|S_IWUSR);
9935 
9936 MODULE_LICENSE("GPL");
9937 MODULE_DESCRIPTION("MD RAID framework");
9938 MODULE_ALIAS("md");
9939 MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR);
9940